Identifying generalizable equilibrium pricing strategies for charging service providers in coupled power and transportation networks

Transportation electrification, involving large-scale integration of electric vehicles (EV) and fast charging stations (FCS), plays a critical role for global energy transition and decarbonization. In this context, coordination of EV routing and charging activities through suitably designed price signals constitutes an imperative step in secure and economic operation of the coupled power-transportation networks (CPTN). This work examines the noncooperative pricing competition between self-interested EV charging service providers (CSP), taken into account the complex interactions between CSPs' pricing strategies, EV users' decisions and the operation of CPTN. The modeling of CPTN environment captures the prominent type of uncertainties stemming from the gasoline vehicle and EV origin-destination travel demands and their cost elasticity, EV initial state-of-charge and renewable energy sources (RES). An enhanced multi-agent proximal policy optimization method is developed to solve the pricing game, which incorporates an attention mechanism to selectively incorporate agents' representative information to mitigate the environmental non-stationarity without raising dimensionality challenge, while safeguarding the commercial confidentiality of CSP agents. To foster more efficient learning coordination in the highly uncertain CPTN environment, a sequential update scheme is also developed to achieve monotonic policy improvement for CSP agents. Case studies on an illustrative and a large-scale test system reveal that the proposed method facilitates sufficient competition among CSP agents and corroborates the core benefits in terms of reduced charging costs for EV users, enhancement of RES absorption and cost efficiency of the power distribution network. Results also validate the excellent generalization capability of the proposed method in coping with CPTN uncertainties. Finally, the rationale of the proposed attention mechanism is validated and the superior computational performance is highlighted against the state-of-the-art methods.