Identification of linear parameter-varying system with missing measurement data and outliers

The robust identification of linear parameter-varying (LPV) finite impulse response (FIR) system with unknown missing output is considered. This paper provides a comprehensive discussion of common outliers and unknown missing measurement problems in practical processes. A Student's t distribution is utilized to data with outliers, and also to automatically identify missing measurements, an indicator variable is introduced for each measurement that follows a Bernoulli distribution. After that, determining whether measurements are missing or not and estimating the unknown parameters by the variational Bayesian (VB) algorithm. A numerical example and the cascaded tank system are provided to exemplify this algorithm and demonstrate its robustness and innovation.