Machine learning-assisted correlations of heat/mass transfer and pressure drop of microchannel membrane-based desorber/absorber for compact absorption cycles

Abstract

Microchannel membrane-based desorbers and absorbers are key components in efficient and compact absorption refrigeration systems. To improve the accuracy of current empirical correlations for describing the heat/mass transfer and solution pressure drop characteristics, more advanced correlation models are urgently needed. This study applies three machine learning algorithms, namely Random Forest (RF), Least-Squares Support Vector Machine (LS-SVM), and Genetic Algorithm-optimized Back Propagation Artificial Neural Network (GABPNN), to develop new models for Nusselt number (Nu), Sherwood number (Sh), and friction factor (Fr) of microchannel membrane-based desorber and absorber, respectively, based on experimental results. These machine learning-assisted models effectively improve the prediction accuracy for both the desorber and absorber. Among them, the Random Forest model performs best, improving the prediction accuracy by 12.37% (Nu), 21.49% (Sh), and 14.24% (Fr) for the desorber and 24.28% (Nu), 27.82% (Sh), and 30.47% (Fr) for the absorber, compared to the conventional empirical correlations. Moreover, in relative to the correlations obtained from the literature, the Random Forest model predicts the overall heat transfer coefficient (U), sorption rate (J), and solution pressure drop (DP) with accuracies higher by 44.75%, 64.66%, and 83.52% for the desorber, and 79.66%, 81.52%, and 69.53% for the absorber. This study is expected to facilitate the simulation, design, and optimization of microchannel membrane-based desorbers/absorbers towards more efficient and compact absorption cycles.