Automatic Modeling with Local Model Networks for Benchmark Processes

Abstract In this paper an automated model generation framework is used to identify three nonlinear dynamic benchmark processes. The nonlinearity is approximated using tree-based local model networks (LMN) with external dynamics, which are represented by three different approaches: NARX, NFIR and NOBF. The automated method assumes no prior knowledge about the process, and aims to be a ready-to-use tool for system identification. Results are given for the different approaches and benchmark processes. The importance of the choice of training data for a good generalizing model performance is discussed.