
一维卷积长短期记忆神经网络的管道泄漏检测方法
聂维, 江竹, 刘伯相, 林豪, 冉义建
一维卷积长短期记忆神经网络的管道泄漏检测方法
One-dimensional Convolutional and Long Short-term Memory Neural Network Method for Pipeline Leak Detection
针对以数据为驱动的管道泄漏检测方法,未能有效同步利用泄漏信号的空间和时序特征的问题,提出一种基于一维卷积神经网络(1D-CNN)和长短期记忆网络(LSTM)相结合的管道泄漏识别方法。该网络模型以去噪处理后的管道压力信号为输入源,先后使用1D-CNN、LSTM提取其空间特征和时间维度特征,利用已提取的空时两种不同维度的特征建立压力信号与管道状况的对应关系,进而实现对管道泄漏的检测。对比分析实验结果表明,1D-CNN-LSTM方法提取的特征参数更具有效性与可靠性,管道泄漏的检测精准度显著提升。
To address the problem that the data-driven pipeline leak detection method fails to effectively utilize the spatial and temporal characteristics of the leak signal simultaneously, a pipeline leak identification method based on a combination of one-dimensional convolutional neural network (1D-CNN) and long short-term memory network (LSTM) is proposed. The network model takes the de-noised pipeline pressure signal as the input source, the spatial and temporal dimensional features are extracted by using 1D-CNN and LSTM successively and the correspondence is established between the pressure signal and the pipeline condition by using the spatio-temporal features extracted in two different dimensions, thus realizing the detection of pipeline leaks. The experimental results show that the features extracted by the 1D-CNN-LSTM method are more effective and reliable, and the accuracy of pipeline leak detection is significantly improved.
管道泄漏 / 检测 / 压力信号 / 卷积神经网络 / 长短期记忆网络 {{custom_keyword}} /
pipeline leaks / detection / pressure signal / convolutional neural network / long short-term memory network {{custom_keyword}} /
表1 神经网络参数Tab.1 Neural network parameters |
参数 | 取值 | 参数 | 取值 |
---|---|---|---|
卷积核尺寸及步长 | 1×3、1×1 | LSTM层数 | 1 |
卷积核1深度 | 8 | LSTM隐藏层节点数 | 64 |
卷积核2深度 | 16 | 全连接层数 | 2 |
卷积核3深度 | 32 | 训练集比例/% | 80 |
卷积核4深度 | 32 | Batch_size | 200 |
池化层尺寸及步长 | 1×2、1×1 | 最大迭代数 | 800 |
学习率 | 初始为0.001,每三轮减半 | 损失函数 | 交叉熵损失函数 |
图11 不同网络结构的准确率与损失变化曲线Fig.11 Accuracy and loss variation curves for different network structures |
表2 不同网络结构的训练结果对比Tab.2 Comparison of training results for different network structures |
序号 | 网络名称 | 训练集准确率/% | 训练集损失 | 测试集准确率/% | 训练时间/min |
---|---|---|---|---|---|
1 | 1D-CNN-LSTM (3C1L) | 98.50 | 0.043 0 | 97.50 | 62.7 |
2 | 1D-CNN-LSTM (4C1L) | 100.00 | 0.009 7 | 99.75 | 125.7 |
3 | 1D-CNN-LSTM (5C1L) | 99.33 | 0.008 4 | 98.92 | 172.0 |
4 | 1D-CNN-LSTM (4C2L) | 99.50 | 0.034 6 | 99.33 | 122.7 |
5 | 1D-CNN | 99.00 | 0.018 2 | 99.42 | 14.0 |
6 | LSTM | 96.00 | 0.111 5 | 96.42 | 655.0 |
表3 各网络对于不同情况的检测准确度Tab.3 Detection accuracy of each network for different situations |
模型 对比 | 泄漏位置A | 泄漏位置B | 泄漏位置C | 泄漏位置D | 无泄漏 | ||||
---|---|---|---|---|---|---|---|---|---|
开度a | 开度b | 开度a | 开度b | 开度a | 开度b | 开度a | 开度b | ||
3C1L | 0.920 | 0.960 | 0.940 | 0.960 | 0.900 | 0.960 | 0.840 | 0.980 | 0.988 |
4C1L | 1.000 | 1.000 | 0.980 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 |
4C2L | 0.980 | 0.980 | 0.960 | 1.000 | 0.980 | 1.000 | 0.960 | 1.000 | 0.996 |
5C1L | 0.980 | 1.000 | 0.960 | 1.000 | 0.980 | 1.000 | 0.980 | 1.000 | 0.995 |
CNN | 0.980 | 0.960 | 0.980 | 0.980 | 0.940 | 1.000 | 0.960 | 0.980 | 0.993 |
LSTM | 0.820 | 0.900 | 0.780 | 0.920 | 0.840 | 0.960 | 0.760 | 0.980 | 0.976 |
1 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
2 |
郎宪明. 基于特征提取与信息融合的管道泄漏检测与定位研究[D]. 西安:西北工业大学, 2018.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
3 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
4 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
5 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
6 |
季舒瑶,袁飞,程恩, 等. 基于声传播特性的供水管道泄漏检测与分类[J]. 南京大学学报(自然科学), 2015,51():64-71.
增刊1
{{custom_citation.content}}
{{custom_citation.annotation}}
|
7 |
郎宪明,李平,曹江涛,等. 长输油气管道泄漏检测与定位技术研究进展[J]. 控制工程, 2018,25(4):621-629.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
8 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
9 |
全胜蓝,江衍铭. 基于神经网络模型的输沙过程推估[J].中国农村水利水电,2020(9):112-116.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
10 |
唐拥军,刘东,肖志怀,等. 基于卷积神经网络与奇异值分解的水电机组故障诊方法研究[J].中国农村水利水电, 2021(2):175-181.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
11 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
12 |
宁方立,韩鹏程,爽段,等. 基于改进CNN的阀门泄漏超声信号识别方法[J]. 北京邮电大学学报, 2020,43(3):38-44.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
13 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
14 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
15 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
16 |
曲建岭,余路,袁涛,等. 基于一维卷积神经网络的滚动轴承自适应故障诊断算法[J]. 仪器仪表学报, 2018,39(7):134-143.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
17 |
孙良,王建林,赵利强. 负压波法在液体管道上的可检测泄漏率分析[J]. 石油学报, 2010,31(4):654-658.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
18 |
唐鸣,雷晓辉,龙岩,等. 基于长短时记忆网络(LSTM)的南水北调中线水位预测[J]. 中国农村水利水电, 2020(10):189-193.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
19 |
王学渊. 基于压力信号辨识的城市供水管道泄漏诊断方法研究[D]. 北京:北京建筑大学, 2018.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
{{custom_ref.label}} |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
/
〈 |
|
〉 |