Recurrent Neural Network based Soft Sensor for Monitoring and Controlling a Reactive Distillation Column

Abstract

Abstract For the real time monitoring of a Reactive Distillation Column (RDC), a Recurrent Neural Network (RNN) based soft sensor has been proposed to estimate the bottoms product composition of the RDC for the synthesis of n-Butyl Acetate using esterification reaction. This soft sensor acts as a measuring element in a closed loop involving a PI controller for the direct control of the RDC’s product concentration. The RNN acts as a dynamic network, which works on the sequential input data and output data with a recurrent connection. While using the RNN based soft sensor in the open loop, it has been observed that the sensor estimated the composition of butyl acetate in the bottoms with such an accuracy that it can be used for the control purpose. Closed loop results demonstrated that the system has been showing precise controlled results and soft sensor is showing small prediction Mean Square Error (MSE) when disturbances in feed flow rate and set point changes are introduced.