基于强化学习的参数自整定及优化算法

传统PID控制算法在非线性时滞系统的应用中，存在参数整定及性能优化过程繁琐、控制效果不理想的问题。针对该问题，提出了一种基于强化学习的控制器参数自整定及优化算法。该算法引入系统动态性能指标计算奖励函数，通过学习周期性阶跃响应的经验数据，无需辨识被控对象模型的具体数据，即可实现控制器参数的在线自整定及优化。以水箱液位控制系统为实验对象，对不同类型的PID控制器使用该算法进行参数整定及优化的对比实验。实验结果表明，相比于传统的参数整定方法，所提出的算法能省去繁琐的人工调参过程，有效优化控制器参数，减少被控量的超调量，提升控制器动态响应性能。

Parameter self-tuning and optimization algorithm based on reinforcement learning

To achieve better control performance in the nonlinear time-delay system, the traditional Proportional-Integral-Derivative (PID) control algorithm requires tuning and optimization, which complicates the controller design. First, we propose a new self-tuning and optimization algorithm for controller parameters based on reinforcement learning. Then, a reward function based on the system dynamic performance index is introduced by this algorithm. This function can learn the empirical data of periodic step response and realize the online optimization of controller parameters without identifying the model data of the controlled object. Finally, the algorithm is tested through experiments on a water tank level control system with different types of PID controllers. Experimental results show that, in contrast to the traditional parameter tuning method, the manual process is eliminated by the proposed algorithm, effectively optimizing the controller parameters, reducing the overshoot of the controlled quantity, and improving the dynamic response performance of the controller.