Contrtol Performance Assessment Of Multivariate Process Based On Data Driven

So far,the calculation of evaluation benchmark based on minimum variance(MV)requires prior knowledge of process transfer function matrix or at least the first several Markov coefficient matrices,and the online performance assessment based on MV needs to use interactor matrix as prior knowledge and separate feedback invariants from routine operation closed-loop output data.Many scholars at home and abroad try to reduce the interactor matrix as a prior knowledge,but still need information essentially equivalent to the interactor matrix.Therefore,this paper aims to solve the following two problems:(1)the calculation of MV benchmark without prior knowledge of process;(2)data-driven MV performance evaluat:ion algorithm and its implementation under routine operating conditions.Based on the known delay matrix,the following solutions are given for different situations in the actual control process.Firstly,when the indicator matrix of process delay in multivariable systems can be changed into a Row-Echelon form matrix through the row(column)shift operations,accurate MV benchmark can be obtained without any other knowledge of process transfer function.Secondly,when the delay matrix can not meet the above requirements,MV benchmark is estimated by the proposed algorithms for three scenarios with different knowledge levels of the process:(1)when only the knowledge of time delay matrix is available,performance index is given as a range with limit upper and lower bounds;(2)when the knowledge of time delay matrix and the range of the pseudo first Markov parameter matrix are available,the performance index is given as a tighter range with the upper and lower bounds;(3)in the case of that both the knowledge of time delay matrix and the pseudo first Markov parameter matrix are exactly known,MV benchmark is calculated exactly.Thirdly,a data-driven parameter estimation method based on common closed-loop conditions is proposed to avoid additional excitation signals to the control system.This method adjust the parameters of the controller to get two sets of stable output data under different controllers for time series modeling,and then the Markov coefficient matrix of process transfer function is obtained by matrix operation.Through numerical examples and Simulation of Shell heavy oil fractionation process,the data-driven MV performance assessment process is realized,and the estimated MV performance index is close to the theoretical value.At the same time,the simulation verifies that the MV benchmark value obtained by this algorithm is identical to the traditional method.With the same prior knowledge,the upper and lower bounds estimated by this algorithm are more compact than those obtained by other methods in the references.