基于LSTM-XGBoost模型的光伏电站短期功率预测

Abstract Currently, accurate prediction of photovoltaic (PV) power generation remains a significant challenge due to the inherent variability and uncertainty of solar energy, which is influenced by factors such as weather conditions, time of day, and seasonal changes. Traditional prediction models often struggle to capture the complex temporal dependencies and nonlinear relationships in the data, leading to suboptimal forecasting accuracy. To address these challenges, this work introduces a novel hybrid approach that leverages the strengths of both deep learning and ensemble learning. Long short-term memory network (LSTM) is employed to extract dynamic features in time series data, providing a robust representation of temporal information by capturing both short- and long-term dependencies. Extreme gradient boosting algorithm (XGBoost) further enhances the model’s predictive capability by utilizing its powerful nonlinear modeling and feature selection techniques to refine the predictions. The experimental results demonstrate that the present hybrid model significantly outperforms standalone LSTM or XGBoost models in terms of prediction accuracy. The validation of the model also requires real-world datasets for support. We validated the model using the top world-ranking solar company’s proprietary dataset and finally conducted an analysis with the SHapley Additive exPlanation (SHAP) algorithm. By combining the complementary strengths of these two methodologies, this work offers an efficient and interpretable solution for energy management and grid optimization, ultimately contributing to more reliable and sustainable energy systems.