BackgroundInterfacial area concentration (IAC) is a key parameter of the interface transfer term in the closed two-fluid model of two-phase flow, which characterizes the strength of the gas-liquid interface transport capacity. There are usually some methods for modeling and predicting the interface area concentration, such as empirical correlation formula and interface area transport equation, but these methods have large data dependence.PurposeThis study aims to provide direction for model revision and improve the prediction accuracy of IAC by adding interpretability to the neural network model.MethodsThe prediction model of IAC based on a neural network was firstly established for better prediction of IAC with two-phase flow. Then, different bubble behavior, physical relationships, and statistical distribution were combined, and the predictive ability of the neural network model with different input feature combinations was compared and analyzed by the post-interpretability method. Finally, based on the structure parameter size of each layer of the neural network, the appropriate data preprocessing method was selected by analyzing the output proportion.ResultsThe post explanatory analysis show that the maximum prediction accuracy of the neural network reaches 95.62% when the inputs of the neural network are the gas superficial velocity (jg), liquid superficial velocity (jf), and void fraction (α). The void fraction is an important factor in IAC prediction, and logarithmic transformation preprocessing of training data can significantly improve the model's predictive ability for real data.ConclusionsThe results of this study provide reference for future interpretability research on interface area concentration.