Terrain- and Battery-Telemetry-Constrained Energy-Saving Route Planning for Electric Buses Using an Improved MegaCRN Model

Electric buses are strongly affected by route topography and battery operating state, yet many energy-saving routing methods still treat segment cost as a static distance- or speed-based attribute. This paper emphasizes an operational extension of MegaCRN in which memory retrieval is conditioned by terrain and live battery telemetry, and the resulting prediction is passed to a feasibility-aware routing decoder. This paper proposes an improved MegaCRN model for energy- saving route planning under terrain and battery telemetry constraints. The method constructs a directed road-segment graph in which gradient, elevation change, curvature, stop density, observed speed, passenger load, state of charge, battery temperature, voltage dispersion, and regenerative braking opportunity are fused through a telemetry-conditioned memory module. A constraint-aware decoder then estimates segment energy and transfers the prediction into a multi-objective path search with timetable, reserve-energy, and thermal-feasibility constraints. Experiments are designed on a 214-segment urban bus network with 38,720 synchronized telemetry records and 1,246 candidate trips. Compared with vanilla MegaCRN, DCRNN, STGCN, ASTGCN, and an energy shortest-path baseline, the proposed model reduces mean absolute energy prediction error from 0.182 to 0.121 kWh/km, lowers route-level energy consumption by 9.8%, and improves reserve-energy violation rate from 7.4% to 1.9%. Ablation results show that terrain memory and battery gating contribute 3.1% and 4.4% energy reduction, respectively, while the complete model keeps average delay within 2.6 min. The results indicate that memory-based graph forecasting can support deployable electric-bus routing when physical terrain features and battery telemetry are treated as coupled constraints rather than auxiliary covariates.