Energy Conversion and Management 164 (2018) 518–532

Contents lists available at ScienceDirect

Energy Conversion and Management

journal homepage: www.elsevier.com/locate/enconman

Multi-objective sizing optimisation of a solar-thermal system integrated with T

a solar-hydrogen combined heat and power system, using genetic algorithm
⁎
Jihane Assaf , Bahman Shabani1
School of Engineering, RMIT University, Melbourne, Australia

A R T I C LE I N FO A B S T R A C T

Keywords: A sizing multi-objective optimisation using the genetic algorithm is performed on a solar-hydrogen combined
Integrated renewable system heat and power system integrated with solar-thermal collectors (SH CHP-ST) to supply both power and heat (i.e.
Multi-objective optimisation hot water demand) to an application. A solar-hydrogen system is a renewable system with hydrogen-based
Solar-hydrogen combined heat and power storage consisting of an electrolyser, a hydrogen tank, and a fuel cell. The fuel cell generates heat while pro-
system
ducing power that can be recovered. The heat collected from the fuel cell can be integrated with the heat supply
Solar-thermal system
of a renewable solar-thermal system consisting of an evacuated tube collector and a hot water storage tank. A
simulation module to model the operation of the whole system is implemented in MATLAB. Energy demands and
meteorological data for a remote household located in southeast Australia are considered. The sizes of the main
components of the system are optimised with the objectives of maximising the overall reliability of the system,
minimising the levelised cost of energy, and minimising the percentage of excess energy from the PV that is not
utilised. The results show that the electric reliability of the optimal solutions in favour is always equal to 100%.
The maximum thermal reliability that could be obtained is around 96%. A trade-off between the cost of energy
and percentage of wasted power from PV is found.

1. Introduction
Renewable systems with hydrogen-based storage have received in-
creasing attention in recent years [1–3]. One of the promising areas of
applications (i.e. as energy storage) is to support intermittent photo-
voltaic (PV)-based standalone power supply systems in standalone ap-
plications: i.e. known as Solar-Hydrogen (SH) system. In a SH system, as
considered in this paper and illustrated in the top section of Fig. 1, the
PV panels that are equipped with a maximum power point tracking
device, supply renewable electricity to the load. The excess of the
electricity generated by the PVs, energises an electrolyser unit that
produces hydrogen [4]. The hydrogen produced by the electrolyser is
stored in the hydrogen storage tank (TKH2) for a later use by the fuel
cell (FC). For both electrolyser and the FC, Proton Exchange Membrane
(PEM) technology is considered. The FC operates when the photovoltaic
panels fall short to meet the demand alone due to low or no solar ra-
diation. In this case, the FC kicks in and draws on hydrogen from TKH2
to produce electricity [5,6]. The key advantage of this arrangement is
its capability to be used as long-term energy storage as opposed to more
conventional technologies such as battery systems [7,8]. This makes it
suitable for areas with considerable seasonal variations in solar
radiation [9] or a good option for large-scale (e.g. utility scale) energy
storage applications [10].
The fuel cell electrical energy efficiency is in the range of
∼30–55%, depending of the current withdrawn from the FC [11].
Hence a substantial amount of energy entering the cell (i.e. the che-
mical energy content of hydrogen) converts to heat (that is about
50–70% of the total energy input) [12]. Cooling the fuel cell (i.e. to a
temperature between 60 °C and 80 °C) by removing this heat is an es-
sential operating requirement for the PEMFC [13,14]. An earlier study,
conducted by Shabani, Andrews [15] for a conservative standalone
household, indicated, that recovering this heat and making a solar-
hydrogen combined heat and power (SH CHP) system, could offer up to
over 30% of saving in the energy normally required for hot water
supply, that is usually supplied through a transported fuel such as LPG.
On the other hand, for supplying thermal energy, a renewable al-
ternative to fossil-fuel-based technologies for hot water supply is using
Solar-Thermal (ST) systems. An ST system consists mainly of solar-
thermal collectors (i.e. flat plat or evacuated tube collectors) and a hot
water storage tank. However, this system, when operating by itself, was
found to be able to supply in its best design only about 60–70% of the
total annual hot water demand of a standalone household in Australia

⁎
Corresponding author at: School of Engineering, RMIT University, 115 Queensberry Street, Carlton 3053, Vic., Australia.
E-mail address: s3467254@student.rmit.edu.au (J. Assaf).
1
School of Engineering, RMIT University, Bundoora East Campus, Bundoora 3083, Vic., Australia.

https://doi.org/10.1016/j.enconman.2018.03.026
Received 16 October 2017; Received in revised form 3 February 2018; Accepted 10 March 2018
0196-8904/ © 2018 Elsevier Ltd. All rights reserved.
J. Assaf, B. Shabani
Fig. 1. Schematic diagram of the integrated SH CHP-ST system.
[16], with only around 54% of the hot water demand met in winter
[17].
When integrated, ST and SH CHP systems can effectively comple-
ment each other in terms of heat supply to the kind of standalone ap-
plications focused by this paper (e.g. a household of a few people lo-
cated in a remote area) [17]. It is because during the absence of
sufficient solar radiation, when the electrical output of the PV system is
insufficient, the thermal output of the ST system is supposed to be in-
sufficient as well. This is when the fuel cell kicks into cover the PVs’
shortage in supplying power. While the FC is in operation, its heat can
be recovered in the SH CHP system, thus, greatly covering the shortage
of supply from the ST collectors [17].
In an integrated solar-hydrogen combined heat and power/solar-
thermal system (SH CHP-ST) (Fig. 1), the fuel cell heat is recovered
through a heat exchanger (HX) that is used to cool down the FC. This
cooling load (i.e. the heat recovered from the FC) is transferred at best
through the HX to the top part of the hot water tank (TKHW) of the ST
subsystem (Fig. 1). The temperature in the TKHW is limited to 65 °C for
domestic use and safety [18,19]. The remaining of the FC cooling load
that cannot be transferred to the tank is dumped to environment. The
integration between the SH CHP and ST was found to be well effective
in covering the shortfall of the ST system, and increased the annual
capacity of the system for meeting the total thermal demand (i.e. hot
519
Energy Conversion and Management 164 (2018) 518–532
water demand in this case) of a typical standalone household, from
about 60–70% to well around 95%, while the SH CHP is sized to fully
meet the electrical demand of the site [17]. The fuel cell heat (cooling
load) that was transferred to the tank was annually on average 83%,
which showed the benefit of this integration. The wasted heat of the
fuel cell (not transferred to the tank) occurred mainly during periods
(i.e. in relatively warm or hot weather), when the hot water demand
could be met at large part from the solar-thermal collector, and the fuel
cell heat was needed partly [17].
2. The focus of this study
In the previous study conducted by the authors on the feasibility of
the SH CHP-ST system, the system was modelled and simulated theo-
retically in TRNSYS [17]. This was followed by a detailed study on the
economics of this integrated system [20]. The sizing strategy adopted
by the authors in those studies was based on an objective to fulfil 100%
of the power load, and by having close to zero-power wasted from the
PVs. This was done by performing repetitive simulations of the system
with different sizes. For instance, the electrolyser was sized to accom-
modate all the excess power from the PVs. A hydrogen tank of un-
constrained size was considered. The PV array was sized in such a way
that all the hydrogen produced by the electrolyser was consumed by the
J. Assaf, B. Shabani
fuel cell throughout the year (i.e. mass of hydrogen in tank at the end of
the year was to be equal to that at the start of the year). Similar sizing
approach for solar-hydrogen systems, was also adopted by other re-
searchers such as in [9,21,22], for meeting the power demand of dif-
ferent standalone applications. The solar-thermal system was sized in
[17] so that to have maximum annual thermal energy output from it
without having the collectors to overheat in summer. This sizing ap-
proach as just explained, has resulted for the SH CHP-ST case modelled
in [17], in meeting 100% of the power demand as primarily designed
for, and 95% of the annual hot water demand, with a levelised cost of
primary energy obtained to be around 0.4 US$/kWhprimary [17,20].
The present paper considers the same system as that detailed in [17]
and illustrated in Fig. 1. However, instead of designing the system to
meet 100% of the power load and having zero-power wasted from the
PV, a multi-objective sizing optimisation approach has been adopted in
order to obtain the best components’ sizes with the objectives of max-
imising the reliabilities (i.e. electrical and thermal), and minimising the
levelised cost of primary energy as well as minimising the power wasted
by the PVs. The sizing optimisation for such a system is complex and
challenging and no such optimisation was yet undertaken for the kind
of the integrated CHP renewable system (i.e. SH CHP-ST). Hence, the
techno-economic results of this multi-objective optimisation are ex-
pected to expand the knowledge base of other sizing options for that
system. For this purpose, a simulation program and an optimisation
model using the genetic algorithm were developed in MATLAB pro-
gramming environment and implemented on a case study of a remote
conservative household in Australia.
The current section will be followed by Section 3, which provides a
brief overview of previously-conducted optimisation studies on in-
tegrated renewable energy systems with hydrogen storage. In Section 4,
a sizing optimisation model for the integrated SH CHP-ST system is
introduced. Then, Section 5 presents the details of the simulation pro-
gram of the system. A case study of the system is covered in Section 6
with all the optimisation results. The main benefits of this optimisation
are highlighted and the sizing results of the system were compared to
the results obtained in [17] for the similar case. Finally the paper is
concluded in Section 7.
3. Multi-objective optimisation of integrated renewable energy
systems: overview of previous studies
Design, sizing, and operating strategy of the multiple components of
such integrated renewable energy systems are generally challenging,
especially when multiple conflicting objectives are aimed. Optimisation
techniques, based on meta-heuristics population approach, such as the
Genetic Algorithm (GA), Particle Swarm Optimisation (PSO),
Simulation Annealing (SA), and etc., can deal with complex optimisa-
tion problems, especially when having one or a combination of the
following conditions: non-linearity in the components’ characteristics,
objectives, and/or constraints; when dealing with dynamic models that
have a variability of input data with time; and when multiple objectives
are targeted [23–26].
There are many examples in the literature of such optimisations for
design and sizing of integrated renewable energy systems with parti-
cular examples of those that covered renewable-hydrogen systems:
Sharafi and Elmekkawy [27] considered a hybrid power supply system
including a wind turbine, PV panels, a diesel generator, a fuel cell, an
electrolyser, and a hydrogen tank. PSO simulation with ε-constraint
method was used to perform a multi-objective optimisation analysis to
simultaneously minimise the total cost, the unmet load, and the fuel
emission. Kashefi Kaviani et al. [28] also considered a sizing optimi-
sation on a similar hybrid system (i.e. including a wind turbine, PVs,
and a fuel cell power generation unit) designed to supply the load taken
from IEEE Reliability Test System. The aim of that optimisation pro-
blem was to minimise the cost subject to a reliable supply of power to
the demand, by considering components outages as well. Brka et al.
520
Energy Conversion and Management 164 (2018) 518–532
[29] used a multi-objective GA method to optimise the components’
size of three stand-alone hydrogen systems with and without batteries
(wind/H2, PV/H2, and wind/PV/H2) under three objective functions,
including: minimising the net present cost (in $), the whole life cycle
emissions (in CO2-eq/year), and the dumped/excess energy EE (in %) at
low demand. A modified formula for the Loss of Power Supply Prob-
ability (LPSP) index, which considered load losses during fuel cell
transient start-up, was also introduced to assess its effect on system
reliability. Giannakoudis et al. [30] addressed the design and sizing
optimisation under uncertainty of a system consisting of PV panels,
wind generators, accumulators, an electrolyser, hydrogen storage tanks,
a compressor, a fuel cell, and a diesel generator, using the Stochastic
Annealing optimisation algorithm. Singh et al. [31] studied the eco-
nomic viability of using a hybrid renewable energy system to supply
power to an academic research building in India with a capacity
shortage of 0%. The system consisted of PV, electrolyser, hydrogen
storage tank, fuel cell, and battery. A fuzzy logic cost optimisation
aiming at cost minimisation was performed. Gao et al. [32] used GA
method to optimise a hybrid system based on integrating electricity and
hot water supply for a residential application. The system consisted of
PV panels, solar thermal collectors, a hot water storage tank, and a
Solid Oxide Fuel Cell (SOFC) fuelled by municipal gas (i.e. not from
electrolysers powered by PVs). The authors built a dynamic model for
this integrated hybrid system and proposed an operation strategy for it,
with heat recovery form the SOFC. Municipal electricity from the power
plant, and hot water supplied from the sub-boiler, could fill the gap in
case of shortage of energy supply from the main system. The objectives
were to reduce the conventional energy consumption and the life cycle
costs for a typical building located in Fukuoka city in Japan.
In the present study, the authors perform a sizing optimisation for a
renewably-integrated combined heat and power system (Fig. 1). The
system involves the integration of well inter-related renewable sub-
systems with both power and heat output. A simulation module de-
veloped in MATLAB programming, models extensively this integration,
which makes the optimisation results more reliable. The sizing opti-
misation is also rendered reliable by covering all the main components
of the system, resulting in six decision variables. Three objectives re-
presenting different dimensions (overall reliability, economics, and
wasted energy from PV) are aimed. The genetic algorithm that can deal
at best with such complexity is used for the multi-objective optimisation
as will be described next in Section 4.
4. Multi-objective sizing optimisation of an integrated SH CHP-ST
system
The decision variables for the optimisation problem of this paper are
the sizes of the main components of the integrated SH CHP-ST system.
They are defined as a vector, including the following dimensions:
[Pr ,PV ,Pr ,EL,Pr ,FC ,VolTK H2,VolTKHW ,ASC ]
where Pr ,PV is the rated power of the PV array (W); Pr ,EL is the rated
power of the electrolyser (W); Pr ,FC is the rated power of the fuel cell
(W); VolTK H2 is the capacity of the hydrogen tank corresponding to the
maximum volume of gas that is capable to store (N m3); VolTKHW is the
volume of the hot water storage tank (litres); and ASC is the collector’s
absorber area (m2).
The aim of the multi-objective optimisation is to simultaneously
maximise the overall reliability Ro (i.e. or alternatively minimise −Ro as
the optimisation tool in MALTAB supports minimisation), to minimise
the levelised cost of primary energy COEprimary (i.e. electric energy and
thermal energy converted back to primary energy), and to minimise the
average percentage of PV energy that is wasted PercE PV waste (i.e. excess
PV power that is dumped):
Minimise (−Ro,COEprimary,PercE PV waste );
where
J. Assaf, B. Shabani Energy Conversion and Management 164 (2018) 518–532

R o = R e × Rt
in which Ro is the overall reliability; R e is the electrical reliability, that
is the annual (average) percentage of electric demand met renewably
by the integrated system; Rt is the thermal reliability, that is the annual
percentage of thermal demand met renewably by the integrated system.
In fact, as both electrical and thermal energy supply are obtained from
the same system, they are both converted back into primary energy that
is used as a hypothetical reference for the calculation of a single cost
figure for energy (i.e. COEprimary ) [20].
The lower and upper bounds for the different sizes of components
are chosen widely as:
0 ⩽ Pr ,PV ⩽ 8000
0 ⩽ Pr ,EL ⩽ 8000
0 ⩽ Pr ,FC ⩽ 1000
(1) set of optimal solutions, known as the Pareto-set or “the Pareto front”,
which is the set of “non-dominated” solutions, where no improvement
is possible in one objective without losing in any other objective [25].
The selection/decision of the appropriate solution can then be made
according to the case’s or user’s preference.
As for the multiple objectives of this optimisation, Ro of the system
is maximised while simultaneously the COEprimary and PercE PV waste are
minimised. The PercE PV waste is of minimisation interest as this is ex-
pected to lead to the best minimal size of the PVs and consequently of
the space occupied by the PVs for different optimal solutions. This also
leads to the best (i.e. maximum) utilisation of the solar energy as the
wasted energy from the PVs is minimised. It is noteworthy that the
(2) multiple objectives aimed here are conflicting. For instance, the
COEprimary that is aimed to be minimised, is expected to increase when
(3)
Ro is maximised; this is while the direction of PercE PV waste with respect
(4) to the variability of the other objectives is ambiguous.

0 ⩽ VolTK H2 ⩽ 500 (5)

5. Simulation program
0 ⩽ VolTKHW ⩽ 400 (6)
5.1. The overall structure of the simulation program
0 ⩽ ASC ⩽ 4 (7)
The operational and technical behaviour of the integrated system
Additionally, in order to consider a proper design (i.e. to avoid over-
throughout the year as well as its economic performance have been
sizing) of the solar-thermal sub-system, and consequently to avoid the
simulated in MATLAB. For a given size of the system represented by the
collector from overheating, especially in summer [16,33], the time the
vector [Pr ,PV ,Pr ,EL,Pr ,FC ,VolTK H2,VolTKHW ,ASC ], the simulation program can
collector is reaching stagnation temperature annually (nbhrsstagSC ) is
generate techno-economic outputs (Fig. 3).
limited to be very low and constrained at: nbhrsstagSC < a.
The simulation program recurs to the component's’ models, con-
The genetic algorithm (GA) has been used for its ability to work in
siders a modular approach, and can therefore be used with any set of
complex (non-linear) multi-objective optimisation problems [25,34]
parameters and data. The input parameters or data used for the simu-
(i.e. especially with such number of decision variables and with mul-
lation program include meteorological data, case study data (i.e. power
tiple distinct objectives). The GA is inspired by the process of natural
and hot water demand profiles), technical parameters (components’
evolution and works as follows (Fig. 2): The GA provides preliminary
performance characteristics) and economic parameters (i.e. interest
random population of solutions satisfying the constraints. Each solution
rate, inflation rate, capital costs of components, components’ lifetime,
is denoted by the vector [Pr ,PV ,Pr ,EL,Pr ,FC ,VolTK H2,VolTKHW ,ASC ], where its
project lifetime, and etc.). These data, parameters, and input variables
elements represents the decision variables (i.e. sizes of the system’s
(i.e. sizes of components) are used by any of the simulation modules of
components). Each vector is run through the simulation program and
the program. In fact, the simulation program consists of three modules:
produces outputs. The produced outputs are evaluated by calculating
the solar-hydrogen CHP module, the thermal module, and the economic
the objectives of each vector solution. Consequently a fitness value is
module. They are interconnected as shown in Fig. 3 and as will be
assigned for each vector solution, representing how close the solution is
explained in detail in Sections 5.2 and 5.3. In summary, the solar-hy-
to the aims set in the optimisation problem (i.e. minimisations). The
drogen CHP module simulates the performance of the solar-hydrogen
“selection” operator selects the predefined percentage of solutions from
CHP system and calculates the available cooling load of the fuel cell,
the initial population based on the fitness value of each. Hence, the best
Qcool,FC that is fed into the thermal module. Other main outputs of this
solutions meeting the aims of the optimisation are selected. Utilising
module are the electrical reliability (R e ) and the average annual per-
these selected solutions, the “crossover” operator provides new possible
centage of PV electric energy output that is wasted (PercE PV waste ) . The
solutions (i.e. offspring) with the aim of achieving better fitness values
second module, that is the thermal module, simulates the thermal in-
[24]. To avoid premature convergence of the solutions, which occurs
tegration between the solar-thermal system and the fuel cell heat re-
when highly fit parents breed many similar offspring, the mutation
covery system (i.e. recovering Qcool,FC ) . This generates a figure for the
operator reintroduces genetic diversity back into the population, thus
thermal reliability Rt as a main output. The third module, that is the
enabling the search to escape from local optima. This whole process of
economic module, mainly calculates the cost of primary energy
the optimisation is done iteratively until a pre-defined termination
(COEprimary ) for the whole integrated system. For this purpose, operation
criterion is reached (i.e. maximum iterations, tolerance function)
hours Ophrs for each of the hydrogen components are fed into the eco-
(Fig. 2).
nomic module for calculating the lifetime of these components in years
MATLAB optimisation toolbox has been used to implement the
and hence determining their replacement costs. The amount of electric
multi-objective GA. The population size is considered to be 200, the
and thermal energy unmet by this integrated renewable energy system
selection is of tournament type, the crossover probability is considered
(denoted respectively as Ee unmet byREN and Et unmet byREN ) are also fed re-
to be 80%, and the mutation is assumed to be 1% (uniform). These
spectively from each of the solar-hydrogen CHP and thermal modules to
parameters are the default recommended by MATLAB and supported by
the economic module (Fig. 3).
the study in [35]. The simulation model has also been developed in
MATLAB programming environment to keep the computation time,
required to complete the optimisation task, down to about 4–5 h (i.e. 5.2. Solar-hydrogen CHP simulation module
for a case study to be discussed later in the present paper). This is
considered short in comparison to 5–7 days that was encountered in 5.2.1. Components models
similarly complicated integrated systems studied by Shirazi et al. [36], The components of this simulation module mainly comprise the PV,
in which the simulation module has been developed in TRNSYS and electrolyser, fuel cell, and hydrogen storage tank.
linked to the multi-objective optimisation tool in MATLAB. The output power of the PV array (PPV ), is calculated through the
In fact, the multi-objective optimisation has no single solution but a following equations [37]:

521
J. Assaf, B. Shabani Energy Conversion and Management 164 (2018) 518–532

Fig. 2. Flow chart of the genetic algorithm.

Tc,NOCT −Ta,NOCT be approximately ∼70%, (i.e. for a low pressure electrolyser)

ηPV = ηPV ,REF ·ηc ·⎡1−α (Ta + Gt −TPV ,REF )⎤
⎢ Gt ,NOCT ⎥ (8)
⎣ ⎦
η sonable assumption for remote areas due to availability of space in such
Gt
PPV (t ) = ·Pr ,PV · PV
1000 ηPV ,REF
where ηPV is the efficiency of the PV in converting the input solar ra-
diation to electricity; ηPV ,REF and TPV ,REF are respectively the reference
(rated) efficiency and temperature of the PV under reference conditions
(1000 W/m2 of solar irradiance and 25 °C of cell temperature); ηc is the
efficiency of auxiliary components for the PVs, also encompassing other
loss factors (e.g. dirt, cables); α is the temperature coefficient of the PV
module (i.e. assumed equal to −0.5%/C); Ta is the ambient temperature
(°C); Gt is the total incident irradiance (W/m2); Tc,NOCT is the open-cir-
cuit PV module temperature at standard Nominal Operating Cell
Temperature (NOCT) with a wind speed of 1 m/s (i.e. assumed equal to
40 °C); Ta,NOCT is the ambient temperature at standard NOCT (i.e. equal
to 20 °C); Gt ,NOCT is the incident solar irradiance at standard NOCT
(equal to 800 W/m2); and Pr ,PV is the rated power of the PV under re-
ference conditions.
The PEM electrolyser has an efficiency of ηe,EL that is considered to
throughout its operational range as suggested by [17,38]. The hydrogen
is assumed to be stored at a relatively low pressure, which is a rea-
application. Hence, no external compression has been considered in the
(9)
model since hydrogen can be generated directly by the electrolyser at
low pressure (i.e. under 5 bar). The hydrogen mass flow rate (ṁ H2 Prod,EL)
supplied by the electrolyser is calculated in kg/h using the following
equation [17]:
ηe,EL × PEL × 2 × 3.6
ṁ H2 Prod,EL =
HHVH2 (10)
where PEL is the power input to the electrolyser (W); and HHVH2 is the
high heating value of hydrogen taken as 285,840 J/mol.
Butler-Volmer equation has been used to model the fuel cell per-
formance (i.e. polarisation curve and the power output at different
current densities throughout the operational ranges). This is similar to
what is used by TRNSYS software tool to model fuel cell performance
characteristics [39]:
Ucell,FC = E −ηact −ηohmic (11)

522
J. Assaf, B. Shabani
Fig. 3. The structure of the simulation program with demonstration of how its three separate modules known as solar-hydrogen CHP, thermal, and economic simulation modules are
interconnected.
where E is the maximum electromotive force of the fuel cell obtained
from the Nernst equation [39]; ηact is the activation overvoltage; and
ηohmic is the ohmic overvoltage. Mass transport overvoltage has been
incorporated in these overpotential terms and was not added as a se-
parate item.
The inlet hydrogen mass flow rate to the fuel cell (ṁ H2 in,FC ) is cal-
culated as follow:
3.6 × n × IFC
ṁ H2 in,FC =
F × μf
where n is the number of cells connected in series in the fuel cell stack;
IFC is the fuel cell current (A); F is the faraday constant (equal to 96,485
C/mol); and μ f is the hydrogen utilisation coefficient. A conservative
average figure of 0.83 based a hydrogen stoichiometry of nearly 1.2 has
been assumed; however, this can be as high as 95% at some operating
points and even less than 80% at some other operating points [38].
By having the mass flow rate of hydrogen entering the fuel cell and
also the polarisation curve of the fuel cell, key parameters such as fuel
cell efficiency and heat generation can be calculated.
The electrical energy efficiency (ηe,FC ) of the fuel cell is calculated
for a given fuel cell power (PFC ) as:
ηe,FC =
2 × 3.6 × PFC
ṁ H2 in,FC × HHVH2
The heat generated by the fuel cell (Qgen,FC ) at a given operating
point PFC is calculated using the following equation [40]:
μf
Qgen,FC = PFC × ⎜⎛ −1⎞⎟
η
⎝ e,FC ⎠
Some of the heat generated by the fuel cell (Qgen,FC ) is absorbed
internally by the fuel cell to evaporate the water product, and few
precent are taken away by the extra air and hydrogen streams and also
through the heat transfer from the external surface body of the FC (i.e.
through convection) [41]. The remaining heat, referred to as fuel cell
cooling load (Qcool,FC ) , is targeted for recovery through the heat ex-
changer [9,17]. Qcool,FC was found to be in the range of ∼60–70% of the
total heat generated by the fuel cell, in the experimental case studied by
Shabani and Andrews [11]. Hence a mid-range figure of 65% is used in
this study.
Qcool,FC = 0.65Qgen,FC
In the context of a solar-hydrogen system operating throughout the
year, at any given time t (i.e. t is the time corresponding to the number
Energy Conversion and Management 164 (2018) 518–532
of time steps Δt , where Δt is defined in hours and t starts from first of
January), the mass of hydrogen stored in the tank (MH2 Stored,TK ) in kg is:
MH2 Stored,TK (t ) = MH2 Stored,TK (t −1) + ṁ H2 Prod,EL (t )−ṁ H2 in,FC (t ) (16)
If t = 1, then MH2 Stored,TK (t −1) = MH2 initial
where MH2 initial is the initial mass of hydrogen in tank (kg) at time t = 0 .
It is important to note that the electrolyser and the fuel cell never op-
erate at the same time as the system either have power deficit (when the
(12) fuel cell kicks in) or power surplus (when the electrolyser can generate
hydrogen). In other words at any time (t) either the hydrogen pro-
duction or the hydrogen consumption rates is zero.
5.2.2. Simulation model for the solar-hydrogen CHP module
The configuration of the integrated system is illustrated in Fig. 1.
For an input vector of [Pr ,PV ,Pr ,EL,Pr ,FC ,VolTK H2,VolTKHW ,ASC ], the simula-
tion model of the solar-hydrogen CHP module is programmed to run as
follows and as shown in the flowchart illustrated in Fig. 4.
In a For loop, for time t = 1 time step to t = h time steps (h corre-
sponding to the total number of time steps in a year, that is 17,520 for
half hourly time steps), and with a time step increase (Δt ) of 0.5 h:
PPV (t ) is calculated (Eqs. (8), (9)) and compared to the power demand
PDem e (t ) . In the first scenario where the former is greater than the latter,
(13) the power demand is fully met by the PV, the fuel cell does not operate,
and the excess PV power goes to the electrolyser, with the condition
that this excess is less or equal to the rated electrolyser power Pr ,EL .
Otherwise, the electrolyser power reaches its maximum allowable
power input, Pr ,EL , and the rest of this excess PV power (i.e. left after
energising the electrolyser with its Pr ,EL ) is wasted (i.e. dumped). The
(14)
mass flow rate of hydrogen produced by the electrolyser (i.e. filled to
the hydrogen tank) is calculated as per Eq. (10). In the second scenario,
where PPV (t ) is less than PDem e (t ) , the load takes all PV power, the
electrolyser does not operate, and the rest unmet of PDem e (t ) is supplied
by the fuel cell with the condition that this unmet power is less than the
rated power of the FC (Pr ,FC ) . Otherwise, the fuel cell power reaches at
maximum Pr ,FC , and the rest of the unmet power demand remains un-
satisfied (Fig. 4). The inlet mass flow rate of hydrogen to the fuel cell
(i.e. consumed from the hydrogen tank) is calculated as per Eq. (12).
Hence, the total mass of hydrogen stored (MH2 Stored,TK (t )) in the tank in
calculated as per Eq. (16). The initial mass of hydrogen in the tank
(15)
(MH2 initial) is considered and calculated in such a way that the mass of
hydrogen in tank at the end of the year is equal to that at the start of the
year (i.e. MH2 StoredTK (h) = MH2 initial ). In this way, the simulation year is
representative for the lifetime of the project. Consequently, all the
523
J. Assaf, B. Shabani Energy Conversion and Management 164 (2018) 518–532

Fig. 4. Flowchart of the simulation model for the Solar-hydrogen CHP module.

hydrogen produced by the electrolyser is consumed by the fuel cell
during the simulation time (i.e. year), and no hydrogen will be pro-
duced as extra or less during the year. Thus, the resulting average an-
nual cost of the system (taking into account operation time and re-
placement cost of components), and the PercE PV waste , would be more
accurate for the year, and the results for different sizing solutions (i.e.
vectors) would therefore be comparable among each other. Hence, for
the purpose of finding MH2 initial for a given vector, an iterative technique
is followed: an initial mass of hydrogen (MH2 initial) in the tank is firstly
assumed to be equal to zero (Fig. 4). MH2 Stored,TK (t ) is calculated and is
hypothetically accepted to be negative here in order to come up with a
starting guess value of MH2 initial taken equal to MH2min (i.e. that is the

524
J. Assaf, B. Shabani Energy Conversion and Management 164 (2018) 518–532

minimum pre-defined mass of hydrogen allowed to be present in the

1−e⎝ Cmin (
⎛ −UA · 1 − Cmin ⎞
Cmax ⎠)
tank) minus minentry (MH2 Stored,TK ) (i.e. that is the minimum value of ε=
MH2 Stored,TK reached in the tank throughout the year). Then, the vector C
1− Cmin ·e⎝ Cmin (
⎛ −UA · 1 − Cmin ⎞
Cmax ⎠ ) (20)
max
MH2 Stored,TK is repetitively calculated in a few iterations (i.e. each
iteration running throughout the year) by taking MH2 initial of one where Cmin and Cmax correspond to:
iteration equal to MH2 StoredTK (h) of the previous iteration. This is stopped
Ch = ṁ h ·cp,h/3.6 (21)
till MH2 StoredTK (h) gets equal to MH2 initial in the same iteration (Fig. 4).
This iterative task simulates the behaviour of the tank after supposedly Cc = ṁ c ·cp,c /3.6 (22)
few consecutive years and its stability at MH2 Stored,TK (h) = MH2 initial .
While this is running, the simulation program also ensures that the such that Cmin is the minimum value of Ch and Cc ; and Cmax is the
electrolyser stops production if the hydrogen in the tank is to exceed the maximum value of Ch and Cc ; cp,h and cp,c are the same for water to
maximum value (MH2max ) , and the fuel cell stops operating if the hy- water heat exchangers; and UA is the overall heat transfer coefficient of
drogen in the tank is to get below a minimum value (MH2min) (Fig. 4). the heat exchanger (W/°C).
MH2max is based on VolTK H2 that is the capacity of the hydrogen tank As in ε−NTU method, the heat transfer though the heat exchanger
corresponding to the maximum volume of gas capable to store in N m3; (QT ) is therefore calculated as follows (W):
and MH2max is calculated as:
QT = ε ·Cmin (Th,in−Tc,in ) (23)
VolTK H2× 101.3 × 1000
MH2max = where Th,in is the inlet temperature at the hot side of the heat exchanger
4127 × 273 (17) (°C); and Tc,in is the inlet temperature at the cold side of the heat ex-
changer (°C). Then the outlet temperature (Tc,out ) at the cold side of the
and, MH2min is based on the number of days of autonomous operation
heat exchanger is in °C:
that the user would like to consider in case of un-expectancies or
maintenance of the system. QT × 3.6 + ṁ c ·cp·Tc,in
Tc,out =
While the key outputs of the solar-hydrogen CHP module that are ṁ c ·cp (24)
used for system optimisation are the electric reliability (R e ) and the
percentage of PV energy that is wasted PercE PV waste (Fig. 4), other key The outlet temperature Th,out at the hot side of the heat exchanger is
efficiency figures (i.e. electric efficiency of the fuel cell (ηe FC ) , overall in °C:
efficiency of the fuel cell (ηoverall,FC ), and the overall combined heat and QT × 3.6 + ṁ h ·cp·Th,in
power efficiency of the FC CHP-SC (ηoverallFCCHP − SC ) ) are also obtained as Th,out =
ṁ h ·cp (25)
per their equations detailed in [20]. The solar-hydrogen CHP module
also provides the vector Qcool,FC of h elements, each calculated at the where ṁ h and ṁ c are respectively the mass flow rates at the hot and the
corresponding t as per Eq. (15). The vector Qcool,FC is then fed to the cold sides of the heat exchanger (kg/h).
thermal module of the simulation program. The outputs Et unmet byREN , The stratified vertical hot water storage tank is assumed to be of N
Ee unmet byREN , and Ophrs for each of the electrolyser and fuel cell, are also equidistant nodes (Fig. 1) with warm water settled on top of cold water.
calculated in the solar-hydrogen CHP module as per their equations The fixed-node model for a stratified hot water storage tank is con-
detailed in [20] and are fed to the economic module of the simulation sidered as in [45–47]. As per that, Ti (t + 1) is calculated based on the
program as was previously stated in Section 5.1 (Fig. 4). following energy balance equation applied for each node i of the tank.
For simplicity, the assumptions of fully mixed streams for each node,
one-dimensional temperature gradients along the vertical direction,
5.3. Thermal simulation module and uniform temperature and density of the fluid in each node are
considered.
5.3.1. Components models
The components of the thermal module mainly comprise the evac- Ti (t + 1)−Ti (t ) (kfluid + k wall ) A c,fluid,i
Mi ·cp· = (Ti + 1 (t )−Ti (t ))
uated-tube collectors, the stratified hot water storage tank as well as the Δt Δx i + 1 → i
fuel cell’s heat exchanger. (kfluid + k wall ) A c,fluid,i
+ (Ti − 1 (t )−Ti (t ))
The thermal efficiency of the collector (ηSC ) is obtained using the Δx i − 1 → i
Hotter-Whillier-Bliss equation corrected with the correction factor κ θ
+ Ui As,i (Ta (t )−Ti (t )) + (ṁ down,i (t )·cp·Ti − 1 (t )
for the incidence angle modifier [42]:
−ṁ down,i (t )·cp·Ti (t )) + (ṁ up,i (t )·cp·Ti + 1 (t )
(TSC,in−Ta) (TSC,in−Ta)2
ηSC = a0 × κ θ−a1 −a2 −ṁ up,i (t )·cp·Ti (t )) + (ṁ in,i (t )·cp·Tin,i (t )
Gt Gt (18)
−ṁ in,i (t )·cp·Ti (t ))) (26)
where a 0 is the intercept (maximum) of the collector efficiency; a1 is the
first-order heat loss coefficient of the collector (W/m2 °C); and a2 is the where Mi is the storage capacity of node i (kg); ṁ up,i and ṁ down,i are
second-order heat loss coefficient of the collector (W/m2 °C2). For respectively the mass flow rates of a flow upwards and downwards node
simplicity, κ θ is assumed to be equal to 0.95 [43]. i (kg/h) (Fig. 5); Ac,fluid,i and As,i are respectively the cross-sectional and
The output temperature (TSC ,out ) from the collector is then calculated surface area of node i (m2); kfluid and k wall are the de-stratification (fluid)
as: conductivity and the tank wall conductivity respectively in kJ/h m °C;
for simplicity, k wall is ignored in this study and assumed to be equal to
Gt ·ηSC ·ASC ·3.6 + ṁ SC ·cp·TSC,in zero; Ui is the node heat loss coefficient per unit area (kJ/h m2 °C); ṁ in,i
TSC,out =
ṁ SC ·cp (19) is the fluid mass flow rate of an entering flow to node i (kg/h) (Fig. 5);
Tin,i is its respective temperature (Fig. 5); Ti + 1, Ti , Ti − 1, are respectively
where ASC is the absorber area of the collector; ṁ SC is the mass flow rate the temperatures located below, at, and above node i . Ta (°C) is the
of water circulating throughout the collector (kg/h); and TSC ,in is the temperature of the environment (i.e. ambient temperature); and Δx i + 1 → i
inlet temperature to the collector (°C). and Δx i − 1 → i are the centre-to-centre distance between node i and the
A counter-flow heat exchanger is used to recover the fuel cell node below and above it respectively (m). Eq. (26) is stated for one
cooling load. The effectiveness (ε ) of a counter-flow heat exchanger (i.e. entering, one upward, and one downward flow to node i as applied to
for the fuel cell) is modelled using ε−NTU method as follows [40,44]: the system studied in this paper. However, generally, any additional

525
J. Assaf, B. Shabani Energy Conversion and Management 164 (2018) 518–532

The input flow to the collector with a mass flow rate of ṁ SC and a
temperature TSC ,in is taken from the bottom of the water tank (i.e. node
5) [17] (Fig. 1). Therefore:
TSC,in (t ) = T5 (t ) (27)
T5 is the temperature at node 5 of the water tank. TSC,out (t ) is cal-
culated as per Eqs. (18) and (19), when Gt is greater than zero. Ad-
ditionally, the controller of the collector circulation pump is set to stop
the pump if the temperature in the tank goes above 65 °C [17]. In this
case, the collector may tend to reach its stagnation temperature with
the efficiency ηSC (t ) decreasing towards zero (Eq. (18)).
The output flow from the collector goes to node 2 of the water tank
(Fig. 1). Therefore:
ṁ in,2 (t ) = ṁ SC (28)
Fig. 5. Flows defining the energy balance of node i of the hot water storage tank and the
control volume used to define the mass flows into and out of node. ṁ in,i,ṁ up,i and ṁ down,i Tin,2 (t ) = TSC ,out (t ) (29)
are respectively the mass flow rates of an inlet flow to node i , a flow upwards, and a flow
As this flow enters at node 2 and leaves at node N of the tank (i.e. to
downwards node i . Ti + 1, Ti , Tbii − 1 , are respectively the temperatures located below, at, and
the collector), it leads to a downward flow for nodes 3 to N of the tank:
above node i .
for i = 3→N
entering, upward, or downward flows to node i can be added in the ṁ down,i (t ) = ṁ in,2 (t ) (30)
same way in this equation with their respective temperatures. All the ṁ SC in kg/h is taken to be equal to 25 × ASC , as recommended in [48].
terms of Eq. (26) (i.e. ṁ in,i , Tin,i , ṁ up,i , and ṁ down,i ) are calculated The water tank is filled up with cold water at node N (bottom) and
whenever they exist for each node i of the tank, based on the thermal hot water is withdrawn from node 1 of the tank to the demand (Fig. 1).
integration of the system, as described next in Section 5.3.2. At the end The hot water is required to be supplied to at the demand at tem-
of each time interval, any temperature inversions that exist at the perature of 40 °C. Therefore, the hot water demand is:
stratified tank are eliminated by total mixing of the appropriate ad-
jacent nodes. QDem t (t ) = ṁ Dem t ·cp·(40−Tcold water (t )) (31)
where ṁ Dem t is the hot water demand (kg/h) that is assumed to be re-
5.3.2. Simulation model for the thermal module quired at 40 °C; and Tcold water is the cold water temperature (°C). If the
The flowchart of the simulation model for the thermal module is temperature at node 1 of the tank is less than 40 °C, then the mass flow
given in Fig. 6. As shown in Fig. 6, the calculation of the water tem- rate of hot water from the tank to demand is equal to ṁ Dem t and the
perature Ti at each node of the tank at all t’s permits the calculation of thermal demand QDem t is not fully met in this case. As the inlet mass
the thermal supply to the demand throughout the year and of the flow rate of cold water to node N of the tank (ṁ in,N ) is equal to the mass
thermal reliability (Rt ) . Accordingly, Ti of the next time step is calcu- flow rate of hot water from node 1 of the tank to demand, therefore:
lated based on Eq. (26), where the initial value for Ti is equal to the If T1 (t ) ⩽ 40
initial cold water temperature for all nodes (i.e. the tank was initially ṁ in,N (t ) = ṁ Dem t (t ) (32)
filled up with cold water). As already stated in the previous Section
5.3.1, the terms of Eq. (26) (i.e. ṁ in,i , Tin,i , ṁ up,i , and ṁ down,i ) are calcu- However, if the temperature at node 1 of the tank is higher than
lated whenever they exist for each node i of the tank, based on the 40 °C, then the hot water flow from the tank to demand is mixed up at
following system’s integration. the user’s side with some cold water in order to reach 40 °C. Therefore
ṁ in,N becomes in this case equal to:
If T1 (t ) > 40
QDem t (t )
min,N (t ) = cp × (T1 (t ) − Tin,N (t )) (33)
In both cases:
Tin,N (t ) = Tcold water (t ) (34)
As this flow enters at node N and leaves at node 1 of the tank, it
leads to an upward flow to nodes N − 1 to 1 of the tank (Fig. 5):
for i = 1 → N −1
ṁ up,i (t ) = ṁ in,N (t ) (35)
As per the configuration of the system illustrated in Fig. 1, the cold
(secondary) part of fuel cell heat exchanger is fed from the top part of
the water tank (node 1) [17]. Therefore:
Tc,in (t ) = T1 (t ) (36)
where T1 (t ) is the water temperature at node 1 of the hot water storage
tank (at time t). The output flow from the secondary (cold) side of the
heat exchanger goes to node 1 of the tank (Fig. 1). Therefore:
ṁ in,1 (t ) = ṁ c (t ) (37)
Fig. 6. Flowchart of the simulation model for the thermal module with the main input/
output data at each time t. Tin,1 (t ) = Tc,out (t ) (38)

526
J. Assaf, B. Shabani
Fig. 7. An iterative search subroutine to find ṁ c , ṁ h of the heat exchanger, and QTrsf ,FC , at each time t .
where Tc,out s calculated as per Eq. (24) based on the heat transferred
(QTrsf ,FC ) through the heat exchanger to the water tank. QTrsf ,FC is found
as described next.The vector Qcool,FC is fed into the thermal module from
the solar-hydrogen CHP module as noted in Section 5.2.2 and as illu-
strated in Fig. 3. Qcool,FC is provided to the iterative search subroutine
for the search of the mass flow rates for the fuel cell heat exchanger
(Fig. 7). The purpose of this subroutine, as shown in Fig. 7, is to find
ṁ h (t ) and ṁ c (t ) at each t that enable Qcool,FC (t ) to be fully or as max-
imum as possible transferred to the hot water storage tank. The con-
dition is to have Th,in (t ) at 65 °C at all t’s (i.e. fuel cell temperature is
aimed to be maintained at 65° C as recommended by [9]). Another
condition is to have the temperature difference at the hot side of the
heat exchanger (ΔTh) at a targeted temperature (ΔTh Tar ) taken as 3 °C, as
also recommended in [9] (i.e. ΔTh to be between 3 °C and 5 °C) for a
proper cooling of the fuel cell. Therefore:
ΔTh = Th,in (t )−Th,out (t )
Th,in (t ) = 65
The heat transferred to the water tank, QTrsf ,FC (t ) , is calculated by
following the flowchart provided in Fig. 7. In fact, Qcool,FC (t ) may not be
all transferred to the water tank due to a temperature at the top of the
tank (node 1) that can inhibit the total heat transfer of Qcool,FC (t )
through the heat exchanger.
5.4. Economic simulation module
The model of the economic module is well detailed in [20]. The
main output of this model is the levelised cost of primary energy
COEprimary . It is calculated based on the net total cost of the system that
includes the initial capital or investment costs, maintenance costs, re-
placement costs, operation costs, and minus any salvage return values
at the end of the project lifetime. It is worth to note that COEprimary
includes, as operation cost, the cost associated with two non-renewable
components (i.e. electric diesel generator, inline gas heater), for sup-
plying the still possibly small percent of electrical and/or thermal en-
ergies that could be unmet by the renewable system (in case R e < 100%
and/or Rt < 100%).
Energy Conversion and Management 164 (2018) 518–532
6. Case study
6.1. An introduction to the case
A description of the input data and parameters used for the case
study of the multi-objective sizing optimisation of the SH CHP-ST
system is given as follows:
• Meteorological data: They consist of the solar irradiance G t on the
tilted surface (PV, and/or collector), as well as the ambient tem-
perature (Ta) , and cold water temperature from municipality
(Tcold water ) . Hourly average meteorological data for the simulation
time (i.e. a year) are taken from TRNSYS database for Melbourne,
Australia for a surface of a tilt angle at 37.8° (i.e. equal to the local
latitude of Melbourne) facing north. Considering the significant
difference between the winter and summer solar radiations in
(39) Melbourne (i.e. requirement for long-term energy storage), this city
has been suggested to be a good candidate to use solar-hydrogen
systems [49].
(40)
• Demand data: These consist of the loads’ profiles (i.e. power de-
mand, hot water demand). They are both taken as daily-repeated
patterns for a conservative remote household in Victoria [17], and
are presented in Fig. 9. The total is of 5 kWh daily power demand
excluding heating and cooling, and of 165 L daily hot water demand
at 40 °C.
• Technical parameters. They represent the components’ character-
istics and some technical assumptions. They are taken as considered
in [17] for the components used, and are shown in summary in
Table 1.
Economic parameters: They consist of the economic parameters
considered such as the interest rate, inflation rate, cost of components,
project lifetime, and etc. They are taken as in [20] and briefly described
in Tables 2 and 3. Same costs and economic assumptions as recently
adopted in [20] are considered so that the techno-economic results of
the sizing optimisation in this paper can be compared to that of the
previous sizing strategy of the system in [17] and its economics [20].
6.2. Results, analysis, and discussion
6.2.1. Verification of the MATLAB simulation model
In order to verify the compatibility of comparing the optimisation
527
J. Assaf, B. Shabani Energy Conversion and Management 164 (2018) 518–532

Table 1
Technical parameters and assumptions for the components of the SH CHP-ST system.

Temperature coefficient of the PV module (α ) −0.5%/C

Efficiency of auxiliary components and other loss 0.9
factors (ηc )
Tilt angle of PV 37.8 degrees facing north
Electrolyser efficiency 70%
Characteristic curve of fuel cell As shown in Fig. 8
Hydrogen utilisation factor for fuel cell (μf ) 0.83
Intercept efficiency (a0) for collector 0.8
Negative of first order efficiency coefficient (a1) for 1.1248 W/m2 K
collector
Negative of second order efficiency coefficient (a2) 0.0279 W/m2 K2
for collector
Incidence angle modifier (κ θ) for collector 0.95
Tilt angle of collector 37.8 degrees facing north
Height of hot water storage tank 1.62 m
Number of nodes considered for hot water storage 5
Fig. 8. Characteristic curve of the PEM fuel cell as considered in [17] based on the Butler- tank
Volmer equation. The current density is the cell current over the active cross-sectional Heat loss coefficients for all nodes of hot water tank 5 kJ/hr m2 K
area of the cell (MEA). The maximum rated power of the cell is 0.78 W.

results with those of the TRNSYS model in [17], the following test has
been done: the sizing vector [3500, 2500, 312, 195, 250, 2] corre-
sponding to the vector [Pr ,PV ,Pr ,EL,Pr ,FC ,VolTK H2,VolTKHW ,ASC ] and that has
been adopted in [17] using TRNSYS software (i.e. following the sizing
strategy considered in [17]) has been entered into the MATLAB simu-
lation program developed for this optimisation study. This has yielded
in MATLAB to very similar overall annual results in terms of reli-
abilities, efficiencies, cost of primary energy, and etc. as those obtained
in TRNSYS (Table 4).
6.2.2. Optimisation results and analysis
As stated in Section 4, the optimisation model considers three ob-
jectives: i.e. Ro to be maximised, while COEprimary and PercE PV waste to be
minimised. This multi-objective sizing optimisation performed on the
SH CHP-ST system result in multiple optimal non-dominated solutions
that constitute the Pareto front, where no improvement can be done on
one objective without losing on the other. The 2D Pareto front plot with
Ro on x-axis and COEprimary on y-axis is shown in Fig. 10, while the 2D
Pareto front plot with PercE PV waste on x-axis and COEprimary on y-axis is
shown in Fig. 11. The conflicting relation between the objective func-
tions is shown in these figures. There is a clear trend that COEprimary
increases with higher Ro (Fig. 10). However, this relationship is not a
properly defined curve as these optimal points plotted in Fig. 10 cor-
respond to different values of PercE PV waste . While in Fig. 11, the general
trend is that the COEprimary increases with lower PercE PV waste (i.e. mainly
due to the increase in the size of the hydrogen components); however,
this relationship is also not as a defined curve as the optimal points
plotted in Fig. 11 correspond to different values of Ro . Fig. 12 that plots
the 2D Pareto front with PercE PV wast on the x-axis and Ro on the y-axis
gives no suggestion about a general trend between these two variables.
It is noted from the MATLAB detailed solutions that, the obtained
Pareto optimal solutions, with a COEprimary less than 0.41 (points below
the horizontal dotted line in Fig. 10), have all an electric reliability of
100%. Only the Pareto optimal solutions in Fig. 10, with a COEprimary
greater than 0.41, have their R e less than 100%. This relatively high
COEprimary for these solutions is caused by the fact that any solution with
an electrical reliability of less than 100% incurs increased additional
costs resulting from the introduction of a diesel generator, as in-
troduced in Section 5.4 (i.e. with all the associated high fixed annual
maintenance costs especially for a remote area). However, these points
fall on the Pareto front despite their high COEprimary due to their mini-
mised PercE PV waste (i.e. close to zero). Another point to note from the
Pareto front (Fig. 10), is that the maximum Ro that could be achieved is
around 96.35%, but with a COEprimary at 0.45 and a PercPV wast of 0.67%
(point F on the graph). The R e is less than 100% in this case and the
COEprimary is relatively high. However, when Ro is just slightly less than
96.35% (i.e. 96.32% at point A) in Fig. 10, the COEprimary for this op-
timal solution falls down to 0.40 with a PercE PV wasted of 0.84% (Fig. 11).
R e for this solution is 100%, and consequently Rt is the same as Ro .
However, it is seen from the solutions that the points falling on the left

Fig. 9. Energy demands of a remote household in Melbourne, Victoria (base case study). (a) Average hourly power demand for a remote household in southeast Australia (Victoria) during
a day [38]; conservative electrical demand excluding heating and air conditioning loads as a passive house is considered. (b) Average hourly hot water demand at 40 °C in a day. The
hourly fractional distribution of the daily hot water demand is taken from ASHRAE 90.2 [47].

528
J. Assaf, B. Shabani Energy Conversion and Management 164 (2018) 518–532

Table 2
Economic data for the components of the SH CHP – ST system and for the non-renewable support components, if needed – [20] Year 2015–2016.

Components for the SH-ST CHP system Capital cost Price after Annual maintenance cost (as percentage of Lifetime
5 years capital cost)

PV panel including invertor 2000 US$/kW 2000 US$/kW 0% 25 years

Electrolyser 5000 US$/KW 2000 US$/KW 2% 30,000 h
Fuel cell 2500 US$/kW 1000 US$/kW 2% 30,000 h
Hydrogen storage tank 50 US$/N m3 50 US$/N m3 0.5% 25 years
Evacuated tube collectors, Hot water storage tank, with control and pump, and gas US$3200 US$3200 0% 15 years
booster if necessary
Balance of system: valves, regulators, safety measures, humidification/ US$2500 US$2500 0% 25 years
dehumidification, control, heat exchanger

Non-renewable support components

Diesel generator; efficiency ηgen : 0.25 1000 US$/kW 1000 US$/kW 300 US$ 20,000 h
Gas water heater; efficiency ηGWH : 0.9 1000 US$/unit 1000 US$/kW 0% 25 years
Diesel tank + balance of system 1000 US$ 1000 US$ 0% 25 years

Table 3
General economic data adopted.

General economic data

Project lifetime 25 years

Nominal interest rate 5%
Inflation rate 2.5%
Price of LPG for remote areas US$150 per 45 kg-LPG bottle
delivered
Unit price of diesel fuel for remote areas 1.9 US$/litre delivered
Annual increase in diesel fuel and gas 3%
prices

Table 4
Annual (average) results obtained by MATLAB and TRNSYS [17] for the sizes considered
in [17].

Annual results Fig. 11. 2D Pareto front obtained from the multi-objective sizing optimisation with
PercE PV waste on x-axis, and COEprimary on y-axis.
Electric reliability (R e ) 100%
Thermal reliability (Rt ) ∼95% (95.3% in MATLAB
and 95.2% in TRNSYS)
PercE PV waste ∼0% (0.5%)
FC electric efficiency (ηe FC ) 33%
FC overall efficiency (electric and thermal) 60%
(ηoverall,FC ) considering the FC heat
transferred to tank
Collector’s thermal efficiency 67%
COEprimary 0.39 US $/kWh primary

Fig. 12. 2D Pareto front obtained from the multi-objective sizing optimisation with
PercE PV waste on x-axis, and Ro on y-axis.

Fig. 10. 2D Pareto front obtained from the multi-objective sizing optimisation with Ro on
x-axis, and COEprimary on y-axis. The description of the indicated points is explained in
Section 6.2.2.
side of the vertical dotted line in Fig. 10 have their fuel cell size around
280 W that is less than 300 W (i.e. maximum hourly power demand as
shown in Fig. 9a). Consequently, if the user would like to design the
system with a minimum size of fuel cell equal to the maximum hourly
power demand, points on the right side of the vertical dotted line shall
only be of interest. Among these points, the maximum overall reliability
that can be obtained is 95.64% (at point G) with a COEprimary of 0.37

529
J. Assaf, B. Shabani Energy Conversion and Management 164 (2018) 518–532

Table 5
Selected solutions from the Pareto front of the multi-objective sizing optimisation for the SH CHP-ST, with all solutions of the Pareto front accessed in excel through this link.

Optimised objectives Solutions (Sizes of components)

Rt (%) R e (%) COEprimary (US$/kWh primary) PercE PV waste (%) Pr ,PV (W) Pr ,EL (W) Pr ,FC (W) VolTK H2 (N m3) VolTKHW (litres) ASC (m2)

1 (point B) 92.4 100 0.307 33 4643 1268 476 155 236 2

2 (point D) 95.2 100 0.339 15 4091 1712 312 182 233 2
3 (point C) 95.3 100 0.396 0.5 3509 2500 312 195 240 2
4 (point G) 95.6 100 0.366 6 3757 2112 304 188 232 2
5 (point F) 96.3 99.8 0.455 0.7 3741 2537 280 206 232 2
6 (point E) 91.9 99.8 0.438 0 3084 2473 554 171 242 2

and a PercE PV waste equal to 6% (The results of point G are detailed in row
4 of Table 5). On the other side, the minimum Ro obtained on the Pareto
front with a minimal COEprimary is achieved at point B with an Ro of
92.4%, a COEprimary of around 0.31, and a PercE PV waste of 33% (Fig. 12).
This figure of PercE PV waste is very close to the highest PercE PV waste ob-
tained in the Pareto optimal solutions (i.e. 33.6%) (Fig. 11). In sum-
mary, Ro varies between around 92% and 96.3% among all obtained
optimal solutions, or between 92% and 95.6% if a minimum size of fuel
cell is adopted.
So, out of the Pareto optimal solutions, the adopted solution for the
system depends on the preferences and criteria of the decision-maker.
For instance, the second objective COEprimary could be preferred to be
given a priority over the third objective PercE PV waste for the different
values of Ro′ s . For instance, one may choose the optimal solution shown
in row 3 in Table 5 and indicated by point C in Figs. 10 and 11. This
solution has an Ro of around 95.3%, with a PercE PV waste of 0.5% and a
COEprimary of 0.396. This solution is nearly the same as the solution
obtained using the sizing approach in TRNSYS in [17] (Table 4). Al-
Fig. 13. Mass of hydrogen stored in tank throughout the year (17,519 time steps; each
ternatively, by increasing the PV array from 3509 W (at point C pre- time step is equal to 0.5 h) for the two optimal solutions: point C (i.e. similar to the
sented in row 3 of Table 5) to 4091 W (at point D presented in row 2 of TRNSYS solution in [17]) in blue, and point D in green. (For interpretation of the re-
Table 5), which is an increase of around 16.5% in PV size, accompanied ferences to colour in this figure legend, the reader is referred to the web version of this
by a decrease in the electrolyser and hydrogen tank sizes as indicated article.)
(i.e. by around 31% for the electrolyser), a more favourite solution in
term of cost (COEprimary ) with around the same reliability (i.e. 95.2%) is The main difference is that the solution of point C has a bigger
obtained. This solution (i.e. at point D) gives approximately the same hydrogen subsystem, which allows for storage of more hydrogen in the
overall reliability as for point C, at a COEprimary (i.e. 0.34) reduced by tank. Consequently, the maximum mass of hydrogen that is reached in
around 14.4% (compared to point C), but at a higher PercE PV waste , which the tank for the solution of point C (in blue) is 17.5 kg, while it is
is ∼16% compared to 0.5% for point C (Table 5). 16.2 kg for point D (in green).
On the other side, taking the extreme lowest COEprimary that can be Another point to note in the Pareto optimal solutions is that the fuel
achieved among the optimal solutions, the COEprimary can go down to cell size is lower in the solutions with higher Ro ′ s (i.e. or alternatively
0.307 (at point B), which is around 22.5% less than that of point C with speaking with higher thermal reliability, as in most of them R e = 100%).
a thermal reliability (i.e. same as overall reliability) of 92.4% and This can be clearly seen for the few solutions presented in Table 5.
PercE PV waste of 33% (Table 5). In this case (i.e. at point B), the PV size is Fig. 14 illustrates this deduction for the different Pareto optimal solu-
increased by 32.2% compared to point C. tions obtained. This is due to the fact that the electric efficiency of the
As observed, ASC was obtained the same in all the solutions due to
the fact that the time the collector is reaching stagnation temperature
annually (nbhrsstagSC ) (particularly in hot sunny periods) is limited
and constrained, as was mentioned in Section 4, for a proper technical
design of the solar hot water system. In fact, nbhrsstagSC is mainly
affected by the collector’s size with respect to hot water demand, par-
ticularly in sunny days of summer. The fuel cell operates at times of low
solar radiation; and the fuel cell heat at different solutions did not affect
the optimal size of ASC .
For more elaboration, Fig. 13 shows the variation of the mass of
hydrogen in the tank taken from MATLAB for both point C (in blue) and
point D (in green). As designed for in the simulation program, the mass
of hydrogen at the end of the year (i.e. for the time step equal to
17,519) is always equal to the mass of hydrogen MH2 initial at the start of
the year, which is obtained at 11.1 kg for point C and at 10.6 kg for
point D. The minimum mass of hydrogen that is supposedly required to
stay in the tank for allowance of an autonomous operation of the system
under any contingency is as pre-defined for all solutions (i.e. 3.5 kg for
around 10 days of autonomous operation). Fig. 14. Overall reliability vs. fuel cell size among the optimal solutions.

530
J. Assaf, B. Shabani
fuel cell is increased when it operates at certain power interval lower
than its rated size, as confirmed experimentally by Shabani and An-
drews [11]. This can also be clearly seen by considering the points with
the two extreme reliabilities in Fig. 10: Point E (fuel cell size: 554 W; Ro :
91.9%) and point F (i.e. fuel cell size: 280 W; Ro : 96.3%). These two
solutions have the two extreme fuel cell sizes and their results are de-
tailed respectively in row 6 and row 5 of Table 5.
7. Conclusion
A multi-objective sizing optimisation was performed in MATLAB
programming environment on a solar-hydrogen combined heat and
power system integrated with a solar-thermal system. In this arrange-
ment, the solar-hydrogen combined heat and power supplies power to
the electrical demand, while the heat recovered from the fuel cell is
transferred to the top part of the water tank of a solar thermal system
that mainly consists of a solar thermal collector (i.e. evacuated-type)
and a hot water storage tank. The aim was to find the optimised sizes
for the main components of the integrated system with the objectives of
maximising the overall reliability (i.e. electric and thermal), minimising
the levelised cost of primary energy, and minimising the percentage of
energy generated by the PVs that is not utilised. The simulation module
of the system guaranteed that the mass of hydrogen in the tank at the
end of the year is equal to that at the start of the year by finding the
required initial mass of hydrogen in the tank, so that no hydrogen is
produced as extra or less during the year. The main findings of this
paper are:
– Only the Pareto optimal solutions that have a COEprimary greater than
0.41 US$/kWh primary, have their R e less than 100%. This rela-
tively high COEprimary for the solutions that have an R e less than
100%, is caused by the fact that such solutions incur increased ad-
ditional costs resulting from the introduction of a diesel generator.
The latter is accompanied with all the associated high fixed annual
maintenance costs especially for a remote area. So, having an elec-
tric reliability of less than 100% will not give favourite solutions
especially in terms of cost.
– A trade-off between COEprimary and PercE PV waste could be seen in the
optimal Pareto solutions for different levels of Ro . For example,
compared to the optimal solution obtained, where Ro is at 95.3%
with PercE PV waste close to zero (i.e. 0.5%) and COEprimary at 0.396 US
$/kWhprimary (i.e. point C that is nearly the same as the solution
obtained using the sizing approach in TRNSYS in [17]), another
optimal solution with the same reliability results, that is more eco-
nomic can be selected: This solution has a COEprimary reduced by
14.4% compared to point C and comes at a PercE PV waste of 15%. In
this solution, the PV size is increased by 16%, and the electrolyser’s
size reduced by around 31% with a decrease of around 6.7% in the
size of the hydrogen tank.
– Taking the extreme lowest COEprimary that can be achieved among
the optimal solutions, the COEprimary can go down to 0.307 US
$/kWhprimary, which is around 22.5% less than that of point C with a
thermal reliability (i.e. same as overall reliability) of 92.4% and
PercE PV waste of 33%.
– The maximum thermal reliability that could be obtained among the
optimal solutions is 96.3%; or it is 95.6% if the user would like to
design the system with a minimum size of fuel cell equal to the
maximum hourly power demand.
– It was also found in the Pareto optimal solutions that the fuel cell
size is lower in the solutions with higher overall reliability.
– Still few percent of the thermal demand are not met. This can lead to
further research by considering this few percent to be met elec-
trically (i.e. by an inline electric heater for example) powered by PV
or FC and guaranteeing a 100% reliability. This would also be a
challenging task that needs further research.
531
Energy Conversion and Management 164 (2018) 518–532
This optimisation helped in finding optimal sizes of this integrated
system, where the effects of the characteristics of the multiple compo-
nents are inter-related and complicated. Additional beneficial results
than those obtained by a straightforward sizing strategy are attained.
Acknowledgement
This work was supported by the “Australian Government Research
Training Program Scholarship”.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in the
online version, at http://dx.doi.org/10.1016/j.enconman.2018.03.026.
References
[1] Uyar TS, Beşikci D. Integration of hydrogen energy systems into renewable energy
systems for better design of 100% renewable energy communities. Int J Hydrogen
Energy 2017;42:2453–6.
[2] Shabani B, Andrews J. Standalone solar-hydrogen systems powering Fire
Contingency Networks. Int J Hydrogen Energy 2015;40:5509–17.
[3] Ozden E, Tari I. Energy–exergy and economic analyses of a hybrid solar–hydrogen
renewable energy system in Ankara, Turkey. Appl Therm Eng 2016;99:169–78.
[4] Sayedin F, Maroufmashat A, Sattari S, Elkamel A, Fowler M. Optimization of
Photovoltaic Electrolyzer Hybrid systems; taking into account the effect of climate
conditions. Energy Convers Manage 2016;118:438–49.
[5] Shabani B, Andrews J. Hydrogen and fuel cells. Energy sustainability through green
energy. Springer; 2015.
[6] Shabani B, Omrani R, Andrews J. Energy security and sustainability for road
transport sector: the role of hydrogen fuel cell technology. Energy security and
sustainability. Taylor & Francis Group; 2017.
[7] Amirante R, Cassone E, Distaso E, Tamburrano P. Overview on recent developments
in energy storage: mechanical, electrochemical and hydrogen technologies. Energy
Convers Manage 2017;132:372–87.
[8] Aris AM, Shabani B. Sustainable power supply solutions for off-grid base stations.
Energies 2015;8:10904–41.
[9] Shabani B, Andrews J, Watkins S. Energy and cost analysis of a solar-hydrogen
combined heat and power system for remote power supply using a computer si-
mulation. Sol Energy 2010;84:144–55.
[10] Moore J, Shabani B. A critical study of stationary energy storage policies in
Australia in an international context: the role of hydrogen and battery technologies.
Energies 2016;9(9):674.
[11] Shabani B, Andrews J. An experimental investigation of a PEM fuel cell to supply
both heat and power in a solar-hydrogen RAPS system. Int J Hydrogen Energy
2011;36:5442–52.
[12] Sajid Hossain M, Shabani B. Metal foams application to enhance cooling of open
cathode polymer electrolyte membrane fuel cells. J Power Sources
2015;295:275–91.
[13] Tetuko AP, Shabani B, Andrews J. Thermal coupling of PEM fuel cell and metal
hydride hydrogen storage using heat pipes. Int J Hydrogen Energy
2016;41:4264–77.
[14] Islam MR, Shabani B, Rosengarten G. Nanofluids to improve the performance of
PEM fuel cell cooling systems: a theoretical approach. Appl Energy
2016;178:660–71.
[15] Shabani B, Andrews J, Badwal S. Fuel cell heat recovery, electrical load manage-
ment, and the economics of solar-hydrogen systems. Int J Power Energy Syst 2010.
[16] Budihardjo I, Morrison GL. Performance of water-in-glass evacuated tube solar
water heaters. Sol Energy 2009;83:49–56.
[17] Assaf J, Shabani B. Transient simulation modelling and energy performance of a
standalone solar-hydrogen combined heat and power system integrated with solar-
thermal collectors. Appl Energy 2016;178:66–77.
[18] Rodríguez-Hidalgo MC, Rodríguez-Aumente PA, Lecuona A, Legrand M, Ventas R.
Domestic hot water consumption vs. solar thermal energy storage: the optimum size
of the storage tank. Appl Energy 2012;97:897–906.
[19] Friedrich Ferrer PA. Average economic performance of solar water heaters for low
density dwellings across South Africa. Renew Sustain Energy Rev 2017;76:507–15.
[20] Assaf J, Shabani B. Economic analysis and assessment of a standalone solar-hy-
drogen combined heat and power system integrated with solar-thermal collectors.
Int J Hydrogen Energy 2016;41:18389–404.
[21] Dou XX, Andrews J. Design of a dynamic control system for standalone solar-hy-
drogen power generation. Proc Eng 2012;49:107–15.
[22] Paola A, Fabrizio Z, Fabio O. Techno-economic optimisation of hydrogen produc-
tion by PV e electrolysis: “RenHydrogen” simulation program. Int J Hydrogen
Energy 2011;36:1371–81.
[23] Al-falahi MDA, Jayasinghe SDG, Enshaei H. A review on recent size optimization
methodologies for standalone solar and wind hybrid renewable energy system.
Energy Convers Manage 2017;143:252–74.
[24] Erdinc O, Uzunoglu M. Optimum design of hybrid renewable energy systems:
overview of different approaches. Renew Sustain Energy Rev 2012;16:1412–25.
J. Assaf, B. Shabani
[25] Alarcon-Rodriguez A, Ault G, Galloway S. Multi-objective planning of distributed
energy resources: a review of the state-of-the-art. Renew Sustain Energy Rev
2010;14:1353–66.
[26] Luna-Rubio R, Trejo-Perea M, Vargas-Vázquez D, Ríos-Moreno GJ. Optimal sizing of
renewable hybrids energy systems: a review of methodologies. Sol Energy
2012;86:1077–88.
[27] Sharafi M, Elmekkawy TY. Multi-objective optimal design of hybrid renewable
energy systems using PSO-simulation based approach. Renewable Energy
2014;68:67–79.
[28] Kashefi Kaviani A, Riahy GH, Kouhsari SM. Optimal design of a reliable hydrogen-
based stand-alone wind/PV generating system, considering component outages.
Renewable Energy 2009;34:2380–90.
[29] Brka A, Al-Abdeli YM, Kothapalli G. The interplay between renewables penetration,
costing and emissions in the sizing of stand-alone hydrogen systems. Int J Hydrogen
Energy 2015;40:125–35.
[30] Giannakoudis G, Papadopoulos AI, Seferlis P, Voutetakis S. Optimum design and
operation under uncertainty of power systems using renewable energy sources and
hydrogen storage. Int J Hydrogen Energy 2010;35:872–91.
[31] Singh A, Baredar P, Gupta B. Techno-economic feasibility analysis of hydrogen fuel
cell and solar photovoltaic hybrid renewable energy system for academic research
building. Energy Convers Manage 2017;145:398–414.
[32] Gao X, Akashi Y, Sumiyoshi D. Installed capacity optimization of distributed energy
resource systems for residential buildings. Energy Build 2014;69:307–17.
[33] IEA SHC Task 49. Overheating prevention and stagnation handling in solar process
heat applications; 2015.
[34] Tezer T, Yaman R, Yaman G. Evaluation of approaches used for optimization of
stand-alone hybrid renewable energy systems. Renew Sustain Energy Rev
2017;73:840–53.
[35] Bernal-Agustín JL, Dufo-López R. Efficient design of hybrid renewable energy sys-
tems using evolutionary algorithms. Energy Convers Manage 2009;50:479–89.
[36] Shirazi A, Taylor RA, Morrison GL, White SD. A comprehensive, multi-objective
optimization of solar-powered absorption chiller systems for air-conditioning
532
Energy Conversion and Management 164 (2018) 518–532
applications. Energy Convers Manage 2017;132:281–306.
[37] Deshmukh MK, Deshmukh SS. Modeling of hybrid renewable energy systems.
Renew Sustain Energy Rev 2008;12:235–49.
[38] Shabani B. Solar-hydrogen combined heat and power systems for remote area
power supply. School of Aerospace, Mechanical and Manufacturing Engineering.
Melbourne. Victoria: RMIT; 2010.
[39] Moseley PT. Fuel cell systems explained. In: James Larminie, Andrew Dicks (Eds.),
Wiley, Chichester, Weinhein, New York, Brisbane, Singapore and Toronto, ISBN 0-
471-49026-1. Journal of Power Sources. 2001;93:285.
[40] TRNSYS 17 Manual. A transient system simulation program. Madison, WI:
University of Wisconsin, Madison, Solar Energy Laboratory; 2012.
[41] Islam MR, Shabani B, Rosengarten G, Andrews J. The potential of using nanofluids
in PEM fuel cell cooling systems: a review. Renew Sustain Energy Rev
2015;48:523–39.
[42] Duffie JA, Beckman WA. Solar engineering of thermal processes. John Wiley &
Sons, Inc; 2013.
[43] Greenland systems. < http://www.greenlandsystems.com/ > .
[44] Kays WM, London AL. Compact heat exchangers. New York: McGraw-Hill Book Co;
1964.
[45] Morrison G, Rosengarten G. Solar thermal energy design: school of mechanical and
manufacturing engineering. University of New South Wales; 2010.
[46] Klein SA. A design procedure for solar heating systems. University of Wisconsin-
Madison; 1976.
[47] Cruickshank CA. Evaluation of a stratified multi-tank thermal storage for solar-
heating applications. Kingston, Ontario, Canada: Queen’s University; 2009.
[48] Gao Y, Zhang Q, Fan R, Lin X, Yu Y. Effects of thermal mass and flow rate on forced-
circulation solar hot-water system: comparison of water-in-glass and U-pipe evac-
uated-tube solar collectors. Sol Energy 2013;98(Part C):290–301.
[49] Andrews J, Shabani B. Dimensionless analysis of the global techno-economic fea-
sibility of solar-hydrogen systems for constant year-round power supply. Int J
Hydrogen Energy 2012;37:6–18.