Unscented transformation-based filters: Performance comparison analysis for the state estimation in polymerization processes with delayed measurements

Abstract

State estimation with delayed measurements is essential to the operation of polymer processes due to the limited availability of reliable online sensors and the unavoidable hold-up time in the acquisition of critical variables data. In this work, a two-timescale approach is applied to three filters based on the Unscented Transformation, the Unscented Kalman Filter, the Unscented Recursive Nonlinear Dynamic Data Reconciliation and the Reformulated Constrained Unscented Kalman Filter, in order to incorporate delayed measurements into their estimation scheme. A comprehensive comparative analysis is performed, which shows that the three of them have very good accuracy and convergence properties. However, the Unscented Kalman Filter performs better in terms of computational time. The handling of delayed measurements is incorporated into three filters of the UT family by means of a two-timescale approach. A comprehensive analysis of the filters' performances in polymer processes is carried out, showing that these filters behave better than classical techniques such as the Extended Kalman Filter. One of them, the Unscented Kalman Filter, stands out from the other two by requiring less computational time.