Study of Random Forest to Identify Wiener–Hammerstein System

Abstract

The Wiener–Hammerstein (W–H) system is the most popular type of the Volterra nonlinear dynamical system. It is a combination of two dynamical subsystems, separated by a static nonlinearity. The best linear approximation (BLA) technique assembles two linear filters and the nonlinearity into a single filter for input and output. The main identification challenge resides in separating two filters. This work proposes an iterative random forest as an alternative to select the dynamics combinatorially. It is like the iterative selection of holiday destinations based on the recommendations of random travelers. The proposed technique supports reasonably high noise level and requires the optimization of a single model. Thus, a speedup in processing time is achieved without any prior knowledge about the model configuration both on simulated examples and benchmark data.