Spatiotemporal digital twin for city-scale EV charging infrastructure using LSTM-GNN fusion and resilience-driven optimization

Abstract

The rapid growth of electric vehicles (EVs) requires intelligent and resilient planning of charging infrastructure under dynamic urban conditions. This paper proposes a city-scale spatiotemporal digital twin (SDT) that integrates LSTM-GNN fusion with resilience-driven hybrid optimization (GA-PSO-deep reinforcement learning) for adaptive EV infrastructure management. The LSTM model captures temporal variations in charging demand, while the graph neural network (GNN) learns spatial dependencies across charging stations, mobility networks, and grid components. Unlike existing approaches, the proposed framework incorporates power electronics-aware modeling, including charger power conversion system (PCS) efficiency, switching losses, and harmonic distortion constraints, ensuring realistic grid interaction. The digital twin also considers energy system metrics such as transformer loading, voltage deviation, and renewable energy variability, along with EV drive-cycle characteristics like fast charging and battery limits. Simulation results show that the proposed model improves demand prediction accuracy by 14-22%, reduces grid overload probability by 35%, lowers operational cost by 18%, and achieves improved power quality performance compared to conventional methods. The system maintains a high resilience index (>0.92) under stress scenarios. Overall, this work presents a holistic AI-driven digital twin framework that enhances grid stability, supports sustainable EV integration, and enables scalable deployment of future smart charging infrastructure.