A Robust State Estimator With Adaptive Factor

This article proposes a robust state estimator with adaptively adjusted observation noise covariance for uncertain linear systems. Since the residuals between observation data and estimation data satisfying the normal distribution, the Mahalanobis distance are known to be Chi-square distributed. We use Chi-square tests to timely distinguish outliers that beyond the confidence interval and adjust the estimated observation noise covariance to a more likely value. Combined with the robust estimation method, an improved algorithm is derived to deal with uncertainties in both system parameters and observation noise covariance. Based on the proposed prediction form, we test the obtained robust state estimator with different deterioration conditions of observation noise covariance and compare it with the estimators without adaptive factor. The simulation results show that the derived state estimator may reduce the accumulation of estimation errors, smooth the estimated state curve, and the performance of the proposed estimator can be significantly improved as the deterioration enlarging.