Scholarlink

<p>Efficient energy management in battery electric vehicles (BEVs) is vital for extending driving range, maintaining battery health, and minimizing energy use. In this paper, we compared five deep reinforcement learning algorithms: Soft Actor–Critic (SAC), Proximal Policy Optimization (PPO), Twin Delayed Deep Deterministic Policy Gradient (TD3), Deep Deterministic Policy Gradient (DDPG), and Advantage Actor Critic (A2C). A detailed powertrain and lithium-ion battery model were incorporated into a custom simulation environment, where each agent was trained and tested across urban, NEDC, WLTP, and aggressive driving cycles. The algorithms were evaluated for energy efficiency, state of charge regulation, convergence speed, and control smoothness. PPO and SAC achieve the best overall performance. TD3 and DDPG offer smooth deterministic control with moderate energy gains, while A2C is computationally fast but exhibits slower convergence and higher performance variability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative Analysis of Deep Reinforcement Learning Algorithms for Energy Management in Battery Electric Vehicles

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Abdelaziz Sahbani

摘要

Efficient energy management in battery electric vehicles (BEVs) is vital for extending driving range, maintaining battery health, and minimizing energy use. In this paper, we compared five deep reinforcement learning algorithms: Soft Actor–Critic (SAC), Proximal Policy Optimization (PPO), Twin Delayed Deep Deterministic Policy Gradient (TD3), Deep Deterministic Policy Gradient (DDPG), and Advantage Actor Critic (A2C). A detailed powertrain and lithium-ion battery model were incorporated into a custom simulation environment, where each agent was trained and tested across urban, NEDC, WLTP, and aggressive driving cycles. The algorithms were evaluated for energy efficiency, state of charge regulation, convergence speed, and control smoothness. PPO and SAC achieve the best overall performance. TD3 and DDPG offer smooth deterministic control with moderate energy gains, while A2C is computationally fast but exhibits slower convergence and higher performance variability.