A Simple Baseline of Pipeline Data Simulation Algorithm Based on DGM

Xu Hu, Hengyu Wang, Hongyang Zhao, Gaoling Liu, Di Wang
{"title":"A Simple Baseline of Pipeline Data Simulation Algorithm Based on DGM","authors":"Xu Hu, Hengyu Wang, Hongyang Zhao, Gaoling Liu, Di Wang","doi":"10.1115/ipc2022-86997","DOIUrl":null,"url":null,"abstract":"\n Data is the cornerstone of data science development. This implies the significance of emphasizing information protection found in oil and gas pipeline operating data generated throughout the development of digitalization in relevant companies. Despite of the abundance of sensitive information contained in real data, the availability of open public data sets for professionals in the industry is considerly scarce. This has in turn become the greatest obstacle in producing any forms of research achievements in the industry. At the same time, the sensor deviation leads to a weak reference of the simulation data obtained by the traditional physical formula. Generative deep learning networks are able to directly learn from data and generate high-fidelity as well as diversified samples, allowing them to complete data set generation and information desensitization. This paper presents the application of two deep generative model algorithms, i.e., VAE (Variational Auto-Encoder) and GAN (Generative Adversarial Networks), for pipeline operating data generation. Three main contributions have been made with this approach: (1) Research and release the pipeline operation simulation algorithm generated by the deep generation model based on the big historical data. (2) the algorithms were found to be capable of exporting operating data as a reference for the pumps according to specific pipeline parameters, e.g., pressure corresponding to a certain flow. A data set with the same probability distribution as the actual pipeline operation data could be generated; (3) data-distribution-based detection of abnormalities in pipeline operating conditions using a GAN algorithm identifier was proposed for the very first time. The experimental results presented < 1% mean square error (MSE) between the optimal model and the normalized real data — 0.4% to be more specific, verifying the algorithm’s capability of generating a desensitized data set similar to the real data. Moreover, a classifier trained with the generated data using the GAN algorithm identifier was able to detect as much as 69% abnormal operating conditions, further verifying the effectiveness of the proposed algorithm. Code are publicly available in https://github.com/ShawnHoo7256/BOGC_Project.","PeriodicalId":183044,"journal":{"name":"Volume 3: Operations, Monitoring, and Maintenance; Materials and Joining","volume":"86 8 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Volume 3: Operations, Monitoring, and Maintenance; Materials and Joining","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1115/ipc2022-86997","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

Abstract

Data is the cornerstone of data science development. This implies the significance of emphasizing information protection found in oil and gas pipeline operating data generated throughout the development of digitalization in relevant companies. Despite of the abundance of sensitive information contained in real data, the availability of open public data sets for professionals in the industry is considerly scarce. This has in turn become the greatest obstacle in producing any forms of research achievements in the industry. At the same time, the sensor deviation leads to a weak reference of the simulation data obtained by the traditional physical formula. Generative deep learning networks are able to directly learn from data and generate high-fidelity as well as diversified samples, allowing them to complete data set generation and information desensitization. This paper presents the application of two deep generative model algorithms, i.e., VAE (Variational Auto-Encoder) and GAN (Generative Adversarial Networks), for pipeline operating data generation. Three main contributions have been made with this approach: (1) Research and release the pipeline operation simulation algorithm generated by the deep generation model based on the big historical data. (2) the algorithms were found to be capable of exporting operating data as a reference for the pumps according to specific pipeline parameters, e.g., pressure corresponding to a certain flow. A data set with the same probability distribution as the actual pipeline operation data could be generated; (3) data-distribution-based detection of abnormalities in pipeline operating conditions using a GAN algorithm identifier was proposed for the very first time. The experimental results presented < 1% mean square error (MSE) between the optimal model and the normalized real data — 0.4% to be more specific, verifying the algorithm’s capability of generating a desensitized data set similar to the real data. Moreover, a classifier trained with the generated data using the GAN algorithm identifier was able to detect as much as 69% abnormal operating conditions, further verifying the effectiveness of the proposed algorithm. Code are publicly available in https://github.com/ShawnHoo7256/BOGC_Project.
一种基于DGM的简单基线管道数据仿真算法
数据是数据科学发展的基石。由此可见,相关企业在数字化发展过程中产生的油气管道运行数据中,重视信息保护的重要性。尽管真实数据中包含了大量的敏感信息,但可供业内专业人士使用的公开数据集却相当稀缺。这反过来又成为该行业产生任何形式的研究成果的最大障碍。同时,传感器的偏差导致传统物理公式得到的仿真数据参照性较弱。生成式深度学习网络能够直接从数据中学习,生成高保真度、多样化的样本,完成数据集生成和信息脱敏。本文介绍了两种深度生成模型算法,即VAE(变分自编码器)和GAN(生成对抗网络)在管道运行数据生成中的应用。该方法主要有三个方面的贡献:(1)研究并发布了基于大历史数据的深度生成模型生成的管道运行仿真算法。(2)算法能够根据特定的管道参数(如某一流量对应的压力)导出泵的运行数据作为参考。生成与实际管道运行数据具有相同概率分布的数据集;(3)首次提出了基于数据分布的基于GAN算法标识符的管道运行工况异常检测方法。实验结果表明,最优模型与归一化真实数据的均方误差(MSE) < 1%,更确切地说是0.4%,验证了算法生成与真实数据相似的脱敏数据集的能力。此外,使用GAN算法标识符对生成的数据进行训练的分类器能够检测到多达69%的异常操作条件,进一步验证了所提出算法的有效性。代码可在https://github.com/ShawnHoo7256/BOGC_Project上公开获得。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信
小红书