Scholarlink

<p>This study constructs a personalized sports training decision-making system using multi-agent reinforcement learning. Integrating IoT sensors for real-time data and serious games for simulation, the system utilizes a centralized training and distributed execution framework combining fuzzy logic and deep residual networks. Physiological limit thresholds are embedded as safety constraints to prevent overtraining. Based on a 12-week dataset from 120 professional athletes and results averaged over five independent runs, empirical results demonstrate that compared to traditional experience-based methods, training efficiency improved by 31.2%, injury risk decreased by 25.6%, and personalized adaptability reached 8.9 points. The MADDPG algorithm achieved a superior average reward of 4,750, converging within 150,000 rounds while handling complex state spaces effectively. This research establishes a quantifiable decision-support paradigm, shifting sports training from discrete fluctuations toward centralized stability.</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

Research on multi-agent reinforcement learning optimization of personalized decision-making system for sports training

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月5日

查看全文
  • Cuicui Guo,
  • Chengcheng Xu

摘要

This study constructs a personalized sports training decision-making system using multi-agent reinforcement learning. Integrating IoT sensors for real-time data and serious games for simulation, the system utilizes a centralized training and distributed execution framework combining fuzzy logic and deep residual networks. Physiological limit thresholds are embedded as safety constraints to prevent overtraining. Based on a 12-week dataset from 120 professional athletes and results averaged over five independent runs, empirical results demonstrate that compared to traditional experience-based methods, training efficiency improved by 31.2%, injury risk decreased by 25.6%, and personalized adaptability reached 8.9 points. The MADDPG algorithm achieved a superior average reward of 4,750, converging within 150,000 rounds while handling complex state spaces effectively. This research establishes a quantifiable decision-support paradigm, shifting sports training from discrete fluctuations toward centralized stability.