Various local deformations inevitably occur during the construction and operation stages of long-distance oil and gas pipelines, which affect their safe operation. IMU internal detection technology is the main method for detecting local deformations in pipelines. This paper establishes an important database containing 33,177 sample pipe sections based on the IMU internal detection data of six rounds of 914 km China-Russia crude oil pipeline with a total of 5023 km. The IMU data characteristics of four typical buried pipeline local deformations, including elbow, dent, curved deformation, and abnormal girth weld are analyzed. A preprocessing method for IMU data based on wavelet denoising is proposed, and a hybrid deep learning model is developed for accurate classification. A systematic multi-index comparison is carried out against mature machine learning models. The application results show that the classification accuracy of the hybrid deep learning model based on the database is close to 96 %, and the classification efficiency is 0.02 min/km. Establishing the long mileage database and accurate identification of local deformed pipe sections provide an effective technical means for the safe operation of oil and gas pipelines.

中文翻译：

---

一种基于 IMU 检测数据和混合深度学习模型的新型局部变形管段识别方法


长输油气管道在建设和运行阶段不可避免地会发生各种局部变形，影响其安全运行。IMU 内部检测技术是检测管道局部变形的主要方法。本文基于六轮 914 公里中俄原油管道总长 5023 公里的 IMU 内部探测数据，建立了一个包含 33,177 个样本管段的重要数据库。分析了 4 种典型埋地管道局部变形的 IMU 数据特征，包括弯头、凹痕、弯曲变形和异常环焊缝。该文提出了一种基于小波去噪的IMU数据预处理方法，并开发了一种用于精确分类的混合深度学习模型。与成熟的机器学习模型进行了系统的多指数比较。应用结果表明，基于数据库的混合深度学习模型的分类准确率接近 96 %，分类效率为 0.02 min/km。建立长里程数据库，准确识别局部变形管段，为油气管道的安全运行提供了有效的技术手段。