A big data-handling machine learning model for membrane-based absorber reactors in sorption heat transformers

Abstract

Membrane-based absorbers have received much attention recently due to their higher absorption rates than conventional absorbers. Several studies have been conducted to analyze membrane-based absorbers’ performance using numerical and analytical modeling. Numerical models are accurate; however, they are computationally expensive. Analytical models are computationally efficient; however, they have limiting assumptions and may exhibit inaccuracy accordingly. This study proposes a novel multi-label, big data-handling machine learning model for membrane-based absorbers used in sorption heat transformers, offering the accuracy of a numerical model with the efficient computation of an analytical model. A comprehensive dataset comprising over one million data points is generated using 2D numerical modeling. The dataset consists of 15 features, input parameters, including operating conditions and geometrical parameters, and four labels, output variables, including absorption rate, outlet concentration, solution outlet temperature, and heat transfer fluid outlet temperature. The Support Vector Regression, Random Forest Regression, and Decision Tree Regression are used and combined to develop the present model. Particle swarm optimization is used to find the optimized hyper-parameters of each model. A map-reduce algorithm is developed to minimize the computational time, and the optimized data chunk number is presented for the highest accuracy and the lowest computational time. The results of the proposed model are validated with experimental data, capturing most data within a relative difference of 15%. The machine learning-based model can predict the four outputs with an accuracy of over 90%. Moreover, it is shown that using the map-reduce algorithm results in a 40-fold decrease in computation time without significantly compromising accuracy.