多尺度信息熵特征的气液二相流流型识别方法

Method for identifying gas-liquid two-phase flow patterns based on multi-scale information entropy feature

为研究垂直上升管中的气液二相流的流型,利用自制的多电导探针的测量系统采集了4种典型流型的电导波动信息。根据小波包变换能将电导波动信号按任意时频分辨率分解到不同频段的特性,对其进行了3层小波包分解后并计算了各个频段的信息熵特征向量,并作为特征参数输入到E lm an神经网络进行训练,实现了与神经网络相结合流型的智能识别。研究结果表明,该方法能够很准确地识别出4种流型,且提取特征比较方便,从而为流型识别提供了一种新的有效方法。

To study the flow regime of gas-liquid two-phase in vertical upward pipe, the conductance fluctuation information of four typical flow regimes was collected by measuring system with self-made multiple conductance probes. Since the wavelet packet transform has the characteristic of decomposing conductance fluctuation signals to any frequency bands, the collected conductance fluctuation signals were decomposed into three-layer wavelet packets, and the information entropy eigenvectors in various frequency range were calculated, then Elman neural network was trained using the eigenvectors as feature parameters and the flow regime intelligent identification was realized. The results show that such method can well identify the four flow regimes, and can abstract characters conveniently. Therefore, a kind of powerful approach is supplied for the identification of flow regimes.