Incremental optimal process excitation for online system identification based on evolving local model networks

In this paper, a methodology for the generation of optimal input signals for incremental data-based modelling of nonlinear static and dynamic systems is presented. For this purpose, an online strategy consisting of an evolving model and an iterative finite horizon input optimization in parallel to the ongoing experiment is pursued. Such an integrated methodology is methodically very efficient since the experiment is only conducted until the desired model quality is obtained. For the process excitation, the compliance with system input and output limits is of great importance. Especially for nonlinear dynamic systems, the compliance with output constraints is challenging since the current input has an impact on all forthcoming outputs. The generation of optimal inputs is based on the iterative optimization of the Fisher information matrix of the current process model subject to input and output constraints. For the identification, an evolving local model network is used that can cope with a growing amount of data. To this end, the parameter adaptation and the automated structure evolution are characteristic of the evolving local model network. The effectiveness of the proposed method is demonstrated on two typical automotive application examples. First, a stationary smoke model of a diesel engine and second, a dynamic exhaust temperature model are identified by use of optimal process excitation.