PV power forecasting using two of the most effective techniques

Abstract

Accurate forecasting of photovoltaic energy production from renewable resources is crucial for economic reasons. In this report we discuss the use of both Machine learning forecasting techniques SVM and ANNs techniques.We compare between the two methods to predict the output of the PV output power, the data used consist of samples covering different weather conditions and error evaluation indexes RMSE MAE are used to determine the most efficient technique.SVM Technique is implemented by three different equations: Linear, Quadratic and Cubic equations, the performance results shows a slight differences between the first two MAE (9.2816% 9.9556%), RMSE (12.562% 12.59%) respectively while the last model outperforms its predecessors MAE (8.7952%) RMSE (11.432 %). The second technique which is implemented by MLPs and Elman shows ever better performances and efficiency than previous models with error indexes RMSE (6.79% 4.75%), MAE (0.21720%, 0.112%), the Elman RNN is more accurate than the Multi-Layer Perceptron and sows better results on good weather conditions wile bot models sow unstable performance is much less suitable conditions. The results of this report is to identify the best forecasting technique to be used in further esearch.