Nonlinear Model Order Selection: A GMM Clustering Approach Based on a Genetic Version of EM Algorithm

Choosing an accurate model order is one of the important sections in system identification. Traditionally, selection a nonlinear depends on predetermined model. However, it requires excess calculation and impossible to get rid trouble structural design model, once specific was determined. A false nearest neighbor (FNN) algorithm that only relies input-output data estimate proposed here. Due FNN sensitive its own threshold which crucial constant for evaluating structure, Gaussian mixture (GMM) clustering based genetic version expectation-maximization (EM) minimum description length (MDL) criterion developed this paper, where can be determined without relying The GMM calculate FNN. Then, MDL criteria are embedded optimize EM as reduce influence initial values not prone fall into local extreme well. Three examples given here indicate superiority technique: simulation strongly system, isothermal polymerization process, Van der Vusse reaction relevant reference. Finally, some typical modeling methods conducted confirm validity approach.