A Scalable Charging Algorithm for Heterogeneous EV Fleets Based on Clustering and Learning Methods

As the number of electric vehicles (EVs) surges, the optimal real-time management of their charging processes poses a significant challenge. While centralized control methods have garnered attention for the potential to achieve global optimality, they are primarily suitable for small or medium-sized EV fleets due to their high computational costs. To harness the benefits of centralized control strategies more effectively, we introduce a novel clustering model, which is trained by learning substantial optimal scheduling datasets backward to categorize heterogenous EVs with similar charging power but distinct states in the dynamic charging processes. Subsequently, a scalable optimal charging strategy is developed by efficiently managing distinct groups of vehicles identified by the clustering model, rather than individual EVs. This approach significantly reduces the control dimension without substantially impairing performance. Simulation results validate that when compared to the centralized charging strategy, the proposed charging strategy achieves similar performance, with only a 1.33% increase in total charging costs in the case where 100 random EVs participate in energy and ancillary electricity markets, and notably, compared to benchmark method, the computational cost of the proposed method is reduced by 99.8%.