EleState: Multi-Dimensional Service-State Forecasting for Electric Vehicle Charging Stations

With the rapid electrification of transportation, electric vehicles (EVs) have become an increasingly important component of sustainable mobility, making the efficient operation and management of charging infrastructure increasingly critical. Accurately forecasting future charging service states is therefore essential for charging resource scheduling, energy management, and intelligent EV charging operations. However, existing EV charging forecasting studies mainly focus on predicting isolated indicators. Such formulations overlook the inherently multi-dimensional nature of charging services, where different indicators describe complementary aspects of future operational conditions. As a result, single-indicator forecasting provides only a partial characterization of charging states and limits its effectiveness for downstream tasks such as risk awareness and intelligent charging management. To address this limitation, this paper reformulates EV charging forecasting as a service-state forecasting problem and proposes EleState, a framework designed to learn and forecast multi-dimensional charging service states. EleState constructs comprehensive service-state representations by jointly modeling temporal evolution, inter-state dependency, and spatial interaction among charging stations. Based on the learned representations, EleState forecasts three key service-state dimensions, including occupancy, charging duration, and charging volume, enabling a more complete characterization of future charging conditions. Extensive experiments on the UrbanEV dataset demonstrate that EleState achieves the best target-wise MAE and RMSE across all three service-state dimensions, reducing target-wise MAE by 1.4–2.4% and RMSE by 1.2–3.2% compared with the strongest baseline. Furthermore, EleState provides stronger operational risk awareness than conventional load-oriented forecasting and maintains robust performance across different forecasting horizons. Overall, this work extends EV charging forecasting beyond conventional single-indicator prediction toward comprehensive service-state forecasting, providing a more effective foundation for intelligent EV charging management.