Research on Photovoltaic Power Generation Energy Forecasting Model Driven by Deep Learning

To improve the accuracy of photovoltaic power generation forecasting, this paper constructs two deep learning forecasting models: LSTM and CNN-LSTM. Historical power generation and meteorological data from photovoltaic power plants are collected and preprocessed for model training and optimization. Comparative experiments are conducted with models such as ARIMA, CNN, and GRU. Results show that the CNN-LSTM hybrid model achieves the best performance, with a MAE of 2.35 kW·h, an RMSE of 3.26 kW·h, a MAPE as low as 3.65%, and an R² of 0.958, significantly outperforming both ARIMA (MAE 6.85 kW·h, MAPE 10.56%) and a single LSTM (MAE 3.56 kW·h, MAPE 5.92%). Simulation curves demonstrate that the model accurately captures power generation fluctuations, particularly during peak and valley periods. Research has shown that a hybrid model that integrates the spatial characteristics of meteorological data with the temporal characteristics of power generation can effectively improve forecasting accuracy and provide a reliable basis for power system scheduling.