A Two-Stage Surrogate Framework for Optimizing EV–PV Integrated Distribution Grids

Abstract:
The increasing penetration of Electric Vehicles (EVs) and Photovoltaic (PV) systems presents significant planning challenges for modern power grids. This paper proposes a two-stage surrogate-assisted optimization framework to address this complexity. A detailed distribution grid was modeled in DIgSILENT PowerFactory to generate a comprehensive dataset under a Bangladeshi tariff structure. In the first stage, a Gradient Boosting Regressor (GBR) was identified as a superior surrogate model—outperforming Random Forest and Neural Network approaches—achieving an R² score exceeding 0.99 for predicting both EV energy cost and PV utilization. In the second stage, this validated GBR surrogate was integrated with the NSGAII algorithm to co-optimize infrastructure investment (PV and BESS capacity) and EV charging policy. A key finding from the optimization is that for the studied system, a policy of prioritizing daytime, solar-aligned EV charging emerged as the dominant strategy across the entire Pareto optimal front. The results provide a quantitative decision-support tool for designing cost-effective and sustainable energy systems, highlighting the critical interplay between physical assets and operational policy.