DYNAMIC NEURAL NETWORKS FOR NON-LINEAR SYSTEM IDENTIFICATION

Abstract

Studies of nonlinear processes indicate that in order to improve the performance, expressing the system dynamics using linear models is insufficient. So the aim of this dissertation is to investigate the process behavior of crystallization process which is strongly nonlinear. In this regard the dynamics of continuous cooling KCL crystallizer were identified using three-input, three output linear and nonlinear model structures. The autoregressive model structure was employed in linear modeling of the process. The nonlinear modeling was performed using different architectures of feedforward and recurrent neural networks. Simulation results demonstrate that the linear modeling, using model for the entire dynamics, is not adequate. Either multimodal or nonlinear modeling is recommended. The performance of different neural network structures in the modeling of the process was illustrated and, based on the results some comparisons were made between these networks. The main objectives of this work are summarised as follows: 1. Study of the crytallisation process in a multivariable framework. This includes investigation of the process stedy state and dynamic behavior. Results from this part will help to select an appropriate model structure and control design techinique. 2. Implementation of system identification using neural network structures to obtain a dynamic model representation for the crystalisation process. Comparison will be made between results obtained from neural network modeling of the process and linear models obtained from system identification using ARX model structures.