Nonlinear state-space system identification with robust laplace model

This paper investigates a robust identification solution for the nonlinear state-space model in which the outputs are polluted by unknown outliers. The problem of outliers is frequently encountered in practical industries that can greatly challenge the modelling of industrial processes. In order to overcome the obstacles brought by the outliers, the heavy-tailed Laplace distribution is applied to describe the output measurement process. Specifically, the Laplace distribution can be decomposed as a scale mixture of Gaussian distributions, which makes it robust for the outliers. The unknown model parameters are estimated with the expectation–maximisation algorithm while the particle smoother is used to solve the latent state estimation problem. The usefulness and robustness of the proposed algorithm are verified through the numerical examples including the model of a common chemical process.