Hybrid modeling approach for terpolymerization reactions in CSTR

Abstract

In order to produce terpolymer of desired quality a model capable of simulating terpolymerization is required. Terpolymerization involves complex reactions, full-scale modeling using the frst principle model is not practical to simulate reaction because there are many parameters to be estimated. In this study, a hybrid model that integrates the frst-principles model and the DNN model is proposed. The proposed hybrid model reduces the parameters that need to be estimated using a cumulative composition model, through the steady-state assumption. Afterward, DNN model in a hybrid model estimates the conversion using measurement data from process sensors, and the terpolymer composition according to conversion is calculated. In the process, by estimating model parameters with error in variables model, hybrid model specifc to the system is constructed. Validation of the hybrid model is performed using measurement data of 600 days and the result shows a good agreement with the actual data. The proposed hybrid model has high fdelity, scalability and robustness to other terpolymerization process.