Energy Conversion and Management 312 (2024) 118588

Contents lists available at ScienceDirect

Energy Conversion and Management

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

Component capacity optimization of a renewable energy system using

data-driven two-stage algorithmic approach
Wenrui Ye a, Munur Sacit Herdem b, Shucheng Huang a, Wei Sun c, Jun Liu d, Jatin Nathwani e,
John Z. Wen a, *
a
Department of Mechanical and Mechatronics Engineering, University of Waterloo, 200 University Ave W, Waterloo, N2L 3G1, Canada
b
Department of Mechanical Engineering, Adiyaman University, Altınşehir, Atatürk Bv No:1, Adiyaman, 02040, Turkey
c
Bugu Solutions Technology, Waterloo, N2L 3G1, Canada
d
Department of Applied Mathematics, University of Waterloo, 200 University Ave W, Waterloo, N2L 3G1, Canada
e
Department of Civil and Environmental Engineering, University of Waterloo, 200 University Ave W, Waterloo, N2L 3G1, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: Navigating the complexities of optimizing Renewable Energy System components requires addressing economic,
Optimization technical, and regulatory challenges. However, existing research often overlooks crucial aspects such as grid
Genetic Algorithm economic interactions, the limited scope of Renewable Energy System objectives, and the scalability of energy
Energy Management
strategies. This study introduces a novel two-stage optimization algorithm integrating a non-dominated sorting
Capacity
Renewable Energy System
algorithm-II for capacity optimization and a multi-integer linear programming model for energy management,
offering a comprehensive solution with diverse decision-making metrics. With implementation of the dynamic
population decay algorithm, the computation time was reduced by 60.42%. The study identified nine Pareto
efficient configurations, with the highest Internal Rate of Return reaching 4.86% and a maximum Energy In­
dependence Score of 0.51. The financial cost associated with improving environmental indicators surged by
839%. Furthermore, the proposed optimization approach outperformed the rule-based non-dominated sorting
algorithm, achieving an 18.12% higher Internal Rate of Return with a comparable energy independence level.
This decision-making framework guides system owners towards medium-sized systems for balanced objectives
while offering flexibility for various sizes tailored to specific local regulations, energy markets, and goals,
extending its applicability to diverse international contexts.

(e.g., converters), and user (e.g., the grid) units. A high-performance

RES requires not only the development of advanced technological
1. Introduction
integration method [4] but also adaptability to practical applications
ranging from household energy supply [5], commercial energy genera­
Driven by the global effort to achieve net-zero emissions by 2050, the
tion [6], to the cutting-edge technologies such as green hydrogen con­
deployment of renewable energy systems (RESs), especially those uti­
version and electrified transportation [7]. It is generally essential to
lizing solar photovoltaic (PV) resources, has become urgent and critical
install an Energy Storage System (ESS) in a RES, to mitigate the inter­
in many countries, while the associated energy management strategies
mittency of PV and wind energy production [8], ensure the grid stability
need to be developed in order to amplify their decarbonization effects
[9], maximize the renewable usage [10], and facilitate energy man­
[1]. It was reported at the end of 2021, the level of global PV installation
agement [11].
reached a significant milestone, with more than 942 GW, which effec­
These two critical components on the topology of a PV based RES are
tively reduces the carbon dioxide emission by approximately 1,100
PV panels and ESS (usually battery based), which play a vital role in the
million tons [2]. In 2022, utilization of solar PV power further surged by
lifetime of a RES. First, their life cycles essentially determine the lifetime
a record breaking 26 % increase, resulting in an approximate production
of a RES. While the PV panel can span the lifetime of a RES, replacement
of 1,300 TWh [3]. In addition to PV technologies, RESs are operated
of ESSs is often necessary, and their efficiency decay should be
with advanced energy management strategies, together with a variety of
accounted for. Secondly, the operations of PV and ESS technologies are
critical components such as energy storage (e.g., batteries), conversion

* Corresponding author.
E-mail address: john.wen@uwaterloo.ca (J.Z. Wen).

https://doi.org/10.1016/j.enconman.2024.118588
Received 13 December 2023; Received in revised form 17 April 2024; Accepted 19 May 2024
Available online 24 May 2024
0196-8904/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Nomenclature EPV PV energy output

EPV Grid Energy from PV to grid
BChar Binary variable − charge EPV Load Energy from PV to load
BDis Binary variable − discharge EPV BESS Energy from PV to BESS
BEx Binary variable − export EPV Curt Curtailed energy from PV
BIm Binary variable − import EPV RE Energy from PV to RE
CO&M Overall O&M cost FBESS age Aging factor of BESS
CO&M,BESS O&M costs for BESS LBESS cal Calendar life of BESS
CO&M,Others O&M costs for other components PConverter_BESS Power rating of the BESS converter
CO&M,PV O&M costs for PV PInverter_BESS Power rating of the BESS inverter
CRep Total replacement cost PInverter_PV Power rating of the PV inverter
CapBESS Capacity of BESS RESav Revenue from energy saved
CapBESS aged Aged capacity of BESS RESol Revenue from energy sold
CapOthers Capacity of other components r Discount rate
CapPV Capacity of PV rBESS Discount rate for BESS unit price
EBESS Grid Energy from BESS to grid (t) Hourly time-domain data
EBESS Load Energy from BESS to load (y) Yearly data
EBESS max Rated power of the BESS ηConverter BESS Converter efficiency for BESS
EChar Charge energy ηInverter BESS Inverter efficiency for BESS
EDis Discharge energy ηInverter PV Inverter efficiency for PV
EEx Energy exported to grid ωO&M,BESS O&M coefficient for BESS
EGrid BESS Energy from grid to BESS ωO&M,Others O&M coefficient for other components
EGrid Load Energy from grid to load ωO&M,PV O&M coefficient for PV
EGrid max Rated power of the grid πCap,BESS BESS price per unit (kWh)
EIm Energy imported from grid πCap,Others Price per unit for other components (kWh)
ELoad Load energy demand πCap,PV PV price per unit (kWh)
ELoss BESS Energy loss in BESS πEx Energy export price
ELoss Converter BESS Energy loss in the BESS converter πIm Energy import price
ELoss Inverter BESS Energy loss in the BESS inverter
Eloss Inverter PV Energy loss in the PV inverter

highly integrated. An ESS stores the excess energy produced from PV for optimization, achieving quicker results without sacrificing the accuracy
both energy supply and economic benefits. For a grid-connected RES, compared to traditional simulation software like HOMER [21]. Addi­
the role of ESS is more significant as it acts as a buffer as well as a source tionally, Hassan et al. combined hardware models with NSGA-II to
[12]. The deployment of a RES requires capacity optimization of its PV identify cost-effective configurations of a grid-connected solar PV sys­
and ESS components, with the targets of maximizing renewable energy tem, reducing energy costs by 30 % [22]. Wu et al. explored the life-
utilization, reducing capital and operation costs, and minimizing the cycle cost minimization in a PV-battery system using a mixed-integer
environmental footprint. In principle, these two tasks are done in ca­ nonlinear programming model, discovering that time-of-use tariffs
pacity optimization in an inherently integrated way: simulation of the were more beneficial than flat rate tariffs [23]. The diversity and
system operation and optimization of the sizes of the critical compo­ effectiveness of these heuristic algorithms underscore the complexity
nents. Note that for both tasks, the energy management strategy must be and innovation in RES optimization. A comparison of recent publica­
implemented using the updated information and data. tions on the application of these methods and the diverse objectives and
Due to the non-convex nature of capacity optimization in renewable performance indicators is summarized in Table 1. However, to further
energy systems (RESs), traditional gradient-based methods such as refine the optimization process and to enhance the computational effi­
linear or quadratic programming are ineffective [13]. Recent research in ciency, a more structured approach should be adopted through the two-
optimizing RESs has employed a variety of heuristic algorithms, each stage methodology.
choosing specific financial or technical indicators as their objective The two-stage approach has been implemented to model the opera­
functions. Genetic Algorithms have been utilized for heuristic optimi­ tion of power systems while the sizes of their individual components are
zation [14], the Non-dominated Sorting Genetic Algorithm (NSGA) decided. In the first stage, the component capacity needs to be optimized
[15], Particle Swarm Optimization (PSO) [16], and Cuckoo Search
Optimization (CSO) [17] have also been among popular choices. These
algorithms have been applied to diverse scenarios. For example, Table 1
Mohamed et al. optimized a grid-connected PV-wind system using PSO, Recent Study on the RES Capacity Optimization.
adapting effectively to system parameter changes [18]. Ali et al. Location Method Objective Year Ref.
employed PSO to segment loads in a RES, observing that prioritizing the Bangladesh NSGA-II NPC, RF 2024 [16]
higher loads increases the cost of energy [19]. Furthermore, Hybrid Cyprus PSO-GA COE, RF 2023 [27]
Optimization of Multiple Energy Resources (HOMER) has been instru­ Pakistan HOMER IRR, LCOE 2023 [39]
Jordan HOMER NPC, REF 2023 [14]
mental in simulating the economic and performance aspects of RESs
Egypt PSO COE 2022 [17]
equipped with PV and Battery Energy Storage Systems (BESS), with Simulated GA ICC 2022 [15]
studies including that of Campana et al. assessing the cost-effectiveness Morocco HOMER LCOE, Emission 2021 [40]
of Li-ion batteries in commercial buildings with PV energy [20]. Simi­ Egypt HOMER LCOE, Emission 2020 [41]
larly, Ali et al. developed a customized program to accelerate RES Pakistan HOMER NPC, COE, Emission 2020 [42]
Sanandaj MOPSO LCC, LPSP 2020 [43]

2
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

based on the pre-determined topology of the system. Then, at the second 2. Methodology
stage, simulation is performed to optimize the system performance and
to achieve the optimal operational efficiencies of major components. This section primarily presents the methodology behind developing
This method creates a strong connection between capacity setting and the two-stage optimization framework for RESs. Section 2.1 provides an
performance optimization, and it has been used to address multiple overview of the target system. Section 2.2 describes the proposed two-
objectives in terms of energy and economic benefits. Liu et al. employed stage RES capacity optimization framework. Section 2.3 outlines the
a two-stage approach to optimize a power system’s capacity first, fol­ performance metrics employed in this study for optimization and
lowed by the implementation of GA for cost minimization and reliability decision-making purposes. Lastly, Section 2.4 provides details on ca­
maximization, supported by a rule-based operational strategy [24]. pacity optimization and scheduling energy management.
Their findings indicated that incorporating multiple ESSs into the power
system can significantly enhance the system’s reliability. He et al. 2.1. System architecture
developed a two-stage optimization process, employing evolutionary
algorithms in the first stage for capacity optimization and a simplified The target system used in this study is demonstrated in Fig. 1, which
rule-based scheduling for operational efficiency in the second stage to presents a detailed schematic illustrating the interactions among its
minimize the levelized cost of energy (LCOE) in energy systems with three main components: the RES with PV arrays, the BESS, and the grid.
different ESSs [25]. Yang et al. utilized a two-stage approach to optimize The energy produced by the PV array (EPV ) is outputted to the RE system
solar-wind RES capacity, implementing NSGA-II for design parameter (EPV RE ), with surplus energy not consumed by the RES being curtailed
generation and applying an off-design model to calculate power output, (EPV Curt ). This energy, once transformed in a DC/AC inverter from
thereby minimizing the loss of power supply probability (LPSP) and Direct Current (DC) to Alternating Current (AC) at a uniform voltage,
LCOE. They highlighted the need to integrate concentrated solar power leads to an energy loss (ELoss Inverter BESS ). After conversion, the energy
plants across most regions of China for enhanced RE utilization effi­ may directly meet the energy demands of the load (EPV Load ), be reserved
ciency [26]. Zahra et al. conducted a capacity optimization for a RES, in the BESS for later usage (EPV BESS ), or be exported to the grid
incorporating a BESS and fuel cells, aimed at minimizing energy costs. (EPV Grid ) for credit. The BESS requires low-voltage DC for charging. As a
They used a hybrid PSO-GA approach for the first optimization stage and result, any AC energy intended for storage in the BESS undergoes an AC/
a rule-based energy management strategy to manage the system’s en­ DC conversion, leading to another loss (ELoss Converter BESS ). The con­
ergy flow in the second optimization stage [27]. Similarly, Al-Quraan verted energy (EChar ) is then stored in the BESS. When discharged, the
et al. investigated the optimal capacity of a standalone RES, employ­ battery energy (EDis ) experiences inherent losses (ELoss BESS ) and another
ing a two-layer model that combined a MILP-based lower layer for DC/AC inversion. EDis can then either supply the building’s energy de­
operation cost minimization and a GA-based upper layer to search for mand (EBESS Load ) or be sold to the grid (EBESS Grid ). In scenarios where
the component capacity. This approach effectively minimizes the total the RE or BESS fails to meet the energy demands, the grid supplies en­
cost and energy loss, with summer weeks demonstrating the best ergy directly to the building (EGrid Load ), or energy can be imported into
outcome [28]. the BESS (EGrid BESS ) during times of lower energy price, optimizing for
Despite the abundant research that has been performed on RES ca­ later consumption. Supply-demand fluctuations influence electricity
pacity optimization, significant gaps still exist in the literature that fail prices in the wholesale electricity market, reflecting the dynamic rela­
to address crucial aspects of their operation and integration into the tionship between demand and electricity supply. Typically, these lower
energy infrastructure. The lack of a comprehensive decision-making prices occur during late night to early morning hours, a period charac­
framework that includes these external constraints and stakeholder terized by a substantial decrease in energy demand relative to supply.
preferences indicates a need for more complex research approaches. This strategic timing allows for the importing of electricity at reduced
Moreover, current studies rely heavily on numerical simulations and costs, optimizing the system for later consumption while ensuring eco­
rule-based energy management strategies, overlooking the benefits of nomic efficiency.
advanced optimization techniques that could enhance system perfor­
mance and efficiency by better addressing regional variability and grid 2.2. Framework overview
interaction complexities.
This paper introduces a novel optimization approach that effectively The concept of the two-stage RES capacity optimization framework
addresses the scalability and adaptability issues prevalent in RES ca­ deployed in this study is demonstrated in Fig. 2. The proposed frame­
pacity optimization, while extending the proven retail-level methodol­ work comprises a system capacity optimization stage on the left side,
ogies to the wholesale domain to enable effective interactions with serving as the primary optimization, and a year-long energy manage­
large-scale grid operations. By capturing the economic interaction be­ ment simulation on the right side, which is used as the secondary opti­
tween a RES and the grid, this modelling approach incorporates the local mization. System capacity optimization is based on NSGA-II and tackles
regulations and wholesale electricity prices into the analysis, refining the computational challenges encountered due to the extensive search­
energy management practices and offering a more realistic design of ing tasks required for favorable component sizes. In Phase one, the
energy systems. Additionally, a modified NSGA-II algorithm integrated primary optimization randomly initializes a group of candidate solu­
with a mixed-integer linear programming (MILP) strategy enhances the tions, containing specific sizes for PV panels and a BESS. These sizes
optimization process. This integration features a dynamic population serve as system parameters, directly influencing the renewable energy
decay (DPD) mechanism designed to accelerate the optimization process generation profiles and energy storage ability used in the energy man­
by efficiently narrowing down the search space. Moreover, the decision- agement simulation. The second phase involves preparing data,
making framework utilizes the multi-objective optimization via the including an enhanced load profile derived from aggregating and
Pareto-adaptive Epsilon dominance algorithm, enabling the custom­ introducing volatility to individual building profiles, making the simu­
ization of energy solutions to meet specific user requirements and tailor lation closer to a real-world scenario. Details on this enhancement are
energy management strategies to individual needs. This adaptation clarified further in a later section. The third phase formulates MILP
significantly enhances the management of power and information flows, problems based on these inputs. During the fourth phase, the MILP
improving efficiency across different scales, and bridges the gap be­ solver optimizes the system using predefined decision variables, objec­
tween retail and wholesale systems, showcasing a significant advance­ tive functions, and constraints. Meanwhile, a BESS aging model simul­
ment in grid interactions. taneously updates the BESS’s available capacity after aging and the state
of the BESS. The fifth phase evaluates the quality of each candidate
solution through calculated indicators and is used as fitness measures

3
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Fig. 1. Schematic of the target system with key variables.

Fig. 2. Framework of the proposed two-stage optimization approach. Note: The center image was created with the assistance of DALL⋅E 2.

based on simulation outcomes. In the last phase, the top-performing peak demand times. This strategy optimizes energy consumption for cost
candidates are selected as parents to form the basis for the next gener­ savings and supports grid stability, exemplifying innovative, cost-
ation of candidate solutions through crossover and mutation processes. effective, and sustainable energy management solutions.
Through this iterative loop, the candidate solutions progressively
evolve, converging towards the optimal solution and achieving the 2.3. Performance metrics
study’s aim of efficient system capacity optimization.
Building upon the two-stage RES capacity optimization framework This study utilizes a set of economic and technical indicators to assess
outlined above, the integration of sophisticated scheduling optimization system performance. The economic indicators are Internal Rate of Re­
within the MILP-based energy management framework plays a crucial turn (IRR), Net Present Value (NPV), and initial investment (Inv). The
role. The proposed system conducts load-shifting and energy arbitrage technical indicators include Energy Independence Score (EIS), Grid
by utilizing 24-hour energy prices and demand forecasts, effectively Energy Interaction Factor (GEIF), Utilization Factor (UF), and Satisfied
managing resources based on predicted market conditions. Active en­ Load Fraction (SLF). IRR and EIS are chosen as the primary optimization
ergy management ensures efficient energy use by charging the BESS objectives for their system-level relevance, with IRR focusing on prof­
during low-price periods and discharging it during high-price periods or itability and EIS on energy independence. The other indicators,

4
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

including NPV, initial investment, GEIF, UF, and SLF, are used to eval­
2.3.2. Technical indicators
uate the performance of the optimized results, providing insights into
The Energy Independence Score (EIS) is a technical indicator used to
financial viability and component-level efficiency.
measure and quantify the degree of energy independence or self-
sufficiency of a RES. The EIS is calculated through Eqn. (10) based on
2.3.1. Economic indicators
the proportion of the total load energy demand that is supplied by the
The economic indicator IRR is used to evaluate financial perfor­
system itself, as opposed to being imported from the grid. The total
mance. It measures a project’s return and is compared against an in­
energy supplied by the system is the difference between ELoad and EIm .
vestor’s expected return, usually close to the investor’s capital cost. Even
The system will reach its upper boundary of EIS when all generated
though the primary limitation of IRR is its lack of scale of investment
renewable energy is utilized. Therefore, the EIS can also indirectly
consideration, it can be supplemented by using IRR along with the NPV
reflect the carbon emission rate and efficiency of the RES.
and initial investment metrics. Therefore, the primary optimization will
use IRR as one of the optimization objectives and use NPV and the ∑T (t)
E
required initial investment to aid decision-making after optimization. EIS = 1 − ∑Tt=1 (t)Im (10)
E
t=1 Load
IRR is the rate at which the NPV of a project equals zero. The NPV
calculation for the proposed RES considers cash inflows, outflows, and Besides the EIS, this study also employs three additional technical
initial investment. The discounted cash flow method determines the indicators [30] to evaluate the performance of the optimized system:
present value of future cash flows. Cash inflows include energy bill GEIF, UF, and SLF. GEIF (Eqn. (11) measures the system’s grid reliance,
savings (RESav ) and revenue from energy sold back to the grid (RESol ). indicating energy surplus or deficit management efficiency. UF (Eqn.
RESav is calculated hourly by subtracting energy imported from the grid (12) assesses renewable energy utilization effectiveness in meeting load
(EIm ) from energy demand (ELoad ), then multiplying by the import energy demand. SLF (Eqn. (13) evaluates the proportion of demand met by
price (πIm ). It is summed up yearly. RESol is the yearly sum of hourly renewable sources, reflecting the system’s sustainability and clean en­
revenues, calculated by multiplying exported energy to the grid (EEx ) by ergy dependence. While EIS focuses on grid independence, GEIF, UF,
the export energy price (πEx ). Both EEx and EIm are determined hourly and SLF provide insights into the system’s operational efficiency,
through the secondary optimization model. The 18-year optimization renewable energy use, and sustainability.
horizon is chosen in this study based on the expected payback period
EPV Grid + EBESS Grid+ EGrid BESS + EGrid Load
[29] for PV panels and their degradation rate, ensuring a realistic GEIF = (11)
ELoad
assessment of financial viability and operational efficiency.
( ) EPV Load
∑y=18 RESav + RESol − CO&M − CRep
(y) (y) (y) (y)
UF = (12)
NPV = − Inv, r = IRR if NPV = 0 EPV BESS + EPV + EPV
Grid Load + EPV Curt
y=1
(1 + r)y− 1
(1) EPV Load
SLF = (13)
ELoad
∑t=8760 ( (t) )
R(y)
ESav = ELoad − E(t)
Im × π(t)
Im (2)
t=1
2.4. Two-Stage optimization
∑t=8760
R(y)
ESol = t=1
E(t) (t)
Ex × πEx (3) This section details capacity optimization and scheduling energy
management, offering a comprehensive insight into each stage. Addi­
The outflow cash consists of the operation and maintenance costs
tionally, it explains the interrelationship between the optimization
(CO&M ) and replacement costs (CRep ). The CO&M shown in Eqn. (4) in­
stages by elucidating how variables influence each other throughout the
cludes three parts: the operation and maintenance costs of PV, BESS, and
optimization process.
other components of the system such as inverters and converters.
Therefore, Eqn. (4) to Eqn. (6) shows the details of how to calculate
2.4.1. Capacity optimization
these costs. The calculation approaches used in Eqn. (5) to Eqn. (7) are
Primary optimization is formulated as a multi-objective optimization
identical for these three equations: the CO&M is a proportion (ωO&M ) of the
problem, integrated with an energy management algorithm, to utilize
initial capital cost of the relative component (capacity of component Cap
the system’s component capacities efficiently. Equations of the primary
multiplied by unit capacity cost πCap ). While CRep in Eqn. (8), which ac­
optimization are formulated to determine the optimal sizes for PV arrays
count for the cost of replacing the battery, are adjusted by the battery
and BESSs. Eqn. (14) defines the objective function which aims to
price and discounted at the battery discount rate (rBESS ) based on his­
maximize two metrics: IRR and EIS. These objective functions are sub­
torical data.
ject to various equality and inequality constraints.
C(y) The first two constraints in Eqn. (14) indicate that both IRR and EIS
O&M = CO&M,PV + CO&M,BESS + CO&M,Others (4)
(y) (y) (y)

are functions of decision variables: capacities of PV (CapPV ) and BESS

(CapBESS ). The third equation sets user-specific budget constraints for
C(y)
O&M,PV = CapPV × πCap,PV × ωO&M,PV (5)
initial investments, while the fourth equation defines thresholds on the
CapPV user energy consumption profile. As shown in Table 2, these
C(y)
O&M,BESS = CapBESS × πCap,BESS × ωO&M,BESS (6)
constraint values are estimated based on the past user’s load profile.
Finally, the last equation indicates that CapBESS exceeding CapPV is not
C(y)
O&M,Others = CapOthers × πCap,Others × ωO&M,Others (7) cost-effective due to diminishing returns, as supported by [31].

πCap,BESS
C(y)
Rep = CapBESS × (8)
(1 + rBESS )y− 1

The initial investment of the system includes the investment in PV,

BESS, and others relatively, calculated in Eqn. (9) by multiplying the
capacity of each component with the relative unit price.
Inv = CapPV × πCap,PV + CapBESS × πCap,BESS + CapOthers × πCap,Others (9)

5
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Table 2 Concurrently, the DPD mechanism operates alongside genetic oper­

Parameter Value. ations. The DPD algorithm assesses the variance in fitness scores across
Variable Name Value Unit Source the population to adjust the population size dynamically. By evaluating
the distribution of fitness scores, the DPD mechanism determines the
πCap,PV 1300.0 kWh/USD From literature [44]
πCap,BESS 150.0 kWh/USD Bloomberg [45]
need for population size adjustments—reducing the population size
ωO&M,BESS 0.2 % From literature [31] when a lower variance is observed. This indicates that the population is
ωO&M,Others 1.0 % From literature [31] converging towards similar fitness values, prompting the DPD to
ωO&M,PV 1.0 % From literature [31] decrease the population size to focus the search on the most promising
rBESS 4.0 % Bloomberg [45]
solutions. This strategy of dynamically adjusting the population size,
r 2.0 % Federal Reserve [46]
CapPV max 1000.0 kWh −
based on the fitness variance, facilitates an optimal balance between
USD/EUR 1.07 − Exchange Rate exploring new solutions and exploiting existing ones, continuously
Crossover rate 0.7 − Experiment refining the solutions until the maximum generation is reached. The
Population 20.0 − Experiment ‘NSGA-II History’ is instrumental throughout this iterative process,
Mutation rate 0.1 Experiment
providing ongoing insights that guide the exploration towards achieving
−
Generation 12.0 − Experiment
the study’s objective of efficient system capacity optimization and
management. This comprehensive approach, integrating multi-objective
max IRR, EIS selection with the DPD mechanism and a thoughtfully randomized
initial population, ensures a thorough exploration of potential solutions
s.t. IRR = f 1 (CapPV , CapBESS )
in alignment with the study’s goal.
EIS = f 2 (CapPV , CapBESS ) The final stage of multi-objective optimization is identifying optimal
Inv ≤ InvReq (14) solutions with the Pareto frontier, as demonstrated in Fig. 3 (b) and
detailed in [32]. This frontier includes optimal solutions where each
CapPV ≤ CapPV max
point is a Pareto optimal. The diagram categorizes solutions into sub-
CapBESS ≤ CapPV optimal solutions (green region), the Pareto frontier, and the Infea­
sible Region. Points 1 and 3, highlighted in orange, illustrate the concept
of Pareto efficiency. Even though the fitness of point 2 equals points 1
In the two-stage optimization framework detailed earlier, this study and 3 in one objective, it is excluded from the frontier because it is
employs NSGA-II for the primary optimization task demonstrated in inferior to another objective.
Fig. 3 (a). The process begins with the initialization of the population, a
critical step where individuals—each representing a potential sol­ 2.4.2. Energy management
ution—are generated randomly following a normal distribution. This The secondary optimization aims to calculate optimal time domain
method of initialization ensures a diverse and representative starting variables, supports the calculation of IRR and EIS, and provides an
point for the energy management system, and it facilitates the effective energy management strategy for RES simulation. As the sec­
comprehensive fitness evaluation for each candidate solution. During ondary optimization in this study, its primary objective is to minimize
optimization, populations are selected for multi-objective evolution the energy cost of the system. To achieve this, the secondary optimiza­
through fitness calculations based on Pareto efficiency and diversity. tion algorithm for energy management has been formulated as a single-
This selection process is driven by a crossover operation that averages objective optimization problem based on Pena-Bello et al.’s work [33].
the capacities of parent solutions, promoting the generation of balanced The secondary optimization is defined by a series of mathematical
offspring and facilitating the effective evolution of the population to­ equations, ranging from 15 to 18. The objective function is subject to
wards optimal solutions. constraints in four categories: basic (Eqn. (16), energy balance (Eqn.

Fig. 3. Algorithm. (a) Algorithm flowchart of the proposed two-stage optimization approach; (b) Concept of Pareto front in multi-objective optimization.

6
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

(17), loss balance (Eqn. (18), and power balance (Eqn. (19), ensuring ELoss = (EGrid + EPV × ηConverter × (1 − ηBESS )
BESS BESS BESS ) BESS
optimal time-domain variable calculation.
Eqn. (15) defines the objective function to minimize the electricity Eqn. (19) demonstrates the power balance constraints. These con­
bill. This is achieved through a balancing act between importing and straints guarantee that the power remains within the physical limits of
exporting energy. Specifically, when energy is imported from the grid at the system components.
import cost, it incurs a cost. Conversely, exporting surplus energy to the
PConverter BESS × dt ≥ EPV BESS + EGrid BESS
grid at the export price at times earns credit, reducing electricity costs.
Therefore, the final electricity bill is computed by the difference be­ PInverter × dt ≥ EBESS + EBESS (19)
BESS Load Grid
tween cost and credit. The πEx used in this study is based on the
(t)

wholesale price from the futures market [34]. And the πIm calculated as a
(t) EPV Load + EPV BESS + EPV Grid
PInverter PV × dt ≥
ηInverter PV
premium added on the Ex ,
π(t) following the approach used in literature
[31]. In addition to the MILP, the energy management algorithm in­
tegrates a BESS aging model following methodologies from Holger et al.
t=8760
∑
min E(t) [35] and Angenendt et al. [36]. This model updates the BESS’s state
Im × πIm − EEx × πEx (15)
(t) (t) (t)

t=0 based on each time step’s optimization outcomes, assessing degradation

by comparing calendar and cycle life, and calculating State of Health
Eqn. (16) outlines the basic constraints, ensuring the system will not
(SOH). BESS replacement is scheduled in the RES when SOH falls below
concurrently charge (C har) and discharge (Dis) the BESS, as well as
70 %, accounting for replacement costs and affecting BESS capacity and
simultaneously import and export energy. In these equations, the terms
performance in charging and discharging.
EBESS max and EGrid max represent the rated power of the BESS and the
The Aging Factor FBESS age of the BESS is calculated as described by
grid, respectively, while B denotes a binary integer that enforces these
Eqn. (20), by taking the maximum between the cycle aging factor and
constraints.
the calendar aging factor. The cycle aging factor is represented as the
− EBESS max × BDis ≤ EDis ≤ EBESS max × (1 − BChar ) inverse of the exponential function of the logarithm of the Depth of
Discharge (DOD) divided by − 2.168. This expression accounts for cycle
− EBESS max × BChar ≤ EChar ≤ EBESS max × (1 − BDis ) aging based on the DOD. The calendar aging factor is the inverse of the
product of the battery’s calendar life LBESS cal , 24 h, and 365 days, which
BChar +BDis = 1 (16) computes the calendar aging considering the total operational hours in a
year. The use of the maximum function ensures that the most significant
− EGrid × BEx ≤ EEx ≤ EGrid × (1 − BIm )
max max
aging effect assumed, whether cycle or calendar, is considered in the
model.
− EGrid max × BIm ≤ EIm ≤ EGrid max × (1 − BEx )
⎛ ⎞
BIm +BEx = 1 ⎜ 1 1 ⎟
FBESS age = max⎜
⎝elog(DOD) ,
⎟ (20)
LBESS × 24 × 365 ⎠
Eqn. (17) represents energy balance constraints. These equations − 2.168 cal

ensure the energy balance across system elements such as PV, BESS,
load, and grid while considering the energy loss. Additionally, they The aged capacity of the battery is computed by reducing the original
restrict the battery’s state of charge (SOC) from exceeding its aged ca­ capacity CapBESS based on the FBESS age and the SOH, as shown in Eqn.
pacity (CapBESS aged ). (21).
( ( ))
EPV = EPV Load + EPV BESS + EPV Grid + EPV Curt + Eloss Inverter PV CapBESS_aged = CapBESS × 1 − 0.3 × 1 − SOH + FBESS age (21)

ELoad = EPV + EBESS × ηInverter + EGrid Finally, the SOH is updated by Eqn. (22) based on FBESS age and
Load Load BESS Load
CapBESS_aged , ensuring that the model accurately reflects the current
EIm = EGrid BESS + EGrid Load health and operational efficiency of the BESS.

1 CapBESS_aged 7
EEx = EPV Grid + EBESS Grid × ηInverter BESS SOH = × − (22)
0.3 CapBESS 3
EChar = EPV BESS + EGrid BESS − ELoss Converter BESS (17)
2.4.3. Interrelation of decision variables and objectives
EChar ≤ SOCmax × CapBESS − SOCt− 1 × CapBESS The IRR and EIS serve as economic and technical indicators,
respectively, as mentioned in the primary optimization section. The
EDis = EBESS Load + EBESS Grid decision variables of optimization are closely related to these indicators,
with their complex relationships demonstrated in Fig. 4. The CapPV
EDis ≤ SOCt− 1 × CapBESS − SOCmin × CapBESS determines the volume of renewable energy EPV available, whereas the
CapBESS sets the upper boundary for the system’s charging and dis­
SOCt × CapBESS aged = SOCt− 1 × CapBESS aged + EChar − EDis − ELoss BESS charging capabilities, influencing both EChar and EDis . These capacities,
Eqn. (18) introduces the loss balance constraints. These equivalent in turn, have chain effects on subsequent parameters within the energy
constraints are employed to compute energy losses specifically in the management phase, ultimately impacting the aggregate energy im­
inverter, converter, and BESS. ported to and exported from the grid. Subsequently, these parameters
directly influence the technical indicator, EIS, and an indirect effect on
ELoss Converter BESS = (EPV BESS + EGrid BESS ) × (1 − ηConverter BESS ) the economic indicator, IRR, through the cost structures and revenue
streams generated by the corresponding system.
Eloss Inverter BESS = (EBESS Load + EBESS Grid ) × (1 − ηInverter BESS ) (18)
3. Result and discussion
1 − ηInverter
Eloss Inverter PV = (EPV Load + EPV BESS + EPV Grid ) × PV
ηInverter PV To validate the proposed system and methodology, a case study has

7
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Fig. 4. Relationship among variables and optimization objectives.

Fig. 5. Energy demand profiles based on the original dataset. (a) Monthly energy demand profile: detailed consumption analysis by equipment for the building; (b)
Annual energy demand profile: proportional analysis by equipment; (c) Daily energy demand profile: zoomed-in view; (d) Hourly energy demand profile: zoomed-in
view; (e) Annual hourly energy demand profile. Note: The simulation assumed the existence of five identical buildings and added white noise. As a result, the energy
demand profile used in the simulation is approximately five times the energy represented in this figure.

8
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

been conducted on a commercial building in the Netherlands, utilizing Wind velocity is consistent year-round, but solar energy is more domi­
real-world data rather than simulated inputs. The data used in this study nant near the building. Temperatures are within a proper range for PV
has two sets: the energy consumption profile of a commercial building panel efficiency [38], and atmospheric pressure remains steady
located in the Netherlands and the corresponding meteorological pro­ throughout the year.
file. The energy consumption profile is crucial and will be directly used In the data preprocessing stage, the primary goal was to improve the
for optimization as the user’s energy consumption after enhancement. accuracy and credibility of the dataset. To achieve this, all missing data
At the same time, meteorological data is employed to estimate renew­ were filled using an interpolation approach. Additional refinement
able energy outputs. The existing literature has collected and validated procedures included identifying and removing outliers, which refer to
the data [37]. those beyond the 99.999th percentile, to maintain data integrity without
Fig. 5 illustrates the energy demand profile of a commercial building compromising its continuity. Finally, the original data, logged on a
in the Netherlands, which is essential for system capacity optimization minute scale, was resampled to an hourly scale. This adjustment was
and energy management simulations in this project. Fig. 5 (a) shows crucial to aligning the data’s resolution with the optimization time ho­
monthly energy consumption by equipment, with four main systems: rizon. As mentioned in the framework section, the target system was
AHU (Air Handling Units), chiller, humidifier, lights, and others. AHU, expanded and scaled up to include five similar buildings. This expansion
lighting, and other demands are consistent year-round. At the same operated under the assumption of consistent energy consumption pat­
time, chillers and humidifiers show seasonal patterns, such as higher terns across these buildings. To capture the variable nature of real-world
chiller use in summer and increased humidifier use in winter due to energy consumption, independent and identically distributed white
temperature-related humidity changes. Fig. 5 (b) displays annual energy noise was incorporated into each building’s data. This study adopted the
demand proportions, with lighting and others being the largest at 47.9 white noise due to its uncorrelated nature and constant variance. The
%, followed by AHU at 38.1 %. Chillers and humidifiers contribute less, amplitude of the white noise was carefully calibrated to be 10 % of the
at 8.1 % and 5.8 % respectively. Fig. 5 (c) and (d) indicate higher dataset’s standard deviation to maintain the balance between simulating
workday usage, peaking during daytime hours and decreasing at night. real-world variances and preserving the integrity of the underlying data
Lastly, Fig. 5 (e) provides a broader view of energy trends. patterns. With the introduction of volatility in the consumption data, the
Fig. 6 offers a comprehensive insight into the meteorological con­ optimization result is more robust and adaptive to the changing and
ditions crucial for renewable energy sources at the nearby weather unpredictable real-world scenario.
station of the building. Fig. 6 (a) shows the monthly solar energy profile
with Global Horizontal Irradiation (GHI), Direct Normal Irradiation
(DNI), and Diffuse Horizontal Irradiation (DIF), highlighting increased 3.1. Optimization on capacity of major components
solar energy from April to August. The data suggest clearer days from
May to July. Box plots in Fig. 6 (b)–(g) display monthly variations in Fig. 7 (a)–(e) offers a detailed visualization of NSGA-II’s optimiza­
GHI, DNI, DIF, wind velocity, temperature, and atmospheric pressure. tion of IRR over five epochs and twelve generations each. The shrinking
distribution of points indicates the NSGA-II’s refinement and

Fig. 6. Meteorological resource overview from the original dataset. (a) Monthly solar energy profile: breakdown of GHI, DNI, and DIF (daytime data only); (b)–(g)
Monthly distributions derived from hourly measurements of: (b) global horizontal irradiation (c) direct normal irradiation (d) diffuse horizontal irradiation (e) wind
speed (f) temperature (g) atmospheric pressure.

9
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Fig. 7. Evolution of NSGA-II Search for IRR Optimization (a) NSGA-II Search Epoch 1: Optimal IRR Achievement (b)–(e) NSGA-II Search Epoch 2–5.

convergence to optimal solutions. Fig. 7 (a) highlights the epoch with IRR, and orange points for EIS. Black diamonds, numbered 1–9, mark
the best fitness. The figure also includes blue ellipses representing the nine configurations lying on the Pareto frontier.
algorithm’s exploration space, derived from the covariance matrix’s The Pareto Frontier in RES optimization highlights a crucial trade-off
eigenvalues and eigenvectors, illustrating search variance and direction. between energy independence and economic returns. A strong negative
Fig. 8 similarly explores NSGA-II’s optimization for EIS; the only dif­ correlation between IRR and technical indicators like EIS, GEIF, and
ference lies in their respective convergence areas. These results validate SLF, as shown in Table 4, confirms this trade-off. The capacity and
the experimental design choice of five epochs and twelve generations, as performance of those nine configurations are detailed in Table 3. The
significant convergence was observed after the eighth generation. correlation between the economic indicator IRR and other technical
A detailed view of fitness across generations for each epoch is pre­ indicators is shown in Table 4.
sented in Fig. 9. Fig. 9 (a) shows the objective IRR and Fig. 9 (b) focuses Moreover, Fig. 10 (b) reveals how initial investment and EIS are
on the objective EIS. Orange lines demonstrate the fitness trends of in­ related. Radar plots (c)–(k) illustrate normalized performance metrics
dividual epochs, with a dashed black line marking the average fitness for these configurations, providing a complete view of each
and a solid black line highlighting the epoch with the best fitness. An configuration.
upward fitness trend stabilizes after 6–8 generations, indicating NSGA-
II’s improvement and convergence. Minor fluctuations in NSGA-II’s
performance are caused by its mutation process, providing insights into 3.2. Decision framework
the algorithm’s exploratory behavior within the search process.
After optimization, the Pareto-adaptive Epsilon-dominance algo­ As demonstrated by the analysis in Fig. 10, the decision framework
rithm was used to select points from the optimization results. Given the guides decision-makers in adopting a structured approach to making
real-world nuances where the objective values of two points might not informed choices. This approach integrates a mathematical-based de­
be the same, a small value, epsilon, was introduced as a tolerance. cision process with accommodating unique preferences, laying the
Fig. 10 (a) visualizes this process and the Pareto frontier. Points from all groundwork for a comprehensive decision-making strategy.
past generations have been plotted, not just those from the last gener­ At the center of this framework is the mathematical-based decision
ation, ensuring that potential Pareto optimal solutions from earlier process, which encompasses regression analysis, marginal analysis, and
generations are not omitted. Blue points show NSGA-II optimizations for radar plot assessments to enable objective decision-making. Initially, the
application of a fourth-degree polynomial regression, illustrated in

10
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Fig. 8. Evolution of NSGA-II Search for EIS Optimization (a) NSGA-II Search Epoch 1: Optimal EIS Achievement (b)–(e) NSGA-II Search Epoch 2–5.

Fig. 9. NSGA-II Convergence: Fitness Across Generations and Epochs (a) IRR and (b) EIS.

Fig. 10 (b), models the relationship between energy independence and independence level of the RES based on their budgetary constraints.
financial performance. Eqn. 23 provides the mathematical expression of Further, the derivative of the regression curve indicates the marginal
the regression model, allowing decision-makers to estimate the energy cost of increasing the EIS per unit of initial investment, highlighting a

11
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Fig. 10. Results of optimization: Pareto front and additional indicators. (a) Epsilon Pareto front in multi-objective optimization; (b) Achievable EIS by given initial
investment; (c)–(k) Comparative evaluation of normalized performance metrics for configurations selected from Pareto front, indexed by 1–9.

Table 3
Pareto-optimal Configurations and Indicators.
Confg. 1 2 3 4 5 6 7 8 9

CapPV 75 120 140 180 225 310 385 515 660

CapBESS 35 60 80 110 145 175 225 245 340
IRR (%) 4.86 4.43 4.22 3.39 2.25 0.46 − 0.73 − 2.38 − 3.75
Diff(%) − − 8.95 − 4.73 − 19.55 –33.73 − 79.45 − 258.2 − 226.1 − 57.15
EIS 0.2 0.29 0.33 0.39 0.44 0.51 0.55 0.59 0.63
Diff(%) − 22.39 18.61 7.46 5.75 3.12 3.23 2.65 2.67
GEIF 0.89 0.97 1.0 1.14 1.28 1.57 1.88 2.35 2.9
Diff(%) − 16.7 15.87 16.98 19.35 12.31 26.25 28.31 40.4
UF 0.83 0.8 0.78 0.73 0.69 0.62 0.56 0.5 0.44
Diff(%) − − 6.79 − 9.24 − 5.24 − 6.51 − 2.74 − 4.89 − 3.46 − 4.84
SLF 0.18 0.23 0.25 0.27 0.28 0.29 0.29 0.3 0.3
Diff(%) − 12.13 7.52 2.25 − 0.5 1.33 0.11 0.72 − 0.55

critical threshold at an EIS of 0.4. Before the identified threshold, min­ allocations against the expected environmental returns.
imal financial sacrifices are shown to result in significant gains in energy
x = 43414.09y4 − 53911.59y3 + 25515.87y2 − 4590.18y + 350.77
independence. The marginal information supports optimizing the
techno-economic performance within RES by balancing budget (24)

12
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Table 4 Through the structured and integrated approach of the framework,

Correlation Matrix of Indicators. decision-makers can achieve a balanced consideration of technical,
IRR EIS GEIF UF SLF economic, and environmental factors, ensuring that final decisions are
both objective and aligned with unique strategic goals.
1.0 − 0.96 − 0.98 1.0 − 0.8
− 0.96 1.0 0.91 − 0.97 0.94 Given a scenario without RES subsidies, stakeholders require a pos­
− 0.98 0.91 1.0 − 0.98 0.72 itive return on investment and expect a balanced system performance.
1.0 − 0.97 − 0.98 1.0 − 0.83 Configuration 4 is selected as the optimal choice for four reasons. First,
− 0.8 0.94 0.72 − 0.83 1.0 Fig. 10 (f) shows the most significant area compared to others in Fig. 10
(c)-(k), indicating the most balanced techno-economic performance
Moreover, the analysis along the Pareto Frontier, as shown in Fig. 10 among the nine configurations. Second, configuration 4 is close to the
(a), reveals that the marginal cost of improving EIS via the Internal Rate threshold where the marginal cost of EIS significantly increases. Third, it
of Return (IRR) surges by 839 %. This underscores the strategic yields a positive IRR and NPV, generating profit while maintaining a
importance of identifying the optimal balance between energy inde­ decent EIS. Fourth, the initial investment is reasonable, considering its
pendence and economic returns during decision-making. This balance is capacity to supply half the energy demands of five commercial buildings
crucial to avoid disproportionate sacrifices when increasing energy in­ for 18 years.
dependence in RES. Furthermore, integrating radar plot analysis into
this framework provides a multifaceted view of RES performance, 3.3. Energy management simulation
emphasizing systems’ efficiency, cost-effectiveness, and environmental
impact. Notably, the shape and balance of a radar plot provide insights Fig. 11 (a)–(d) demonstrates the detailed energy management
into a system’s all-around performance, highlighting well-rounded op­ simulation outcomes for configuration 4, examining the temporal dy­
tions that meet a broad spectrum of project goals. namics of energy distribution. Fig. 11(a) shows an hourly view of the
In addition to providing a comprehensive performance overview, energy supply mix from PV, BESS, and the grid, while Fig. 11(b) offers
radar plot analysis uniquely supports aligning RES choices with specific an annual overview of the energy mix. Weekly variations in energy
preferences and strategic objectives. An unbalanced radar plot can sourcing are described in Fig. 11(c), and Fig. 11(d) presents monthly
highlight outstanding systems in particular areas, meeting unique variations of the energy mix.
project requirements or priorities. This analysis, coupled with the From Fig. 11, it is evident that the PV system serves as the primary
overall area within the radar plot, quantifies system performance, energy supplier during the day. In the early evening, the BESS plays a
facilitating objective comparisons. Such detailed assessments ensure crucial role in supplying energy by discharging stored energy, which is
decision-makers can identify systems that meet broad strategic objec­ particularly cost-effective during high electricity prices in the evening.
tives and sustainability goals and adhere to specific preferences or On weekends, the combined capacity of these two systems is generally
constraints. sufficient to meet the lower energy demands of commercial buildings.
This decision framework offers comprehensive support to human Seasonal variations also significantly influence the system’s reliance,
experts, enabling them to make informed and strategic final decisions. with a greater dependency on solar energy during summer due to

Fig. 11. Simulation results for configuration 4: load demand energy supply breakdown. (a) Hourly energy supply dynamics: PV, BESS, Grid (zoomed-in view); (b)
Yearly energy supply percentages; (c) Weekly energy supply percentages; (d) Monthly energy supply percentages.

13
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

abundant sunlight and a shift towards grid dependency during winter. high, it prefers exporting energy to the grid. This observation again
Fig. 12 shows the RE utilization within a RES, including interactions highlights the importance of optimizing PV and BESS capacity and
with the BESS, load, and grid. Fig. 12(a) shows a detailed hourly RE effective energy management.
distribution, focusing on immediate load requirements during daylight,
surplus energy storage, and grid exportation, with a noted decrease at
3.4. Comparative analysis
night. Due to decreased load demand on weekends, a large proportion of
RE is stored at the BESS and exported to the grid. This exportation is
The comparative analysis in Fig. 13 indicates that expanding the
feasible while adhering to regulatory limits on grid sell-back, balanced
system’s capacity does not proportionally enhance energy efficiency,
by the system’s PV and BESS capacities. Fig. 12(b) provides weekly data,
and the system’s primary goal is self-sufficiency rather than selling
showing successful grid exportation for most of the year, while Fig. 12(c)
electricity. This conclusion is drawn from the following observations: as
highlights monthly variations in energy production and grid
the system size expands, the utilization of RE sources and dependence on
dependency.
the grid increase logarithmically, as shown in Fig. 13 (a), while the
Observations from Fig. 12(a) indicate that during weekdays, the
proportion of RE stored in the BESS remains constant and the supply to
energy generated by PV panels is primarily used to meet the energy
buildings decreases linearly, as indicated in Fig. 13 (b). This suggests a
demands of buildings, followed by charging the battery system and
diminishing return on energy efficiency as system size increases. A
selling excess energy to the grid. On weekends, despite a reduction in
detailed exploration of the dynamics reveals that larger systems, though
energy demand, the generation of RE remains significant, leading to
being capable of a greater energy generation, do not proportionally in­
most of this power being sold to the grid. Fig. 12(b) shows that the
crease the direct energy supply to buildings. Instead, the excess energy,
proportion of RE sold to the grid remains relatively stable, most of the
which could potentially be sold to the grid or stored, leads to a scenario
time ranging between 20 % and 25 %. This finding underscores the
in which the system’s capability to enhance itself sufficiency without
critical role of the battery system as a buffer between the solar panels
making external energy purchase is undermined. In addition, observa­
and the grid. However, Fig. 12(c) shows that approximately 14 % of the
tions from Fig. 13 offer evidence to support the idea that RESs should
electricity is directed to the battery system, which differs from the
prioritize supplying the renewable energy to users over profiting from
previously analyzed figure of 12 %. This discrepancy is primarily due to
the sale of surplus energy.
efficiency losses incurred during the energy migration to the BESS. This
Fig. 14 (a) and (b) present the average monthly utilization rate of the
observation aligns with the initial conclusion, suggesting that the
BESS and the State of Health (SOH) of the BESS, respectively. In smaller
generated RE is typically utilized immediately, thereby providing
systems, where the PV capacity is limited, the BESS shows lower utili­
additional evidence to support this conclusion. The preference for
zation rates, as demonstrated in Fig. 14(a). This underutilization can be
directly utilizing RE to avoid efficiency losses is evident. Yet, the deci­
attributed to the minimal surplus energy generation, particularly under
sion to store or export energy to the grid depends on current and future
conditions such as an increased cloud cover that further reduces output.
energy prices and protentional cost related to BESS charging, such as
This operational characteristic is beneficial from a lifecycle perspective
efficiency loss and degradation. Consequently, when energy prices are
of the battery system, as lesser cycling of the BESS extends its SOH, as
low, the system stores energy in the BESS; conversely, when prices are
indicated in Fig. 14(b). Conversely, larger systems expedite BESS

Fig. 12. Energy management simulation results for configurations 4: renewable energy utilization. (a) Hourly renewable energy utilization dynamics: PV, BESS, Grid
(zoomed-in view); (b) Weekly utilization percentages; (c) Monthly utilization percentages.

14
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Fig. 13. Comparison of technical performance of Pareto-optimal configurations. (a) Load demand energy supply and (b) RE utilization.

Fig. 14. Energy management simulation results for Pareto-optimal configurations: BESS utilization. (a) Monthly BESS utilization rate for optimized system
configuration; (b) Monthly BESS SOH for optimized system configuration.

degradation due to increased charge and discharge cycles necessitated Fig. 15 (a) and (b) demonstrate the variability of the GEIF across
by higher energy outputs. Such insights emphasize the need for a different system configurations, underscoring these larger systems,
balanced approach in system design, prioritizing direct energy con­ despite their increased interaction with the grid, are less cost-effective
sumption and minimal battery usage to optimize both economic and than smaller systems. The primary reason for this is that in configura­
environmental benefits. tions 1 to 4, most of the generated electricity is utilized to meet the
In contrast, BESSs in smaller PV systems are often underutilized. energy demands of buildings, resulting in lower GEIF values because a
Storing and retrieving energy leads to losses, making direct RE con­ smaller proportion is exported to the grid. Conversely, larger systems
sumption more efficient in systems with small PV capacity. Minimizing produce more surplus electricity, leading to more significant grid in­
battery use in such systems extends battery life and aligns with economic teractions. Nevertheless, these larger configurations require substantial
considerations, especially when energy imports from the grid are expansion in BESS capacity. Constructing such systems primarily to sell
affordable. This strategy highlights the need for system design optimi­ excess electricity to the grid without subsidies proves to be economically
zation, balancing energy storage, direct usage, and overall efficiency. infeasible. Nonetheless, these findings support the initial conclusions
Fig. 15 evaluates the effectiveness of the optimized RES, showing that the generated RE tends to be consumed immediately. For instance,
how the capacity of system components affects grid interaction and RE in comparison to configuration 1, configuration 9 exhibits an 880 %
utilization. An increase in PV capacity changes the energy distribution increase in PV capacity, yet the corresponding rise in GEIF is less than
among the grid, BESS, and load. Fig. 15 (a) and (b) display the vari­ 300 %. This highlights that while larger systems can supply more elec­
ability of the GEIF with different system sizes. Fig. 15 (c) and (d) show tricity, the negative IRR indicates that they are economically unfeasible
that smaller systems use generated energy more efficiently. Addition­ without subsidies, particularly when their primary role is to sell
ally, the SLF in Fig. 15 (e) and (f) increases with system size, suggesting electricity.
that larger systems benefit from a larger BESS for stable RE distribution. Fig. 15 (c) and (d) illustrate that smaller systems have a higher UF,

15
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

Fig. 15. Technical performance analysis of system configurations determined by primary optimization/Genetic Algorithm. (a) GEIF: Monthly statistical distribution;
(b) GEIF: Monthly trend analysis; (c) UF: Monthly statistical distribution; (d) UF: Monthly trend analysis; (e) SLF: Monthly statistical distribution; (f) SLF: Monthly
trend analysis. *GEIF: Grid Energy Interaction Factor, UF: Utilization Factor, SLF: Satisfied Load Fraction

indicating a tendency to consume electricity directly due to lower RE and economic benefits, is expected to diminish social resistance to new
generation. Direct consumption of RE can avoid efficiency losses typi­ renewable projects. The validity of applying predictive and optimization
cally associated with energy storage or conversion processes, thereby methods hinges on the stability of future user energy consumption
enhancing energy efficiency and economic viability. Conversely, larger patterns, energy prices, policies, regulations, and local meteorological
systems face challenges in managing surplus energy. When this surplus conditions, which should not deviate significantly from historical trends.
cannot be sold to the grid, options like curtailment or storage in the BESS The limitations of this study can be categorized into two key areas:
bring additional costs, consequently reducing the economic attractive­ assumptions that introduce uncertainties in the models and simplifica­
ness of large-scale systems. tions that may not accurately capture the dynamics of energy systems.
The SLF across various configurations, as shown in Fig. 15 (e) and (f), Firstly, the assumptions about electricity prices and energy consumption
demonstrates a notable increase from configurations 1 to 4, yet this patterns are based on historical trends, susceptible to unpredictable
trend reaches a plateau between configurations 5 and 9. This pattern factors such as changes in global energy production, policy shifts,
confirms that smaller systems tend to meet load demands directly. In climate change, and technological advancements. These uncertainties
contrast, larger systems result in energy waste, primarily because sur­ above could lead to significant volatility in energy prices and shifts in
plus energy generation is stored in the BESS or exported to the grid. Such consumption patterns. This limitation could affect the optimization re­
practices diminish the profit of the system, underscoring the in­ sults if future data do not align with the current ones. For example, an
efficiencies associated with large-scale energy production in the context unexpected rise in energy prices might cause an underestimation of the
of this configuration. necessary RES capacity, while a decrease could result in overestimation.
Building on this, another notable limitation of our study arises from
modelling the BESS charging and discharging processes, which are
3.5. Assumptions and limitations
represented with hourly granularity. This approach simplifies the
intricate dynamics of energy flow into and out of the BESS by averaging
This study assumes that the existing electrical grid can integrate the
over an hour, potentially overlooking the nuances of shorter-duration
proposed RES without significant upgrades. An expected rise in demand
fluctuations and transitions within each hour. Such simplifications
for renewable energy, driven by increasing environmental conscious­
may not fully capture rapid changes in energy demand or supply. This
ness and supportive policies, is anticipated to ensure a stable market for
limitation could be solved by integrating another optimizer for the BESS
the produced energy. Technological advancements will likely reduce
to resolve during operation. Although this simplification introduces
costs for solar panels and batteries, making renewable energy produc­
limitations by not fully accounting for minute variations, it is crucial to
tion costs competitive with traditional fossil fuel sources. High public
note that the primary goal of this study is not to devise a precise energy
and stakeholder support, influenced by the recognized environmental

16
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

management strategy but to support capacity optimization over de­ CRediT authorship contribution statement
cades. In this extended time frame, these micro-level fluctuations
become less significant. Integrating an additional optimizer for BESS Wenrui Ye: Writing – original draft, Software, Validation, Method­
operations could offer a more detailed resolution during operation to ology, Investigation, Formal analysis, Conceptualization. Munur Sacit
address this limitation. However, given our study’s focus on long-term Herdem: Writing – review & editing, Methodology, Investigation,
capacity optimization, the impact of such micro-level fluctuations is Conceptualization. Shucheng Huang: Writing – review & editing,
deemed marginal. Conceptualization, Software, Investigation. Wei Sun: Writing – review
Moreover, policy and regulatory frameworks are assumed to be & editing, Investigation, Conceptualization. Jun Liu: Methodology,
steady, but changes here would affect the profit model and require new Conceptualization. Jatin Nathwani: Writing – review & editing, Su­
energy management strategies. Additionally, based on historical data, pervision, Resources, Conceptualization. John Z. Wen: Writing – review
variables like the discount rate might not align with future value due to & editing, Supervision, Resources, Project administration, Methodology,
the dynamic nature of technology, economics, and socio-political fac­ Investigation, Funding acquisition, Conceptualization.
tors. Periodic reassessment of the optimization model is necessary to
maintain its relevance and accuracy due to the unpredictable nature of
policy and regulatory changes. Declaration of competing interest

4. Conclusion The authors declare the following financial interests/personal re­

lationships which may be considered as potential competing interests:
This research develops a novel methodology for optimizing the ca­ [John Z. Wen reports financial support was provided by Waterloo Insi­
pacities of a renewable energy system consisting of photovoltaic panels titute for Sustainable Aeronautics (WISA). If there are other authors,
and battery energy storage to achieve a critical balance between tech­ they declare that they have no known competing financial interests or
nological advancement and economic feasibility. By integrating pa­ personal relationships that could have appeared to influence the work
rameters corresponding to user-specific demands, dynamic policy reported in this paper].
landscapes and volatile energy markets, this approach adopts an
advanced energy management algorithm alongside stochastic modelling Data availability
to enhance the optimization’s accuracy, adaptability, and robustness.
This study demonstrates that implementing a dynamic population decay Data will be made available on request.
algorithm within the non-dominated sorting genetic algorithm II
framework can significantly accelerate the convergence speed (reducing Acknowledgement
the search times by 60.42 %).
A case study has been conducted on optimizing the capacities of The author(s) thanks Waqas Khan for providing essential data for the
components in a renewable energy system in the Netherlands to validate research. This research has been partially supported by the Waterloo
the proposed methodology. Performance analysis reveals that this Institute for Sustainable Aeronautics (WISA) and the Natural Sciences
optimal configuration achieves a substantial 95 % improvement in the and Engineering Research Council of Canada (NSERC).
energy independence score and a 28.1 % enhancement in the grid en­
ergy interaction factor. Meanwhile, the sustainable internal rate of re­ References
turn remains at 3.39 %. This configuration outperforms the reference by
216 % in the Battery Energy Storage System utilization. It is also [1] IEA. Net zero by 2050: A roadmap for the global energy sector, 2021. https://www.
iea. org/reports/net-zero-by-2050.
demonstrated that a balance exists between lower capital costs and
[2] IEA. Snapshot of global pv markets 2022. 2022.
significant advancements in key performance indicators, establishing its [3] IEA. Solar pv, n.d. https://www.iea.org/energy-system/renewables/solar-PV.
cost-effectiveness, efficiency, and technical robustness. Compared to the [4] Zhu Hongyu, Goh Hui Hwang, Zhang Dongdong, Ahmad Tanveer, Liu Hui,
Wang Shuyao, Li Shenwang, Liu Tianhao, Dai Hang, Thomas Wu. Key technologies
rule-based non-dominated sorting genetic algorithm, the proposed
for smart energy systems: recent developments, challenges, and research
approach shows an 18.12 % improvement in the Internal Rate of Return opportunities in the context of carbon neutrality. Journal of Cleaner Production
(3.39 % vs. 2.87 %) while maintaining a similar level of energy inde­ 2022;331:129809.
pendence (0.39 vs. 0.38). The study concludes that smaller Renewable [5] Petrucci A, Barone G, Buonomano A, Athienitis A. Modelling of a multi-stage
energy management control routine for energy demand forecasting, flexibility, and
Energy System configurations offer greater economic viability and en­ optimization of smart communities using a recurrent neural network. Energ Conver
ergy efficiency through direct consumption of generated renewable Manage 2022;268:115995.
energy. In contrast, larger systems encounter financial difficulties and [6] Zhang B, Weihao Hu, Ghias AM, Xu Xiao, Chen Zhe. Multi-agent deep
reinforcement learning based distributed control architecture for interconnected
operational inefficiencies due to the challenges of managing excess en­ multi-energy microgrid energy management and optimization. Energ Conver
ergy, often requiring subsidies and policy support. This finding un­ Manage 2023;277:116647.
derscores the importance of system capacity optimization in [7] Herdem MS, Mazzeo D, Matera N, Wen JZ, Nathwani J, Hong Z. Simulation and
modeling of a combined biomass gasification-solar photovoltaic hydrogen
determining economic and environmental RES outcomes. This approach production system for methanol synthesis via carbon dioxide hydrogenation. Energ
ensures a comprehensive system performance evaluation, aligns eco­ Conver Manage 2020;219:113045.
nomic interests with sustainable energy practices, and contributes [8] Yousri D, Farag HEZ, Zeineldin H, El-Saadany EF. Integrated model for optimal
energy management and demand response of microgrids considering hybrid
significantly to the advancement of RESs. Future research will aim to
hydrogen-battery storage systems. Energ Conver Manage 2023;280:116809.
identify the economic-environmental balance for RESs in different lo­ [9] Curto Domenico, Favuzza Salvatore, Franzitta Vincenzo, Guercio Andrea, Am- paro
cations and enhance system robustness through scenario design and Navarro Navia Milagros, Telaretti Enrico, Zizzo Gaetano. Grid stability
improvement using synthetic inertia by battery energy storage systems in small
pressure testing.
islands. Energy 2022;254.
Declaration [10] Liu L, Wang R, Wang Y, Li W, Sun J, Guo Y, et al. Comprehensive analysis and
During the preparation of this work, the author(s) used GPT4 in optimization of combined cooling heating and power system integrated with solar
order to improve readability and language. After using this tool, the thermal energy and thermal energy storage. Energ Conver Manage 2023;275:
116464.
author(s) reviewed and edited the content as needed and take(s) full [11] Dong W, Sun H, Mei C, Li Z, Zhang J, Yang H. Forecast-driven stochastic
responsibility for the publication’s content. optimization scheduling of an energy management system for an isolated hydrogen
microgrid. Energ Conver Manage 2023;277:116640.
[12] Ra N, Ghosh A, Bhattacharjee A. Iot-based smart energy management for solar
vanadium redox flow battery powered switchable building glazing satisfying the
hvac system of ev charging stations. Energ Conver Manage 2023;281:116851.

17
W. Ye et al. Energy Conversion and Management 312 (2024) 118588

[13] Lasemi MA, Arabkoohsar A, Hajizadeh A, Mohammadi- ivatloo B. A comprehensive International Conference on Computer, Control, Electrical, and Electronics Engineering
review on optimization challenges of smart energy hubs under uncertainty factors. (ICCCEEE), pages 1–6. IEEE, 2018.
Renew Sustain Energy Rev 2022;160:112320. [30] Mazzeo D, Herdem MS, Matera N, Wen JZ. Green hydrogen production: Analysis
[14] Hassan A, Bass O, Al-Abdeli YM, Masek M, Masoum MAS. A novel approach for for different single or combined large-scale photovoltaic and wind renewable
optimal sizing of stand-alone solar pv systems with power quality considerations. systems. Renew Energy 2022;200:360–78.
Int J Electr Power Energy Syst 2023;144:108597. [31] Zhang Y, Lundblad A, Campana PE, Benavente F, Yan J. Battery sizing and rule-
[15] Ahmed MM, Das BK, Pronob Das Md, Hossain S, Kibria MG. Energy management based operation of grid-connected photovoltaic-battery system: a case study in
and sizing of a stand-alone hybrid renewable energy system for 21 community sweden. Energ Conver Manage 2017;133:249–63.
electricity, fresh water, and cooking gas demands of a remote island. Energ Conver [32] Ying Xu, Zhang H, Huang L, Rong Qu, Nojima Y. A pareto front grid guided multi-
Manage 2024;299:117865. objective evolutionary algorithm. Appl Soft Comput 2023;136:110095.
[16] Mahmoud FS, Zaki AA, Diab ZM, Ali A-H, El-Sayed TA, Ahmed M, et al. Optimal [33] Pena-Bello A, Barbour E, Gonzalez MC, Patel MK, Parra D. Optimized pv-coupled
sizing of smart hybrid renewable energy system using different optimization battery systems for combining applications: impact of battery technology and
algorithms. Energy Rep 2022;8:4935–56. geography. Renew Sustain Energy Rev 2019;112:978–90.
[17] Mohamed MA, Eltamaly AM, Alolah AI, Hatata AY. A novel framework-based [34] Ember. European wholesale electricity price data. Data sourced from ENTSO-e and
cuckoo search algorithm for sizing and optimization of grid-independent hybrid cleaned, October 2023. Available at https://ember-climate.org/data-catalogue/
renewable energy systems. Int J Green Energy 2019;16(1):86–100. european-wholesale-electricity-price-data/.
[18] Mohamed MA, Eltamaly AM, Alolah AI. Swarm intelligence-based optimization of [35] Hesse Holger C, Martins Rodrigo, Musilek Petr, Naumann Maik, Truong Cong Nam,
grid-dependent hybrid renewable energy systems. Renew Sustain Energy Rev Jossen Andreas. Economic optimization of component sizing for residential battery
2017;77:515–24. storage systems. Energies 2017;10(7):835.
[19] Eltamaly AM, Mohamed MA, Al-Saud MS, Alolah AI. Load management as a smart [36] Georg Angenendt, Bahram Ashrafinia, Sebastian Zurmu¨hlen, Kevin Jacqué, Julia
grid concept for sizing and designing of hybrid renewable energy systems. Eng Badeda, Dirk Uwe Sauer. (2017) Influence of the battery voltage level on the
Optim 2017;49(10):1813–28. efficiency and cost of a pv battery energy storage system. In 32. Symposium
[20] Campana PE, Cioccolanti L, Franc¸ois B, Jurasz J, Zhang Y, Varini M, Yan J. Li-ion Photovoltaische Solareergie.
batteries for peak shaving, price arbitrage, and photovoltaic self-consumption in [37] Khan W, Liao JY, Walker S, Zeiler W. Impact assessment of varied data
commercial buildings: a monte carlo analysis. Energ Conver Manage 2021;234: granularities from commercial buildings on exploration and learning mechanism.
113889. Appl Energy 2022;319:119281.
[21] Eltamaly AM, Mohamed MA. A novel software for design and optimization of [38] Ye Wenrui, Herdem Mu¨nu¨r Sacit, Li Joey Z, Nathwani Jatin, Wen John Z.
hybrid power systems. J Braz Soc Mech Sci Eng 2016;38:1299–315. Formulation and data-driven optimization for maximizing the photovoltaic power
[22] Hassan A, Al-Abdeli YM, Masek M, Bass O. Optimal sizing and energy scheduling of with tilt angle adjustment. Energies 2022;15(22):8578.
grid-supplemented solar pv systems with battery storage: Sensitivity of reliability [39] Bukhari Muhammad Haroon, Javed Adeel, Kazmi Syed Ali Abbas, Ali Majid,
and financial constraints. Energy 2022;238:121780. Chaudhary Mateeb Talib. Techno-economic feasibility analysis of hydrogen
[23] Wu Y, Liu Z, Liu J, Xiao H, Liu R, Zhang L. Optimal battery capacity of grid- production by ptg concept and feeding it into a combined cycle power plant
connected pv-battery systems considering battery degradation. Renew Energy leading to sector coupling in future. Energy Conversion and Management 2023;
2022;181:10–23. 282:116814.
[24] Liu T, Yang Z, Duan Y, Shuozhuo Hu. Techno-economic assessment of hydrogen [40] El-Houari H, Allouhi A, Salameh T, Kousksou T, Jamil A, El Amrani B. Energy,
integrated into electrical/thermal energy storage in pv+ wind system devoting to economic, environment (3e) analysis of wt-pv-battery autonomous hybrid power
high reliability. Energ Conver Manage 2022;268:116067. plants in climatically varying regions. Sustainable Energy Technol Assess 2021;43:
[25] Yi He Su, Guo JZ, Feng Wu, Huang J, Pei H. The quantitative techno-economic 100961.
comparisons and multi-objective capacity optimization of wind-photovoltaic [41] Ibrahim MM, Mostafa NH, Osman AH, Hesham A. Performance analysis of a stand-
hybrid power system considering different energy storage technologies. Energ alone hybrid energy system for desalination unit in Egypt. Energ Conver Manage
Conver Manage 2021;229:113779. 2020;215:112941.
[26] Yang J, Yang Z, Duan Y. Capacity optimization and feasibility assessment of solar- [42] Al-Ammar EA, Habib HUR, Kotb KM, Wang S, Ko W, Elmorshedy MF, et al.
wind hybrid renewable energy systems in China. J Clean Prod 2022;368:133139. Residential community load management based on optimal design of standalone
[27] Medghalchi Z, Taylan O. A novel hybrid optimization framework for sizing hres with model predictive control. IEEE Access 2020;8:12542–72.
renewable energy systems integrated with energy storage systems with solar [43] Sadeghi D, Naghshbandy AH, Bahramara S. Optimal sizing of hybrid renewable
photovoltaics, wind, battery and electrolyzer-fuel cell. Energ Conver Manage 2023; energy systems in presence of electric vehicles using multi-objective particle swarm
294:117594. optimization. Energy 2020;209:118471.
[28] Al-Quraan A, Al-Mhairat B. Sizing and energy management of standalone hybrid [44] Sodhi M, Banaszek L, Magee C, Rivero-Hudec M. Economic lifetimes of solar
renewable energy systems based on economic predictive control. Energ Conver panels. Procedia CIRP 2022;105:782–7.
Manage 2024;300:117948. [45] Bloomberg New Energy Finance. Top 10 energy storage trends in 2023, 2023.
[29] Elaff ALfadil Ismail and Samah Mohammed Hashim. An economic evaluation of [46] Why does the federal reserve aim for inflation of 2 percent over the longer run?
grid connected photovoltaic system for a residential house in khartoum. In 2018 Board of Governors of the Federal Reserve System, 2020. Available at https://www.
federalreserve. gov/faqs/economy_14400.htm.

18