Simulation-Based Artificial Neural Network Predictive Control of BTX Dividing Wall Column

Abstract

For the separation of ternary liquid mixture, use of dividing wall column is one of the nonconventional techniques in the field of liquid separation by thermal process. Benzene, toluene, and o-xylene have been selected as a ternary system for this study. As it is difficult to control the product purity directly due to delay time in the composition analyzer, temperatures of the appropriate trays were selected as the controlled variables. Reflux rate, sidestream flow rate, and reboiler heat duty were selected as manipulated variables to control sixth tray temperature of rectifying section, 11th tray temperature of the main column, and 12th tray of stripping section. Back-propagation algorithm was used as a training algorithm to tune the connection weights for the function of each neuron. The control performance of ANNPC was investigated for ±10 % load changes in feed flow rate, feed composition, and liquid split factor. The control performance was analyzed by using performance criteria indexes and performance parameters such as IAE, ITAE, ISE, ITSE, rise time, and settling time. It is observed that these performance parameters are less for ANNPC as compared to PID control. The settling time in case of PID varies from 2.77 to 4.55 h, significantly higher than that in ANNPC (0.37–0.66 h). The rise time is 0.66–1.20 h for PID and 0.03–0.27 h for ANNPC. These results indicate that ANNPC performs better than PID controller.