Research Of Performance Evaluation For Control Systems Based On Multivariate Statistical Methods

Now the researches of control engineering domain mainly focus on model identification, design of controller and robustness of control systems, etc. Little has been done to the performance assessment of the running control systems. In fact, many control systems can’t achieve the design index. So, it is a concerned problem for local technicians and control engineers that how to obtain the performance information of the control systems from the operating data so that the performance problem can be recognized and diagnosed early. Recently, the demand of high control system performance from industry promotes the development of the research of control system performance assessment and diagnosis rapidly.First, the minimum variance benchmark which is the most extensive useful performance evaluation method is introduced in this paper. The minimum variance benchmark can provide a practical reference to the performance of control systems, and this method only need the routine data and the time delay of the process. In order to evaluate the control systems, we use the method to compare the output variance with the minimum variance.The minimum variance benchmark which is the most extensive useful performance evaluation method has two problems for multivariable process control system, which needs correlation matrix and the white noise covariance∑ya of the number of the output sequence which is i. In order to receive both the amount may need to go through complex calculations, thus the application of this method in actual industrial process is limited.Because the multivariate projection technology is the method which can make the high-dimensional variables to a few low-dimensional variables, and then we can use a single variable minimum variance performance evaluation method for performance evaluation.Therefore, this paper uses the method based on multivariate statistical methods to evaluate the multi-variable control system.This paper first describes multivariate statistical methods in detail, then combines the criteria of minimum variance performance evaluation.According to the situation of multi-variable process control system, using multivariate statistical methods to reduce the dimension of the variables, then using a single variable minimum variance performance evaluation method to evaluate the variables, the dimesion of which has been reduced.For the nature of multivariable process control system, this paper introduces the performance evaluation method based on independent component analysis and the improved calucation of the independent component anlysis, and because the special nature of batch process, this paper also gives the performance evaluation method based on ICA-MPCA. The result of simulation shows that the above method is effective.