基于深度强化学习改进的Smith预估器温度控制  

Improved smith predictor temperature control based on deep reinforcement learning

作　　者：高东祥 张洪 修伟杰 张林 GAO Dongxiang;ZHANG Hong;XIU Weijie;ZHANG Lin(School of Mechanical Engineering,Jiangnan University,Jiangsu Wuxi 214122,China;Jiangsu Provincial Key Laboratory of Advanced Food Manufacturing Equipment Technology,Jiangsu Wuxi 214122,China;Jiangsu Huilin Environmental Protection Technology Co.,Ltd.,Jiangsu Wuxi 214122,China)

机构地区：[1]江南大学机械工程学院,江苏无锡214122 [2]江苏省食品先进制造装备技术重点实验室,江苏无锡214122 [3]江苏惠霖环保科技有限公司,江苏无锡214122

出　　处：《工业仪表与自动化装置》2024年第3期54-59,99,共7页Industrial Instrumentation & Automation

摘　　要：针对牛粪发酵过程具有惯性大、时滞性、参数变化非线性的特点,提出了一种基于深度确定性策略梯度(DDPG)改进Smith模糊PID控制器的温度控制方法。首先,针对传统模糊PID不能对时滞系统有效控制的问题,建立Smith预估模糊PID控制器。其次,使用DDPG算法改进温度控制器,对设计的智能体进行离线训练。最后,通过仿真对所设计控制器进行实验验证。实验结果表明:DDPG改进的Smith模糊PID控制器能有效消除时滞对温度控制的影响,减少超调量和误差,且能避免被控对象参数随时间变化产生动态偏离时造成的系统不稳定。Aiming at the characteristics of large inertia,time lag and nonlinear parameter change in the fermentation process of cow manure,a temperature control method based on deep deterministic strategy gradient to improve Smith fuzzy PID is proposed.Firstly,to address the issue that traditional fuzzy PID cannot effectively control time-delay systems,a Smith predictive fuzzy PID controller is established.Secondly,use the DDPG algorithm to improve the temperature controller and conduct offline training on the designed intelligent agent.Finally,the designed controller is experimentally validated through simulation.The results show that the Smith PID controller improved by DDPG can eliminate the influence of time delay on temperature control,reduce overshoot and errors,and avoid system instability caused by dynamic deviation of controlled object parameters over time.

关 键 词：温度控制 SMITH预估器 强化学习 神经网络 时滞系统 

分 类 号：TP23[自动化与计算机技术—检测技术与自动化装置]