Investigating the performance of support vector machine and artificial neural networks in predicting solar radiation on a tilted surface: Saudi Arabia case study Makbul A.M. Ramli a,c,⇑ , Ssennoga Twaha b , Yusuf A. Al-Turki a,c a Department of Electrical and Computer Engineering, King Abdulaziz University, Jeddah 21589, Saudi Arabia b Energy and Sustainability Division, Faculty of Engineering, University of Nottingham, NG7 2RD, United Kingdom c Renewable Energy Research Group, King Abdulaziz University, Jeddah 21589, Saudi Arabia article info Article history: Received 28 May 2015 Accepted 30 July 2015 Keywords: Solar radiation Prediction Super vector machine Artificial neural networks abstract In this paper, investigation of the performance of a support vector machine (SVM) and artificial neural networks (ANN) in predicting solar radiation on PV panel surfaces with particular tilt angles was carried out on two sites in Saudi Arabia. The diffuse, direct, and global solar radiation data on a horizontal surface were used as the basis for predicting the radiation on a tilted surface. The amount of data used is equiv- alent to 360 days, averaged from the 5-min basis data. By solving the tilt angle equation, an optimum value of solar radiation was obtained using a tilt angle of 16° and 37.5° for Jeddah and Qassim locations, respectively. The evaluation of performance and comparison of results of ANN as well as SVM and the measured/calculated data are made on the basis of statistical measures including the root mean square error (RMSE), coefficient of correlation (CC), and magnitude of relative error (MRE). The speed of compu- tation of the algorithms is also considered for comparison. Results indicate that for Jeddah, the CC for SVM is between 0.918 and 0.967 for training and in the range of 0.91981–0.97641 for testing while that of ANN is in the range of 0.517–0.9692 for training and 0.0361–0.0961 for testing. For Qassim, the results are even better with CC of 0.999 for training and 0.987 for testing ANN showed higher values of MRE ranging between 0.19 and 1.16 and SVM is between 0.33 and 0.51 for training and testing respectively. In terms of speed of computation, it is observed that SVM is faster than ANN in predicting solar radiation data with a lower speed of 2.15 s compared to 4.56 s for ANN during training. Moreover, SVM has lower values of RMSE indicating that it is robust and has the capability to minimize errors during computations. Therefore, SVM has significantly higher accuracy, robust during computation and is faster in predicting the radiation on the tilted surfaces in comparison with ANN. Ó 2015 Elsevier Ltd. All rights reserved. 1. Introduction Solar energy is a major source of renewable energy in which solar cells and other technologies are used to convert solar energy to electrical energy. Renewable energy, solar energy in particular, is becoming increasingly popular due to several factors including increasing energy demand and continued depletion of conven- tional energy sources [1]. Fossil fuels are almost exhausted, which is having a devastating effect on the environment, a factor that has placed RE in the spotlight around the world [2]. In solar energy applications like photovoltaic (PV), solar thermal as well as passive solar designs, solar radiation data are vital [3]. A multitude of terms are used to describe the solar radiation received from the sun by the converting device, including diffuse, direct, and global solar radiation. Diffuse and global radiation intensities are computed on horizontal planes, while fixed solar systems such as PV panels and flat plate solar collector are oriented to track the sun purposely, in order to extract the greatest quantity of radiation incident on the solar panels [4]. Therefore, it is always practical to ensure that a proper tilt angle is determined for a given station where the solar systems are to be mounted. Researchers and engineers have devised numerous tools and methodologies to aid in determining the optimal tilt angle of PV panel. Besides estimating the radiation of the tilted surface, some methods mea- sure radiation on a horizontal surface [5,6]. This is due to the fact that such information regarding horizontal radiation is always http://dx.doi.org/10.1016/j.enconman.2015.07.083 0196-8904/Ó 2015 Elsevier Ltd. All rights reserved. ⇑ Corresponding author at: Department of Electrical and Computer Engineering, King Abdulaziz University, Jeddah 21589, Saudi Arabia. Tel.: +966 126952000; fax: +966 126952686. E-mail address: mramli@kau.edu.sa (M.A.M. Ramli). Energy Conversion and Management 105 (2015) 442–452 Contents lists available at ScienceDirect Energy Conversion and Management journal homepage: www.elsevier.com/locate/enconman