Towards Sustainable Electric Bus Fleet Electrification: A Rolling Stock Digital Twin for Pre-Investment Charging Infrastructure Planning in a Resource-Constrained Environment

Sizing charging infrastructure for battery electric bus (BEB) fleets requires an accurate estimate of per-trip energy demand before capital is committed. Accurate sizing is itself a sustainability decision in resource-constrained settings: undersized infrastructure delays the shift away from diesel fleets, while oversized infrastructure diverts capital that could otherwise fund a wider electrification programme. Yet most pre-investment planning still relies on a single fleet-average consumption rate that cannot capture how route, operating conditions, and climate interact. This paper presents a rolling stock digital twin for the Golden Arrow Bus Services (GABS) electric fleet at the Arrowgate depot in Cape Town, South Africa, one of the first large BEB fleets in sub-Saharan Africa. A LightGBM gradient-boosted model is trained on 596 trips drawn from two five-day measurement periods in different seasons (winter and early spring) to predict the per-trip consumption rate (kWh/km) from route-geometry and operating-condition features derived from GPS traces, duty schedules, and public weather data, requiring no vehicle specification beyond a single usable-battery-capacity figure. Under two grouped cross-validation schemes representing known and genuinely unseen routes, the model predicts the per-trip state-of-charge drop, and hence arrival state of charge given a known departure SoC, to within 2.95 and 3.46 percentage points, respectively. SHAP analysis shows that departure time window and ambient temperature, rather than route geometry, dominate the prediction, and the model removes a structured time-of-day bias that the flat rate cannot represent: a roughly 4% underestimate of the morning peak and a 6–7% overestimate of midday and afternoon demand. Independent OCPP charging records corroborate the reconstructed depot demand profile. By replacing an assumed consumption figure with a validated, empirically grounded one, this work supports capital-efficient, lower-risk electrification pathways for transit operators in the Global South, where infrastructure budgets are especially constrained. The twin supplies an empirically grounded, transferable demand layer for depot-level pre-investment planning in resource-constrained operating contexts.