Identification of batch reactor process using Extreme Learning Machine based Hammerstein and Hammerstein-Wiener Models

This paper concentrates on system identification using neural networks, specifically avoiding the need for gradient calculation. In our proposed approach, the parameters of the nonlinear static blocks in both Hammerstein and Hammerstein-Wiener models are represented as Single Hidden Layer Feed forward Networks (SLFNs). The identification of both nonlinear and linear parameters is accomplished through the application of Extreme Learning Machine (ELM). In ELM, the hidden layer weights and biases are generated randomly and remain fixed, whereas the output weights are computed using linear regression. The training process involves a forward pass, in which the hidden layer activations are computed, followed by the computation of the output weights and backward pass is exempted, where iterative optimization based weight updates are done. To show the efficacy of the ELM based block oriented models, intense nonlinear batch reactor process is identified with its input output data. One can take away the best model fit for controller design from this ELM based modelling.