Hierarchical identification of nonlinear hybrid systems in a Bayesian framework

This paper presents a hierarchical framework for the identification of nonlinear hybrid systems in form Switched Nonlinear AutoRegressive models with eXogenous variables (SNARX). The is done via three levels inference, using Bayes' rule. In first level, model parameters are computed Maximum Posteriori (MAP) estimator. posterior distribution therein involved depends on hyper-parameters that tuned second level inference. Such terms determine complexity, and Bayesian key returning values trade off complexity accuracy by automatically embodying Occam's razor principle. Lastly, third compares different structures means quality measure encompasses data fitness, classification. proposed compared existing relevant methods tested numerical models, showing promising performance.