基于数据驱动的抽油机井实时故障诊断方法研究

项目名称： 基于数据驱动的抽油机井实时故障诊断方法研究

项目编号： No.61273160

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 田学民

作者单位： 中国石油大学（华东）

项目金额： 80万元

中文摘要： 石油资源的安全开采具有重要的战略意义，抽油机井运行故障的实时检测、识别和预报是石油工业安全监控中亟待解决的关键科学问题。针对目前抽油机井故障诊断方法存在的检测实时性差、识别准确率低等问题，本项目立足于复杂的实时测控数据，拟研究基于数据驱动的抽油机井实时故障诊断方法。研究内容包括：提出一种基于非线性慢特征分析的特征提取方法，解决基于快速时变数据的故障实时检测问题；建立一种基于多核单类支持向量机的故障识别技术，解决抽油机井故障识别准确率低的问题；研究一种基于增量式非线性典型变量分析的动态多变量时间序列分析方法，解决油井缓变故障的早期预报问题；设计一种具有进化机制的多智能体技术，探索井群监控方法；通过仿真分析、实验验证和现场测试等手段对所研究的方法进行改进完善。项目研究成果将有助于保障采油过程安全性，建立一套基于数据驱动的抽油机井故障诊断新机制，进一步完善故障诊断理论研究.

中文关键词： 故障诊断；数据驱动；抽油机井；；

英文摘要： Safety of petroleum production has important strategic significance. Realtime fault detection, recognition and prediction for rod pumped wells have become a key and urgent scientific problem in petroleum process safety monitoring. Aiming at the problems of bad performance of realtime fault detection and low correct rate of fault recognition from traditional fault diagnosis methods, this project utilizes the complex realtime measured data and is to study the data-driven realtime fault diagnosis methods for rod pumped wells. Research points are given as follows: A nonlinear slow feature analysis method for feature extraction is proposed to solve the problem of realtime fault detection based on fast temporal data. Multi-kernel one-class support vector machine is built to improve the correct rate of rod pumped well fault recognition. Dynamic multivariate time series analysis method using incremental nonlinear canonical variate analysis is studied for the early fault prediction. A multi-agent method with evolutional mechanism is designed for the monitoring of rod pumped well group.Simulations, experimentations and field testings are used to improve and perfect the studied methods. The project would be useful for ensuring the safety of petroleum production, build a new mechanism for rod pumped well fault diagnosis an

英文关键词： fault diagnosis；data-driven；rod pumped well；；