Agent-based Pricing Strategy for Fast-Charging Stations Considering Price Responsiveness of Electric Vehicles at Large Scales

The widespread adoption of electric vehicles (EVs) presents urgent demands for the efficient operation of fast-charging stations (FCSs), which must accommodate diverse user preferences, high mobility, and spatiotemporal traffic variability. As these systems scale, one critical challenge lies in dynamically responding to the price-sensitive behavior of EV users under complex spatiotemporal conditions and pricing strategies. This paper proposes a data-driven, agent-based pricing framework for FCSs that captures both individual decision-making dynamics and collective behavior of EV fleets. Compared to conventional studies that rely on macroscopic traffic flow models (e.g., user equilibrium formulations) or regression-based charging load forecasts, this paper introduces a fine-grained, data-driven, agent-based pricing framework that captures both the microscopic decision-making dynamics of EVs and the emergent collective behaviors across large fleets. We first formulate a graph-embedded Markov Decision Process to simulate EV drivers’ route and charging choices, incorporating transportation and charging network features. A scalable multi-agent deep reinforcement learning (MA-DRL) algorithm with count-based abstraction is then proposed to jointly learn decentralized EV policies and enable collective coordination. To bridge behavioral modeling and demand forecasting, a deep Graph Attention Network (GAT) is trained to approximate the nonlinear, high-dimensional mapping between FCS attributes. Finally, an agent-based pricing model is designed to optimize FCS profits while accounting for users’ price sensitivity and electricity market constraints, with nodal prices obtained from a lower-level power distribution network optimization. Simulation results validate the framework’s effectiveness in modeling price-responsive EV charging behaviors and achieving economically efficient station operation under dynamic conditions.