Short-Term Photovoltaic Power Generation Prediction Based on LightGBM-LSTM Model

Abstract

Predicting photovoltaic power generation is one of the emphases of energy management system. In recent years, with the increasingly serious environmental pollution, the use of artificial intelligence technology to study the world's largest clean energy solar energy is of great significance. Other than the existing work, this paper proposes a fusion model based on LightGBM and LSTM to forecast short-term photovoltaic power generation. For obtaining the best prediction results, MLP is applied for dynamic weight allocation and combination to generate the prediction results. Based on the real photovoltaic power generation data collected by a photovoltaic power station in China, our model was compared with three traditional models, including XGBoost model, LightGBM model and LSTM model. Besides, the prediction results of four models in a day with enough light and four models in a day with less light are compared. The experiment result indicates that the model proposed in this paper has higher accuracy in short-term photovoltaic power generation forecast and It has good adaptability to the nonlinear feature of illumination.