基于子空间LQG的高速列车预测控制器性能监控

Performance Monitoring of High Speed Train Predictive Controller Based on Subspace LQG

针对高速列车在复杂多变环境运行时子空间预测控制器出现性能下降的问题,提出一种基于子空间线性二次高斯(linear quadratic Gaussian,简称LQG)基准的列车预测控制器性能监控算法。首先,使用子空间辨识算法处理列车历史运行数据获得子空间矩阵,设计基于子空间LQG的高速列车预测控制器性能评价基准;其次,通过在线求解列车实时性能指标并与已建立的性能基准进行比较得到评价指标后,对列车预测控制器进行在线评估;最后,对评估结果为列车控制性能下降进行诊断,即建立控制器性能下降模式库,设计基于支持向量机的分类器,对噪声方差变化、过程模型失配、输出约束饱和及控制参数设置不当这4类性能下降源进行训练学习。将测试集输入分类器进行仿真,得到的准确率分别为95.63%,92.49%,90.52%和97.56%,表明该分类器可靠性强,准确率高。

In view of the performance degradation of subspace predictive controller of high-speed train in complex and changeable environment,a performance monitoring algorithm of train predictive controller based on subspace linear quadratic Gaussian(LQG)benchmark is proposed.Firstly,the performance benchmark based on LQG is designed by the subspace matrix,which can be obtained during using subspace identification to process the historical train operation data.By solving the real-time performance index of the train online and then comparing with the established performance benchmark,the evaluation index of the train is obtained,and the train predictive controller can be evaluated on line.Then,when the evaluation result is degradation,it needs to diagnose the concrete type,that is,to establish the performance degradation mode database of controller,and a classifier based on support vector machines is designed to train and study the four performance degradation sources,which are noise variance change,process model mismatch,output constraint saturation and control parameter setting improperly.The accuracy of the test set input to the classifier is 95.63%,92.49%,90.52% and 97.56%,which shows that the classifier has high reliability and accuracy.