Comparison of prediction methods of photovoltaic power system production using a measured dataset

At present, generating energy from renewable sources is an important topic and is attracting significant attention because of its many benefits. Recent technological developments have made generating renewable energy from various sources such as the sun, wind, geothermal energy, and many other sources a commercially viable process. In this study, a photovoltaic (PV) system has been designed and installed for energy production. In addition, the PV output was measured for a period of one year. Neural mathematical models such as generalized feedforward networks (GFF), multilayer perceptron (MLP), self-organizing feature maps (SOFM) and support vector machines (SVM) are implemented for simulating and predicting the output of solar energy systems. The practical implementation of the proposed models achieves excellent results in comparison with results found by other researchers. The SOFM attains the lowest MSE value in the training phase (0.0638) compared to the MLP model (0.0645), GFF model (0.0658) and SVM model (0.0693). The accuracy percentage of the proposed models were found to be 80.28% for the GFF and MLP models, 78.55% for the SOFM model and 77.1% for the SVM model. GFF achieved a higher accuracy percentage in comparison with the MPL, SOFM and SVM models and also compared to those found in other studies. The proposed SOFM and MLP models achieved a smaller MAPE value of 5.339 and 5.718 respectively. All of the MLP, GFF, SOFM and SVM models accomplished a low value of RMSE of about 0.25. The proposed models scored excellent NMSE results, especially SVM, which achieved a value of 0.0039. (C) 2017 Elsevier Ltd. All rights reserved.