Conformal Digital-Twin Planning for Pre-Dispatch Electric Vehicle Fleet Routing Under Communication Uncertainty

Abstract

Electric-vehicle fleets plan from digital-twin traffic, energy, charging, and station-availability state before dispatch. Communication loss, backend delay, and sensor noise make that state stale or incomplete. We study whether a conservative planning state, rather than update frequency alone, improves electric vehicle routing with time windows (EVRPTW). CP-World maps information age and communication regime to row-level conformal upper bounds. Because timestamped calibration rows repeat within synthetic trajectories, we claim empirical coverage only: held-out continuous and joint inclusion are both 92.0%, with trajectory-cluster 95% intervals 89.1–94.7% and 89.1–94.8%. The primary experiment pairs Raw-6, deterministic Heuristic-6, Pooled-CP-6, and Mondrian-CP-6 on the same six epochs, physical worlds, packet/noise draws, and 6-s end-to-end solver-call deadline. Across 432 cells from 24 unseen 100-customer instances and three seeds, strict release feasibility is 1.2%, 36.1%, 23.6%, and 20.4%, respectively. A six-request feasibility/risk gate is evaluated secondarily. Direct final-world inclusion, common-feasible routing cost, deadline compliance, expanded search, and exact five-customer solutions separate predictive, operational, and solver effects. Together, these endpoints distinguish prediction quality from routing quality and communication-policy behavior under matched conditions. The contribution is a dependence-explicit audit and planning layer, not a new routing metaheuristic. Results remain synthetic pre-dispatch evidence and are not a deployment or safety certificate for fleet routing under communication uncertainty.