基于LSTM的振动台参数整定方法

Parameters Tuning of Shaking Table Based on LSTM

地震模拟振动台的控制多采用三参量控制作为底层控制算法,三参量控制参数多,整定耗时费力。提出一种基于LSTM(长短时记忆网络)的振动台三参量控制参数整定算法。将振动台系统的测试加速度输入和输出数据分为训练集、测试集和验证集,建立并训练一个LSTM深度网络用以模拟振动台的系统模型;针对LSTM深度网络模型引入新的三参量控制环节,采用梯度下降法进行控制参数的离线整定;最后将整定参数与控制系统原参数进行合并用于实机验证。结果表明:所提出的整定方法可以达到优于手动调参的结果,整定过程通过系统模型离线完成,不需实机运行,具有效率高、效果好的优点.

The control of shaking table mostly adopts three-parameter control as the underlying basic control algorithm,and there are many three-parameter control parameters,the parameters tuning is time-consuming and laborious.A three-parameter control parameter tuning algorithm of shaker based on LSTM(long short-term memory network)was proposed.The test acceleration input and output data of the shaking table system were divided into training set,test set and verification set,and an LSTM deep network was established and trained to simulate the system model of the shaking table.For the LSTM deep network model,a new three-parameter control link was introduced,and the gradient descent method was used to perform offline tuning of the control parameters.Finally,the parameters tuning were combined with the original parameters of the control system for real machine verification.The simulation results show that the proposed tuning method can achieve better results than manual parameter adjustment,and the tuning process is completed offline through the system model,without the need for real machine operation,which has the advantages of high efficiency and good effect.