Machine learning algorithm for mapping computational data of water reservoir with air bubble flow column reactor

Abstract

Analysis of CO₂ absorption by water-based bubble column reactors is of great importance and computational methods help understand the process and improve its efficiency. Numerical evaluation of CO₂ absorption using water in a bubble column was investigated by analysis of mass transfer in the process. The results showed that the CO₂ absorption in water was increased from 0 to around 0.53 L after 450 s and the rate of CO₂ absorption in water was decreased from 0.28 L/min to around 0 after 450 s. Then, the obtained results from the model were used for understanding these parameters in controlled environments using machine learning methodologies. We explored the predictive accuracy of regression models to estimate the concentration of CO₂ (mol/m³) across spatial (z) and temporal (t) dimensions in a controlled environment. The dataset comprises measurements collected over 451 s at varying depths, structured as a regression task to model CO₂ based on t(s) and z(m). Data preprocessing involved Z-score normalization and Isolation Forest-based outlier detection, optimizing data integrity. The methodology incorporated the Whale Optimization Algorithm (WOA) to refine model hyperparameters, enhancing performance metrics across Decision Tree (DT), K-Nearest Neighbors (KNN), and Multilayer Perceptron (MLP) models. Evaluation metrics such as R2, RMSE, and MAE indicated KNN’s superior predictive capability, demonstrating strong generalization across training, cross-validation, and testing phases. The KNN model accurately captured the non-linear spatial–temporal relationships inherent in the dataset, achieving a near-perfect R2 of 0.9991 on the training set and 0.9979 on the test set, with low RMSE (0.291) and MAE (0.042) values on the test data. These results underscore the model’s high precision in predicting concentration levels across varying depths and time, supporting its potential for applications requiring precise concentration estimations in similar contexts.