Long Short-Term Memory Network PV Power Prediction Incorporating Extreme Extreme Gradient Boosting Algorithm

Abstract

The proportion of photovoltaic (PV) power generation to the total global power generation is increasing. Accurate prediction of PV power generation is crucial to the real-time dispatch of PV power generation. In response to the problems that traditional prediction modeling methods can overfit the situation and also have low prediction accuracy for complex and high-dimensional data, a long short-term memory (LSTM) PV power prediction model incorporating extreme gradient boosting (XGBoost) and attention mechanism is proposed. Firstly, the XGBoost and Pearson correlation coefficient method are used to feature select the data to remove the redundant and unimportant features; secondly, the attention mechanism is set to increase the weight of important features to enhance the model's understanding of features and feature values. The XGBoost-LSTM model is obtained for PV power prediction using trial-and-error method and cyclic selection method, and the experimental results show that the PV power prediction by this method is more efficient and accurate than the traditional LSTM and support vector machine methods.