Method Of Model Predictive Control System Performance Assessment

Model predictive control is widely used in complicated industrial processes, but the performance will become bad gradually because of various factors in actual processes, so it is much valuable to monitor the performance online and diagnose the root cause of model predictive control system. An overview of the current status in control system performance assessment is presented and minimum variance control benchmark is introduced. The performance assessment of model predictive control system is discussed mainly. The work includes following aspects:(i) The minimum variance control and generalized minimum variance control are investigated. The method using minimum variance control and generalized minimum variance control benchmark to assess the performance of univariate and multivariate control system is introduced. A calculation example is given.(ii) The performance assessment benchmark of model predictive control system is studied. The Cumulative Sum (CUSUM) chart of residual is put forward to monitor the performance of model predictive control system online. The method has solved the problem that the (X|-)- chart can hardly detect the change of performance index with small change or severe autocorrelation.(iii) Reasons of poor performance of model predictive control system are divided into two groups based on the Cumulative Sum (CUSUM) chart monitoring results of historical and design performance index residuals. These reasons include model mismatch, effect of unmeasured noise, input saturation and effect of measured disturbance. The type of reasons can be distinguished finally by manipulated variables observation and model validation.(iv) Case studies with model of a Shell tower is made to test the method of model predictive control performance assessment. It illustrates that the Cumulative Sum (CUSUM) chart of residual can monitor the model predictive control performance online effectively. Historical and design performance index can distinguish the type of root cause correctly.