See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/270885637

Solar Power forecasting: The State-Of-The-Art

Article · June 2014

CITATIONS READS
3 4,291

2 authors:

Kumar Vaibhav Vikas pratap Singh

Amity University Rajasthan. Eaton Technologies
12 PUBLICATIONS   29 CITATIONS    17 PUBLICATIONS   88 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Solar PV power generation forecasting Using Neural Network Based approaches View project

Currently working in Holograms. View project

All content following this page was uploaded by Kumar Vaibhav on 22 January 2015.

The user has requested enhancement of the downloaded file.

 Solar Power forecasting: The State-Of-The-Art
Vikas Pratap Singha,b and Kumar Vaibhavc
a-Central Power Research Institute
b-Indian Institute of Technology Jodhpur
c-Amity University Jaipur

ABSTRACT:
In 21st century Renewable energy sources, especially Solar Energy, are to play a larger role in
Hybrid Generation. There exist a number of technological, environmental and political challenges linked
to supplementing existing electricity generation capacities with solar energy. Solar power forecasting can
avoid many of the balancing issues, if accurate forecasts of solar output are available. Anybody trading
in solar energy can negotiate a better price, if they have precise information about the volume they have to
sell at any particular time.

Keywords: Solar power forecasting, PV power

1 INTRODUCTION
Solar power is play is an important role in The forecasting challenges involve a
distributed power system. Photovoltaic systems combination of Weather parameters, Solar
have been increasingly installed worldwide in Radiation, Mathematical modeling, and Solar PV
recent years. Because it produces clean energy, Power generation monitoring data. For the day-
moreover the development of technology is ahead period, the challenges involve forecasting
continued therefore the reliability is increasing and the dynamic changes in weather parameters and
the price is decreasing in opposite. To implement cloud cover. One challenge affecting solar power
the PV system, however, a significant limitation of forecasts for all timescales is the need for solar
PV system is the uncertainty of power from the plant models that accurately convert irradiance
sun. This will affect the quality of the electrical values to power forecasts. In order to effectively
system that connected [1]. support the solar power providers and grid
Solar power forecasting is playing a key role in managers, the meteorological community needs to
solar PV park installation, operation and accurate prioritize the current shortfalls in solar forecasting
solar power dispatch ability as well as scheduling. and identify opportunities to improve them.
Solar Power is varies with solar radiation and
weather parameters such as ambient temperature, This short term power forecasting helps in the
wind velocity, cloud coverage [2] [3]. In order to following power system areas such as: [4]
effectively achieve a high-penetration of
commercial solar power into the grid, useful a) Control
forecasting methodologies must be developed for b) Unit Commitment
integration of solar power in electric grid. The c) Security Assessment
existing electric grid was designed and created to d) Optimum planning of power generation
safely and reliably distribute power from a few e) Energy exchange
concentrated power generation sources through f) Grid integration
highly monitored and controlled transmission lines
to typically radial distributed loads. So we better The detrimental impact of large volumes of PV on
monitoring and control system for performance an unprepared grid involve (1) voltage violations
analysis and operation of SPV system. (2) flicker and other power quality issues, (3)
reverse power flow and protection coordination
issues, (4) increased wear on utility equipment,

*Senior Research Fellow, ERED, Central Power Research Institute Bangalore Email:vikasforsmile@gmail.com
Phone no:7829140894
**Assistant Professor, Amity University, Jaipur Rajasthan, Email: kumarvaibhav2003@gmail.com
and (5) real and reactive power imbalances. Solar 2.1.1 CS Model 1
forecasting methods in concert with distribution Polynomial Fit: The first clear sky model that
circuit operating controls can help mitigate the we are considering is a third degree polynomial
impact and compensate for relevant conditions and function of cosine of the solar zenith angle
anomalies on distribution circuits including cos(θ):
transient and dynamic voltage fluctuations, reverse I𝑐𝑙𝑟.𝑝𝑙𝑜𝑦 = C3 (cosθ)3 + C2 (cosθ)2 +
power flow, unintentional islanding and local
C1 (cosθ) + C0 ……………………………………….(1)
thermal impacts [5].
This model was introduced by University of
This paper is organized as follows: Section II California Merced solar observatory station.
explains the literature review on solar power
forecasting. In Section III, the technologies for 2.1.2 CS Model 2: ESRA
Solar power forecasting Section IV shows the Another clear sky model, it is computed using the
methodologies used for conventional forecasting ESRA model adopted from Refs. [8, 9]. The
and finally, the conclusion of this paper is ESRA model only depends on a site dependent
provided in Section V. Linke-Turidity Factor (TL).

2.2 Persistence models

2 CONVENTIONAL MODELS FOR
SOLAR POWER FORECASTING 2.2.1 Clear Sky Persistence Models

In distributed power generation solar power In Ref [7] clear sky persistence models are
forecasting play an important role and it has defined as having the clear sky conditions (ratio
amplified day by day. Several recent studies deal between the measured irradiance to the clear sky
with the problem. Many of these consider irradiance) persist for the next time-step:
forecasts of the global irradiance which is
essentially the same problem as forecasting solar 𝐼(𝑡)
𝑘̂(𝑡 + ∆𝑡) = 𝑘 ∗ (𝑡) = 𝐼(𝑡) ………..…….…….(2)
power. For solar power forecasting following 𝑐𝑙𝑟

conventions models are discussed here:

2.2.2 Clearness Persistence Model
 Clear sky models As a slight variation of the persistence models just
 Persistence models described, we also apply a clearness persistence
model which is based on the extraterrestrial solar
2.1 Clear-Sky Models
radiation (I0).
Here we only choose two clear sky models, one I0=G0 cos(θz)…………………………………..(3)
which is a based on a polynomial regression and
the other based on the European Solar
Here
Radiation Atlas (ESRA) clear sky model, so that
the sensitivity of the methods for forecasting of G0=1367 W/m2 is the solar constant.
clear sky days can be examined[6]. The θz
sensitivity should be small since short-term
forecasting involves the use of previous The Characteristics of inputs for Solar-forecasting
irradiance values that can be used to adjust the Techniques and Important variables in solar power
clear sky forecasting [7]. forecasting is shown in table. [1-2]. in above
models inputs parameters monitoring in particular
sampling rate for particular application [5].
 TABLE -1
CHARACTERISTICS OF AND INPUTS FOR SOLAR-FORECASTING TECHNIQUES
Sl. No. Techniques Sampling Spatial Spatial Suitable Application
rate resolution extent forecast
horizon
1. Persistence High 1 point 1 point Minutes Baseline
2. Total sky 30 s 10s -100s 2-5m radius Tense of Short term
imager minutes ramps
regulation
3. Satellite 15 Minutes 1 km U.S. 5 Hours Load following
imagery
4. Weather model 1 Hour 12 km U.S. 10 Day Unit
commitments

TABLE-2
IMPORTANT VARIABLES IN SOLAR POWER FORECASTING
Sl. No. Forecasting Application Primary Important to Current
Variable determinants market forecast skill
1. GI PV Clouds, Solar High Medium
geometry
2. Cell temperature PV GI, Air Low High
temperature,
wind
3. DNI Concentrating solar Clouds, Medium low
power aerosols, water
vapour

3 APPROACHES FOR SOLAR POWER electrical energy of PV systems. For long term
FORECASTING forecasting the methods are based on time series
weather parameters [10] or solar power plant
We can classify the forecasting approaches into performance data historical data, as well as the
three categories, time series based approaches, typical meteorological year (TMY) data which
and Model Output Statistics (MOS) and soft provide the data for typical months of a
computing. synthetic year are only useful to predict the
average monthly or even daily PV array
performance. A time scale smaller than the day
3.1 Time series approach
requires knowledge of the cloud cover and their
expected instantaneous changes [11].
The traditional forecast of PV energy is based on
Autoregressive Models (AR) [12], Moving
the time series of solar energy and weather
conditions, which are used to calculate the
Average (MA) and Autoregressive Moving 3.3.2 Hybrid models approach
Averages (ARMA) models are frequently used to
model linear dynamic structures [13], to depict Other classes of models combining neural
linear relationships among lagged variables and networks with other techniques have been
to serve as vehicles for linear forecasting. Non- proposed in order to overcome the shortcomings
stationary processes can be modeled by of neural nets. They are known as hybrid models.
differencing the original process to obtain a We can find the wavelet decomposition, genetic
stationary process. Multiple differencing may algorithm based model that combine NN [22].
sometimes be required in order to achieve
stationary. This results in an autoregressive 4 TIME ATTRIBUTE
integrated moving average ARIMA (p,d,q) In solar power forecasting, needful step is to
model [14]. specify the required time attributes. These include:

3.2 Model Output Statistics (MOS) approach 4.1 Time period

The Model Output Statistics (MOS) [15] was also Total amount of data included in the analysis. This
used in the early attempts in irradiance can range from a few minutes to many years. This
forecasting. MOS is a post-processing technique paper focuses on one year of data.
used to objectively interpret numerical model
output. It determines statistical relationship
between observed weather elements and variables 4.2 Time interval
forecast by a numerical model at some projection How the data in the time period is binned. This
time [16, 17]. In ref. [18] was the first application can range from a few seconds to annually. For
of MOS technique to predict daily solar radiation example, if the time period is one year and the
for the short term, up to two days in advance time interval is one hour, the time period would be
binned into 8,760 increments. This paper
3.3 Soft Computing approach examines one-minute to one-year time intervals.

3.3.1 Neural networks

4.3 Time perspective
Neural Networks (NN) are a class of nonlinear When the predicted observation is reported. This
functional forms which have been developed can range from historical, (backward looking) to
separately from standard regression techniques. forecasted a few hours ahead, to forecasted
Fitting the network involves training the model multiple days ahead (forward looking).
over known input and output values; the algorithm
adjusts the hidden and output node weights until
the output approximates the actual data within a 5 ASSESSMENT OF SOLAR
given threshold. Training is accomplished using a FORECASTING
back-propagation algorithm, which is analogous to The uncertainty and high variability in nature of
the steepest descent algorithms used in nonlinear solar power generation due to solar radiation and
regression, except that the derivatives for each weather parameters the choice of error metric is
weight are adjusted separately. For this reason, the very important for the evaluation of solar forecast
time involved in training can be considerable. models. Recent studies suggest that in order to
Forecast studies with NN are found in [19,20, 21]. facilitate higher market and grid penetration of
solar power, ISOs need accurate forecasts of solar
irradiance on multiple time horizons [23, 24, 25].
Traditional statistical metrics to characterize through highly monitored and controlled
model quality include the coefficient of transmission lines to typically radial distributed
determination which compares the variance of the loads.. As solar PV systems thrive, synergistic PV
errors to the variance of the data which is to be forecasting and control strategies can help not only
modeled is given by: integrate solar PV into the grid but can also help
optimize local distribution grid performance. To
𝑉𝑎𝑟(𝐼̂−𝐼) cope with energy demand, forecasting plays a
𝑅2 = 1 − …………………………….(4)
𝑉𝑎𝑟(𝐼) critical role in power-system management,
scheduling, and dispatch operations. System fault
the Root-Mean Squared Error (RMSE) which is a detection, minimizing extravagant consumption
measure of the average spread of the errors: with energy-efficient buildings and saving on peak
load energy generation for plants and systems are
1 the merits of energy forecasting.
𝑅𝑀𝑆𝐸 = √𝑁 ∑𝑁 ̂
𝑡=1(𝐼𝑡 − 𝐼𝑡 ) ……….……….….(5)
2

REFERENCES
The Mean Bias Error (MBE) which is a measure
of the average bias of the model: [1] A. Woyte, V. Van Thong, R. Belmans, and J.
Nijs, "Voltage fluctuations on distribution
1
𝑀𝐵𝐸 = 𝑁 ∑𝑁 ̂
𝑡=1(𝐼𝑡 − 𝐼𝑡 )……………………….(6) level introduced by photovoltaic systems,"
IEEE Trans. on Energy Conversion, vol. 21,
The Mean Absolute Error (MAE) which considers pp. 202-209, 2006.
only the absolute value of the errors:
[2] E. Lorenz, J. Hurka, D. Heinemann, and H. G.
1
̂ Beyer, "Irradiance Forecasting for the Power
𝑀𝐴𝐸 = 𝑁 ∑𝑁
𝑡=1 |𝐼𝑡 − 𝐼𝑡 |………………………..(7) Prediction of Grid-Connected Photovoltaic
Systems," IEEE Journal of Selected Topics in
The Mean Absolute Percentage Error (MAPE) Applied Earth Observations and Remote
Sensing, vol. 2, pp. 2-10,2009.
100% 𝐼̂𝑡 −𝐼𝑡
𝑀𝐴𝑃𝐸 = ∑𝑁
𝑡=1 | |……………….……(8)
𝑁 𝐼𝑡
[3] Mellit A., Arab A.H., Khorissi N., Salhi H, “an
ANIF based forecasting for solar radiation data
and the correlation coefficient
from sunshine duration and ambient
(𝐶𝑜𝑣(𝐼̂,𝐼)2 )
temperature” Power Engineering Society
𝜌 = 𝑉𝑎𝑟(𝐼̂)𝑉𝑎𝑟(𝐼)…………………………...……(9) General Meeting,2007. IEEEDigital Object
Identifier:10.1109/PES.2007.38631Publication
As a solution to these obstacles, a novel approach Year: 2007, Page(s): 1 – 6.
for the evaluation of the quality of forecast models
based on the comparison of solar resource [4]D. K. Chaturvedi. "Short-Term Load
variability and the forecast uncertainty is defined Forecasting Using Soft Computing
in ref. [26]. Techniques" International Journal of
Communications, Network and System
Sciences 3.1 (2010): 270-279)
6 CONCLUSION
[5]Marquez R, Coimbra CFM. “Forecasting of
The existing electric grid was designed and Global and Direct Solar Irradiance Using S
created to safely and reliably distribute power tochastic Learning Methods, Ground Experime
from a few concentrated power generation sources nts and the NWS Database”, Solar Energy. 201
1 (85), pp. 746-756
[6] Marquez R, Coimbra CFM. “Proposed Metric [16] D. Heinemann, E. Lorenz, and M. Girodo.
for Evaluation of Solar Forecasting Models” Forecasting of solar radiation. In E. Dunlop,
Journal of Solar Energy Engineering February L. Wald, and M. Suri, editors, Solar Energy
2013, Vol. 135, pp.011016-1-9 Resource Management for Electricity
Generation from Local Level to Global
[7] Ineichen, P., 2006, “Comparison of Eight Clear Scale, Nova Science Publishers, Hauppauge,
Sky Broadband Models Against 16 2006.
Independent Data Banks,” Sol. Energy, 80, pp.
468–478. [17] B. Chowdhury and S. Rahman. Forecasting
sub-hourly solar irradiance for prediction of
[8] Zarzalejo, L. F., Polo, J., and Ramirez, L., photovoltaic output. In IEEE Phtovoltaic
2004, “Gc_model5_irradiance,” (Matlab Specialist Conference, pages 171–176, New
computer program) . Orleans (USA), 1987.

[9] Badescu, V., 2008, Modeling Solar Radiation [18] J. Jensenius and G. Cotton. The development
at the Earth Surface, SpringerVerlag, and testing of automated solar energy
Berlin/Heidelberg. forecasts based on the Model Output
Statistics (MOS) technique. In 1st Workshop
[10]V. Puri. Estimation of half-hour solar on terrestrial solar resource forecasting and
radiation values from hourly values. Solar on use of satellites for terrestrial solar
Energy, 21:409–414, 1978. resource assessment, Washington, D.C.,
1981.
[11] B. Chowdhury and S. Rahman. Forecasting
[19] G. Reikard. Predicting solar radiation at high
sub-hourly solar irradiance for prediction of resolutions: A comparison of time series
photovoltaic output. In IEEE Phtovoltaic forecast. Solar Energy, 83:342–349, 2009.
Specialist Conference, pages 171–176, New
Orleans (USA), 1987. [20] M. Santamouris, G. Mihalakakou, B.
[12] P. Bacher, H. Madsen, and H. A. Nielsen. Psiloglou, G. Eftaxias, and D. N.
Asimmakopoulos. Modeling the global solar
Online short-term solar power forecasting.
radiation on the earth surface using
Solar Energy, 83(10):1772– 1783, 2009. atmospheric deterministic and intelligent data
driven techniques. Journal of Climate,
[13] A. Sfetsos and A. H. Coonick. Univariate and 12:3105–3116, 1999.
multivariate forecasting of hourly solar
radiation with artificial intelligence [21] A. Sfetsos and A. H. Coonick. Univariate and
techniques. Solar Energy, 68(2):169–178, multivariate forecasting of hourly solar
2000. radiation with artificial intelligence
[14] G. Reikard. Predicting solar radiation at high techniques. Solar Energy, 68(2):169–178,
resolutions: A comparison of time series 2000.
forecast. Solar Energy,83:342–349, 2009.
[22] Woyte, A., Belmans, R., Nijs, J. 2007.
[15] H. Glahn and D. Lowry. The use of Model “Fluctuations in instantaneous clearness
Output Statistics (MOS) in objective weather index: Analysis and statistics.” Solar Energy
forecasting. Journal of Applied Meteorology, 81 (2), 195-206.
11:1203–1211, 1972.
[23] D. Lew and R. Piwko, \Western wind and
solar integration study," tech. rep., National
Renewable Energy Labratories, 2010.

[24]V. Sundar, \Integration of renewable

resources: Operational requirements and
feneration eet capability at 20 percent rps,"
tech. rep., California Independent System
Operator (CAISO), 2010.

[25] G. D. Rodriguez, \A utility perspective of the

role of energy storage in the smart grid," in
Power and Energy Society General Meeting,
IEEE, pp. 1-2,2010.

[26] R. Marquez and C. F. M. Coimbra, \A

Proposed Metric for Evaluation of Solar
Forecasting Models," ASME Journal of Solar
Energy Engineering (in press), 2012.

View publication stats