OPTIMAL DESIGN OF MODEL PREDICTIVE CONTROLLER

Abstract

Many real industrial processes are nonlinear in nature, or have characteristics, which change over time, due to gradual degradation, variation in raw material, changes in operating conditions etc. For such kind of application if a fixed gain linear controller like PID is utilized, then closed-loop performance could degrade even to the point of process instability. The focus of this dissertation work is to develop an efficient on-line computer based control scheme for such kind of industrial process. In this work Nonisothermal reactor (CSTR) have been taken as a test process for controller design and simulated with the help of Simulink / MATLAB® 7.0.1 software. Pseudo-random Binary sequence (PRBS) signal test and parameter estimation techniques have been used for process identification. Then identified process model has been used for the design of Model based Predictive Controller. Relay feedback experiment has been used for auto tuning of PID controller. Performance of Model Predictive Controller (MPC) has been tested for different objective functions by incorporating input rate and output rate weights. Later part of the work is dedicated to optimal design of Model Predictive Controller for getting optimal performance.Evolutionary computational techniques like Genetic Algorithm and Particle Swarm Optimization have been used to optimize the performance MPC. The results are compared with the conventional PID controller. Thus in this work an effort has been made to develop an efficient control scheme which can give excellent performance for set point tracking and disturbance rejection. The work presented in this dissertation report is done with the help of MATLAB® 7.0.1 software.