Abstract: Flow pattern is a very important hydrodynamic state parameter in the study of flow process, and the traditional flow pattern identification method generally adopts observation methods, which often require harsh experimental observation conditions, which is difficult to apply in engineering. The artificial neural network can establish a completely data-based model, so as to directly establish the connection between the state quantity of the flow pattern and the physical quantity (such as flow velocity) that is easier to measure and obtain in the project, and the purpose of observing the two-phase flow pattern in real time and analyzing the current hydrodynamic state in the complex two-phase flow can be realized in the project. Based on this, this paper proposed a data-based stream pattern recognition possibility. Using the black box characteristics of the artificial neural network, the experimental data of the flow pattern of the gas-liquid two-phase flow in the horizontal pipeline under different flow velocities measured in the experiment were selected, and the data were identified and iteratively trained and learned through the artificial neural network, so that the accuracy of the recognition was continuously improved until convergence through the learning and training of the data, so as to establish a model for flow pattern recognition based on the two-phase flow velocity, which could achieve a high accuracy of flow pattern prediction. The model can identify the flow pattern state of the two-phase flow at different flow velocities, and the effect is good, and the error rate can be maintained at a very low level. At the same time, the structure irrelevance analysis of the artificial neural network model was carried out, the influence of the number of neurons in the hidden layer of the neural network on the accuracy of the model was studied, and the minimum number of neurons in the hidden layer that met the accuracy requirements was obtained, so that the artificial neural network model could be optimized, and the optimal artificial neural network structure with high precision and no waste of computing power was designed.