STMO: A spatio-temporal multimodal optimization algorithm for optimizing electric vehicle charging scheduling and station layout

Traditional approaches to EV charging scheduling and station layout optimization often rely on single-modal data and static planning, overlooking the dynamic interactions among traffic patterns, user behavior, meteorological factors, and grid constraints, leading to suboptimal resource utilization and grid instability. We propose the Spatio-Temporal Multimodal Optimization (STMO) framework, which integrates Long Short-Term Memory (LSTM) networks for spatio-temporal charging demand prediction, Non-dominated Sorting Genetic Algorithm II (NSGA-II) for multi-objective station layout optimization, and a dynamic load balancing algorithm for real-time scheduling. STMO fuses heterogeneous data sources—including traffic flow, user behavior, meteorological conditions, and grid load—via a novel meteorological-battery coupling feature weighting mechanism and a closed-loop feedback architecture to enable dynamic adaptation to varying urban conditions. Experimental evaluations, using datasets with up to 100,000 EV behaviors across diverse scenarios (e.g., varying densities and weather conditions), demonstrate that STMO reduces average user travel distance by 15.8 %, increases charging station utilization by 18.3 %, decreases grid peak-to-valley difference by 10.2 %, and shortens user waiting time by 25.4 %. Compared to baselines like STGCN-MOEA, DRL Pricing, and FedCharge, STMO achieves superior performance, with a 23.3 % lower mean squared error in demand prediction and a 74.2 % reduction in computational time.