A protocol-aware learning framework for harmonizing and translating official electric vehicle range labels across regulatory standards

Abstract Official electric vehicle (EV) driving-range labels are generated under distinct regulatory protocols, including the U.S. Environmental Protection Agency (EPA), Worldwide Harmonised Light Vehicle Test Procedure (WLTP), and New European Driving Cycle (NEDC), and therefore are not directly interchangeable. This study develops a protocol-aware learning framework for harmonizing and translating official range labels from static vehicle specifications while explicitly modeling the reporting protocol. The compiled database contains 319 EV configurations, with 176 EPA, 88 WLTP, and 98 NEDC labels; 227 configurations have at least one of these regulatory labels, and paired coverage varies substantially by protocol. The framework evaluates single-standard estimation, pooled learning with and without protocol information, a two-stage latent-baseline-plus-correction formulation, protocol-gap learning, cross-standard translation, interpretability, robustness, and uncertainty. In held-out pooled testing, explicit protocol encoding reduced RMSE from 54.21 to 43.20 km (R² = 0.925). Across 30 vehicle-grouped repeated holdout splits, mean RMSE decreased from 46.97 km without protocol information to 32.14 km with protocol information, with zero vehicle overlap between training and testing. Translation models were benchmarked against identity, median-ratio, and linear baselines and further assessed using bootstrap and nested validation; the low-overlap WLTP–NEDC tasks are therefore treated as exploratory rather than validated conversion rules. Grouped split-conformal intervals achieved 93.4% empirical coverage for a nominal 90% target. The framework is intended for cross-market comparison and regulatory-label harmonization, not prediction of actual on-road EV range, which requires independent real-world operating and energy-consumption data.