.png){kind=icon}
.png){kind=icon}
Return
Outperforming Classical PI Tuning Methods via AlphaZero Algorithm
.png){kind=icon}
PRE
AI
DOI：10.1007/s40313-025-01225-x.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper introduces AlphaTune, a novel method for tuning proportional-integral (PI) controllers based on the AlphaZero reinforcement learning algorithm. The method formulates the controller tuning problem as a two-player game, where an agent plays against itself to optimize the controller parameters kp\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_p$$\end{document} and ki\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_i$$\end{document}. The search space is constrained using the Signature Method to ensure that exploration occurs only within the stabilizing set of gains. AlphaTune is designed to meet time-domain performance specifications, specifically, settling time, overshoot, and control signal constraints, which are often unaddressed by analytical methods. Simulation results demonstrate that AlphaTune outperforms established classical tuning techniques such as Ziegler-Nichols, CC, CHR, IMC, and SIMC. To the best of our knowledge, this represents the first application of the AlphaZero algorithm to the problem of controller tuning, offering a powerful and flexible new approach for control system design.
Keywords:
PI controllers
Reinforcement learning
AlphaZero
Tuning method
Artificial intelligence
Journal
J
IF:
1.3
Papers:
103
Citations:
1.1K
