Short-term power load forecasting based on spatial-temporal dynamic graph and multi-scale Transformer

Abstract

Listen

Translate article

Abstract Short-term power load forecasting is crucial for power system stability and market planning, yet the multi-periodic nature of load data and its complex correlations with external factors pose significant challenges for accurate prediction. To address these, we propose the spatial-temporal dynamic graph Transformer (SDGT). SDGT incorporates a multi-scale Transformer module to capture multiple periodic patterns, with a patch-based multi-scale encoder layer for extracting temporal features. Simultaneously, It leverages a spatial-temporal correlation graph (STCG) integrating shape similarity and semantic relevance, coupled with a graph convolution module, to model dynamic spatial correlations between load data and external factors. Experiments on two public datasets validate that SDGT surpasses state-of-the-art models, demonstrating its effectiveness and robustness in improving power load forecasting accuracy.