参数优化CatBoost的输油管道缺陷识别方法  

Defectidentification method for oil pipelines based on parameter-optimized CatBoost

作　　者：吴倩玉 孙志刚[2] 王佳琦 李策[2] 吕冰泽 王国涛[1,2] WU Qianyu;SUN Zhigang;WANG Jiaqi;LI Ce;L Bingze;WANG Guotao(College of Electronic Engineering,Heilongjiang University,Harbin 150080,China;Electrical and Electronic Reliability Research Institute,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]黑龙江大学电子工程学院,哈尔滨150080 [2]哈尔滨工业大学电器与电子可靠性研究所,哈尔滨150001

出　　处：《黑龙江大学自然科学学报》2024年第5期613-622,共10页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(51607059);黑龙江省重点研发计划项目(2022ZX03A06);黑龙江省自然科学基金资助项目(QC2017059,JJ2020LH1310);黑龙江省省属高等学校基本科研业务费项目(HDRCCX-201604,2020-KYYWF-1006);黑龙江省教育厅科技成果培育项目(TSTAU-C2018016);七台河市科技计划项目(20308C);黑龙江大学研究生创新科研项目(YJSCX2021-067HLJU)。

摘　　要：针对传统信号处理技术仅能实现初步的输油管道缺陷检测,无法提供具体缺陷类型的识别问题,提出了一种结合CatBoost机器学习算法的输油管道缺陷识别方法。通过磁致伸缩导波检测系统采集两种缺陷类型的输油管道信号,并在原有的14个时域和频域特征基础上,提出特征变换方法构建了新的时频域特征,扩充了原始特征域的表示。经过特征筛选,选取了性能优异的3个特征加入到现有特征集中,形成了扩展的特征数据集。采用4种常见的基于集成学习的分类器进行训练,比较4种分类器精度,得到最优的CatBoost取得的分类精度是92.78%。通过随机搜索算法对CatBoost的5个参数进行优化后,其精度进一步提升至95.15%,比默认参数的CatBoost提升了2.37%。新采集的含有2种缺陷类型的输油管道测试数据集得到分类精度为94.74%,验证了所提方法的实用性、可行性和稳定性。The traditional signal processing techniques can only realize the preliminary oil pipeline defect detection problem and cannot provide specific defect type identification.An automatic identification method of oil pipeline defects combined with CatBoost machine learning algorithm is proposed.Specifically,the oil pipeline signals of two defect types are collected by magnetostrictive guided wave inspection system,and based on the original 14 time and frequency domain features,a feature transformation method is proposed to construct new time and frequency domain features,which expands the representation of the original feature domain.After feature screening,three features with excellent performance are selected and added to the existing feature set to form an extended feature dataset.Four common integrated learning-based classifiers are used for training,and the accuracy of the four classifiers is compared,thus obtaining the optimal CatBoost and achieving a classification accuracy of 92.78%.After optimizing the five parameters of CatBoost by random search algorithm,its accuracy is further improved to 95.15%,which is 2.37% higher than that of CatBoost with default parameters.The classification accuracy of 94.74% is obtained on a newly collected test dataset of oil pipeline containing two defect types,which verifies the practicality,feasibility and stability of the proposed method.

关 键 词：CatBoost 缺陷识别 磁致伸缩导波 特征工程 输油管道 

分 类 号：TN98[电子电信—信息与通信工程]