石油钻井过程井漏异常的预警技术研究

发布时间：2018-01-07 01:13

本文关键词：石油钻井过程井漏异常的预警技术研究　出处：《郑州大学》2015年硕士论文　论文类型：学位论文

【摘要】：整个石油钻井工程预警系统结构复杂,操作环境恶劣且工况多变,对钻井过程中各工况的实时监测控制与故障诊断是目前钻井工程预警系统研究的重要方向;其中实时监测数据的采样、处理、传输、野值点剔除、工况分类等技术是系统故障检测与诊断的核心技术。本文以核主元分析方法为理论基础,以实现石油钻井工程的自动化预警为目的,通过建立多个核主元分析模型,完成钻井过程的信息自动获取、特征量提取和故障诊断。本文针对石油钻井这一复杂的多工况过程,在所选取“井漏事故”的故障点已知的情况下对工程预警系统的故障检测与诊断进行了深入研究。主要研究内容如下:(1)数据处理:从钻井现场获得的原始运行数据中存在大量的过程变量,比如立管压力、总池体积、出口流量等,彼此之间具有很强的相关性,所以必须对这些数据进行预处理才能用于过程分析。对样本数据进行标准化处理是数据处理的基础;野值点剔除算法主要是用来剔除数据中出现的孤立野值点和连续野值点,验证了以剔除率的拐点为标准进行滑动窗口长度自适应确定的可靠性,能够有效剔除过程中的一些毛刺,以免影响检测结果;除此之外,还综合分析了各个变量间的相关性,提取出一些能够描述变量变化趋势的主要特征量(短期方差、长期方差、离差等)。(2)钻井工况分类:针对石油钻井工程预警系统,本文提出了一种新的能将钻井中的各个工况正确分类的门限值分类算法。该分类法不需要进行繁琐的计算,只需依据综合录井仪所记录的钻井过程数据,准确预置门限参量和参考数值,便可实现对各稳态工况的正确分类。钻井过程复杂多变,变量间存在很强的相关性,故障类型也呈现出多样性,如果利用常用的K均值聚类方法对样本数据进行分类,无法根据钻井数据准确的计算出系统的稳定度因子、分类指数以及隶属度;门限值分类算法不需要计算这些量,它是通过预置过程变量的门限值来划分工况的。(3)实例仿真:鉴于研究对象是非线性过程,需考虑将基于单一核主元分析模型的故障检测方法扩展为可以应用于石油钻井过程的多个核主元分析模型故障检测方法。钻井工程预警系统采用的多模核主元分析方法,不仅构造了单一核主元分析模型贯穿整个钻井过程,也构造了多个核主元分析模型对应过程中不同的工况;如果哪一个工况发生故障,那么基于门限值分类算法的多个核主元分析模型故障检测方法能够迅速的将发生故障的工况分离出来,并引入相对应的故障检测模块。因此,可以实现在不同的变量空间和相应的故障检测统计量控制图中的过程监测,虽然它们不能直接的判断出故障出现的原因,却能够通过统计图显示出过程变量是否超出了正常控制限,然后将检测结果跟经验相结合,最终可以判定故障的类型和产生故障的原因,实现准确、灵敏的故障检测。
[Abstract]:The whole system of petroleum drilling engineering prediction of complex structure, poor operating environment and changing conditions of different conditions in the process of drilling, the real-time monitoring and fault diagnosis is an important direction of the present system of drilling engineering warning; including sampling, real-time monitoring data processing, transmission, eliminating outliers, condition classification technology is the core technology fault detection and diagnosis system. Based on kernel principal component analysis method as the theoretical basis, in order to achieve the automatic early warning of petroleum drilling engineering for the purpose, through the establishment of multiple kernel principal component analysis model, complete the drilling process information automatic acquisition, feature extraction and fault diagnosis. In this paper a more complicated process oil drilling, in the selected "fault point known well leakage accidents" in the case of fault detection and diagnosis of the early-warning system is deeply studied. The main research contents such as : (1) data processing: the existence of a large number of process variables of original operation data obtained from drilling site, such as standpipe pressure, total pool volume, export volume, has a strong correlation between each other, it is necessary to preprocess the data can be used to process analysis of the sample data standardization. Processing is the basis of data processing; outliers elimination algorithm is mainly used to isolate outliers in the data points and eliminate appear continuous outliers, to verify the inflection point rejection rate for the standard length of sliding window is adaptively determined by, can effectively eliminate the burr in the process, so as not to affect the test results; in addition, also analyzed the correlation between the variables, extract some main features which can describe the change trend of quantity variables (short term variance, variance, deviation, etc.). (2): according to the classification of drilling conditions Petroleum drilling engineering warning system, this paper proposes a new classification algorithm can correctly classify all conditions in drilling the threshold. The method does not require tedious calculations, only on the basis of drilling data of comprehensive logging instrument recorded accurately, the preset threshold parameter and a reference value, it can achieve correct the classification of the steady state conditions. The drilling process is complicated, there is a strong correlation between variables, the fault type is also showing diversity, if using K means clustering method used to classify the sample data, not according to the calculation of drilling data accurate system stability factor, classification index and membership threshold value; the classification algorithm does not need to compute these quantities, it is through the preset process variable threshold division condition. (3) the simulation: in view of the research object is nonlinear process, must be considered based on single A fault detection method of kernel principal component analysis model for the expansion of oil drilling process can be applied to a plurality of kernel principal component analysis model of fault detection method. The early warning system for drilling multi kernel principal component analysis method, not only the structure of single kernel principal component analysis model through the whole process of drilling, but also constructed a multiple different conditions a kernel principal component analysis model corresponding to the process; if a fault condition occurs, then the threshold condition classification algorithm of multiple kernel principal component analysis model of fault detection method can quickly turn fault separation based on fault detection module and introduce corresponding. Therefore, can be implemented in different variable space and the corresponding fault detection statistics process control monitoring map, although they can not directly determine the cause of the failure, but can through the statistical chart shows the process variable Whether it exceeds the normal control limit, and then combine the test results with experience, we can ultimately determine the type of failure and cause the cause of failure, so as to achieve accurate and sensitive fault detection.

【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TE28;TP277

【参考文献】