Modeling-on-demand-based multivariable control performance monitoring

Abstract

The current work presents a modeling-on-demand-based multivariable control performance monitoring (CPM) strategy. The suggested approach addresses several challenging topics in multi input-multi output CPM area such as data-driven characteristics, recursive/adaptive philosophy, modeling, plant-wide control (PWC) performance/feasibility indicator, simple structure, minimum interference with the process operation and recovering actions suggestions. The new model-on-demand (MoD) algorithm combines partial least squares (block-wise and moving-window), model quality index, and the reference matrix concepts to improve the robustness characteristics of the recursive process, i.e. the model adaptation is only performed when it is needed. Recovery actions may involve changes on tuning parameters as well as control structure modifications. The proposed strategy allows to identify static and dynamic abnormal events in the process which could produce strong degradation of the PWC performance. The robustness of the proposed methodology is tested by dynamic simulations on the well-known Shell fractionator process and compared with some classical recursive algorithms.