针对目前两相流流型识别率不高且通常依赖精密仪器获取流型特征等问题，提出一种基于深度神经网络的流型识别方法。通过文献报告中已收集的流型数据集，分析影响流型的关键变量，利用粒子群优化后的深度神经网络结合Softmax分类器在Tensorflow平台上进行训练，并将其分类结果与统一模型进行对比。结果表明：流型识别的最终综合识别准确率在97.44%；流型分类结果与流型统一模型基本一致；与目前流型识别的主流方法相比，具有特征易提取、神经网络模型收敛速度快等优点。

At present，the flow pattern recognition rate of twophase flow is not high，and it usually depends on precise instruments to obtain flow pattern characteristics.To solve these problems，a flow pattern recognition method based on deep neural network is proposed.Through the flow pattern data set collected in the literature reports，the key variables affecting the flow pattern are analyzed，the combination of the deep neural network optimized by particle swarm and the Softmax classifier is used for training on the Tensorflow platform，and its classification results and the unified model are compared.The experimental results show that the final comprehensive recognition accuracy of flow pattern recognition is 97.44%，and the classification result of flow pattern is basically consistent with the unified flow pattern model.Compared with the current mainstream methods of flow pattern recognition，it has the advantages of easy feature extraction and fast convergence speed of neural network model.

O359

国家自然科学基金（61976083）