Predictive Constraint-Verified EV Routing: A Machine-Learning-Guided Weighted Partial MaxSAT Framework for On-Demand Charging-Stop Selection on Long-Distance Indian Routes

Range anxiety stems less from battery size than from distrust of the dashboard range and from the difficulty of choosing among chargers that differ in speed, price and connector. We present PC-EVR, a charging-stop selection framework for long-distance EV trips that combines (i) an XGBoost range predictor, scaled to each vehicle, with (ii) a Weighted Partial MaxSAT choice of the next stop that is re-solved at every decision point with the core-guided RC2 solver. A three-level state-of-charge (SOC) policy (normal operation at 20% or more, controlled relaxation at 15–20%, emergency operation below 10% in exceptional cases) is applied to the estimated arrival SOC, and realized arrivals are reported by band. XGBoost had the lowest error of four range models on simulated data and the lowest journey time in an ablation, although that advantage comes from one long trip and a higher cost. On five interstate Indian trips against a 39,641-station national registry, with a common energy rate, PC-EVR averaged 28.7 h against 42.7 h (nearest station) and 49.4 h (cheapest station), and was faster on four and five of the five trips. Controls show that the saving comes from preferring fast chargers: baselines restricted to DC chargers close most of the gap. RC2 selected the same station as ranking the screened candidates in all 539 MaxSAT decisions, and we list the limits of a simulation-based evaluation.