Constrained Latent Demand Prediction for Public EV Charging Stations

Public charging infrastructure planning is shifting from spatial coverage expansion to demand-oriented siting and sizing. However, candidate sites often lack station-specific charging records, and the observed demand of existing stations is often constrained by capacity, price, and local competition, which limits the potential of charging demand prediction for future charging stations. This study develops a regime-aware constrained latent demand framework for pre-deployment evaluation of public charging station candidate sites. The framework integrates multi-source spatial features, including road-network accessibility, urban activity, traffic exposure, ride-hailing charging demand, surrounding charging supply, and price conditions. It first identifies heterogeneous demand regimes, and then estimates total charging demand and connector-level charging intensity through the regime-enhanced learning approach. Observed charging demand is further represented through a model-based conditional demand benchmark and a corresponding realization rate associated with capacity, price, and surrounding competition. Using public charging stations and parking-lot candidate sites in Nanjing, China, the empirical results show that the proposed framework improves predictive performance and provides interpretable information on the associations of local activity, capacity configuration, price position, and competition with predicted demand and demand realization. By jointly considering conditional demand, expected realized demand, and connector-level efficiency, the framework supports more sustainable siting and sizing of public charging infrastructure across alternative locations and capacity configurations.