A REVIEW ON SOLAR RADIATION PREDICTION USING DIFFERENT MACHINE LEARNING TECHNIQUES

(1)

iJournals: International Journal of Software & Hardware Research in Engineering ISSN-2347-4890 Volume 6 Issue 6 June, 2018

Ms. Sachin Kale_,Mr. Sunil Kumar Bhatt, vol 6 Issue 6, pp 12-16, June 2018

A REVIEW ON SOLAR RADIATION PREDICTION USING DIFFERENT MACHINE LEARNING TECHNIQUES

Authors: Mr. Sachin Kale

¹,

Mr. Sunil Kumar Bhatt

²

1 M. Tech Scholar in Power System CIIT, Indore (M.P). E-mail: ([email protected]) 2 Assistant Professor in Electrical & Electronics Dept. CIIT, Indore (M.P). E-mail: ([email protected])

Abstract:

In this work a study is done on understanding the role of ANN in predicting hourly solar radiation value of any location. What make predicting solar radiation important is the fact that solar energy has a great potential along with wind energy to fulfill energy demands of growing India. By looking at past decades record a continuous growth can be seen in new plants installation every year.

This is due to the supportive Government policies and interest of investment companies. The other factor which catalyzed this process is the new development in technology which is making it cheap and reliable source of energy to compete with conventional source of energy.

India itself is have installed around 10 GW of solar plants in 2017 which is a rise of about 70% than that of 2016 which is a great leap. Also after this year India is third biggest solar market globally. The prime objective of this work is to review Artificial Neural Network based techniques for finding best methods available in the literature for solar radiation prediction. The work shows that Artificial Intelligence are far more accurate than conventional methods.

Keywords: Solar irradiation, Solar PV generation, Artificial Neural Network, Artificial intelligence, Backpropagation, mean absolute percentage error,MATLAB.

I. Introduction:

It has been very well known from the past several years that a due to expected increase in world population and development there will be exponential increase in energy requirement in every part of the world. This high demand need to be fulfilled by new power plants. Now since considering most of the power plants present day run on fossil fuels any new addition will lead to increase in greenhouse gas emission which leads to global warming.

Also, it is expected that fossil fuels reserves will end in

coming few decades hence they may work as a temporary alternative only. Hence permanent source has to be found and research must be done to make those sources more efficient and reliable. Considering India’s geographical conditions solar and wind energy are two such major alternative sources of energy which has the potential of behaving as backhand to supply additionally required energy. Now problem with both this source is that of inconsistency. Solar being a priceless source require a device call solar pv cell to convert it to electrical energy directly. The conversion can also take place using concentrated solar power plant.

The problem that made solar less popular in past is less technical advancement and irregular performance due to uncertain output generation which cause problems for grid connection and high cost of energy storage system.

These problems are addressed affectively in the past decade or so which can be clearly seen from the continuous rise in installed solar capacity throughout the world. The output of PV cell is severely affected by two main factors one day night cycle and other cloud structure. Hence before integrating solar plant with grid one need to perform accurate availability evaluation. To effectively integrate enlarged levels of solar power generation while keeping reliability is the biggest test for solar energy supply and makes the accessibility of accurate information an important requirement.

II. Literature review

Ravinesh C. Deoa, Mehmet Şahin in [1] discussed how prior knowledge of solar radiation will be very helpful in may engineering application, especially in predicting energy generation from solar pv system and in site selection for solar power plant. Author concluded that it is due to various fiscal issues and policies which drive this study to find a method which can provide input parameters for the sake of predicting solar radiation using model which is cheap easily available and maintenance free. This study mainly focused on preparing model using ANN which is capable of

(2)

© 2018, IJournals All Rights Reserved www.ijournals.in Page 13

predicting solar irradiation with the help of MODIS-LST

as input parameters.

Maria Malvoni, Paolo Maria Congedo, Maria Grazia de Giorgi in [2] proposed that the continuous altering nature of weather directly affect the PV cell output. In this work author has checked the power of data driven techniques in which input data are preprocessed before supplying to model. Author has developed model to forecast day ahead pv power generation using meteorological data used as inputs along with Wavelet analysis and other mathematical tools. For the sake of forecasting a day ahead PV power generation a time series forecasting method as the GLSSVM is used which is the combination of LS-SVM and GMDH is applied.

Cyril Voyant, Gilles Notton, en. al. in [3] has propounded that having a prior knowledge of output of solar pv system will be very helpful in efficient management of power grid. There are many techniques which are developed for forecasting solar irradiation which can be roughly classified as physical methods and Neural network methods. Author mainly focused on different intelligent techniques and their comparison is made. Author mainly focused his study on hybrid model’s regression tree, random forest, gradient boosting and many others to check their efficiency.

Daniel O’Leary and Joel Kubby in [4] presented a new approach based on image processing and ANN for solar power prediction for supporting small scale energy markets and individuals. Hence for the aim to reduce losses and expenses of producing solar power without affecting natural elements for both commercial and domestic fields require a good forecasting system. In this study author has utilized image edge detection technique for the acoustic signal classification.

L. Saad Saoud, F. Rahmoune, V. Tourtchine, K. Baddari in [5] has proposed that due to importance of forecasting solar irradiation in planning pv power plant it is very necessary to develop a good forecasting model. In this work authors have developed a Complex Valued Wavelet Network (FCWN) for this purpose. The data for training and testing this improved wavelet NN is meteorological data of Tamanrasset city, Algeria. Bothn daoly and hourly solar irradiation value is forecasted and the results shown to has significant improvement.

III. Renewable Energy Status

It has been clearly seen that Indian economy is continuously growing and is still expected to remain on the same path unlike Global economy which is following a low-level Growth. The main reason behind this is consider to be the power sector of India which is rising at

fast rate. As on 30-09-2017 the total installed capacity of India has reached to 330 GW. This power cumulative contribution of different sources. Thermal is still the prime source of energy with a total installed capacity of 218 GW contributing to around 66% of total power.

Then comes share of renewable energy sources which contribute to around 60GW which is around 18% of power in India. Hydro power on the other hand contributes to around 45GW of power generation which is about 13.5% and remaining power is generated using Nuclear power which is about 2% of total generation. It is evident that the renewable power has secured 2nd position after Thermal and is spreading its wings rapidly in India [18]. At present Indian government has made a target to achieve an install capacity of renewables in India to about 175 GW by the year 2022, out of which the main contributor will be solar power which will contribute to around 65% of total capacity due to high potential of solar power in Indian subcontinent.

Figure 1 Source wise Power Installed Capacity as on 31.12.2017.

IV. Proposed Methodology

Artificial Neural Networks (ANN): ANN are computing systems or technique that mimic the learning processes of the brain to discover the relations between the variables of a system. They process input data information to learn and obtain knowledge for forecasting or classifying patterns etc. type of work.

ANN consists of number of simple processing elements called neurons. All information processing is done within this neuron only.

Levenberg–Marquardt (LM) Algorithm: Reducing error function is the main reason to use this algorithm.

Levenberg-Marquardt algorithm [8] [9] is a very efficient technique for minimizing a nonlinear function. The algorithm includes many different variables like in

(3)

© 2018, IJournals All Rights Reserved www.ijournals.in Page 14

present study we have output data, weight between

neurons and error function, that determine efficiency and success rate of model. The ideal values of these variables are very dependent on the test function.

Levenberg-Marquardt algorithm is fast [10] and has stable convergence. When the performance function has the form of a sum of squares, which contains first order derivatives of the network errors with respect to the weights and biases, � is a vector of network errors. The Jacobian matrix can be computed through a standard backpropagation technique that is much less complex than computing the Hessian matrix [11]. The following is the relation for LM algorithm computation,

𝑊

_𝑘+1=

𝑊

_𝑘− [

𝐽

_𝐾^𝑇

𝐽

_𝑘+ µ

𝐼

]⁻¹

𝐽

_𝐾^𝑇

𝑒

_𝑘 (1)

where I is the identity matrix, W_k is the current weight, W_k+1is the next weight, e_k+1is the current total error, and e_k is the last total error, μ is combination coefficient. It tries to combine the advantages of both the methods hence it inherits the speed of the Gauss–Newton algorithm and the stability of the steepest descent method.

Fuzzy Logic: The word “fuzzy” usually used for cases in which there is no clear answer or boundary like there is a vague situation. Fuzzy logic resembles the human decision taking methodology and deals with vague and imprecise information unlike classical set where precise and clear information is used for decision making. Fuzzy unlike Boolean logic where things can be either 0 or 1 i.e. true or false represent value with the degree of truth.

This fuzzy logic is applicable to all real-world problems because in none of them the boundary is clear. The value in fuzzy logic comprises of value between 0 & 1 Including two.

Fuzzy logic was proposed in the USA by Prof L. A.

Zadeh, in the early 1960s. FL is a superset of conventional (Boolean) logic that has been extended to handle the concept of partial truth-truth values between

“completely true” and “completely false”. Lofti A.

Zadeh, who is considered as father of fuzzy logic introduced fuzzy logic in 1965 in his research paper”

fuzzy Sets” Fuzzy logic is a way to make machines more intelligent, enabling them to reason in a fuzzy manner like humans. Fuzzy models “think” the way humans do and include verbal expressions instead of numbers. This method mimics the way operation of expert’s opinion over any decision making.

The structure of neural network both biological and artificial is such that each neuron is connected with other

neuron through a link. This link holds an information in the form of what we call weights based upon the relationship between input and output data provided to model. Now we will discuss in this section how fuzzy logic can be beneficial to be used in Neural network:

● Fuzzy logic is used mainly to optimize weights based on fuzzy set theory.

● Fuzzy is used when conventional set theory can’t be used.

Training validation and learning help neural networks do well in unpredicted circumstances. In those cases, fuzzy values would be more applicable than convention binary values.

Figure 2: Working model of an ANN

The Neural Network in present study consist of three layers. The first one is Input layer (consist of 4 Neutrons for 4 input element) from where inputs are feed to model for further computation. Then comes second layer called Hidden Layer (consist of 10 Neurons), this is where activation function is used to limit value of output Neuron. At last we have output layer (1 Neuron) form which we take output result for comparison with actual result to calculate error and the feeding it back to model to vary weight accordingly for improving Performance.

V. Conclusion:

In this paper, we have discussed different solar radiation forecasting techniques. From the work reported by different researchers, it can be concluded that the artificial intelligence-based forecasting algorithms are proved to be potential techniques for this challenging job of nonlinear time series prediction. Further a use of fuzzy logic-based ANN model could result in much better results much close prediction of mon linear time series.

Hence the two techniques are studied.

(4)

© 2018, IJournals All Rights Reserved www.ijournals.in Page 15 REFERENCES:

[1] Ravinesh C. Deoa, Mehmet Şahin,“Forecasting long- term global solar radiation with an ANN algorithm coupled with satellite-derived (MODIS) land surface temperature (LST) for regional locations in Queensland”, Renewable and Sustainable Energy Reviews (2017), pp 828–848, Elsevier, 2017.

DOI: http://dx.doi.org/10.1016/j.rser.2017.01.114 [2] Maria Malvoni, Paolo Maria Congedo, Maria Grazia de Giorgi,“Forecasting of PV Power Generation using weather input data‐ preprocessing techniques”, 72nd Conference of the Italian Thermal Machines Engineering Association, ATI2017,Energy Procedia, pp 651-658, 6-8 September 2017,Lecce Italy, Elsevier, 2017.

DOI: 10.1016/j.egypro.2017.08.293

[3] Cyril Voyant, Gilles Notton, Soteris Kalogirou, Marie-Laure Nivet, Christophe Paoli, Fabrice Motte, Alexis Fouilloy, “Machine learning methods for solar radiation forecasting: A review”, Renewable Energy 105, pp 569-582, Elsevier, 2017.

DOI: http://dx.doi.org/10.1016/j.renene.2016.12.095 [4] Daniel O’Leary and Joel Kubby, “Feature Selection and ANN Solar Power Prediction”, Journal of Renewable Energy, Hindawi, 2017.

DOI: https://doi.org/10.1155/2017/2437387

[5] L. Saad Saoud, F. Rahmoune, V. Tourtchine, K.

Baddari, “Fully Complex Valued Wavelet Network for Forecasting the Global Solar Irradiation”, Neural Process Lett, Springer, 2016.

DOI: 10.1007/s11063-016-9537-7

[6] S.P.Sharma et al., Prediction of Solar Energy Based on Intelligent ANN Modeling , International Journal Of Renewable Energy Research, Vol.6, No.1, 2016

[7] Emre Akarslan, Fatih Onur Hocaoglu, “A novel adaptive approach for hourly solar radiation forecasting”, Renewable Energy 87 (2016) pp 628-633, Elsevier, 2015.

DOI: http://dx.doi.org/10.1016/j.renene.2015.10.063 [8] Hakan Kutucu, Ayad Almryad, “Modeling of Solar Energy Potential in Libya using an Artificial Neural Network Model”, IEEE First International Conference on Data Stream Mining & Processing, 23-27 August 2016, Lviv, Ukraine. IEEE, 2016.

[9] Premalatha Neelamegama, Valan Arasu Amirtham,

“Prediction of solar radiation for solar systems by using ANN models with different back propagation algorithms”, Journal of Applied Research and Technology 14 (2016) pp 206–214, 2016.

DOI: http://dx.doi.org/10.1016/j.jart.2016.05.001 [10] John Boland, Mathieu David, Philippe Lauret,

“Short term solar radiation forecasting: Island versus continental sites”, Energy 113, pp 186-192, Elsevier, 2016.

DOI: http://dx.doi.org/10.1016/j.energy.2016.06.139 [11] Masoud Vakilia, Saeed-Reza Sabbagh-Yazdib, Koosha Kalhorb, Soheila Khosrojerdic, “Using Artificial Neural Networks for Prediction of Global Solar Radiation in Tehran Considering Particulate Matter Air Pollution”, Energy Procedia 74, pp 1205 – 1212, International Conference on Technologies and Materials for Renewable Energy, Environment and Sustainability, TMREES15, 2015.

DOI: 10.1016/j.egypro.2015.07.764

[12] João Faceira, Paulo Afonso, and Paulo Salgado,

“Prediction of Solar Radiation Using Artificial Neural Networks”, CONTROLO’2014 - Proc. of the 11th Port.

Conf. on Autom. Control, Springer 2015 DOI: 10.1007/978-3-319-10380-8_38

[13] Amit Kumar Yadav, S.S. Chandel, “Solar radiation prediction using Artificial Neural Network techniques: A review”, Renewable and Sustainable Energy Reviews (2013), Elsevier,2013.

DOI: http://dx.doi.org/10.1016/j.rser.2013.08.055i [14] Hong-Tzer Yang, Chao-Ming Huang, Yann-Chang Huang, and Yi-Shiang Pai, “A Weather-Based Hybrid Method for 1-Day Ahead Hourly Forecasting of PV Power Output”, IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 3, JULY 2014.

DOI: 10.1109/TSTE.2014.2313600

[15] Jompob Waewsaka, Chana Chanchama, Marina Mania and Yves Gagnonb, “Estimation of Monthly Mean Daily Global Solar Radiation over Bangkok, Thailand using Artificial Neural Networks”, Energy Procedia 57, pp 1160 – 1168, 2013 ISES Solar World Congress.2013.

DOI: 10.1016/j.egypro.2014.10.103

(5)

[16] Rich H. Inman, Hugo T.C. Pedro, Carlos F.M.

Coimbra, “Review Solar forecasting methods for renewable energy integration”, Progress in Energy and Combustion Science 39 (2013) 535-576, Elsevier, 2013.

DOI: http://dx.doi.org/10.1016/j.pecs.2013.06.002 [17] E. G. Kardakos, M. C. Alexiadis, S. I. Vagropoulos, C. K. Simoglou, P. N. Biskas, and A. G. Bakirtzis,

“Application of Time Series and Artificial Neural Network Models in Short-term Forecasting of PV Power Generation”, IEEE 2013.

[18] Ministry of power, Government of India.

https://powermin.nic.in/en/content/power-sector-glance- all-india.

[19] R. Hyndman, and A, Koehler "Another look at measures of forecast accuracy”, International Journal of Forecasting 22, pp. 679 – 688, 2005.

[20] Alexiadis, M.C., Dokopoulos, P.S., Sahasamanoglou, H.S., Manousaridis, I.M.: Short-term forecasting of wind speed and related electrical power.

Solar Energy Volume 63, Pages 61–68, July 1998.

[21] Giorgi, M.G.D., Ficarella, A., Tarantino, M. “Error analysis of short term wind power prediction models”.

Appl. Energy 88, 1298–1311, 2011.

A REVIEW ON SOLAR RADIATION PREDICTION USING DIFFERENT MACHINE LEARNING TECHNIQUES