基于特征向量子空间距离的MPC控制器性能诊断

Performance diagnosis of model predictive controller based on eigenvector subspace distance

针对当前MPC控制器性能评价方法无法定位性能下降源的问题,提出一种基于子空间距离聚类的控制器性能诊断新方法。新方法引入特征向量子空间描述各性能类别的特征,建立子空间距离来度量当前实时数据和已知类别数据的相似性,以距离为度量函数确定监控数据对应的类别,定位引起MPC控制器性能下降的原因。在Wood-Berry塔上的仿真结果验证了新方法的有效性。

Aiming at the shortcoming that current research on controller performance assessment can＇t isolate the root causes for the poor performance, a novel method of model predictive controller performance diagnosis based on distance clustering was proposed. The concept of eigenvector subspace which can describe the characteristic of various subspace was presented, classification could be made by calculating the distances between the current subspace and the predefined ones, and then it can correctly locate the causes contributed to the performance variation. The simulation results on the Wood-Berry validate the efficiency of the novel method.