基于IHBA-VMD和多特征提取的不同管道工况识别技术  

Recognition Technology for Different Pipeline Working Conditions Based on IHBA-VMD and Multi-feature Extraction

作　　者：郑颢 ZHENG Hao(Process Research Institute of No.3 Oil Production Plant,Daqing Oilfield Co.,Ltd.)

机构地区：[1]大庆油田有限责任公司第三采油厂工艺研究所,黑龙江省大庆市163000

出　　处：《油气田地面工程》2025年第3期44-50,共7页Oil-Gas Field Surface Engineering

摘　　要：管道工况信号具有非线性、非平稳、大尺度特性,如何从信号中提取和识别有效信息已成为完整性管理的重要环节。基于此,提出一种结合IHBA-VMD和多特征提取的管道工况识别技术。采用混沌策略和非线性密度因子对蜜獾算法(HBA)进行改进,利用改进蜜獾算法(IHBA)寻找VMD的最佳分解模态个数和惩罚因子;根据豪斯多夫距离的增量水平判断有效模态和噪声模态的分界点,完成信号重构;提取与时频域、熵值和波形相关的特征向量,采用概率神经网络(PNN)实现工况的自适应识别。研究结果表明:与其余信号降噪方法相比,IHBA-VMD算法的重构信号信噪比最大,均方误差最小,最佳分解模态个数为6,惩罚因子为1046;单变量无法提取到负压波信号的全部特征,分类效果不佳;基于均值、标准差、云模型特征熵和裕度因子等多特征变量在管道工况上具有良好的区分性,PNN模型可实现不同工况的分类,训练阶段和预测阶段的分类准确率分别可达97.50%、96.25%。该研究结果可为油气田管道安全管理水平的提升提供理论依据和实际参考。Pipeline working condition signal has nonlinear,non-stationary,and large-scale characteristics.How to extract and identify effective information from the signal has become an important part of integrity management.Based on this,a pipeline condition recognition technology combining IHBA-VMD and multi-feature extraction is proposed.Firstly,chaos strategy and nonlinear density factor are used to improve the Honey Badger algorithm(HBA),and IHBA algorithm is used to find the best decomposition mode number and penalty factor of VMD.Then,according to the increment level of the Hausdorff distance,the boundary point between the effective mode and the noise mode is determined to complete the signal reconstruction.Finally,feature vectors related to time-frequency domain,entropy,and waveform are extracted,and the probabilistic neural network(PNN)is used to realize adaptive recognition of working conditions.The results show that compared with other denoising methods,the IHBA-VMD algorithm has the largest signal-to-noise ratio,the smallest mean square error,the best decomposition mode number is 6,and the penalty factor is 1046.The single variable cannot extract all the characteristics of the negative pressure wave signal,and the classification effect is not good.Based on multiple-feature variables such as mean value,standard deviation,cloud model feature entropy,and margin factor,the PNN model has good discriminability in pipeline working conditions.It can achieve the classification of different working conditions,and the classification accuracy in the training and prediction stages can reach 97.50%and 96.25%,respectively.The research results can provide theoretical basis and practical reference for improving the safety management level of oil and gas pipeline.

关 键 词：IHBA VMD 豪斯多夫距离 工况识别 PNN 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]