Prediction of photovoltaic power output using artificial neural networks

Accuracy in the prediction of the performances of photovoltaic plants is indispensable in power-generating industries. This has made manufacturers of photovoltaic cells place a high premium on the precision of the forecast of the power output. Artificial neural networks have been proven to be highly effective for forecasting outputs in many technologies. In this study, a feedforward backpropagation neural network model and a radial basis network model were introduced to predict or forecast the power generated by a photovoltaic plant for industrial applications. The inputs and outputs of the models for the training were selected based on the objective of the study, correlation analysis, and analysis of variance test. The results of the simulations of the proposed feedforward backpropagation artificial neural model indicated a mean absolute error of 0.0446, a mean square error of 0.0099, and a mean square error of 0.095. The results of the simulation of the developed radial basis network model indicated a mean absolute error of 0.114, a mean square error of 0.0375, and a root mean square error of 0.196. The comparative analysis of the study shows that the accuracy of the feedforward backpropagation neural network model is 3.79 times that of the radial basis function network, in terms of mean square error. The accuracy and correlation of the proposed feedforward backpropagation neural network were 98.27% and 99.97%, respectively. The proposed feedforward backpropagation neural network model is suitable for industrial applications.