Photovoltaic Power Generation Forecasting Based on Optimized-CNN-LSTM

As the global energy transition accelerates, photovoltaic (PV) power generation plays a crucial role in renewable energy systems. However, accurate PV power forecasting remains a challenge due to its strong dependence on weather conditions. This study addresses the issue of fluctuating prediction accuracy under varying weather conditions by identifying five key meteorological features: solar irradiance, air pressure, radiation, air temperature, and relative humidity, while excluding less relevant factors such as wind speed. A classification-based approach is adopted to distinguish weather conditions, and multiple forecasting models, including Convolutional Neural Network(CNN), Long Short-Term Memory(LSTM), Transformer, CNN-Transformer, CNN-LSTM, and the proposed Optimized-CNN-LSTM, are evaluated. Experimental results demonstrate that the Optimized-CNN-LSTM model effectively captures temporal and nonlinear dependencies, achieving superior prediction accuracy across all weather conditions compared to other models.