An FPM-DM hybrid model for yield prediction of gas–liquid micro-sulfonation

Abstract

The industrial application of gas–liquid micro-sulfonation is hindered by the lack of prediction models for product yield that balance mechanistic interpretability with accuracy. This work proposed a novel hybrid model for predicting the yield of Linear Alkylbenzene (LAB) sulfonation products in microreactors, achieving both accuracy and interpretability. Recognizing the direct relationship between the theoretical molar ratio of reactants and product yield, we first developed a mass transfer empirical equation tailored for the annular flow pattern in microreactors. This equation, derived from surface renewal theory and validated through numerical simulations and SO₃ absorption experiments, exhibited an average error of 11.48 %. To account for reaction kinetics not captured by the mass transfer model, an ensemble model consisting of 3D-CNN and 3D-DenseNet-SE was established. This ensemble model learned the spatiotemporal characteristics of the reaction process, effectively bridging the gap between mass transfer efficiency and product yield. The resulting parallel hybrid model, termed FPM-DM, demonstrated high accuracy in predicting product yield (test set: average MSE = 0.206, average R² = 0.983), paving the way for its application in industrial gas–liquid micro-sulfonation processes.