Dynamic Energy Management Strategy for Vehicle-Grid Interaction Based on Deep Reinforcement Learning

Aiming at the problem that large-scale access of electric vehicles to the power grid leads to aggravated load peak-valley differences and increased charging costs for users, this paper proposes a dynamic energy management strategy for vehicle-grid interaction based on deep reinforcement learning (DRL). In terms of methods, this paper first models charging and discharging scheduling as a high-dimensional multi-factor Markov decision process, integrating electricity price fluctuations, grid load, user demand and battery dynamic characteristics. Then, this paper adopts the Deep Deterministic Policy Gradient (DDPG) algorithm and combines it with the attention mechanism to dynamically optimize charging and discharging decisions; finally, a cloud-edge collaborative architecture is designed to achieve real-time deployment of strategies. Experiments show that in the simulation scenario of electric vehicles, this strategy reduces peak load by 7.26%, suppresses load fluctuations by 43.90%, and saves users 27.6% of electricity costs. At the same time, the battery health retention rate is increased by 1.33%, which is significantly better than baseline methods such as rule-driven and model predictive control. The study verifies the effectiveness of DRL in collaboratively optimizing grid stability and user economy.