Neural Network Modeling and Control of an Etherification Hybrid Reactor
Systematic methods and tools for managing the complexity
Process Control (T4-8P)
Keywords: neural network modeling, neural network control, internal model control, hybrid reactor
The etherification processes are improved by a hybrid reactor which combine separation with chemical reaction. As a result, the process yields are higher while their process dynamic behaviors are more complex. A conventional Proportional Integral Derivative (PID) controller, a linear controller, which has been widely used in the industrial chemical processes, is able to control the non-linear and complex processes but with slow responses, low performances, limited operating ranges. In addition, its performance is not guaranteed in the cases of disturbance changes and plant-model mismatches as well. In recent years, the neural network control techniques have been successfully applied to those highly non-linear and complex systems due to the present availability of advanced computer technology. However, those investigations did not apply to any hybrid reactors. Moreover, the purpose of those previous researches was to control the reactor temperature. On the other hand, to set up the product specifications, the reactant concentration needs to be controlled rather than the reactor temperature. The objective of this study is aimed at controlling the reactant concentration in the reactor during the etherification by manipulating the jacket temperature to obtain a high yield of the desired product. In this paper, neural networks are used to not only develop a black box model but also formulate neural network internal model controller in a Proportional Integral-Nonlinear Internal Model Control (PI-NIMC) cascade strategy. The simulation study involves the use of PI-NIMC cascade control for set point tracking and disturbances rejection in both nominal and plant-model mismatches conditions and its performance is compared with that of the conventional PID cascade control. Simulation results indicate that PI-NIMC cascade control strategy provide better control performance than the conventional PID cascade control in all cases. Besides, the results justify the use of neural network control technique in a highly non-linear and complex process.
Presented Tuesday 18, 13:30 to 15:00, in session Process Control (T4-8P).
