Scholarlink

<p>Reinforcement learning has been explored to solve challenges of DC microgrids when the model is unknown, addressing issues such as voltage, current, and power instabilities. It learns optimal control policies directly from system interactions without explicit models. This paper provides comprehensive reviews of reinforcement learning approaches for dynamic stabilization in DC microgrids, focusing on the strategic integration of various RL methodologies. The study analyzes value-based methods (Q-learning, SARSA, DQN) and policy-based approaches (DDPG, A2C, PPO, SAC) and evaluates their effectiveness in maintaining system stability across different microgrid configurations. The research examines multiple control frameworks, including online/offline learning, multi-agent systems, and various exploration policies (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\epsilon \)</EquationSource> <EquationSource Format="MATHML"><math> <mi>ϵ</mi> </math></EquationSource> </InlineEquation>-greedy, softmax, and UCB) for optimizing voltage regulation. This paper covers model-free approaches, particularly to actor-critic architectures for policy gradient algorithms. The investigation shows different learning approaches and control policies within the DC microgrid structure. This review establishes a comprehensive framework for RL-based control strategies in DC microgrid applications and discusses future possible research directions.</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

Reinforcement learning strategies for dynamic stabilization in DC microgrid: a comprehensive survey

  • 期刊论文

  • 发表日期：

    2026年1月9日

查看全文
  • Rinki Maurya,
  • Rahul Meshram,
  • Surya Prakash,
  • Ashutosh Kumar Singh

摘要

Reinforcement learning has been explored to solve challenges of DC microgrids when the model is unknown, addressing issues such as voltage, current, and power instabilities. It learns optimal control policies directly from system interactions without explicit models. This paper provides comprehensive reviews of reinforcement learning approaches for dynamic stabilization in DC microgrids, focusing on the strategic integration of various RL methodologies. The study analyzes value-based methods (Q-learning, SARSA, DQN) and policy-based approaches (DDPG, A2C, PPO, SAC) and evaluates their effectiveness in maintaining system stability across different microgrid configurations. The research examines multiple control frameworks, including online/offline learning, multi-agent systems, and various exploration policies ( \(\epsilon \) ϵ -greedy, softmax, and UCB) for optimizing voltage regulation. This paper covers model-free approaches, particularly to actor-critic architectures for policy gradient algorithms. The investigation shows different learning approaches and control policies within the DC microgrid structure. This review establishes a comprehensive framework for RL-based control strategies in DC microgrid applications and discusses future possible research directions.