A probabilistic forecasting framework for neighbourhood-level disaggregation of electric vehicle adoption scenarios

Abstract

Abstract The rapid growth of electric vehicle (EV) adoption presents significant challenges for electricity networks, particularly at the low-voltage level, where clustered neighbourhood demand risks overloading infrastructure. Existing scenario-based planning approaches typically assume uniform EV uptake across neighbourhoods within a region, failing to capture the heterogeneity in historical EV registration data. They provide limited uncertainty quantification, despite the difficulty of predicting future adoption at fine spatial scales. This paper introduces a Gaussian process (GP)-based forecasting framework that combines granular historical EV registration data with top-down regional scenarios to generate probabilistic neighbourhood-level forecasts. The GP captures how local adoption deviates from regional trends, encoded in the GP’s mean function, ensuring consistency with broader scenarios while accounting for local variation and uncertainty. We validate the framework using ten representative local authority districts in England and Wales, covering 1,294 neighbourhoods. The framework demonstrates improved performance compared to baseline methods (scaled scenario, logistic growth, linear extrapolation) in normalised mean absolute error, with statistically significant improvements at horizons of three years and beyond. It also delivers well-calibrated prediction intervals, providing reliable uncertainty estimates. This framework offers a practical tool for network operators, policymakers, and planners to support targeted decision-making and investment.