A modified Hammerstein modeling by the differential evolution algorithm

Abstract This paper focuses on the nonlinear system modeling based on using a modified Hammerstein system model. The proposed Hammerstein structure is composed of a bilinear neural network (BNN) and a recursive digital system in the cascaded form. The former is taken to be the nonlinear function part of the Hammerstein model, and the latter is used as the linear dynamic subsystem. The BNN is then constructed by the bilinear digital system and the recurrent neural network, which already possesses a satisfactory modeling capacity. To update all the adjustable parameters within the proposed Hammerstein model, a popular and powerful evolutionary computation called the differential evolution (DE) is utilized so that the model output can be very close to the actual nonlinear system output. Finally, a simulated nonlinear chemical process system, continuously stirred tank reactor (CSTR), is illustrated with the modeling phase and testing phase. Some numerical results as compared with a different method from subject literature are provided to show the feasibility of the proposed method and its good modeling.