[关键词]
[摘要]
针对海洋油气采输工业中集输-立管内气液两相流流量测量问题,在大型集输-立管实验环路上采集立管底部、顶部压力波动信号,提取其绝对值均值、绝对值方差、偏态系数、峰态系数,结合本征模函数(IMF)的高、中、低3个频段上的能量分数,构建了一个包含7个特征参数的BP神经网络测量模型。设计了集成化网络结构,在宽广的流型范围内,以最小均方误差算法(LMS)为基础,引入动量因子α和学习率自适应调节进行算法优化。集成网络预测的气相平均相对误差EMR和均方根相对误差ERMS分别为4.67%、4.91%,液相分别为5.83%、5.87%。
[Key word]
[Abstract]
To solve the problem of gas-liquid two phase flow metering in pipeline riser system,pressure signals at bottom and top of the riser were measured in large pipeline-riser experimental loop.The Mean of absolute value,variance of absolute value,skewness and kurtosis of pressure signals were calculated as well as the energy fractions of intrinsic mode function(IMF)of high,medium,and low frequency bands to construct a neural network including seven feature vector parameters.Based on least mean square algorithm(LMS),an integrated neural network was designed,which was optimized by momentum factor and an adaptive adjustment learning rate in a wide range of flow regime.The mean relative error(EMR) and the standard error(ERMS) of the integrated network for gas flow rate were 4.67% and 4.91%,respectively.For liquid flow,they were 5.83% and 5.87%,respectively.
[中图分类号]
TK313
[基金项目]
国家自然科学基金(51527808)