Joint Scheduling of EV Routing and Battery Charging in BSSs via Hypergraph-Enhanced MADRL with Kolmogorov-Arnold Networks

This work studies the joint scheduling problem of electric vehicle (EV) routing and the battery charging scheduling for multiple battery swapping stations (BSSs) in a transportation network, considering stochastic EV swapping demands, time-varying electricity prices, and renewable generation. The problem is formulated as a decentralized partially observable Markov decision process (Dec-POMDP) to minimize total system costs comprising battery charging costs and EV drivers' travel/waiting costs. However, solving the formulated Dec-POMDP involves two critical challenges: (1) complex spatiotemporal dependency of heterogeneous agents under dynamic system, and (2) the curse of dimensionality induced by high-dimensional joint state-action spaces. To overcome these difficulties, we integrate hypergraph attention (HGAT) mechanisms for modeling high-order spatio-temporal interdependencies and interactions among heterogeneous agents, and leverage Kolmogorov-Arnold Networks (KANs) to mitigate overfitting and training instability caused by the curse of dimensionality. This results in the proposal of a heterogeneous multi-agent hypergraph attention KAN-based actor-critic (HMA-HGA-KAC) framework. Numerical experiments using real-world data and a 44-node transportation network with different numbers of EVs demonstrate that the proposed approach achieves significant system cost reductions compared to state-of-the-art methods.