.png){kind=icon}
.png){kind=icon}
Return
Stabilizing and synchronizing the chaotic PMSM systems via improving A2C algorithm
.png){kind=icon}
PRE
AI
DOI：10.1088/1402-4896/ae15c4.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper presents a model-free deep reinforcement learning (DRL) framework for the stabilization and synchronization of chaotic permanent magnet synchronous motor (PMSM) systems, addressing the limitations of traditional control methods that rely on precise system models. The proposed approach employs a modified Advantage-Actor-Critic (A2C) algorithm, featuring a novel nonlinear error-sensitive reward function that significantly enhances convergence speed by amplifying learning signals in the vicinity of the target state. Unlike conventional DRL applications, the control strategy utilizes only two control inputs, reducing implementation complexity and energy cost. Experimental results demonstrate that the proposed method achieves single-system stabilization within 0.28 s and synchronization between two PMSMs within 0.32 s—improving convergence speed by over 40% compared to standard A2C and other reward designs. Under Gaussian noise, the trained policy achieves significantly lower mean absolute errors, demonstrating high control precision and strong robustness. These results highlight the practical potential of DRL in real-world motor control applications, where system uncertainties and external disturbances are inevitable. The integration of a mechanism-driven reward design with reduced control effort represents a significant advancement in intelligent control of nonlinear electromechanical systems.
Journal
IF:
2.6
Papers:
4.3K
Citations:
2.5W
Organization
View more organizations