A Comprehensive Framework for Adaptive Electric Vehicle Charging Station Siting and Sizing Based on Transferable Spatio-Temporal Demand Prediction

Under the global trend of promoting green development, the electric vehicle (EV) industry has entered a period of rapid growth. However, the disorderly access of large-scale electric vehicles into the power grid threatens stability and security, and simultaneously rational planning of charging infrastructure is increasingly needed to meet the region-specific and time-varying demands. Thus, this work introduces a systematic framework combining spatiotemporal demand calculation and transferable prediction with Electric Vehicle charging station (EVCS) siting and sizing optimization. By leveraging partial EV trip data, the method can capture users’ driving patterns and supplement the citywide complete EV trip data, which is then fed into the established mathematical model to compute spatiotemporal charging demand. The derived best-performing ConvLSTM model, trained on basic features and calculated charging demand, can be applied to cross-city demand prediction by inputting identical basic features. This methodology is particularly tailored for prediction with limited or no corresponding data, mitigating data scarcity in underdeveloped small cities. Building upon this, a bi-level planning model is designed for EVCS siting and sizing, considering both user satisfaction and investment cost. The optimal coordinated EVCS construction strategy is obtained using the NSGA-II algorithm. Finally, this framework was employed in Shenzhen and Shanwei for empirical analysis, confirming its feasibility and accuracy.