Professional Documents
Culture Documents
net/publication/305321699
CITATIONS READS
51 779
3 authors:
C. Ekanayake
The University of Queensland
91 PUBLICATIONS 787 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Muhammad Qamar Raza on 28 December 2017.
Solar Energy
journal homepage: www.elsevier.com/locate/solener
Review
a r t i c l e i n f o a b s t r a c t
Article history: In last decade, the higher penetration of renewable energy resources (RES) in energy market was encour-
Received 25 December 2015 aged by implementing the energy polices in several developed and developing countries due to increas-
Received in revised form 28 June 2016 ing environmental concerns. Among wide range of RES, Photovoltaic (PV) electricity generation get higher
Accepted 29 June 2016
attention by researcher, energy policy makers and power production companies due to its economic and
environmental benefits. Therefore, a large PV penetration was observed in energy market with rapid
growth in the last decade. The PV output power is highly uncertain due to several meteorological factors
Keywords:
such as temperature, wind speed, cloud cover, atmospheric aerosol levels and humidity level. The inher-
Photovoltaic
Artificial intelligence (AI)
ent variability of PV output power creates different issues directly or indirectly for power grid such as
Regressive techniques power system control and reliability, reserves cost, dispatchable and ancillary generation, grid integra-
PV output forecasting tion and power planning. Therefore, there is need to accurately forecast the PV output over the spectrum
Hybrid forecast models of forecast horizon at different chronological scales. In this paper, a comprehensive and systematic review
Artificial neural network (ANN) of PV output power forecast models were provided. This review covers the different factors affecting PV
Auto-regressive moving average forecast, PV output power profile and performance matrices to evaluate the forecast model. The critical
Fuzzy Logic analysis regressive and artificial intelligence based forecast models are also presented. In addition, the
Persistence method
potential benefits of hybrid techniques for PV forecast models are also thoroughly discussed.
Statistical methods
Ó 2016 Elsevier Ltd. All rights reserved.
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
2. Factors affecting PV output power forecast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
2.1. Forecast horizon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
2.2. Forecast model performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
2.3. Preprocessing of input data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
2.4. Forecast model inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3. Solar output power profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.1. UQ center array PV output profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4. Classification of forecast techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.1. Persistence forecast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.2. Physical models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.3. Statistical techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.4. Linear models or time series models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.4.1. ARMA model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.4.2. ARIMA techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.4.3. CARDS model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Abbreviations: AI, artificial intelligence; AM, air mass; ANN, artificial neural network; AR, auto-regressive; ARIMA, Auto-Regressive Integrated Moving Average; ARIMAX,
ARIMA exogenous; ARMA, Auto-Regressive Moving Average; ARMAX, ARMA-exogenous; DR, Demand response; BP, back propagation; CPV, concentrated PV; DHI, diffuse
horizontal irradiance; DNI, direct normal irradiance; GA, genetic algorithm; GHI, global horizontal irradiance; HS, hybrid system; IEA, International Energy Agency; ISO,
independent system operator; LMS, least means square; MA, moving average; MAE, Mean Absolute Error; MAPE, mean absolute percent error; MBE, Mean Bias Error; MLP,
multi-layer perceptron; NAR, non-linear AR; NARMAX, non-linear ARMA exogenous; NARX, non-linear AR exogenous; NWP, numerical weather prediction; PV, Photovoltaic.
⇑ Corresponding author.
E-mail addresses: Qamar.raza@uq.edu.au (M.Q. Raza), mithulan@itee.uq.edu.au (M. Nadarajah), chandima@itee.uq.edu.au (C. Ekanayake).
http://dx.doi.org/10.1016/j.solener.2016.06.073
0038-092X/Ó 2016 Elsevier Ltd. All rights reserved.
126 M.Q. Raza et al. / Solar Energy 136 (2016) 125–144
1. Introduction Atacama Desert, the Sahara and Kalahari Deserts (Africa) and
North-western Australia are potential locations for large power
In last decades, the global energy demand increased steadily generation from solar PV technology.
with rapid growth in world population. The energy demand is at Another research study (R. Shah et al., 2015) highlights that, the
the higher level ever before and most of the fossil resources are electricity demand of Mediterranean, North African region and
the edge of depletion due to excess usage. Therefore, ‘‘how to entire Europe can be fulfilled by developing solar plants in Sahara
meet the 21 century energy demand” is the hot topic of discussion Desert. The red sea including different areas of Saudi Arabia and
among the governments, researcher, scientists and energy policy Egypt are also among the highest potential sources for solar
makers in developed and developing countries. In addition, other energy. In addition, United States and Australia also have greater
concern is rapid changes in environmental and climatic conditions potential to get benefit form solar energy than the world average.
(i.e. global warming, depletion of ozone layer, etc.). Keep in view Due to potential of solar energy, large penetration of solar PV is
the energy issue, International Energy Agency (IEA) raised the expected in Australian energy sector in terms of rooftop PV, large
concerned namely energy security, economic efficiency, and envi- and small scale solar PV units. In last decade, the higher penetra-
ronmental protection, which are known as ‘‘3Es” ((IEA), 2007). tion of PV technology in energy market of different countries is
Therefore, concentrated efforts have made to reduce the emission observed due its environmental and economic benefits
of CO2 and minimum reliance on fossil fuels in order to achieve (Photovoltaics, 2012). These benefits are reduction in CO2 emis-
the 3Es objectives. Serval countries have been made efforts to meet sion, minimum refinance of fossil fuel resources and Solar Photo-
the 3E objectives which are align with IEA guidelines and their voltaic (PV) plants consist of solar panels, which directly convert
national energy targets. the sunlight into electricity unlike power generation using rotating
European union (EU) was decided to meet the energy targets by generators. PV becomes more popular due to different promising
2020 (Council, 2010). First target is to reduce the EU Green House features such as modularity, low maintenance and operational
Gas (GHG) emission by 20% below the level of 1990. Secondly, the cost, longer lifetime, CO2 reduction and environmental cleanliness.
contribution of renewable energy resources (RES) raise up to 20%. The energy generation capacity of solar plant varies due different
Thirdly, reduction of energy usage by 20% in contrast with pro- factors such as PV plant site, meteorological variables, solar tech-
jected levels through energy efficiency measures. European Union nology and installation capacity.
renewable energy directive set the RES production targets. It states Fig. 2 highlights the growth of world solar PV capacity in the
that, 30% total energy will be produced from RES generation by last decade. An exponential growth can be observed in global solar
2030. The target of RES generation contribution will be climbed PV capacity from 2004 to 2014. Global capacity was increased from
up to 100% by 2050 (Zervos et al., 2010). In addition, a high RES 3.7 GW to 7 GW in three years (2004–2007). In contrast, it was
contribution in existing power grid network is also expected by increased from 7 to 40 GW in next three years. In addition, a huge
the energy regulators of USA, Canada, Australia, China and India. growth in global solar PV capacity was observed in next couple of
Among number of RES resources, solar and wind power generation years. For example, global was more than double in 2010 as com-
are more promising sources. They have higher potential for pene- pared to 2008. Overall, global PV energy capacity was increased
tration in energy market with greater degree of success. However, from 3.7 to 177 GW in the last decade. A research study reports
solar generation get much more attention by the energy player, that, the PV module have individual capacity from 100 W to
investors and Government funding agencies in the last decade 320 W (Omran). The PV technology still facing lot of challenges
because of its economic and environmental benefits. for large penetration, in which intermittent and uncertain nature
Solar energy is feasible solution in order to meet the world of solar PV is more prominent. The PV output power is variable
energy demand. A research study highlights that, earth received mainly due to variations in solar radiations and amount received
approximately 1.8 ⁄ 1011 MW power from solar radiation at instant solar by the solar panels.
(B.M. Shah et al., 2015). However, the present world energy con- With a remarkable growth of PV in last decade, the integration
sumption requirement is less than the amount of energy received of photovoltaic plants in current power network raise the different
from solar (Wengenmayr and Bührke, 2011). Fig. 1 depicts the technical and stability issues for the power system directly or indi-
world solar energy map with solar hotspots. A huge potential for rectly. These distress arises due to continuous change in solar
solar power generation in different countries can be observed resource, temperature, PV output power, high energy storage cost,
those are above 45°N or below latitude 45°S. It can also grid reliability, seasonal and environmental changes (Denholm and
be observed, the Middle East, Mojave Desert (USA), the Chilean Margolis, 2007; Dixon et al., 2010). The implementation of PV
M.Q. Raza et al. / Solar Energy 136 (2016) 125–144 127
120
power plant operational management. ISO’s and utilities are more
100 interested in relatively longer forecast horizons for unit commit-
80
ment, load balancing and scheduling. The distribution and trans-
mission grids operational planning and balancing require the
60
spectrum of solar forecast for efficient management. It helps to grid
40 to reduce the ancillary costs associated with weather dependency
and deliver quality of energy. In addition, power gird stability of PV
20
integrated can be ensured with accurate PV forecast. It will
0
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
increase PV penetration in existing power grid network and help
Year to reduce of CO2 emission. From the grid prospective, reduction
of the power system operational costs is a main factor to design
Fig. 2. Global solar PV capacity from 2004 to 2014 (Zervos, 2015). accurate PV forecast models. PV output power forecast may also
part of smart grid (future generation power grids) energy manage-
ment system along with wind and load forecast.
power network as back up supply without storage devices is not a A number of factors affecting on PV output power, which leads
technical viable solution as it affects the grid stability due variable to uncertain and unpredictable PV output pattern. Therefore, this is
output power. It is due to large variations in meteorological condi- essential to design a robust, intelligent and adaptive forecast
tions, which increase the uncertainty of PV output power. model which can accommodate the factors affecting on PV output
Therefore, an accurate PV output forecast over the spectrum of for higher forecast accuracy. The complexity of forecast model will
forecast horizon is required for independent power producing (IPP) be increased by accommodating the different factors affecting on
and managing companies or equivalent grid balancing authorities. the output such wind speed, irradiance, temperature, cloud cover,
The accurate PV forecast will help IPP or power authorities for bet- and seasonal variations. Therefore, there is always need to trade off
ter energy planning and management. In addition, accurate fore- the number of forecast model inputs and complexity by keeping in
cast will be beneficial in terms of smart integration the PV view of forecast accuracy.
generation with current grid with higher system reliability As a reminder the organization of the paper as fellows: Section 2
(Dixon et al., 2010; Rodriguez, 2010; Helman et al., 2010). There- describes fundamental considerations for solar power generation.
fore, the importance of PV output forecast is vital in order to Section 3 briefly illustrate the, solar output power profile of The
achieve higher penetration of solar power technology. It will also University of Queensland (UQ), Australia St. Lucia campus center
contribute to minimize the reliance on fossil fuels resources. In array. The preprocessing of input data and forecast model
addition, grid regulation, power scheduling, unit commitment performance evaluation matrices are elaborated in Sections 4 and
and energy management system can be designed in effective 5. Section 6 reviews the regressive and artificial intelligence based
128 M.Q. Raza et al. / Solar Energy 136 (2016) 125–144
forecasting techniques with help of comprehensive tables. The logical conditions. A research study (Oudjana et al., 2013), purpose
potential benefits of hybrid system for PV forecast application a methodology to compare the ANN and regression model perfor-
are discussed in Section 7. mance for PV plant in the Ghardaia province of Algeria. ANN model
demonstrate higher forecast accuracy than the regression model
2. Factors affecting PV output power forecast with solar radiation and temperature as independent variables.
In Almeida et al. (2015), authors analyze the PV output
A number of variable are affecting on the output of PV forecast forecast performance of five different models named as the
model such as forecast horizon, forecast model input, performance k-Nearest-Neighbour, Persistent model, the Autoregressive
of prediction and data preprocessing applied as forecast model Integrated Moving Average (ARIMA), ANNs and the hybrid Genetic
input data. The forecast accuracy can be enhanced by accommo- algorithm based ANN model for 1 MW PV plant in California. ANN
dating the factors affecting on the PV output as forecast model model produces less forecast error up to RMSE of 11.42 for 1-h ahead
inputs. These factors are described below. forecast. However, the forecast accuracy further can be enhanced by
optimizing the ANN parameters using population based optimiza-
2.1. Forecast horizon tion techniques. It can be concluded that, the forecast accuracy of
PV output forecast model varies by changing the forecast model. It
The span of time into the future for which forecasts are to be is due to performance of individual forecast model.
prepared called forecast horizon. There are no well-defined criteria
to classify the forecasting in different categories based on forecast 2.3. Preprocessing of input data
horizon. However, some researcher reports that, forecast can be
divided in three based on time horizon. According to Raza and A number of PV output forecast model utilized the historical PV
Khosravi (2015), electrical load forecasting can be divided into output data as forecast model input. The historical PV output data
three categories by most of researchers but some of them divided may contain different spikes and non-stationary components due
it into four categories (Amral et al., 2007). Broadly, forecasting can to uncertain and variable meteorological conditions. As a result,
be divided in four major’s categories/types based on forecast hori- these non-stationary and spikes in data will leads to higher fore-
zon as given below: cast error due to improper training. The historical PV output data
can be processed for smoothing. In addition, missing input data
I. Long term forecast (1–10 year ahead). points in historical data will also play a role to increase the forecast
II. Medium term forecast (1 month to 1 year ahead). error. Therefore, the forecast accuracy of model can be consider-
III. Short term forecast (1 h or several hours ahead to 1 day or ably improved by input preprocessing. A number of techniques
1 week ahead). were applied to preprocess the inputs of forecast model. The pre-
IV. Very short term forecast (1 min to several min ahead). processed input data will significantly reduce computational cost
of forecast model by learning the historical pattern in better way.
Some of the researcher consider the few seconds or 1 min to 1 h Stationary, trend free time series, historical lag identification and
or several hours (maximum 6 h) ahead forecast as very short term normalization are useful techniques to preprocess the input data
forecast. However, some researcher considers as subclass of short for accurate PV output forecast. Time series of the clearness or
term forecast. However, majority of them enlist the very short clear-sky index are used in different research studies for prepro-
forecast as a separate class. In case of PV output forecast, short cessing as reported in Bacher et al. (2009) and Kemmoku et al.
term or very short term forecast is useful in order to design PV (1999).
integrated better energy management system, unit commitment, However, some of research doesn’t agree with huge impact of
power scheduling and dispatching. It is reported that for PV output clearness or clear-sky index in enhancement of forecast accuracy
forecast, the prediction accuracy of the model varies by changing (Sfetsos and Coonick, 2000). This study reports that, time series
the forecast horizon even with identical forecast model parame- of the clearness or clear-sky index is random in nature and varies
ters. In Lipperheide et al. (2015), research study analyze the perfor- due to different meteorological factors. Therefore, it does not pro-
mance of PV output forecast model over the different forecast vide strong learning basis to prediction model. It may lead to poor
horizon time such as 20, 40, and 60 . . . up to 180 s. The proposed forecast accuracy along with increased computational cost. The
forecast model produce prediction error (rRMSE) in the range of forecast results of this study demonstrates that, the preprocessed
3.2–15.5% for forecast horizon from 20 to 180 s. In Lonij et al. solar irradiance data is more effective to use as forecast model.
(2013) authors design a forecast model which produces the error In Cao and Cao (2006), Wavelets transform (WT) technique was
4.6% and 2.4% for 15 and 30 min forecast horizons, respectively. used for preprocessing of input data. The inputs of forecast model
It can be observed from above reported that the forecast error var- are split in different frequency components. These components
ies with change in forecast horizon. were used forecast model inputs. In Reikard (2009), solar irradi-
ance was used as input of model due to its impact on the PV out-
2.2. Forecast model performance put. In addition, statistical tool was used to remove the seasonality
trend in input data and it helps to increase the learning perfor-
The PV output forecast accuracy is dependent on prediction mance of the model. However, it is difficult to find the accurate
model performance. It is due to capability of individual forecast trend of daily solar irradiance data series due unpredictable noise.
model to handle the meteorological uncertainties. A number of sta- Some other research studies, analyze the performance of forecast
tistical and artificial intelligence based forecast models are model by applying a trend and de trend techniques for solar irradi-
designed to accurately forecast the PV output. Several research ance data (Baig et al., 1991; Kaplanis, 2006). Generally, the PV out-
studies have been reported to compare the performance of conven- put power follow the solar irradiance patter up to certain extent.
tional, statistical and ANN based forecast models. In Almonacid Cyclic behavior of solar radiation can be predicted with high
et al. (2010) authors designed three different conventional mathe- degree of accuracy by using Fourier series as predictor. This tech-
matical models for PV output forecast and compare with artificial nique is able predict the solar irradiance by combing the different
neural network (ANN). According to their findings, ANN based significant frequencies. According to Boland (1995, 2008), the daily
model outperform than the conventional mathematical models in time series profile of solar irradiance can be constructed effectively
terms of forecast accuracy and adaptability in uncertain meteoro- constructed by capturing the yearly and intraday cycles. In order to
M.Q. Raza et al. / Solar Energy 136 (2016) 125–144 129
X t ¼ aX t þ Z t ð1Þ
where a = 1 and Z t WN 0; r 2Z , WN represents the time series
white noise. Preprocessing techniques were also applied to different
other forecast application such as load, electricity price and wind
forecast. These results demonstrate the potential benefits of prepro-
cessing in terms of forecast accuracy.
Table 2
Characteristics of Selected UQ PV sites.
Sites Longitude Latitude Height above sea level (M) Module area (M2) Modules Inverters Nominal output (kWp)
0 00 0 00
GCI 153°00 52 E 27°29 51 S 27+ BH 941.2 575 9 138
LEB 153°290 4400 E 27°000 4900 S 43 612 374 7 89.76
CP1 153°000 3700 E 27°290 4200 S 23 2305 1412 26 338.9
CP2 153°000 3500 E 27°290 4200 S 23 2305 1412 26 338.9
UQC 153°000 54.800 E 27°290 4500 S 28 2956 1806 32 433.44
AEB 153°000 5300 E 27°290 5800 S 18 + BH 640 383 10 95.75
GSRFDA 152°200 14.100 E 27°330 41.500 S 88 5784 7200 – 684
UQ Centre (UQC), Sir Llew Edwards Building (LEB), Car Park 1 (CP1), Car Park 2 (CP2), Global Change Institute (GCI), Building, Advanced Engineering Building (AEB) and Gatton
Solar Research Facility Dual Axis (GSRFDA), BH = Building Height.
5
10
3
Power (W)
2.5
1.5
0.5
0
Jan. Feb. March April May June July Aug. Sep. Oct. Nov. Dec.
Fig. 7. The per unit curves of PV output power and air temperature.
The amount of generated power is increased with raise in air during the partially cloudy day as shown in Fig. 9. The solar irradi-
temperature and vice versa. This indicates the correlation between ance is less than the 400 W/m2 during the major portion of day
PV output power and air temperature. Some other research studies time except to hour 10–13. The solar irradiance goes up to
also reported that, there is strong correlation between temperature 1250 W/m2 during the hour 10–13 and PV output power also fol-
and PV output power (Chen et al., 2011). Therefore, Air tempera- low the same pattern. The sharp changes in PV output power is
ture can be applied as forecast model input in order to predict observed during the hour 10–13 and solar irradiance pattern is fol-
the PV output correctly. lowed by the PV output power, which indicates the strong positive
Fig. 8 depicts the variations in PV output power with change in correlation among them. It can be observed from Fig. 11 that, the
wind speed, which is recorded data in solar data management sys- solar irradiance peak goes up to maximum 400 W/m2 in cloudy
tem. The wind speed plays a role in heat dissipation and as result it day, which is less half of clear day. Therefore, average solar PV out-
leads to reduction in PV cell temperature. Therefore, PV output put power is quite lower than the clear and cloudy day. In addition,
power will be reduced with lower the cell temperature. Fig. 8 indi- higher level fluctuations in solar irradiance are observed during the
cates, the PV output power pattern does not follow the exactly the cloudy day than the clear and partially cloudy day. As a result, solar
wind speed pattern. In day, the PV output power is at higher level PV output power is also variable throughout the day.
as compare the wind speed. After that, the per unit PV output It can be concluded that from the above figures, PV output
power was reduced with increase in wind speed. However, similar power have very strong correlation with solar irradiation as com-
PV output power pattern with wind speed was not observed for pared to air temperature and wind speed. Therefore, these influen-
reset of year. Therefore, relatively a weak correlation is observed tial variables are recommended as forecast model to accurately
between PV output power and wind speed in comparison of forecast the PV output power in variable meteorological condition.
temperature. In addition, these variables add more uncertainty in PV output
Figs. 9–11 show the relationship between PV output power and power, which may create some critical issue directly or indirectly
solar irradiance during clear day (CD), partially cloudy (PCLD) and in PV integrated power grid.
cloudy day (CLD), which are selected from year 2014. It can be Some other meteorological related parameters are also affected
observed that in clear day graph, the PV output power fairly follow the PV output power such as air sole index, cloud cover, shading,
the solar irradiance curve. It can be observed that, solar irradiance humidity and wind speed. The forecast accuracy PV output forecast
fluctuates with high ramp rate. As a results, similar pattern is model may be enhanced using large number of inputs. However,
observed for PV output power profile. A similar trend is observed forecast model computational cost and complexity will be also
Fig. 10. Relationship between PV output and solar radiation for partially cloudy
Fig. 8. The per unit curves of PV output power and wind speed. day.
Fig. 9. Relationship between PV output and solar radiation for clear day. Fig. 11. Relationship between PV power and solar radiation for cloudy day.
132 M.Q. Raza et al. / Solar Energy 136 (2016) 125–144
increased due to aggregating the large number of input parame- persistence based forecast model shows higher forecast error and
ters. Therefore, it is utmost important to design a forecast model it is used to compare with proposed in order to analyze the perfor-
with optimal number of forecast model inputs to deal with com- mance. In Perez et al. (2010) research study, authors proposed
plexity issue. forecast model is evaluated with comparison of persistence model
results.
Forecast
Techniques
Numeric Arficial
Weather Neural Time Series
Predicon network
HIRM
MM5 GFS FNN BPNN RBF ARX ARIMA ARMA
LAM
NWP NN+
ANFIS
Ens. Fuzzy WT +NN T.S
non-stationary time-series and autoregressive (integrated) moving white noise. Some of the inherent features of time series can be
average with exogenous inputs (ARIMAX), etc. (Mohamed and extracted by using backshift operator and white noise process.
Bodger, 2004; Bacher et al., 2009; Alfares and Nazeeruddin, The backshift operator and the white noise process describe the
2002). The different tine series for forecast models are discussed intrinsic features of the time series. In addition, the adjacent obser-
below. vations are dependent to independent time step t of the white
noise process. In Hansen (1995), describes the working of ARIMA
4.4. Linear models or time series models (Auto-Regressive Integrated Moving Average) methods can used
as reference in prediction models domain. In research study
It is observed that the, time series forecasting has been applied (Reikard, 2009), ARIMA model is used to predict the solar irradi-
for different applications with higher degree of success since the ance and forecast results are compared ANN based forecast model.
last decade. In statistical techniques, variables are used as model The results highlights that, the ARIMA based forecast model trace
input having correlation with output and predictors. Several stud- the Sharpe changes of solar irradiance pattern and produces higher
ies has been published for time series modeling (Rodriguez, 2010). forecast accuracy than the benchmark techniques. There is poten-
In this research study, the comparison of time series techniques is tial to apply the ARIMA model stand along or hybridize with other
presented. In Kasten and Young (1989), AR model is used to inves- models for PV output power forecast.
tigate the forecasting performance of PV power with other predic-
tion techniques. 4.4.3. CARDS model
Autoregressive (AR) and dynamical system (DS) models are
4.4.1. ARMA model coupled together. It is used to predict to time series data called
ARMA model is combination of two basic models which are CARDS model. This model is also used to forecast the time series
autoregressive (AR) and moving average (MA) model as given in solar irradiance (Huang et al., 2013). The dynamical system equa-
Eq. (3). tion can be derived using Lucheroni model as given in Eqs. (7)
and (8).
X
p X
q
XðtÞ ¼ ai Xðt 1Þ þ bYðt jÞ ð3Þ
R_ ¼ x ð7Þ
i¼1 j¼1
Eq. (3) represents the forecasted PV output as it is represented ex_ ¼ kðx þ RÞ kð3R2 x þ R3 Þ ex cR b f ð8Þ
in form of function XðtÞ. P represent the number of AR process in
the model and ai is ith AR coefficient of the forecasting model. where f is noise term and k; k; c; e; b are tunable parameters. In
Where second part of the equation eðtÞ represents the jth coefficient above equation R_ represents the derivative of R with respect to time.
of MA model. Y(t) represents the white noise and it is not corre- Where x denotes the double derivative of R w.r.t. time as given in
lated forecast model output variable (Hamilton, 1994). Auto cor- Eqs. (9) and (10).
rected time series data can be treated by Autoregressive Moving
Average (ARMA) model. ARMA is considered as good prediction Rtþ1 ¼ Rt þ X t Dt þ xt ð9Þ
model to forecast the future values of provided time series with h i
stable input variables. ARMA model refer as (p, q), where p repre- X tþ1 ¼ X t þ kðX t þ Rt Þ k 3R2t X t þ R3t eX t cRt b
sent the order of AR model and q denotes the order of MA model.
Dt
The ability to extract the statistical properties and Box Jenkins þ at ð10Þ
adoption are the main reasons to the popularity of ARMA model e
(Boland, 2008). In addition, several types of time series can be where Dt represents the time step change and by using ordinary
characterized by using ARMA model with set of equations of differ- least square method other parameter can be determined such as
ent order. The ARMA model more suitable for stationary time ser- k; k; e; c; b. In Boland (1995), authors utilized the Fourier series tech-
ies data (Hansen, 1995). niques to de-seasoned the time series data. It is due to partial
inability of autoregressive process that it cannot model it alone.
4.4.2. ARIMA techniques Therefore, Fourier series is subtracted to from original series in
Auto-Regressive Integrated Moving Average (ARIMA) time ser- order to get the residual series. However, at the peak reversion,
ies model an extension to ARMA model. It is widely used for differ- the AR model is efficient enough to capture the series peaks. In
ent modeling and forecasting application with acceptable level of Lucheroni (2009), first who apply resonating model on power sector
forecast accuracy (Box et al., 2011). ARIMA for (p, q) for the time application. Therefore, a good level of fitting is achieved for residual
ðX 1 X 2 ; X 3 ; . . .Þ can be defined as given in Eqs. (4)–(6). series by using curvature’s proxy. The CARDS model shows good
results in comparison evaluated techniques in Kostylev and
Up ðZÞDd X t ¼ UðZÞat ð4Þ Pavlovski (2011). The results of this study highlights that, CARDS
model produce RMSE 16.5% as compared to other implemented
Up ðZÞ ¼ 1 /1 Z /2 Z 2 /p Z p ð5Þ model with RMSE 17% in clear day and 32% in cloudy at 1 h time
step.
Hq ðZÞ ¼ 1 h1 Z h2 Z 2 hp Z q ð6Þ Number of time series techniques are applied to different fore-
casting application along with different model inputs and forecast
where Z is backward shift operator. Therefore, back difference oper- horizons. There is potential to utilize the other time series tech-
ators are ZX y ¼ ZX y1 , D ¼ 1 Z. Up and Hq are the polynomials of niques for PV output power forecast such as stochastic time series
degree p and q. As mention earlier, ARIMA is product of AR model, (Chakhchoukh et al., 2011), autoregressive moving average (Chen
integrating part I and MA model. Therefore, ARIMA (p, q, q) model is et al., 1995), linear regression (Amral et al., 2007), general expo-
the combination of AR (p), integrating part IðdÞ ¼ Dd and moving nential technique (Christiaanse, 1971; Marín and Sandoval,
average MA (q). In order to avoid the unbounded process, the poly- 1997). Generally, statistical techniques provide higher forecast
nomials parameters such as U and H are chosen in such a way that, accuracy. Majority of time techniques provides the higher forecast
the both polynomials lies outside the unit circle. The variation from accuracy, if the forecast model input pattern is smooth. However,
the fixed distribution with zero mean and variance ra is called sharp or abrupt changes in meteorological variables such as
134 M.Q. Raza et al. / Solar Energy 136 (2016) 125–144
temperature, irradiance, wind speed and humidity leads to ferent layers based on the type and complexity of the problem. The
increase in forecast error. network is trained using training data, which is applied as forecast
model inputs. A post processing techniques can be used before cal-
culating and analyzing the performance of prediction model.
4.5. Artificial intelligence techniques
Artificial intelligence (AI) based techniques popular among the 4.6. Significance of ANN approach
researcher since three decades for forecast applications. Among
the AI techniques, neural network (NN) is more poplar and used Artificial Neural network based model has better capability to
as power computational tool for different predication applications map the input as model output without formulating the complex
with higher degree of success since 1980. Generally, NN models relationship between input and output (Fausett, 2006). McCulloh
produces higher forecast accuracy. It is due to its ability to capture and Pitts conduct a research to model the nets based bio-system
the sharp changes in the output with help of intelligent training for simple logical operation in 1942 (McCulloch and Pitts, 1943).
process of the network. An adaptive and robust NN training meth- They attempt to model the simplest form of nonlinear model of a
ods can further improve the capability of the network to learn the neurons. This research come up with new world of computational
complex relationship between input and output variable calculations. ANN models are robust in nature due to dynamically
(Patterson, 1998). respond to rapid changes with help of interconnected neurons and
A number of learning techniques are designed to train the NN learning methods. Several neural network learning techniques of
effectively such gradient and population based techniques. NN NN are available with different computational complexity and con-
model tries to predict the future output pattern using the different vergence rate for global optimum solution. The computational
set of input data. The forecast accuracy of the model is calculated complexity of the network, convergence rate and training time
with real time data and predicted values. The forecast accuracy can be reduced by optimizing the different parameters (Shekhar
of model can be further enhanced by carefully selecting the differ- and Amin, 1992). These parameters are correlated forecast model
ent influential parameters. These parameters are the normalized inputs, preprocessed training data of the network, optimal NN
and suitable number of forecast model inputs, appropriate training structure and robust NN training techniques, etc. ANN based mod-
algorithm, optimized network structure, appropriate learning algo- els have been implemented on several fields of life such as aero-
rithm, suitable training data and optimized network structure may space, Bio-medical, research and development of socio economic
increase the overall performance of network and reduce the com- applications, automotive industry, electronics industry, stock mar-
plexity (Ho et al., 1992). The primary steps ANN based load forecast ket and finance industry (Lai, 1998).
model are provided in Fig. 13.
Input selection of PV output forecast model is one of most crit- 4.6.1. Artificial neural network
ical part in the design process. The forecast accuracy varies with In the last decade, ANN models are utilized as power full tool for
change in type and number of model inputs. It is due to large different real life and computational applications. ANN have ability
dependence of PV output power on meteorological variables. Sec- to draw complex relationship between input and output using
ondly, the real time meteorological and PV output recorded data learning mechanism of the network. Through learning algorithm
is used to train the forecast model and compare the model perfor- of neural network, ANN model can map the complex input output
mance. This data may contains several missing data points, sharp relationship by using feedback error system. ANN based models
peaks and variations. Therefore, input preprocessing technique is are robust and adaptive in nature and performed well up to certain
need to be apply on data for smoothing before applying the data level even under the noisy environment. The summary of early
as forecast model input. After that, the historical PV output power stages developments of ANN model are given below in Table 3.
data is divided in two groups named as training and testing data.
The training data of forecast model is used for learning of the net- 4.6.2. Artificial neural network architecture
work to forecast the future values. Testing data is used to analyze The artificial neural network (ANN) is network of connected
the performance of forecast model by comparing the actual and artificial neurons in different layers such as input, hidden and out-
predicted values. After that, neural network is initialized with dif- put layers. The basic artificial neural network as shown in Fig. 14.
The neurons are connected in different layers with synaptic
weights values. The learning algorithm of neural network tries to
map the input and output relationship by updating the synaptic
Selection Of Forecast Model Inputs
weights values. The network generated output is compared with
desired output and then error is calculated. Therefore, the weights
Pre-processing of Input Data for and biases values of NN are updated based on the error. The acti-
Forecast Models vation function is applied to weighted input for output of the
network.
This cycle will continue until the desired output achieved.
Split Data Into Testing And Training
The weighted sum of inputs can express in the form mathemat-
Sets
ical relationship as given below:
!
X
n
Initialize and Train ANN Model Ai ¼ g W ij ai ð11Þ
i¼0
Table 3
Summary of major ANN developments.
of the network varies with change in activation function, architec- 1998). The first and second layer of the network can be differenti-
ture and inputs of neural network. Different types of activation ate on the basis of synaptic weights. The first layer synaptic
functions are discussed in next section. weights can be determined by using input data set. The weight val-
ues of input layer are fixed at first stage and the second layer
4.6.3. Activation function weights are determined. The RBFNN network is trained using
The output of the network is generated by using activation unsupervised learning method as only input data is provided to
function by summing of weighted inputs. Therefore, activation the network. The supervised learning method is utilized for second
function of network acts as squeezing function to transfer the input layer weights determination. Classic least mean squares is used for
in the form of output. Different activation functions are available optimization. The basic form RBFNN mapping is given in Eq. (11).
for neural network and output of the network also varies with The RBFNN architecture is shown in Fig. 16.
change of activation function (Zhang et al., 2007). There are num-
ber of transfer functions are available such as Gaussian radial basis, X
M
uni and bipolar polar step function, linear function and sigmoid Y k ðxÞ ¼ W kj /j ðxÞ þ W ko ð12Þ
j¼1
function respectively. Table 4 summaries the different types of
activation function along with derivative and diagrams. There are
hard and fast rules for transfer function selection. The activation 4.6.6. Recurrent neural network
function of neural network can be selected based on nature of Recurrent neural network (RNN) demonstrate the higher capa-
application (Mellit and Kalogirou, 2008). bility to learn different complex relationships and computational
structures. A real-time recurrent learning network (RTRL) contains
4.6.4. Multi layer perceptrons neural network input layers, processing layers (feed forward and connections lay-
A number of complex problems from different research ers) and some additional node elements. The successful application
domains, which cannot be solved by using single layer neural net- of RNN model is discussed comprehensively in Hertz et al. (1991).
work. It is due to complex input and output relationship between In Elman (1990), authors proposed RNN architecture with feedback
different variables. Multilayer perceptron’s neural network back loop. It is taken from hidden layer of the network to input
(MLPNN) have the ability to map the input output relationship layer. However, another research studies the proposed the feed-
using proper training of network. There are one or more than hid- back from output layer to input layer of the network (Jordan,
den layers between input and output layer of the network. These 1997). These feedback loops are used to minimize the network
layers are connected with each other. The multilayer perceptrons learning error using training process.
neural network (MLPNN) was for different forecast applications In fully connected recurrent neural network (RNN) every pro-
such as load and electricity price forecasting (STLF) (Tasre et al., cessing node is connected with other processing node and with
2011; Raza et al., 2014). A typical MLPNN architecture for PV fore- itself as well. Therefore, the output of RNN at any time is depends
cast is represent in Fig. 15. on the two parameters, which are feedback signal at previous time
step and input signals. In Williams and Zipser (1989), authors ana-
4.6.5. Radial basis function network (RBFNN) lyze and perform experiments on the learning of real time of RNN.
The radial basis function neural network (RBFNN) is considered The network was trained in every processing cycle. The activation
as two-layer neural network. The learning process of RBFNN can be function of the network is weighted sum of feedback and current
divided in two different stages (Madan et al., 2003; Jain and Martin, input signals. Therefore, the activation function is given in Eq. (13).
136 M.Q. Raza et al. / Solar Energy 136 (2016) 125–144
Table 4
Artificial neural network transfer functions.
Bipolar step function f ðxÞ ¼ sinðxÞ ¼ 2HðxÞ 1 1 if ; x – 0
dðxÞ ¼
1 if ; x ¼ 0
8
Unipolar linear function <0 if x < 1 dðxÞ ¼ 1=2½Hðx þ 1Þ Hðx 1Þ
f ðxÞ ¼ HðxÞ ¼ 1=2ðx þ 1Þ if jxj < 1
:
1 if x > 1
8
Bipolar linear function <1 if x < 1 dðxÞ ¼ ½Hðx þ 1Þ Hðx 1Þ
f ðxÞ ¼ HðxÞ ¼ y if jxj < 1
:
1 if x > 1
Gaussian radial basis f ðxÞ ¼ exp kx mk2 =r2 2ðx mÞf ðxÞ=r2
X
pþ1 X
q 4.6.7. Feed forward neural network
Sk ðtÞ ¼ ðW ik X p ðtÞÞ ðV kq Y q ðtÞÞ ð13Þ Feed forward neural network (FFNN) is relatively less complex
p¼1 q¼1
NN architecture. In FNN, the information moves from input to out-
put layer in forward direction. Network can be single layer or
where Sk ðtÞ is activation function at the time of processing node k
multi-layer but information moves only on one direction. There
and V kq is the connection weight of the node q which is connected
is no feedback loop or cycle for information to process. In feed for-
with node k. W pþ1 is the value of the bias. The output of node k is
ward neural network, the information reach at output layer
given below in Eq. (14).
through input and hidden layer of the network. Fig. 18 highlight
Y k ðt ¼ 1Þ ¼ f k ðSk ðtÞÞ ð14Þ the neural network having input, hidden and output layer with 3,
2, 3 neurons respectively. Feed forward neural network was also
where f is sigmoid activation function. The Recurrent neural net- applied for several forecasting and pattern recognition application
work (RNN) architecture is shown in Fig. 17. (Ahmad et al., 2009; Malki et al., 2004).
M.Q. Raza et al. / Solar Energy 136 (2016) 125–144 137
Neural Network
Historical PV 11 Architecture
Output Data
Temperature 2 1
PV
Forecast
Humidity 3 2 1
Wind Speed 4 20
Cloud Cover 38
8
Fig. 16. Radial basis function network (RBFNN) architecture (Rivas et al., 2004).
Ref. Year Country of Journal/conference Forecast horizon Forecast error Forecast model Comments
PV data set
Chen et al. (2011) 2011 China Solar Energy 24 h ahead MAPE < 9.28% Self-organizing map (SOM) The performance of different test case studies on sunny, cloudy day and
trained ANN model rainy day. However, forecast model performance varies with day type
Ding et al. (2011) 2011 Ashland, Procedia 24 Hours ahead MAPE 10.06% Artificial Neural Network (ANN) Backpropagation learning algorithm based ANN is used to forecast the 24 h
USA Environmental PV output
Sciences
Cococcioni et al. 2011 Italy IEEE conference 24 h ahead MAPE 5% Artificial neural networks (ANN) Time-series analysis model NARX with feedforward neural network is used
(2011) to predict 24 h PV output with varying number of hidden layer, number of
neurons and training window
Kang et al. (2011) 2011 Korea IEEE conference 24 h ahead Approximately k-means clustering method K means clustering technique is used at first stage and five years historical
MAPE 11% data was analyzed to classify them based on cloudiness
Mori and Takahashi 2012 Japan IEEE conference 24 h ahead Error 0.228 pu Generalized Radial Basis Generalized Radial Basis Function Network (GRBFN), Deterministic
(2012) [error is Function Network (GRBFN) Annealing (DA), and Evolutionary Particle Swarm Optimization (EPSO)
quantized in based Neural Netwrok
standard method]
Ref. Year Country of Journal/conference Forecast horizon Forecast error Forecast model Comments
PV data set
Chupong and Plangklang 2011 Thailand Energy Procedia 11 h MAPE 16.83% Elman neural network The proposed Elman neural network based forecast model
(2011) than Recurrent Neural Network based model
Pedro and Coimbra (2012) 2012 California, Solar Energy 1 and 2 h-ahead 35.1% Auto-Regressive Integrated Moving GA optimized Artificial Neural Networks (ANNs) outperform
USA Improvement in Average (ARIMA), and ANN Genetic than the other forecast models
forecast results Algorithm (GAs/ANN)
Mandal et al. (2012) 2012 USA Procedia 1 h ahead MAPE from Wavelet transform (WT) and artificial Radial basis function neural network (EBFNN) forecast model
Computer Science 2.38% to 4.08% intelligence (AI) is used with wavelet transform was tested for spring,
summer, winter and autumn’s sunny and cloudy day
Lonij et al. (2013) 2013 USA Solar Energy 30–90 min RMSE Persistence model Forecast results show that, the prediction performance
improvement increased by numerical weather Model
23%
Marquez and Coimbra 2013 USA Solar Energy 30, 60, 90, 120 min RMSE < 2.95% Artificial neural network (ANN) ANN model inputs are satellite image analysis including
(2013) velocimetry and cloud indexing to enhance the forecast
accuracy
139
140 M.Q. Raza et al. / Solar Energy 136 (2016) 125–144
Historical power and weather data is applied to support vector machines (SVM) model
forecast models. However, the forecast accuracy of both model are close to each other
such as UCMA model, ARIMA, UCM model, a transfer function
ANN based model provides lower forecast error by using pre-processing techniques
Solar irradiance, Air temperature, cloud, humidity and sun position is used for both
model, hybrid model and neural network based model. The perfor-
mance comparison analysis highlights that, ARIMA model gives
better results over the 60, 30 and 15 min window. However, other
to predict the PV output power for sunny, foggy, rainy and cloudy day
studies reports that neural network based model outperform than
other comparative models (Mueller et al., 2004). It is reported that,
there is potential to improve forecast accuracy using robust fore-
cast techniques as PV (Reikard, 2009). It can be conclude that from
time series forecast techniques, it provides higher forecast accu-
racy under relatively smooth meteorological conditions. However,
the forecast error is increased under uncertain and sharp meteoro-
logical changes. NN models also not fully generalize over input
data of model inputs. Therefore, there is potential to apply the
hybrid techniques for PV output power forecast for precise
forecast.
ing application. Therefore, forecast output can be used for real time
Forecast model
(SVM)
nRMSE 10%
MRE 8.64%
ahead forecast
120 and 160 h
China
2010
Year
Ref.
than the comparative forecast model. In Voyant et al. (2011), models tries to forecast PV output power accurately. Due to bad
hybrid forecast model was proposed by using optimized MLP net- performance single model in hybrid architecture may leads may
work architecture. The exogenous and meteorological variable are to higher forecast error. Therefore, there is a need to design model,
applied as model input. Hybrid forecast models were also utilized in which each predictor doesn’t affect the performance of each
for electrical load forecasting in order to enhance the prediction other. It is reported that, the multi predictor based forecast frame-
performance. In Raza and Khosravi (2015) and Raza and work organized in ensemble network and combined the output of
Baharudin (2012) provide the detailed analysis of hybrid models each predictor in intelligent way can enhance the overall forecast
for electrical load forecast. These hybrids forecast models have a output (Li et al., 2011). In conclusion, there is potential to apply
potential to apply on PV output forecasting. There are several the ensemble based forecast framework to accurately predict to
hybrid techniques reported for load forecasting as given below. PV output power.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN
output forecast for different forecast horizon is required for higher 2
t¼1 ðZ t Z t Þ
1
N
penetration of solar plants, which is able facilitate grid, ISOs and to RMSE=Capacity ¼ ð23Þ
achieve high grid stability. C
As discussed earlier, output of PV unit is largely dependent on
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the solar irradiance. In addition, a variation in solar irradiance 1 1 XN 2
can be observed as it is fundamentally dependent several factors RMSE=Capacity ¼ pffiffiffiffi ðZ t Z t Þ ð24Þ
C N N t¼1
such as time of the day and year, climatic conditions, geographic
location and elevation techniques. In order to measure the quality
of forecasted data, correlation coefficient is measured which mea-
sure the error variance to the variance of modeled data as given in 7. Conclusions and future work
Eq. (15).
In this paper global solar PV status and potential has been ana-
varðZ 1Þ
R2 ¼ 1 ð15Þ lyzed to meet the current global energy requirements. In the last
varðZÞ decade, large pentation of PV was observed due to tremendous
Different parameters are used to analyze and compare the fore- potential of solar energy in the different regions of the world in
cast model accuracy. Some of these parameters are discussed in terms of rooftop PV, large, medium and small scale solar plants.
this study. However, solar PV energy is can create different issues for modern
The absolute of value of the error is called Mean Absolute Error power systems directly or indirectly due to uncertain and intermit-
(MAE) which be measured by using given in Eq. (16). tent nature. Therefore, it is utmost important to accurately forecast
the PV output power.
1X N
This study provides systematic and comprehensive literature of
MAE ¼ jZ t Zj ð16Þ
N t¼1 different solar forecasting technique. In addition, the factors affect-
ing on solar output were identified, which could be applied as fore-
where Z t is forecasted and Z is actual PV output power. A similar cast model inputs for higher accuracy. Literature indicates that, a
notation is used later equations for forecasted and actual value. certain level of forecast accuracy can be achieved by applying dif-
The Mean Absolute Percentage Error (MAPE) can be calculated by ferent stand alone and hybrid models for PV output forecast appli-
using this formula in Eq. (17). cation. It is identified that, solar irradiance, temperature, wind
N speed and direction, humidity cloud cover and aerosol index are
100% X
Z t Z
MAPE ¼ ð17Þ major parameters to change of PV output power. It is also con-
N t¼1 Z
cluded that, solar irradiance is highly correlated with PV output
The Average spread and average bias of error can be measured power and follows the similar pattern. Therefore, the forecast accu-
by using Root mean square error (RMSE) and Mean Bias Error for- racy of prediction models can be enhanced by optimizing and bet-
mula as given in Eqs. (18) and (19) respectively. ter selection of these correlated variables.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Several attempts have been made to precisely forecast the out-
1 XN put power using different techniques in past few years. Among
RMSE ¼ ðZ t Z t Þ2 ð18Þ
N t¼1 these techniques, regressive methods take the advantage of corre-
lated nature of meteorological variables, which are used prediction
1X N model as inputs. It is concluded that, endogenous stochastic meth-
MBE ¼ ðZ t Z t Þ ð19Þ ods such as AR, MA, ARMA and/or ARIMA can be used where less
N t¼1
number of meteorological parameters are available as model input.
Correction coefficient can be calculated according to Eq. (20). In addition, different classification and clustering methods can be
applied for improved training the forecasting model to enhance
Cov ðx; yÞ
qx;y ¼ ð20Þ the forecast model performance. Intelligent Leaning techniques
rðxÞrðyÞ such as artificial neural network (ANN) and fuzzy logic can applied
where Cov is covariance and r indicates the standard deviation. in dynamic environment to forecast the PV output, if adequate his-
Statistical distributions observation ability of a model is known as torical patterns are available to train the network. ANN based
Kolmogorov Smirnov Integral (KSI) (Hoyer-Klick et al., 2009; model offer improved nonlinear approximation performance and
Espinar et al., 2009). KSI can calculated as given in Eq. (21). better capability to handle the uncertainty using better training
Z data and algorithm. In order to achieve the higher prediction accu-
Ymax
racy of PV output power forecast model, it is required to removed
KSI ¼ Dn dy ð21Þ
Ymin sharp changes and fluctuations in data of meteorological variables
and historical PV output power. The forecast accuracy of prediction
The cumulative distribution function difference can be repre-
models can be enhanced by pre-processing and post processing of
sented by Dn. In addition, Trapezoidal integration can used for dis-
historical and forecasted PV output power. There is also a need to
crete value of Dn. Frequency distribution reproduction can be
investigate the performance of forecast models during different
provided by using KSI (Hoff et al., 2013; Hansen, 1995). In Perez
days such as clear day, cloudy day and rainy day for further appli-
et al. (2011), also utilized KSI method to analyze the performance.
cation use. It is concluded that, the forecast performance of
A research study (Hoff et al., 2013) reports that, better relative dis-
stochastic models also affected due to individual poor perfor-
persion error can be measured by using normalized Mean Absolute
mance, model learning capability for certain PV system data, nat-
Error by using Eq. (22).
ure of meteorological events and their ramp rates. Therefore,
PN
t¼1 jZ t Zt j
1 there is a need to design multi predictor based forecast model to
MAE=Avg ¼ N 1
PN ð22Þ mitigate the above mentions factors up to certain level. Therefore,
N t¼1 Z t
it is recommended to explore more on ensemble forecast models
This above mentioned research study also suggests that the nor- with various combination of predictors to achieve the higher pre-
malization of forecast RMSE over the maximum nominal irradiance diction accuracy. In future, accurate PV output forecast has a
cane calculated as provided in Eqs. (23) and (24). potential to utilized for different power system applications such
M.Q. Raza et al. / Solar Energy 136 (2016) 125–144 143
smart grid, efficient energy management system, PV integrated Hamilton, J.D., 1994. Time Series Analysis. Princeton University Press Princeton.
Hansen, B.E., 1995. Time series analysis James D. Hamilton Princeton University
smart buildings and eclectic vehicle charging, etc.
Press, 1994. Economet. Theory 11, 625–630.
Haque, A.U., Nehrir, M.H., Mandal, P., 2013. Solar PV power generation forecast
References using a hybrid intelligent approach. In: 2013 IEEE Power and Energy Society
General Meeting (PES). IEEE, pp. 1–5.
Hecht-Nielsen, R., 1987. Counterpropagation networks. Appl. Opt. 26, 4979–4984.
Ahmad, I., Abdullah, A.B., Alghamdi, AS., 2009. Application of artificial neural
Helman, U., Loutan, C., Rosenblum, G., Rothleder, M., Xie, J., Zhou, H., 2010.
network in detection of probing attacks. IEEE Symposium on Industrial
Integration of Renewable Resources: Operational Requirements and Generation
Electronics & Applications, ISIEA. IEEE, pp. 557–562.
Fleet Capability at 20% rps. California Independent System Operator.
Alfares, H.K., Nazeeruddin, M., 2002. Electric load forecasting: literature survey and
Hernández, L., Baladrón, C., Aguiar, J.M., Calavia, L., Carro, B., Sánchez-Esguevillas, A.,
classification of methods. Int. J. Syst. Sci. 33, 23–34.
Cook, D.J., Chinarro, D., Gómez, J., 2012. A study of the relationship between
Almeida, M.P., Perpiñán, O., Narvarte, L., 2015. PV power forecast using a
weather variables and electric power demand inside a smart grid/smart world
nonparametric PV model. Sol. Energy 115, 354–368.
framework. Sensors 12, 11571–11591.
Almonacid, F., Rus, C., Hontoria, L., Muñoz, F.J., 2010. Characterisation of PV CIS
Hertz, J., Krogh, A., Palmer, R.G., 1991. Introduction to the Theory of Neural
module by artificial neural networks. A comparative study with other methods.
Computation. Basic Books.
Renew. Energy 35, 973–980.
Ho, K.-L., Hsu, Y.-Y., Yang, C.-C., 1992. Short term load forecasting using a multilayer
Amral, N., Ozveren, C.S., King, D., 2007. Short term load forecasting using multiple
neural network with an adaptive learning algorithm. IEEE Trans. Power Syst. 7,
linear regression. In: 42nd International Universities Power Engineering
141–149.
Conference, UPEC, 4–6 Sept. 2007, pp. 1192–1198.
Hoff, T.E., Perez, R., Kleissl, J., Renne, D., Stein, J., 2013. Reporting of irradiance
Bacher, P., Madsen, H., Nielsen, H.A., 2009. Online short-term solar power
modeling relative prediction errors. Prog. Photovoltaics Res. Appl. 21, 1514–
forecasting. Sol. Energy 83, 1772–1783.
1519.
Baharudin, Z., Nallagownden, P., Raza, MQ., 2014. A hybrid neuro-genetic approach
Hopfield, J.J., 1982. Neural networks and physical systems with emergent collective
for STLF: a comparative analysis of model parameter variations. In: Power
computational abilities. Proc. Natl. Acad. Sci. 79, 2554–2558.
Engineering and Optimization Conference (PEOCO), 2014 IEEE 8th
Hoyer-Klick, C., Beyer, H.-G., Dumortier, D., Schroedter-Homscheidt, M., Wald, L.,
International. IEEE, pp. 526–531.
Martinoli, M., Schilings, C., Gschwind, B., Ménard, L., Gaboardi, E., 2009.
Baig, A., Akhter, P., Mufti, A., 1991. A novel approach to estimate the clear day global
MESoR-.anagement and exploitation of solar resource knowledge. In: Solar
radiation. Renew. Energy 1, 119–123.
Paces.
Boland, J., 1995. Time-series analysis of climatic variables. Sol. Energy 55, 377–388.
Huang, J., Korolkiewicz, M., Agrawal, M., Boland, J., 2013. Forecasting solar radiation
Boland, J., 2008. Time series modelling of solar radiation. In: Badescu, V. (Ed.),
on an hourly time scale using a Coupled AutoRegressive and Dynamical System
Modeling Solar Radiation at the Earth’s Surface. Springer, Berlin, Heidelberg.
(CARDS) model. Sol. Energy 87, 136–149.
Box, G.E., Jenkins, G.M., Reinsel, G.C., 2011. Time Series Analysis: Forecasting and
Huang, Y., Lu, J., Liu, C., Xu, X., Wang, W., Zhou, X., 2010. Comparative study of
Control. John Wiley & Sons.
power forecasting methods for PV stations. In: 2010 International Conference
Cao, J.C., Cao, S.H., 2006. Study of forecasting solar irradiance using neural networks
on Power System Technology (POWERCON). IEEE, pp. 1–6.
with preprocessing sample data by wavelet analysis. Energy 31, 3435–3445.
(IEA), I.E.A., 2007. International Organization for Standardization (IEA–ISO).
Chaabene, M., Ben Ammar, M., 2008. Neuro-fuzzy dynamic model with Kalman
International standards to develop and promot energy efficiency and
filter to forecast irradiance and temperature for solar energy systems. Renew.
renewable energy sources. Special ISO focus—world energy congress, 5–10.
Energy 33, 1435–1443.
Jain, L.C., Martin, N., 1998. Fusion of Neural Networks, Fuzzy Systems and Genetic
Chakhchoukh, Y., Panciatici, P., Mili, L., 2011. Electric load forecasting based on
Algorithms: Industrial Applications. CRC Press.
statistical robust methods. IEEE Trans. Power Syst. 26, 982–991.
Ji, W., Chee, K.C., 2011. Prediction of hourly solar radiation using a novel hybrid
Chen, C., Duan, S., Cai, T., Liu, B., 2011. Online 24-h solar power forecasting based on
model of ARMA and TDNN. Sol. Energy 85, 808–817.
weather type classification using artificial neural network. Sol. Energy 85,
Jordan, M.I., 1997. Serial order: a parallel distributed processing approach. Adv.
2856–2870.
Psychol. 121, 471–495.
Chen, J.-F., Wang, W.-M., Huang, C.-M., 1995. Analysis of an adaptive time-series
Junior, J.G.D.S.F., Oozeki, T., Ohtake, H., Shimose, K.-I., Takashima, T., Ogimoto, K.,
autoregressive moving-average (ARMA) model for short-term load forecasting.
2014. Forecasting regional photovoltaic power generation – a comparison of
Electr. Power Syst. Res. 34, 187–196.
strategies to obtain one-day-ahead data. Energy Procedia 57, 1337–1345.
Christiaanse, W.R., 1971. Short-term load forecasting using general exponential
Kang, M.-C., Sohn, J.-M., Park, J.-Y., Lee, S.-K., Yoon, Y.-T., 2011. Development of
smoothing. IEEE Trans. Power App. Syst. PAS-90, 900–911.
algorithm for day ahead PV generation forecasting using data mining method.
Chu, Y., Urquhart, B., Gohari, S.M.I., Pedro, H.T.C., Kleissl, J., Coimbra, C.F.M., 2015.
In: 2011 IEEE 54th International Midwest Symposium on Circuits and Systems
Short-term reforecasting of power output from a 48 MWe solar PV plant. Sol.
(MWSCAS). IEEE, pp. 1–4.
Energy 112, 68–77.
Kaplanis, S.N., 2006. New methodologies to estimate the hourly global solar
Chua, L.O., Yang, L., 1988. Cellular neural networks: applications. IEEE Trans. Circ.
radiation; comparisons with existing models. Renew. Energy 31, 781–790.
Syst. 35, 1273–1290.
Kasten, F., Young, A.T., 1989. Revised optical air mass tables and approximation
Chupong, C., Plangklang, B., 2011. Forecasting power output of PV grid connected
formula. Appl. Opt. 28, 4735–4738.
system in Thailand without using solar radiation measurement. Energy Proc. 9,
Kaur, A., Pedro, H.T.C., Coimbra, C.F.M., 2013. Impact of onsite solar generation on
230–237.
system load demand forecast. Energy Convers. Manage. 75, 701–709.
Cococcioni, M., D’andrea, E., Lazzerini, B., 2011. 24-hour-ahead forecasting of energy
Kemmoku, Y., Orita, S., Nakagawa, S., Sakakibara, T., 1999. Daily insolation
production in solar PV systems. In: 2011 11th International Conference on
forecasting using a multi-stage neural network. Sol. Energy 66, 193–199.
Intelligent Systems Design and Applications (ISDA). IEEE, pp. 1276–1281.
Khotanzad, A., Enwang, Z., Elragal, H., 2002. A neuro-fuzzy approach to short-term
Council, T.E., 2010. Communication from the Commission to the European
load forecasting in a price-sensitive environment. IEEE Trans. Power Syst. 17,
Parliament. The European Economic and Social Committee and the
1273–1282.
Committee of the Regions.
Kostylev, V., Pavlovski, A., 2011. Solar power forecasting performance–towards
de Giorgi, M.G., Congedo, P.M., Malvoni, M., 2014. Photovoltaic power forecasting
industry standards. In: 1st International Workshop on the Integration of Solar
using statistical methods: impact of weather data. IET Sci. Meas. Technol. 8, 90–
Power into Power Systems Aarhus, Denmark.
97.
Lai, L.L., 1998. Intelligent System Applications in Power Engineering: Evolutionary
Denholm, P., Margolis, R.M., 2007. Evaluating the limits of solar photovoltaics (PV)
Programming and Neural Networks. John Wiley & Sons Inc.
in traditional electric power systems. Energy Policy 35, 2852–2861.
Leva, S., Dolara, A., Grimaccia, F., Mussetta, M., Ogliari, E., 2015. Analysis and
Ding, M., Wang, L., Bi, R., 2011. An ANN-based approach for forecasting the power
validation of 24 hours ahead neural network forecasting of photovoltaic output
output of photovoltaic system. Proc. Environ. Sci. 11 (Part C), 1308–1315.
power. Math. Comput. Simul., 0378-4754. http://dx.doi.org/10.1016/
Dixon, R.K., McGowan, E., Onysko, G., Scheer, R.M., 2010. US energy conservation
j.matcom.2015.05.010.
and efficiency policies: challenges and opportunities. Energy Policy 38, 6398–
Li, G., Shi, J., Zhou, J., 2011. Bayesian adaptive combination of short-term wind speed
6408.
forecasts from neural network models. Renew. Energy 36, 352–359.
Dolara, A., Leva, S., Manzolini, G., 2015. Comparison of different physical models for
Lipperheide, M., Bosch, J., Kleissl, J., 2015. Embedded nowcasting method using
PV power output prediction. Sol. Energy 119, 83–99.
cloud speed persistence for a photovoltaic power plant. Sol. Energy 112, 232–
Elman, J.L., 1990. Finding structure in time. Cogn. Sci. 14, 179–211.
238.
Energy, G., 2010. Western Wind and Solar Integration Study. Citeseer.
Liu, J., Fang, W., Zhang, X., Yang, C., 2015. An Improved photovoltaic power
Espinar, B., Ramírez, L., Drews, A., Beyer, H.G., Zarzalejo, L.F., Polo, J., Martín, L., 2009.
forecasting model with the assistance of aerosol index data. IEEE Trans. Sustain.
Analysis of different comparison parameters applied to solar radiation data
Energy 6, 434–442.
from satellite and German radiometric stations. Sol. Energy 83, 118–125.
Lonij, V.P.A., Brooks, A.E., Cronin, A.D., Leuthold, M., Koch, K., 2013. Intra-hour
Fausett, L., 2006. Fundamentals of Neural Networks: Arquitectures, Algorithms, and
forecasts of solar power production using measurements from a network of
Applications. Pearson Education India.
irradiance sensors. Sol. Energy 97, 58–66.
Gohari, M.I., Urquhart, B., Yang, H., Kurtz, B., Nguyen, D., Chow, C.W., Ghonima, M.,
Lorenz, E., Scheidsteger, T., Hurka, J., Heinemann, D., Kurz, C., 2011. Regional PV
Kleissl, J., 2014. Comparison of solar power output forecasting performance of
power prediction for improved grid integration. Prog. Photovoltaics Res. Appl.
the total sky imager and the University of California, San Diego Sky Imager.
19, 757–771.
Energy Proc, 49, 2340–2350.
Lucheroni, C., 2009. A Resonating Model for the Power Market and its Calibration.
Hahn, H., Meyer-Nieberg, S., Pickl, S., 2009. Electric load forecasting methods: tools
SSRN: <http://ssrn.com/abstract>.
for decision making. Eur. J. Oper. Res. 199, 902–907.
144 M.Q. Raza et al. / Solar Energy 136 (2016) 125–144
Madan, M., Liang, J., Noriyasu, H., 2003. Static and Dynamic Neural Networks: From Rivas, V.M., Merelo, J.J., Castillo, P.A., Arenas, M.G., Castellano, J.G., 2004. Evolving
Fundamentals to Advanced Theory. John Wiley & Sons, Canada. RBF neural networks for time-series forecasting with EvRBF. Inf. Sci. 165, 207–
Malki, H.A., Karayiannis, N.B., Balasubramanian, M., 2004. Short-term electric power 220.
load forecasting using feedforward neural networks. Expert Syst. 21, 157–167. Rodrigyuez, G.D., 2010. A utility perspective of the role of energy storage in the
Mandal, P., Madhira, S.T.S., Haque, A.U., Meng, J., Pineda, R.L., 2012. Forecasting smart grid. In: Power and Energy Society General Meeting, 2010 IEEE, 25–29
power output of solar photovoltaic system using wavelet transform and July 2010, pp. 1–2.
artificial intelligence techniques. Proc. Comput. Sci. 12, 332–337. Rosenblatt, F., 1962. Principles of Neurodynamics.
Marín, F.J., Sandoval, F., 1997. Short-term peak load forecasting: statistical methods Sansa, I., Missaoui, S., Boussada, Z., Bellaaj, N.M., Ahmed, E.M., Orabi, M., 2014. PV
versus artificial neural networks. In: Mira, J., Moreno-Díaz, R. (Eds.), From power forecasting using different artificial neural networks strategies. In: 2014
Neuroscience to Technology Biological and Artificial Computation. Springer, International Conference on Green Energy. IEEE, pp. 54–59.
Berlin, Heidelberg. Sfetsos, A., Coonick, A.H., 2000. Univariate and multivariate forecasting of hourly
Marquez, R., Coimbra, C.F., 2013. Proposed metric for evaluation of solar forecasting solar radiation with artificial intelligence techniques. Sol. Energy 68, 169–178.
models. J. Sol. Energy Eng. 135, 011016. Shah, B.M., Syahiman, A., Yokoyama, H., Kakimoto, N., 2015. High-precision
Martín, L., Zarzalejo, L.F., Polo, J., Navarro, A., Marchante, R., Cony, M., 2010. forecasting model of solar irradiance based on grid point value data analysis
Prediction of global solar irradiance based on time series analysis: application for an efficient photovoltaic system. IEEE Trans. Sustain. Energy 6, 474–481.
to solar thermal power plants energy production planning. Sol. Energy 84, Shah, R., Mithulananthan, N., Bansal, R., Ramachandaramurthy, V., 2015. A review of
1772–1781. key power system stability challenges for large-scale PV integration. Renew.
Mathiesen, P., Kleissl, J., 2011. Evaluation of numerical weather prediction for intra- Sustain. Energy Rev. 41, 1423–1436.
day solar forecasting in the continental United States. Sol. Energy 85, 967–977. Shekhar, S., Amin, M.B., 1992. Generalization by neural networks. IEEE Trans.
McCulloch, W.S., Pitts, W., 1943. A logical calculus of the ideas immanent in nervous Knowl. Data Eng. 4, 177–185.
activity. Bull. Math. Biophys. 5, 115–133. Shi, J., Lee, W.-J., Liu, Y., Yang, Y., Wang, P., 2012. Forecasting power output of
Mellit, A., Kalogirou, S.A., 2008. Artificial intelligence techniques for photovoltaic photovoltaic systems based on weather classification and support vector
applications: a review. Prog. Energy Combust. Sci. 34, 574–632. machines. IEEE Trans. Ind. Appl. 48, 1064–1069.
Mohamed, Z., Bodger, P.S., 2004. Forecasting Electricity Consumption: A Smith, C., Jin, Y., 2014. Evolutionary multi-objective generation of recurrent
Comparison of Models for New Zealand. neural network ensembles for time series prediction. Neurocomputing 143,
Mori, H., Takahashi, M., 2012. Development of GRBFN with global structure for PV 302–311.
generation output forecasting. In: 2012 IEEE Power and Energy Society General Soman, S.S., Zareipour, H., Malik, O., Mandal, P., 2010. A review of wind power and
Meeting. IEEE, pp. 1–7. wind speed forecasting methods with different time horizons. In: North
Morris, R., 1999. DO Hebb: the organization of behavior, Wiley: New York; 1949. American Power Symposium (NAPS), 2010, 26–28 Sept. 2010, pp. 1–8.
Brain Res. Bull. 50, 437. Tao, Y., Chen, Y., 2014. Distributed PV power forecasting using genetic algorithm
Mueller, R.W., Dagestad, K.F., Ineichen, P., Schroedter-Homscheidt, M., Cros, S., based neural network approach. In: 2014 International Conference on Advanced
Dumortier, D., Kuhlemann, R., Olseth, J.A., Piernavieja, G., Reise, C., Wald, L., Mechatronic Systems (ICAMechS). IEEE, pp. 557–560.
Heinemann, D., 2004. Rethinking satellite-based solar irradiance modelling: the Tasre, M.B., Bedekar, P.P., Ghate, V.N., 2011. Daily peak load forecasting using ANN.
SOLIS clear-sky module. Rem. Sens. Environ. 91, 160–174. In: Nirma University International Conference on Engineering (NUiCONE). IEEE,
Nilsson, N.J., 1965. Learning Machines. pp. 1–6.
Omran, W. Performance Analysis of Grid-connected Photovoltaic System. PhD Tuyishimire, B., Mccann, R., Bute, J., 2013. Evaluation of a Kalman predictor
Dissertation. PhD, University of Waterloo. approach in forecasting PV solar power generation. In: 2013 4th IEEE
Oudjana, S., Hellal, A., Mahammed, I.H., 2013. Power forecasting of photovoltaic International Symposium on Power Electronics for Distributed Generation
generation. Int. J. Electr. Electron. Sci. Eng., 7 Systems (PEDG). IEEE, pp. 1–6.
Paoli, C., Voyant, C., Muselli, M., Nivet, M.-L., 2010. Forecasting of preprocessed daily UQ solar <http://solar-energy.uq.edu.au/> [Online]. [Accessed].
solar radiation time series using neural networks. Sol. Energy 84, 2146–2160. Voyant, C., Muselli, M., Paoli, C., Nivet, M.-L., 2011. Optimization of an artificial
Patterson, D.W., 1998. Artificial Neural Networks: Theory and Applications. Prentice neural network dedicated to the multivariate forecasting of daily global
Hall PTR. radiation. Energy 36, 348–359.
Pedro, H.T.C., Coimbra, C.F.M., 2012. Assessment of forecasting techniques for solar Wengenmayr, R., Bührke, T., 2011. Renewable Energy: Sustainable Energy Concepts
power production with no exogenous inputs. Sol. Energy 86, 2017–2028. for the Future. John Wiley & Sons.
Pelland, S., Galanis, G., Kallos, G., 2013. Solar and photovoltaic forecasting through Widrow, B., Hoff, M.E., 1960. Adaptive Switching Circuits.
post-processing of the global environmental multiscale numerical weather Widrow, B., Winter, R., 1988. Neural nets for adaptive filtering and adaptive pattern
prediction model. Prog. Photovoltaics Res. Appl. 21, 284–296. recognition. Computer 21, 25–39.
Perez, R., Beauharnois, M., Hemker, K., Kivalov, S., Lorenz, E., Pelland, S., Schlemmer, Williams, R.J., Zipser, D., 1989. Experimental analysis of the real-time recurrent
J., Van knowe, G., 2011. Evaluation of numerical weather prediction solar learning algorithm. Connect. Sci. 1, 87–111.
irradiance forecasts in the US. In: ASES Annual Conference. Raleigh, NC, USA, p. Yang, C., Thatte, A., Xie, L., 2015. Multitime-scale data-driven spatio-temporal
011. forecast of photovoltaic generation. IEEE Trans. Sustain. Energy 6, 104–112.
Perez, R., Kivalov, S., Schlemmer, J., Hemker, K., Renné, D., Hoff, T.E., 2010. Validation Yang, H.-T., Huang, C.-M., Huang, Y.-C., Pai, Y.-S., 2014. A weather-based hybrid
of short and medium term operational solar radiation forecasts in the US. Sol. method for 1-day ahead hourly forecasting of PV power output. IEEE Trans.
Energy 84, 2161–2172. Sustain. Energy 5, 917–926.
Photovoltaics, S., 2012. Renewable Energy Technologies: Cost Analysis Series. Yona, A., Senjyu, T., Funabashi, T., Chul-Hwan, K., 2013. Determination method of
International Renewable Energy Agency (IRENA) Working Paper. insolation prediction with fuzzy and applying neural network for long-term
Ramsami, P., Oree, V., 2015. A hybrid method for forecasting the energy output of ahead PV power output correction. IEEE Trans. Sustain. Energy 4, 527–533.
photovoltaic systems. Energy Convers. Manage. 95, 406–413. Zagouras, A., Pedro, H.T.C., Coimbra, C.F.M., 2015. On the role of lagged exogenous
Raza, M.Q., Baharudin, Z., 2012. A review on short term load forecasting using variables and spatio–temporal correlations in improving the accuracy of solar
hybrid neural network techniques. In: Power and Energy (PECon), 2012 IEEE forecasting methods. Renew. Energy 78, 203–218.
International Conference on, 2–5 Dec. 2012, pp. 846–851. Zervos, A., 2015. Renewables 2015 Global Status Report. REN 21.
Raza, M.Q., Baharudin, Z., Badar-Ul-islam, B.-U.-I., Nallagownden, P., 2014. A Zervos, A., Lins, C., Muth, J., 2010. RE-thinking 2050: a 100% Renewable Energy
comparative analysis of neural network based short term load forecast models Vision for the European Union, EREC.
for anomalous days load prediction. J. Comput. 9, 1519–1524. Zhang, H.L., Baeyens, J., Degrève, J., Cacères, G., 2013. Concentrated solar power
Raza, M.Q., Khosravi, A., 2015. A review on artificial intelligence based load demand plants: review and design methodology. Renew. Sustain. Energy Rev. 22, 466–
forecasting techniques for smart grid and buildings. Renew. Sustain. Energy 481.
Rev. 50, 1352–1372. Zhang, J.-R., Zhang, J., Lok, T.-M., Lyu, M.R., 2007. A hybrid particle swarm
Reikard, G., 2009. Predicting solar radiation at high resolutions: a comparison of optimization-back-propagation algorithm for feedforward neural network
time series forecasts. Sol. Energy 83, 342–349. training. Appl. Math. Comput. 185, 1026–1037.
Rezk, H., Tyukhov, I., Raupov, A., 2015. Experimental implementation of Zhang, J., Florita, A., Hodge, B.-M., Lu, S., Hamann, H.F., Banunarayanan, V.,
meteorological data and photovoltaic solar radiation monitoring system. Int. Brockway, A.M., 2015. A suite of metrics for assessing the performance of
Trans. Electr. Energy Syst. 25, 3573–3585. solar power forecasting. Sol. Energy 111, 157–175.