Power prediction of photovoltaic power generation based on LSTM model with additive Attention mechanism

Photovoltaic power generation is affected by meteorological factors, characterized by intermittency, readiness and fluctuation, which brings challenges to the stable operation of PV power generation systems. Therefore, accurate prediction of PV power generation is of great significance. In this study, the LSTM (Long Short-Term Memory Network) model based on additive attention mechanism aims to improve the prediction accuracy of PV power. Firstly, the Pearson correlation coefficient method is used to pre-screen the meteorological data and select the meteorological parameters with higher correlation with PV power. Then Principal Component Analysis was applied to reduce the data dimensionality, and the most relevant parameters affecting PV power were successfully mined as inputs to the prediction model. Next, the additive Attention mechanism is introduced into the LSTM model, which is conducive to reducing the influence of irrelevant variables and paying more attention to useful input parameters. It is verified experimentally that the Attention-LSTM-based prediction model has higher accuracy in different weather scenarios when compared with single LSTM model, BP (BackPropagation Neural Network) model, and Convolutional Neural Network (CNN) model. Therefore, the model proposed in this study can effectively improve the prediction of PV power.