基于VMD和多特征融合的管道信号特征提取方法

Feature extraction method of pipeline signal based on VMD and multi-feature fusion

针对管道泄漏检测过程中泄漏特征信息提取困难、泄漏检测准确率低的问题,提出了基于VMD和多特征融合的特征提取方法。首先利用变分模态分解(Variational mode decomposition, VMD)对采集的实验室管道信号进行分解,得到若干个IMFs,利用提出的WCC算法计算相邻模态之间的相似度来确定VMD分解的模态个数;根据有效模态分量与原始信号的相似程度确定特征分量,然后提取特征分量的有效特征参数,构建成基于多特征融合的特征向量组;最后,将特征向量输入到概率神经网络(Probabilistic neural network, PNN)进行工况识别;实验结果表明,与单一特征构成的特征向量相比,本文提出的多特征融合的特征提取方法能够有效地识别出不同的工况信号。

For the problems of difficulty in leakage feature extraction in the process of pipeline leak detection and low accuracy of leak detection, a feature extraction method based on VMD and multi-feature fusion was proposed.Firstly, the collected laboratory pipeline signals were decomposed by using variational modal decomposition(VMD),and several IMFs were obtained.The proposed WCC algorithm was used to calculate the similarity between adjacent modes to determine the number of modes decomposed by VMD;The feature component was determined according to the similarity between the effective modal component and the original signal, and then the effective feature parameters of the feature component were extracted to construct the feature vector group based on multi-feature fusion;Finally, the feature vector was input into the probabilistic neural network(PNN) for condition recognition;The experimental results show that compared with the feature vector composed of single feature, the feature extraction method for multi-feature fusion proposed herein can effectively identify different working condition signals.