Multi-Phase Focused PID Adaptive Tuning with Reinforcement Learning

The Proportional-Integral-Derivative (PID) controller, a fundamental element in industrial control systems, plays pivotal role regulating an extensive array of controlled objects. Accurate and rapid adaptive tuning PID controllers holds significant practical value fields such as mechatronics, robotics, automatic control. three parameters the controller exert substantial influence on performance, rendering these area interest within related research fields. Numerous techniques are widely employed to optimize its functionality. Nonetheless, their adaptability stability may be constrained situations where prior knowledge is inadequate. In this paper, multi-phase focused method introduced, leveraging deep deterministic policy gradient (DDPG) algorithm automatically establish reference values for tuning. This constrains agent actions multiple phases based reward thresholds, allowing output focus stable region, which provides enhanced maintains even with limited knowledge. To counteract potential issue vanishing following action constraints, residual structure incorporated into actor network. results experiments conducted both first-order second-order systems demonstrate that proposed can reduce tracking error by 16–30% compared baseline methods without loss stability.