基于多元经验模态分解与卷积神经网络的气液两相流流型识别

Flow Pattern Recognition Method of Gas-Liquid Two-Phase Flow Based on Multiple Empirical Mode Decomposition and Convolution Neural Network

提出了一种基于多元经验模态分解(MEMD)与卷积神经网络(CNN)的垂直管道气液两相流流型识别方法。该方法基于数字化电阻层析成像(ERT)系统采集的测量数据,预处理后进行MEMD分析,通过求取各分量与原始信号的皮尔逊相关系数选取本征模函数(IMFs)并求解Hilbert边际谱,提取Hilbert边际谱的标准差与均值作为卷积神经网络(CNN)输入以识别流型。结果表明,该方法能够有效识别泡状流、弹状流、段塞流,平均识别准确率可达96.43%。

A flow pattern identification method of gas-liquid two-phase flow in vertical pipeline based on multiple empirical mode decomposition(MEMD)and convolution neural network(CNN)is proposed.Based on the measurement data collected by the digital electrical resistance tomography(ERT)system,MEMD analysis is carried out after preprocessing.By calculating the Pearson correlation coefficient between each component and the original signal,the eigenmode function(IMFs)is selected and the Hilbert marginal spectrum is solved.The standard deviation and mean value of Hilbert marginal spectrum are extracted as convolution neural network(CNN)input to identify the flow pattern.The results show that the method can effectively identify bubbly flow,slug flow and slug flow,and the average recognition accuracy can reach 96.43%.