Spatial-temporal load prediction of electric bus charging station based on S2TAT
Guangnian Xiao, Hailin Tong, Yaqing Shu, Anning Ni
Abstract
In recent years, electric buses have advanced rapidly due to their green and low-carbon attributes. To address range anxiety and optimize charging strategies, accurately predicting charging load has become essential. This paper introduces a synchronous spatial-temporal attention transformer (S 2 TAT) model that models temporal and spatial dependencies simultaneously, utilizing operational data from new energy electric buses in Shanghai. To improve charging event prediction, we propose two key improvements: an adaptive adjacency matrix for dynamic spatial dependencies learning and a periodicity extraction mechanism for capturing cyclical patterns. These enhancements significantly boost prediction accuracy over baseline models. An ablation study further verifies the contributions of the model’s components.