Energy 161 (2018) 737e743

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

The impact of complementarity on power supply reliability of small

scale hybrid energy systems
Jakub Jurasz a, *, Alexandre Beluco b, Fausto A. Canales c
a
AGH University, Faculty of Management, Department of Engineering Management, 30 Mickiewicza Ave., 30-059 Krako
w, Poland
b
Instituto de Pesquisas Hidraulicas, Universidade Federal do Rio Grande do Sul, Av Bento Goncalves, 9500, Caixa Postal 15029, Bairro Agronomia, 91570-
901 Porto Alegre, Rio Grande do Sul, Brazil
c
Department of Civil and Environmental, Universidad de la Costa, Calle 58 #55-66, 080002 Barranquilla, Atla
ntico, Colombia

a r t i c l e i n f o a b s t r a c t

Article history: Small scale hybrid power systems gain popularity around the world as a viable way of reducing power
Received 10 April 2018 generation environmental impact, reducing energy cost and increasing power supply reliability. Hybrid
Received in revised form systems which are based on variable renewable sources usually utilize the effect of resources temporal,
16 July 2018
and to a smaller extent, spatial complementarity. Although there is already an extensive body of liter-
Accepted 26 July 2018
Available online 27 July 2018
ature investigating the concept of resources complementarity, they rarely addressed the impact of
complementarity on the reliability of a given hybrid system. In this paper we simulate the operation of
wind and solar hybrid energy system (with and without battery) for evenly distributed 86 locations in
Keywords:
Photovoltaics
Poland over the period 2010e2016 based on 15 min’ time step data. We analyze the impact of resources
Wind turbine complementarity (on various time scales: 15 min, hourly and monthly) on the system reliability. To
Hybrid energy source remove the capacity factor (resources availability depends on location) on the results, we select the
Complementarity installed capacity in wind and solar sources in such a way that on an annual scale they generate evenly
Reliability 50% of the observed demand (which is assumed to be constant ¼ 1 kW). We investigate the impact on
complementarity in the system reliability for hybrids with and without energy storage. The second part
of the paper deals with the problem of simulating the system reliability (in terms of Loss of Load
Parameter) based on multiple linear regression and artificial neural networks. The results indicate that
both temporal complementarity (expressed as coefficient of correlation) and storage capacity has non-
linear impact on the hybrid system capacity to cover the load. Generated relations between
mentioned factors show how complementarity indices may be used to size the solar-wind hybrids. The
follow up studies should concentrate on analyzing the operation of hybrids utilizing more than two
energy sources and juxtaposing complementarity based reliability assessment with other methods.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction literature are the ones that combine: solar-wind, solar-hydro,

Due to the rapid technology improvements and progress in the
manufacturing processes (mainly an effect of scale) over the recent
years the hybrid renewable energy systems (HRES) attracted sig-
nificant attention as reliable and affordable energy sources, which
are not only capable of providing electricity on a small scale by
utilizing locally available resources but also as viable means of
improving both life and environment quality.
The HRES usually exploits two or more renewable energy
sources (RES). To the most commonly configurations addressed in
* Corresponding author.
E-mail address: jakubkamiljurasz@gmail.com (J. Jurasz).
hydro-wind and solar-wind-hydro. A detailed analysis/review of so
far investigated concepts can be found in Refs. [1,2]. Often such
hybrids will be equipped with dispatchable sources such as diesel
generators, or with a form of energy storage like battery banks or
fuel cells. Additionally, they would be often a part of a microgrid
which facilitates the integration of various distributed generators
and also provides multiple end-use needs. Microgrids are charac-
terized by a low application and design costs and therefore they can
be adapted by a locally available technology [3,4]. The need for
utilizing the above-mentioned sources comes from the fact that the
underlying assumption which is a foundation of the HRES is that
the climate/weather driven variable renewable energy sources
(VRES) tend to exhibit either temporal or spatial or spatial-

https://doi.org/10.1016/j.energy.2018.07.182
0360-5442/© 2018 Elsevier Ltd. All rights reserved.
738 J. Jurasz et al. / Energy 161 (2018) 737e743
Abbreviations
ANN artificial neural network
BFGS BroydeneFletchereGoldfarbeShanno
CF capacity factor
HRES hybrid renewable energy system
LOLP loss of load probability
MLR multiple linear regression
PV photovoltaics
RES renewable energy source
STD standard deviation
VRES variable renewable energy source
WT wind turbine
Indices
I unit of time
K site/location
temporal complementarity. However, this phenomenon herein-
after referred to as “complementarity” is not perfect and tend to
vary in its range depending on considered time scale. As an
example, it is worth to imagine two energy sources which power
generation follows sine functions with the same amplitudes (see
Fig. 1). If they are out of phase by p/2, then they would exhibit a
perfect complementarity (which also can be expressed as a coef-
ficient of correlation equal to 1. If the correlation between those
two-time series is different than mentioned “-1” then the aggre-
gated energy output of both sources will start to exhibit significant
variation on time domain.
The power output fluctuation aforementioned, might signifi-
cantly limit the HRES ability to cover the demand load, thereby
reducing the system reliability and in consequence leading to
financial losses or other undesirable events like spoiled food or
vaccines. There are several approaches which can be used to
overcome this problem. Either together or separately one can
apply:
 already mentioned energy storage;
 swiftly ramping up and down dispatchable sources;
 another solution is to oversize the whole system in such a
manner which will ensure that the considered load is always
covered.
All approaches have their drawbacks and advantages and the
optimal choice will be usually based on a multi-criteria
optimization.
The concept of resources complementarity has been addressed
in several research papers. Authors have investigated comple-
mentarity in order to assess its potential to reduce required energy
storage, impact on the net load or the changes in complementarity
as a result of climate variability. For example, Agelidis and Shriv-
astava [5] investigated the correlation of complementary solar and
wind resources with the electricity load. Hoicka and Rowlands [6],
have shown solar and wind resources complementarity as a po-
tential advancing option for renewables integration into Ontario
power system. Monforti et al. [7], investigated the wind and solar
resources complementarity in Italy based on a Monte Carlo
approach. De Jong et al. [8], analyzed the relationship between solar
and wind energy production and its relationship to the electric load
and hydropower availability in Brazil, stating that solar and wind
power can increase water savings. Solomon et al. [9], performed an
analysis of wind and solar resources complementarity impact on
Parameters, Constants & Variables
CFWT wind turbine capacity factor [%]
CFPV photovoltaics capacity factor [%]
CIq 150 complementarity index []
CIh hourly complementarity index []
CId daily complementarity index []
CIm monthly complementarity index []
EB battery state of charge [kWh]
EB_Max battery rated capacity [kWh]
EG energy generation [kWh]
ED energy demand
EDef not covered demand [kWh]
EPV energy generation from PVs [kWh]
EWT energy generation from wind turbine [kWh]
PS source rated capacity [kW]
hChar battery charging efficiency [%]
hDis battery discharging efficiency [%]
the storage requirements. Liu et al. [10], scrutinized the solar and
wind resource spatiotemporal complementarity on an hourly basis
in China whereas Xu et al. [11], created a solar-wind complemen-
tarity map for China. For the case study of China also Zheng et al.
[12], used reanalysis data for extraction of solar-wind spatiotem-
poral synergy patterns based on k-clustering method. Thomaidis
et al. [13], have shown that complementary nature of solar and
wind resources can increase the system reliability; dos Anjos et al.
[14], have analyzed long-term correlations and cross-correlation
between wind speed and solar radiation on a selected island near
Brazil. Santos-Alamillos et al. [15], presented how spatiotemporal
correlation between solar and wind smooths their combined power
output in case of Iberian Peninsula. Li et al. [16], aimed at modelling
and assessing the wind and solar resources complementarity in
Oklahoma (USA). Prasad et a. [17], analyzed solar and wind re-
sources synergy in case of Australia and have shown the MERRA
(Modern-Era Retrospective analysis for Research and Applications)
datasets are suitable tool for such research. Migilietta et al. [18],
analyzed the solar and wind resources complementarity from the
meteorological perspective. Whereas Bett and Thorton [19] per-
formed a similar analysis from a climatological perspective in case
of Britain. Castro and Crispim [20] analyzed the variability and
correlation between renewable sources in case of Portugal and
found strong complementarity between solar and wind energy for
typical summer and winter days. Gutierrez et al. [21], performed a
spatial complementarity analysis between generators using only
one energy resource (solar irradiation) and created complementary
clusters for the whole Iberian Peninsula.
In addition to the above mentioned works other authors have
addressed the complementarity phenomena between other re-
sources, like: solar-hydro [22e26], hydro-offshore wind [27],
hydro-onshore wind [28,29], solar/wind-hydrokinetic [30], solar/
wind-hydro [31,32]. Also recently Risso et al., proposed a new
method for assessing the spatial complementarity between wind
and hydro energy (also applicable to other sources) based on
complementarity roses [33]. As can be seen the body of literature
on this topic has significantly grown over the recent years. This is
undoubtedly a result of an increasing role of renewables in modern
power systems.
Considering state of the art presented above and the aspects of
the HRESs described in earlier paragraphs, this paper has two ob-
jectives, the first one is to investigate the impact of wind and solar
resources temporal complementarity on their reliability. In other
words, we want to assess how the tendency of solar (photovoltaics)
 J. Jurasz et al. / Energy 161 (2018) 737e743 739

and wind (wind turbine) to complement each other impacts the

HRES capability of covering the load. The second objective is to
assess if it’s possible to model the HRES reliability, given a demand
load, and based on capacity factors for wind and solar generation,
complementarity indices and storage capacity.
The main novelty of our contribution lays in approaching the Wind
problem of small HRES reliability from the perspective of solar and
turbine
wind resources temporal complementarity. We investigate the
impact of complementarity indices (on different time scales) on the
system capability to cover the load.
The remainder of this paper is organized as follows. Section 2 Constant
briefly describes the investigated problem, summarizes consid- load
ered data and made assumptions made regarding HRES simulation
and optimization. Section 3 presents and discusses obtained re- PV
sults. In Section 4 we provide a short summary along with several
prospective future research directions.
installation

2. Problem description, data and methods

2.1. Hybrid system configuration and simulation

Battery bank
Fig. 2. Conceptual design of PV-WT-battery system. The red square indicates regula-
In our analysis we have considered the most common type of tion and conversion components.
HRES, specifically one which consists of wind and solar generators
as well as an energy storage in form of a battery bank, Fig. 2. The
energy yield from photovoltaics has been calculated as in Ref. [34]
P
n
whereas in case of the wind turbine we have used a power curve o EiG
1 kW turbine manufactured by Kestrel (http://www.kestrelwind. CF ¼ i¼1
*100% (1)
co.za) as in Ref. [35]. The management of energy flow and the nP S
operation of battery bank has been simulated as shown on Fig. 3. To
where: EiG e energy generated from given source [kWh], P S e
simulate the behavior of battery bank (charging and discharging
installed capacity [kW], n e number of considered sub periods
processes) it was a used modified version of formula found at [36].
usually accounting to whole year, namely 8760 h
Potential energy surpluses (energy neither used to cover the load
nor stored) are disregarded and are not subject of our analysis. P
n
Def
Ei
i¼1
2.2. Research procedure LOLP ¼ *100% (2)
Pn
EiD
The procedure represented in Fig. 3, has been applied at 86 i¼1

(k ¼ 86) selected locations in order to determine the impact of

temporal complementarity on the system reliability. For k ¼ 1 the
system was sized under fixed hourly load (1 kW). The installed
capacity of wind turbines and photovoltaics has been selected
based on an assumption that each of them should generate 50% of
the energy demand on an annual scale. In case of each location k we
have gathered data on: WTs and PVs capacity factors (CF, Eq. (1)),
solar and wind energy complementarity on a monthly, daily, hourly
and quarterly (150 ) time scale, LOLP value (Eq. (2)) and installed
capacity in PVs and WTs. The capacity of the battery bank was
arbitrarily assumed to range within 0e96 kWh.
where: EiDef e energy deficit [kWh], EiD e energy demand [kWh]
The complementarity index was calculated based on Pearson
correlation coefficient due to the ease of its interpretation, although
other approaches can be used [22].
2.3. Input data
Irradiation time series has been downloaded from Ref. [37],
wind speed from Ref. [38] whereas air temperature from Ref. [37]
(temperature was used to simulate PV energy yield). Time series
describing wind speed and irradiation came from satellite mea-
surements, however our previous research [39] has shown their
good fit with ground measurements. All three above mentioned
time series covered period 2010e2016 with a 15-min time step.
Data has been downloaded for 86 evenly distributed locations in
Poland (excluding maritime regions). The charge and discharge
efficiency of the battery bank was set to be linear and equal to 90%.
The self-discharge effect was neglected.

3. Results and discussion

A significant wind resource variation was observed for the sites

considered (as shown on Fig. 4). The resulting capacity factors for
wind generation ranged from slightly over 2.5% to almost 23%. The
Fig. 1. Various degrees of two renewable energy sources (green and yellow) temporal ones for PV generation were fairly stable and their average value
complementarity. was 10.9%.
740 J. Jurasz et al. / Energy 161 (2018) 737e743

NO YES
IF
EPV+EWT-ED < 0

LOAD COVERED
LOAD NOT COVERED,
CHARGE
USE BATTERY
BATTERY

NO YES
IF
NO YES
EB + (EPV+EWT-ED) CHAR
> EB_MAX
IF
E + (E +E -ED) (1/
B PV WT DIS
) 0

NEW BATTERY LEVEL LOAD NOT COVERED

EB + (EPV+EWT-ED) CHAR EB = EB_MAX LOAD COVERED COMPLETELY
EDEF = ED -EPV-EWT - DIS EB

Fig. 3. HRES simulation procedure flowchart.

Analyzed solar and wind resources exhibited almost lack of
complementarity on a 15 min and hourly time scale in terms of
correlation coefficient (CC). Observed values ranged from 0.15 to
0.08. The CC values for daily and monthly complementarity were
significantly higher and ranged respectively from 0.42 to 0.24
and 0.87 and 0.67. Although the complementarity on a daily and
monthly basis was higher from the perspective of satisfying the
load, it is crucial to maintain a stable level of power supply on a
short time scale. Following scatter plots (Fig. 5) visualize the rela-
tionship between resources complementarity and the system
(considering no storage) reliability in terms of LOLP parameter. It is
easily observed that the higher the resources temporal comple-
mentarity (lower values on the vertical axis) the higher the system
reliability. In other words, the higher the tendency of resources to
complement each other the higher the probability that the load will
be covered. It is worth to point out that the resources comple-
mentarity indices (especially on a 15 min and hourly time scale)
tend to be significantly correlated. This known relation may be used
to estimate the complementarity indices on short time scale when,
for example, complementarity values on a monthly time scale are

Fig. 4. Box-plot for capacity factors of wind and solar generations in 86 considered
sites. Fig. 5. Complementarity index impact on the system loss of load probability.
 J. Jurasz et al. / Energy 161 (2018) 737e743 741

known.
In order to get a better understanding of the relationship be-
tween resources complementarity and system reliability we have
created supply synthetic time series (by combination of sine
functions depicting solar and wind yields) which exhibited various
complementarity indices ranging from 1 to 1. Our analysis
focused on hourly time series as the selected load did not exhibit a
variation on a daily or monthly time scale. The use of synthetic data
shows that the relationship between resources complementarity
and system reliability is not linear (Fig. 6). Perfectly complementary
sources (complementarity index equal to 1) are capable of always
covering the load (assuming that non-appearance of equipment
failure), whereas those which do not complement each other Fig. 7. Impact of energy storage potential on the system reliability for all considered
(perfectly correlated, complementarity index ¼ 1) will not cover locations.
over 32% of the observed load.
The results presented at Fig. 6 are representative for a solar-
and the system reliability (LOLP). This relation was evaluated for all
wind hybrid not equipped with energy storage. As the power
86 sites and for various storage capacities. As can be seen, an in-
output of PVs and wind turbines changes almost simultaneously
crease in storage capacity weakens the correlation between hourly
with varying availability of the primary energy, therefore there is a
complementarity index and the LOLP parameter. It shows that the
very limited potential to increase their natural complementarity.
greater the storage capacity the lower impact on system reliability
Therefore, unlike in case of hydro-solar or hydro-wind systems, it is
has the complementarity. However, in case of a daily and monthly
not possible to artificially increase their complementarity by the
complementarity, adding up to 14 kWh in storage makes the rela-
available storage in form of pondage/reservoir. The problem of
tionship between resources complementarity index and system
relatively low tendency of solar-wind resources to complement
reliability stronger and after exceeding 14 kWh of storage it oscil-
each other on a short time scale is usually addressed by the addition
lates at a high range between 0.8 and 0.7.
of battery banks, diesel generators, fuels cells or other dispatchable
Undoubtedly this is a complex phenomenon which can be better
sources. Here, a battery bank was considered. The Fig. 7 displays the
understood by means of the results shown on Fig. 9. It presents the
impact of adding storage capacity to the solar-wind hybrid on the
reduction of the loss of load probability parameter as a conse-
observed reliability. Box and whiskers plots present results for all
quence of adding storage capacity. To calculate the impact on LOLP
considered 86 locations. It can be seen this relation is also not
a solar-wind hybrid without storage was considered as a bench-
linear, and adding up to 16 kWh of storage tends to significantly
mark. It can be observed that for the lowest storage capacity the
reduce the loss of load probability (each 1 kWh of storage reduces
reduction of the LOLP parameter is almost the same for all
LOLP by 3% points). Adding more storage than 16 kWh does not
considered values of the daily complementarity index. The situa-
significantly affect the system reliability. Interestingly the addi-
tion changes with an increase of storage capacity (as the premises
tional storage capacity does not only reduce the mean observed
from the Fig. 8 initially indicated) and it can be seen that not only a
value of LOLP but also the range of observed values within all sites.
decrease in complementarity index makes the LOLP reduction
Based on the results described in the previous paragraphs and in
values slightly smaller but also their variability is greater within the
Figs. 6 and 7, it can be stated that both factors affect the HRES
same storage capacity level.
reliability in nonlinear way. Fig. 8 shows the relationship between
complementarity indices for hourly, daily and monthly time scales
3.1. Reliability modelling

Considering the complexity of the combined impact of resources

complementarity and storage capacity this subsection aims at
modelling the system reliability based on available HRES parame-
ters and characteristics of the renewable sources.
The input variables selection for statistical models such as MLR

Fig. 8. Values of correlation between observed Loss of Load Probability for given
Fig. 6. Impact of the complementarity index on the system reliability. storage capacity and values of complementarity indices.
742 J. Jurasz et al. / Energy 161 (2018) 737e743

complex issue. Considering five different activation functions, a

Fig. 9. Impact of adding given storage capacities (chart legend) on the reduction of
observed values of LOLP parameter depending on the daily complementarity index.
total of 200 different variants of the ANN architecture were
analyzed. During the ANN creation an available algorithm in Sta-
tistica of the type BroydeneFletchereGoldfarbeShanno was used.
The best performing ANN model had 6 neurons in the input
layer, 10 neurons in the hidden layer and 1 neuron in the output
layer. In both cases, hyperbolic tangent activation function was
used. Based on the testing subset the ANN performed very well, and
it was possible to model the LOLP parameter with a mean absolute
percentage error (MAPE) less than 1.5%. The resulting MLR equation
(Eq. (3)) clearly was not (unlike ANN) able to grasp the non-linear
dependence between battery capacity/complementarity index
and the loss of load probability. In terms of MAPE criterion it per-
formed relatively poorly, and was characterized by an average error
of 15%. Fig. 10 visualizes in form of scatter plot the modelling ac-
curacy of both models. As can be seen the ANN model significantly
outperformed traditional MLR.

LOLP ¼ 0:167CF WT  0:035CF PV  7:996CIq þ 7:782CI h

and ANN is a complex task, and its outcome strongly impacts the
quality and performance of the final model. When the number of
potential input variables is limited, and there is enough computing
power, an exhaustive search can be applied which will deliver
models based on all potential configurations of input variables and
the best performing model will usually be selected. According to
May et al. [40], the optimal input variables set (exogenous vari-
ables) will have the smallest sufficient number variables capable of
describing the changes in the output variable, and be characterized
by a minimal level of redundancy (strongly correlated input vari-
ables) and missing non-informative variables. May et al. [40],
pointed to the existence of a multitude of methods for input vari-
ables selection, although only exhaustive search ensures finding
the optimal configuration.
Based on the stated above we have applied the following
approach: apart from capacity factor of PVs (CFPV), all input vari-
ables are significantly correlated with output variable (Table 1);
complementarity indices on hourly and 15 min’ time scale are
perfectly correlated so one can be removed, so the authors decided
to not include 15 min’ complementarity indices.   The remaining
variables did not exhibit a strong collinearity ðrxy  > 0:9Þ and were
considered in the MLR and ANN models creation.
Statistica (version 12) software was used to create the MLR and
ANN models. In case of the MLR model the whole set was used to
optimize the model. Whereas during the ANN model formulation, it
is crucial to follow a strict procedure and divide the input sets into
training, validation and testing subsets which will be used to design
the most appropriate architecture of the neural network. The ANN
architecture refers to, foremost, the number of neurons in the
hidden layer and the activation functions of the hidden and the
output layer. Here an assumption was made that the number of
neurons in the hidden layer will range from 3 to 10, based on rule
provided by Ref. [41] - it is important to note that this problem is a
þ 0:213CId þ 0:101CIm  0:002EB Max
þ 0:452 (3)
Where: LOLP e loss of load probability, CF WT e capacity factor for
wind generation, CF PV e capacity factor for PV generation, CI q e
quarterly complementarity index, CI h - hourly complementarity
index, CI d e daily complementarity index, CI m e monthly
complementarity index, EB Max e battery bank capacity.
4. Conclusions
The objective of this article was to investigate the impact of solar
and wind resources temporal complementarity impact on the mi-
cro hybrid system reliability in terms of covering fixed load. Addi-
tional, analysis was carried out to see the relationship between
energy storage, resources complementarity and reliability. Multiple
linear regression and artificial neural network models were applied
to simulate the loss of load probability based on known ahead ca-
pacity factors, complementarity indices and battery bank capacity.
The conducted analysis enables us to formulate the following
conclusions:
 the impact of resources complementarity on the system reli-
ability is similar to the impact of battery storage, and is non-
linear;
 synchronously operating solar and wind sources (correlation
coefficient of “1”), characterized by the same performance in
terms of power and energy output will cover slightly above 70%
of the load;
 storage capacity increases the system reliability but only to a
certain extent this impact is significant, additional storage over

Table 1
Correlation coefficients between input variables, marked correlations are significant
at p < 0.05.

CFWT CFPV CIq CIh CId CIm EB_Max LOLP

WT
CF 1 ¡0.554 ¡0.463 ¡0.463 ¡0.349 ¡0.322 0.000 ¡0.093
CFPV 1 0.014 0.015 ¡0.247 ¡0.094 0.000 0.001
q
CI 1 1.000 0.871 0.722 0.000 0.099
CIh 1 0.872 0.723 0.000 0.099
CId 1 0.804 0.000 0.128
CIm 1 0.000 0.127
EB_Max 1 ¡0.778
Fig. 10. Scatter plots presenting reliability modelling tests based on artificial neural
LOLP 1
network model and multiple linear regression.
 J. Jurasz et al. / Energy 161 (2018) 737e743
certain value will have minor impact on the loss of load
parameter;
 adding storage capacity weakens the relation between loss of
load parameter and the complementarity index (coefficient of
correlation) on an hourly time scale, showing that the storage
capacity in case of such hybrids has a significant impact on the
system reliability and the lower is the impact of resources
complementarity;
 multiple linear regression model is not able to grasp the de-
pendencies between complementarity, battery storage and
system reliability, but this task can be more precisely performed
by means of artificial neural networks.
The findings of this study can be used during the design stages
of an autonomous power supply station which will be utilizing
solar and wind sources. The inclusion of complementarity index in
the sizing process (apart from commonly used energy availability
per source in given time and energy demand) may increase the
system reliability and reduce renewable energy curtailment due to
the appropriate design of the hybrid energy source.
Considering the limitations of this study and formulated con-
clusions we highlight the possible future research directions as:
 investigating more complex hybrid sources (more than two
energy sources) and their resources complementarity on the
system reliability;
 performing test studies on large datasets which will include
various energy sources and be representative for different
climate zones, in order to prove the validity of results based on
synthetic time series;
 designing a system sizing method/rules based on known re-
sources complementarity values and local capacity factors.
References
[1] Bajpai P, Dash V. Hybrid renewable energy systems for power generation in
stand-alone applications: a review. Renew Sustain Energy Rev 2012;16(5):
2926e39.
[2] Nehrir MH, Wang C, Strunz K, Aki H, Ramakumar R, Bing J, Salameh Z.
A review of hybrid renewable/alternative energy systems for electric power
generation: configurations, control, and applications. IEEE Trans Sustain En-
ergy 2011;2(4):392e403.
[3] Mirez J, Hernandez-Callejo L, Horn M, Bonilla LM. Simulation of direct current
microgrid and study of power and battery charge/discharge management.
DYNA 2017;92(6):673e9.
€
[4] Yoldaş Y, Onen A, Muyeen SM, Vasilakos AV, Alan I. _ Enhancing smart grid with
microgrids: challenges and opportunities. Renew Sustain Energy Rev 2017;72:
205e14.
[5] Li Y, Agelidis VG, Shrivastava Y. Wind-solar resource complementarity and its
combined correlation with electricity load demand. In: Industrial electronics
and applications, 2009. ICIEA 2009. 4th IEEE conference on. IEEE; 2009.
p. 3623e8.
[6] Hoicka CE, Rowlands IH. Solar and wind resource complementarity:
advancing options for renewable electricity integration in Ontario, Canada.
Renew Energy 2011;36(1):97e107.
[7] Monforti F, Huld T, Bo
dis K, Vitali L, D’isidoro M, Lacal-Ara

ntegui R. Assessing
complementarity of wind and solar resources for energy production in Italy. A
Monte Carlo approach. Renew Energy 2014;63:576e86.
[8] De Jong P, Sa
nchez AS, Esquerre K, Kalid RDA, Torres EA. Solar and wind en-
ergy production in relation to the electricity load curve and hydroelectricity in
the northeast region of Brazil. Renew Sustain Energy Rev 2013;23:526e35.
[9] Solomon AA, Kammen DM, Callaway D. Investigating the impact of
windesolar complementarities on energy storage requirement and the cor-
responding supply reliability criteria. Appl Energy 2016;168:130e45.
[10] Liu Y, Xiao L, Wang H, Dai S, Qi Z. Analysis on the hourly spatiotemporal
complementarities between China’s solar and wind energy resources
spreading in a wide area. Sci China Technol Sci 2013;56(3):683e92.
[11] Xu L, Wang Z, Liu Y. The spatial and temporal variation features of wind-sun
complementarity in China. Energy Convers Manag 2017;154:138e48.
[12] Zhang H, Cao Y, Zhang Y, Terzija V. Quantitative synergy assessment of
regional wind-solar energy resources based on MERRA reanalysis data. Appl
Energy 2018;216:172e82.
[13] Thomaidis NS, Santos-Alamillos FJ, Pozo-V
azquez D, Usaola-García J. Optimal
743
management of wind and solar energy resources. Comput Oper Res 2016;66:
284e91.
[14] dos Anjos PS, da Silva ASA, Stosi

c B, Stosi

c T. Long-term correlations and cross-
correlations in wind speed and solar radiation temporal series from Fernando
de Noronha Island, Brazil. Phys Stat Mech Appl 2015;424:90e6.
[15] Santos-Alamillos FJ, Pozo-V
azquez D, Ruiz-Arias JA, Lara-Fanego V, Tovar-
Pescador J. Analysis of spatiotemporal balancing between wind and solar
energy resources in the southern Iberian Peninsula. J Appl Meteorol Climatol
2012;51(11):2005e24.
[16] Li W, Stadler S, Ramakumar R. Modeling and assessment of wind and inso-
lation resources with a focus on their complementary nature: a case study of
Oklahoma. Ann Assoc Am Geogr 2011;101(4):717e29.
[17] Prasad AA, Taylor RA, Kay M. Assessment of solar and wind resource synergy
in Australia. Appl Energy 2017;190:354e67.
[18] Miglietta MM, Huld T, Monforti-Ferrario F. Local complementarity of wind
and solar energy resources over Europe: an assessment study from a mete-
orological perspective. J Appl Meteorol Climatol 2017;56(1):217e34.
[19] Bett PE, Thornton HE. The climatological relationships between wind and
solar energy supply in Britain. Renew Energy 2016;87:96e110.
[20] Castro R, Crispim J. Variability and correlation of renewable energy sources in
the Portuguese electrical system. Energy Sustain Develop 2018;42:64e76.
[21] Gutierrez C, Gaertner MA,
Perpin ~a

n O, Gallardo C, Sa
nchez E. A multi-step
scheme for spatial analysis of solar and photovoltaic production variability
and complementarity. Sol Energy 2017;158:100e16.
[22] Beluco A, de Souza PK, Krenzinger A. A method to evaluate the effect of
complementarity in time between hydro and solar energy on the performance
of hybrid hydro PV generating plants. Renew Energy 2012;45:24e30.
[23] Francois B, Borga M, Creutin JD, Hingray B, Raynaud D, Sauterleute JF.
Complementarity between solar and hydro power: sensitivity study to
climate characteristics in Northern-Italy. Renew Energy 2016;86:543e53.
[24] Kougias I, Szabo
S, Monforti-Ferrario F, Huld T, Bo
dis K. A methodology for
optimization of the complementarity between small-hydropower plants and
solar PV systems. Renew Energy 2016;87:1023e30.
[25] Beluco A, de Souza PK, Krenzinger A. Influence of different degrees of
complementarity of solar and hydro energy availability on the performance of
hybrid hydro PV generating plants. Energy Power Eng 2013;5(04):332.
[26] François B, Zoccatelli D, Borga M. Assessing small hydro/solar power
complementarity in ungauged mountainous areas: a crash test study for hy-
drological prediction methods. Energy 2017;127:716e29.
[27] Silva AR, Pimenta FM, Assireu AT, Spyrides MHC. Complementarity of Brazil‫ ׳‬s
hydro and offshore wind power. Renew Sustain Energy Rev 2016;56:413e27.
[28] Denault M, Dupuis D, Couture-Cardinal S. Complementarity of hydro and
wind power: improving the risk profile of energy inflows. Energy Pol
2009;37(12):5376e84.
[29] Palfi GC, Zambon RC. Hydro and wind power complementarity and scenari-
zation in Brazil. In: World environmental and water resources congress 2013:
showcasing the future; 2013. p. 2414e24.
[30] Jurasz J, Piasecki A, Wdowikowski M. Assessing temporal complementarity of
solar, wind and hydrokinetic energy. In: E3S web of conferences, vol. 10. EDP
Sciences; 2016. p. 00032.
[31] Jurasz J, Mikulik J, Krzywda M, Ciapała B, Janowski M. Integrating a wind-and
solar-powered hybrid to the power system by coupling it with a hydroelectric
power station with pumping installation. Energy 2018;144:549e63.
[32] Kies A, Schyska BU, von Bremen L. The effect of hydro power on the optimal
distribution of wind and solar generation facilities in a simplified highly
renewable european power system. Energy Procedia 2016;97:149e55.
[33] Risso A, Beluco A, de Ca
ssia Marques Alves R. Complementarity roses evalu-
ating spatial complementarity in time between energy resources. 2018.
Accepted. To be published in Energies.
[34] Duffie JA, Beckman WA. Solar engineering of thermal processes. John Wiley &
Sons; 2013.
[35] Jurasz J, Piasecki A. A simulation and simple optimization of a wind-solar-
hydro micro power source with a battery bank as an energy storage device.
In: E3S web of conferences, vol. 14. EDP Sciences; 2017. p. 01017.
[36] Elia CP, Yang Z, Anders L, Hailong L, Jinyue Y. An open-source platform for
simulation and optimization of clean energy technologies. Energy Procedia
2017;105:946e52.
[37] http://www.soda-pro.com/web-services/radiation/cams-radiation-service
Accessed 02 02 2018.
[38] Global Modeling and Assimilation Office (GMAO). MERRA-2 tavg1_2d_slv_Nx:
2d,1-hourly, time-averaged, single-level, assimilation, single-level diagnostics
V5.12.4, Greenbelt, MD, USA, Goddard earth sciences data and information
services center (GES DISC). 2015. https://doi.org/10.5067/VJAFPLI1CSIV.
Accessed [02.02.2018].
[39] Jurasz J, Wdowikowski M, Ka
zmierczak B, Da˛ bek P. Temporal and spatial
complementarity of wind and solar resources in Lower Silesia (Poland). In:
E3S web of conferences, vol. 22. EDP Sciences; 2017. p. 00074.
[40] May R, Dandy G, Maier H. Review of input variable selection methods for
artificial neural networks. In Artificial neural networks-methodological ad-
vances and biomedical applications. InTech; 2011.
[41] Shibata K, Ikeda Y. Effect of number of hidden neurons on learning in large-
scale layered neural networks. In: ICCAS-SICE, 2009. IEEE; 2009, August.
p. 5008e13.