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

作　　者：路敬祎[1,2,4] 李禹琦 褚丽鑫 宋南南 胡仲瑞 LU Jingyi;LI Yuqi;CHU Lixin;SONG Nannan;HU Zhongrui(Sanya Offshore Oil&Gas Research Institute,Northeast Petroleum University,Sanya572024,China;Artificial Intelligence Energy Research Institute,Northeast Petroleum University,Daqing163318,China;School of Electrical Information Engineering,Northeast Petroleum University,Daqing163318,China;Key Laboratory of Networking and Intellectual Control system in Heilongjiang Province,Daqing163318,China)

机构地区：[1]东北石油大学三亚海洋油气研究院,海南三亚572024 [2]东北石油大学人工智能能源研究院,黑龙江大庆163318 [3]东北石油大学电气信息工程学院,黑龙江大庆163318 [4]黑龙江省网络化与智能控制重点实验室,黑龙江大庆163318

出　　处：《压力容器》2022年第11期69-77,共9页Pressure Vessel Technology

基　　金：国家自然科学基金项目(61873058,62103096);海南省科技专项(ZDYF2022SHFZ105);黑龙江省自然科学基金项目(LH2020F005)。

摘　　要：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.

关 键 词：变分模态分解 多特征融合 概率神经网络 管道泄漏检测 

分 类 号：TH49[机械工程—机械制造及自动化] TE973.6[石油与天然气工程—石油机械设备]