A new hybrid support vector machine–wavelet transform approach for estimation of horizontal global solar radiation. (1st March 2015)

• Record Type: Journal Article
• Authors: Mohammadi, Kasra; Shamshirband, Shahaboddin; Tong, Chong Wen; Arif, Muhammad; Petković, Dalibor; Ch, Sudheer
• Is Part Of: Energy conversion and management. Volume 92(2015)
• Page Start: 162
• Page End: 171
• Publication Date: 2015-03-01
• Subjects: Global solar radiation estimation -- Support vector machine -- Wavelet transform algorithm -- Meteorological parameters -- Statistical indicators; Direct energy conversion -- Periodicals; Energy storage -- Periodicals; Energy transfer -- Periodicals; Énergie -- Conversion directe -- Périodiques; Direct energy conversion; Periodicals; 621.3105
• DOI: 10.1016/j.enconman.2014.12.050
• Languages: English
• ISSNs: 0196-8904

Abstract

Highlights: Horizontal global solar radiation (HGSR) is predicted based on a new hybrid approach. Support Vector Machines and Wavelet Transform algorithm (SVM–WT) are combined. Different sets of meteorological elements are used to predict HGSR. The precision of SVM–WT is assessed thoroughly against ANN, GP and ARMA. SVM–WT would be an appealing approach to predict HGSR and outperforms others. Abstract: In this paper, a new hybrid approach by combining the Support Vector Machine (SVM) with Wavelet Transform (WT) algorithm is developed to predict horizontal global solar radiation. The predictions are conducted on both daily and monthly mean scales for an Iranian coastal city. The proposed SVM–WT method is compared against other existing techniques to demonstrate its efficiency and viability. Three different sets of parameters are served as inputs to establish three models. The results indicate that the model using relative sunshine duration, difference between air temperatures, relative humidity, average temperature and extraterrestrial solar radiation as inputs shows higher performance than other models. The statistical analysis demonstrates that SVM–WT approach enjoys very good performance and outperforms other approaches. For the best SVM–WT model, the obtained statistical indicators of mean absolute percentage error, mean absolute bias error, root mean square error, relative root mean square error and coefficient of determination for daily estimation are 6.9996%, 0.8405 MJ/m 2, 1.4245 MJ/m 2, 7.9467% and 0.9086, respectively. Also, for monthly mean estimation the values are 3.2601%, 0.5104 MJ/m 2, 0.6618 MJ/m 2, 3.6935% and 0.9742, respectively. Based upon relative percentage error, for the best SVM–WT model, 88.70% of daily predictions fall within the acceptable range of −10% to +10%.