A Bayesian robust identification method for piecewise affine autoregressive exogenous model from outlier-contaminated data

This paper presents a robust identification method for piecewise affine autoregressive exogenous(PWARX) model from outlier-contaminated data. The piecewise affine (PWA) system representation is useful for process control due to its capability of describing both different characteristics depending on changes in operational conditions and non-linear characteristics. The PWARX model, which is a subclass of the PWA system, is often used for such identification. Outliers, however, which are irregular observations, are rarely dealt with simultaneously even though they can lead to misidentification. Therefore, we propose a hierarchical Bayesian model for identifying both PWARX model and possibility of outlier for each sample, and give the solution with the Variational Bayes method. Furthermore, a model selection method is presented within our framework. The capability of our approach is examined through a benchmark system.