基于DDPG的振动台控制参数整定方法  

Parameter Tuning of Shaking Table Based on DDPG

作　　者：纪金豹[1] 黄飞 张文鹏 JI Jinbao;HUANG Fei;ZHANG Wenpeng(Beijing Key Lab of Earthquake Engineering and Structural Retrofit,Beijing University of Technology,Beijing 100124,China;Power China Hubei Electric Engineering Co.,Ltd.,Wuhan Hubei 430040,China)

机构地区：[1]北京工业大学,工程抗震与结构诊治北京市重点实验室,北京100124 [2]湖北省电力规划设计研究院有限公司,湖北武汉430040

出　　处：《机床与液压》2024年第14期160-166,共7页Machine Tool & Hydraulics

基　　金：国家自然科学基金面上项目(51978015)。

摘　　要：三参量控制是地震模拟振动台的底层控制算法,其参数整定过程中涉及的参数多,传统的参数整定方法存在效率低、过程繁琐等问题。为了提高整定效率和准确性,提出一种基于确定性策略梯度(DDPG)算法的振动台三参量控制参数整定算法。此方法通过将振动台三参量控制系统作为强化学习环境,利用DDPG算法对系统的状态-动作-奖励数据进行学习和训练;训练好的智能体则可以输出最优的控制参数,然后将整定完成的控制参数放在实际振动台系统模型中进行测试。结果表明:DDPG算法可以有效优化振动台控制性能,提高试验结果的准确性和可靠性,具有实际应用价值。Three variable control is commonly used as the underlying control algorithm in earthquake simulation shaking table,involved numerous parameters in the process of parameter tuning,and traditional parameter tuning methods suffer from problems such as low efficiency and complicated processes.In order to improve tuning efficiency and accuracy,a novel parameter tuning method for three variable control of shaking table based on the deep deterministic policy gradient(DDPG)algorithm was proposed.Taking the three variable control system as a reinforcement learning environment,the DDPG algorithm was used to learn and train the state-action-reward of the system,the optimal control parameters were obtained.The tuning parameters were then tested in shaking table and compared with traditional tuning methods.The results show that the DDPG algorithm can effectively optimize the control performance of the shaking table and improve the accuracy and reliability of experimental results,which has practical application value.

关 键 词：地震模拟振动台 三参量控制 参数整定 深度强化学习 确定性策略梯度 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]