基于Q-学习的底盘测功机自适应PID控制模型  

Adaptive PID Control Model of Chassis Dynamometer Based on Q-Learning

作　　者：乔通 周洲[1,2] 程鑫 郭兰英[1] 王润民 QIAO Tong;ZHOU Zhou;CHENG Xin;GUO Lan-ying;WANG Run-min(School of Information Engineering,Chang’an University,Xi’an 710064,China;Shaanxi Engineering Research Center of Internet of Vehicles and Intelligent Vehicle Testing Technology,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,陕西西安710064 [2]陕西省车联网与智能汽车测试技术工程研究中心,陕西西安710064

出　　处：《计算机技术与发展》2022年第5期117-122,共6页Computer Technology and Development

基　　金：国家重点研发计划项目(2018YFB1600800);陕西省重点研发计划(2020GY-018);西安市科技计划项目(20RGZN0008);中央高校基本科研业务费专项资金项目(300102241305)。

摘　　要：为了解决汽车底盘测功机控制系统在动态控制时出现延迟较高和误差大的问题,提出了一种基于强化学习的底盘测功机控制策略。以PID控制算法为基础,扭力偏差为控制器输入,调节电压控制量为输出,选择扭力差变化为智能体奖惩的学习策略,通过Q学习算法对PID参数进行在线自适应整定;在底盘测功机仿真试验中验证了控制器的调控性能,并与传统PID控制以及神经网络PID控制的结果进行了对比;实验结果表明,基于Q学习的自适应PID控制模型较传统PID算法控制周期缩减至40.7%,相较于神经网络PID算法控制周期缩短至27.9%。相对于传统PID控制模型与神经网络PID模型,基于Q学习的自适应PID控制模型输出力上升过程稳定且快速。提出的基于Q学习的自适应PID控制模型能够有效提升底盘测功机控制精度,满足其使用的工业要求。In order to solve the problems of high delay and large error in dynamic control of chassis dynamometer control system, a chassis dynamometer control strategy based on reinforcement learning is proposed. Based on the PID control algorithm, the torque deviation is the input of the controller, the control quantity is the output, and the selection of the torque difference is the learning strategy of the intelligent body reward and the penalty, and the PID parameter is adjusted by the Q learning algorithm. In the simulation test of chassis dynamical machine, the control performance of the controller is verified, and the results of the traditional PID control and the neural network PID control are compared. The experimental results show that the control cycle of the adaptive PID control model based on Q learning is reduced to 40.7% compared with the traditional PID algorithm, and the control cycle is shortened by 27.9% compared with the neural network PID algorithm. Compared with the traditional PID control model and neural network PID model, the process of output force rising of the adaptive PID control model based on Q-Learning is stable and fast. The proposed adaptive PID control model based on Q learning can effectively improve the control accuracy of the chassis dynamometer and meet the industrial requirements of the chassis dynamometer.

关 键 词：强化学习 PID控制 Q学习 控制策略 底盘测功机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]