Support Vector Regression Based Controller for Non-Linear CSTR Tank Processes

Abstract

Support Vector Regression (SVR) based predictive controller is proposed for nonlinear Continuous Stirrer Tank Reactor (CSTR) processes. SVR based prediction control algorithm is recommended for complex non-linear processes, having very high dimensional input spaces. The proposed control algorithm is compared with Neural Network Predictive Control (NNPC) algorithm to show its effectiveness and observed that, local minima are hidden units selection issues involved in Neural predictive controller are solved. The same algorithm is used in nonlinear Continuous Stirrer Tank Reactor (CSTR) tank process to observe the effectiveness. In CSTR, concentration is controlled by manipulating its feed flow rate. From results, it is concluded that SVR ensures robustness and control accuracy under servo and regulatory operations.