Article

A Sizing Method for PV–Battery–Generator Systems for

Off-Grid Applications Based on the LCOE
Ioannis E. Kosmadakis and Costas Elmasides *

Department of Environmental Engineering, Democritus University of Thrace, 67100 Xanthi, Greece;

ikosmada@env.duth.gr
* Correspondence: kelmasid@env.duth.gr; Tel.: +30-25410-79876

Abstract: Electricity supply in nonelectrified areas can be covered by distributed renewable energy
systems. The main disadvantage of these systems is the intermittent and often unpredictable nature
of renewable energy sources. Moreover, the temporal distribution of renewable energy may not
match that of energy demand. Systems that combine photovoltaic modules with electrical energy
storage (EES) can eliminate the above disadvantages. However, the adoption of such solutions is
often financially prohibitive. Therefore, all parameters that lead to a functionally reliable and self-
sufficient power generation system should be carefully considered during the design phase of such
systems. This study proposes a sizing method for off-grid electrification systems consisting of pho-
tovoltaics (PV), batteries, and a diesel generator set. The method is based on the optimal number of
PV panels and battery energy capacity whilst minimizing the levelized cost of electricity (LCOE) for
a period of 25 years. Validations against a synthesized load profile produced grid-independent sys-
tems backed by different accumulator technologies, with LCOEs ranging from 0.34 EUR/kWh to
Citation: Kosmadakis, I.E.;
0.46 EUR/kWh. The applied algorithm emphasizes a parameter of useful energy as a key output
Elmasides, C. A Sizing Method for parameter for which the solar harvest is maximized in parallel with the minimization of the LCOE.
PV-Battery-Generator Systems for
Off-Grid Applications Based on the Keywords: photovoltaics; battery; diesel generator; sizing method; renewable energy sources; off-
LCOE. Energies 2021, 14, 1988. grid power systems; levelized cost of electricity; distributed energy resources; electrical energy stor-
https://doi.org/10.3390/en14071988 age

Academic Editor: Praveen

Cheekatamarla

1. Introduction
Received: 9 March 2021
Accepted: 1 April 2021
Renewable energy sources can, in the long term, ensure a sustainable energy supply
Published: 3 April 2021
and reduce local and global air pollutant emissions. In addition, it is the most promising
solution for supplying electrical load to remote and rural areas that are not served by an
Publisher’s Note: MDPI stays neu- electrical grid. According to a recent study [1], the population without access to electricity
tral with regard to jurisdictional was estimated to be 861 million in 2018. Moreover, in the same study, it was also men-
claims in published maps and insti- tioned that, in 2018, about 2.65 billion people were living without access to clean cooking,
tutional affiliations. which means that they did not have access to fuels and technologies such as natural gas,
liquefied petroleum gas (LPG), electricity, and biogas.
Reducing the cost of photovoltaics while increasing their efficiency has made off-grid
solar systems more economically attractive, resulting in increased use of these systems to
Copyright: © 2021 by the authors. Li- meet the energy needs of the abovementioned populations. Specifically, since 2010, more
censee MDPI, Basel, Switzerland. than 180 million off-grid solar systems have been installed in these kinds of applications.
This article is an open access article
However, only 17% of these systems were used to meet household needs, whereas the
distributed under the terms and con-
peak power (up to 10 W) of the other 83% was mainly used for lighting and charging
ditions of the Creative Commons At-
mobiles [1]. In order to further increase the number of systems that meet household needs
tribution (CC BY) license (http://crea-
while improving the quality of these people’s lives, methods should be developed which
tivecommons.org/licenses/by/4.0/).
lead to optimum techno-economic solutions. Some of the efforts that have been made in
this direction are described below.

Energies 2021, 14, 1988. https://doi.org/10.3390/en14071988 www.mdpi.com/journal/energies

Energies 2021, 14, 1988 2 of 28

Several research teams dealt with optimal dimensioning of PV–battery systems using
appropriate software tools. For example, a widely used software is the Hybrid Optimiza-
tion Model for Electric Renewables (HOMER) developed in order to determine the opti-
mal size of system components by carrying out techno-economic analysis. In El-Houari et
al. [2], real monthly electrical demands and hourly site-specific meteorological conditions
were used to calculate, via HOMER Pro software, the PV and battery requirements for
rural African regions. A very comprehensive review [3] describes how this software
works. Input data including meteorological data, load profile, equipment characteristics,
and economic and technical data are required for simulation and optimization. Optimiza-
tion occurs by finding an optimal value for an objective function which is the present value
of the sum of costs minus the sum of revenues. The final step concerns the sensitivity
analysis where an evaluation of the variation in uncertain parameters (such as fuel cost,
wind speed, solar radiation, electricity price, and component cost) on optimal sizing takes
place.
Other available software tools for optimal sizing of standalone PV–battery systems
are Improved Hybrid Optimization by Genetic Algorithms (IHOGA), Transient Systems
Simulation Program (TRNSYS), RETScreen, and PVSYST [4]. All of these programs use
almost the same input data as those reported for HOMER in the previous paragraph.
However, the results of each software and their application differ, as described in detail
in [4].
In addition to software, many research groups developed optimization tools and
techniques to approach an optimum techno-economic solution. These attempts can be
classified into various categories: genetic algorithms, particle swarm optimization (PSO),
simulated annealing, and ant colony algorithm. Information on the development philos-
ophy of these methods and how to use them can be found in a previous review [5]. In their
work, Dufo-López et al. [6] studied the multi-objective design of PV–wind–diesel config-
urations coupled with electrical energy storage (EES) in order to minimize both the lev-
elized cost of electricity (LCOE) and CO2 emissions, employing relatively fast evolution-
ary algorithms. Results showed that the best pareto fronts included a diesel generator that
contributes to the overall economic and environmental performance. In Maleki et al. [7],
a combinatorial optimization method based on the harmony search algorithm for sizing
off-grid PV–battery and generator systems was presented. In this study, the proposed
method outperformed a simulated annealing method, displaying extremely fast runtimes,
although battery technology and various battery specifics such as depth of discharge and
cycle life were not considered. Simulated annealing genetic algorithms were employed by
Wei et al. [8] to optimize coal-fired boiler operation so as to reduce NOx emissions,
thereby exhibiting accurate solutions in a low computing time. Similarly, in [9], Ghafoor
et al. presented a deterministic and straightforward off-grid PV–battery configuration
model for a residential case study in Pakistan accompanied by a lifecycle cost analysis
without considering battery technology.
In [10], Mandelli et al. proposed an off-grid PV–battery sizing method for rural areas
of developing countries introducing the loss of load probability and the electricity unit
cost as key aspects in the sizing process. In a case study, they displayed the effectiveness
of their numerical method and showed that accurate results are required in order to opti-
mize the energy and economic cost of the system. Furthermore, in [11], Mandelli et al.
investigated the impact of load profile uncertainty on PV and battery sizing and intro-
duced an algorithm for implementing stochastic load profile formulation.
In a paper by Shank et al. [12], a dispatch-coupled sizing method based on a PSO-
enabled algorithm for batteries in systems with different penetration levels of renewables
was introduced. Results showed that their metaheuristic modifications of the particle
swarm optimization avoided premature convergence to local optima and was effective in
large-scale energy systems with different penetration levels of RES. Moreover, in [13],
Rodríguez-Gallegos et al. employed a multi-objective optimization approach, considering
Energies 2021, 14, 1988 3 of 28

the LCOE, CO2 emissions, and grid voltage for sizing interconnected PV–battery–diesel
combinations.
Recently, Li [14] studied the PV–battery optimization problem of grid-connected
households by incorporating load demand uncertainty and time-of-use tariffs into a sizing
genetic algorithm involving a time-series simulation, showing more realistic results com-
pared to algorithms using average data from chosen sites. Reinforcement learning and
evolutionary algorithms have an extended scope and have also been reported to be effec-
tive in various applications such as battery-related energy management systems and elec-
trical vehicles [15,16].
A common feature of sizing methods is the construction of an objective function and
the attempt to optimize it through changes in operational, technical, and economic param-
eters. The most widely used objective function is the LCOE [10,17,18]. The LCOE provides
an indication of the cost per unit of energy throughout the operation of the sizing system.
In general, LCOE aggregates all costs over the lifetime of an energy system and divides
them by the total energy production over the lifetime.
In this work, we propose a design method for non-interconnected photovoltaic sys-
tems with batteries and generator sets (PVBG). This method is based on the synthesis of
the objective function expressed by the LCOE. Its aim is the selection of an effective com-
bination of PV panel quantity and nominal energy capacity of the battery, so that the cost
of the system per unit of energy for a period of 25 years is minimized. The sizing method
considers a detailed electricity demand profile; therefore, the optimized PVBG configura-
tion matches with the energy consumption more realistically, in contrast to algorithms
using average data from specific sites. Additionally, this enumerative method not only
produces optimal PVBG configurations but also determines the actually utilized solar en-
ergy harvest, thereby calculating a more precise LCOE.
After this introduction, the off-grid design of a PVBG system is discussed in Section
2. In this section, the hourly PV power output is estimated and linked to a typical house-
hold load profile. The operation and sizing algorithms of the PVBG system, as well as the
system cost analysis based on the LCOE, are presented and described in detail. The results
of the applied sizing method are validated and presented in Section 3, followed by a dis-
cussion in Section 4. Concluding remarks and suggestions for future work are given in the
last section.

2. Materials and Methods

2.1. Off-Grid PV–Battery–Generator System
Distributed energy resources (DERs) such as PV modules, power electronics, and
batteries can be integrated into a unified and autonomous electrification system in several
ways. A key consideration when planning PV–battery systems is the selection of the al-
ternating current (AC) or direct current (DC) system architecture, i.e., the preference
among AC or DC bus utilization, which differs in power converter arrangements and con-
version steps. Along with the electrical energy generation and storage, an auxiliary power
source in the form of a generator has to be added to the system to achieve total grid inde-
pendence.
In this section, the sizing method for PV–battery–generator systems oriented towards
off-grid applications is presented in detail. The key points of the method are depicted in
Figure 1.
Energies 2021, 14, 1988 4 of 28

Figure 1. Flow diagram depicting the basic steps of the proposed photovoltaic (PV)–battery sizing method. Steps 7 to 10
are iterated for a specified range of PV modules and battery energy capacities, subsequently highlighting optimal values
for N and Bnec at minimum levelized cost of electricity (LCOE).

2.1.1. Preference of the DC-Coupled Architecture

In DC-coupled systems, the PV arrays are connected to the DC busbars via unidirectional
DC/DC converters, commonly known as maximum power point tracking (MPPT) solar
charge controllers. The battery is directly connected to the DC busbars, forming together with
the PV output a parallel connection to the DC side of an inverter, as shown in Figure 2.

Figure 2. Off-grid system architecture of direct current (DC)-coupled PV–battery–generator system.

In this setup, the PV modules power the load on the consumption side through the
DC/DC converter and then through a DC/AC inverter. PV power also charges the battery
through the DC/DC converter in a single conversion step, and the discharging battery
power also contributes to load supply via the DC/AC inverter. In off-grid systems, a
standby power source on the AC side of the inverter can act as a backup for energy short-
ages and supply power directly to the load. Battery charging by the standby power source
occurs through an additional AC/DC rectifier, usually, a charger incorporated into the
inverter, as shown in Figure 2.
Energies 2021, 14, 1988 5 of 28

In AC-coupled systems, the typical AC bus is used as the conductive medium linking
the power source, energy storage, and load. The PV arrays and the battery are connected to
the AC bus via unidirectional and bidirectional DC/AC inverters, respectively. Since, in this
setup, the load is directly connected to the AC bus, the power supply of the load is done
from the PV modules and the discharging battery in a single conversion step. Again, in off-
grid systems, the standby power source supplies power directly to the load when needed.
For the most part, off-grid PVBG systems on a smaller scale follow a DC-coupled
system architecture, primarily for practical reasons, since electrochemical accumulators
work with DC. This simple and robust setup facilitates the reliable and efficient charging
of batteries, especially in instances where the surplus renewable energy harvest needs to
be maximized. Another critical issue is the fact that, when batteries are too far discharged,
a battery inverter shuts down the AC output to protect the battery from damage. In the
case of a DC architecture, this is relatively inconsequential as solar energy eventually re-
charges the battery through the charge controller. On the contrary, when batteries are too
far discharged in an AC architecture, special use of a backup generator is required to pro-
vide power references, such as voltage and frequency, to allow renewable solar and bat-
tery inverters to reconnect.
Furthermore, DC-coupled setups are very cost-effective for small to medium-size off-
grid systems. Solar charge controllers are highly modular and scalable, i.e., additional PV
arrays can be easily added to the DC bus, if required, using comparatively low-cost solar
charge controllers. Consequently, for these reasons, a DC architecture was preferred for
simulation in this work.

2.1.2. Selection of the System Components

The decisive criteria for the selection of PV panels are the degree of efficiency as a
function of the module temperature coefficients, the product and performance module
warranties, and the purchasing cost of PV in EUR per installed capacity. In the present
work, only photovoltaic modules from crystalline silicon were considered since they are
the most prevalent PV type in the PV market.
Industrially produced crystalline silicon PV cells currently achieve efficiencies rang-
ing from 18% to 22% under standard test conditions (STCs) [19]. The industry average of
solar panel manufacturer warranties varies from 10 to 25 years. Considering the global
declining trend in module average selling price, the reported average crystalline silicon
module prices in 2020 ranged from approximately 0.40 EUR/Wp to 0.25 EUR/Wp [20].
For the simulation purposes of the present work, the module selected was SolarCall
SCM310, a passivated emitter and rear contact (PERC) PV module made of monocrystal-
line silicon, from a European manufacturer [21]. The module’s installed capacity is 310
Wp with a surface area of 1.627 m2, resulting in a module energy efficiency of approxi-
mately 19%, under STCs. According to the manufacturer’s datasheet, the peak power tem-
perature coefficient is −0.40%/°C. The manufacturer product and linear power warranties
are 15 years and 30 years, respectively. More specifically, the performance warranty for
15 years is 91.2% and that for 30 years is 80.6% of rated power. Cost of the module was
inquired online, and offers ranging from 0.45 EUR/Wp to 0.35 EUR/Wp were collected
from selected distributors.
For the power electronics simulation requirements, features were taken from data
sheets of established power electronics manufacturers. More specifically, in every simu-
lation run, the energy flows originating from a PV array block were handled by an MPPT
solar charge controller, and its efficiency level was considered relatively stable throughout
the whole power range (approximately 95%). Nevertheless, efficiency estimations of in-
verters are more complicated since inverter efficiency is a function of apparent power out-
put. Inverters are generally designed to achieve a high power factor when operating at
full power. Actual conditions vary, however. Therefore, the power factor of the inverters
used in the simulations was considered greater than 90% when the power output was
greater than 20% of the nominal inverter power [22].
Energies 2021, 14, 1988 6 of 28

The solar charge controllers and the battery inverters considered for simulation in
this work were from Victron Energy. The charge controllers exhibit a PV input voltage,
ranging from 75 V to 150 V and a current output up to 70 A. The inverters supply up to
5000 VA apparent power and also feature a battery charger and generator triggering. The
battery charging current and automatic generator triggering can be programmed with a
computer. Although manufacturer product warranties are limited to 5 years, the power
converters are expected to last 25 years. Costs of the specific items were inquired online,
and offers were collected from selected distributors, which were 20% lower than the offi-
cial prices published quarterly by established manufacturers [23].
In self-sufficient power systems, the EES is heavily used to facilitate power deficits from
RES and to operate as a short-term storage medium. In recent years, lithium-ion batteries
have surged in popularity and overtaken lead–acid batteries as the preferred EES medium
[24]. Nevertheless, in many instances, the lead–acid battery is still the first choice when it
comes to EES, especially when the initial capital cost is a critical point of consideration [25].
Another point of consideration in the EES selection process is the end-of-life management
of the anode and cathode material of the accumulators. In contrast to the lithium-ion battery,
the lead–acid battery exhibits a very high recovery rate [26]. Furthermore, lithium-ion bat-
teries must operate in a strictly defined voltage and temperature window. Safe and efficient
charging and discharging of lithium-ion accumulators is assured only through the use of a
battery management system (BMS), which again increases cost.
Consequently, both battery types were studied in this work and simulated as an EES
resource. Complying with the DC-coupled architecture, the battery was directly con-
nected to the DC busbars and, therefore, defined the system voltage. Commonly, higher
power levels require greater system voltage, since delivering more power at a given volt-
age takes more current, which in turn leads to higher power losses. Here, the considered
nominal system voltage was 48 V for both battery types. Equally important is the nominal
battery capacity since the product of voltage and capacity defines the nominal energy ca-
pacity of the battery used in the simulation. Furthermore, significant battery features such
as nominal battery cycles at maximum depth of discharge (DOD) and calendar battery life
were extracted from datasheets of renowned battery manufacturers. More specifically, in
this study, the lead–acid battery selected was the RES SOPzV from Systems Sunlight S.A.
[27] and the lithium-ion battery selected was the LFP-Smart from Victron Energy [28],
both labelled as EES solutions used in conjunction with RES. Both batteries were sized at
approximately 5 kWh energy capacity intervals, and, according to the manufacturers, the
cycle life at 50% DOD was 2500 cycles for the lead–acid battery and 5000 cycles for the
lithium-ion battery. Moreover, different theoretical maximum calendar lives of 5 years
and 10 years, respectively, were included in the simulation, in line with [29] and [30]. The
cost per kWh for each battery type was inquired online, and offers ranging from 154
EUR/kWh to 574 EUR/kWh, for lead–acid and lithium-ion batteries, respectively, were
collected from selected distributors.
Generator sets are fuel power systems comprising an engine and an alternator and
are broadly categorized into primary and standby generators. The latter are used as
backup generators (BGs) that instantly supply power to the load when the primary power
supply, in this work the PV–battery system, fails. Optimal sizing of diesel generator sets
in hybrid energy systems is an essential step in planning economical off-grid electrifica-
tion systems [31], with resiliency being a key aspect to consider [32]. Currently, natural
gas- and diesel-powered engines are the industry standard, although biomass and biogas
are gaining ground. Nevertheless, there are several other crucial factors to consider before
choosing a BG. These systems run, to a high degree, infrequently and remain in a cold
state for most of their cycle life. Therefore, it is important to carefully define the required
function and the operating time intervals of BGs in an off-grid system. The sizing proce-
dure for BGs must take into account the maximum continuous time use (CTU), the esti-
mated annual operation time (APT), and the average load value (ALV) during BG opera-
tion as a function of the rated generator power. Furthermore, the amount of electricity
Energies 2021, 14, 1988 7 of 28

supplied to the load by the BG depends on the quantity of fuels consumed, which in turn
is a function of generator efficiency and operation time. The total operational lifespan of
such systems can reach 20 years [33] as long as the generator is well maintained and CTU
and APT are not exceeded.
Once again generator specifications were taken from manufacturer datasheets and
used to simulate standby BG operation on an hourly basis, corresponding to PV, battery,
and power converter components. The considered generator in this study was the Hyun-
dai DHY6000SE diesel standby generator, capable of supplying up to 4.5 kW of power
over a single phase, with maximum CTU and APT values limited to 5 and 500 h, respec-
tively. The standard warranty for this product is 2 years, yet the manufacturer claims that
a total lifespan of above 10,000 h can be achieved with correct maintenance and operation
(that is, if the average load is not lower than 30% or higher than 80% of the rated generator
output). The capital cost for this generator set varies from 1390 EUR to 1500 EUR per item
depending on location and shipping. Annual operation and maintenance cost estimations
range from 235 to 260 EUR.

2.2. Annual Electricity Generation Estimation

A method for accurately estimating the annual energy winnings from PV systems
was examined in previous work [25,34] in more detail. In summary, simulating the annual
PV energy output requires the use of the PV datasheet and hourly sequenced meteorolog-
ical data for a specified position. Time series on an hourly basis of global horizontal irra-
diance (GHI), onsite atmospheric temperature (T), and horizontal windspeed (WS) are
used for at least one calendar year to produce analogous sequences of direct normal irra-
diance (DNI) and diffuse horizontal irradiance (DHI). This was done using the DISC
model from [35,36]. Given the PV surface and PV azimuth angles, sequences of the solar
angle incidence (AOI) as a function of solar position with respect to latitude, longitude,
and altitude were calculated. An acceptable albedo of 0.25 was used for all sizing simula-
tions in this study and, lastly, hourly sequences of the total solar irradiance instance (Epoa)
hitting the PV modules were calculated. The above-simulated sequences were fed at first
into the Sandia Cell and Module Temperature Model [37], to compute PV cell (TC) and
PV module temperatures (TM), and subsequently into the five-parameter model [38] lead-
ing to the five corresponding outputs of the model: (i) the photocurrent (IL), (ii) the diode
reverse saturation current (I0), (iii) the series resistance (Rs), (iv) the shunt resistance (Rsh),
and (v) the modified diode ideality factor (nNsVth). These outputs comprise the inputs
for the widely accepted single-diode model [39,40] for photovoltaic modules from which
power estimations are obtained.
Losses because of PV module reflectivity and soiling were taken into consideration.
According to PV manufacturers, PV modules are typically guaranteed to deliver 80% of the
rated power output at the end of their lifetime. Hence, the annual PV energy output was
corrected further with a degradation factor of 0.9, optimizing mean power output estima-
tions over a typical 25 year period. Model implementation and data processing were done
in Python using the PVLIB solar simulation library for PV energy systems [41–45].
Using a typical meteorological year (TMY) [46] for a specified location (Xanthi,
Greece) in conjunction with the above-described electricity generation estimation method,
a total of 8760 hourly sequenced data points were generated, representing the annual elec-
trical energy output from one PV module. This referential output was then multiplied by
the number of PV modules used in each sizing step to determine the total annual energy
winnings from a certain PV capacity instance of the algorithm.

2.3. Reference Load Profile

A consumer load profile of an average four-person household was synthesized by
approximating the electrical appliance usage over a whole year, which was then used as
a reference for the sizing algorithm developed in this study. The load profile was scaled
to the annual electricity consumption of approximately 5500 kWh with the winter months
Energies 2021, 14, 1988 8 of 28

(in the northern hemisphere) from December to February acting as the baseload months
with a baseload of approximately 300 kWh, as one can observe in Figure 3.

Annual Electricity Consumption

700
600
500
400
kWh

300
200
100
0
January

March

May

August
September

December
February

April

June
July

November
October
Figure 3. The synthesized electricity consumption of an average four-person household. The esti-
mated baseload of the lowest 3 months was approximately 300 kWh.

A 1 h resolution was used to match the resolution of the PV production profile and
to generate 8760 data points of inelastic electricity demand. The electricity profile exhib-
ited seasonal variations and followed the daily human activity in a household peaking
twice in a day, i.e., once at midday and once in the evening, similar to [47]. Figure 4 shows
the annual load profile in more detail illustrating the hourly peak loads of 2 kWh, 1.4 kWh,
and 1.2 kWh generated in the summer, spring/autumn, and winter months, respectively.

Figure 4. Hourly sequence of the annual load profile, based on the daily electrical appliance usage
from a four-person household.

Electricity demand for heating and transportation purposes, flexible loads, and work-
day/ weekend demand variations were not considered since this would require a more
detailed modeling effort; this will be addressed in future work.
Energies 2021, 14, 1988 9 of 28

2.4. System Operation and Sizing Algorithm

In this section, an operation and sizing algorithm (SA) for autonomous off-grid
PVBAT systems is introduced. After the system specification (Step 1 and Step 2 of Figure
1), the SA used time-series data from a TMY of a specified location and consistent data
from datasheets of PV, converter, battery, and generator manufacturers to generate an-
nual time series of the power output for one PV module with a 1 h resolution. Initially,
numbers for the required PV modules (N) and the required battery nominal energy ca-
pacity (Bnec) were set, leading to the determination of the annual PV output power of the
entire PV array in hourly steps (Ppv) and to the initial battery state of energy (SOE) avail-
able. The Ppv at this point included converter losses. A matching load profile was synthe-
sized, as described in Section 2.3, leading to an hourly power demand (Pload) for an entire
year. Figure 5 illustrates the system operation flow in terms of Ppv usage, Pload coverage,
and hourly battery SOE.

Figure 5. System operation flow according to the to the hourly battery state of energy and PV energy usage.

We can distinguish the following cases of operation depending on whether excess

energy is available or not at every hourly step.
Case A. Surplus energy case and a fully charged battery
In this case, available surplus energy, defined here as Esur, should either be used to
charge the battery or be discarded.
𝑃𝑙𝑜𝑎𝑑
𝐸𝑠𝑢𝑟 = [𝑃𝑃𝑉 − ( )]𝛥𝑡 ≥ 0, (1)
𝑛𝑖𝑛𝑣
where ninv is the inverter efficiency and can be acquired from the inverter’s efficiency curve
[26]. The SOEt-1 is greater than or equal to 100%, which means that the battery is fully
charged, and the excess of PV energy remains unused; thus, potential Esur at this step is
wasted and characterized as Ewaste. With the SOE remaining at the higher level (100%), the
algorithm proceeds to the next hourly step with the following equations:
𝐸𝑤𝑎𝑠𝑡𝑒 = 𝐸𝑠𝑢𝑟 , (2)
and
𝑆𝑂𝐸𝑡 = 𝑆𝑂𝐸𝑡−1 = 100%. (3)
Τhe t − 1 indicator means that the calculation of the SOE was made taking into ac-
count the data obtained in the previous hourly step.
Case B. Surplus energy case and a partially charged battery
Energies 2021, 14, 1988 10 of 28

If the SOEt-1 is not greater than or equal to 100%, the battery is in a partially charged
state and keeps charging by Esur (Equation (1)). Then, SOEt which is initially scaled at 100%
is defined as follows:
min(𝐸𝑏𝑎𝑡𝑡 , 𝐵𝑛𝑒𝑐 ) × 100
𝑆𝑂𝐸𝑡 = . (4)
𝐵𝑛𝑒𝑐
If 𝐸𝑏𝑎𝑡𝑡 < 𝐵𝑛𝑒𝑐 with 𝐸𝑏𝑎𝑡𝑡 calculated with
𝐸𝑏𝑎𝑡𝑡 = 𝐸𝑏𝑎𝑡𝑡−1 + 𝐸𝑠𝑢𝑟 × 𝑛𝑏𝑎𝑡 , (5)
then the surplus energy (Esur) can be totally absorbed by the battery, where nbat is the charg-
ing efficiency of the battery.
If for Ebatt, calculated by Equation (5), Ebatt > Bnec applies, then the amount of energy
that cannot be absorbed is discarded.
𝐸𝑤𝑎𝑠𝑡𝑒 = 𝐸𝑏𝑎𝑡𝑡 − 𝐵𝑛𝑒𝑐 . (6)
Monitoring the above two cases using the algorithm ensures that the SOE is always
≤100%. This is the reason why Equation (4) has this form. Afterward, the algorithm pro-
ceeds to the next hourly step.
Case C. Energy shortage case
In this case, the PV power output cannot cover the complete inelastic power demand
and requires backing to compensate for the energy shortage, defined here as Eshort.
𝑃𝑙𝑜𝑎𝑑
𝐸𝑠ℎ𝑜𝑟𝑡 = [𝑃𝑃𝑉 − ( )]Δ𝑡 < 0. (7)
𝑛𝑖𝑛𝑣
The SOEt-1 is greater than the threshold percentage (100%—DOD) which means that
the battery is in a partially charged state and keeps discharging as it contributes to the
load coverage. Consequently, the SOEt is defined again by Equation (4), using this time
the remaining amount of energy Ebatt, defined by the following equations:
𝐸𝑏𝑎𝑡𝑡 = 𝐸𝑏𝑎𝑡𝑡−1 + 𝐸𝑠ℎ𝑜𝑟𝑡 (Eshort < 0), (8)
and
𝐸𝑤𝑎𝑠𝑡𝑒 = 0. (9)

Case D. Backup case

If the SOEt-1 is less than the threshold value, which means that the battery is in a
discharged state, the backup source needs to start charging the battery in the next hourly
step and keep the entire load covered until the battery is fully charged again. This recur-
ring case consists of a loop in which the hourly load is primarily covered by the generator,
for loads up to the maximum generator output (PbackupMAX). Secondarily, the battery is
charged with the sum of the power of the remaining generator output on the DC side
(Pbackup->battery) and the available PV power output. Hence, no energy is being discarded (Εwaste
= 0). In this case, the accumulated Ebatt is described in Equation (13).
𝑃𝑏𝑎𝑐𝑘𝑢𝑝 = 𝑃𝑙𝑜𝑎𝑑 + (𝑃𝑏𝑎𝑐𝑘𝑢𝑝𝛭𝛢𝛸 − 𝑃𝑙𝑜𝑎𝑑 ), (10)
𝑃𝑏𝑎𝑐𝑘𝑢𝑝→𝑏𝑎𝑡𝑡𝑒𝑟𝑦 = 𝑛𝑖𝑛𝑣 × (𝑃𝑏𝑎𝑐𝑘𝑢𝑝𝛭𝛢𝛸 − 𝑃𝑙𝑜𝑎𝑑 ), (11)
and
𝐸𝑠𝑢𝑟 = 𝑃𝑝𝑣 Δ𝑡, (12)
𝛦𝑏𝑎𝑡𝑡 = 𝛦𝑏𝑎𝑡𝑡−1 + 𝑃𝑏𝑎𝑐𝑘𝑢𝑝→𝑏𝑎𝑡𝑡𝑒𝑟𝑦 + 𝑛𝑏𝑎𝑡 × 𝐸𝑠𝑢𝑟 . (13)
The SOEt variation is kept track of by using Equation (4). Breaking out of the loop
and returning to normal operation requires that the SOEt reaches approximately 80%, a
safe state that prevents overcharging by an overshooting charge. More specifically, during
Energies 2021, 14, 1988 11 of 28

battery charging by the backup, all the energy generated by the PV system is directed to
the battery due to the particular architecture of the system. If we charge the battery further
than SOEt > 80% with the generator, then there is a serious possibility that the contribution
of the PV-generated energy gets wasted. Our main concern is always to avoid situations
where the energy produced by RES has to be rejected.
On the other hand, energy storage technologies, such as lead–acid batteries, must be
fully charged at regular intervals to avoid premature aging, e.g., after a certain number of
charge/discharge cycles. This will be integrated into the algorithm in future work.
When the end of the time series is reached, which means that the last hourly step of
the simulation has been examined, the results of the simulation are saved. A brief descrip-
tion of the results is shown in Table A1 of Appendix B. Subsequently, a new simulation
instance is initiated with incremental N and Bnec.
The repetition of this process produces multiple sums of annual discarded energy
and annual energy supplied by the generator (ΣΕwaste + ΣPbackupΔt) for every N and Bnec
simulation instance. Therefore, a distinct minimum of ΣΕwaste + ΣPbackupΔt signifies an opti-
mal N and Bnec arrangement for which the renewable harvest is maximized. Furthermore,
the annually utilized PV output, i.e., the useful energy Eus, is defined in Equation (14).
𝐸𝑢𝑠 = 𝛴(𝑃𝑝𝑣 𝛥𝑡) − 𝛴(𝐸𝑤𝑎𝑠𝑡𝑒 ). (14)
Datapoints of the economic unit cost of useful energy are produced in each simula-
tion instance. The distinct minimum of these datapoints signifies an optimal N and Bnec
arrangement for which the cost of every electrical kilowatt-hour generated is essentially
minimized.

2.5. System Cost Analysis and the Levelized Cost of Electricity (LCOE)
The performance of different electricity generation technologies can be evaluated by
calculating the LCOE [48]. This estimate considers all costs incurred during the lifetime of
a power resource with reference to the power resource’s total electricity output through-
out its complete lifecycle. In this method, the capital cost, as well as the lifecycle cost,
which includes operation, maintenance, and replacement expenditures during the power
resource’s lifetime, are calculated considering an estimated discount factor. Thus, the
LCOE is the ratio of the total discounted lifetime cost of a power resource divided by its
discounted energy production, and it can be calculated in every simulation instance ac-
cording to Equation (15).
𝐶𝑚 + 𝐶𝑒𝑏 + 𝐶𝑒𝑠 (1 + 𝑖)𝑛 − 1
𝐶𝑝𝑖 + 𝐶𝑏,𝑟 + ∑24
𝑛=0 𝐶𝑝𝑖 + 𝐶𝑏,𝑟 + (𝐶𝑚 + 𝐶𝑒𝑏 + 𝐶𝑒𝑠 ) [ ]
(1 + 𝑖)𝑛 𝑖 (1 + 𝑖)𝑛
𝐿𝐶𝑂𝐸 = = . (15)
𝐸𝑢𝑠 (1 + 𝑖)𝑛 − 1
∑24
𝑛=0 (1 + 𝑖)𝑛 𝐸𝑢𝑠 [ ]
𝑖 (1 + 𝑖)𝑛
In this work, the LCOE was calculated in EUR/kWh, where Cpi is the total expenditure
cost, Cb,r is the discounted battery replacement cost, Cm is the annual system maintenance
cost, Ceb is the annual backup energy cost, Ces is the cost of the discarded renewable energy,
and Eus is the annual useful energy. Similar to the total economic costs at the numerator,
the Eus at the denominator is also multiplied by a discount factor which is described in
more detail in Appendix A. The annual real interest rate, here i, is estimated using the
Fisher equation [49].
𝑖′ − 𝑓
𝑖= , (16)
1+𝑓
where 𝑖 ′ the nominal interest rate and 𝑓 the annual inflation rate.

2.5.1. Cost of Battery Replacement (Cb,r)

The replacement time of batteries (TOR) varies across different accumulator technol-
ogies and depends, among other factors, on the number of charge–discharge cycles, as
Energies 2021, 14, 1988 12 of 28

well as on the battery calendar life [50]. Thus, an upper limit calendar life of 5 years was
chosen as the maximum TOR for the lead–acid technology, while, for lithium-ion technol-
ogy, an upper limit calendar life of 10 years was chosen.
Furthermore, in the context of the system’s operating cost calculation and to evaluate
if the battery needs to be replaced before the abovementioned maximum TOR (TORmax),
the number of charge–discharge cycles of the batteries at 100% DOD was estimated ac-
cording to the total electrical energy delivered by the battery (Pbatd) in a year, using Equa-
tion (17).
∑ 𝑃𝑏𝑎𝑡𝑑
𝑁𝐶 = , (17)
𝐵𝑛𝑒𝑐
where NC is the number of cycles performed by the battery in 1 year. The time of replace-
ment of each battery, expressed in years, was determined using Equation (18).
𝑇𝑜𝑡𝑎𝑙 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐶𝑦𝑐𝑙𝑒𝑠 𝑢𝑝 𝑡𝑜 𝑡ℎ𝑒 𝐸𝑛𝑑 𝑜𝑓 𝐿𝑖𝑓𝑒 𝑎𝑡 100% 𝐷𝑂𝐷
𝑇𝑂𝑅 = . (18)
𝑁𝐶
The information concerning the numerator of Equation (17) can be retrieved from the
battery manufacturer datasheet. In this work, it took the value 500 for lead–acid batteries
and the value 2500 for lithium-ion batteries.
When the battery needs to be replaced sooner than the maximum battery calendar
life (TOR < TORmax), the estimated TOR is taken into account as the replacement time of
the battery. In cases where the calculated TOR exceeds the maximum battery calendar life
(TOR > TORmax), TORmax is considered as the battery replacement time.
Since the guaranteed power output of the PV modules is 25 years, the entire PVBAT
system is considered to have the same overall cycle life. Therefore, batteries are conserva-
tively expected to be replaced multiple times over this period. The total cost of battery
replacement, at net present value, is calculated using the following equation:
𝑘=𝑛×𝑇𝑂𝑅<25 1
𝐶𝑏,𝑟 = 𝐶𝑏 ∑ , (19)
𝑘=0 (1 + 𝑖)𝑘
where Cb is the initial battery purchasing cost, k is the year of battery replacement which
takes integer values from 0 up to n ×TOR, with n receiving the values 1, 2, 3, … to the
point where n × TOR < 25. As mentioned above, the product n × TOR is only used if
TOR<TORmax. Otherwise, the product n ×TORmax is used.

2.5.2. Estimations of Annual Cm, Ceb, and Ces

The annual PVBAT maintenance cost Cm, is the sum of the individual operation and
maintenance costs of each system component during a year, including the cost of potential
services (e.g., equipment insurance, monitoring services).
The annual backup energy (Efuel), defined in Equation (20), is the required electrical
energy to preserve load supply and to charge the batteries via the generator set. This en-
ergy is converted from chemical energy contained in diesel fuel and expressed in kWh.
∑ 𝑃𝑏𝑎𝑡
𝐸𝑓𝑢𝑒𝑙 = , (20)
𝑛𝑔𝑒𝑛
where ngen is the generator efficiency. The annual backup energy cost Ceb, defined in Equa-
tion (21), is derived from the product of Efuel and the fuel cost (Fuelcost) expressed in EUR/lt,
divided by the lower heating value (LHV) of diesel fuel, i.e., 9.85 kWh/lt.
𝐸𝑓𝑢𝑒𝑙 × 𝐹𝑢𝑒𝑙𝑐𝑜𝑠𝑡
𝐶𝑒𝑏 = . (21)
𝐿𝐻𝑉
The cost of the discarded surplus energy Ces described in Equation (22) is derived
from the product of the annual sum of Ewaste in kWh and the average LCOE of PV–battery
systems expressed in EUR/kWh.
Energies 2021, 14, 1988 13 of 28

𝐶𝑒𝑠 = 𝛴𝐸𝑤𝑎𝑠𝑡𝑒 × 𝐿𝐶𝑂𝐸𝑝𝑣𝑏𝑎𝑡 . (22)

3. Results
3.1. Sizing Simulations
The sizing method proposed in this work is based on the selection of the appropriate
number of photovoltaic panels (Noptimal) and the necessary nominal energy of the energy
storage system (Bnecoptimal), so as to satisfy, without interruptions, the energy consumption
profile of an average household for a typical meteorological year in a specified location,
whilst minimizing the LCOE expressed by Equation (15).
The initial data required for two distinct and autonomous PV–battery–generator sim-
ulation examples are illustrated in Table 1. Both off-grid systems are equal except for as-
pects concerning the EES technology used. In simulation A, the load supply is facilitated
by a lead–acid battery, and, in simulation B, it is facilitated by a lithium-ion battery. In the
lead–acid case, the calendar life, cycle life, DOD, and charging efficiency were set to 5
years, 500 cycles, 50%, and 85%, respectively. In the lithium-ion case, the same parameters
were set to 10 years, 2500 cycles, 80%, and 98%, respectively. In both cases, the nominal
energy content Bnec and the PV module number N were gradually increased in every
simulation instance by steps of 5 kWh and two modules, respectively.

Table 1. Input data of two off-grid PV–battery–generator sizing simulations employing different electrical energy storage
(EES) technologies. DOD, depth of discharge; SOE, state of energy; BMS, battery management system; LCOE, levelized
cost of electricity; LHV, lower heating value.

Input Parameters Sim Data A Sim Data B Units Input Type

1 Number of PV modules, N 6 to 34 6 to 34
3 Nominal PV module power, Pmpp 310 310 Wp
2 Nominal energy capacity, Bnec 5 to 30 5 to 30 kWh
Cycles (100%
4 Nominal battery cycle life 500 2500
DOD)
5 Calendar battery life 5 10 Years
6 Depth of discharge 50 80 %
Physical model
7 SOEt=0 100 100 %
assumptions
8 Inverter efficiency, ninv 95 95 %
9 Battery charging efficiency, nbat 85 98 %
10 Pbackup_max 4000 4000 W
11 Number of inverters 1 1
12 Pload_max 4000 4000 W
13 Battery type Lead–acid Lithium-ion
14 Generator efficiency, ngen 80 80 %
15 Battery cost 154 574 EUR/kWh
16 PV module cost 110 110 EUR
17 Inverter cost 1600 1600 EUR
18 Monitoring and BMS cost 250 500 EUR
19 PV mounting system cost 50 50 EUR/mod
20 Electrical installation cost 1000 1000 EUR Economic cost as-
21 Generator cost 1300 1300 EUR sumptions
22 LCOEpvbat 0.26 0.26 EUR/kWh
23 LHV 9.85 9.85 kWh/lt
24 Cost of fuel 1.175 1.175 EUR/lt
25 Real interest rate 0.06919 0.06919 %
26 Maintenance cost, Cm 160 160 EUR/year
Energies 2021, 14, 1988 14 of 28

27 Project years 25 25 Years

3.2. Performance of Systems

The sizing model described in Section 2.4 of this work was initially validated by ex-
amining the change over time of the most important parameters of the PVBAT system for
four designated weeks of the year, one for each season. On the basis of the component
analysis in Section 2.1.2, the two systems were selected for the presentation of the curves
in Figures 6 and 7. Ceteris paribus, one includes lead–acid batteries and the other includes
lithium-ion batteries, albeit with different multiples of the nominal energy content. More-
over, both systems exhibit the same PV module type, albeit with different installed capac-
ities.
As the operating model predicts, when the power output from the photovoltaics
(black curve) is high and the corresponding SOE (blue curve) is capped at 100%, there is
surplus energy that cannot be absorbed (green curve) by the system to charge the batteries
or to meet the needs of the load (red curve).
The range in which the SOE varies depends on the battery technology used. In the
case of lead–acid technology, a lower limit of 50% was set. We notice that this limit was
approached quite often in the first week of the year (winter), while, in the 26th week (sum-
mer), the SOE did not receive values less than 60%. On the contrary, since the electricity
generation from photovoltaics in winter was lower than in summer months, in winter, the
SOE reached its maximum only with energy coming from the backup source, while, in
summer, the photovoltaic output alone could fully charge the battery. However, during
weeks 13 (spring) and 39 (autumn), the SOE approached its maximum with combined
photovoltaic and backup power. Furthermore, on day 5 of week 13 and day 7 of week 39,
a significantly reduced PV output at periods with an elevated load causes a temporal un-
dershooting of the lower SOE limit of 50%, thereby briefly draining the battery without
triggering the backup source.
Energies 2021, 14, 1988 15 of 28

Figure 6. System performance using lead-acid battery.

A similar behavior is observed in the system where a lithium-ion battery was used,
depicted in weeks 1 and 13 of Figure 7. Even though a lower SOE threshold of 20% was
set, in the first week (winter), the SOE briefly received values below this threshold. Alt-
hough this situation does not particularly affect the specific technology, this fact suggests
that, to avoid low values for SOE, the control should take place in less than 1 h.
Energies 2021, 14, 1988 16 of 28

Figure 7. System performance using lithium-ion battery.

3.2.1. Maximizing the Contribution of Solar Energy

The nominal energy content and the module quantity were gradually increased by
steps of 5 kWh and 2 modules respectively, forming distinct PV–battery arrangements for
every simulation instance of the sizing algorithm.
As the number of photovoltaic panels increases, the solar energy input increases, thus
reducing the backup energy required by the generator to cover the load and charge the
batteries when the SOE is below the cutoff setting. Moreover, the discarded surplus en-
ergy of the system increases, i.e., the solar energy that cannot be absorbed by the system.
As evident in Figure 8, maximizing the contribution of renewable energy is achieved
when the sum of the backup energy and the discarded surplus energy is minimized.
This was the case for the simulation instance where the lead–acid battery had a nom-
inal energy content of 30 kWh, where the quantity of PV panels to be used in the system
should be 12, as illustrated in Figure 9a. Figure 9b shows the optimal lithium-ion arrange-
ment which happens to be the same as the lead–acid arrangement. The simulations were
performed from 6–32 PV panels, the range of which was derived from the ratio of the
annual electricity demand to the annual electricity generation, for each N, in every simu-
lation instance, as shown in Equation (23). When considering this ratio, the number of
possible simulation instances is reduced significantly, thereby saving computing time in
the enumeration process.
∑𝑛 𝐸𝑙𝑜𝑎𝑑
= {0.4, … ,2}. (23)
∑𝑛 𝑃𝑝𝑣𝑁
The range of Bnec was limited from 5 kWh to 30 kWh, in view of the modular char-
acteristics of the batteries and according to the daily minimum and maximum energy con-
sumption of 1 year.
Energies 2021, 14, 1988 17 of 28

Figure 8. Marker chart depicting 14 simulation instances with constant Bnec and varying PV mod-
ule number. The descending change of the backup energy and the ascending change of discarded
surplus energy indicate the optimal PV panel quantity for which the solar generation is maxim-
ized.

Figure 9. Each marker chart unfolds the optimal N–Bnec arrangement in terms of maximal solar energy usage (a,c) and
LCOE minimization (b,d). Charts 9a and 9b show all the simulation instances of the lead–acid system, while charts 9c and
9d show the results of the lithium-ion system.
Energies 2021, 14, 1988 18 of 28

In both cases, the same behavior was observed: a change in the minimum of each
curve to lower values as the nominal battery energy increased. In fact, in the case of lead–
acid technology (Figure 9a), the rate of minimum reduction was greatly reduced with the
value of 20 kWh of nominal battery energy, while, in the case of lithium-ion technology,
this was observed after the nominal energy of 15 kWh. According to the above, the maxi-
mization of the solar harvest and, consequently, the minimization of the environmental
footprint were achieved using 12 photovoltaic panels and 30 kWh of nominal battery en-
ergy in both cases. However, this picture changed when we took into account the financial
data of the system for the calculation of the LCOE.

3.2.2. Minimizing the LCOE

The estimation of the LCOE, as defined in Equation (15), was done following the
same procedure described previously in Section 2.4. To be more specific, the estimation of
the LCOE was performed for two equal PV–battery–generator systems with distinct EES
technologies for a nominal battery energy range from 5 kWh to 30 kWh and for 6–32 PV
panels. The variation of the LCOE in relation to the quantity of photovoltaic panels is
presented in Figures 9b,d for the systems equipped with lead–acid and lithium-ion bat-
teries, respectively.
It is observed that the LCOE curves passed through a minimum. Regarding the sys-
tem with lead–acid battery, this minimum decreased by increasing the nominal energy of
the battery until it reached the value of 20 kWh. After this value, an increase in the LCOE
curve’s minimum was observed. The same behavior was also noted in the case of lithium-
ion batteries with the exception that the nominal energy of the battery, after which the
minimum of the LCOE curve began to receive higher values, was 15 kWh. Figure 10 de-
picts how the minimum value of each curve presented in Figure 9 varied as a function of
the nominal energy capacity of the lead–acid and lithium-ion batteries. The same graph
includes the optimum number of photovoltaic panels derived by the model.

Figure 10. In this marker chart, six simulation instances per EES technology are presented and
their calculated LCOEs are directly compared.

One would expect that an increase in the number of PV panels would be accompa-
nied by a corresponding increase in useful energy. This is not the case, since, after a certain
point, solar energy cannot be absorbed anymore due to stocked storage resources and
limited consumption. This is illustrated in Figure 11 where the annual accumulated useful
Energies 2021, 14, 1988 19 of 28

energy of all simulation instances, in both storage technologies, are shown in contrast to
the theoretically optimal photovoltaic electricity generation. Here, after the 14th PV panel,
the utilized photovoltaic electricity somewhat stabilized and significantly affected the
LCOE of the system, considering that the discounted sum of the generated energy by the
system is inversely proportional to the LCOE.

Figure 11. Comparison of the annually utilized PV output Eus, as defined in Equation (14), of every
lithium-ion (a) and lead–acid (b) simulation instance and the corresponding theoretically optimal
photovoltaic output Ppv (black marks).

The calculation of the LCOE, i.e., the unit cost of electricity, requires at first consistent
information regarding the capital cost and the operation, maintenance, and replacement
expenditures. Subsequently, these financial data can be used to prepare the essential cash
flow over a period of 25 years. Table 2 shows the detailed costs for every simulation in-
stance considered in Figure 10. Value added tax was not included, and cost figures refer
to the second part of 2020.

Table 2. The values of the parameters used to calculate the LCOE at the point where the minimum appeared.
Component Cost Maintenance Cost Operation Cost
Installation Eus LCOE
Battery N Bnec Cbr Cm Ces Ceb
PV Battery PE&M * Generator Cost (kWh/yea (EUR/kW
Type (mods) (kWh) (Discounted) (EUR/yea (EUR/year (EUR/yea
(EUR) (EUR) (EUR) (EUR) (EUR) r) h)
(EUR) r) ) r)
10 5 1100 770 2600 1300 3125 4445 100 369 503 3487.4 0.61
12 10 1320 1540 2700 1300 3350 4405 100 389 341 4409.8 0.48
Lead–Acid

18 15 1980 2310 3000 1300 4025 4289 100 834 215 5557.0 0.47
14 20 1540 3080 2800 1300 3575 5719 100 247 143 6021.8 0.34
14 25 1540 3850 2800 1300 3575 7149 100 215 126 6157.9 0.36
14 30 1540 4620 2800 1300 3575 8579 100 199 117 6226.7 0.38
10 5 1100 2870 2600 1300 3125 4972 100 282 449 3856.2 0.57
Lithium-Ion

16 10 1760 5740 2900 1300 3800 4446 100 655 205 5306.1 0.51
14 15 1540 8610 2800 1300 3575 6669 100 301 129 5792.4 0.46
14 20 1540 11,480 2800 1300 3575 8892 100 265 107 5944.1 0.51
14 25 1540 14,350 2800 1300 3575 11,115 100 241 92 6047.0 0.57
14 30 1540 17,220 2800 1300 3575 13,338 100 227 82 6106.0 0.63
* Power Electronics and Monitoring Equipment.
Energies 2021, 14, 1988 20 of 28

Each PV–battery–generator system was considered to have an overall cycle life of 25

years. Standby generator operation was kept below 500 h per year in all instances, with
maximal continuous operation limited to 4 h. Premium power electronics were conserva-
tively expected to be replaced every 10 years. The EES calendar life for lithium-ion batter-
ies was set to 10 years, while it was set to 5 years for lead–acid batteries. The determined
EES replacement points, ranging from 5–10 years for lithium-ion technology and 1.6–5
years for lead–acid technology, were incorporated in the discounted maintenance cost,
listed in the Cbr column of Table 2.

4. Discussion
In this study, a sizing method for autonomous non-interconnected PV–battery sys-
tems backed by a standby power source was presented. Emphasis was placed on mono-
crystalline photovoltaics in conjunction with DC-coupled EES systems supplying electric-
ity to a typical inelastic load profile. The standby power was simulated in the form of a
diesel generator set, taking into account the CTU, APT, and ALV thresholds. The opera-
tion and sizing algorithm described in Section 2.4 was studied and validated by examining
the variation of important system parameters for four designated weeks of the year, one
for each season. The algorithm calculated, among other factors, the useful energy Eus, a
key output parameter for which the solar harvest is maximized in parallel with the mini-
mization of the LCOE.
The results showed that the LCOE varied in different configurations with respect to
system component selection and design principles. The results also indicated a fairly
downward trend of the LCOE of PV–battery systems. More specifically, the LCOE was
calculated using Eus in the denominator of Equation (15). Furthermore, for every increase
in the nominal battery energy capacity Bnec in every simulation instance, an increase in
useful energy output was observed (Figure 11). Hence, if one disregarded the LCOE, the
sizing method would eventually point to a configuration which, on the one hand, would
fully exploit the available solar energy but, on the other hand, would lead to an oversized
battery. This is a weak point in most sizing algorithms since an oversized battery entails
a higher carbon footprint and increased costs. However, in this study, the lowest LCOE
was found to be 0.34 EUR/kWh for systems using lead–acid batteries, whereas, for systems
using lithium-ion batteries, the LCOE was found to be 0.46 EUR/kWh, which are both
considerably lower estimations than those which appeared in several other studies [25,51].
Another critical aspect to consider is the backup energy source to be used. In this
study, a standby diesel generator set was selected as a backup to compensate for energy
shortages due to intermittencies from renewables. Low battery SOE in the absence of solar
irradiance initiates the generator set, which in turn commences the charging loop until the
SOE is equal to or greater than 80%. The upper charging limit was set to 80% to ensure
that the PV and the generator would not overcharge the battery and discard energy. This
of course means that, especially in winter when the generator needs to be used more often,
the battery may have to operate at a partial state of charge. However, this is not a deter-
rent, as pointed out in [52,53]. The total generator output supplies power to the prioritized
load on the AC bus, and the remaining power is used to charge the battery on the DC bus.
Thus, CTU is equal to the battery charging time. Since the nominal power output of the
generator set was fixed, battery charging time increased in line with the increase in Bnec,
which can be confirmed from the CTU column in Table 3. Simulation instances where
CTU exceeded 5 h were ignored. The ALV was kept constant, disregarding battery charg-
ing profiles for simplification. Therefore, the generator was expected to operate invariably
at high-efficiency levels constantly providing the maximum power output, as is seen in
the ALV column of Table 3. In instances of lithium-ion batteries, elevated charging times
were mainly attributable to the greater DOD of this accumulator technology. On the con-
trary, lead–acid batteries require more annual generator operation time than lithium-ion
batteries, with APT ranging from 873 h to 196 h.
Energies 2021, 14, 1988 21 of 28

Table 3. The maximum continuous time use (CTU), the estimated annual operation time (APT), and the average load
value (ALV) of the standby diesel genrator in diverse simulation arrangemnets.

N Bnec CTU APT ALV

Battery Type
(mods) (kWh) (h) (h/year) (kW)
10 5 1 873 4000
12 10 2 571 4000
Lead–Acid

18 15 2 360 4000
14 20 3 239 4000
14 25 3 211 4000
14 30 4 196 4000
10 5 1 784 4000
Lithium-Ion

16 10 3 343 4000
14 15 4 216 4000
14 20 5 180 4000
14 25 5 158 4000
14 30 6 137 4000

Furthermore, battery aging calculated in this work provides a more realistic way to
find battery replacement time than in the literature [7,9,13]. This is due to the fact that it
takes into account the total energy amount removed from the battery compared to the
manufacturer’s specifications. However, a weak point of the method is the use of SOE
instead of state of charge (SOC). More specifically, the use of SOE does not take into ac-
count the rate at which the battery is charged and discharged as is the case with SOC
[16,54]. This means that, at high rates of battery charging and discharging, the SOE
method allows us to transfer power to and from the battery without any restriction except
for the maximum and minimum limit whereas, in the case of the SOC, under the same
conditions, the energy amounts may be smaller. Although the appearance of high currents
in applications using renewable energy sources in non-interconnected systems is not fre-
quent, future study of this method using SOC is one of the objectives of our research team.

5. Conclusions
One reason for an increase in global energy demand is economic growth, which is
strongly related to higher energy use. Power systems that take advantage of renewable
energy sources, in conjunction with energy storage systems, are effectively addressing the
challenge of rural and remote electrification while mitigating greenhouse gas emissions.
This study proposes a sizing method for off-grid electrification systems consisting of
photovoltaics, batteries, and a diesel generator set, which are based on the selection of the
appropriate quantity of PV panels and battery energy capacity whilst minimizing the
LCOE. An operation and sizing algorithm was presented, which used TMY time series of
a specified location and consistent data from datasheets of PV, converter, battery, and
generator manufacturers, to generate annual time series of the power output for one PV
module, eventually determining the annual PV output of an entire PV array in hourly
steps. The algorithm calculates, among other factors, the useful energy Eus, a key output
parameter for which the solar harvest is maximized in parallel with the minimization of
the LCOE. A load profile was synthesized and matched against the computed PV power
output. Optimal operation is validated by examining the change over time of the most
important parameters for different PVBAT systems utilizing lead–acid and lithium-ion
batteries.
Sizing was done for two single-phase DC-coupled PVBG systems differing only in
the applied battery technology, i.e., lead–acid and lithium-ion. Both systems used one in-
verter (5000 VA) and one diesel generator (6000 VA). Test results showed that, for a given
Energies 2021, 14, 1988 22 of 28

annual load of approximately 5570 kWh, the optimal PV array size for both systems con-
sisted of 14 modules (310 Wp/mod). The optimal nominal energy capacity for the lead–
acid system was 20 kWh and that for the lithium-ion system was 15 kWh. The lead–acid
system utilized 6021.8 kWh of electrical energy and wasted approximately 922.2 kWh re-
newable energy, whereas the lithium-ion system utilized 5792.4 kWh and wasted approx-
imately 1117.9 kWh. The solar harvest was maximized in both cases in a different config-
uration, i.e., using 12 photovoltaic panels and 30 kWh of battery nominal energy capacity.
The estimated LCOE for systems using lead–acid batteries was 0.34 EUR/kWh, while it
was 0.46 EUR/kWh for systems using lithium-ion batteries. Further consideration of tech-
nical aspects of the auxiliary generator set and EES technologies led to the determination
of capital costs, replacement costs, operation and maintenance costs, and fuel costs. The
tests also revealed a weak point. The sizing algorithm applied in this study constitutes an
enumerative rule-based method and is, therefore, central processing unit (CPU)-intensive.
This means that the total number of simulation instances (N and Bnec combinations) is a
tradeoff between accuracy and runtime speed. Nevertheless, runtime remains relatively
brief for small-scale applications. In large-scale applications, this can be overcome by in-
tuitively limiting the N and Bnec values.
Applying the methodology used in the present study, a proposal for future work
involves mapping the LCOE, using geographic information systems, while addressing the
meteorological uncertainties of various isolated areas such as islands. Especially in areas
without a power grid, comparing different backup energy systems and their uncertainties
could give some additional insights into varying electricity generation costs. Another rec-
ommendation for future work includes the development of a method for generating dis-
tributions of load profiles according to consumption category, which would lead to a more
general application of the methodology proposed in this work.

Author Contributions: Conceptualization, methodology, software, validation, and writing—re-

view and editing, C.E. and I.E.K. Both authors have read and agreed to the published version of the
manuscript.
Funding: This research received no external funding.
Conflicts of Interest: The authors declare no conflicts of interest.

Nomenclature
ALV Average load value of diesel generator
AOI Angle of incidence
APT Annual operation time of diesel generator
BMS Battery management system
Bnec The required battery energy capacity
Cb,r Discounted battery replacement cost,
Ceb Annual backup energy cost
Ces Cost of the discarded renewable energy
Cm Annual system maintenance cost
Cpi Total capital cost
CTU Maximum continuous time use of diesel generator
DER Distributed energy resource
DHI Diffuse horizontal irradiance
DNI Direct normal irradiance
DOD Depth of discharge
Ebatt The remaining amount of energy that is transferred to the battery at time t
EES Electrical energy storage
Eshort Energy shortage when the PV output cannot cover the power demand
Esur Surplus energy
Energies 2021, 14, 1988 23 of 28

Eus Annually utilized PV energy output

Ewaste Excess solar power that remains unused
GHI Global horizontal irradiance
I0 Diode reverse saturation current
IL Photocurrent
LCOE Levelized cost of electricity
LPG Liquefied petroleum gas
MPPT Maximum power point tracking
N The required PV module number
nbat Charging efficiency of the battery
ninv Inverter efficiency
nNsVth Modified diode ideality factor
Pbackup->battery The remaining generator output on the DC side charging the battery
PbackupMAX Maximum diesel generator output
PERC Passivated emitter and rear contact PV module
Pload The hourly power demand for in a year.
Ppv Hourly PV power output in a year
PVBG Photovoltaic system with batteries and generator set
PVLIB Solar simulation library for PV energy systems
Rs Series resistance
Rsh Shunt resistance
SA Operation and sizing algorithm
SOE Battery state of energy
SOE Battery state of energy (SOE) available
STC Standard test conditions
T Atmospheric temperature
TC PV cell temperature
TM PV module temperature
TMY Typical meteorological year
WS Horizontal windspeed

Appendix A
Proof of relationship:
𝑛
1 (1 + 𝑖)𝑛 − 1
∑ = . (A1)
(1 + 𝑖)𝑛 (1 + 𝑖)𝑛 𝑖
1

The left-hand side of the A.1 becomes

𝑛
1 1 1 1
∑ = + +. . . . + .
(1 + 𝑖)𝑛 (1 + 𝑖)1 (1 + 𝑖)2 (1 + 𝑖)𝑛
1

Setting (1 + i) = x yields

𝑛
1 1 1 1
∑ 𝑛
= 1 + 2 +. . . . + 𝑛 .
𝑥 𝑥 𝑥 𝑥
1

(𝑥−1)
Multiplying with (𝑥−1)
yields
(𝑥 − 1) 1 1 1 1 1 1 1 1 1
( 1 + 2 +. . . . + 𝑛 ) = (1 + 1 +. . . . + 𝑛−1 − 1 − 2 −. . . . − 𝑛 ) =
(𝑥 − 1) 𝑥 𝑥 𝑥 (𝑥 − 1) 𝑥 𝑥 𝑥 𝑥 𝑥
Energies 2021, 14, 1988 24 of 28

1 1 𝑥𝑛 − 1 (1 + 𝑖)𝑛 − 1 (1 + 𝑖)𝑛 − 1
(1 − 𝑛 ) = 𝑛 = = .
(𝑥 − 1) 𝑥 𝑥 (𝑥 − 1) (1 + 𝑖)𝑛 (1 + 𝑖 − 1) (1 + 𝑖)𝑛 𝑖

Appendix B

Table A1. Description of simulation parameters acquired on every simulation instance.

Parameter Description
The minimum multiple battery nominal energy capac-
p.1 Bnec
ity, in Wh.
p.2 N The PV module number in an array.
p.3 Σ(Ppv) The sum of hourly PV power output in a year, in Wh.
The sum of the hourly power demand on the consump-
p.4 Σ(Pload)
tion side in a year, in Wh.
The annual sum of energy discharging the battery, in
p.5 Σ(Pbat,d)
Wh.
p.6 Σ(Pbat,ch) The annual sum of energy charging the battery, in Wh.
p.7 Σ(Ewaste) The annual sum of unused (discarded) energy, in Wh.
p.8 Σ(Pbackup) The annual sum of the generator output, in Wh.
The annual sum of energy flowing from the generator
p.9 Σ(Pbackup->bat)
to the battery, in Wh.
The sum of operational hours of the generator in a
p.10 Σ(Backup_Operation)
year.
The ratio of the annual PV power output to the annual
p.11 Σ(Pload)_Σ(Ppv)_ratio
power demand.
Battery_Cycles = abs(Σ(Pbat,d))/Bnec
p.12 Battery_Cycles The total charge–discharge cycles of the battery for a
year.
The sum of annual discarded energy and annual en-
p.13 Σ(Εwaste) + Σ(Pbackup)
ergy supplied by the generator, in Wh.
= Σ(Ppv) − Σ(Ewaste)
p.14 Eus
Useful_Energy The utilized PV output, in Wh.
= NBC/Battery_Cycles
p.15 Bat_Replac_Years Where NBC is the nominal battery cycle life stated by
the battery manufacturer in the datasheet.
= N × PV_Module_Cost
p.16 PV_Cost(N)
The capital cost of PV panels.
= (Bnec/1000) × Bat_Cost_Lead–Acid,
p.17 Battery_Cost_Lead where Bat_Cost_Lead–Acid is the battery capital cost
per kWh.
= (Bnec/1000) × Bat_Cost_LiFePO4,
p.18 Battery_Cost_Lithium where Bat_Cost_LiFePO4 is the battery capital cost per
kWh.
= Inverter_Cost + Monitoring_Cost + N × 50, in EUR.
The capital cost for power electronics depends on the
p.19 Power_Electr_Cost(N) number of PV panels installed. The above function is
an approximation and can usually be derived by the
power electronics distributor pricelist.
p.20 Mounting_Cost(N) = N × 1_Panel_Roof_Mounting_Cost.
Energies 2021, 14, 1988 25 of 28

The capital cost for PV mounting systems can vary sig-

nificantly depending on site-specific individualities.
Although, in this work, a simple dependency on the
amount of PV panels was acceptable, a more precise
cost estimation must be considered in demanding in-
stallation sites.
= Electrical_Install_Cost + Mounting_Cost(N),
where Electrical_Install_Cost is the service cost for the
indoor electrical construction, i.e., the power electron-
ics–battery–wiring setup. Installation_Service_Cost(Ν)
Installation_Ser-
p.21 is also dependent on the amount of installed PV panels
vice_Cost(Ν)
since more modules generally mean more converters,
batteries, cabling, and mounting stands to install. This
is an empirically determined quantity and scaled to the
size of the total PV installation.
= N × 12.5 + Electrical_Install_Cost, in Euro. The material
purchasing and installation cost for the electrical wiring,
and construction, consisting mainly of low voltage
Electr&Install_Materi- protection and control equipment (Miniature circuit
p.22
als(N) breakers (MCBs), wires, enclosure, switchboards, etc.).
This is also an empirically determined quantity, scaled
to the number of PV panels and dependent on the
service cost for the indoor electrical construction.
= [PV_Cost(N) + Battery_Cost_Lead +
p.23 Σ_COST_(PV&Lead-Acid) Power_Electr_Cost(N) + Mounting_Cost(N) +
Installation_Service_Cost + Electr&Install_Materials]
= [PV_Cost(N) + Battery_Cost_Lithium +
p.24 Σ_COST_(PV&LiFePO4) Power_Electr_Cost(N) + Mounting_Cost(N) + Installa-
tion_Service_Cost + Electr&Install_Materials]
Accumulation of net present value battery replacement
p.25 Σ_Bat_Repl_Cost_Lead
cost for lead–acid batteries.
Accumulation of net present value battery replacement
p.26 Σ_Bat_Repl_Cost_Lithium
cost for LiFePO4 batteries.
= Σ(Pbackup)/ngen
p.27 Efuel The fuel energy required by the generator to cover the
load and battery demand, in kWh per year.
= Σ(Εwaste) × LCOEPVBAT/1000
The unit cost of the discarded energy, based on the es-
p.28 Ces
timated LCOE of a typical PV-Battery setup, approxi-
mately 0.26 EUR/kWh.
= Efuel × Cost_of_Fuel/Fuel_energy_vol
p.29 Ceb Cost_of_Fuel = 1.175 EUR/L and Fuel_energy_vol =
9850
= [(Ces + Ceb) × ((1 + Real_Interest) ^ (Project_Years + 1)
p.30 Cost_of_Energy − 1)/(Real_Interest × (1 + Real_Interest) ^ Pro-
ject_Years)]
= T + Generator_Cost + Σ_Bat_Repl_Cost + Cost_of_En-
p.31 Total_Cost
ergy,
Energies 2021, 14, 1988 26 of 28

where T = [Σ_COST_(PV&Lead–Acid) or
Σ_COST_(PV&LiFePO4)]
p.32 TC_UE_ratio = Total_Cost/ΣUseful_Energy (EUR/kWh)

References
1. REN21. Renewables 2020 Global Status Report; REN21 Secretariat: Paris, France, 2020; ISBN 978-3-948393-00-7.
2. El-houari, H.; Allouhi, A.; Rehman, S.; Buker, M.S.; Kousksou, T.; Jamil, A.; El Amrani, B. Design, Simulation, and Economic
Optimization of an Off-Grid Photovoltaic System for Rural Electrification. Energies 2019, 12, 4735, doi:10.3390/en12244735.
3. Bahramara, S.; Parsa Moghaddam, M.; Haghifam, M.R. Optimal planning of hybrid renewable energy systems using HOMER:
A review. Renew. Sustain. Energy Rev. 2016, 62, 609–620, doi:10.1016/j.rser.2016.05.039.
4. Khatib, T.; Ibrahim, I.A.; Mohamed, A., A review on sizing methodologies of photovoltaic array and storage battery in a
standalone photovoltaic system. Energy Convers. Manag. 2016, 120, 430–448, doi:10.1016/j.enconman.2016.05.011.
5. Hina Fathima, A.; Palanisamy, K. Optimization in microgrids with hybrid energy systems—A review. Renew. Sustain. Energy
Rev. 2015, 45, 431–446, doi:10.1016/j.rser.2015.01.059.
6. Dufo-López, R.; Bernal-Agustín, J.L.; Yusta-Loyo, J.M.; Domínguez-Navarro, J.A.; Ramírez-Rosado, I.J.; Lujano, J.; Aso, I. Multi-
objective optimization minimizing cost and life cycle emissions of stand-alone PV-wind-diesel systems with batteries storage,
Appl. Energy 2011, 88, 4033–4041, ISSN 0306-2619, doi:10.1016/j.apenergy.2011.04.019.
7. Maleki, A.; Askarzadeh, A. Optimal sizing of a PV/wind/diesel system with battery storage for electrification to an off-grid
remote region: A case study of Rafsanjan, Iran. Sustain. Energy Technol. Assess. 2014, 7, 147–153, ISSN 2213-1388,
doi:10.1016/j.seta.2014.04.005.
8. Wei, Z.; Li, X.; Xu, L.; Cheng, Y. Comparative study of computational intelligence approaches for NOx reduction of coal-fired
boiler. Energy 2013, 55, 683–692, ISSN 0360-5442, doi:10.1016/j.energy.2013.04.007.
9. Ghafoor, A.; Munir, A. Design and economics analysis of an off-grid PV system for household electrification. Renew. Sustain.
Energy Rev. 2015, 42, 496–502, ISSN 1364-0321, doi:10.1016/j.rser.2014.10.012.
10. Mandelli, S.; Brivio, C.; Colombo, E.; Merlo, M. A sizing methodology based on Levelized Cost of Supplied and Lost Energy for
off-grid rural electrification systems. Renew. Energy 2016, 89, 475–488, ISSN 0960-1481, doi:10.1016/j.renene.2015.12.032.
11. Mandelli, S.; Brivio, C.; Colombo, E.; Merlo, M. Effect of load profile uncertainty on the optimum sizing of off-grid PV systems
for rural electrification. Sustain. Energy Technol. Assess. 2016, 18, 34–47, ISSN 2213-1388, doi:10.1016/j.seta.2016.09.010.
12. Shang, C.; Srinivasan, D.; Reindl, T. An improved particle swarm optimization algorithm applied to battery sizing for stand-
alone hybrid power systems. Int. J. Electr. Power Energy Syst. 2016, 74, 104–117, ISSN 0142-0615, doi:10.1016/j.ijepes.2015.07.009.
13. Rodríguez-Gallegos, C.D.; Yang, D.; Gandhi, O.; Bieri, M.; Reindl, T.; Panda, S.K. A multi-objective and robust optimization
approach for sizing and placement of PV and batteries in off-grid systems fully operated by diesel generators: An Indonesian
case study. Energy 2018, 160, 410–429, ISSN 0360-5442, doi:10.1016/j.energy.2018.06.185.
14. Li, J. Optimal sizing of grid-connected photovoltaic battery systems for residential houses in Australia. Renew. Energy 2018, 136,
1245–1254, ISSN 0960-1481, doi:10.1016/j.renene.2018.09.099.
15. Wu, J.; Wei, Z.; Liu, K.; Quan, Z.; Li, Y. Battery-Involved Energy Management for Hybrid Electric Bus Based on Expert-Assis-
tance Deep Deterministic Policy Gradient Algorithm. IEEE Trans. Veh. Technol. 2020, 69, 12786–12796,
doi:10.1109/TVT.2020.3025627.
16. Wu, J.; Wei, Z.; Li, W.; Wang, Y.; Li, Y.; Sauer, D.U. Battery Thermal- and Health-Constrained Energy Management for Hybrid
Electric Bus Based on Soft Actor-Critic DRL Algorithm. IEEE Trans. Ind. Inform. 2021, 17, 3751–3761,
doi:10.1109/TII.2020.3014599.
17. Aldersey-Williamsa, J.; Rubert, T. Levelized cost of energy—A theoretical justification and critical assessment. Energy Policy
2019, 124, 169–179, doi:10.1016/j.enpol.2018.10.004.
18. Lai, C.S.; McCulloch, M.D. Levelized cost of electricity for solar photovoltaic and electrical energy. Appl. Energy 2017, 190, 191–
203, doi:10.1016/j.apenergy.2016.12.153.
19. Available online: https://www.energy.gov/eere/solar/crystalline-silicon-photovoltaics-research (accessed on 20 January 2021).
20. Available online: https://www.nrel.gov/docs/fy20osti/77010.pdf (accessed on 20 January 2021).
21. Available online: https://www.solarcall.it/home-en/download (accessed on 20 January 2021).
22. Kosmadakis, I.; Elmasides, C. Towards performance enhancement of hybrid power supply systems based on renewable energy
sources. Energy Procedia Elsevier 2018, 157, 977–991.
23. Available online: https://www.victronenergy.gr/media/pricelist/WEB_WithoutIP65_PricelistVictron2020-Q4.pdf (accessed on).
24. Ayeng’o, S.P.; Schirmer, T.; Kairies, K.; Axelsen, H.; Sauer, D.U. Comparison of off-grid power supply systems using lead-acid
and lithium-ion batteries. Sol. Energy 2018, 162, 140–152, ISSN 0038-092X, doi:10.1016/j.solener.2017.12.049.
Energies 2021, 14, 1988 27 of 28

25. Kosmadakis, I.E.; Elmasides, C.; Eleftheriou, D.; Tsagarakis, K.P. A Techno-Economic Analysis of a PV-Battery System in
Greece. Energies 2019, 12, 1357.
26. Liu, N.; Senthil, R.A.; Zhang, X.; Pan, J.; Sun, Y.; Liu, X. A green and cost-effective process for recovery of high purity α-PbO
from spent lead acid batteries. J. Clean. Prod. 2020, 267, 122107, ISSN 0959-6526, doi:10.1016/j.jclepro.2020.122107.
27. Available online: https://www.systems-sunlight.com/wp-content/uploads/2018/10/sun-
light_res_sopzv_eu_us_2_19_6p_web_final.pdf (accessed on 20 January 2021).
28. Available online: https://www.victronenergy.com/upload/documents/Datasheet-12,8-&-25,6-Volt-lithium-iron-phosphate-bat-
teries-Smart-EN.pdf (accessed on 20 January 2021).
29. Keil, P.; Simon; Schuster, F.; Wilhelm, J.; Travi, J.; Hauser, A.; Karl, R.C.; Jossen, A. Calendar Aging of Lithium-Ion Batteries. J.
Electrochem. Soc. 2016, 163, A1872, doi:10.1149/2.0411609jes.
30. Li, D.; Danilov, D.L.; Xie, J.; Raijmakers, L.; Gao, L.; Yang, Y.; Notten, P.H.L. Degradation Mechanisms of C6/LiFePO4 Batteries:
Experimental Analyses of Calendar Aging. Electrochim. Acta 2016, 190, 1124–1133, ISSN 0013-4686,
doi:10.1016/j.electacta.2015.12.161.
31. Yilmaz, S.; Dincer, F. Optimal design of hybrid PV-Diesel-Battery systems for isolated lands: A case study for Kilis, Turkey.
Renew. Sustain. Energy Rev. 2017, 77, 344–352, ISSN 1364-0321, doi:10.1016/j.rser.2017.04.037.
32. Marqusee, J.; Jenket, D. Reliability of emergency and standby diesel generators: Impact on energy resiliency solutions. Appl.
Energy 2020, 268, 114918, ISSN 0306-2619, doi:10.1016/j.apenergy.2020.114918.
33. Benton, K.; Yang, X.; Wang, Z. Life cycle energy assessment of a standby diesel generator set, Journal of Cleaner Production,
2017, 149, 265–274, ISSN 0959-6526, doi:10.1016/j.jclepro.2017.02.082.
34. Kosmadakis, I.E.; Elmasides, C.; Koulinas, G.; Tsagarakis, K.P. Energy Unit Cost Assessment of Six Photovoltaic-Battery Con-
figurations. Renew. Energy 2021, ISSN 0960-1481, doi:10.1016/j.renene.2021.03.010.
35. Maxwell, E.L. A Quasi-Physical Model for Converting Hourly Global to Direct Normal Insolation. Aug. 1987 (SERI/TR-215-
3087). pp. 35–46. Available online: http://rredc.nrel.gov/solar/pubs/PDFs/TR-215-3087.pdf (accessed on 21 August 2018).
36. Loutzenhiser, P.G.; Manz, H.; Felsmann, C.; Strachan, P.A.; Frank, T.; Maxwell, G.M. Empirical validation of models to compute
solar irradiance on inclined surfaces for building energy simulation. Sol. Energy 2007, 81, 254–267.
37. King, D.L.; Boyson, W.E.; Kratochvil, J.A. Photovoltaic Array Performance Model. Sandia Report 2004 (SAND2004-3535). Avail-
able online: http://www.osti.gov/servlets/purl/919131-sca5ep/ (accessed on 17 August 2018).
38. De Soto, W.; Klein, S.A.; Beckman, W.A. Improvement and Validation of a Model for Photovoltaic Array Performance. Sol.
Energy 2006, 80, 78–88.
39. Hongmei, T.; Mancilla-David, F.; Ellis, K.; Muljadi, E.; Jenkins, P. A Cell-to-Module-to-Array Detailed Model for Photovoltaic
Panels. Sol. Energy 2012, 86, 2695–2706.
40. Gray, J.L. The Physics of the Solar Cell; John Wiley & Sons, Ltd.: Hoboken, NJ, USA, 2003; Chapter 3; pp. 62–112, ISBN 0-471-
49196-9, doi:10.1002/9780470974704.ch3.
41. Holmgren, W.F.; Hansen, C.W.; Mikofski, M.A. pvlib python: A python package for modeling solar energy systems. J. Open
Source Softw. 2018, 3, 884, doi:10.21105/joss.00884.
42. Stein, J.S.; Holmgren, W.F.; Forbess, J.; Hansen, C.W.H. PVLIB: Open Source Photovoltaic Performance Modeling Functions for
Matlab and Python. In Proceedings of the IEEE 44th Photovoltaic Specialist Conference, Portland, OR, USA, 5–10 June 2017; pp.
3425–3430.
43. Andrews, R.W.; Stein, J.S.; Hansen, C.W.H.; Riley, D. Introduction to the Open Source PV LIB for Python Photovoltaic System
Modelling Package. In Proceedings of the IEEE 40th Photovoltaic Specialist Conference, Denver, CO, USA, 8–13 June 2014; pp.
170–174.
44. Holmgren, W.F.; Groenendyk, D.G. An Open Source Solar Power Forecasting Tool Using PVLIB-Python. In Proceedings of the
IEEE 43rd Photovoltaic Specialists Conference, Portland, OR, USA, 5–10 June 2016; pp. 972–975.
45. Stein, J.S. The Photovoltaic Performance Modeling Collaborative (PVPMC). In Proceedings of the 2012 38th IEEE Photovoltaic
Specialists Conference, Austin, TX, USA, 3–8 June 2012; pp. 3048–3052.
46. Available online: https://ec.europa.eu/jrc/en/PVGIS/tools/tmy (accessed on 17 August 2018).
47. Andersen, F.M.; Baldini, M.; Hansen, L.G.; Jensen, C.L. Households’ hourly electricity consumption and peak demand in Den-
mark. Appl. Energy 2017, 208, 607–619, ISSN 0306-2619, doi:10.1016/j.apenergy.2017.09.094.
48. Short, W.; Packey, D.; Holt, T. A Manual for the Economic Evaluation of Energy Efficiency and Renewable Energy Technologies.
National Renewable Energy Laboratory—March 1995. Available online: https://www.nrel.gov/docs/legosti/old/5173.pdf (ac-
cessed on 20 January 2021).
49. Blejer, M.I.; Eden, B. A note on the specification of the fisher equation under inflation uncertainty. Econ. Lett. 1979, 3, 249–255,
ISSN 0165-1765, doi:10.1016/0165-1765(79)90126-5.
Energies 2021, 14, 1988 28 of 28

50. Dufo-López, R.; Cortés-Arcos, T.; Artal-Sevil, J.S.; Bernal-Agustín, J.L. Comparison of Lead-Acid and Li-Ion Batteries Lifetime
Prediction Models in Stand-Alone Photovoltaic Systems. Appl. Sci. 2021, 11, 1099, doi:10.3390/app11031099.
51. Kost, C.; Shammugam, S.; Juelch, V.; Nguyen, Hu.; Schlegl, T. Study: Levelized Cost of Electricity—Renewable Energy Technologies;
Fraunhofer ISE: Freiburg, Germany: 2018.
52. Newnham, R.H.; Baldsing, W.G.A. Benefits of partial-state-of-charge operation in remote-area power-supply systems. J. Power
Sources 2002, 107, 273–279, ISSN 0378-7753, doi:10.1016/S0378-7753(01)01007-2.
53. de Vries, H.; Nguyen, T.T.; Veld, B.O.H. Increasing the cycle life of lithium ion cells by partial state of charge cycling. Micrroe-
lectronics Reliability 2015, 55, 2247-2253, ISSN 0026-2714, doi:10.1016/j.microrel.2015.08.014.
54. Wei, Z.; Zhao, J.; He, H.; Ding, G.; Cui, H.; Liu, L. Future smart battery and management: Advanced sensing from external to
embedded multi-dimensional measurement. J. Power Sources 2021, 489, 229462, ISSN 0378-7753, doi:10.1016/j.jpow-
sour.2021.229462.