Robust LPV System Identification with Asymmetric Measurement Noise

In this paper, we consider the identification problem for linear parameter varying (LPV) systems with skewed and asymmetric measurement noise. The shifted asymmetric Laplace (SAL) distribution and the expectation maximization (EM) algorithm are combined to build the robust system identification framework. To overcome various types of outliers in the data including the skewed and asymmetric noise, the skewness and tails of the SAL distribution are adaptively adjusted based on the collected identification data. The maximum likelihood estimates (MLEs) of the parameters are produced by using the expectation maximization algorithm, in which the system model and noise parameters are extracted directly from the collected identification data. The effectiveness of the proposed algorithm is exhibited in the verification section.