Correction of Kalman filter in the presence of outlier

In this paper, a new method of detecting outlier in data is proposed. The new method is based on the identification of the ARMA model of system output. The outlier can be detected by a detection function. The recursive extended least-squares (RELS) method is used to identify the ARMA model of system output. Since the method is very sensitive to changes of coefficients of the ARMA model, an outlier can be detected quickly. Because the performance of Kalman filter will be deteriorated by the outlier, therefore, after the detection of the outlier, the residual of the Kalman filter is smoothed. Using this correction, the performance of the Kalman filter is improved. As an example of application, a simulation of guidance for semi-active radar homing missile is conducted. The result of the simulation proves that the outlier can be detected correctly, and the correction of Kalman filter is efficient and practical