Scholarlink

<p>Extending the lifespan and performance of batteries depends on cell balancing in battery management system (BMS). In contrast to passive cell balancing, which causes energy to be dissipated, active balancing increases capacity and lifetime by transferring uniform charge across cells. However, a few challenges present in active cell balancing, such as complexity, cost, and energy fatalities. Intelligent decision-making, optimized charge transfer, and improved accuracy can lead to enhancing the performance of active cell balancing. In this regard, this paper presents a comprehensive study on deep deterministic policy gradient (DDPG), proximal policy optimization (PPO), multi-agent reinforcement learning (MARL), hierarchical reinforcement learning (HRL), graph neural networks (GNNs), and generative adversarial networks (GANs) algorithms. Training, validation, and testing of deep learning-based active cell balancing (DLACB) can be performed using MATLAB/Simulink platform. After complete evaluation, it is observed that DDPG and PPO provide better accuracy (95% and 90%) and predict maximum accuracy error (MAE) in quick instances for DDPG and PPO at 146th and 165th as compared to MARL, HRL, GNN, and GAN (170th, 175th, 180th, and 185th), respectively.</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

An advanced optimal active cell equalization algorithm for electric vehicle battery system using deep learning evaluation: a comparative survey

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Sairaj Arandhakar,
  • Jayaram Nakka

摘要

Extending the lifespan and performance of batteries depends on cell balancing in battery management system (BMS). In contrast to passive cell balancing, which causes energy to be dissipated, active balancing increases capacity and lifetime by transferring uniform charge across cells. However, a few challenges present in active cell balancing, such as complexity, cost, and energy fatalities. Intelligent decision-making, optimized charge transfer, and improved accuracy can lead to enhancing the performance of active cell balancing. In this regard, this paper presents a comprehensive study on deep deterministic policy gradient (DDPG), proximal policy optimization (PPO), multi-agent reinforcement learning (MARL), hierarchical reinforcement learning (HRL), graph neural networks (GNNs), and generative adversarial networks (GANs) algorithms. Training, validation, and testing of deep learning-based active cell balancing (DLACB) can be performed using MATLAB/Simulink platform. After complete evaluation, it is observed that DDPG and PPO provide better accuracy (95% and 90%) and predict maximum accuracy error (MAE) in quick instances for DDPG and PPO at 146th and 165th as compared to MARL, HRL, GNN, and GAN (170th, 175th, 180th, and 185th), respectively.