An adaptive real-time outlier detection algorithm based on ARMA model for radar's health monitoring

To guarantee the data quality is of the first importance in the research of monitoring and management of the health of large complex electronic systems such as radars, and tracking the working performance. Influenced by the combined effect of all kinds of interfering factors, the observation series reflecting the health of a radar usually includes variety of outliers and noise, which ought to be detected and eliminated in time. After modeling the radar's health observation series, analyzing the correlation of neighbored observations based on Autoregressive and Moving Average Model (ARMA), a real-time outlier detection and replacement algorithm was put forward, which could rectify the employed ARMA model self-adaptively when radar's heath evolvement stage changed. The promoted algorithm was suitable to detect and replace the outliers dynamically, overcoming the shortcomings of the classical threshold comparison detection methods which could not deal with the outliers belongs to the heath evolution process characterized by trend and self-correlation. Finally, the experiment result indicated that the algorithm could detect the outliers validly and satisfy the application's demand for real time.