Subspace Identification-based Control Performance Assessment With Lqg Benchmark

There are usually tens of thousands of controllers in modern factory. These controllers generally have a good performance in the early stages of operation. However, due to various reasons lacking of regular maintenance over time will result in poor performance. Poor controller performance will reduce the control effectiveness, increase disqualified products and operating costs.Currently, Minimum Variance benchmark is the most frequently used benchmark for feedback control performance assessment, whereas in practice, the control objective is not just minimizing process output variance but also keeping the input variability minimized. The LQG benchmark compromises the requirements and is fit for actual industrial control. However, calculation of the LQG benchmark or Minimum Variance control benchmark requires process knowledge or model identified from data with external excitation, which is harsh in practice.This paper focuses on subspace identification without excitation and LQG benchmark controller performance assessment. The main research results are as follows:1. Calculation of the common control performance benchmark needs process model but the process operates without sufficient excitation, we discussed the identifiability of the fast-sampling-based subspace identification, and proposed a new fast-sampling-based closed-loop subspace identification method for multivariate systems without excitation. While the system is unidentifiable or partly identifiable with traditional sampling rate, the fast-sampling based subspace identification method can improve the model accuracy. Simulation example was included to show the effectiveness of the proposed method.2. Currently, control systems'objective is not only minimizing process output variance but also keeping the input variability minimized. In this paper, we used LQG benchmark to evaluate control performance, and gave corresponding steps in identification and performance evaluation. Based on the subspace matrix obtained from fast-sampling subspace identification without excitation, LQG performance benchmark was calculated and extended to the measurable disturbance processes.3. The identifiability and feasibility of the proposed performance evaluation method was verified through water tank pilot experiments.