Machine learning based system identification of a realistic heat integrated distillation column using particle swarm optimization

Abstract

In the petrochemical industry, heat integration is one of the most commonly used methods for saving energy in the distillation column. The component of the liquid mixture can be estimated by correctly modeling or identifying the respective heat integrated distillation column (HIDC). Previously, system identification for HIDC has been performed using steady state data in an open loop configuration. However, the data obtained from steady state does not characterize realistic HIDC. Complex nonlinear modeling is required to account for the nonlinearities present in HIDC. In this study, a unique approach is presented to identify a dynamic closed-loop HIDC system using Support Vector Regression (SVR). Here, HYSYS is used to simulate HIDC in a fairly realistic environment and the resulting data is used in MATLAB for system identification. Particle Swarm Optimization (PSO) is used to optimize the SVR parameters, which gives better results than conventional metaheuristic optimization approaches. The proposed approach is also verified by accurately estimating the mole fraction of benzene-toluene mixtures in a dynamic HIDC. In terms of root mean square error (RMSE) and regression coefficients (R), the proposed strategy outperforms artificial neural networks (ANN) and radial basis function (RBF) models. The proposed technique reduces the RMSE by 80% and 47% compared to ANN for the mole fractions of benzene and toluene, respectively.