基于改进集合经验模态分解与稀疏表示的连续钻井液压力波信号处理与识别方法研究

The information transmission rate of MWD(measurement while drilling) is so low that it ishardly able to obtain the real-time down-hole information accurately, which have restricted the safe and efficientexploitationof oil and gas resources. The theory of continuous-wave drilling fluid pressure wave signal transmission can significantly improve the transmission rate of down-hole data , but in the transmission, the drilling fluid pressure wave signal can be easily affected by complex drilling channels and operating environment. The current techniques for signal processing and recognition is still the bottleneck for real application, so the theory and application in this field need to be perfected. This project focuses on studying the continuous-wave drilling fluid pressure wavesignal characteristics expressed by modal component and residual quantity of EEMD(ensemble empirical mode decomposition), demonstrating the noise coupling mechanism in the signal transmission process, revealing the characteristics and rules of changes caused by the variation of drilling pump and drilling depthin signal baseline nonlinear drift and residual , and exploring an effective processing method of continuous-wave drilling fluid pressure wave signal. On this basis, according to the encoding rules of continuous-wave drilling fluid pressure wave signal, no training and training methodsare jointly employed to constructan over-complete dictionary for signal representation;then to develop a highly effective and robust recognition method for continuous-wave drilling fluid pressure wave signal based on sparse representation. Finally, the experimental analysis of processing and recognition of continuous-wave drilling fluid pressure wave signal is conducted to validate the theory and method. The research results of this project will eventually form the drilling fluid pressure wave signal processing and identification method in the complex drilling environment, which will provide a guarantee for the rapid identification and early warning of the underground complex accident.

当前无线随钻测量信息传输速率低、无法实时获得井下信息，制约了油气资源安全高效开发。连续钻井液压力波信号传输 虽然理论上可以大幅提高井下数据传输速率，但传输过程中易受复杂钻柱信道及现场环境影响，信号处理与识别仍是技术瓶颈，理论和应用还需完善。本项目将重点研究连续钻井液压力波信号改进集合经验模态分解的各模态分量及残余量所表示的信号特征，阐明信号在钻柱传输 过程中噪声耦合机理，揭示钻井泵、钻井深度变化引起的信号基线非线性漂移与残余量变化特点规律，建立一种有效的连续钻井液压力波信号处理方法；在此基础上，根据连续钻井液压力波信号编码规则，采用有训练与无训练结合方法建立信号识别的超完备字典，并通过稀疏重构误差阈值的研究，得到一种强鲁棒性与高准确率的基于稀疏表示的连续钻井液压力波信号识别方法。本项目研究成果最终形成复杂钻井环境下钻井液压力波信号处理与识别方法，将为井下复杂事故的快速识别和预警提供保障。

石油资源在国民经济发展中扮演着至关重要的作用。然而，石油资源埋藏深、地质情况复杂，若不能准确掌握井下信息，易导致井下复杂事故，从而制约油气资源安全高效开发。因此，井下数据获取与快速处理可以减少石油勘探深井井下复杂事故的发生，保证钻井作业安全，实现对井下事故的快速识别和预警，具有重要的现实和经济意义。本项目主要研究了连续钻井液压力波信号识别方法及其监督范式的机器学习信号识别算法，包括连续钻井液压力波信号训练样本数据库构建、不确定训练样本标签分类研究、空-谱分类方法研究等，具体的研究工作总结如下：.为了准确识别连续钻井液压力波信号，采集不同井深条件下的连续钻井液压力波信号建立了相应不同井深条件下的连续钻井液压力波信号训练样本数据库。.提出了基于峰值密度聚类算法的不确定训练样本标签检测方法。实验结果表明，该方法可减少低置信样本对分类性能的影响。同时通过引入局部异常因子，提出了光谱角局部异常因子的不确定训练样本标签检测算法，有效提高了不确定训练样本标签噪声的高光谱遥感图像分类性能。.通过在联合稀疏模型上设计基于信息散度的正则化项，克服联合稀疏空-谱分类中非同质化区域导致的分类问题，进一步提升了联合稀疏分类精度。.提出了基于空间峰值密度聚类的不确定样本检测方法，得到了空间峰值密度、超像素加权峰值密度聚类的高光谱遥感图像不确定样本分类表达模型，揭示了不确定样本对监督学习过程的影响机理，实现了高光谱遥感图像不确定样本的检测与去除，解决了监督学习中训练样本的不确定标注问题。.设计了稀疏与类依属融合的高光谱遥感图像分类方法，揭示了高光谱遥感图像中同质区域类依赖的特性关系，实现了高光谱遥感同质区域类别之间的有机关联。.提出了高光谱遥感图像空-谱自适应分类新模型，实现了高光谱遥感空间信息和光谱信息的有机关联，解决了小样本训练样本下高光谱遥感图像地物分类问题