NONLINEAR TIME-SERIES MODELS FOR MULTIVARIABLE DYNAMIC PROCESSES

Several paradigms are available for developing nonlinear dynamic input -output models of processes. Polynomial models, threshold models, mode ls based on spline functions, and polynomial models with exponential a nd trigonometric functions can describe various types of nonlinearitie s and pathological behavior observed in many physical processes. A uni fied nonlinear model development framework is not available, and the s earch of the appropriate nonlinear structure is part of the model deve lopment effort. Various artificial neural network structures and nonli near time series model structures are presented and illustrated by dev eloping a model from data sets generated by a series of example system s. The use of a nonlinear model development paradigm which is not comp atible with the types of nonlinearities that exist in the data can hav e a significant effect on model development effort and model accuracy.