基于卷积胶囊网络的油套管变形程度评估方法

Eestimation Method of Oil Pipeline Casing Deformation Degree Based on Convolution Capsule Network

针对当前油气田管道电磁检测中存在的管道形变量化方法单一、精度不足的问题,引入深度学习的思路,提出一种基于卷积胶囊网络的油套管变形程度估计方法,通过设计多个卷积层对不同探头涡流信号进行特征提取;设计胶囊网络的输出层,构建基于模长的约束函数,对最小臂值进行量化,实现对油套管形变程度的估计。本方法考虑了不同类型探头信息之间的联系,并构建量化模型,以提高模型非线性映射能力,适用于多个探头同时检测的设备。通过对实际井下管道的脉冲涡流检测数据进行验证,相比于常见的深度学习方法,本方法具有更好的量化精度。

At present,there are some problems in the electromagnetic detection of oil and gas field pipeline,such as single quantitative method of pipeline deformation and insufficient accuracy.Aiming at the issue,this paper introduced the idea of deep learning and proposed a quantitative estimation method for the deformation degree of oil pipeline casing based on convolution capsule network.Firstly,several convolutions were designed to extract the characteristics of eddy current signals from different probes.Then the output layer based on capsule network was designed to construct the constraint function based on module length to quantify the minimum arm value.Finally,the degree of casing deformation was estimated.This method considered the relationship between different probes to build a quantitative model,which improved the nonlinear mapping ability of the model,and was suitable for equipment with multiple probes simultaneous detection.Through the verification of the actual downhole pipeline pulse eddy current testing data,this method had better quantization accuracy than the common methods.