Photovoltaic Station Power Forecasting for Integrated Energy Systems Based on Spatiotemporal Features

With the increasing energy shortages and environmental degradation, the proportion of traditional fossil fuel-based power generation is gradually declining, being replaced by integrated energy systems that incorporate photovoltaic (PV) power generation, wind power generation, battery energy storage, and hydrogen-based power generation. However, PV power generation systems are highly influenced by environmental factors, exhibiting strong volatility and randomness, which creates challenges for grid dispatching. Conducting power forecasting for PV stations is thus essential to optimize grid operations. Yet, due to outdated communication systems in existing PV stations, integrated energy system cannot directly access operational data from all PV stations, making power forecasting difficult to implement. This paper proposes a method for predicting regional photovoltaic power generation using limited operational data from a few photovoltaic stations, based on long short-term memory (LSTM) network and graph neural networks (GNN). By deeply exploring the spatiotemporal characteristic relationships between these stations and the overall regional PV power generation, the method achieves data-driven, high-precision, and rapid power forecasting.