Forecast of photovoltaic generated power based on WOA-LSTM

Abstract

With the injection of a large amount of electricity from photovoltaic (PV) panels, there is an increasing impact on power girds. In order to lessen the uncertainties of photovoltaic output power, a Long-Short Term Memory (LSTM) Neural Networks, based on Whale Optimization Algorithm (WOA), is proposed to predict the photovoltaic generated power. First of all, the data is pre-processed to analyze gray relativity, which helps reduce the dimensionality of variables. Then the similar day samples are given by selected input variables and Gray Relativity Analysis. Moreover, the global optimization is improved, uncertainties reduced, for WOA is used to optimize network levels and learning rate of LSTM Neural Networks. Finally, optimized LSTM Neural Networks are used to predict PV output power compared to selected samples on similar days. The prediction results prove the model efficient.