With the advance of industrial automation, more and more attention have been focused on how to evaluate the system performance using actual operating data in industrial process. Minimum variance benchmark was first used by Harris to evaluate the performance of single input single output (SISO) system. Because the estimation of minimum variance benchmark only use the knowledge of time delay and actual operating data without additional experiments for SISO system. Therefore, this evaluation method based on minimum variance benchmark is widely used in industrial process.The minimum variance benchmark can be estimated by Filtering and Correlation Analysis (FCOR) of the input and output data under the condition of the process time delay known. FCOR algorithm is intuitively simple and computationally efficient, however, it should be noticed that FCOR algorithm is mainly used to analyze models established by stationary series. In actual systems, the disturbance input signal is non-stationary series due to the complexity and variability of the disturbance signal, which makes the output data is non-stationary and the FCOR algorithm can not be directly applied. Therefore, the non-stationary output signals should be smooth processing first, then the FCOR algorithm can be used to obtain the estimates value of the minimum variance benchmark. For the general control system, there are two cases of non-stationary disturbance signals:one is random walk disturbance, corresponding to the integrated moving-average model, the other is a combination of stationary signal and periodic signal. In this paper, repeat difference method are used to processing the non-stationary output signals, and the FCOR algorithm are used to estimate the value of minimum variance benchmark. Finally, via MATLAB simulation, the comparison between the estimated variance of the non-stationary series by FCOR algorithm and that of the actual output data verifies the effectiveness of FCOR algorithm. This paper first introduces the control system performance assessment methods, and the stochastic method based on minimum variance benchmark of performance assessment is mainly concerned in this paper. Secondly, knowledge of the time series analysis is presented. Then, under the constrained actuator and the unconstrained actuator conditions, the derivation process of the minimum variance control law are given respectively. The steps and methods of performance assessment of the SISO system are provided under the unconstrained minimum variance control. Finally, two cases are studied to analyze the causes of non-stationary series, and a non-stationary data processing method is presented. In view of the two cases, simulations are given respectively, in which MATLAB simulations are utilized to estimate the minimum variance benchmark. |