Resolving the V2G pricing trilemma: A multi-objective reinforcement learning framework for differentiated tariffs based on user and station heterogeneity

Conventional time-of-use (TOU) tariffs for Vehicle-to-Grid (V2G) systems rely on a flawed “one-size-fits-all” homogeneity assumption. This oversimplification ignores the diverse behaviors of users and the functional differences of charging stations, creating a trilemma of conflicting interests among grid stability, user welfare, and operator revenue. This paper introduces a multi-objective deep reinforcement learning (RL) framework to resolve this challenge by autonomously designing adaptive, differentiated TOU tariffs. Our framework integrates quantified price elasticities, derived from a large-scale dataset of 386,456 charging transactions, to learn optimal pricing policies that navigate these multi-stakeholder trade-offs. The results demonstrate the framework's effectiveness and flexibility. A holistic dual-dimensional strategy, differentiating by both user and station, achieved the highest grid stability (1.1% peak-valley load reduction) while concurrently delivering a 17.9% revenue uplift and ¥499,730.91 in user welfare gains. Concurrently, a user-differentiated strategy prioritizing economic outcomes boosted operator revenue by 26.0% and user welfare by ¥931,230.75. These findings provide a validated, data-driven blueprint for granular “per-station, per-user” pricing, offering a practical pathway to balance stakeholder objectives and advance the synergistic integration of the power and transportation sectors.