Multi-Agent Deep reinforcement learning for EV aggregator bidding in Renewable-Dominated electricity markets

• A unified multi-agent market model simulates strategic interactions under high renewable penetration. • The framework combines fleet scheduling and bidding via MADDPG with Prioritized Experience Replay. • Adaptive bidding strategies are learned in continuous action spaces. • The simulation on a 30-bus system proves improved efficiency and reliability versus baselines. The increasing penetration of renewable energy into electricity markets has introduced significant volatility and uncertainty, presenting new challenges for flexible demand-side resources such as Electric Vehicle Aggregators (EVAs). However, existing approaches to EVA bidding strategies often relied on static pricing forecasts or single-agent optimization frameworks, which failed to capture the dynamic, multi-agent nature of modern electricity markets. To address these limitations, this paper proposes an intelligent bidding strategy framework based on Multi-Agent Deep Reinforcement Learning tailored for heavy-duty EV fleets engaged in freight operations. A multi-agent game-theoretic market model is constructed, incorporating renewable generators, traditional suppliers, industrial and commercial users, and EVAs to replicate real-time interactions within spot markets. A joint optimization strategy is developed to align vehicle scheduling constraints with market participation, while the Multi-agent Deep Deterministic Policy Gradient algorithm is employed to enable adaptive, decentralized learning of optimal bidding behaviors. Simulation results demonstrate that the proposed framework effectively reduces peak loads, smooths load profiles, and enhances EVA revenues under volatile market conditions. The findings offer actionable insights into integrating flexible electric mobility into future renewable-dominated power systems.