Pilot scale distillation column control using neurol networt

Abstract

We propose a new feedforward control scheme combining neural network with Model Predictive Control (MPC), and review the theories of neural network, DMC and feedforward DMC briefly. The disturbance error learning techniques is proposed for on-line training of the neural network. Dynamic Matrix Control (DMC) is chosen as an example MPC. The proposed control scheme is efficient to the systems where the disturbance dynamics and complex and uncertain. The neural feedforward controller is implemented on a pilot scale distillation column in order to disturbance rejection. The performance of the neural controller is compared with that of DMC and feedforward DMC. The neural feedforward controller shows good performance. Also the controller appears to be relatively robust in the face of process/model mismatch in DMC. The results show that the neural controller can perform well even in situation with significant nonlinearities, strong interactions, and complex dynamics. results of implementation show good possibilities to industrial applications of the neural controller. We believe the possibility of the neural feedforward controller in industrial applications.