摘要
对垂直上升管中油水两相流流动测得的电导波动信号,从时频域内提取了11个反映油水两相流流动特性的特征量.频域内特征量提取采用了语音信号处理中的线性预测方法,时域内特征量提取采用了时间序列统计分析方法.将这些特征量作为人工神经网络的输入量,在总流量10～60 m3·d-1及含水率51%～91%范围内,采用基于Levenbery-Marquardt算法的BP人工神经网络作为相含率预估模型,较好地实现了油水两相流含水率预测,为两相流相含率测量提供了一种新的软测量途径.
For the conductance fluctuating signals measured from oil/water two-phase upward flow in a vertical pipe, 11 feature selections reflecting flow characteristics of oil/water two phase flow were extracted. The feature selections in frequency domain were derived by using the linear prediction method of speech signal processing and the feature selections in time domain were derived by the time series statistical analysis. All extracted features were taken as the input of artificial neural network, at the total flow rate ranging from 10 m^3·d^-1 to 60 m^3·d^-1 and water cut from 51% to 91%, the prediction model of artificial neural network based on the Levenbery-Marquardt algorithm was used and good water cut prediction results of oil/water two-phase flow were obtained. This study provided a new way to measure the phase volume fraction of two phase flow by soft sensor.
出处
《化工学报》
EI
CAS
CSCD
北大核心
2005年第10期1875-1879,共5页
CIESC Journal
基金
国家自然科学基金项目(60374041)
教育部留学回国人员科研启动基金项目~~
关键词
油水两相流
相含率
特征提取
软测量
oil/water two-phase flow
phase volume fraction
feature extraction
soft measurement
