Photovoltaic power prediction based on CNN-LSTM-XGBoost

Abstract

Accurate household photovoltaic power forecasting is important for energy management and renewable energy utilization. Because PV output is affected by irradiance, temperature, cloud cover, and other factors, it shows significant fluctuations and temporal dependence. This study uses preprocessed household PV data and meteorological variables to build three forecasting models: CNN, CNN-LSTM, and CNN-LSTM-XGBoost. Results under sunny and cloudy conditions show that CNN-LSTM-XGBoost achieves the best prediction accuracy and robustness.