基于多域特征提取的气液两相流流型识别  

Gas-Liquid Two-Phase Flow Pattern Recognition Based on Multi-domain Feature Extraction

作　　者：张立峰[1] 王智 ZHANG Li-feng;WANG Zhi(Department of Automation,North China Electric Power University,Baoding,Hebei 071003,China)

机构地区：[1]华北电力大学自动化系,河北保定071003

出　　处：《计量学报》2023年第10期1509-1516,共8页Acta Metrologica Sinica

基　　金：国家自然科学基金(61973115)。

摘　　要：针对气液两相流的准确识别问题提出了一种多域特征处理方案。利用电阻层析成像(ERT)系统获取垂直上升管道流动数据,从测量数据与截面电导率分布图像两方面分析,对高维测量数据降维处理后提取时域特征,同时提取线性反投影(LBP)算法重建图像空域特征,进一步对图像进行Walsh-Hadamard变换后提取列率域特征。使用统一流形逼近与投影(UMAP)算法对量化的多域特征降维处理,最后搭建支持向量机(SVM)实现流型识别。结果表明,该流型分类框架对泡状流、泡状-段塞过渡流型、段塞流及严重段塞流的分类准确率分别为98.1%、96.3%、95.2%、94.8%。A multi-domain feature processing scheme is proposed for the accurate identification of gas-liquid two-phase flow.The electrical resistance tomography(ERT)system is used to obtain the flow data of vertical rising pipeline.From the perspective of measurement data and cross-sectional conductivity distribution image,the time-domain features are extracted after dimensionality reduction of high-dimensional measurement data,and the spatial features of the reconstructed image are extracted by linear back projection(LBP)algorithm.Further,Walsh-Hadamard transform is performed on the image to extract column rate domain features.The uniform manifold approximation and projection(UMAP)algorithm is used to reduce the dimension of the quantized multi-domain features,and finally a support vector machine(SVM)is built to realize flow pattern recognition.The results show that the classification accuracy of bubble flow,bubble-slug transition flow,slug flow and severe slug flow are 96.2%,95.0%,93.5%and 95.8%,respectively.

关 键 词：计量学 流型识别 电阻层析成像 Walsh-Hadamard变换 统一流形逼近和投影 多域特征 气液两相流 

分 类 号：TB937[一般工业技术—计量学]