Prediction interval-based controller for Chemical Reactor

Abstract

Chemical Reactors, such as continuous stir tank reactor (CSTR) and polymerization reactors (PR) are nonlinear dynamics in nature. Moreover, in practice, uncertainties and disturbances always exist in operation of such kind of reactors. Therefore, the smooth tracking of these reactors is still a challenging task for the researchers. As a core contribution, a prediction interval (PI)-based controller (PIC) is proposed for nonlinear chemical reactors in this paper. PIs have been vigorously used in forecasting problems to quantify the uncertainties and disturbances of the nonlinear systems. It is well established that PI is more informative than point forecast, and hence, it is expected that the integration of PIs in a control system can accelerate the performance of the controller. In the proposed PIC system, neural network (NN) inverse model is used as a controller. PI-based model is used in the loop of the proposed control system to construct the PIs for the controller. PIs are then used as additional inputs for the PIC to predict control signal. An exothermic chemical reactor, CSTR is used to examine the performance of the proposed PIC. Simulation study demonstrated that the setpoint tracking performance of the PIC is acceptable, and superior compared to traditional NN controller.