Development of support vector regression (SVR)-based model for prediction of circulation rate in a vertical tube thermosiphon reboiler

Abstract

The hydrodynamics and heat transfer in a thermosiphon reboiler interact with each other making the process very complex. Prediction of the rates of heat transfer and thermally induced flow are the primary requirements for the design of thermosiphon reboilers. The objective of this study was to develop, for the first time, a unified data-driven model, for the prediction of circulation rate in a thermosiphon reboiler for different pure components with wide variation in thermo-physical properties and operating parameters, using support vector regression (SVR)-based modeling technique. In the present work, 148 experimental data points from accessible sources, including the author's own study were used. First, a multiple regression (MR) model for circulation rate (in the form of Reynolds number) was developed as a function of dimensionless parameters namely, Peclet number for boiling ( Pe b ), Subcooling number ( K sub ), and the Lockhart–Martinelli parameter ( X tt ), followed by the formulation of an SVR-based model. Statistical analysis revealed that the proposed generalized SVR-based model had high prediction accuracy with an average absolute relative error (AARE) of 3.82%, root mean square error (RMSE) of 0.0717, leave-one-out cross validation ( Q 2 LOO ) of 0.9975 and mean relative error (MRE) of 0.0288 on the training data. Corresponding values of 6.11% AARE, 0.0816 RMSE, 0.9991 leave-one-out cross validation on test data ( Q 2 ext ) and 0.0541 MRE were obtained for the test data. A comparison of the SVR-based correlation was made with the MR model and with some selected empirical correlations in the literature. It was observed that the proposed SVR-based model significantly exhibited an enhanced prediction and generalization performance.