State of The Art of Machine PDF

energies
Review
State of the Art of Machine Learning Models in
Energy Systems, a Systematic Review
Amir Mosavi 1,2,3 , Mohsen Salimi 4 , Sina Faizollahzadeh Ardabili 5 , Timon Rabczuk 6 ,
Shahaboddin Shamshirband 7,8, * and Annamaria R. Varkonyi-Koczy 9
1 School of the Built Environment, Oxford Brookes University, Oxford OX3 0BP, UK;
amirhosein.mosavi@qut.edu.au
2 Centre for Accident Research Road Safety-Queensland, Queensland University of Technology,
Brisbane QLD 4059, Australia
3 Institute of Automation, Kando Kalman Faculty of Electrical Engineering, Obuda University, Budapest,
Hungary
4 Department of Renewable Energies, Niroo Research Institute, Tehran, Iran; msalimi@nri.ac.ir
5 Biosystem Engineering Department, University of Mohaghegh Ardabili, Ardabil 5619911367, Iran;
sina_fa1990@yahoo.com
6 Department of Computer Engineering, College of Computer and Information Sciences,
King Saud University, Riyadh, Saudi Arabia; timon.rabczuk@uni-weimar.de
7 Department for Management of Science and Technology Development, Ton Duc Thang University,
Ho Chi Minh City, Vietnam
8 Faculty of Information Technology, Ton Duc Thang University, Ho Chi Minh City, Vietnam
9 Department of Mathematics and Informatics, J. Selye University, Komarno 94501, Slovakia;
varkonyi-koczy@uni-obuda.hu
* Correspondence: shahaboddin.shamshirband@tdtu.edu.vn

Received: 12 February 2019; Accepted: 28 March 2019; Published: 4 April 2019
Abstract: Machine learning (ML) models have been widely used in the modeling, design and
prediction in energy systems. During the past two decades, there has been a dramatic increase in
the advancement and application of various types of ML models for energy systems. This paper
presents the state of the art of ML models used in energy systems along with a novel taxonomy
of models and applications. Through a novel methodology, ML models are identified and further
classified according to the ML modeling technique, energy type, and application area. Furthermore,
a comprehensive review of the literature leads to an assessment and performance evaluation of the
ML models and their applications, and a discussion of the major challenges and opportunities for
prospective research. This paper further concludes that there is an outstanding rise in the accuracy,
robustness, precision and generalization ability of the ML models in energy systems using hybrid
ML models. Hybridization is reported to be effective in the advancement of prediction models,
particularly for renewable energy systems, e.g., solar energy, wind energy, and biofuels. Moreover,
the energy demand prediction using hybrid models of ML have highly contributed to the energy
efficiency and therefore energy governance and sustainability.
Keywords: energy systems; machine learning; artificial neural networks (ANN); support vector
machines (SVM); neuro-fuzzy; ANFIS; wavelet neural network (WNN); big data; decision tree (DT);
ensemble; hybrid models; deep learning; blockchain; renewable energy systems; energy informatics;
internet of things (IoT); smart sensors; remote sensing; prediction; forecasting; energy demand
Energies 2019, 12, 1301; doi:10.3390/en12071301 www.mdpi.com/journal/energies

Energies 2019, 12, 1301 2 of 42
1. Introduction
An energy system is a group of organized elements designed for the purpose of generation,
control and/or transformation of energy [1,2]. Energy systems may incorporate combinations of
mechanical, chemical, thermal, and electromagnetical components, covering a wide range of energy
categories including renewables and alternatives [3–5]. The advancement of energy systems is facing
critical decision-making tasks to satisfy numerous demanding and contradictory objectives considering
functional performance, efficiency, financial burden, environmental impact, etc. [6].
The growing utilization of data collectors in energy systems has resulted in a massive amount
of data accumulated. Smart sensors are now extensively used in energy production and energy
consumption [7–9]. Such big data has created a vast number of opportunities and challenges for
informed decision-making [10,11]. ML models have contributed to the implementation of big data
technologies in various applications [12–16]. Since prediction methods based on ML models simplify
the extraction of functional dependencies from observations, such data-driven models have gained
popularity in the energy realm [17–19]. Today, ML models in energy systems are essential for predictive
modeling of production, consumption, and demand analysis due to their accuracy, efficacy and
speed [20,21]. ML models also provide an understanding on energy system functionality in the context
of complex human interactions [22,23]. The use of ML models for conventional energy systems, along
with alternative and renewable energy systems, has been promising [24,25]. Due to the popularity
of the field, many review papers have emerged that present insight into present applications and
future challenges and opportunities [26]. However, the existing review papers either survey the
applications of a single ML model, e.g., ANNs [17], or cover only one energy domain, e.g., solar
radiation forecasting [24]. Consequently, the advancements of ML models and their progress in
various energy systems have not yet been addressed in the literature. Therefore, a comprehensive
review of essential ML models is the main objective of this paper. Consequently, the contribution of
this paper is to present the state of the art of ML models in energy systems and discuss their likely
future trends.
The rest of this paper is organized as follows. In Section 2 the methodology of the research is
presented. In the section three state of the art of ML models in energy systems is presented with an
initial analysis of the database search. The ML models are categorized and the original papers with
high relevance have been reviewed. Each subsection contains a brief discussions and outlook on the
results related to each subject. Section 3 also includes an overview of recently emerged hybrid ML
models. Section 4 focuses on the latest advancement of hybrid ML models in the highly demanding
application areas e.g., solar, wind, and demand energy systems. Finally, in Section 4, an overall
discussion and conclusions are presented.
2. Methodology of Survey
The purpose of the research methodology is to identify, classify and review the notable ML
and DL models used in energy systems. In our comprehensive review, using the Thomson Reuters
Web-of-Science and Elsevier Scopus for implementation of the search queries would ensure that any
paper in the database would meet the essential quality measures, originality, high impact, and high
h-index. Furthermore, to present an in-depth review and understanding of each modeling technique
and its progress, we aimed at having four different categories for the models used in energy systems,
i.e., single ML models, hybrid models, ensemble models, and DL.
Figure 1 demonstrates the methodology of this review. In step 1 of the methodology the initial
database of the relevant articles is identified based on the search queries of: “energy system” and
“machine learning” or “neural network” or “support vector” or “ANFIS” or “WNN” or “DT” or “MLP”
or “ELM” or “ensemble” or “deep learning.” However, for every ML method, we applied a new search
query to suit that search well. These queries will identify the relevant articles, yet the queries are
uncertain whether the ML model belongs to either ensemble or hybrids. Also, some articles in the
initial database might not be relevant at all. For instance, a hybrid or ensemble model of ML may
Energies 2019, 12, 1301 3 of 42
Energies 2019, 12, x FOR PEER REVIEW 3 of 43

include single model(s). For that reason, steps 2 and 3 of the methodology are designed in such a way
to in
in
such athe
classify way to classify
such a wayML modelsthe
to classify the
ML
ML
into themodels
models
into the right
forcategories
into the right
right categories categories
forIn
the
for the
the review. review.
4 theIn
review.
step In
step 4 the
are models
step 4 the
models models
classified
are classified into the four categories and arranged in separate tables to be reviewed individually.
arethe
into classified into the four
four categories and categories
arranged and arranged
in separate in separate
tables tables to be
to be reviewed reviewed individually.
individually.
Figure 1.
Figure 1. Methodology
Methodologyofofresearch.
research.
Figure 1. Methodology of research.
The
The database
database includes
includes 2601
2601 relevant
relevant documents.
documents. During
During stepstep
two two
of theofmethodology
the methodology the
the number
The database includes 2601 relevant documents. During step two of the methodology the
of number
relevantofdocuments
relevant documents
drops to drops
240. Intostep
240. three
In step three
the the documents
documents are reduced
are reduced to 70 to 70 original
original papers
number of relevant documents drops to 240. In step three the documents are reduced to 70 original
to papers to be reviewed
be reviewed in this
in this paper. paper. Analyzing
Analyzing the initialthe initial database
database shows a shows a great increase
great increase
papers to be reviewed in this paper. Analyzing the initial database shows a great increase in the
in the of
in the number
number
articles of articles
using using ML
ML models. models.
Figures Figures
2 and 2 and 3 demonstrate
3 demonstrate the growth the growth in the number
in the number of papers
of papers during
number of articles using ML models. Figures 2 and 3 demonstrate the growth in the number of papers
theduring the past two decades on energy systems that utilized ML and different subjectareas
during the past two decades on energy systems that utilized ML and different subject areas using ML in
past two decades on energy systems that utilized ML and different subject areasusing
using ML
ML
in energy systems, respectively. The increase in the number of documents in energy systems has also
in energy
energy systems,
systems, respectively.
respectively. The The increaseininthe
increase thenumber
number of of documents
documents in inenergy
energysystems
systems hashas
also
also
been duetoto theimplementation
implementation of smartsmart grid systems
systemsand andthe theIoT.
IoT.During
Duringthe past decade, novel MLML
been
been due
due tothe
the implementationof of smart grid systems and the IoT. During the past
the decade,
past decade,novel ML
novel
models includingdeepdeep learning,ensembles,
ensembles, andand hybrids
hybridshave haveemerged
emergedin energy systems. Through
models
models including deeplearning,
including learning, ensembles, and hybrids have emerged ininenergy
energysystems.
systems.Through
Through
introducing novel ML models to energy systems, the increasing number of papers represents a huge
introducing
introducing novelML
novel MLmodels
modelsto toenergy
energy systems,
systems, the
the increasing
increasingnumber
numberofofpapers
papers represents
representsa huge
a huge
experimental effort to explore new opportunities.
experimentaleffort
experimental efforttotoexplore
explorenew
newopportunities.
opportunities.
Figure 2. The growth in the number of articles during the past two decades.
the quantity of the literature on the use of ML in various energy systems (see Figure 2). Considering
the application areas of the database, ML models have been extensively used in diverse applications
of energy systems, especially for predicting electrical energy and renewable energies demand and
consumption. The continued growth of literature can also confirm the great potential of ML models
in energy
Energies systems.
2019, 12, 1301 4 of 42
Figure 3. Subject areas of ML in energy systems.

Figure 3. Subject areas of ML in energy systems.
The initial consideration of the database of identified articles shows an exponential increase in
3. State of theofArt
the quantity theof ML Models
literature inuse
on the Energy
of MLSystems
in various energy systems (see Figure 2). Considering
The methodology identifies 10 major ML models
the application areas of the database, ML models havefrequently used inused
been extensively energy systems,
in diverse i.e., ANN,
applications
MLP, ELM, SVM, WNN, ANFIS, decision trees, deep learning, ensembles, and advanced hybrid and
of energy systems, especially for predicting electrical energy and renewable energies demand ML
models. The notable
consumption. manuscripts
The continued growthareofaccordingly categorized
literature can into
also confirm thethe relevant
great groups
potential and
of ML further
models in
reviewed in this section. Note that, in the presented taxonomy, deep learning as an emerging
energy systems.
modeling technique has been categorized under the ML models. Furthermore, it is worth mentioning
3. State of the Art of ML Models in Energy Systems
that WNN and ANFIS are of a hybrid nature. However, the category of “advanced hybrid ML
models”
The includes only recently
methodology identifiesdeveloped
10 major MLalgorithms.
models frequently used in energy systems, i.e., ANN,
MLP,InELM,
this section the accuracy
SVM, WNN, ANFIS,ofdecision
the ML trees,
modeldeep is furtherlearning, evaluated. Sinceand
ensembles, different ML models
advanced are
hybrid ML
available,
models. The choosing
notableand demonstrating
manuscripts the superiority
are accordingly categorized of theintomodels requires
the relevant comparing
groups their
and further
performance.
reviewed in this Hence, some
section. Notecomparative performance
that, in the presented taxonomy, parameters are required.
deep learning The mostmodeling
as an emerging popular
comparative
technique hasperformance parameters
been categorized under theareMLthe root Furthermore,
models. mean squareit error (RMSE)
is worth and correlation
mentioning that WNN
coefficient
and ANFIS(r), arewhich are used
of a hybrid to indicate
nature. However,the the
error and precision
category of the models
of “advanced [27,28];
hybrid ML Equations
models” (1)
includes
and
only(2) represent
recently these parameters.
developed algorithms.
In this section the accuracy of the ML model is further evaluated. Since different ML models
∑( )
1 N 2
are available, choosing and demonstrating= RMSE the superiority xt i − x pi of the models requires comparing their (1)
performance. Hence, some comparative performance n i =1 parameters are required. The most popular
comparative performance parameters are the root mean square1/2error (RMSE) and correlation coefficient
 2 
(r), which are used to indicate the error and precision ( xt i − x piof) the
n
∑  models [27,28]; Equations (1) and (2)

represent these parameters. r= 1 −  v
i =1
  (2)
 u n  
∑ ( )
2
  N x  
 = t i1=1∑ xti − x pi , 2
u t i
RMSE (1)
n i =1
where xti represents the target value, xpi represents the predicted value, and n is the number of data points.
  n 2 1/2
 i∑
x ti − x pi  
 =1

r= 1 − n
 , (2)
2
  
∑ ( xti )
i =1
where xti represents the target value, xpi represents the predicted value, and n is the number of
data points.
Energies 2019, 12, 1301 5 of 42
3.1. ANN
3.1. ANN
Artificial neural
Artificial neural networks
networks are are frameworks
frameworks for for different
different machine
machine learning
learning algorithms
algorithms to to process
process
complex data
complex data inputs.
inputs. ANNs
ANNscan canbebeutilized forfor
utilized several
severalpurposes
purposes such as forecasting,
such as forecasting, regression and
regression
curve
and fitting.
curve An An
fitting. artificial neural
artificial neuralnetwork
network fundamental
fundamental unit
unitisisa aneuron
neuron that
that utilizes
utilizes aa transfer
transfer
function for
function for the
the output
output formulation.
formulation. The The main
main advantage
advantage of of ANN
ANN models
models is is their
their lower
lower complexity
complexity
for multiple-variable
for multiple-variable problems.
problems. Table
Table 11 demonstrates
demonstrates some some critical
critical papers
papers inin this
this field.
field. Instead
Instead ofof
complicated rules,
complicated rules, ANN
ANN can can learn
learn patterns
patterns ofof crucial
crucial information
information within
within an
an intricate
intricate information
information
domain [4].
domain [4]. Also,
Also, due
due toto the
the noise-immune
noise-immune and and fault-tolerant
fault-tolerant characteristics
characteristics of of ANNs,
ANNs, they they can
can be
be
successfullyused
successfully usedfor
forinherently
inherentlynoisynoisydata
datafrom
fromenergy
energysystems.
systems.
Table1.1.Notable
Table NotableANN
ANNmodels
modelsused
usedin
inenergy
energysystems.
systems.
Year
Year Reference
Reference Journal
Journal Application
Application
Optimization
Optimization ofof renewable
renewable energy
2018 Abbas Abbas et al. [29] Electronics
et al. [29] (Switzerland) energy
2018 Electronics (Switzerland) generation
generation capacities
capacities
IEEE Transactions on Power Mitigation of wind power fluctuation and
2017 Anwar et al. [30] IEEE Transactions on Mitigation of wind power fluctuation and
2017 Anwar et al. [30] Systems
Power Systems scheduling
schedulingstrategies
strategies for
for power
power generation
generation
Boukelia et al.
2017
2017 Boukelia et al. [31] Renewable Energy
Renewable Energy Prediction
Predictionof
of levelized
levelized cost
cost of
of electricity
electricity
[31]
2016 Chatziagorakis et et al. [32]NeuralNeural
Computing and and
Computing AAforecasting
forecasting model
model for
for wind
wind speed and
speed and
2016 Chatziagorakis
Applications hourlyand
and daily
daily solar
solar radiation
radiation
al. [32] Applications hourly
Gallagher et al.
Gallagher et al. [33] EnergyEnergy
Measurement
Measurementand and verification
verification of
of energy
energy
2018
2018 and Buildings
and Buildings
[33] savings in industrial buildings
savings in industrial buildings
Abbasetetal.al.
Abbas (2018)
(2018) [29][29]
usedused
geneticgenetic algorithms
algorithms to optimize
to optimize the generation
the generation capacities
capacities of renewable of
renewable energy systems integrated with storage systems. This study evaluated
energy systems integrated with storage systems. This study evaluated the economic feasibility of the the economic
feasibility ofofthe
introduction introduction
energy of energy
storage systems storage
to the systems
electric to the
grid. The electric
artificial grid.network
neural The artificial
was usedneural
to
network was used to validate the predicted load model. The uncertainties
validate the predicted load model. The uncertainties related to the renewable energy systems were related to the renewable
energy
dealt systems
with usingwere dealt with using amodel.
a chance-constrained chance-constrained
Then, the problem model.wasThen, the problem
solved by geneticwas solved by
algorithms.
The robustness of the proposed model was verified by its application to a case in westerna China.
genetic algorithms. The robustness of the proposed model was verified by its application to case in
western China. Figure 4 reports the results of the proposed method in comparison
Figure 4 reports the results of the proposed method in comparison with the base case in terms of total with the base case
in terms of total
expenditure expenditure
(billion (billion
$), fuel cost $),$),
(billion fuel costenergy
clean (billioncontribution
$), clean energy
(%), contribution (%),ofthe
the average cost average
electricity
cost of electricity (cents $/kWh), and CO
(cents $/kWh), and CO2 emissions (million Tons). Based on Figure 4, the use of clean energyof
2 emissions (million Tons). Based on Figure 4, the use clean
nearly
energy nearly doubled, which led to a reduction in CO 2 emissions from about
doubled, which led to a reduction in CO2 emissions from about 109 million tons to 38 million tons. 109 million tons to 38
million tons.
Therefore the Therefore
proposed case the proposed case has
has an effective ancompared
role effective rolewithcompared with the base case.
the base case.
Figure 4. Results for study by Abbas et al. (2018). Reproduced from [29], Elsevier: 2018.
Figure 4. Results for study by Abbas et al. (2018). Reproduced from [29], Elsevier: 2018.
Anwar et al. (2017) [30] presented a novel strategy for generation scheduling and power
smoothing for a hybrid system of marine current and wind turbines. In this study, innovative
Energies 2019, 12, 1301 6 of 42
Anwar
Energies et al.
2019, 12, x FOR(2017)
PEER[30] presented a novel strategy for generation scheduling and power smoothing
REVIEW 6 of 43
for a hybrid system of marine current and wind turbines. In this study, innovative strategies were
strategiesto
proposed were proposed
mitigate windtointermittency
mitigate wind intermittency
effects. Contraryeffects. Contrary to the
to the randomness randomness
of wind, of wind,
marine currents
marine
are highly currents are highly
predictable. predictable.
The presented The presented
methodologies methodologies
incorporate incorporate
an optimal strategy anforoptimal
sizingstrategy
for this
for sizing
hybrid for this
system. hybrid system.
Bootstrapped Bootstrapped
artificial artificialwere
neural networks neural networks
developed towere
predictdeveloped
intervalsto forpredict
wind
intervals for wind
speed—speeds speed—speeds
of marine currents modeledof marine currents
utilizing modeledAnalysis
the Harmonic utilizingMethod.
the Harmonic
The resultsAnalysis
of the
Method.
model showThetheresults of the of
robustness model show the methodology,
the presented robustness of which the presented
can be usedmethodology, which
to successfully can be
decrease
used tofluctuations,
power successfullymake decrease power fluctuations,
considerable cost savings, make
and considerable
ensure reliable cost savings,
dispatch and ensure
scheduling forreliable
power
dispatch scheduling
generation. Boukelia et foral.power
(2017)generation.
[31] used anBoukelia
ANN-based et al.approach
(2017) [31]
to used
assessana ANN-based
parabolic trough approach
solar
to assess
power a parabolic
plant. troughan
In this study, solar
ANN power
modelplant.
wasIndeveloped
this study,to anpredict
ANN model was developed
the levelized electricityto predict
cost of
the levelized
two parabolicelectricity
trough solar costthermal
of two parabolic
power plants trough solar thermal
coupled power
with a fuel plantssystem
backup coupled with
and a fuel
thermal
backupstorage.
energy system and The thermal energy storage.
techno-economic study wasThe techno-economic
performed comparing studymolten
was performed comparing
salt and thermic oil
moltentosalt
usage and thermic
optimize thermal oilplants’
usage hourly
to optimize thermalperformance.
and annual plants’ hourly and annual performance.
Chatziagorakis et
Chatziagorakis et al.
al. (2016)
(2016) [32]
[32] studied
studied the the control
control of of hybrid
hybrid renewable
renewable energyenergy systems,
systems, using
using
recurrentneural
recurrent neuralnetworks
networks to to forecast
forecast weather
weather conditions.
conditions. A forecasting
A forecasting modelmodel
using using a recurrent
a recurrent neural
neural network
network for prediction
for prediction of hourly of and
hourly
dailyand daily
solar solar radiation
radiation and wind and windwas
speed speed was presented.
presented. The resultsThe
results
of of the simulation
the simulation indicated that indicated that neural
a recurrent a recurrent
network neural network
was capable was capable
of delivering of delivering
acceptable future
acceptable future
estimation estimation
to evaluate to evaluate
the available the available
renewable energyrenewable
safely. energy safely.
Gallagher et
Gallagher et al.
al. (2018)
(2018)[33][33]studied
studiedthe thesuitability
suitability of of
machine
machine learning
learning forfor
thetheoptimization
optimization of
uncertainty
of uncertainty in energy
in energy savings
savingsmeasurement
measurement and verification. In this In
and verification. paper,
this the newthe
paper, usenew
of machine
use of
learning learning
machine algorithms for energy
algorithms forsavings
energy in industrial
savings buildingsbuildings
in industrial was studied. The applied
was studied. The machine
applied
learning learning
machine techniques consistedconsisted
techniques of k-nearest neighbors,
of k-nearest supportsupport
neighbors, vector machines, artificialartificial
vector machines, neural
networks,
neural decision
networks, trees, and
decision trees,bi-variable and multi-variable
and bi-variable and multi-variable ordinary least least
ordinary squares regression.
squares The
regression.
model’s prediction performances were validated to optimize
The model’s prediction performances were validated to optimize model parameters. The results model parameters. The results
demonstrated that
demonstrated that models
models based based on on ML
ML algorithms
algorithms were were more
more precise
precise than
than conventional
conventional methods.
methods.
Results of
Results RMSE have
of RMSE have beenbeen presented
presented in inFigure
Figure5.5.
RMSE
Figure5.5.RMSE
Figure valuesfor
values forstudy
studybybyGallagher
Gallagheretetal.al.(2018).
(2018).Reproduced
Reproduced from
from [33],Elsevier:
[33], Elsevier:2018.
2018.
3.2. MLP
3.2. MLP
MLP is an advanced version of ANN for engineering applications and energy systems; it is
MLP is an advanced version of ANN for engineering applications and energy systems; it is
considered a feed-forward neural network and uses a supervised and back-propagation learning
considered a feed-forward neural network and uses a supervised and back-propagation learning
method for training purposes [34–36]. This is a simple and popular method for the modeling and
method for training purposes [34–36]. This is a simple and popular method for the modeling and
prediction of a process, and, in many cases, it is considered as the control model. Table 2 demonstrates
prediction of a process, and, in many cases, it is considered as the control model. Table 2 demonstrates
some important papers in this field.
Table 2. Notable MLP models used in energy systems.
Year Reference Journal Application

Energies 2019, 12, 1301 7 of 42
Table 2. Notable MLP models used in energy systems.
Year
Energies 2019, 12, x FOR Reference
PEER REVIEW Journal Application 7 of 43
A day ahead prediction of hourly
2015 Ahmad et al. [37] Solar Energy
A day ahead
global prediction
solar of hourly
irradiation
2015 Ahmad et al. [37] Solar Energy
2017 Chahkoutahi et al. [38] Energy global solar irradiation
Electricity load forecasting
2017 Chahkoutahi et al. [38] Energy Electricity load forecasting
Energy Conversion and Prediction of solar system
2017 Kazem et al. [39] Energy Conversion and Prediction of solar system power
2017 Kazem et al. [39] Management power output
Management output
International Journal of Hourly global solar radiation
2017 Loutfi et al. [40] International Journal of Hourly global solar radiation
2017 Loutfi et al. [40] Renewable Energy Research prediction
Renewable Energy Research prediction
Computational
Computational Intelligence
Intelligence and Ranking
Ranking of different
of different potential
potential power
2017
2017 Shimray et
Shimray et al. [41]al. [41]
and Neuroscience
Neuroscience power plant projects
plant projects
Ahmedetetal.al.(2015)
Ahmed (2015)
[37][37] performed
performed a study
a study on forecasting
on forecasting hourlyhourly solar irradiation
solar irradiation for New for New
Zealand.
Zealand.
In In this
this paper, thepaper,
abilitythe ability to24-h-ahead
to provide provide 24-h-ahead hourly
hourly global global
solar solar irradiation
irradiation forecasts
forecasts was was
assessed
assessedseveral
utilizing utilizing several
methods, methods,
especially especially autoregressive
incorporating incorporating recurrent
autoregressive recurrent Hourly
neural networks. neural
networks.
time series Hourly
were used timeforseries were
training used
and for training
testing and testing
the forecasting methods.the forecasting
MLP, NARX, methods.
ARMA,MLP,and
NARX, ARMA,
persistence and were
methods persistence
comparedmethods
usingwere
RMSE.compared
Figure 6using RMSE.
presents Figure 6results.
the related presents the related
Based on the
results.the
results, Based
NARXon the results,
method thethe
with NARX method
lowest valuewith the lowest
of RMSE value
presented a of RMSE presented
precision about 49%,a 22%,
precision
and
about 49%, 22%, and 52% higher than that of the MLP, ARMA, and
52% higher than that of the MLP, ARMA, and Persistence methods, respectively.Persistence methods, respectively.
Figure 6. RMSE
Figure values
6. RMSE forfor
values study by by
study Ahmed et al.
Ahmed (2015).
et al. Reproduced
(2015). from
Reproduced [37],
from Elsevier:
[37], 2015.
Elsevier: 2015.
Chahkoutahi
Chahkoutahietetal.al. (2017) [38][38]
(2017) introduced a seasonal
introduced optimal
a seasonal hybrid model
optimal hybridtomodel
forecasttothe electricity
forecast the
load. In this study, a direct optimum parallel hybrid model was presented using multi-layer
electricity load. In this study, a direct optimum parallel hybrid model was presented using multi- perceptron
neural network, Seasonal
layer perceptron neuralAutoregressive
network, SeasonalIntegrated Moving Average,
Autoregressive and Adaptive
Integrated MovingNetwork-based
Average, and
Fuzzy Inference System to forecast the electricity load. The main reason
Adaptive Network-based Fuzzy Inference System to forecast the electricity load. The for using thismain
model was for
reason to
utilize these models’ advantages for modeling complex systems. The validation
using this model was to utilize these models’ advantages for modeling complex systems. The of the presented model
implies thatofit the
validation waspresented
more accurate
modelthan its components.
implies Figureaccurate
that it was more 7 presents theitsresults
than of the proposed
components. Figure 7
DOPH method against SARIMA, MLP, ANFIS, DE-based, and GA-based
presents the results of the proposed DOPH method against SARIMA, MLP, ANFIS, DE-based, models. The output of each
and
method
GA-based was compared
models. with target
The output values
of each using
method RMSE.
was Basedwith
compared on the results,
target valuestheusing
proposed
RMSE. method
Based
could
on theimprove
results,the theprediction
proposedcapability by 51.4%,
method could 33.18%,
improve the31.10%, 16.44%,
prediction and 12.8%,
capability by compared with
51.4%, 33.18%,
the SARIMA, MLP, ANFIS, DE-based, and GA-based models, respectively,
31.10%, 16.44%, and 12.8%, compared with the SARIMA, MLP, ANFIS, DE-based, and GA-based in the test stage.
Kazem
models, et al. (2013)
respectively, in [39] designed
the test stage. and installed a photovoltaic system for electricity production.
The output of the system was measured for one year. The photovoltaic system output was simulated
and predicted by self-organizing feature maps, feed-forward networks, support vector machines,
and multi-layer perceptron. Ambient temperature and solar radiation data were these model’s inputs,
Energies 2019, 12, 1301 8 of 42
and the PV array current and current were the outputs. The outputs of each model were compared
with the target values using RMSE factor. The results have been presented in Figure 8. Based on
the results, the SOFM generates the lowest RMSE value compared to the MLP model, GFF model,
Energies
and SVM 2019, 12, x FOR
model. PEER REVIEW
Therefore the SOFM model is suitable for this purpose. 8 of 43
Figure 7. RMSE values for study by Chahkoutahi and Khashei (2017). Reproduced from [38], Elsevier
2017.
Kazem et al. (2013) [39] designed and installed a photovoltaic system for electricity production.
and predicted by self-organizing feature maps, feed-forward networks, support vector machines, and
multi-layer perceptron. Ambient temperature and solar radiation data were these model’s inputs,
with the target values using RMSE factor. The results have been presented in Figure 8. Based on the
results, the SOFM generates the lowest RMSE value compared to the MLP model, GFF model, and
7. RMSE
SVMFigure
Figure 7. values
thefor study by Chahkoutahi and Khashei (2017).(2017).
Reproduced from [38], Elsevier
RMSE values for study by Chahkoutahi and Khashei Reproduced from [38],
model. Therefore SOFM model is suitable for this purpose.
2017.
Elsevier 2017.
Kazem et al. (2013) [39] designed and installed a photovoltaic system for electricity production.
and predicted by self-organizing feature maps, feed-forward networks, support vector machines, and
multi-layer perceptron. Ambient temperature and solar radiation data were these model’s inputs,
with the target values using RMSE factor. The results have been presented in Figure 8. Based on the
results, the SOFM generates the lowest RMSE value compared to the MLP model, GFF model, and
SVM model. Therefore the SOFM model is suitable for this purpose.
Figure 8. RMSE
Figure8. RMSEvalues
valuesfor
forstudy
studyby
byKazem
Kazemand
and Yousif
Yousif (2013).
(2013). Reproduced
Reproduced from
from [39],
[39], Elsevier:
Elsevier: 2013.
2013.
Loutfi
Loutfi etet al. (2017) [40]
al. (2017) [40] presented
presented an an analysis
analysis of
of the
the design
design ofof solar
solar energy
energy systems.
systems. InInthis
thisstudy,
study,
a
a comparison between multilayer perceptron and neural autoregressive with exogenous inputs was
comparison between multilayer perceptron and neural autoregressive with exogenous inputs was
presented.
presented. The proposed model has excellent ability to produce hourly solar radiation forecasts for
The proposed model has excellent ability to produce hourly solar radiation forecasts for
cheaper data such
cheaper data such asas relative
relative humidity
humidity and and temperature.
temperature. The The results
results of
of the
the best
best model
model areare presented
presented
in
in Table
Table 33 for
for the
the developed
developed model.
model. The
The study
study proposes
proposes the the NARX
NARX method
method in in conjunction
conjunction withwith the
the
MLP method. As is clear from Table 3, the proposed method has the best prediction
MLP method. As is clear from Table 3, the proposed method has the best prediction capability with capability with
reference
reference to
to nRMSE
nRMSE and and correlation
correlation coefficient
coefficient values.
values.
Shimray et al. (2017) [41] performed a study on the installation of hydropower plant sites ranking
usingFigure
Table RMSE values
3.8.Results
a Multi-layer of for study
Perceptron
evaluations of by
Neural Kazem
models byand
Network. Yousifal.(2013).
Inetthis
Loutfi Reproduced
paper,
(2017). a model was
Reproduced from [39],
[40],Elsevier:
fromdeveloped for2013.
Elsevier: decision
2017.
makers to rank potential power plant sites based on water quality, air quality, energy delivery cost,
Loutfi
natural et al.ecological
hazard, (2017) [40]impact,
presentedandanproject
analysis of the design
duration. of solar
The case energy
in this paper systems. In thisseveral
was ranking study,
a comparison between multilayer
potential plant sites in India. perceptron and neural autoregressive with exogenous inputs was
presented. The proposed model has excellent ability to produce hourly solar radiation forecasts for
cheaper data such as relative humidity and temperature. The results of the best model are presented
in Table 3 for the developed model. The study proposes the NARX method in conjunction with the
MLP method. As is clear from Table 3, the proposed method has the best prediction capability with
reference to nRMSE and correlation coefficient values.
ML Model The Best Structure r f

MLP 5-30-1 0.938 23.31
Energies 2019, 12, 1301 9 of 42
NARX 1 5-10-1 0.974 15.1
Shimray
Table et al. (2017)
3. Results [41] performed
of evaluations of modelsa by
study onetthe
Loutfi al.installation of hydropower
(2017). Reproduced from [40],plant sites
Elsevier: ranking
2017.
using a Multi-layer Perceptron Neural Network. In this paper, a model was developed for decision
ML Model The Best Structure r f
makers to rank potential power plant sites based on water quality, air quality, energy delivery cost,
natural hazard, ecological impact,MLP and project5-30-1duration. The 0.938
case in this paper was ranking several
23.31
NARX 1 5-10-1 0.974 15.1
potential plant sites in India.
3.3. ELM
3.3. ELM and
and Other Advanced ANNs
Other Advanced ANNs
In order
In order toto find
find an
an advanced
advanced version
version of
of ANNs,
ANNs, the
the keywords
keywords of
of the
the search
search here
here were
were extreme
extreme
learning machine, Feed-forward neural networks, Back-propagation neural networks, functional
learning machine, Feed-forward neural networks, Back-propagation neural networks, functional
neural network,
neural network, Feedforward,
Feedforward, and
and back
back propagation.
propagation. The ELM has
The ELM has aa high
high speed
speed ofof learning
learning and
and the
the
proper ability of generalization. Table 4 lists some important papers in this field.
proper ability of generalization. Table 4 lists some important papers in this field.
4. Notable
Table 4.
Table Notable ELM
ELM models
models used
used in
in energy
energy systems.
systems.
Year
Year Reference
Reference Journal
Journal Application
Application
AppliedThermal
Applied Thermal Optimization
Optimization of theofdistrict
the district heating
heating system
2017
2017 Arat
Aratand
andArslan
Arslan[42]
[42]
Engineering
Engineering system
aidedaided
with with geothermal
geothermal heat pump
heat pump
Electricalconsumption
consumptionforecasting
forecastingmodel
model
2015
2015 Bagnasco
Bagnascoetetal.
al.[43]
[43] Energyand
Energy andBuildings
Buildings Electrical
of a building
of a building
2018 Li, Q et al. [44] Applied Energy Forecasting of PV power generation
2018 Li, Q et al. [44] Applied Energy Forecasting of PV power generation
Premalatha and Valan Journal of Applied Monthly average global radiation
2016 Premalatha and Valan Journaland
of Applied Monthly average global radiation
2016 Arasu [45] Research Technology prediction
Arasu [45] Research and Technology prediction
Applied Thermal Performance forecasting of a solar
2014 Yaïci and Entchev [46]
2014 Yaïci and Entchev [46] Engineering
Applied Thermal Performance thermal energy
forecasting ofsystem
a solar thermal
Engineering energy system
Arat and Arslan (2017) [42] presented an optimum design for a geothermal heat pump for a
district
Aratheating system.
and Arslan Three
(2017) different an
[42] presented back propagation
optimum learning
design for algorithms
a geothermal was used.
heat pump These
for a district
algorithms
heating wereThree
system. Pola–Ribiere
differentConjugate Gradient,
back propagation Levenberg‒Marquardt,
learning algorithms was used.and Scaled Conjugate
These algorithms
Gradient.
were The presented
Pola–Ribiere ConjugateELMGradient,
model was composed of two stages.
Levenberg-Marquardt, The second
and Scaled stage
Conjugate of the ELM
Gradient. The
structure was
presented ELM composed
model was of three levels of
composed of two
ANN models
stages. in second
The this proposed
stage ofmodel.
the ELMThe structure
results of was
this
study showed
composed that levels
of three the values
of ANNobtained
models from ANN
in this were very
proposed closeThe
model. to results
the analytical data. Figure
of this study showed9
demonstrates
that the valuesthe correlation
obtained coefficient
from ANN values,
were very and
close Table
to the 5 presents
analytical data.the RMSE
Figure values for each
9 demonstrates the
parameter separately.
correlation coefficient values, and Table 5 presents the RMSE values for each parameter separately.
9. rr values
Figure 9.
Figure values of
of study
study of
of Arat
Arat and
and Arslan
Arslan (2017). Reproduced from
(2017). Reproduced from [42],
[42], Elsevier:
Elsevier: 2017.
2017.
Table 5. RMSE values of the study of Arat and Arslan (2017). Reproduced from [42], Elsevier: 2017.
Energies 2019, 12, 1301 10 of 42
EnergiesTable 12,RMSE
2019, 5. x FOR values of the study of Arat and Arslan (2017). Reproduced from [42], Elsevier: 2017.
PEER REVIEW 10 of 43
Parameter Model RMSE Parameter Model RMSE Parameter Model RMSE

Parameter Model RMSE Parameter Model RMSE Parameter Model RMSE
LM-28 0.07506 LM-28 229,835 LM-50 58.3957
LM-28 0.07506 LM-28 229,835 LM-50 58.3957
COP SCG-28 0.79114 Cinst SCG-28 335,283 WP2 SCG-50 606.962
COP SCG-28 0.79114 Cinst SCG-28 335,283 WP2 SCG-50 606.962
CGP-28 0.77328 CGP-28 345,108.6 CGP-50 862.943
CGP-28 0.77328 CGP-28 345,108.6 CGP-50 862.943
LM-28
LM-28 0.00335
0.00335 LM-28
LM-28 32,357.74
32,357.74 LM-50
LM-50 1.14269
1.14269
εsys SCG-28 0.01616 Cop SCG-28 183,385.5 WP3 SCG-50 13.3365
εsys SCG-28 0.01616 Cop SCG-28 183,385.5 WP3 SCG-50 13.3365
CGP-28 0.01963 CGP-28 384,134.8 CGP-50 22.8781
CGP-28 0.01963 CGP-28 384,134.8 CGP-50 22.8781
LM-50
LM-50 98.07454
98.07454 LM-20
LM-20 45,456.24
45,456.24 LM-50
LM-50 25.2462
25.2462
WcWc SCG-50
SCG-50 1044.513
1044.513 NPV
NPV SCG-20
SCG-20 3,104,498
3,104,498 Ηr
Hr SCG-50
SCG-50 294.616
294.616
CGP-50 1939.327 CGP-20 20,565,006 CGP-50 498.103
CGP-50 1939.327 CGP-20 20,565,006 CGP-50 498.103
LM-50
LM-50 87.0961
87.0961 LM-28
LM-28 0.03824
0.03824 LM-50
LM-50 2.208349
2.208349
Qcon SCG-50
Qcon SCG-50 1027.196
1027.196 COPsys
COPsys SCG-28
SCG-28 0.31434
0.31434 Qevp
Qevp SCG-50
SCG-50 465.1691
465.1691
CGP-50
CGP-50 1754.37
1754.37 CGP-28
CGP-28 0.29927
0.29927 CGP-50
CGP-50 575.9355
575.9355
Bagnasco et
Bagnasco et al.
al. (2015)
(2015) presented
presented aa study
study ofof power
power consumption
consumption forecasting
forecasting (load
(load forecasting
forecasting
model) in
model) in hospitals.
hospitals. TheThe presented
presented artificial neural network
artificial neural network utilizing
utilizing aa backpropagation
backpropagation training
training
algorithm can
algorithm cantake
takeloads,
loads,time
timeofofthe
the day,
day, data
data concerning
concerning the the
typetype of day
of day (e.g.,(e.g., weekday/holiday)
weekday/holiday) and
and weather data. The proposed forecast algorithm can be easily integrated into
weather data. The proposed forecast algorithm can be easily integrated into the Building Management the Building
Management
Systems Systems
real-time real-time
monitoring monitoring
system. system. no
Case A includes Case A includes
weather no weather
data, and data, and
Case B includes Case B
day-ahead
includes day-ahead temperature data. RMSE values have
temperature data. RMSE values have been presented in Figure 10. been presented in Figure 10.
Figure
Figure 10. RMSE values
10. RMSE values for
for study
study by
by Bagnasco
Bagnasco et
et al. (2015). Reproduced
al. (2015). Reproduced from
from [43],
[43], Elsevier:
Elsevier: 2015.
2015.
Li et al. (2018) [44] presented a multi-clustered echo state network model to directly forecast
Li et al. (2018) [44] presented a multi-clustered echo state network model to directly forecast PV
PV electricity generation. Measured and estimated PV electricity generation data characteristic such
electricity generation. Measured and estimated PV electricity generation data characteristic such as
as stationarity (or non-stationarity), seasonality and complexity analysis were investigated through
stationarity (or non-stationarity), seasonality and complexity analysis were investigated through data
data mining approaches. The simulation results showed that the presented multi-clustered echo state
mining approaches. The simulation results showed that the presented multi-clustered echo state
network model can accurately forecast photovoltaic power output one hour ahead. The one-day
network model can accurately forecast photovoltaic power output one hour ahead. The one-day
forecast has 91–98% correlation coefficient for cloudy days and 99% for sunny days. Figure 11 presents
forecast has 91–98% correlation coefficient for cloudy days and 99% for sunny days. Figure 11
the correlation coefficient values for the study based on day and method.
presents the correlation coefficient values for the study based on day and method.
Energies 2019, 12, 1301 11 of 42
Figure
Figure 11. Correlation coefficient
11. Correlation coefficient values
values for
for the
the study
study by
by Li
Li et
et al.
[44],
Elsevier: 2018.
Elsevier: 2018.
Premalatha and Valan Arasu (2016) [45] utilized ANN models to predict global solar radiation.
Premalatha and Valan Arasu (2016) [45] utilized ANN models to predict global solar radiation.
The main purpose of this study was to develop an ANN model for accurate prediction of solar radiation.
The main purpose of this study was to develop an ANN model for accurate prediction of solar
Two different ANN models based on four algorithms were considered in this paper. The last 10 years
radiation. Two different ANN models based on four algorithms were considered in this paper. The
of meteorological data were collected from five different sites in India to train the models. The criteria
last 10 years of meteorological data were collected from five different sites in India to train the models.
for determining best ANN algorithm were a minimum mean absolute error, root mean square error,
The criteria for determining best ANN algorithm were a minimum mean absolute error, root mean
and maximum linear correlation coefficient of 3.028, 3.646, and 0.927, respectively.
square error, and maximum linear correlation coefficient of 3.028, 3.646, and 0.927, respectively.
Yaïci and Entchev (2014) [46] utilized artificial neural networks to predict the performance of a
Yaïci and Entchev (2014) [46] utilized artificial neural networks to predict the performance of a
solar thermal energy system. This system was used for space heating and domestic hot water—two
solar thermal energy system. This system was used for space heating and domestic hot water—two
different variants of the back-propagation learning algorithm (scaled conjugate gradient algorithms
different variants of the back-propagation learning algorithm (scaled conjugate gradient algorithms
and the Levenberg-Marquardt). The presented model was applied for predicting several performance
and the Levenberg‒Marquardt). The presented model was applied for predicting several
parameters of the system such as the temperature stratification of the preheat tank, the derived solar
performance parameters of the system such as the temperature stratification of the preheat tank, the
fractions, the solar
derived solar collectors’
fractions, heatcollectors’
the solar input to the heat
heat exchanger,
input andexchanger,
to the heat auxiliary propane-fired
and auxiliary tank heat
propane-
input. This methodology can be used for condition monitoring and fault detection of a solar
fired tank heat input. This methodology can be used for condition monitoring and fault detection of thermal
energy system. energy system.
a solar thermal
3.4. SVM
3.4. SVM
SVMs are machine learning algorithms built on statistical learning theory for structural risk
SVMs are machine learning algorithms built on statistical learning theory for structural risk
minimization. In pattern recognition, classification, and analysis of regression, SVMs outperform other
minimization. In pattern recognition, classification, and analysis of regression, SVMs outperform
methodologies. The significant range of SVM applications in the field of load forecasting is due to its
other methodologies. The significant range of SVM applications in the field of load forecasting is due
ability to make generalizations. Also, local minima lead to no problems in SVM. Table 6 presents some
to its ability to make generalizations. Also, local minima lead to no problems in SVM. Table 6 presents
important papers in this field.
Arabloo et al. (2015) [47] introduced a novel methodology for the optimization of the
oxygen-steam ratio in the gasification process of coal. A methodology utilizing support vector machine
Table 6. Notable SVM models used in energy systems.
algorithm was presented for estimation of proper steam-oxygen ratio to balance the heat requirement
andYear Authors
released heat Journal A comparison of experimental
in the coal gasification process. Application
data and predicted
values Estimation of optimum oxygen-
2015showed theetprecision
Arabloo al. [47] of the predictive Fuelmodel that can be used for commercial implications
steam ratios
in the coal gasification process. Arikan et al. (2013) [48] performed a study to classify power quality
International Review of Electrical Classification of power quality
2013
disturbances Arikan et al.support
utilizing [48] vector machines. In this paper, five kinds of power quality disturbances
Engineering disturbances
and pure sine was utilizing support vector machines that were based on wavelets. The proposed
Estimation of irradiance levels from
method IEEE Transactions on Circuitsdata
and derived from the mathematical model
2017 performance was validated
Ma et al. [49] utilizing synthetic photovoltaic electrical
Systems I: Regular Papers
and real-time measurements. Support vector machines, artificial neural network, and the same future
characteristics
vector and data Bayes classifier were compared. It was observed that Harmonic estimationmachines
support vector of power gave
2016 Özdemir et al. [50] Neural Network World
the best results both for synthetic data and in real time. quality in electrical energy systems
An electricity market price
2016 Pinto et al. [51] Neurocomputing
prediction in a fast execution time
Energies 2019, 12, 1301 12 of 42
Energies 2019, 12, x FOR PEER REVIEW

Table 6. Notable SVM models used in energy systems. 12 of 43
Arabloo et al.
Year (2015) [47] introduced Journal
Authors a novel methodology for the optimization
Application of the oxygen‒
steam ratio in the gasification process of coal. A methodology utilizing
Estimation support oxygen-steam
of optimum vector machine
2015 Arabloo et al. [47] Fuel
algorithm was presented for estimation of proper steam‒oxygen ratio to balance the heat requirement
ratios
and released heat in the coal gasification process.
International A comparison
Review of of experimental
Classification data and
of power predicted
quality
2013 Arikan et al. [48]
values showed the precision of the predictive model that can be used for commercial implications in
Electrical Engineering disturbances
the coal gasification process. Arikan et al. (2013)on[48]
IEEE Transactions performed
Circuits a studyofto
Estimation classify levels
irradiance power quality
from
2017 Ma et al. [49]
disturbances utilizing support vector machines. In this paper, five kinds of power quality
and Systems I: Regular Papers photovoltaic electrical characteristics
disturbances and pure sine was utilizing support vector machines that were
Harmonic basedofon
estimation wavelets.
power qualityThe
2016 Özdemir et al. [50] Neural Network World
proposed method performance was validated utilizing synthetic data derived from
in electrical energythe mathematical
systems
model and real-time measurements. Support vector machines,An artificial neural
electricity network,
market and the same
price prediction in
2016 Pinto et al. [51] Neurocomputing
future vector and data Bayes classifier were compared. It was observeda that support vector
fast execution time machines
gave the best results both for synthetic data and in real time.
Ma et
Ma et al.
al. (2017)
(2017) [49]
[49] developed
developed aa soft
soft sensor
sensor (a(a field-support
field-support vector
vector regression)
regression) to to upgrade
upgrade the the
estimation accuracy of solar irradiance levels from photovoltaic electrical parameters.
estimation accuracy of solar irradiance levels from photovoltaic electrical parameters. The soft sensor The soft sensor
collected its
collected its input
input data
data into
into several
several groups
groups based
based onon ambient
ambient temperature.
temperature. The The introduced
introduced soft
soft sensor
sensor
can be
can be implanted
implanted in in aa photovoltaic
photovoltaic module,
module, aa current
current sensor,
sensor, or
or aa thermometer.
thermometer. It It was
was validated
validated by by
experimental prototype and simulations utilizing measured outdoor conditions.
experimental prototype and simulations utilizing measured outdoor conditions. Figure 12 presents the Figure 12 presents
the RMSE
RMSE values values for study.
for the the study.
Figure 12.
Figure RMSE values
12. RMSE values of
of the
the study
study of
of Ma
Ma et
et al.
al. (2018).
(2018). Reproduced
Reproduced from
from [49],
[49], Elsevier:
Elsevier: 2018.
2018.
Özdemir
Özdemir et et al.
al. (2016)
(2016)[50]
[50]used
used Support
Support Vector
Vector Machine
Machine for for harmonic
harmonic distortion
distortion estimation.
estimation. The
The
power distribution network was studied, and the predicted results were compared with the
power distribution network was studied, and the predicted results were compared with the
quantified
quantified real
real data.
data. The
The presented
presented approach
approach was
was compared
compared with with the
the artificial
artificial neural
neural network
network and
and
linear
linear regression
regression methods.
methods. The
The validation
validation of
of the
the predicted
predicted results
results demonstrated
demonstrated that that Support
Support Vector
Vector
Machine
Machine is robust for total harmonic distortion in the power network. Pinto et [51]
is robust for total harmonic distortion in the power network. Pinto et al. (2016) developed
al. (2016) [51]
adeveloped
multi-agent system for system
a multi-agent modeling forcompetitive electricity markets.
modeling competitive electricityThis studyThis
markets. proposed applying
study proposed
support
applyingvector machines
support vector to lay out to
machines decision
lay outsupport
decision forsupport
electricity
for market players.
electricity marketThe presented
players. The
model was coupled with an Adaptive Learning Strategic Bidding System
presented model was coupled with an Adaptive Learning Strategic Bidding System to be used to be used as a decision
as a
support
decisionsystem.
support This methodology
system. was validated
This methodology and
was then compared
validated and thenwithcompared
the artificial neural
with the network.
artificial
The results were encouraging: a robust price forecast for electricity market was
neural network. The results were encouraging: a robust price forecast for electricity market was obtained quickly.
obtained quickly.
3.5. WNN
3.5. WNN
WNN takes the benefits of both the theory of wavelets and neural networks and combines them.
This method contains an FFNN with one hidden layer. One of the missions of WNNs is to estimate the
WNN takes the benefits of both the theory of wavelets and neural networks and combines them.
This method contains an FFNN with one hidden layer. One of the missions of WNNs is to estimate
the function of a process or a trend or computing. A WNN can train the structure of a function using
a series of data and generate or compute an expected output value for a given input value [52]. WNN
Energies 2019, 12, 1301 13 of 42
function of a process or a trend or computing. A WNN can train the structure of a function using a
series of data and generate or compute an expected output value for a given input value [52]. WNN
has several advantages over other neural networks. WNN needs a smaller training amount than
the
13 of 43
MLP method, and has fast convergence. Table 7 presents some important papers in this field.
has several advantages over other neural networks. WNN needs a smaller training amount than the
MLP method, and has fastTable 7. Notable WNN
convergence. Table models usedsome
7 presents in energy systems.papers in this field.
important
Year Authors Journal Application
Table 7. Notable WNN models used in energy systems.
Prediction of time series for renewable energy
2016 Doucoure et al. [53] Renewable Energy
sources
2016
2018 Doucoure
Gu etet al.
al. [54]
[53] RenewableEnergy
Energy Prediction
Heatofload
time series forinrenewable
perdition energy
district heating sources
systems
2018 Gu et al. [54] Energy Heat load perdition in district heating systems
Wind speed forecasting (reduction of the influence
2018 He et al. [55] Applied Energy Wind speed forecasting (reduction of the influence of
2018 He et al. [55] Applied Energy of noise in the raw data series)
noise in the raw data series)
Simultaneous optimization of fuel economy and
2018 Qin et al. [56] Algorithms Simultaneous optimization of fuel economy and battery
2018 Qin et al. [56] Algorithms battery state of charge
state of charge
2017
2017 Sarshar et al.
Sarshar et al. [57] [57] Energy
Energy Uncertainty reduction in in
Uncertainty reduction wind
wind power
power prediction
prediction
Doucoure et
Doucoure et al.
al. (2016)
(2016) [53]
[53] developed
developed aa prediction
prediction methodology
methodology forfor renewable
renewable energy
energy sources
sources
to promote the use of renewable energy isolated and grid-connected power systems.
to promote the use of renewable energy isolated and grid-connected power systems. The presented The presented
method was
method wasbased
basedononartificial
artificial neural
neural networks
networks andand wavelet
wavelet decomposition.
decomposition. The predictability
The predictability of
of every
every component
component of thedata
of the input input data utilizing
utilizing the Hurstthe Hurst coefficient
coefficient was analyzed
was analyzed in this
in this study. study. the
To ensure To
ensure the predictability
predictability of the information,
of the information, some components
some components with low predictability
with low predictability potential
potential were were
eliminated
eliminated
to to computational
reduce the reduce the computational algorithm Figure
algorithm complexity. complexity. Figurethe
13 presents 13 RMSE
presents the RMSE
values for thevalues
study
in three terms, with all data components, without random data and persistence data related todata
for the study in three terms, with all data components, without random data and persistence the
related to method.
proposed the proposed method.
Figure 13. RMSE

values for
for study
study by
by Doucoure
Doucoure et
et al.
[53], Elsevier:
Elsevier: 2016.
2016.
Gu et al. (2018) [54] studied heat load prediction utilizing different prediction models such as
Gu et al. (2018) [54] studied heat load prediction utilizing different prediction models such as
extreme learning machine, wavelet neural network, support vector machine, back propagation neural
extreme learning machine, wavelet neural network, support vector machine, back propagation neural
network optimized using a genetic algorithm, and wavelet neural network. Historical loads and indoor
network optimized using a genetic algorithm, and wavelet neural network. Historical loads and
temperature were assumed to be influential. The support vector machine demonstrated smaller errors
indoor temperature were assumed to be influential. The support vector machine demonstrated
compared to the three other neural network algorithms. Figure 14 presents the values of RMSE and
smaller errors compared to the three other neural network algorithms. Figure 14 presents the values
correlation coefficient for the study.
of RMSE and correlation coefficient for the study.
Energies 2019,
Energies 12, 1301
2019, 12, x FOR PEER REVIEW 14
14 of 43
of 42
(a)
(b)
Figure 14. RMSE
Figure 14. and rr values
RMSE and values for
for study
study by
by Gu
Gu et
et al.
al. (2018).
(2018). Reproduced
Reproduced from
from [54],
[54], Elsevier:
Elsevier: 2018.
2018.
He
He et
et al.
al. (2018)
(2018) [55]
[55] proposed
proposed aa hybrid
hybrid forecasting
forecasting model
model which
which was
was composed
composed of of three
three modules:
modules:
data clustering, data
data clustering, data preprocessing
preprocessingand andforecasting
forecastingmodules.
modules.TheThedecomposing
decomposing technique
technique was
was used
used to
to decrease the noise influence within the raw data series to achieve a more stable sequence to extract
decrease the noise influence within the raw data series to achieve a more stable sequence to extract
traits from the
traits from theoriginal
originaldata.
data.AAsimilar
similarfluctuation
fluctuation pattern
pattern waswas selected
selected forfor
thethe training
training database
database in
in the
the forecasting module to improve the forecasting accuracy. The experimental data demonstrate that
forecasting module to improve the forecasting accuracy. The experimental data demonstrate that the
the presented
presented model model outperforms
outperforms other
other discussed
discussed forecasting
forecasting models
models in the
in the paper.
paper. Figure
Figure 15 presents
15 presents the
the RMSE
RMSE values
values in cases
in cases of non-season
of non-season and and season
season datasets.
datasets.
Qin et al. (2018) [56] introduced online energy management control of hybrid electric vehicles
based on NDP to optimize battery state of charge and fuel economy at the same time. In the proposed
NDP method, the action network was a conventional wavelet neural network, and the critic network
was a multi-resolution wavelet neural network. The action network was constructed on the Morlet
function, and the critic network was stem from the Meyer wavelet function. Based on results, the
NDP EMS have a high capability the same as the NDP in online applications. Based on comparisons,
the RBFNN-based NDP EMS supports the efficiency of the CWNN and MRWNN.
Energies 2019, 12, 1301 15 of 42
Figure 15. RMSE

Figure15. RMSE values
values for
for study by He
study by He et
et al.
al. (2018).
(2018).Reproduced
Reproducedfrom
from[55],
[55],Elsevier:
Elsevier:2018.
2018.
Sarshar et al.
Qin et al. (2017)
(2018) [56][57] presented
introduced an adaptive
online probabilisticcontrol
energy management conceptofofhybrid
the confidence interval
electric vehicles
for addressing randomness of wind speed. To increase forecasting accuracy, wavelet
based on NDP to optimize battery state of charge and fuel economy at the same time. In the proposed decomposition
was
NDP utilized
method,forthe
time series
action of wind
network power,
was and the results
a conventional waveletwere used
neural in the artificial
network, and the neural network.
critic network
Then
was apredicted wind power
multi-resolution dependable
wavelet levels adaptive
neural network. The actionprobabilistic
network was concept of confidence
constructed interval
on the Morlet
function, and the critic network was stem from the Meyer wavelet function. Based on results, theon
were calculated. An energy storage system was used to decrease the impact of forecasting errors
NDP
the EMS have
micro-grid anda high capability
to increase the same
planning as the NDP
flexibility. in online
Finally, applications.
the presented Basedwas
algorithm on comparisons,
validated with
athe RBFNN-based
typical micro-gridNDP case EMS
study.supports the efficiency
The results of the
indicated that CWNN
the and adaptive
presented MRWNN.probabilistic concept
Sarshar et
of confidence al. (2017)
interval [57] presented
worked an adaptive probabilistic
well and demonstrated concept
the superiority of thetoconfidence
of WNN ANN. interval
for addressing randomness of wind speed. To increase forecasting accuracy, wavelet decomposition
3.6.
wasANFIS
utilized for time series of wind power, and the results were used in the artificial neural network.
Then predicted
ANFIS is awind power dependable
modeling method that levels adaptive
employs anprobabilistic conceptnetwork
artificial neural of confidence
basedinterval
on the
were calculated.fuzzy
Takagi–Sugeno An energy storage
inference system
system. Thiswas used to decrease
technique benefits the
fromimpact of forecasting
the capabilities errorsfuzzy
of both on
the micro-grid and to increase planning flexibility. Finally, the presented algorithm
logic and neural network. ANFIS is a method with five main layers. This method is considered as an was validated
withform
early a typical micro-grid
of the hybrid ML case study.[28].
method The results
Table 8indicated that the
presents some presented
important adaptive
papers probabilistic
in this field.
concept of confidence interval worked well and demonstrated the superiority of WNN to ANN.
Table 8. Notable ANFIS models used in energy systems.
3.6. ANFIS
ANFIS is a modeling method that employs an artificial neural network based on the Takagi–
A protection device for a reverse
Sugeno
2018 fuzzy inference system.
Abdulwahid et al. [58]This technique benefits
Sustainability from the capabilities of both fuzzy logic and
(Switzerland)
power protection system
neural network. ANFIS is a method with five main layers. This method is considered as an early form
Module temperature estimation of
of2017
the hybrid ML method
Bassam et al.[28].
[59] Table 8Sustainability
presents some important papers in this field.
(Switzerland)
PV systems
IEEE Transactions on Prediction of the power demand of a
2018 Kampouropoulos et al. [60]
Smart Grid plant and optimization of energy flow
Identification of the most relevant
Renewable and Sustainable
2016 Mohammadi et al. [61] parameters for forecasting of daily
Energy Reviews
global solar radiation
Journal of the Taiwan
Transesterification yield estimation
2016 Sajjadi et al. [62] Institute of Chemical
and prediction of biodiesel synthesis
Engineers
Abdulwahid and Wang (2018) [58] introduced a novel protection method for preventing reverse
power flow developed on neuro-fuzzy networks for utilization in the smart grid. This study presented
an upgraded protection device using a newly developed intelligent decision support system (IDSS).
The presented IDSS was a decision system support system that coupled the robust specification for
fuzzy inference systems and neural networks. The proposed methodology can monitor extreme
environmental conditions.
Abdulwahid and Wang (2018) [58] introduced a novel protection method for preventing reverse
power flow developed on neuro-fuzzy networks for utilization in the smart grid. This study
presented an upgraded protection device using a newly developed intelligent decision support
system (IDSS). The presented IDSS was a decision system support system that coupled the robust
specification for fuzzy inference systems and neural networks. The proposed methodology can
Energies 2019, 12, 1301 16 of 42
monitor extreme environmental conditions.
Bassam et al. (2017) [59] developed an adaptive neuro-fuzzy inference to estimate the
temperature
Bassam et ofal.photovoltaic systems. Experimental
(2017) [59] developed measurements
an adaptive neuro-fuzzy for the
inference learning
to estimate theprocess were
temperature
comprised
of photovoltaic of six environmental
systems. Experimentalvariables, namely wind
measurements for thevelocity,
learning temperature, wind direction,
process were comprised of six
irradiance,
environmental atmospheric
variables, pressure,
namely wind and velocity,
relative temperature,
humidity and PV direction,
wind power output as an atmospheric
irradiance, operational
variable,
pressure, which were used
and relative for training
humidity and PVparameters.
power output Theasmodel was validated
an operational with
variable, experimental
which were useddata
for
from a photovoltaic
training parameters. system
The modelwithwas
a high value for
validated fitness
with correlation
experimental dataparameter. The resultssystem
from a photovoltaic of the
model
with showvalue
a high that the
forpresented methodology
fitness correlation was a reliable
parameter. toolof
The results forthe
modules’
model show temperature
that theestimation
presented
using environmental
methodology variables.
was a reliable toolFigure 16 presents
for modules’ the RMSE
temperature (a) and correlation
estimation coefficient (b)
using environmental values
variables.
for the 16
Figure study. Thisthe
presents comparison
RMSE (a) isandincorrelation
line with choosing
coefficientthe
(b)best
valuestypeforofthe
membership
study. This function
comparison for
ANFIS.
is in lineAswithis choosing
clear from theFigure 16, of
best type a Gbell membership
membership functionfunction with As
for ANFIS. a low RMSE
is clear fromand a high
Figure 16,
a Gbell membership value has
function withthe highest
a low RMSE accuracy compared
and a high with coefficient
correlation the other types.
value has the highest
accuracy compared with the other types.
(a)
(b)
Figure 16.
Figure 16.(a) RMSE
(a)RMSE andand (b) correlation
(b) correlation coefficient
coefficient values
values for studyfor
by study
Bassamby Bassam
et al. (2017).etReproduced
al. (2017).
Reproduced
from from [59],
[59], Elsevier: 2017.Elsevier: 2017.
Kampouropoulos et
Kampouropoulos et al.
al. (2018)
(2018) [60]
[60] introduced
introduced aa novel
novel approach
approach for
formulti-carrier
multi-carrier energy
energy systems’
systems’
energy optimization. In this study, an adaptive neuro-fuzzy inference system was applied to forecast
the power
the powerdemand
demandofof a factory,
a factory, andand a genetic
a genetic algorithm
algorithm wastoused
was used to its
model model itsflow.
energy energy
Theflow. The
objective
objective of the optimization algorithm was to fulfill the power demand of the factory to reduce
of the optimization algorithm was to fulfill the power demand of the factory to reduce optimization
optimization
criteria. criteria. The
The proposed proposed
method method was
was validated validated at SEAT.
at SEAT.
Mohammadi et al. (2016) [61] presented a method to identify the essential parameters for
forecasting global solar radiation utilizing an ANFIS selection procedure. In this study, a
methodology based on ANFIS was applied to identify the most related parameters for daily
prediction of global solar radiation. Three different cities were considered as case studies. Nine
parameters of extraterrestrial radiation, sea level pressure, relative humidity, water vapor pressure,
minimum, average and maximum air temperatures, maximum possible sunshine duration, and
Reproduced from [59], Elsevier: 2017.
Kampouropoulos et al. (2018) [60] introduced a novel approach for multi-carrier energy systems’
the power demand of a factory, and a genetic algorithm was used to model its energy flow. The
Energies 2019, 12, 1301 17 of 42
objective of the optimization algorithm was to fulfill the power demand of the factory to reduce
optimization criteria. The proposed method was validated at SEAT.
Mohammadietetal.
Mohammadi al. (2016)
(2016) [61]
[61] presented
presented aa method
method to to identify
identify the
the essential
essential parameters
parameters for for
forecastingglobal
forecasting global solar
solar radiation
radiation utilizing
utilizing an selection
an ANFIS ANFIS selection
procedure. procedure. In athis
In this study, study, a
methodology
methodology
based on ANFIS based on ANFIS
was applied was the
to identify applied
most to identify
related the most
parameters related
for daily parameters
prediction for daily
of global solar
prediction
radiation. of global
Three solar
different radiation.
cities Three different
were considered as casecities wereNine
studies. considered as case
parameters studies. Nine
of extraterrestrial
parameters
radiation, seaoflevel
extraterrestrial
pressure, radiation, sea level pressure,
relative humidity, water vaporrelative humidity,
pressure, water vapor
minimum, pressure,
average and
minimum, average and maximum air temperatures, maximum possible
maximum air temperatures, maximum possible sunshine duration, and sunshine duration were sunshine duration, and
sunshine duration
considered wereinconsidered
for selection ANFIS process.for selection in ANFIS
The results indicated process. Theoptimal
that the results sets
indicated thatwere
of inputs the
optimal sets of inputs were different for different case studies. This study
different for different case studies. This study demonstrated the significance of the selection of inputdemonstrated the
significance
parameters forofpredicting
the selection
dailyof global
input parameters for predicting
solar radiation. Figure 17daily global
indicates thesolar radiation.
value of RMSEFigure 17
in terms
ofindicates
two andthe threevalue of RMSE
inputs. Basedinon terms of two and
the results, threethree
inputsinputs.
provideBased on the
better results, performance
prediction three inputs
provide better prediction performance
compared with that for the two inputs. compared with that for the two inputs.
Figure 17. RMSE values for study by Mohammadi et al. (2016). Reproduced from [61], Elsevier: 2016.
Figure 17. RMSE values for study by Mohammadi et al. (2016). Reproduced from [61], Elsevier: 2016.
Sajjadi et al. (2016) [62] performed a sensitivity analysis using catalyzed-transesterification as a
renewable energy production system [63] by an ANFIS-based methodology. Influential parameters on
transesterification yield should be analyzed and predicted. ANFIS was applied in this paper to select
the most critical parameters based on operational variables. Experimental results were used to extract
training data for an adaptive neuro-fuzzy inference system network. The robustness of the presented
method was verified by the simulation results.
3.7. Decision Trees

The decision tree method is used to approximate discrete-valued target functions that the learned
function is illustrated by a decision tree. These methods are among the most powerful inductive
inference algorithms and are successfully used in many different energy systems. Table 9 presents
Table 9. Notable decision trees models used in energy systems.

IEEE Transactions on Railway electric energy systems optimal
2018 Aguado et al. [64]
Smart Grid operation
Electric Power Systems Security dispatch method for coupled natural
2016 Costa et al. [65]
Research gas and electric power networks
Prediction of the risk of a blackout in electric
2017 Kamali et al. [66] Applied Energy
energy systems
2016 Moutis et al. [67] Applied Energy energy storage planning and energy controlling
Total cost minimization in energy systems for
2016 Ottesen [68] Energy
the prosumers’ buildings
Energies 2019, 12, 1301 18 of 42
Aguado et al. (2018) [64] introduced a methodology for railway electric energy systems optimal
operation considering PV panels and wind turbines as renewable energy sources, hybrid electric
energy storage systems, and regenerative braking capabilities. The uncertainties related to renewable
energies were considered through a scenario tree methodology. All the compliments were coupled
into a multi-period optimal power flow problem. Results were reported for different cases for different
operation modes.
Costa et al. (2016) [65] presented a security dispatch method based on decision trees, which can
be applied to coupled natural gas and electric power networks against contingencies that may cause
interruptions. Preventive measures to the optimal gas production and electric energy generation were
performed based on boundaries of controllable variables and security regions determined by decision
trees. The decision tree’s rules give details of the security regions were tractable constraints and
were included in the optimization procedures of gas production and power generation rescheduling.
Kamali et al. (2017) [66] presented a novel two-stage method to predict the risk of a blackout in an
electric power network. Firstly, electric islands’ boundaries were determined to utilize a mixed integer
nonlinear programming method that optimized the cost of load curtailment and power generation
re-dispatch. Secondly, a data-mining method was completed to forecast the risk of an electric island
separation from the rest of the network. Several scenarios such as island and non-island situations
were analyzed and then used by the decision tree classification method to forecast a possible blackout.
Moutis et al. (2016) [67] presented a novel tool for utilization of decision trees for planning storage
systems in microgrids and controlling energy resources to balance energy for planned community
microgrids. The presented methodology was validated by sensitivity analysis for several case studies.
A test implementation was introduced for the utilization of distributed controller hardware to run
the algorithm of energy balancing in real-time. Ottesen et al. (2016) [68] used decision tree for energy
balancing and planning for planned community microgrids. The goal of this paper was to minimize the
total cost by trading in an electricity market and considering grid tariffs costs, imbalance penalization
and fuels use. The flexibility properties of the energy systems in the buildings of prosumers was
modeled bidding rules and handling the interrelations between hours were considered. Uncertain
parameters’ information structure was captured through scenario trees. Therefore, a two-stage
stochastic mixed-integer linear program was applied for bidding decisions and scheduling.
3.8. Deep Learning

Deep learning aims at modeling the hierarchical characterization behind data prediction patterns
through stacking multi-layer information processing modules. Increasing computing power and
increasing data size resulted in the popularity of deep learning. Table 10 presents some important
papers in this field.
Table 10. Notable deep learning models used in energy systems.

2018 Chemali et al. [69] Journal of Power Sources Battery State-of-charge estimation
2017 Coelho et al. [70] Applied Energy Household electricity demand forecasting
Computational Intelligence Estimation of the power consumption of
2017 Kim et al. [71]
and Neuroscience individual appliances in the distribution system
Sustainable Energy, Grids,
2016 Mocanu et al. [72] Prediction of building energy consumption
and Networks
Energy Conversion and
2017 Wang et al. [73] PV power forecasting
Management
Energies 2019, 12, 1301 19 of 42
Chemali et al. (2018) [69] introduced a machine learning methodology for state of charge (SOC)
estimation in Li-ion batteries utilizing deep neural networks. In this study, a new approach utilizing
deep neural networks was presented for estimating battery SOC. Training data were generated in
the laboratory by applying drive cycle loads for different ambient temperatures to a Li-ion battery.
As a result, the battery can be exposed to variable dynamics. The ability of deep neural networks
to encode dependencies into the network weights in real time was demonstrated. Coelho et al.
(2017) [70] presented a deep learning model for Graphics Processing Unit for time series forecasting.
A new parallel methodology for time series learning was designed. The presented methodology
was applied in a hybrid metaheuristic model for mini/microgrid forecasting problem (household
electricity demand). Calculated results demonstrated that the presented graphics processing unit
learning methodology was a robust deep learning tool to be used in smart sensors. Kim et al. (2017) [71]
presented a nonintrusive load monitoring method based on advanced deep learning. In this study, an
energy disaggregation utilizing advanced deep learning and long short-term memory recurrent neural
network model was proposed. Then, a new signature to upgrade the proposed model classification
performance in multistate appliance case was designated. It was demonstrated that the combination
between novel signature and advanced deep learning could be a robust solution for improving load
identification performance.
Mocanu et al. (2016) [72] introduced two novel models, namely Factored Conditional Restricted
Boltzmann Machine and Conditional Restricted Boltzmann Machine, for prediction of energy
consumption time series. The models were evaluated by four years of one-minute electric consumption
data gathered from a residential building. The results demonstrated that Factored Conditional
Restricted Boltzmann Machine outperformed ANN, SVM, RNN, and CRBM. In this study, predictions
were performed in various scenarios. RMSE and correlation coefficient were employed to compare the
results to choose the best model. Figure 18 presents the RMSE (a) and correlation coefficient (b) values
in case of prediction for a year with a weekly resolution.
Wang et al. (2017) [73] introduced a new method for photovoltaic electricity forecasting utilizing
wavelet transform and deep convolutional neural network. The deep convolutional neural network
was used to extract invariant structures and the nonlinear features exhibited in each frequency.
Numerical results demonstrated that the introduced methods could improve the accuracy of forecasting
for various seasons and different prediction horizons. Figure 19 presents the average RMSE values in
two scenarios, 45- and 75-min-ahead PV power forecasting.
Energies 2019, 12, 1301 20 of 42
(a)
(b)
Figure18.
Figure 18.(a)(a) RMSE
RMSE andand (b) correlation
(b) correlation coefficient
coefficient values
values for studyfor
by study
Mocanubyet Mocanu
al. (2016).etReproduced
al. (2016).
Reproduced
from from [72],
[72], Elsevier: 2016.Elsevier: 2016.
Wang et al. (2017) [73] introduced a new method for photovoltaic electricity forecasting utilizing
wavelet transform and deep convolutional neural network. The deep convolutional neural network
was used to extract invariant structures and the nonlinear features exhibited in each frequency.
Numerical results demonstrated that the introduced methods could improve the accuracy of
forecasting for various seasons and different prediction horizons. Figure 19 presents the average
RMSE values in two scenarios, 45- and 75-min-ahead PV power forecasting.
Figure 19.RMSE
Figure19. RMSEvalues
valuesfor
forthe
thestudy
studyby
byWang
Wang et
et al.
al. (2017).
(2017). Reproduced
Reproducedfrom
from [73],
[73], Elsevier:
Elsevier: 2017.
2017.
3.9. Ensemble Methods

Ensemble methods employ multiple learning algorithms in machine learning and statistics, in
order to reach the best modeling performance compared any other single learning algorithms. In
statistical mechanics, the ensemble method contains only a concrete finite set of alternative models
Energies 2019, 12, 1301 21 of 42
3.9. Ensemble Methods

Ensemble methods employ multiple learning algorithms in machine learning and statistics, in
order to reach the best modeling performance compared any other single learning algorithms. In
statistical mechanics, the ensemble method contains only a concrete finite set of alternative models
but allows for a flexible architecture to exist among alternative models [74]. Table 11 presents some
important papers in this field.
Table 11. Notable ensemble models used in energy systems.

forecasting of building
2015 Burger and Moura [75] Energy and Buildings
electricity demand
International Journal of Control and Non-linear fault features
2017 Changfeng et al.
Automation extraction
Cooling load forecasting in
2018 Fu, G. [76] Energy
buildings
Human energy expenditure
2015 Gjoreski et al. [77] Applied Soft Computing Journal
estimation
International Journal of Innovative Prediction of the underground
2016 Hasan and Twala [78]
Computing, Information, and Control water dam level
Burger and Moura (2015) [75] worked on the generalization of electricity demand forecasting by
formulating an ensemble learning method to perform model validation. By learning from data streams
of electricity demand, this method needed little information about energy end use, which made it
desirable for real utilization.
Changfeng et al. (2017) used ensemble empirical mode decomposition (EEMD) and multiclass
relevance vector machine for diagnosis of faults of self-validating air data sensing system. The
EEMD working principle was highlighted for distinct faults features extraction. The multiclass
relevance vector machine was utilized for fault diagnosis in a self-validating air data sensing system.
By the failure mode analysis and prototype design of the self-validating air data sensing system,
an experimental system was designed to verify the execution of the presented methodology.
Fu (2018) [76] presented an ensemble approach for forecasting of the cooling load of the
air-conditioning system. The presented approach was used for deterministic forecasting of the cooling
load with high precision. In this approach a deep belief network, empirical mode decomposition, and
the ensemble technique were utilized. The data series of the original cooling load was decomposed
into several components. The ensemble method was used to mitigate the influence of uncertainties,
such as data noise and model uncertainty, on forecasting precision. Figure 20 presents the RMSE
values of the study for each season by the employed models.
Gjoreski et al. (2015) [77] utilized the ensemble method for estimation of human energy expenditure.
In this paper, a multiple contest ensemble method was presented to extract multiple features from
the sensor data. Every feature was utilized as a context for building multiple regression models and
applying other features as training data. The models related to the feature values in the evaluated
sample were assembled regression models ensemble to estimate the energy expenditure of the user.
Figure 21 presents the RMSE values of the study for each activity by the employed models.
Hasan and Twala (2016) [78] presented an ensemble technique to monitor and predict the
underground water dam level. Six different classifier methods were applied for this goal. The paper
introduced a new method to select the most appropriate classifiers to construct the most accurate
ensemble. This methodology was based on the determination of the amount of mutual information
between pairs of classifiers and was utilized to find the optimum number of classifiers to build the
most precise ensemble.
Energies 2019, 12, 1301 22 of 42
Figure 20. RMSE values for study by Fu (2018). Reproduced from [76], Elsevier: 2018.
Gjoreski et al. (2015) [77] utilized the ensemble method for estimation of human energy
expenditure. In this paper, a multiple contest ensemble method was presented to extract multiple
features from the sensor data. Every feature was utilized as a context for building multiple regression
models and applying other features as training data. The models related to the feature values in the
evaluated sample were assembled regression models ensemble to estimate the energy expenditure of
the user. Figure 21 presents the RMSE values of the study for each activity by the employed models.
Figure20.
Figure RMSE values
20.RMSE values for
for study
study by
by Fu
Fu (2018).
(2018). Reproduced
Reproducedfrom
from[76],
[76],Elsevier:
Elsevier:2018.
2018.
Gjoreski et al. (2015) [77] utilized the ensemble method for estimation of human energy
expenditure. In this paper, a multiple contest ensemble method was presented to extract multiple
features from the sensor data. Every feature was utilized as a context for building multiple regression
models and applying other features as training data. The models related to the feature values in the
evaluated sample were assembled regression models ensemble to estimate the energy expenditure of
the user. Figure 21 presents the RMSE values of the study for each activity by the employed models.
Figure21.
Figure RMSEvalues
21.RMSE valuesfor
forstudy
studyby
by Gjoreski
Gjoreski et
et al.
al. (2015).
(2015). Reproduced
Reproducedfrom
from[77],
[77],Elsevier:
Elsevier:2015.
2015.
3.10. Hybrid ML Models

underground
Hybrid modelswater benefit
dam level.fromSixmultiple
differentMLclassifier
methodsmethods
and/orwere applied
other for this goal.and
soft computing Theartificial
paper
introduced methods.
intelligence a new method Datatopreprocessing
select the most appropriate
and classifiers
optimization to construct
tools have now becomethe most accurateto
common
ensemble.
produce This methodology
high-accuracy hybrid was
modelsbased
foron the determination
improved predictionofcapabilities.
the amountInofthesemutual information
models, usually,
between pairs of classifiers and was utilized to find the optimum number of classifiers
one part is for prediction or acts as an estimator, and the other part acts as an optimizer. These models to build the
most precise ensemble.
are mainly employed when there is a need for an accurate estimation. ANFIS and WNN are among
Figure 21. RMSE values for study by Gjoreski et al. (2015). Reproduced from [77], Elsevier: 2015.
the early generation of hybrid models [36,79]. Table 12 presents some important papers in this field.
Hybrid models
underground benefit
Table 12.from
water dam level. Six multiple
Notable papersML methods
of hybrid
different classifier and/or
models
methods other
in energy
were soft computing
systems.
applied for this goal.andTheartificial
paper
intelligenceamethods.
introduced new methodDatatopreprocessing
select the most and optimization
appropriate tools have
classifiers now become
to construct the most common
accurate to
produce high-accuracy hybrid
ensemble. This methodology was based models for improved prediction capabilities. In these models, usually,
Journalonof the determination
Renewable and of the amount load
Short-term of mutual information
forecasting in
one part is
2018
between forDeng
pairs prediction or acts
et al. [80]
of classifiers and as
wasanutilized
estimator,to and
find the
theother part acts
optimum as anof
number optimizer. These models
Sustainable Energy microgrids to build the
classifiers
most precise ensemble. Electric Power Components prediction of renewable energy loads in
2016 Dou et al. [81]
and Systems microgrids
2016 Peng et al. [82] Energies Electric load forecasting
Hybrid models
2016 benefit
Qu et al. [83] from multiple
Advances ML methods and/or other
in Meteorology softwind
Reliable computing and artificial
speed prediction
intelligence methods. Data preprocessing and optimization tools have now become common to
2017 Yang and Lian [84] Applied Energy Electricity price prediction
produce high-accuracy hybrid models for improved prediction capabilities. In these models, usually,
2019 Dehghani et al. [85] Energies Hydropower generation forecasting
one part is for prediction or acts as an estimator, and the other part acts as an optimizer. These models
2017 Mosavi et al. [86] Intelligent Systems General energy sectors
2016 Peng et al. [82] Energies Electric load forecasting
2016 Qu et al. [83] Advances in Meteorology Reliable wind speed prediction
2017 Yang and Lian [84] Applied Energy Electricity price prediction
Hydropower generation
2019 Dehghani et al. [85] Energies
forecasting
Energies 2019, 12, 1301 23 of 42
2017 Mosavi et al. [86] Intelligent Systems General energy sectors
Deng etetal.al.(2018)
Deng (2018)
[80][80] presented
presented a hybrid
a hybrid short-term
short-term load-predicting
load-predicting model optimized
model optimized by switchingby
switching
delayed delayed
particle particle
swarm swarm optimization.
optimization. In this
In this study, thisstudy,
method this
wasmethod was based
proposed proposed based on
on switching
switching
delayed delayed
particle particle
swarm swarm optimization,
optimization, extrememachine
extreme learning learningwith
machine withkernels
different differentandkernels and
empirical
empirical
mode mode decomposition.
decomposition. At the first At the the
stage, firstload
stage, the load
database database
history history was decomposed
was decomposed into independentinto
independent
intrinsic modeintrinsic
functions,mode
andfunctions, and
the intrinsic the intrinsic
mode function mode
sample function
entropysample
values entropy values were
were computed. The
computed. The intrinsic mode function was categorized into three groups. Then
intrinsic mode function was categorized into three groups. Then the extreme learning machine was the extreme learning
machinetowas
applied applied
predict to predict
the three theLastly,
groups. three thegroups. Lastly,results
prediction the prediction resultstowere
were gathered gathered
achieve to
the final
achieve the final prediction result. The experimental results showed that the
prediction result. The experimental results showed that the presented perdition model was robust. presented perdition
model 22
Figure was robust.the
presents Figure
RMSE 22 values
presentsof the RMSEfor
the study values
each of thebystudy
load for each load
the employed by the
models. employed
In this figure,
models.
L1, In this
L2, and L3 arefigure, L1, L2,datasets
user-load and L3 for
are three
user-load datasetslocated
micro-grids for threein micro-grids located
Beijing, Yanqing, in Beijing,
Guangdong
Yanqing, Guangdong
Province, Dong’ao Island, Province, Dong’aoTurpan,
and Xinjiang, Island, respectively.
and Xinjiang, Turpan, respectively.
Figure 22. RMSE values for study by Deng et al. (2018). Reproduced from [80], Elsevier: 2018.
Figure 22. RMSE values for study by Deng et al. (2018). Reproduced from [80], Elsevier: 2018.
Dou et al. (2016) [81] presented energy management strategies for a microgrid by utilizing a
Dou et al. (2016) [81] presented energy management strategies for a microgrid by utilizing a
renewable energy source and load prediction. An energy management system based on a two-level
renewable energy source and load prediction. An energy management system based on a two-level
multi-agent was built. Then, in the upper-level EMA, strategies of the energy management were
multi-agent was built. Then, in the upper-level EMA, strategies of the energy management were
constructed by using a PSO method based on renewable energy sources and load probabilistic
constructed by using a PSO method based on renewable energy sources and load probabilistic
forecasting. Ensemble empirical mode decomposition coupled with sparse Bayesian learning was
forecasting. Ensemble empirical mode decomposition coupled with sparse Bayesian learning was
used for forecasting of the lower-level renewable energy source and load agents. Simulation results
used for forecasting of the lower-level renewable energy source and load agents. Simulation results
examined the validity of the proposed method.
examined the validity of the proposed method.
Peng et al. (2016) [82] introduced a method to hybridize differential empirical mode decomposition
Peng et al. (2016) [82] introduced a method to hybridize differential empirical mode
and quantum particle swarm optimization algorithm with support vector regression in electric load
decomposition and quantum particle swarm optimization algorithm with support vector regression
forecasting. The differential empirical mode decomposition method was applied for decomposing
the electric load to several parts related to high frequencies and an approximate part related to low
frequencies. The quantum particle swarm optimization algorithm was utilized for optimizing the
parameters of support vector regression. The validation of the method demonstrated that it could
provide forecasting with good precision and interpretability. Qu et al. (2016) [83] introduced a hybrid
model for wind speed forecasting based on fruit fly optimization algorithm and ensemble empirical
mode decomposition. The original data of wind speed was divided into a set of signal components
using ensemble empirical mode decomposition. Then, the fruit fly optimization algorithm was used
to optimize parameters of prediction artificial intelligence models. The final prediction values were
acquired by reconstructing the refined series. The empirical results demonstrate that the presented
hybrid model was better than some of the existing forecasting models. Figure 23 presents the RMSE
values of the study by the employed models.
ensemble empirical mode decomposition. The original data of wind speed was divided into a set of
signal components using ensemble empirical mode decomposition. Then, the fruit fly optimization
signal components using ensemble empirical mode decomposition. Then, the fruit fly optimization
algorithm was used to optimize parameters of prediction artificial intelligence models. The final
algorithm was used to optimize parameters of prediction artificial intelligence models. The final
prediction values were acquired by reconstructing the refined series. The empirical results
prediction values were acquired by reconstructing the refined series. The empirical results
demonstrate that the presented hybrid model was better than some of the existing forecasting models.
demonstrate that the presented hybrid model was better than some of the existing forecasting models.
Figure 2019,
Energies 12, 1301 the RMSE values of the study by the employed models.
23 presents 24 of 42
Figure 23 presents the RMSE values of the study by the employed models.
Figure 23. RMSE

RMSE values
values for
for study
study by
by Qu
Qu et al. (2016).
(2016). Reproduced
Reproduced from
from [83],
[83], Elsevier:
Elsevier: 2016.
Figure 23. RMSE
Figure 23. values for study by Qu et al. (2016).
et al. Reproduced from [83], Elsevier: 2016.
2016.
Yang et al.
Yang et al. (2017)
(2017) [84]
[84] presented
presented aaa hybrid
hybrid model
model for
for electricity
electricity price
price forecasting
forecasting utilizing
utilizing ARMA,
ARMA,
Yang et al. (2017) [84] presented hybrid model for electricity price forecasting utilizing ARMA,
wavelet transform,
wavelet transform, and and KELM
KELM methods.
methods. SAPSO
SAPSO was
was applied
applied for
for searching
searching the
the optimal kernel
optimal kernel
wavelet transform, and KELM methods. SAPSO was applied for searching the optimal kernel
parameters. After
parameters. After the
the test
testofofthe
the wavelet
waveletdecomposition
decomposition components,
components, the ARMA
the ARMA model
model waswasused to
used
parameters. After the test of the wavelet decomposition components, the ARMA model was used to
predict
to stationary
predict stationary series.
series. SAPSO-KELM
SAPSO-KELM model.
model. The
Theproposed
proposed method
method performance
performance is
is validated
validated
predict stationary series. SAPSO-KELM model. The proposed method performance is validated
utilizing electricity price
utilizing electricity price data fromfrom several cities.
cities. The
The real
real data demonstrated that that the presented
utilizing electricity price datadata from several
several cities. The real data demonstrated
data demonstrated that the presented
the presented
method
method was more
was more accurate
accurate than
than individual
individual methods. Figure
methods. Figure 24 presents
24 presents the RMSE
the RMSE values
values of
of the
the study
study
method was more accurate than individual methods. Figure 24 presents the RMSE values of the study
for each
for each season by the employed models.
for each season
season byby the
the employed
employed models.
models.
Figure 24. RMSE

values for
for the
the study
study by
by Yang et al.
Yang et al. (2017).
(2017). Reproduced
Reproduced from [84], Elsevier:
from [84], Elsevier: 2017.
2017.
Figure 24. RMSE values for the study by Yang et al. (2017). Reproduced from [84], Elsevier: 2017.
Renewable energy systems such as wind and solar are site-dependent and highly difficult to
predict [7,87,88]. The prediction model using hybrid ML models effectively contributes to increased
solar energy production [89]. The economic and environmental aspects of solar photovoltaic as a
renewable energy source have caused a significant rise in the number of PV panels in recent years. The
high level of computational power and data has empowered ML models for more precise predictions.
Due to the significance of prediction in solar photovoltaic power output for decision makers in the
energy industry, ML models are employed extensively. Table 13 lists some critical papers in this field.
Energies 2019, 12, 1301 25 of 42
Table 13. Notable hybrid ML models in solar energy systems.
Year Reference Journal ML Model Application

Hybrid Forecasting of the solar
2016 David et al. [90] Solar Energy
ARMA-GARCH model irradiance
GRNN, RF, ELM and
International Journal
2017 Feng et al. [91] optimized back Estimating daily Hd
of Hydrogen Energy
propagation GANN
Gradient boosting, RF
2017 Hassan et al. [92] Applied Energy Modeling solar radiation
and bagging
Estimation of daily global
Salcedo-Sanz et al. A hybrid CRO-ELM
2018 Applied Energy solar radiation in
[93] model.
Queensland, Australia.
Salcedo-Sanz et al. A hybrid CCRO-ELM Global solar radiation
2017 Renewable Energy
[94] model. prediction at a given point
Forecasting the output
Hybrid of MA, AR and
2017 Touati et al. [95] Renewable Energy power of PV panels in
ARMA modeling
environmental conditions.
linear quadratic
2017 Voyant et al. [96] Energy Prediction of solar yields
estimation
Hybrid of multilayer Forecasting of global
2017 Voyant et al. [97] Energy
perceptron radiation time series
David et al. (2016) [90] evaluated performances of a combination of ARMA and GARCH models
in econometrics to establish solar irradiance probabilistic forecasts. A testing procedure has been
utilized to evaluate probabilistic forecasts and point forecasts. The results are presented in Table 14
and Figure 25.
Table 14. Results related to the study by David et al. (2016). Reproduced from [90], Elsevier 2016.
Method RMSE
Recursive ARMA 20.8%
SVR 20.8%
NN 20.6%
Energies 2019, 12, x FOR PEER REVIEW AR 21.3% 26 of 43
Figure 25.
Figure RMSE values
25. RMSE values for
for study
study by
by David
David et
et al.
al. (2016).
(2016). Reproduced
Reproduced from
from [90],
[90], Elsevier:
Elsevier: 2016.
2016.
As is clear from Table 14 and Figure 25, Recursive ARMA has a low value for RMSE compared
with other models. Therefore it can be claimed that the presented model can carry out point forecasts
as accurately as other models established on machine learning techniques, and the accuracy of the
proposed model is same as the other machine learning techniques for both point and probabilistic
forecasts.
Energies 2019, 12, 1301 26 of 42
As is clear from Table 14 and Figure 25, Recursive ARMA has a low value for RMSE compared
with other models. Therefore it can be claimed that the presented model can carry out point
forecasts as accurately as other models established on machine learning techniques, and the accuracy
of the proposed model is same as the other machine learning techniques for both point and
probabilistic forecasts.
Feng et al. (2017) [91] incorporated GRNN, RF, ELM, and optimized back propagation GANN to
estimate daily Hd for two stations in northern China. All presented artificial models were compared
with the empirical model (Table 15 and Figure 26).
Table 15. Results related to the study by Feng et al. (2017). Reproduced from [91], Elsevier: 2017.
Station ML Model RMSE r

ELM 17.3 0.9196
GANN 17.1 0.9209
Beijing RF 18.3 0.9102
GRNN 19.2 0.8902
Iqbal 32.9 0.8865
ELM 13.8 0.947
GANN 13.4 0.9515
Zhengzhou RE 15 0.9408
Energies 2019, 12, x FOR PEER REVIEW GRNN 16.5 0.9072 27 of 43
Iqbal 35.8 0.929
(a) (b)
Figure (a) RMSE
Figure26.26. and (b)
(a) RMSE andr values for study
(b) r values bystudy
for Feng etbyal.Feng
(2017).etReproduced
al. (2017). from [91], Elsevier:
Reproduced from 2017.
[91],
Elsevier: 2017.
Hassan et al. (2017) [92] presented ensemble models for solar radiation modeling. Gradient
Based
boosting, RFon
andFigure
bagging26 and
wereTable 15, theto
developed GANN model
estimate presents
radiation the best
in hourly andaccuracy
daily timeduescales.
to its These
lowest
RMSE and highest r values compared with those for others for both Beijing and
novel ensemble models were developed to generate synthetic radiation data to be utilized to simulate Zhengzhou stations.
Hassan et al.
the performance (2017)
of solar [92] presented
energy systems with ensemble models
different for solar radiation
configurations. Figure 27 modeling.
presents theGradient
results
boosting, RF and bagging were developed to estimate radiation in hourly and daily
of the study in detail. In this study, D1 is a daily model with day number, sunshine fraction as inputs time scales. These
novel ensemble models were developed to generate synthetic radiation data to be
and horizontal global irradiation as the output of the model. D2 is a daily model with global clearnessutilized to simulate
the performance
index, day numberofas solar energy
inputs andsystems
a diffusewith different
fraction as theconfigurations.
output of the Figure
model.27D3 presents themodel
is a daily results
of the
with studyclearness
global in detail.index,
In thisday
study, D1 is as
number a daily
inputs,model
and with
normaldayclearness
number, index
sunshine fraction
as the outputasof
inputs
the
and horizontal global irradiation as the output of the model. D2 is a daily
model. H1 is an hourly model with horizontal global irradiation, sunshine time, day number asmodel with global clearness
index,and
inputs, dayhorizontal
number asglobal
inputs and a diffuse
irradiance as thefraction as the
output of themodel.
outputH2 of the
is anmodel.
hourlyD3 is a with
model dailyglobal
model
clearness index, sunshine fraction, day number as inputs, and a diffuse fraction as the output ofofthe
with global clearness index, day number as inputs, and normal clearness index as the output the
model. H3 is the hourly model with global clearness index, sunshine time, day number as inputs, andas
model. H1 is an hourly model with horizontal global irradiation, sunshine time, day number
inputs,clearness
normal and horizontal
index global irradiance
as the output as the
of the output of the model. H2 is an hourly model with global
model.
clearness index, sunshine fraction, day number as inputs, and a diffuse fraction as the output of the
model. H3 is the hourly model with global clearness index, sunshine time, day number as inputs, and
normal clearness index as the output of the model.
with global clearness index, day number as inputs, and normal clearness index as the output of the
model. H1 is an hourly model with horizontal global irradiation, sunshine time, day number as
inputs, and horizontal global irradiance as the output of the model. H2 is an hourly model with global
clearness index, sunshine fraction, day number as inputs, and a diffuse fraction as the output of the
model. H3 is the hourly model with global clearness index, sunshine time, day number as inputs, and
Energies 2019, 12, 1301 27 of 42
normal clearness index as the output of the model.

(a)
(b)
Figure 27. (a)
Figure 27. (a) RMSE
RMSE and
and (b)
(b) rr values
values for
for study by Hassan
study by Hassan et
et al.
al. (2017).
(2017). Reproduced
Reproduced from
from [92],
[92],
Elsevier: 2017.
Elsevier: 2017.
Based
Based onon Figure
Figure 27,27, generally, SVR has
generally, SVR has the
the best
best prediction
prediction ability
ability compared
compared with
with the
the other
other
techniques because it has a high correlation coefficient and a low average RMSE compared with the
other
other models
models employed
employed by by Hassan et al.
Hassan et al. [14].
[14]. Salcedo-Sanz
Salcedo-Sanz et et al.
al. (2018)
(2018) [93]
[93] integrated
integrated the
the CRO
CRO with
with
the ELM model in their study. The presented algorithm was applied in two stages. An ELM algorithm
was used
was used for
for the
the feature
feature selection
selection process,
process, and
and solar
solar radiation
radiation waswas estimated
estimated using
using the
the optimally
optimally
screened variables by the CRO‒ELM model (Figure 28).
screened variables by the CRO-ELM model (Figure 28).
Based on Figure 28, the hybrid CRO-(ELM)-ELM model has the highest accuracy compared with
that for hybrid CRO-(ELM)-MLR, CRO-(ELM)-MARS, and CRO-(ELM)-SVR and the GGA models.
Generally, the CRO-based hybrid system is carefully screened through a wrapper-based modeling
system. The hybrid CRO-(ELM)-ELM model presents clearer advantages compared with the alternative
machine learning approaches.
Salcedo-Sanz et al. (2017) [94] studied the prediction of global solar radiation at a given point
incorporating a multilayer perceptron trained with extreme learning machines. A coral reefs
optimization algorithm with species was used to reduce the number of significant predictive variables.
Based on the results (Figure 28), the proposed model (CRO-SP) has been tested by Toledo (Spain)
data. The average best result of RMSE was equal to 69.19 (W/m2 ), which led to higher accuracy of
predictions compared with other machine learning techniques. This claim is evident in Figure 29,
which presents the average values of RMSE for the four developed techniques.
(a)
other models employed by Hassan et al. [14]. Salcedo-Sanz et al. (2018) [93] integrated the CRO with
was used for the feature selection process, and solar radiation was estimated using the optimally
screened
Energies variables
2019, 12, 1301 by the CRO‒ELM model (Figure 28). 28 of 42

(a) 29 of 43
Figure 28. (a) RMSE and (b) r values for study by Salcedo-Sanz et al. (2018). Reproduced from [93],
Elsevier: 2018.
Based on Figure 28, the hybrid CRO-(ELM)-ELM model has the highest accuracy compared with
that for hybrid CRO-(ELM)-MLR, CRO-(ELM)-MARS, and CRO-(ELM)-SVR and the GGA models.
Generally, the CRO-based hybrid system is carefully screened through a wrapper-based modeling
system. The hybrid CRO-(ELM)-ELM model presents clearer advantages compared with the
alternative machine learning approaches.
Salcedo-Sanz et al. (2017) [94] studied the prediction of global solar radiation at a given point
incorporating a multilayer perceptron trained with extreme learning machines. A coral reefs
optimization algorithm with species was used to reduce the number of significant predictive
variables. Based on the results (Figure 28), the proposed model (CRO-SP) has been tested by Toledo
(Spain) data. The average best result of RMSE was(b) equal to 69.19 (W/m2), which led to higher accuracy
of predictions compared with other machine learning techniques. This claim is evident in Figure 29,
Figure 28. (a) RMSE and (b) r values for study by Salcedo-Sanz et al. (2018). Reproduced from [93],
which presents the average values of RMSE for the four developed techniques.
Elsevier: 2018.
Figure 29.
Figure RMSE values
29. RMSE values for
for study
study by
by Salcedo-Sanz
Salcedo-Sanz et
et al.
al. (2017).
Reproduced from [94], Elsevier:
Elsevier: 2017.
2017.
Touati
Touatietetal.al.
(2017) [95] [95]
(2017) predicted outputoutput
predicted power from
power photovoltaic panels under
from photovoltaic different
panels underatmospheric
different
conditions. This study’s goal was to investigate photovoltaic performance in the harsh environmental
atmospheric conditions. This study’s goal was to investigate photovoltaic performance in the harsh
conditions of Qatar.
environmental The MLofmodel
conditions Qatar.was
Theused
ML to relatewas
model various
usedenvironmental factors
to relate various such as irradiance,
environmental factors
PV surface temperature, wind speed, temperature, relative humidity, dust, and cumulative
such as irradiance, PV surface temperature, wind speed, temperature, relative humidity, dust, dustand
to
power production.
cumulative dust to Figure
power 30 presents the
production. results
Figure 30 of the analysis
presents with correlation
the results coefficient.
of the analysis with correlation
coefficient.
Touati et al. (2017) [95] predicted output power from photovoltaic panels under different
atmospheric conditions. This study’s goal was to investigate photovoltaic performance in the harsh
environmental conditions of Qatar. The ML model was used to relate various environmental factors
such as irradiance, PV surface temperature, wind speed, temperature, relative humidity, dust, and
cumulative dust to power production. Figure 30 presents the results of the analysis with correlation
Energies 2019, 12, 1301 29 of 42
coefficient.
Figure 30. Correlation coefficient values for study by Touati et al. (2017). Reproduced from [95], the
Elsevier: 2017.
As is clear from Figure 30, Linear Regression and M5P tree decision algorithms have been
developed for prediction proposes equipped with CFS and RelifF to select subsets of relevant and
high-quality features. Based on the results, the M5P model equipped with RelifF creates more accurate
predictions due to its high correlation coefficient value; on the other hand, the developed models are
relatively simple and can be readily equipped to predict PV power output.
Voyant et al. (2017) [96] proposed models based on the Kalman filter to forecast global radiation
time series without utilizing historical data. These methodologies were compared with other
data-driven models with different time steps using RMSE values. The results claimed that the
proposed model improved the prediction purposes. Voyant et al. (2017) [97] presented a method to
better understand the propagation of uncertainty in the global radiation time series. In this study, the
reliability index has been defined to evaluate the validity of predictions. The presented method has
been applied to several meteorological stations. The comparisons were performed using RMSE factor.
The results were promising for successfully applying in these stations in the Mediterranean area.
There are many novel hybrid ML models proposed to forecast solar radiation. Hybrids of
the ANN method have often been used for this purpose and SVM and SVR are being used more
extensively nowadays. SVM and SVR usually have the same forecasting performance. Also, the
ensemble models were reported to generally deliver higher performance. SVR, GP, and NN have
better forecasting performance than AR in forecasting solar radiation. The RMSE values of ELM,
GANN, RF, and GRNN [18] shows that there is no meaningful difference between them in terms of
forecasting performance.
In order to integrate highly volatile wind power in a power grid, precise forecasting of wind
speed is crucial. This would result in less of a need to control the energy provided by wind, having
battery loading strategies and planning reserve plants. ML models can predict a time interval from
seconds to hours and, as a result, are essential for energy grid balancing. Table 16 shows some
critical papers in this field. The estimation of the total power collected from wind turbines in a wind
farm depends on several factors such as the location, hub height, and season. Cornejo-Bueno et al.
(2017) [98] applied different machine learning regression techniques to predict WPREs. Variables from
atmospheric reanalysis data were used as predictive inputs for the learning machine. RMSE was
employed as a comparison factor among the developed models. The results have been presented in
Figure 31. In general, GPR followed by MLP has the lowest RMSE compared with SVR and ELM for
each farm. This shows the high prediction capability of GPR and MLP models, in line with the purpose
of the study.
Energies 2019, 12, 1301 30 of 42
Table 16. Notable hybrid ML models in wind energy systems.
Year Reference Journal ML model Application

Cornejo-Bueno et al. SVR, MLP and ELM, GPs, Accurate prediction of
2017 Energies
[98] ERA-Interim reanalysis Wind Power Ramp Events
MLP, SVR, fuzzy inference
Prediction of wind speed
2018 Khosravi et al. [99] Applied Energy system, ANFIS, and group
data for Osorio wind farm
model of data handling
International Journal
Accurate wind power
2017 Burlando et al. [100] of Renewable Energy Hybrids of ANNs
forecast
Energies 2019, 12, x FOR PEER REVIEW Research 31 of 43
Predictive condition
2018 Pandit et al. [101] Energies Hybrid of GP
The wind power
monitoring
Renewable
2018 Sharifian et al. [102] accurate
PSO, the Type-2 fuzzy NN The wind power accurate
2018 Sharifian et al. [102] Energy Energy
Renewable PSO, the Type-2 fuzzy NN
forecasting
forecasting
Figure 31. RMSE

Figure 31. RMSEvalues
valuesfor
forstudy byby
study Cornejo-Bueno et al.
Cornejo-Bueno et (2017). Reproduced
al. (2017). fromfrom
Reproduced [98], Elsevier: 2017.
[98], Elsevier:
2017.
Accurate forecasting of WPREs is necessary for the efficient integration of a wind farm into an
electricity system
Accurate [103]. Khosravi
forecasting of WPREset al.
is (2018) [99] for
necessary developed models
the efficient based onof
integration a group
a windmodel of data
farm into an
handling type neural network, adaptive neuro-fuzzy inference system, ANFIS optimized
electricity system [103]. Khosravi et al. (2018) [99] developed models based on a group model of data with an
ant colony,
handling ANFIS
type neuraloptimized
network,with particle
adaptive swarm optimization
neuro-fuzzy algorithm,
inference system, ANFISANFIS optimized
optimized with anwith
ant
genetic algorithm, and multilayer feed-forward neural network. Day, month,
colony, ANFIS optimized with particle swarm optimization algorithm, ANFIS optimized withaverage air temperature,
minimum and maximum
genetic algorithm, air temperature,
and multilayer air pressure,
feed-forward neuralwind speed, relative
network. humidity,
Day, month, latitude,
average air
longitude, and top of atmosphere insolation. The group method of data handling-type
temperature, minimum and maximum air temperature, air pressure, wind speed, relative humidity, neural network
was the best-developed
latitude, longitude, andmodel.
top ofFigure 32 demonstrates
atmosphere insolation.the RMSE
The groupandmethod
of data for each
handling-type
model
neural for making
network thethe
was bestbest-developed
comparison. model. Figure 32 demonstrates the RMSE and correlation
Burlando et al. (2017) [100]
coefficient for each model for makingcompared a pure
the best ANN model and a hybrid model. Both models had
comparison.
similar performance. Both models were validated against the wind farm SCADA data. However, the
hybrid model made better predictions during high and low ranges of wind speed, and ANN better
predicted medium wind speed ranges. The results were compared using the normalized root mean
square error and the normalized mean absolute error. The best results (the lowest value of comparison
factors) were calculated for NWP height of 100 and 200 m for both layout 1 and 2.
Pandit and Infield (2018) [101] performed a study to reduce the costs of operation and maintenance
of the wind turbine. Predictive condition monitoring based on SCADA was applied to identify early
failures, boost production, limit downtime, and lower the energy cost. A Gaussian Process algorithm
was presented to roughly calculate operational curves, which can be utilized as a reference model to
recognize critical failures of the wind turbine and enhance power performance. Figure 33 presents the
correlation coefficient for the prediction results of four variables using Gaussian process compared with
(a)
Figure 31. RMSE values for study by Cornejo-Bueno et al. (2017). Reproduced from [98], Elsevier:
Energies 2019, 12, 1301 31 of 42
2017.
Accurate
the target forecasting
values. Based onofFigure
WPREs 33,isthis
necessary for the efficient
model successfully integration
estimated of a wind
the power curvefarm into an
compared
with other variables. Sharifian et al. (2018) [102] presented a new model based on the fuzzy of
electricity system [103]. Khosravi et al. (2018) [99] developed models based on a group model data
neural
handlingtotype
network neural
forecast network,
wind poweradaptive neuro-fuzzy
under uncertain inference system,
data conditions. ANFIS model
The proposed optimized
was with an ant
established
colony, ANFIS optimized with particle swarm optimization algorithm, ANFIS optimized
using a particle swarm optimization algorithm. This model was based on the neural network’s learning with
genetic algorithm, and multilayer feed-forward neural network. Day, month,
and expert knowledge of the fuzzy system. The presented model was validated against a real wind average air
temperature,
farm. minimum
The results and maximum
are presented in Figure air 34
temperature,
using RMSE airvalues
pressure,
for wind speed,
each case relative
study. As ishumidity,
evident,
latitude, longitude, and top of atmosphere insolation. The group method of data handling-type
RMSE for the first case study has the lowest value and for the fifth case study has the highest value.
neural network
Therefore, wasclaimed
it can be the best-developed
that the precisionmodel. Figure
of the 32 demonstrates
employed model for the RMSE
thefirst caseand
studycorrelation
is higher
coefficient
than that offor eachcase
other model for making the best comparison.
studies.

(a)
(b)
Figure
Figure 32. (a)RMSE
32.(a) RMSEand
and r values
(b)(b) for study
r values by Khosravi
for study et al. (2018).
by Khosravi Reproduced
et al. (2018). from [99],from
Reproduced Elsevier:
[99],
2018.
Elsevier: 2018.
Burlando
For properetintegration
al. (2017) [100] compared
of wind a pure
power into theANN model
power grid, and a hybrid model.forecasting
a high-performance Both models had
model
similar performance.
to compute Both models
the forecasting of wind were
speedvalidated againstspeed
at a reasonable the wind farm SCADA
is needed. It can bedata. However,
concluded that the
the
hybrid model made better predictions during high and low ranges of wind speed, and ANN better
multilayer perceptron ANN model has better forecasting performance for wind speed than SVM and
predicted
regression medium
trees [25].wind
Also,speed
hybridranges.
modelsThesuchresults
as thewere
ANFIScompared using
model have the normalized
better performanceroot
thanmean
SVR
square
models. error andspeed
For wind the normalized mean
forecasting, the absolute error.
hybridization Thewith
of ANFIS bestGP
results (the performance
has better lowest valuethan
of
comparison factors) were calculated for NWP height of 100 and 200 m for both layout 1 and 2.
its hybridization with PSO and GA [26].
Pandit
ML modelsand can
Infield (2018)
provide [101] energy
accurate performed a study and
consumption to reduce
demandthe costs ofand
prediction, operation and
can be used
maintenance of thelevel
at the managerial wind turbine.
such as by Predictive condition monitoring
building commissioning based on SCADA
project managers, was applied
utility companies, to
and
identify early failures, boost production, limit downtime, and lower the energy cost. A Gaussian
Process algorithm was presented to roughly calculate operational curves, which can be utilized as a
reference model to recognize critical failures of the wind turbine and enhance power performance.
Figure 33 presents the correlation coefficient for the prediction results of four variables using
Gaussian process compared with the target values. Based on Figure 33, this model successfully
Gaussian process compared with the target values. Based on Figure 33, this model successfully
estimated the power curve compared with other variables. Sharifian et al. (2018) [102] presented a
new model based on the fuzzy neural network to forecast wind power under uncertain data
conditions. The proposed model was established using a particle swarm optimization algorithm. This
model was based on the neural network’s learning and expert knowledge of the fuzzy system. The
presented model was validated against a real wind farm. The results are presented in Figure 3432
Energies 2019, 12, 1301
using
of 42
RMSE values for each case study. As is evident, RMSE for the first case study has the lowest value
and for the
facilities fifth case
managers study has energy-saving
to introduce the highest value. Therefore,
policies. it demonstrates
Table 17 can be claimed thatcritical
some the precision of
papers in
the employed
this field. model for the first case study is higher than that of other case studies.
Figure 33.
Figure 33. Correlation
Correlationcoefficient
coefficientvalues
valuesforfor study
study by Pandit
by Pandit andand Infield
Infield (2018).
(2018). Reproduced
Reproduced from
from [101],
[101], Elsevier: 2018.
Elsevier: 2018.
Figure 34. RMSE

values for
for study
study by
by Sharifian et al.
Sharifian et al. (2018).
(2018). Reproduced
Reproduced from
from [102],
[102], Elsevier: 2018.
Elsevier: 2018.
For proper
Table integration of wind
17. Notable hybrid ML power into
models in the consumption
energy power grid,and a high-performance
demand prediction. forecasting
model to compute the forecasting of wind speed at a reasonable speed is needed. It can be concluded
thatYear
the multilayer perceptron Journal
Reference ANN model has better forecasting performance for
ML Model wind speed than
Application
SVM 2016and regression

Albert and trees [25]. Also,
Applied hybrid models such as the
predictive segmentation technique ANFIS model
Predictive have better
segmentation
Maasoumy [104] Energy technique for energy companies
performance than SVR models. For wind speed forecasting, the hybridization of ANFIS with GP has
Predicting the average daily
better
2018
performance than its hybridization
Alobaidi et al. [105]
Applied with PSO and GA [26].
Hybrids energy consumption on a
ML models can provide accurate energy consumption and demand prediction,
Energy andlevel
household can be used
at the managerial level
Benedetti et al.such asApplied
by building commissioning project managers, Controlutility companies,
of energy consumption and
2016 Hybrids of ANN
facilities managers
[106]to introduce energy-saving policies. Table 17 demonstrates
Energy some critical
in energy-intensive papers
industries
in this field. ensemble learning technique
Prediction of the household
2018 Chen et al. [107] Energy (feedforward deep networks and
electricity consumption
extreme gradient boosting forest)
Table 17. Notable hybrid ML models in energy consumption and demand prediction.
Support Vector Regression,
Energy and Improving thermal comfort and
Year
2018 Reference
Kuroha et al. [108] Journal
Buildings
Particle ML Model
Swarm Optimization, Application
reduction of electricity costs
predictive Vote
Predicted Mean
Albert and Predictive segmentation technique
2016 Applied Energy
Sustainable segmentation
2018 Maasoumy
Torabi et al.[104]
[7,88] Hybrid ML models for energy
Solar companies
radiation forecasting
Energy technique
Predicting the average daily
Alobaidi et al.
2018
Albert and Applied Energy Hybrids energytechnique
Maasoumy (2016) [104] presented a predictive segmentation consumption on a the
to create
[105]
household level
targeting process and highly-interpretable segmentation for energy companies. The presented model
Benedetti et al. Control of energy consumption in
2016 Applied Energy Hybrids of ANN
[106] energy-intensive industries
ensemble learning
technique
(feedforward deep Prediction of the household
2018 Chen et al. [107] Energy
networks and electricity consumption
Energies 2019, 12, 1301 33 of 42

utilized demographics, consumption and program enrollment data to make predictive patterns. This
consumption
2019, 12, forecasting.
Energies displayed
model x FOR In this
PEER REVIEW
homogeneous paper, athat
segments prediction
were 2- framework was
to 3-fold more presentedfor
productive to targeting.
predict individual
34 of 43
Alobaidi
household
et al. (2018)average daily energy
[105] proposed consumption.
an ensemble The framework
learning results showed the robustness
for household energy of the proposed
consumption
consumption
ensemble
forecasting. model
Inforecasting.
paper,Inaprediction
to provide
this this paper,performance
prediction aframework
predictionusing
framework
was limitedwas
presented topresented
data. 35topresents
predict household
Figure individual
predict individual
the results
household
of RMSEdaily
average foraverage
each daily
model
energy energy consumption.
separately.
consumption. The resultsThe resultsthe
showed showed the robustness
robustness of the ensemble
of the proposed proposed
model to provide prediction performance using limited data. Figure 35 presents the results of results
ensemble model to provide prediction performance using limited data. Figure 35 presents the RMSE
of RMSE
for for each
each model model separately.
separately.
Figure 35. RMSE values for study by Alobaidi et al. (2018). Reproduced from [105], Elsevier: 2018.
Benedetti
Figure 35.et
Figure 35. al. (2016)
RMSE
RMSE values[106]
values for introduced
for study
study by a new
by Alobaidi
Alobaidi al.methodology
et al.
et (2018). for control
(2018). Reproduced
Reproduced from automation
from [105],
[105], Elsevier: of energy
2018.
Elsevier: 2018.
consumption utilizing adaptive algorithms and artificial neural networks. Three neural network
Benedetti et al. (2016) [106] introduced a new methodology for control automation of energy
Benedetti
structures wereetpresented
al. (2016) and
[106]trained
introduced a new
to deal with methodology
an enormous for control
amount automation
of data. of energy
Three indicators
consumption utilizing adaptive algorithms and artificial neural networks. Three neural network
consumption utilizing adaptive algorithms and artificial neural networks.
were used to identify the best structure for creating a control tool for energy consumption. Three neural network
The
structures were presented and trained to deal with an enormous amount of data. Three indicators were
structuresofwere
accuracy presented
the model and trained toFinally,
was investigated. deal with an enormous
the model amount
was applied to aofcase
data. Three
study of aindicators
building
used to identify the best structure for creating a control tool for energy consumption. The accuracy of
were
in used
Rome, to identify the best structure for creating a control tool for energy consumption. The
Italy.
the model was investigated. Finally, the model was applied to a case study of a building in Rome, Italy.
accuracy
Chen ettheal.model
of (2018)was investigated.
[107] worked on Finally,
a novelthe model
approachwasfor
applied to a case
predicting study of aelectricity
residential building
Chen et al. (2018) [107] worked on a novel approach for predicting residential electricity
in Rome, Italy.using ensemble learning. In this study, a data-driven framework was introduced to
consumption
consumption using ensemble learning. In this study, a data-driven framework was introduced to
Chen
forecast theetannual
al. (2018) [107] consumption
electricity worked on aofnovelhouseholdapproach for predicting
utilizing residential
ensemble learning electricity
model. Ridge
forecast the annual electricity consumption of household utilizing ensemble learning model. Ridge
consumption
regression wasusing
usedensemble
to combine learning. In thisdeep
feed-forward study, a data-driven
networks framework
and extreme gradient wasboosting
introduced to
forest.
regression was used to combine feed-forward deep networks and extreme gradient boosting forest.
forecast the annual electricity consumption of household utilizing ensemble learning
Figure 36 presents the results of the study in comparison with those of the other models. As is clear model. Ridge
Figure 36 presents the results of the study in comparison with those of the other models. As is clear
regression
from Figurewas36, used to combine
the proposed feed-forward
models deep networks
have the highest accuracyandwithextreme gradient
the lowest RMSEboosting forest.
in comparison
from Figure 36, the proposed models have the highest accuracy with the lowest RMSE in comparison
Figure
with the36other
presents the results of the study in comparison with those of the other models. As is clear
models.
with the other models.
from Figure 36, the proposed models have the highest accuracy with the lowest RMSE in comparison
with the other models.
Figure
36. RMSE values for
for the
the study
study by
by Chen et al.
Chen et al. (2018).
(2018). Reproduced
Reproduced from
from [107],
[107], Elsevier:
Elsevier: 2018.
2018.
Figure 36. RMSE values for the study by Chen et al. (2018). Reproduced from [107], Elsevier: 2018.
Energies 2019, 12, 1301 34 of 42
Kuroha et al. (2018) [108] presented an operational planning model for residential air conditioners.
In this study, the focus was on automatic air conditioners for thermal comfort improvement and
electricity cost reduction. An energy management methodology was introduced to provide an air
conditioner operation plan by learning the installation environment characteristics from result data of
the historic operation. Based on the results, the proposed model could reduce the electricity cost about
39.7% compared with that for the benchmark method.
The type of data that is available today is continuously evolving. Some data already encode
information that is used as proxy metrics to predict energy consumption in buildings. For example,
geometry, size, and height can be used to predict energy consumption in buildings. Wang et al. [109]
developed the Unige Building Identifier to correspond attribute data and building energy to smooth
the way for corresponding across datasets. Depecker et al. [110] matched the consumption of heating
of the buildings and their shape. In this study, the criterion for the shape of buildings was presented.
Fourteen buildings were chosen based on their shape varieties. The results demonstrated that the
energy consumption of buildings is inversely proportionate to the building’s compactness. Qi and
Wang [111] introduced a novel model for calculation of shape coefficient of buildings utilizing Google
Earth. Astronomy principles, geometry, and GIS slope analysis were used for calculation of shape
coefficient of buildings. This new model can be used for energy-saving measures in existing buildings.
ML and big data have led to believe that personally identifiable information is released when
predicting energy patterns, and so forth. However, this is not often the case. The following studies show
how to protect data privacy while predicting and disclosing information about energy in commercial
buildings. Livingston et al. [112] presented a solution to measure the impact of modifying the utility
meter aggregation threshold for dweller privacy and on buildings that are qualified for energy usage
reporting. As the threshold rises, lesser buildings are qualified for disclosure of energy use data. This
paper’s goal was to study the resemblance between whole-building totals and individual utility meters
at various aggregation levels. Sweeney et al. [113] proposed a solution for data privacy. The solution
included a formal protection model titled k-anonymity as a series of accompanying policies. For this
definition of privacy, in a k-anonymized dataset, every record is identical from at least k-1 other records.
Machanavajjhala et al. [114] studied two problems about k-anonymity; little diversity in sensitive
attributes and background knowledge of attackers. They introduced a new privacy criterion called
l-diversity that can shield against such attacks. The hybridization of ML models in energy demand
field demonstrated that the accuracy of energy demand forecasting could improve significantly. Also,
an ensemble model has significantly higher generalization ability than ANN and SVM models, and it
has a lower uncertainty of forecasting [7,32].
3.11. Comparative Analysis of ML Models

Table 18 provides a comparative study of ML models and deep learning for prediction in
different energy systems. In this table, the complexity of ML models, user-friendliness, accuracy,
speed, and dataset type are summarized. Hybrid ML models are reported to be superior in terms of
user-friendliness, accuracy, and speed. However, the complexity of their methodologies has increased.
There is not any meaningful difference in the forecasting performance of SVR, GP, and NN
models [17]. Therefore, their quality of forecasting is the same. Considering the reviewed papers,
these models have the same statistics of error. Also, SVM and SVR have similar performance, with
no statistical forecasting performance difference among them. SVR models have lower forecasting
performance than ELM, GPR, and MLP models [20]. Decision Tree models have a lower fluctuation
of performance than SVM and NN models for quarter-hourly, daily, and weekly periods [46]. The
RMSE value of SVM models are higher than the MLP models, and therefore it has lower forecasting
performance [52]. The hybrid models have a lower RMSE value than ARIMA, EEMD-FOASVR, and
EEMD-FOAGRNN [96]. Hybridizing existing methodologies and algorithm ensemble are among the
most effective ways to improve the ML models.
Energies 2019, 12, 1301 35 of 42
Table 18. Comparative study of ML models in energy systems.
ML Model Complexity User-Friendliness Accuracy Speed Dataset Type

ANN Reasonably high Low High Reasonable Historical
MLP Reasonable Reasonable Reasonably high High Historical
ELM Reasonable Reasonably high Reasonable Reasonably high Historical
SVM Reasonably high Low High Low Historical
DT Reasonable Low Reasonable Reasonable Historical
DL High Reasonable High Reasonable Historical
Ensemble High Low Reasonable High Historical
WNN Reasonable Low High Low Historical
ANFIS Reasonable Reasonable Reasonable High Historical
Hybrids Reasonable High High High Historical
4. Conclusions
Smart sensors, smart grid, and IoT technologies have enabled big data with new challenges and
opportunities to the energy systems. The ML models have gained popularity in putting big data in the
constructive use for informed-decision and more efficient models. The state of the art of ML models in
energy systems identified 10 groups of ML algorithms with the highest popularity for model building
in energy systems, i.e., ANN, MLP, ELM, SVM, WNN, ANFIS, decision trees, deep learning, ensembles,
and advanced hybrid ML models. Diverse applications of each ML models have been reviewed
showing the popularity and effectiveness of ML models almost in any energy domain. The example
application of ML models include energy consumption prediction, demand prediction, cost prediction,
wind speed estimation, load forecasting, solar radiation prediction, ranking, various optimization
tasks, power generation forecasting, power quality estimation, time series forecasting, load perdition,
wind speed forecasting, power demand prediction, risk prediction, storage planning, energy control,
peak load management, dynamic energy pricing, cost minimization, battery charge estimation, price
prediction and more. Further investigation of the accuracy and generalization ability of the various
ML models suggested that novel hybrid models are superior to the more conventional ML models.
The trend in advancement of hybrid ML models will continue to deliver higher performance and
sophisticated energy models.
The progress of renewable energy systems is dependent on the advancement of ML models.
Note that renewable energy systems have recently gained popularity due to climate protection and
sustainability concerns, so it is more acceptable to extract energy from the wind, sun, and other
renewable energy resources. The central aspect of all renewable energy resources is their dependence on
the environment and huge barriers to controlling and planning. Furthermore, due to the expansion of
the electricity grid, it is necessary to forecast power generation and demand. Nevertheless, fluctuations
in wind and solar energies and energy demand have a negative impact on the grid and its users if not
managed properly. Therefore, ML models have become essential in such energy systems. The review
of these particular applications has demonstrated the promising robustness of the ML models. Based
on the results, the research direction is moving toward personalized ML models designed for a
particular application. This trend has shown that the highest degree of accuracy can be achieved
though case-based ML model development. Although the hybrid ML models are reported as the
optimal choice and the future trend in energy systems, further evaluation and comparison of various
hybrid ML models are essential, considering alternative comparative performance parameters besides
the RMSE and r.
Author Contributions: Review, data curation and analyzing the results, M.S., A.M., and S.F.A.; energy expertise,
M.S., T.R., S.S., and A.V.-K.; machine learning expertise, A.M., S.S., A.V.-K., and S.F.A.; management, database,
conceptualization, writing, administration and methodology, A.M.; visualization, S.F.A.; supervision, resources,
software, expertise, revision, and verifying the results, S.S., A.V.-K. and T.R.
Conflicts of Interest: The authors declare no conflict of interest.
Energies 2019, 12, 1301 36 of 42
Acronyms
AR Autoregressive
ARMA Autoregressive Moving Average
ANNs Artificial Neural Networks
ANFIS Adaptive Neuro-Fuzzy Inference System
BANN Back Propagation Neural Network
BP Back Propagation
CFS Correlation based Feature Selection
CGP Cartesian Genetic Programming
CRBM Conditional Restricted Boltzmann Machines
CRO Coral Reefs Optimization
CWNN Convolutional-Wavelet Neural Networks
DL Deep Learning
DOPH Direct Optimum Parallel Hybrid
DT Decision trees
EANN Evolutionary Artificial neural networks
EEMD Ensemble Empirical Mode Decomposition
ELM Extreme Learning Machine
EMD Empirical Mode Decomposition
FCRBM Factored Conditional Restricted Boltzmann Machine
FFNN Feed Forward Neural Network
FIS Fuzzy inference system
FOARBF Fruit Fly Optimization Algorithm Radial Basis Function
FOAGRNN Fruit Fly Optimization Algorithm Generalized Regression Neural Networks
FOASVR Fruit Fly Optimization Algorithm Support Vector Regression
GA Genetic Algorithm
GANN Neural Networks by Genetic Algorithm
GARCH Generalized Autoregressive Conditional Heteroskedasticity
GHG Greenhouse Gas
GFF General Factorization Framework
GP Gaussian Processes
GPR Gaussian Processes Regression
GRNN Generalized Regression Neural Networks
IDSS Intelligent Decision Support System
IoT Internet of Things
KELM kernel-based extreme learning machine
KFCM Kernel Fuzzy C Means
KNN K-Nearest Neighbors
LM Levenberg-Marquardt
MARS Multivariate Adaptive Regression Splines
ML Machine Learning
MLP Multilayer Perceptron
MR Multi-Resolution
MLR Multiple linear regression
MTL Method of transmission lines
MRWNN Multi-Resolution Wavelet Neural Network
NARX Nonlinear Auto-Regressive with external input
NDP Neuro-Dynamic Programming
NN Neural Networks
NWP Numerical Weather Prediction
OLR Outgoing long-wave radiation
PR Persistence model
PSO Particle Swarm Optimization
PV Photovoltaic
Energies 2019, 12, 1301 37 of 42
r Correlation coefficient
RBF Radial Basis Function
RBFNN Radial Basis Function Neural Networks
RF Random Forests
RMSE Root Mean Squared Error
RNN Recurrent Neural Network
SANN Subsequent Artificial Neural Networks
SAPSO Self-Adaptive Particle Swarm Optimization
SARIMA Seasonal Autoregressive Integrated Moving Average
SCADA Supervisory Control and Data Acquisition
SCG Scaled Conjugate Gradient
SE Solar radiation
SEAT A car manufacturing plant in Spain
SOC State of Charge
SOFM Self-Organizing Feature Map
SVM Support Vector Machine
SP Coral Reefs Optimization algorithm with species
SVR Support vector regression
WPRE Wind Power Ramp Events
WNN Wavelet Neural Network
References
1. Deshmukh, M.; Deshmukh, S. Modeling of hybrid renewable energy systems. Renew. Sustain. Energy Rev.
2008, 12, 235–249. [CrossRef]
2. Elgerd, O.I. Electric Energy Systems Theory: An introduction; McGraw-Hill: New York, NY, USA, 1982.
3. Gomez-Exposito, A.; Conejo, A.J.; Canizares, C. Electric Energy Systems: Analysis and Operation; CRC Press:
Boca Raton, FL, USA, 2018.
4. Quaschning, V. Understanding Renewable Energy Systems; Routledge: Abingdon, UK, 2016.
5. Isabella, O.; Smets, A.; Jäger, K.; Zeman, M.; van Swaaij, R. Solar Energy: The Physics and Engineering of
Photovoltaic Conversion, Technologies and Systems; UIT Cambridge Limited: Cambridge, UK, 2016.
6. Palensky, P.; Dietrich, D. Demand side management: Demand response, intelligent energy systems, and
smart loads. IEEE Trans. Ind. Inform. 2011, 7, 381–388. [CrossRef]
7. Torabi, M.; Hashemi, S.; Saybani, M.R.; Shamshirband, S.; Mosav, A. A Hybrid clustering and classification
technique for forecasting short-term energy consumption. Environ. Prog. Sustain. Energy 2018, 38, 66–76.
[CrossRef]
8. Najafi, B.; Faizollahzadeh Ardabili, S.; Mosavi, A.; Shamshirband, S.; Rabczuk, T. An Intelligent Artificial
Neural Network-Response Surface Methodology Method for Accessing the Optimum Biodiesel and Diesel
Fuel Blending Conditions in a Diesel Engine from the Viewpoint of Exergy and Energy Analysis. Energies
2018, 11, 860. [CrossRef]
9. Hosseini Imani, M.; Zalzar, S.; Mosavi, A.; Shamshirband, S. Strategic behavior of retailers for risk reduction
and profit increment via distributed generators and demand response programs. Energies 2018, 11, 1602.
[CrossRef]
10. Dineva, A; Mosavi, A.; Ardabili, S.F.; Vajda, I.; Shamshirband, S.; Rabczuk, T.; Chau, K.-W. Review of soft
computing models in design and control of rotating electrical machines. Energies 2019, 12, 1049. [CrossRef]
11. Chong, L.W.; Wong, Y.W.; Rajkumar, R.K.; Rajkumar, R.K.; Isa, D. Hybrid energy storage systems and control
strategies for stand-alone renewable energy power systems. Renew. Sustain. Energy Rev. 2016, 66, 174–189.
[CrossRef]
12. Curry, N.; Pillay, P. Biogas prediction and design of a food waste to energy system for the urban environment.
Renew. Energy 2012, 41, 200–209. [CrossRef]
13. Amarasinghe, K.; Marino, D.L.; Manic, M. Deep neural networks for energy load forecasting. In Proceedings
of the 2017 IEEE 26th International Symposium on Industrial Electronics (ISIE), Edinburgh, UK, 19–21 June
2017; pp. 1483–1488.
Energies 2019, 12, 1301 38 of 42
14. Quej, V.H.; Almorox, J.; Arnaldo, J.A.; Saito, L. ANFIS, SVM and ANN soft-computing techniques to estimate
daily global solar radiation in a warm sub-humid environment. J. Atmos. Sol.-Terr. Phys. 2017, 155, 62–70.
[CrossRef]
15. Daut, M.A.M.; Hassan, M.Y.; Abdullah, H.; Rahman, H.A.; Abdullah, M.P.; Hussin, F. Building electrical
energy consumption forecasting analysis using conventional and artificial intelligence methods: A review.
Renew. Sustain. Energy Rev. 2017, 70, 1108–1118. [CrossRef]
16. Yildiz, B.; Bilbao, J.I.; Sproul, A.B. A review and analysis of regression and machine learning models on
commercial building electricity load forecasting. Renew. Sustain. Energy Rev. 2017, 73, 1104–1122. [CrossRef]
17. Kalogirou, S.A. Applications of artificial neural-networks for energy systems. Appl. Energy 2000, 67, 17–35.
[CrossRef]
18. Faizollahzadeh Ardabili, S.; Najafi, B.; Alizamir, M.; Mosavi, A.; Shamshirband, S.; Rabczuk, T. Using
SVM-RSM and ELM-RSM Approaches for Optimizing the Production Process of Methyl and Ethyl Esters.
Energies 2018, 11, 2889. [CrossRef]
19. Al-Jarrah, O.Y.; Yoo, P.D.; Muhaidat, S.; Karagiannidis, G.K.; Taha, K. Efficient machine learning for big data:
A review. Big Data Res. 2015, 2, 87–93. [CrossRef]
20. Amasyali, K.; El-Gohary, N.M. A review of data-driven building energy consumption prediction studies.
Renew. Sustain. Energy Rev. 2018, 81, 1192–1205. [CrossRef]
21. Peng, Y.; Rysanek, A.; Nagy, Z.; Schlüter, A. Using machine learning techniques for occupancy-prediction-
based cooling control in office buildings. Appl. Energy 2018, 211, 1343–1358. [CrossRef]
22. Tsanas, A.; Xifara, A. Accurate quantitative estimation of energy performance of residential buildings using
statistical machine learning tools. Energy Build. 2012, 49, 560–567. [CrossRef]
23. Robinson, C.; Dilkina, B.; Hubbs, J.; Zhang, W.; Guhathakurta, S.; Brown, M.A.; Pendyala, R.M. Machine
learning approaches for estimating commercial building energy consumption. Appl. Energy 2017, 208,
889–904. [CrossRef]
24. Voyant, C.; Notton, G.; Kalogirou, S.; Nivet, M.-L.; Paoli, C.; Motte, F.; Fouilloy, A. Machine learning methods
for solar radiation forecasting: A review. Renew. Energy 2017, 105, 569–582. [CrossRef]
25. Mosavi, A.; Lopez, A.; Varkonyi-Koczy, A.R. Industrial applications of big data: State of the art survey.
In Proceedings of the International Conference on Global Research and Education, Ias, i, Romania, 25–28
September 2017; pp. 225–232.
26. Qasem, S.N.; Samadianfard, S.; Nahand, H.S.; Mosavi, A.; Shamshirband, S.; Chau, K.W. Estimating Daily
Dew Point Temperature Using Machine Learning Algorithms. Water 2019, 11, 582. [CrossRef]
27. Najafi, B.; Ardabili, S.F. Application of ANFIS, ANN, and logistic methods in estimating biogas production
from spent mushroom compost (SMC). Resour. Conserv. Recycl. 2018, 133, 169–178. [CrossRef]
28. Faizollahzadeh Ardabili, S.; Najafi, B.; Ghaebi, H.; Shamshirband, S.; Mostafaeipour, A. A novel enhanced
exergy method in analyzing HVAC system using soft computing approaches: A case study on mushroom
growing hall. J. Build. Eng. 2017, 13, 309–318. [CrossRef]
29. Abbas, F.; Habib, S.; Feng, D.; Yan, Z. Optimizing Generation Capacities Incorporating Renewable Energy
with Storage Systems Using Genetic Algorithms. Electronics 2018, 7, 100. [CrossRef]
30. Anwar, M.B.; El Moursi, M.S.; Xiao, W. Novel Power Smoothing and Generation Scheduling Strategies for a
Hybrid Wind and Marine Current Turbine System. IEEE Trans. Power Syst. 2017, 32, 1315–1326. [CrossRef]
31. Boukelia, T.; Arslan, O.; Mecibah, M. Potential assessment of a parabolic trough solar thermal power plant
considering hourly analysis: ANN-based approach. Renew. Energy 2017, 105, 324–333. [CrossRef]
32. Chatziagorakis, P.; Ziogou, C.; Elmasides, C.; Sirakoulis, G.C.; Karafyllidis, I.; Andreadis, I.; Georgoulas, N.;
Giaouris, D.; Papadopoulos, A.I.; Ipsakis, D. Enhancement of hybrid renewable energy systems control with
neural networks applied to weather forecasting: The case of Olvio. Neural Comput. Appl. 2016, 27, 1093–1118.
[CrossRef]
33. Gallagher, C.V.; Bruton, K.; Leahy, K.; O’Sullivan, D.T. The suitability of machine learning to minimise
uncertainty in the measurement and verification of energy savings. Energy Build. 2018, 158, 647–655.
[CrossRef]
34. Faizollahzadeh Ardabili, S.; Mahmoudi, A.; Mesri Gundoshmian, T. Modeling and simulation controlling
system of HVAC using fuzzy and predictive (radial basis function, RBF) controllers. J. Build. Eng. 2016, 6,
Energies 2019, 12, 1301 39 of 42
35. Karballaeezadeh, N.; Mohammadzadeh, S.D.; Shamshirband, S.; Hajikhodaverdikhan, P.; Mosavi, A.;
Chau, K.W. Prediction of remaining service life of pavement using an optimized support vector machine.
Eng. Appl. Comput. Fluid Mech. 2019, 13, 188–198.
36. Faizollahzadeh Ardabili, S.; Najafi, B.; Shamshirband, S.; Minaei Bidgoli, B.; Deo, R.C.; Chau, K.-w.
Computational intelligence approach for modeling hydrogen production: A review. Eng. Appl. Comput.
Fluid Mech. 2018, 12, 438–458. [CrossRef]
37. Ahmad, A.; Anderson, T.; Lie, T. Hourly global solar irradiation forecasting for New Zealand. Sol. Energy
2015, 122, 1398–1408. [CrossRef]
38. Chahkoutahi, F.; Khashei, M. A seasonal direct optimal hybrid model of computational intelligence and soft
computing techniques for electricity load forecasting. Energy 2017, 140, 988–1004. [CrossRef]
39. Kazem, H.A.; Yousif, J.H. Comparison of prediction methods of photovoltaic power system production
using a measured dataset. Energy Convers. Manag. 2017, 148, 1070–1081. [CrossRef]
40. Loutfi, H.; Bernatchou, A.; Tadili, R. Generation of Horizontal Hourly Global Solar Radiation From
Exogenous Variables Using an Artificial Neural Network in Fes (Morocco). Int. J. Renew. Energy Res.
(IJRER) 2017, 7, 1097–1107.
41. Shimray, B.A.; Singh, K.; Khelchandra, T.; Mehta, R. Ranking of Sites for Installation of Hydropower Plant
Using MLP Neural Network Trained with GA: A MADM Approach. Comput. Intell. Neurosci. 2017, 2017,
4152140. [CrossRef] [PubMed]
42. Arat, H.; Arslan, O. Optimization of district heating system aided by geothermal heat pump: A novel
multistage with multilevel ANN modelling. Appl. Therm. Eng. 2017, 111, 608–623. [CrossRef]
43. Bagnasco, A.; Fresi, F.; Saviozzi, M.; Silvestro, F.; Vinci, A. Electrical consumption forecasting in hospital
facilities: An application case. Energy Build. 2015, 103, 261–270. [CrossRef]
44. Li, Q.; Wu, Z.; Xia, X. Estimate and characterize PV power at demand-side hybrid system. Appl. Energy 2018,
218, 66–77. [CrossRef]
45. Premalatha, N.; Valan Arasu, A. Prediction of solar radiation for solar systems by using ANN models with
different back propagation algorithms. J. Appl. Res. Technol. 2016, 14, 206–214. [CrossRef]
46. Yaïci, W.; Entchev, E. Performance prediction of a solar thermal energy system using artificial neural networks.
Appl. Therm. Eng. 2014, 73, 1348–1359. [CrossRef]
47. Arabloo, M.B.A.; Ghiasi, M.M.; Lee, M.; Abbas, A.; Zendehboudi, S. A novel modeling approach to optimize
oxygen-steam ratios in coal gasification process. Fuel 2015, 153. [CrossRef]
48. Arikan, C.; Ozdemir, M. Classification of power quality disturbances at power system frequency and out of
power system frequency using support vector machines. Prz. Elektrotech 2013, 89, 284–291.
49. Ma, J.; Jiang, H.; Huang, K.; Bi, Z.; Man, K.L. Novel Field-Support Vector Regression-Based Soft Sensor
for Accurate Estimation of Solar Irradiance. IEEE Trans. Circuits Syst. I Regul. Pap. 2017, 64, 3183–3191.
[CrossRef]
50. Özdemir, S.; Demirtas, M.; Aydin, S. Harmonic estimation based support vector machine for typical power
systems. Neural Netw. World 2016, 26, 233–252. [CrossRef]
51. Pinto, T.; Sousa, T.M.; Praça, I.; Vale, Z.; Morais, H. Support Vector Machines for decision support in electricity
markets’ strategic bidding. Neurocomputing 2016, 172, 438–445. [CrossRef]
52. Zhang, J.; Walter, G.G.; Miao, Y.; Lee, W.N.W. Wavelet neural networks for function learning. IEEE Trans.
Signal Process. 1995, 43, 1485–1497. [CrossRef]
53. Doucoure, B.; Agbossou, K.; Cardenas, A. Time series prediction using artificial wavelet neural network and
multi-resolution analysis: Application to wind speed data. Renew. Energy 2016, 92, 202–211. [CrossRef]
54. Gu, J.; Wang, J.; Qi, C.; Min, C.; Sundén, B. Medium-term heat load prediction for an existing residential
building based on a wireless on-off control system. Energy 2018, 152, 709–718. [CrossRef]
55. He, Q.; Wang, J.; Lu, H. A hybrid system for short-term wind speed forecasting. Appl. Energy 2018, 226,
56. Qin, F.; Li, W.; Hu, Y.; Xu, G. An Online Energy Management Control for Hybrid Electric Vehicles Based on
Neuro-Dynamic Programming. Algorithms 2018, 11, 33. [CrossRef]
57. Sarshar, J.; Moosapour, S.S.; Joorabian, M. Multi-objective energy management of a micro-grid considering
uncertainty in wind power forecasting. Energy 2017, 139, 680–693. [CrossRef]
58. Hadi Abdulwahid, A.; Wang, S. A Novel Method of Protection to Prevent Reverse Power Flow Based on
Neuro-Fuzzy Networks for Smart Grid. Sustainability 2018, 10, 1059. [CrossRef]
Energies 2019, 12, 1301 40 of 42
59. Bassam, A.; May Tzuc, O.; Escalante Soberanis, M.; Ricalde, L.; Cruz, B. Temperature estimation for
photovoltaic array using an adaptive neuro fuzzy inference system. Sustainability 2017, 9, 1399. [CrossRef]
60. Kampouropoulos, K.; Andrade, F.; Sala, E.; Espinosa, A.G.; Romeral, L. Multiobjective optimization of
multi-carrier energy system using a combination of ANFIS and genetic algorithms. IEEE Trans. Smart Grid
2018, 9, 2276–2283. [CrossRef]
61. Mohammadi, K.; Shamshirband, S.; Kamsin, A.; Lai, P.; Mansor, Z. Identifying the most significant input
parameters for predicting global solar radiation using an ANFIS selection procedure. Renew. Sustain. Energy
Rev. 2016, 63, 423–434. [CrossRef]
62. Sajjadi, B.; Raman, A.A.A.; Parthasarathy, R.; Shamshirband, S. Sensitivity analysis of catalyzed-transesterification
as a renewable and sustainable energy production system by adaptive neuro-fuzzy methodology. J. Taiwan
Inst. Chem. Eng. 2016, 64, 47–58. [CrossRef]
63. Nosratabadi, S.; Mosavi, A.; Shamshirband, S.; Kazimieras Zavadskas, E.; Rakotonirainy, A.; Chau, K.W.
Sustainable Business Models: A Review. Sustainability 2019, 11, 1663. [CrossRef]
64. Aguado, J.A.; Racero, A.J.S.; de la Torre, S. Optimal operation of electric railways with renewable energy and
electric storage systems. IEEE Trans. Smart Grid 2018, 9, 993–1001. [CrossRef]
65. Costa, D.C.; Nunes, M.V.; Vieira, J.P.; Bezerra, U.H. Decision tree-based security dispatch application in
integrated electric power and natural-gas networks. Electr. Power Syst. Res. 2016, 141, 442–449. [CrossRef]
66. Kamali, S.; Amraee, T. Blackout prediction in interconnected electric energy systems considering generation
re-dispatch and energy curtailment. Appl. Energy 2017, 187, 50–61. [CrossRef]
67. Moutis, P.; Skarvelis-Kazakos, S.; Brucoli, M. Decision tree aided planning and energy balancing of planned
community microgrids. Appl. Energy 2016, 161, 197–205. [CrossRef]
68. Ottesen, S.Ø.; Tomasgard, A.; Fleten, S.-E. Prosumer bidding and scheduling in electricity markets. Energy
2016, 94, 828–843. [CrossRef]
69. Chemali, E.; Kollmeyer, P.J.; Preindl, M.; Emadi, A. State-of-charge estimation of Li-ion batteries using deep
neural networks: A machine learning approach. J. Power Sources 2018, 400, 242–255. [CrossRef]
70. Coelho, I.M.; Coelho, V.N.; Luz, E.J.d.S.; Ochi, L.S.; Guimarães, F.G.; Rios, E. A GPU deep learning
metaheuristic based model for time series forecasting. Appl. Energy 2017, 201, 412–418. [CrossRef]
71. Kim, J.; Le, T.-T.-H.; Kim, H. Nonintrusive load monitoring based on advanced deep learning and novel
signature. Comput. Intell. Neurosci. 2017, 2017. [CrossRef] [PubMed]
72. Mocanu, E.; Nguyen, P.H.; Gibescu, M.; Kling, W.L. Deep learning for estimating building energy
consumption. Sustain. Energy Grids Netw. 2016, 6, 91–99. [CrossRef]
73. Wang, H.; Yi, H.; Peng, J.; Wang, G.; Liu, Y.; Jiang, H.; Liu, W. Deterministic and probabilistic forecasting of
photovoltaic power based on deep convolutional neural network. Energy Convers. Manag. 2017, 153, 409–422.
[CrossRef]
74. Liu, Y.; Yao, X. Ensemble learning via negative correlation. Neural Netw. 1999, 12, 1399–1404. [CrossRef]
75. Burger, E.M.; Moura, S.J. Gated ensemble learning method for demand-side electricity load forecasting.
Energy Build. 2015, 109, 23–34. [CrossRef]
76. Fu, G. Deep belief network based ensemble approach for cooling load forecasting of air-conditioning system.
Energy 2018, 148, 269–282. [CrossRef]
77. Gjoreski, H.; Kaluža, B.; Gams, M.; Milić, R.; Luštrek, M. Context-based ensemble method for human energy
expenditure estimation. Appl. Soft Comput. 2015, 37, 960–970. [CrossRef]
78. Hasan, A.N.; Twala, B. Mine’s Pump Station Energy consumption and Underground Water Dam Levels
Monitoring System Using Machine Learning Classifiers and Mutual Information Ensemble Technique. Int. J.
Innov. Comput. Inf. CONTROL 2016, 12, 1777–1789.
79. Fotovatikhah, F.; Herrera, M.; Shamshirband, S.; Chau, K.-w.; Faizollahzadeh Ardabili, S.; Piran, M.J. Survey
of computational intelligence as basis to big flood management: Challenges, research directions and future
work. Eng. Appl. Comput. Fluid Mech. 2018, 12, 411–437. [CrossRef]
80. Deng, B.; Peng, D.; Zhang, H.; Qian, Y. An intelligent hybrid short-term load forecasting model optimized by
switching delayed PSO of micro-grids. J. Renew. Sustain. Energy 2018, 10, 024901. [CrossRef]
81. Dou, C.-X.; An, X.-G.; Yue, D. Multi-agent system based energy management strategies for microgrid by
using renewable energy source and load forecasting. Electr. Power Compon. Syst. 2016, 44, 2059–2072.
[CrossRef]
Energies 2019, 12, 1301 41 of 42
82. Peng, L.-L.; Fan, G.-F.; Huang, M.-L.; Hong, W.-C. Hybridizing DEMD and quantum PSO with SVR in
electric load forecasting. Energies 2016, 9, 221. [CrossRef]
83. Qu, Z.; Zhang, K.; Wang, J.; Zhang, W.; Leng, W. A hybrid model based on ensemble empirical mode
decomposition and fruit fly optimization algorithm for wind speed forecasting. Adv. Meteorol. 2016, 2016.
[CrossRef]
84. Yang, Z.; Ce, L.; Lian, L. Electricity price forecasting by a hybrid model, combining wavelet transform,
ARMA and kernel-based extreme learning machine methods. Appl. Energy 2017, 190, 291–305. [CrossRef]
85. Dehghani, M.; Riahi-Madvar, H.; Hooshyaripor, F.; Mosavi, A.; Shamshirband, S.; Zavadskas, E.K.;
Chau, K.-W. Prediction of Hydropower Generation Using Grey Wolf Optimization Adaptive Neuro-Fuzzy
Inference System. Energies 2019, 12, 289. [CrossRef]
86. Mosavi, A.; Rituraj, R.; Varkonyi-Koczy, A.R. Review on the Usage of the Multiobjective Optimization
Package of modeFrontier in the Energy Sector. In Proceedings of the International Conference on Global
Research and Education, Ias, i, Romania, 25–28 September 2017; pp. 217–224.
87. Moeini, I.; Ahmadpour, M.; Mosavi, A.; Alharbi, N.; Gorji, N.E. Modeling the time-dependent characteristics
of perovskite solar cells. Sol. Energy 2018, 170, 969–973. [CrossRef]
88. Mosavi, A.; Torabi, M.; Ozturk, P.; Varkonyi-Koczy, A.; Istvan, V. A hybrid machine learning approach for
daily prediction of solar radiation. In Proceedings of the International Conference on Global Research and
Education, Kaunas, Lithuania, 24–27 September 2018.
89. Ijadi Maghsoodi, A.; Ijadi Maghsoodi, A.; Mosavi, A.; Rabczuk, T.; Zavadskas, E. Renewable Energy
Technology Selection Problem Using Integrated H-SWARA-MULTIMOORA Approach. Sustainability 2018,
10, 4481. [CrossRef]
90. David, M.; Ramahatana, F.; Trombe, P.-J.; Lauret, P. Probabilistic forecasting of the solar irradiance with
recursive ARMA and GARCH models. Sol. Energy 2016, 133, 55–72. [CrossRef]
91. Feng, Y.; Cui, N.; Zhang, Q.; Zhao, L.; Gong, D. Comparison of artificial intelligence and empirical models for
estimation of daily diffuse solar radiation in North China Plain. Int. J. Hydrogen Energy 2017, 42, 14418–14428.
[CrossRef]
92. Hassan, M.A.; Khalil, A.; Kaseb, S.; Kassem, M. Exploring the potential of tree-based ensemble methods in
solar radiation modeling. Appl. Energy 2017, 203, 897–916. [CrossRef]
93. Salcedo-Sanz, S.; Deo, R.C.; Cornejo-Bueno, L.; Camacho-Gómez, C.; Ghimire, S. An efficient neuro-evolutionary
hybrid modelling mechanism for the estimation of daily global solar radiation in the Sunshine State of
Australia. Appl. Energy 2018, 209, 79–94. [CrossRef]
94. Salcedo-Sanz, S.; Jiménez-Fernández, S.; Aybar-Ruiz, A.; Casanova-Mateo, C.; Sanz-Justo, J.; García-Herrera, R.
A CRO-species optimization scheme for robust global solar radiation statistical downscaling. Renew. Energy
2017, 111, 63–76. [CrossRef]
95. Touati, F.; Chowdhury, N.A.; Benhmed, K.; Gonzales, A.J.S.P.; Al-Hitmi, M.A.; Benammar, M.; Gastli, A.;
Ben-Brahim, L. Long-term performance analysis and power prediction of PV technology in the State of Qatar.
Renew. Energy 2017, 113, 952–965. [CrossRef]
96. Voyant, C.; Motte, F.; Fouilloy, A.; Notton, G.; Paoli, C.; Nivet, M.-L. Forecasting method for global radiation
time series without training phase: Comparison with other well-known prediction methodologies. Energy
2017, 120, 199–208. [CrossRef]
97. Voyant, C.; Notton, G.; Darras, C.; Fouilloy, A.; Motte, F. Uncertainties in global radiation time series
forecasting using machine learning: The multilayer perceptron case. Energy 2017, 125, 248–257. [CrossRef]
98. Cornejo-Bueno, L.; Cuadra, L.; Jiménez-Fernández, S.; Acevedo-Rodríguez, J.; Prieto, L.; Salcedo-Sanz, S.
Wind Power Ramp Events Prediction with Hybrid Machine Learning Regression Techniques and Reanalysis
Data. Energies 2017, 10, 1784. [CrossRef]
99. Khosravi, A.; Machado, L.; Nunes, R. Time-series prediction of wind speed using machine learning
algorithms: A case study Osorio wind farm, Brazil. Appl. Energy 2018, 224, 550–566. [CrossRef]
100. Burlando, M.; Meissner, C. Evaluation of Two ANN Approaches for the Wind Power Forecast in a
Mountainous Site. Int. J. Renew. Energy Res. (IJRER) 2017, 7, 1629–1638.
101. Pandit, R.; Infield, D. Gaussian process operational curves for wind turbine condition monitoring. Energies
2018, 11, 1631. [CrossRef]
102. Sharifian, A.; Ghadi, M.J.; Ghavidel, S.; Li, L.; Zhang, J. A new method based on Type-2 fuzzy neural network
for accurate wind power forecasting under uncertain data. Renew. Energy 2018, 120, 220–230. [CrossRef]
Energies 2019, 12, 1301 42 of 42
103. Baranyai, M.; Mosavi, A.; Vajda, I.; Varkonyi-Koczy, A.R. Optimal Design of Electrical Machines: State of the
Art Survey. In Proceedings of the International Conference on Global Research and Education, Ias, i, Romania,
25–28 September 2017; pp. 209–216.
104. Albert, A.; Maasoumy, M. Predictive segmentation of energy consumers. Appl. Energy 2016, 177, 435–448.
[CrossRef]
105. Alobaidi, M.H.; Chebana, F.; Meguid, M.A. Robust ensemble learning framework for day-ahead forecasting
of household based energy consumption. Appl. Energy 2018, 212, 997–1012. [CrossRef]
106. Benedetti, M.; Cesarotti, V.; Introna, V.; Serranti, J. Energy consumption control automation using Artificial
Neural Networks and adaptive algorithms: Proposal of a new methodology and case study. Appl. Energy
2016, 165, 60–71. [CrossRef]
107. Chen, K.; Jiang, J.; Zheng, F.; Chen, K. A novel data-driven approach for residential electricity consumption
prediction based on ensemble learning. Energy 2018, 150, 49–60. [CrossRef]
108. Kuroha, R.; Fujimoto, Y.; Hirohashi, W.; Amano, Y.; Tanabe, S.-I.; Hayashi, Y. Operation planning method
for home air-conditioners considering characteristics of installation environment. Energy Build. 2018, 177,
109. Wang, N.; Vlachokostas, A.; Borkum, M.; Bergmann, H.; Zaleski, S. Unique Building Identifier: A natural
key for building data matching and its energy applications. Energy Build. 2019, 184, 230–241. [CrossRef]
110. Depecker, P.; Menezo, C.; Virgone, J.; Lepers, S. Design of buildings shape and energetic consumption.
Build. Environ. 2001, 36, 627–635. [CrossRef]
111. Qi, F.; Wang, Y. A new calculation method for shape coefficient of residential building using Google Earth.
Energy Build. 2014, 76, 72–80. [CrossRef]
112. Livingston, O.V.; Pulsipher, T.C.; Anderson, D.M.; Vlachokostas, A.; Wang, N. An analysis of utility meter
data aggregation and tenant privacy to support energy use disclosure in commercial buildings. Energy 2018,
159, 302–309. [CrossRef]
113. Sweeney, L. k-anonymity: A model for protecting privacy. Int. J. Uncertain. Fuzziness Knowl.-Based Syst. 2002,
10, 557–570. [CrossRef]
114. Perez, A.J.; Zeadally, S. PEAR: A privacy-enabled architecture for crowdsensing. In Proceedings of the
International Conference on Research in Adaptive and Convergent Systems, Krakow, Poland, 20–23
September 2017; pp. 166–171.
© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

State of The Art of Machine PDF

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

State of The Art of Machine PDF

Hochgeladen von

Copyright:

Verfügbare Formate

energies

Energies 2019, 12, 1301; doi:10.3390/en12071301 www.mdpi.com/journal/energies

Energies 2019, 12, x FOR PEER REVIEW 3 of 43

Figure 3. Subject areas of ML in energy systems.

∑  models [27,28]; Equations (1) and (2)

Table 2. Notable MLP models used in energy systems.

Year Reference Journal Application

Table 2. Notable MLP models used in energy systems.

Energies 2019, 12, 1301 8 of 42

ML Model The Best Structure r f

Parameter Model RMSE Parameter Model RMSE Parameter Model RMSE

Energies 2019, 12, x FOR PEER REVIEW

Figure 13. RMSE

Figure 15. RMSE

Energies 2019, 12, x FOR PEER REVIEW 17 of 43

3.7. Decision Trees

Table 9. Notable decision trees models used in energy systems.

Year Authors Journal Application

3.8. Deep Learning

Table 10. Notable deep learning models used in energy systems.

Year Authors Journal Application

3.9. Ensemble Methods

3.9. Ensemble Methods

Table 11. Notable ensemble models used in energy systems.

Year Authors Journal Application

3.10. Hybrid ML Models

Figure 23. RMSE

Figure 24. RMSE

Table 13. Notable hybrid ML models in solar energy systems.

Year Reference Journal ML Model Application

Station ML Model RMSE r

Energies 2019, 12, x FOR PEER REVIEW 28 of 43

Energies 2019, 12, x FOR PEER REVIEW

Table 16. Notable hybrid ML models in wind energy systems.

Year Reference Journal ML model Application

Figure 31. RMSE

Energies 2019, 12, x FOR PEER REVIEW 32 of 43

Figure 34. RMSE

SVM 2016and regression

Energies 2019, 12, x FOR PEER REVIEW 34 of 43

3.11. Comparative Analysis of ML Models

Table 18. Comparative study of ML models in energy systems.

ML Model Complexity User-Friendliness Accuracy Speed Dataset Type

Das könnte Ihnen auch gefallen