Control of Continuous Stirred Tank Reactor using neural networks

Abstract

Background/Objectives: This paper presents the design of neuro controller NARMA-L2 for composition control in an isothermal Continuous Stirred Tank Reactor (CSTR) by manipulating the input feed composition. Methods/Statistical Analysis: The NARMA-L2 controller design is implemented in two stages in which the first stage is system identification to model the process and the second stage is designing the process controller. For controlling the product composition in the CSTR, the neuro controller NARMA-L2 is implemented in MATLAB Simulink environment. Findings: The simulation results show the superiority of the NARMA-L2 in accurately tracking the composition set-point changes in the CSTR and control the system better as compared to that of the conventional PID. The neuro controller NARMA-L2 can handle non-linear aspects of the CSTR by transforming its non-linear dynamic into an implicit algebraic model which can control the trajectory of the CSTR efficiently. Application/Improvements: The advantage of using the neuro controller NARMA-L2 is that it requires the minimal online computation compared to other neural network architecture for control such as model reference control and model predictive control.