Robust extended Kalman filter based state estimation for nonlinear dynamic processes with measurements corrupted by gross errors

Abstract

Abstract The quality of on-line measurement data usually cannot meet the demands of practical process monitoring and control due to the influence of measurement noise. Although extended Kalman filter (EKF) is able to improve the quality by reconciling the measurement data to fit the dynamic process model describing nonlinear and Gaussian processes, it rarely considers the presence of different types of gross errors, such as outliers, bias and drift. However, the three types of gross errors often simultaneously appear in dynamic process systems. We propose a robust EKF combined with measurement compensation (MC-REKF) approach, in which a statistical test method is used to detect the abnormal measurement data, where gross error identification is accomplished within a moving window. Subsequently, the magnitudes of gross errors are estimated and used for measurement compensation. Finally, the compensated measurements are updated to re-estimate the accurate states via EKF. The effectiveness of the proposed MC-REKF is demonstrated through a complex nonlinear dynamic chemical process system, namely the free radical polymerization of styrene. With three different types of gross errors, the mean squared error (MSE) of reconciled measurements based on the MC-REKF decreases 37 fold compared to EKF. The magnitude of the residuals between the estimated states and the true states falls below 1.0E−6 when using the MC-REKF. The implementation results imply that the proposed MC-REKF can identify and estimate different types of gross errors and finally decreases their influence on state estimation and measurement reconciliation.