基于TPE-XGBoost模型的输气管道泄漏定位方法

doi:10.11949/0438-1157.20250296

摘要/Abstract

摘要：

精准定位天然气管道泄漏点对事故应急处置具有重要意义。针对现有泄漏定位方法依赖瞬时信号、易受管道运行噪声干扰、定位精度不足的缺陷，提出了基于TPE(tree-structured Parzen estimator)-XGBoost（eXtreme gradient boosting）模型的管道泄漏定位方法。通过采集持续200 s的压降速率时间序列数据构建训练集，对比了TPE、PSO（粒子群算法）、BOA（贝叶斯优化）和Optuna四种优化算法对XGBoost模型精度的影响；分析了TPE-XGBoost与SVM、CNN及CNN-LSTM-Attention模型的定位准确性；探究了数据时间窗（30~200 s）与叠加运行噪声对定位准确性的影响规律。研究结果表明：（1）经TPE优化后，XGBoost模型的性能显著提升，在测试集上决定系数R²达0.9835；（2）在泄漏定位任务中，TPE-XGBoost模型的定位偏差仅为3.77%，优于SVM、CNN及CNN-LSTM-Attention等对比模型；（3）当数据时间窗长度在30~200 s变化时，模型在测试集上的R²始终高于0.96；（4）在叠加管道运行噪声场景下，模型定位偏差为7.58%。

关键词: 天然气, 管道, 神经网络, 算法, 预测, 泄漏定位

Abstract:

Accurate localization of natural gas pipeline leakage points is of great significance for emergency response to accidents. In view of the defects of existing leakage location methods that rely on instantaneous signals, they are easily disturbed by pipeline operation noise, and have insufficient positioning accuracy, a pipeline leakage location method based on TPE (tree-structured Parzen estimator)-XGBoost (eXtreme gradient boosting) model is proposed. This method utilizes pressure drop rate data collected at 5-second intervals after leakage to train and predict the model. The hyperparameter optimization performance of TPE is compared with PSO (particle swarm optimization), BOA (Bayesian optimization algorithm), and Optuna methods, addressing the challenge of hyperparameter search in the model. Additionally, the prediction accuracies of the TPE-XGBoost model are compared with those of SVM (support vector machine), CNN (convolutional neural network), and CNN-LSTM-Attention models. The effects of dataset time series length and superimposed running noise on localization accuracy are also analyzed. The results demonstrate that the TPE optimization algorithm enhances the accuracy of the XGBoost model, outperforming PSO, BOA, and Optuna in hyperparameter optimization. Compared to other localization models, the TPE-XGBoost model achieves the highest prediction accuracy, with an R² value of 0.9835 and a localization error of only 3.77% on the test set. In contrast, the R² values for SVM, CNN, and CNN-LSTM-Attention models are 0.3684, 0.9285, and 0.9821, respectively. Analysis of the impact of time series length reveals that, within the data length range of 30 to 150 seconds, model decision coefficients increase with longer time series, and significant differences in model hyperparameters across time lengths are observed. When the experimental group with pipeline running noise is introduced, all models' accuracy decreases, while the model configuration adapts by adjusting hyperparameters to extract valid information through increased segmentation. In this noise-added experimental group, the R² of the XGBoost model is 0.9580 and the localization error is 7.58%.

Key words: natural gas, pipeline, neural networks, algorithm, prediction, leak localization

中图分类号:

TE 88

贾文龙, 陈浚哲, 李长俊, 刘勇, 谢玟. 基于TPE-XGBoost模型的输气管道泄漏定位方法[J]. 化工学报, 2025, 76(10): 5510-5521.

Wenlong JIA, Junzhe CHEN, Changjun LI, Yong LIU, Wen XIE. Leak localization method for gas pipeline based on TPE-XGBoost modeling[J]. CIESC Journal, 2025, 76(10): 5510-5521.

图/表 15

参考文献 30

[1]	孙良, 王建林. 基于泄漏瞬变模型的管道泄漏检测与定位方法[J]. 应用基础与工程科学学报, 2012, 20(1): 159-168.
	Sun L, Wang J L. Release transient modeling based leak detection and location method for pipelines[J]. Journal of Basic Science and Engineering, 2012, 20(1): 159-168.
[2]	孟令雅. 基于瞬态模型法的输气管道泄漏监测与定位技术[J]. 北京交通大学学报, 2008, 32(3): 73-77.
	Meng L Y. Research on leak detection and position for natural gas pipeline based on transient model method[J]. Journal of Beijing Jiaotong University, 2008, 32(3): 73-77.
[3]	王辰, 刘庆文, 陈典, 等. 基于分布式光纤声波传感的管道泄漏监测[J]. 光学学报, 2019, 39(10): 1006005.
	Wang C, Liu Q W, Chen D, et al. Monitoring pipeline leakage using fiber-optic distributed acoustic sensor[J]. Acta Optica Sinica, 2019, 39(10): 1006005.
[4]	付士崇, 张丹, 罗群. 基于光纤分布式声波传感的输气管道泄漏识别与定位方法[J]. 激光与光电子学进展, 2025, 62(3): 111-117.
	Fu S C, Zhang D, Luo Q. Gas pipeline leakage identification and localization method based on fiber-optic distributed acoustic sensing[J]. Laser & Optoelectronics Progress, 2025, 62(3): 111-117.
[5]	韩宝坤, 何景涛, 王金瑞, 等. 天然气管道泄漏声学检测及定位方法[J]. 噪声与振动控制, 2021, 41(5): 58-64, 146.
	Han B K, He J T, Wang J R, et al. Acoustic detection and location method for natural gas pipeline leakage[J]. Noise and Vibration Control, 2021, 41(5): 58-64, 146.
[6]	Yan Z, Cui X, Gao Y. Acoustic injection method based on weak echo signals for leak detection and localization in gas pipelines[J]. Applied Acoustics, 2023, 211: 109577.
[7]	杨红英, 华科, 叶昊, 等. 基于Hilbert-Huang变换的输气管道泄漏诊断方法[J]. 化工学报, 2011, 62(8): 2095-2100.
	Yang H Y, Hua K, Ye H, et al. Leak diagnosis of gas transport pipelines based on Hilbert-Huang transform[J]. CIESC Journal, 2011, 62(8): 2095-2100.
[8]	周芷怡, 尹冠雄, 王彬, 等. 考虑基准点: 修正点两阶段的天然气管道泄漏定位新方法[J]. 天然气工业, 2023, 43(5): 88-99.
	Zhou Z Y, Yin G X, Wang B, et al. A new two-stage natural gas pipeline leakage locating method considering base point and correction point[J]. Natural Gas Industry, 2023, 43(5): 88-99.
[9]	王春园. 可燃气体管道泄漏检测方法的研究[D]. 大连: 大连理工大学, 2019.
	Wang C Y. Research on method for leakage detection of flammable gas pipeline[D]. Dalian: Dalian University of Technology, 2019.
[10]	李玉星, 刘翠伟. 基于声波的输气管道泄漏监测技术研究进展[J]. 科学通报, 2017, 62(7): 650-658.
	Li Y X, Liu C W. Advances in leak detection and location based on acoustic wave for gas pipelines[J]. Chinese Science Bulletin, 2017, 62(7): 650-658.
[11]	贾文龙, 孙溢彬, 汤丁, 等. 基于支持向量机的输气管道泄漏压降信号智能识别方法[J]. 化工进展, 2022, 41(9): 4713-4722.
	Jia W L, Sun Y B, Tang D, et al. Intelligent recognition method for pressure drop signals of gas pipeline leakage based on support vector machine[J]. Chemical Industry and Engineering Progress, 2022, 41(9): 4713-4722.
[12]	Zhao J, Li J, Bai Y, et al. Research on leakage detection technology of natural gas pipeline based on modified Gaussian plume model and Markov chain Monte Carlo method[J]. Process Safety and Environmental Protection, 2024, 182: 314-326.
[13]	Hou Q, Ren L, Jiao W, et al. An improved negative pressure wave method for natural gas pipeline leak location using FBG based strain sensor and wavelet transform[J]. Mathematical Problems in Engineering, 2013(1): 1-8.
[14]	Tian X, Jiao W, Liu T, et al. Leakage detection of low-pressure gas distribution pipeline system based on linear fitting and extreme learning machine[J]. International Journal of Pressure Vessels and Piping, 2021, 194: 10553.
[15]	Yang L, Li Y, Di C. Application of XGBoost in identification of power quality disturbance source of steady-state disturbance events[C]//2019 IEEE 9th International Conference on Electronics Information and Emergency Communication (ICEIEC). 2019: 1-6.
[16]	Khan M S, Salsabil N, Alam M G R, et al. CNN-XGBoost fusion-based affective state recognition using EEG spectrogram image analysis[J]. Scientific Reports, 2022, 12(1): 14122.
[17]	张忠义, 张磊, 王宇, 等. 基于数据挖掘的玉米淀粉果糖生产流程的关键位点筛选[J]. 化工学报, 2023, 74(10): 4208-4217.
	Zhang Z Y, Zhang L, Wang Y, et al. Data mining-based screening of key points for corn starch and sugar production process[J]. CIESC Journal, 2023, 74(10): 4208-4217.
[18]	Chakraborty D, Elzarka H. Advanced machine learning techniques for building performance simulation: a comparative analysis[J]. Journal of Building Performance Simulation, 2018, 12(2): 193-207.
[19]	Rong G, Li K, Su Y, et al. Comparison of tree-structured parzen estimator optimization in three typical neural network models for landslide susceptibility assessment[J]. Remote Sensing, 2021, 13(22): 4694.
[20]	Bergstra J, Bardenet R, Bengio Y, et al. Algorithms for hyper-parameter optimization[C]//Proceedings of the 24th International Conference on Neural Information Processing Systems. Red Hook, NY, USA: Curran Associates Inc., 2011: 2546-2554.
[21]	Ramachandram D, Lisicki M, Shields T J, et al. Bayesian optimization on graph-structured search spaces: optimizing deep multimodal fusion architectures[J]. Neurocomputing, 2018, 298: 80-89.
[22]	Lago J, De Ridder F, Vrancx P, et al. Forecasting day-ahead electricity prices in Europe: the importance of considering market integration[J]. Applied Energy, 2018, 211: 890-903.
[23]	Lago J, De Brabandere K, De Ridder F, et al. Short-term forecasting of solar irradiance without local telemetry: a generalized model using satellite data[J]. Solar Energy, 2018, 173: 566-577.
[24]	Chen T Q, Guestrin C. XGBoost: a scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco: ACM, 2016: 785-794.
[25]	Guo R, Zhao Z Q, Wang T, et al. Degradation state recognition of piston pump based on ICEEMDAN and XGBoost[J]. Applied Sciences, 2020, 10(18): 6593.
[26]	Ghanbari-Adivi F, Mosleh M. Text emotion detection in social networks using a novel ensemble classifier based on Parzen Tree Estimator (TPE)[J]. Neural Computing and Applications, 2019, 31(12): 8971-8983.
[27]	Shen K Y, Qin H, Zhou J Z, et al. Runoff probability prediction model based on natural gradient boosting with tree-structured parzen estimator optimization[J]. Water, 2022, 14(4): 545.
[28]	孔凡超, 刘懿晗, 刘士吉, 等. 基于智能优化算法和Bi-LSTM神经网络的盾构隧道刀盘扭矩智能预测方法[J]. 中国科学: 技术科学, 2025, 55(1):171-186.
	Kong F C, Liu Y H, Liu S J, et al. Intelligent prediction method for cutterhead torque of shield tunnel based on intelligent optimization algorithm and Bi-LSTM neural network[J]. Scientia Sinica (Technological), 2025, 55(1): 171-186.
[29]	Sameen M I, Pradhan B, Lee S. Application of convolutional neural networks featuring Bayesian optimization for landslide susceptibility assessment[J]. Catena, 2020, 186: 104249.
[30]	马飞虎, 雷皓安, 孙翠羽, 等. 基于RFE-OPTUNA-XGBoost模型的高速公路逃费模式识别[J]. 应用科学学报, 2024, 42(5): 857-870.
	Ma F H, Lei H A, Sun C Y, et al. Highway toll evasion patterns identification based on RFE-OPTUNA-XGBoost model[J]. Journal of Applied Sciences, 2024, 42(5): 857-870.

参数	参数的解释	搜索范围
n_estimators	决策树的数量	[100, 500]
max_depth	决策树最大深度	[3, 10]
learning_rate	学习率，用于控制每棵树对最终预测的贡献	[0.01, 0.3]
subsample	子采样比例，训练每棵树时使用的样本比例	[0.6, 1.0]
colsample_bytree	特征采样比例，训练每棵树时使用的特征比例	[0.6, 1.0]
alpha	L1正则化项系数	[0, 10]
lambda	L2正则化项系数	[0, 10]
min_child_weight	叶子节点最小样本权重	[1, 10]

参数	参数的解释	搜索范围
n_estimators	决策树的数量	[100, 500]
max_depth	决策树最大深度	[3, 10]
learning_rate	学习率，用于控制每棵树对最终预测的贡献	[0.01, 0.3]
subsample	子采样比例，训练每棵树时使用的样本比例	[0.6, 1.0]
colsample_bytree	特征采样比例，训练每棵树时使用的特征比例	[0.6, 1.0]
alpha	L1正则化项系数	[0, 10]
lambda	L2正则化项系数	[0, 10]
min_child_weight	叶子节点最小样本权重	[1, 10]

优化方法	最优的XGBoost超参数	R²	MAPE/%	RMSE	MAE
未优化	[100, 5, 0.1, 1.0, 1.0, 1.0, 1.0, 10]	0.9766	82.0688	0.9334	0.4219
Optuna	[500, 10, 0.2406, 0.7196, 0.6002, 9.9782, 2.8999, 3]	0.9820	4.3397	0.8185	0.3722
PSO	[110, 8, 0.0527, 0.6674, 0.8353, 4.0561, 0.0, 7]	0.9827	4.5012	0.8036	0.3746
BOA	[364, 6, 0.1846, 1.0, 0.9079, 2.8133, 9.4656, 10]	0.9822	3.9499	0.8145	0.3523
TPE	[300, 10, 0.0322, 0.7473, 0.8959, 1.2055, 2.9552, 10]	0.9835	3.7669	0.7850	0.3257

优化方法	最优的XGBoost超参数	R²	MAPE/%	RMSE	MAE
未优化	[100, 5, 0.1, 1.0, 1.0, 1.0, 1.0, 10]	0.9766	82.0688	0.9334	0.4219
Optuna	[500, 10, 0.2406, 0.7196, 0.6002, 9.9782, 2.8999, 3]	0.9820	4.3397	0.8185	0.3722
PSO	[110, 8, 0.0527, 0.6674, 0.8353, 4.0561, 0.0, 7]	0.9827	4.5012	0.8036	0.3746
BOA	[364, 6, 0.1846, 1.0, 0.9079, 2.8133, 9.4656, 10]	0.9822	3.9499	0.8145	0.3523
TPE	[300, 10, 0.0322, 0.7473, 0.8959, 1.2055, 2.9552, 10]	0.9835	3.7669	0.7850	0.3257

对比模型	评价指标
对比模型	R²	MAPE/%	RMSE	MAE
SVM模型	0.3684	44.1430	4.6039	3.4720
CNN模型	0.9285	70.8405	7.8271	1.1014
CNN-LSTM-Attention模型	0.9821	5.4649	0.8161	0.4868
TPE-XGBoost模型	0.9835	3.7669	0.7850	0.3257