Decoupling nonlinear state-space models: case studies

Identifying nonlinear models becomes more important in a wide range of engineering applications. There are two promising approaches, namely black-box models, of which the internal workings are often very complicated, and block-oriented models, which consist of simple interconnections of linear dynamical systems and static nonlinearities. Black-box models are relatively easy to estimate, and provide a good system description, but only offer a limited correspondence to the underlying nature of the system. Block-oriented models use fewer parameters, and can represent the underlying nature of a system, but require tailored identification methods. We propose to combine the two model classes: from a given initial black-box state-space model, we derive a so-called decoupled representation, which uses fewer parameters and may offer insight to the nonlinear nature of the system. A significant reduction in the number of model parameters is thus obtained, while maintaining a low model error.

ISBN:9789073802940

Jaar van publicatie:2016