引用本文: | 刘喆,郑宏伟,郭哲,彭世亮,张佩颖,李振林,等. 天然气超声流量计健康状态评价方法初探[J]. 石油与天然气化工, 2024, 53(2): 112-118, 138. |
| |
|
摘要: |
目的 天然气计量用超声流量计性能诊断主要采用设定参数阈值的诊断方法,为降低误报、漏报等情况,建立了基于生成对抗神经网络(generated adjoint neural network,GAN)和高维非线性无监督学习的超声流量计健康状态系统诊断方法。方法 采用GAN对原始数据进行学习、生成和扩充,保障超声流量计健康状态诊断建模的数据基础,提取超声流量计在运行过程中的健康状态参数并对其进行时序分析,采用高维非线性无监督聚类学习方法,结合超声流量计失效模式分析,对超声流量计设备进行在线的健康状态诊断。结果 结合超声流量计工作原理和现场实际采集数据,对生成的故障数据集进行了验证。结论 能够准确识别超声流量计当前状态,显著解决传统阈值法误报率、漏报率高的问题,为超声流量计健康诊断的统一管理与开发给予支撑。 |
关键词: 超声流量计 健康状态 特征提取 无监督学习 故障诊断 |
DOI:10.3969/j.issn.1007-3426.2024.02.017 |
分类号: |
基金项目:国家管网集团科技项目“天然气管道超声流量计性能提升技术研究”(CLZB202108) |
|
Study on evaluation method for health status of natural gas ultrasonic flowmeter |
LIU Zhe1,2, ZHENG Hongwei2, GUO Zhe2, PENG Shiliang1, ZHANG Peiying3, LI Zhenlin1, SU Huai1
|
1. College of Mechanical and Storage Engineering, China University of Petroleum-Beijing, Beijing, China;2. PipeChina West East Gas Pipeline Company, Shanghai, China;3.Research Institute of Natural Gas Technology, PetroChina Southwest Oil & Gasfield Company, Chengdu, Sichuan, China
|
Abstract: |
Objective The performance diagnosis of ultrasonic flowmeter for natural gas metering mainly adopts the diagnostic method of setting parameter threshold value, which has some problems such as false alarm and missing alarm. This study establishes a set of ultrasonic flowmeter health state system diagnosis method based on generated adjoint neural network (GAN) and high-dimensional nonlinear unsupervised learning. Methods GAN is used to learn, generate and expand the original data, so as to ensure the data basis of ultrasonic flowmeter health state diagnosis and modeling. Health state parameters of ultrasonic flowmeter during operation are extracted and time sequence analysis is carried out. Based on high-dimensional nonlinear unsupervised clustering learning method and combined with ultrasonic flowmeter failure mode analysis,the online health status diagnosis of ultrasonic flowmeter equipment was carried out. Results The fault data set was generated for verification based on the mechanism of ultrasonic flowmeter and the actual data collected on site. Conclusion The method can accurately identify the current state of ultrasonic flowmeter, significantly solve the problems of high false alarm rate and missalarm rate of traditional threshold method, and support the unified management and development of ultrasonic flowmeter health diagnosis. |
Key words: ultrasonic flowmeter state of health feature extraction unsupervised learning fault diagnosis |