Scholarlink

<p>This study presents a data-driven controller (DDC) for stabilizing cardiac rhythm dynamics. A total disturbance observer (TDO) is integrated to estimate unmodeled dynamics and external disturbances, thereby enhancing robustness to uncertainty. To eliminate manual gain tuning, an actor–critic smart tuning agent is trained offline using randomized cardiac-model simulations. During closed-loop operation, the trained actor weights remain fixed while the policy generates bounded, state-dependent controller gains online. In addition, a control-effort penalty is embedded within the reward function to discourage excessive actuation, thereby reducing control effort and discouraging unnecessarily abrupt actuation in the simulations. The proposed strategy is evaluated under three robustness dimensions: (i) synthetic band-limited disturbances spanning 0.8–30&#xa0;Hz and grouped into Delta-, Theta-, Alpha-, and Beta-labeled frequency intervals, (ii) additive stochastic noise at multiple power levels, and (iii) reference-heartbeat variation and actuator-saturation constraints. Simulation results demonstrate accurate reference tracking and improved robustness compared with non-optimal iPID, ultra-local model (ULM), and conventional sliding-mode control (SMC). Across all tested conditions, the STA-tuned DDC strategy consistently achieves lower RMSE and MAE values and improved performance index while maintaining minimal tracking delay relative to the reference heartbeat.</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

Artificial intelligence-based adaptive gain tuning for robust data-driven cardiac rhythm control under uncertainty

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月5日

查看全文
  • V. T. Mai,
  • Hoang Nguyen-Huy,
  • Hoang Quoc Dong,
  • Thanh-Nam Tran,
  • Ardashir Mohammadzadeh,
  • Osman Taylan,
  • Abdulaziz S. Alkabaa

摘要

This study presents a data-driven controller (DDC) for stabilizing cardiac rhythm dynamics. A total disturbance observer (TDO) is integrated to estimate unmodeled dynamics and external disturbances, thereby enhancing robustness to uncertainty. To eliminate manual gain tuning, an actor–critic smart tuning agent is trained offline using randomized cardiac-model simulations. During closed-loop operation, the trained actor weights remain fixed while the policy generates bounded, state-dependent controller gains online. In addition, a control-effort penalty is embedded within the reward function to discourage excessive actuation, thereby reducing control effort and discouraging unnecessarily abrupt actuation in the simulations. The proposed strategy is evaluated under three robustness dimensions: (i) synthetic band-limited disturbances spanning 0.8–30 Hz and grouped into Delta-, Theta-, Alpha-, and Beta-labeled frequency intervals, (ii) additive stochastic noise at multiple power levels, and (iii) reference-heartbeat variation and actuator-saturation constraints. Simulation results demonstrate accurate reference tracking and improved robustness compared with non-optimal iPID, ultra-local model (ULM), and conventional sliding-mode control (SMC). Across all tested conditions, the STA-tuned DDC strategy consistently achieves lower RMSE and MAE values and improved performance index while maintaining minimal tracking delay relative to the reference heartbeat.