基于1D-CNN-AdaBoost及电阻层析成像的两相流流型辨识  

Identification of Two-phase Flow Pattern Based on 1D-CNN-AdaBoost and Electrical Resistance Tomography

作　　者：张立峰[1] 肖凯 华回春 ZHANG Li-feng;XIAO Kai;HUA Hui-chun(Department of Automation,North China Electric Power University,Baoding,Hebei 071003,China;Department of Mathematics and Physics,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 [2]华北电力大学数理系,河北保定071003

出　　处：《计量学报》2022年第12期1622-1626,共5页Acta Metrologica Sinica

基　　金：国家自然科学基金(61973115)。

摘　　要：基于电阻层析成像(ERT)系统采集的垂直管气液两相流测量数据,将一维卷积神经网络(1D-CNN)与AdaBoost(Adaptive Boosting)相结合,构建1D-CNN-AdaBoost算法,进行了气液两相流流型辩识研究。该算法使用5个1D-CNN作为弱分类器,通过实验数据样本训练,结合AdaBoost形成最终的强分类器。将1D-CNN-AdaBoost算法与BP神经网络、支持向量机及决策树算法进行比较,结果表明,1D-CNN-AdaBoost算法辨识正确率高于其他算法,平均辨识正确率可达97%。Based on the vertical tube gas-liquid two-phase flow measurement data collected by the electrical resistance tomography(ERT)system,the one-dimensional convolutional neural network(1D-CNN)and AdaBoost(Adaptive Boosting)are combined to construct the 1D-CNN-AdaBoost algorithm,a study on the flow pattern identification of gas-liquid two-phase flow has been carried out.The algorithm uses five 1D-CNNs as weak classifiers,trained on experimental data samples,and combined with AdaBoost to form the final strong classifier.Comparing 1D-CNN-AdaBoost algorithm with BP neural network,support vector machine and decision tree algorithm,the results show that the recognition accuracy of 1D-CNN-AdaBoost algorithm is higher than other algorithms,and the average recognition accuracy can reach 97%.

关 键 词：计量学 流型辨识 电阻层析成像 卷积神经网络 自适应提升 

分 类 号：TB937[一般工业技术—计量学]