Intelligence predictions: enhancing bubble column efficiency with advanced machine learning (ML) approaches

In chemical, petrochemical, and biochemical sectors, bubble columns are essential for facilitating critical reactions like methanation, oxidation, and hydrogenation. Gas hold-up is one of the most important operational parameters that has a significant impact on the performance and efficiency of these multiphase reactions. In order to improve the design and operational efficiency of bubble columns, it is imperative that the gas hold-up be precisely estimated. Creating a sophisticated predictive model with machine learning techniques is a promising solution for obtaining accurate and dependable estimations because of the intricate relationships and the unpredictable impact of numerous factors on gas hold-up. Therefore, this study explores the efficiency of four deep learning models, including the Recurrent Neural Network, Convolutional Neural Network, Long Short-Term Memory, and Gated Recurrent Unit in gas hold-up estimation. This study utilizes the Harris Hawk Optimization algorithm for hyperparameter fine-tuning and enhancing the model’s performance. The optimized Recurrent Neural Network model achieves markable results, attaining a test correlation coefficient of 0.995, root mean square error of 0.005 alongside a mean absolute error of 0.0054. Additionally, this study demonstrate that liquid height is the most influential variable in controlling gas hold-up by sensitivity indices approaching 1, which suggests that even slight variations in liquid height can lead to significant changes in gas retention.