基于隐马尔科夫模型的石油钻井过程异常检测的研究

发布时间：2018-07-31 11:20

【摘要】：石油钻井是一项非常庞大的工程,其过程中存在很多不确定因素并且这些因素具有较强的隐蔽性。在钻井过程中非常容易发生各种钻井事故,这严重威胁着钻井平台安全地作业。因此,在复杂的钻井施工过程中,如果能够捕获和运用各种有用的技术信息,对将有可能发生的钻井事故能够给出相对准确的预警或者能够给出一定程度上的警示,对于防止和控制钻井事故的扩大、最大限度地减少经济损失具有非常重大的现实意义。本文从信号处理的角度全面的分析了钻井工程中各种钻井参数的变化情况,并且总结出当钻井事故发生时所有与该钻井事故相关的钻井参数的变化趋势。对钻井事故进行预警时相关钻井参数的趋势变化是关键,并且不同钻井甚至是同一钻井不同井深石油钻井参数的基准值会发生变化,传统的非平稳信号的处理方法不适用于钻井参数的趋势分析。因此,本文构建了一种能反映钻井参数趋势变化特征,并且不受基准值影响的趋势特征量。由于HMM能够对一个时间跨度上的信号进行统计分析建模和分类是一种动态模式识别工具,因此本文构建了基于CHMM的钻井参数异常预警模型。并且通过经验统计出CHMM模型输出匹配概率异常的阈值,实验表明CHMM异常预警模型能够及时有效的对钻井参数异常进行预警。凭经验选取或统计出固定阈值的方法虽然很容易实现,但是实质上它是不太合理的,当选择的阈值太大预警的漏报率就会很高,反之,选取的阈值太小误报率就会升高,所以最好是能使阈值自适应地改变其大小,从而降低预警的漏报率和误报率。通过分析两种常用自适应阈值方法并将其用于CHMM模型输出匹配概率异常的阈值确定中,实验表明,自适应阈值方法能够取得比固定阈值更高的预警准确率。当钻井过程中发生故障时,其特征能够通过几个钻井参数的异常变化表现出来,因此,对钻井故障进行预警时需要对几个钻井参数进行综合决策。
[Abstract]:Petroleum drilling is a very large project, there are many uncertain factors in the process and these factors have strong concealment. Drilling accidents are very easy to occur in drilling process, which seriously threaten the safe operation of drilling platform. Therefore, in the complex drilling process, if we can capture and use all kinds of useful technical information, we can give a relatively accurate warning or a certain degree of warning to the possible drilling accidents. It is of great practical significance to prevent and control the expansion of drilling accidents and to minimize economic losses. In this paper, the variation of various drilling parameters in drilling engineering is analyzed comprehensively from the point of view of signal processing, and the variation trend of all drilling parameters related to the drilling accident is summarized when the drilling accident occurs. The trend change of relevant drilling parameters is the key to early warning of drilling accidents, and the reference values of drilling parameters of different drilling wells and different depth wells will change. The traditional non-stationary signal processing method is not suitable for the trend analysis of drilling parameters. Therefore, this paper constructs a trend characteristic quantity which can reflect the trend change of drilling parameters and is not affected by the reference value. As HMM is a dynamic pattern recognition tool for statistical analysis and classification of signals over a time span, a drilling parameter anomaly warning model based on CHMM is constructed in this paper. The threshold of outputting probability anomaly of CHMM model is calculated by experience. The experimental results show that the CHMM anomaly early warning model can forewarn drilling parameter anomalies in time and effectively. The method of selecting or statistics fixed threshold based on experience is easy to realize, but in essence it is unreasonable. When the threshold is too large, the false alarm rate will be very high, otherwise, if the threshold is too small, the false positive rate will increase. So it is better to change the threshold adaptively to reduce the false alarm rate and false alarm rate. By analyzing two commonly used adaptive threshold methods and applying them to determine the threshold of output-matching probability anomaly of CHMM model, the experimental results show that the adaptive threshold method can achieve higher early warning accuracy than fixed threshold. When the drilling failure occurs, its characteristics can be shown by the abnormal variation of several drilling parameters. Therefore, it is necessary to make a comprehensive decision on several drilling parameters when the drilling fault is forewarned.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TE28;O211.62

【相似文献】