Adaptive Graph Neural Network-Based Hybrid Approach for Long-Term Photovoltaic Power Forecasting

Photovoltaic power generation prediction is crucial for the effective integration of renewable energy into the grid, real-time grid balancing, and the optimization of energy storage systems. However, PV power generation is highly dependent on environmental factors such as weather conditions. Photovoltaic power generation prediction is crucial for the effective integration of renewable energy into the grid, real-time grid balancing, and the optimization of energy storage systems. However, PV power generation is highly dependent on environmental factors such as weather conditions. Effectively integrating environmental information remains a major challenge for photovoltaic power forecasting. This study proposes a hybrid deep learning model that incorporates an adaptive neural network to capture the latent relationships between PV power generation and environmental variables, thereby enhancing forecasting accuracy. The adaptive graph neural network employs a data-driven directed graph structure, where TCN and variable interaction layers are alternately stacked to better model the spatiotemporal coupling among variables for long-term PV output forecasting. The proposed model was evaluated on three sites located in different regions, with a fixed input length of 96 and output horizons ranging from 96 to 768 steps. Compared with state-of-the-art baselines, the model achieved average improvements of 2.19% and 1.57% in MSE and MAE at a 384-step horizon, and 2.81% and 2.47% at a 768-step horizon, respectively, demonstrating superior performance in long-term PV output forecasting tasks.