基于强化学习的热连轧智能PID厚度控制

Hot continuous rolling intelligent PID gauge control based on reinforcement learning

针对某钢厂2250热连轧产线是一个变时滞、强耦合的非线性系统,传统的PID控制在处理此类非线性系统时存在着抗干扰性差以及实时控制能力不足等问题,为了提高轧制成品质量,提出一种基于强化学习优化PID控制器参数的设计办法,并在钢铁热连轧领域首次应用。该控制器通过强化学习与PID控制相结合,应用基于执行器⁃评价器结构以及径向基网络的自适应PID控制器在线优化PID控制器参数。实验结果表明,在外部存在干扰的情况下,该控制器能快速回到稳态,且在存在变时延的情况下响应迅速,能快速地作出反应,具有良好的动态性能。通过该控制器的设计结果证明,该智能优化控制器与传统PID控制器相比具有超调量小、响应时间短的优点,具有良好的鲁棒性以及自适应性。

The 2250 hot continuous rolling line of a certain steel plant is a nonlinear system with time⁃varying delay and strong coupling.The traditional PID control has some deficiencies,for example,poor anti⁃interference,inadequate real⁃time control ability when dealing with this kind of nonlinear system,etc.Therefore,a design method of PID controller parameters optimized by reinforcement learning is proposed to improve the quality of rolled products.This method is applied in the field of hot continuous rolling of iron and steel for the first time.By combining reinforcement learning with PID control,the adaptive PID controller based on actuator⁃evaluator structure and radial basis function network is applied to optimize PID controller parameters on line.The experimental results show this controller can quickly return to the steady state in the presence of external disturbances,and respond rapidly in the presence of time⁃varying delay,so it has good dynamic performance.In comparison with the traditional PID controller,the design result shows that the intelligent controller has the advantages of small overshoot,short response time,good robustness and self⁃adaptability.