Short-Term Load Forecasting of Integrated Energy System Based on Spatio-temporal Graph Neural Network

Accurate short-term load forecasting is crucial for the efficient planning, dispatching, and economic management of integrated energy systems. Despite its significance, current methods face two key limitations: (1) insufficient exploration of intricate relationships between different loads and their interaction with meteorological factors, and (2) inadequate extraction of time-dependent features of different loads at finer temporal granularity. To alleviate these issues, this paper introduces a novel short-term load forecasting method based on a multi-granularity adaptive spatio-temporal graph neural network. Firstly, a multi-granularity temporal learning module is introduced to enhance temporal feature extraction across various loads, allowing for finer temporal resolution. Additionally, an adaptive spatio-temporal graph neural network module is constructed to capture the complex coupling relationships of diverse loads and meteorological factors. Finally, numerous experiments are carried out using integrated energy system datasets that are openly available. The findings reveal the proposed method exhibits superior prediction precision over current approaches.