Sie sind auf Seite 1von 5

2017 the 5th IEEE International Conference on Smart Energy Grid Engineering

Fault Diagnosis of Micro Energy Grids Using Bayesian Belief Network and Adaptive
Neuro-Fuzzy Interference System

Yahya Koraz1, Hossam A. Gabbar1,2*


1
Faculty of Engineering and Applied Science, University of Ontario Institute of Technology, 2000 Simcoe Street North,
Oshawa L1H7K4 ON, Canada
2
Faculty of Energy Systems and Nuclear Science, University of Ontario Institute of Technology, 2000 Simcoe Street North,
Oshawa L1H7K4 ON, Canada
e-mail: Hossam.Gaber@uoit.ca

Abstract—Safety assessment of complex systems such as micro Bayesian belief network (BBN) layer and
energy grids has lately become an interesting open research adaptive-network-based fuzzy inference system (ANFIS)
field. In this article, fault diagnosis for a micro energy grid in layer. In addition, it considers the following data sets which
the occurrence of incomplete data and expert knowledge is essential for safety analysis:
discussed. A hybrid technique of Bayesian belief networks and x Deterministic data-set of credible information such as
adaptive-network-based fuzzy inference system is proposed for
system topology, operation parameters, units
fault diagnosis and safety assessment of micro energy grid
under uncertainty conditions and incomplete system’s specification, etc.;
information. Merging adaptive-network-based fuzzy inference x Statistical data-historical observation of the system
system with Bayesian belief networks contributes to a reduction operation life cycle; and
of the information required for micro energy grid fault x Linguistic data–defines the system behavior by expert’s
diagnosis when compared with each method separately. Where knowledge contribution.
each method has different capability on capturing safety related The main challenges associated with MEG safety
information. The proposed hybrid approach helps operation assessment are dealing with randomness, vagueness and
crew to make the optimum decision. The approach depends on uncertainties. Classical methods based on the stochastic
expert’s knowledge more than the data from instrumentation theory, such as fault tree analysis (FTA) and hazard matrix
and control system. The demonstrative example of a micro
method, have been employed in Ref. [5] to deal with
energy grids safety assessment is validated in this study.
hazardous events in MEGs.
Keywords-micro energy grid (MEG); risk analysis; Many fuzzy models were presented to deal with
adaptive-network-based fuzzy inference system (ANFIS); vagueness [6] and many reasoning approaches were
bayesian belief network (BBN) illustrated to deal with uncertainties [7]. However,
integration of different safety assessment methods for
I. INTRODUCTION complex systems is still in the early stages.
Energy and environment safety concerns are becoming Bayesian belief network (BBN) is extensively used in
the most significant worldwide issues as the prompt safety assessment for systems with uncertainty and
depletion of fossil fuels and global warming reaches incomplete knowledge. Therefore, BBN is the base of
unprecedented levels. Micro energy grids (MEGs) were different types of expert diagnosis systems in numerous
presented as advanced energy systems by increasing energy fields such as nuclear power systems operation monitoring
efficiency and reducing the greenhouse gas emissions [1]. [8], oil and gas pipelines safety assessment [9], wind turbine
The rapid development of MEGs technologies allows fault diagnosis [10], risk assessment of complex systems
integration of more renewable energy sources (RES). [11].
Utilizing more RES in the energy system leads to further This study introduces a hybrid-safety assessment
clean, efficient and reliable energy systems [2]. On the other approach of MEG using a combination of ANFIS and BBN
hand, it leads to more complex control and management of techniques.
energy systems. Therefore, safety assessment of MEGs is II. METHODOLOGY
one of the topics currently are under research. MEGs’ safely
operation is paramount for modern societies (i.e. residential, The proposed approach consists of two cascaded layers
commercial and industries) development and to the welfare i.e. BBN layer and ANFIS layer. Where the output of the
of its customers. The great northeast blackout of 1965 [3] BBN layer is the input to the ANFIS layer. Therefore, safety
exposed that reliability and safety level of the energy system assessment process runs in consequence from top to bottom
has a direct impact on the public interest, safety and heavy as given in Fig. 1.
financial liabilities [4]. Deterministic data of MEG has sufficient information to
The proposed safety assessment approach for MEGs create BBN qualitative structure for MEG diagnosis
splits analysis process into two main disciplines, i.e. approach. The linguistic data is mainly used to build the

978-1-5386-1776-2/17/$31.00 ©2017 IEEE 


ANFIS structure. The quantitative term of each node in the The inputs to the BBN layer are MEG’s condition
BBN and ANFIS structures can be illustrated from static data measurements, which is extracted from the deterministic data
analysis, which is the conditional probability tables (CPT) of the MEG. Fig. 2 illustrates the BBN structure of a MEG.
and the membership function (MF) respectively. The BBN consists of five parameters of condition
measurements that form the observation level and the
MEG’s Condition Measurment
parameters are: demand not served (DNS), high-temperature
instrument (I&C) alarm, trip alarm, pollution level and fire alarm. The BBN
structure also consists of the output parameters which form
the causes level and include: over load, lack of DER,
Observation intermittency of RES, integration of multi DERs, faults in the
transmission line, faults in distribution line, faults in
transformer and utility grid failure.
The main role of ANFIS layer is to process the BBN
output values to provide an accurate decision of which
BBN
Causes parameter(s) is (are) causing the fault event.
A. Bayesian Belief Networks
BBN is an effective tool for modeling systems that have
incomplete information and/or uncertainty of input data. The
ANFIS uncertainty may be caused due to different reasons. For
instance, the experts’ knowledge may have inherent
Decision uncertainty factors on the situation being modeled. It may
also be the uncertainty of accurate and available information.
Diagnosis Result BBN offers a probabilistic inference network for
representing an intuitive graphical model of numerous causes
Figure 1. Hybrid MEG safety assessment approach. and consequences.

Figure 2. BBN structure of MEG.

Fig. 2 illustrates the BBN structure of MEG, which conditional probability of x given y and the conditional
consists of three layers (causes, consequences and probability of y given x respectively.
observation). Each layer comprises several variables named The prior information about the relationships among
nodes. Each node has a set of probable values that form the nodes located in different layers are stored in the CPT, which
conditional probability table (CPT). Arrows between some can be utilized for a diagnostic decision-making.
nodes in successive layers demonstrate causality flow of the
B. Adaptive Neuro-Fuzzy Interference System
influences. The conditional probability can be defined by
using Bayesian theorem presented in [12]. The Bayesian The modern expert systems are utilizing fuzzy logic
theorem states the conditional probability of x given y in the theory for reasoning the input data instead of Boolean logic
following equation: [13]. The fuzzy expert system converts a set of user-supplied
human language rules to their mathematical equivalents.
( ) ( | ) ANFIS is an integration of neural network (NN) and
( | )= (1)
( ) fuzzy logic (FL) [14]. Fuzzy logic has the capability to
convert human knowledge and insights into a quantitative
where x and y are events while P(x)and P(y) are the process and rules. Nevertheless, there is no defined rule
probabilities of x and y occurrences, P(x|y) and P(y|x) are the governing the converting process of human knowledge to


rule-based fuzzy inference system (FIS), further to a long 500 MWh thermal energy storage (TES) tank is used to store
process time to refine the shapes and ranges of the the surplus cooling and heating energy.
membership functions (MFs). The NN has a greater In accordance with the literature, numerous faults may
capability in the learning process. Thus, the NN was used to cause an energy blackout. The most common fault events and
refine the MFs automatically [15]. their observation parameters were presented in ref. [5]. The
BBN structure in Fig. 2 is proposed to reasoning links
III. APPLICATION OF HYBRID BBN-ANFIS METHOD FOR between fault observation and fault causes layers (diagnosis
MEG SAFETY ASSESSMENT symptoms) as shown in Table I.
In this research, a MEG structure that illustrated in ref. [1] Fig. 3 shows an ANFIS structure of the MEG safety
is selected as a case study. The MEG comprises 16 MW assessment decision-making stage. The ANFIS architecture
consists of five main layers, each layer consists of a number
co-generator, 2MW photovoltaic power (PV) and 2 MW
of nodes distributed as follows: 8-112-14-141. The first and
wind turbine (WT) for the electricity discipline. A set of six
fourth layers consist of adaptive nodes while fixed nodes are
identical electrical chillers of 2100 KWe size capacity and used among the other layers.
6.0 performance are used to satisfy the cooling demand. A

Figure 3. The structure of adaptive neuro-fuzzy interference system.

The eight input nodes that form the first layer of ANFIS
are the diagnosis symptoms of a MEG illustrated in Table I Gaussian(x, c, σ) = e (2)
and each node in this layer has three Gaussian membership
functions where c is the center of the membership function and σ its
width.
TABLE I. CONDITIONAL PROBABILITY OF FAULT DIAGNOSIS OF ONE AND TWO COMBINED FAULTS OBSERVATION OF A MEG

Diagnosis result
Intermitt Integratio Fault in Fault in Fault in Utility
# Fault Observation Nodes Over Lack of
ency of n of multi Transmis distributi transform grid
Load DER
RES DERs sion line on line er failure
1 2 3 4 5 6 7 8
15 Demand not served (DNS) 0.2267 0.2309 0.1464 0.1595 0.1122 0.2006 0.2605 0.1027

16 High Temp. Alarm 0.2369 0.2 0.173 0.2922 0.4638 0.1378 0.2 0.1

17 Trip Alarm 0.2113 0.2 0.1559 0.1945 0.2549 0.2288 0.3116 0.1

18 Pollution Alarm 0.1956 0.2 0.1317 0.1317 0.1804 0.216 0.2851 0.1

19 Fire 0.2163 0.2 0.1493 0.1 0.1991 0.3044 0.3826 0.2183


Demand not served (DNS)-High Temp.
15-16 0.2556 0.2152 0.1712 0.4504 0.3103 0.1518 0.1953 0.0952
Alarm
15-17 Demand not served (DNS)-Trip Alarm 0.2116 0.1725 0.0832 0.2444 0.1361 0.2593 0.342 0.0951


Demand not served (DNS)-Pollution
15-18 0.1982 0.1948 0.1112 0.1726 0.1111 0.2597 0.3398 0.098
Alarm
15-19 Demand not served (DNS)-Fire 0.209 0.187 0.1059 0.1116 0.1132 0.3336 0.4166 0.1781

16-17 High Temp. Alarm-Trip Alarm 0.209 0.2 0.1689 0.3596 0.5037 0.1125 0.1825 0.1

16-18 High Temp. Alarm-Pollution Alarm 0.2123 0.2 0.1877 0.3288 0.444 0.0938 0.1579 0.1

16-19 High Temp. Alarm-Fire 0.2883 0.2 0.1923 0.2528 0.5295 0.2498 0.3188 0.2005

17-18 Trip Alarm-Pollution Alarm 0.1929 0.2 0.13 0.1773 0.2455 0.2505 0.3406 0.1

17-19 Trip Alarm-Fire 0.2087 0.2 0.1459 0.0905 0.1214 0.3526 0.4543 0.1602

18-19 Pollution Alarm-Fire 0.1961 0.2 0.1334 0.0621 0.1029 0.3592 0.4499 0.1675

Unlikely, the output layer consists of one node that The scheme of fuzzy reasoning mechanism for ANFIS is
provides a decision of which input node(s) is (are) causing illustrated in Fig. 4. Fourteen if-then rules are governing the
the hazardous event. process where the columns represent the eight inputs and one
output data.

Figure 4. Fuzzy reasoning ANFIS scheme of MEG.

Fifteen cases with complete symptoms, see Table I, were perform fault diagnosis with complete or incomplete
utilized to evaluate the hybrid BBN-ANFIS performance for symptoms. The main strength of the proposed approach is
fault diagnosis of a MEG. The first case in the table is due to its dependency on experts’ knowledge than the data
presented in Fig. 4. The summation of symptom nodes states from measurement instrumentation (I&C) in its
is assumed to be a unity and they are medium, medium, decision-making process. The results show a robust
medium, medium, medium, medium, medium and medium, performance of the hybrid technique proposed of fault
respectively. Based on this evidence, the posterior diagnosis of a MEG that would potentially provide a solution
probability of each fault can be computed to obtain the most to the reasoning problem of complex systems. This may be an
candidate fault among all is “Fault in transformer”, with a interest of the authors for future works.
probability of 98.31%.
REFERENCES
IV. CONCLUSIONS AND FUTURE WORK
[1] Y. Koraz and H. A. H. A. Gabbar, “Hierarchical Safety Control for
Failure in MEGs has negative influences on the public Micro Energy Grids using adaptive neuro-fuzzy decision making
interest and safety. As well as, it has critical consequences on method,” in IEEE International conference on Smart Energy Grid
Engineering, 2016, pp. 131–136.
the financial liabilities. In this paper, a study of safety
[2] H. A. Gabbar and Y. Koraz, “Safety Design of Resilient Micro Energy
assessment of MEGs under uncertainty conditions and Grids,” in Smart Energy Grid Engineering, 1st ed., Elsevier/Academic
incomplete system’s information was proposed. A hybrid Press, 2017, pp. 101–150.
technique, using BBN and ANFIS based technologies, [3] C. Sulzberger, “History - When the lights went out remembering 9
contributes an efficient tool for MEGs fault diagnosis. Where November 1965,” IEEE Power Energy Mag., vol. 4, no. 5, pp. 90–95,
the results demonstrate that the hybrid BBN-ANFIS can Sep. 2006


[4] Z. Bo, O. Shaojie, Z. Jianhua, S. Hui, W. Geng, and Z. Ming, “An [9] G. Kabir, R. Sadiq, and S. Tesfamariam, “A fuzzy Bayesian belief
analysis of previous blackouts in the world: Lessons for China‫׳‬s power network for safety assessment of oil and gas pipelines,” Struct.
industry,” Renew. Sustain. Energy Rev., vol. 42, pp. 1151–1163, Feb. Infrastruct. Eng., vol. 12, no. 8, pp. 874–889, 2016.
2015. [10] G. Li and J. Shi, “Applications of Bayesian methods in wind energy
[5] Y. Koraz, H. A. Gabbar, and A. Gabbar, “Risk Analysis and Self-Healing conversion systems,” Renew. Energy, vol. 43, pp. 1–8, Jul. 2012.
Approach for Resilient Interconnect Micro Energy Grids,” Sustain. [11] P. Trucco, E. Cagno, F. Ruggeri, and O. Grande, “A Bayesian Belief
Cities Soc., vol. 32, no. February, pp. 638–653, 2017. Network modelling of organisational factors in risk analysis: A case
[6] J. Wang and O. Kieran, “Off shore Safety Assessment and Safety-Based study in maritime transportation,” Reliab. Eng. Syst. Saf., vol. 93, no. 6,
Decision-Making-The Current Status and Future Aspects,” J. Offshore pp. 845–856, Jun. 2008.
Mech. Arct. Eng., vol. 122, no. 2, p. 93, 2000. [12] M. Vileiniskis, R. Remenyte-Prescott, D. Rama, and J. Andrews, “Fault
[7] L. Yang and J. Lee, “Bayesian Belief Network-based approach for detection and diagnostics of a three-phase separator,” J. Loss Prev.
diagnostics and prognostics of semiconductor manufacturing systems,” Process Ind., vol. 41, pp. 215–230, 2016.
Robot. Comput. Integr. Manuf., vol. 28, no. 1, pp. 66–74, Feb. 2012. [13] D. McNeill and P. Freiberger, Fuzzy logic. Simon & Schuster, 1994.
[8] C. Kang, “A Bayesian belief network-based advisory system for [14] J.-S. R. Jang, “ANFIS: adaptive-network-based fuzzy inference system,”
operational availability focused diagnosis of complex nuclear power IEEE Trans. Syst. Man. Cybern., vol. 23, no. 3, pp. 665–685, 1993.
systems,” Expert Syst. Appl., vol. 17, no. 1, pp. 21–32, Jul. 1999.
[15] W. Z. Chmielowski, Fuzzy Control in Environmental Engineering, vol.
31. Cham: Springer International Publishing, 2016.



Das könnte Ihnen auch gefallen