用神经网络分析两相流横掠三角形柱体的信号

Unstable vortex signal sended from triangle pole in gas-liquid two-phase cross flow using neural network

对气液两相横掠柱体旋涡脱落信号数据进行了三层小波分解,并提取第三层小波分解的8个重构信号的能量作为人工神经网络的输入特征向量,训练了人工BP神经网络,并应用该网络实现了气液两相绕流中含气率大小的定位.解决了用传统的傅里叶变换进行信号分析时所不能解决的问题,即当含气率过大时,无法从功率谱图上确定含气率的大小.试验范围内最大可确定的截面含气率可达35.9%.研究表明,小波分解的特征提取技术和人工神经网络的模式识别技术联合应用可以作为测量两相流组分的一种新方法,也可作为对大含气率气液两相流横掠柱体旋涡脱落的分析方法.

In this paper, we decomposed the signal of the vortex sended from a triangle pole in gas-liquid twophase cross flow to the 3^rd layer by wavelet function, and extracted the energy of 8 recomposed signals. Afterward, we designed a BP neural network and trained it by the power feature vector of the 8 recomposed signals, and then used the network to identify the void fraction in the gas-liquid two-phase flow. The experiment shows that the technique of feature extraction based on wavelet decomposition and the technique of pattern recognition based on the artificial neural network can be combined as a new method that can be used to identify the components in gas-liquid two-phase cross flow, or that can be used to analyse the signal of a vortex sended from a pole in gas-liquid two-phase cross flow.