Neural Network Modeling of Nonlinear Systems Based on Extreme Learning Machine

This paper studies the intelligent modeling method of the whole nonlinear systems, and proposes an improved modeling method based on the forgetting factor recursive least squares method and extreme learning machine. First, establish a nonlinear discrete general-purpose systems model with universality the higher-order and lower-order separation and then rolling optimization identification are established. The unmodeled part of extreme learning machine (ELM) is backward calculated using low-order recursive least squares identification (FFRLS). The linear error is compensated by extreme learning machine. Finally, the alternating identification is carried out under the external error criterion, so as to realize the hybrid intelligent modeling of nonlinear system. This method can overcome the influence of the modeling error of the controlled object and the uncertainty of the structure. Dual networks make the identification of complex systems more organized and simple, and make the identification process is faster and more accurate. Experimental comparative analysis results prove the effectiveness and university of the proposed identification method.