System identification and control of heat integrated distillation column using artificial bee colony based support vector regression

• Published in: Chemical engineering communications Vol. 209; no. 10; pp. 1377 - 1396
• Main Authors: Jaleel, E. Abdul, Anzar, S. M., Rehannara Beegum, T., Mohamed Shahid, P. A.
• Format: Journal Article
• Language: English
• Published: Philadelphia Taylor & Francis 03.10.2022; Taylor & Francis Ltd
• Subjects: Algorithms; Artificial bee colony algorithm; Artificial neural networks; Benzene; control; Control systems; Distillation; First principles; fluid separation; heat integrated distillation column; Heuristic methods; Hydrocarbons; Mixtures; Modelling; Nonlinearity; Parameter identification; Proportional integral derivative; Regression coefficients; Root-mean-square errors; Support vector machines; support vector regression; Swarm intelligence; System identification; Toluene
• ISSN: 0098-6445; 1563-5201
• DOI: 10.1080/00986445.2021.1974409

Distillation is a high-energy process widely employed in separating fluid mixtures in the oil and gas industries. Heat integration is one of the practical approaches for energy saving in the distillation columns. Proper identification or modeling of heat-integrated distillation column (HIDC) is employed to predict the composition of fluid mixtures. The nonlinear modeling of HIDC is highly challenging, and methods based on the first principles are not effective in coping with the nonlinearities. Hence, a novel, non-parametric support vector regression (SVR) approach is proposed for system identification and control of HIDC in this work. SVR parameters were optimized using artificial bee colony (ABC) algorithm, which resulted in better performance over other meta-heuristic algorithms. Moreover, the SVR model demonstrated better performance than the artificial neural network models in root mean square error (RMSE) and regression coefficient (R). RMSE and R values for ABC-SVR were found to be 0.0010 and 0.99992, respectively, with the validation dataset. The performance of the SVR and PID controllers are also compared. Integral square error (ISE), integral average error (IAE), integral time square error (ITSE), and integral time average error (ITAE) are the comparison metrics employed, which yielded minimal values of and respectively, for the SVR controller. The proposed model outperforms all other related methods, and it can be used to predict the mole fraction of Benzene in Benzene-Toluene HIDC accurately.