Scholarlink

<p>This paper proposes a hydraulic press slider leveling control method based on the Lyapunov Distributional Soft Actor–Critic (LDSAC) algorithm, aiming to simultaneously address the critical issues of <i>Q</i>-value overestimation in reinforcement learning and system stability assurance. The LDSAC algorithm innovatively integrates the distributional reinforcement learning framework with Lyapunov stability theory, effectively suppressing <i>Q</i>-value overestimation through distributional <i>Q</i>-value estimation and ensuring global stability of the control system policy by introducing Lyapunov constraints. This paper first establishes a dynamic model for hydraulic press slider leveling, then designs an LDSAC-based control framework, and validates the algorithm’s effectiveness by simulations. The results demonstrate that, compared to traditional methods, the LDSAC algorithm exhibits significant advantages in leveling accuracy and <i>Q</i>-value overestimation suppression. In addition, we provide a theoretical analysis showing that the system can achieve finite-time convergence under the learned policy under specific assumptions. This study offers theoretical insights and a potential framework for the intelligent control of complex industrial systems, though further validation in real-world settings is needed.</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

A Novel Reinforcement Learning-Based Leveling Control Method for Multi-cylinder Hydraulic Press: Lyapunov Distributional Soft Actor–Critic

  • 期刊论文

  • 发表日期：

    2026年4月2日

查看全文
  • Chao Jia,
  • Tao Yu

摘要

This paper proposes a hydraulic press slider leveling control method based on the Lyapunov Distributional Soft Actor–Critic (LDSAC) algorithm, aiming to simultaneously address the critical issues of Q-value overestimation in reinforcement learning and system stability assurance. The LDSAC algorithm innovatively integrates the distributional reinforcement learning framework with Lyapunov stability theory, effectively suppressing Q-value overestimation through distributional Q-value estimation and ensuring global stability of the control system policy by introducing Lyapunov constraints. This paper first establishes a dynamic model for hydraulic press slider leveling, then designs an LDSAC-based control framework, and validates the algorithm’s effectiveness by simulations. The results demonstrate that, compared to traditional methods, the LDSAC algorithm exhibits significant advantages in leveling accuracy and Q-value overestimation suppression. In addition, we provide a theoretical analysis showing that the system can achieve finite-time convergence under the learned policy under specific assumptions. This study offers theoretical insights and a potential framework for the intelligent control of complex industrial systems, though further validation in real-world settings is needed.