Evolving Neuro-Fuzzy Design of Experiments: A Novel Approach of Nonlinear Process Identification

This paper presents a novel model-oriented sequential online design of experiments for model identification. An evolving neuro-fuzzy model identification of nonlinear dynamical systems is combined with two approaches to experiment design. First, a sequential design based on the maximin space-filling criterion was used to select an input space with the least model coverage. Second, an optimal experiment design method was used to select the optimal input signal online near the operating point selected with the first step. The use of an evolving model allows online generation of a model-based optimal input signal for identification of nonlinear dynamical processes without the need for multiple iterations or an initial model. The method was validated on a multiple-input single-output theoretical model of a plate heat exchanger.