基于随机森林的钻井工程预警研究

Study on Drilling Engineering Prewarning Based on Random Forests

摘要: 针对因录井传感器工作性能不稳定、安装位置受限等原因造成数据失真和丢失导致的钻井工程异常预报准确率不高，因传感器传输问题无法获知钻井状态导致工程事故预报准确率不高，因新研制的岩屑流量监测仪系统单一无法准确预警等问题，以录井传感器、井下传感器和岩屑流量监测仪为对象，结合其数据失真、数据不完整、数据传输困难、动态数据流、单一系统等特点，从欧氏距离、曼哈顿距离、GMBR距离、马氏距离4个维度，以无异常、异常上升、异常下降为目标空间，采用随机森林算法设计参数异常判断。以各个参数的欧氏距离为维度，以各种钻井事故复杂程度为目标空间，采用随机森林算法设计钻井工程事故复杂预警模型。利用该模型在现场真实数据集上进行仿真，结果表明工程参数异常和事故复杂预报准确率均明显提升。

Abstract: In consideration of the problems of instability in operation, limitation in installation position that caused low accuracy of drilling abnormal prewarning for data distortion and loss, low accuracy of drilling abnormal prewarning from unstable operation of sensors and newly developed cuttings flow monitoring instrument, the mud logging sensors, down hole sensors and cuttings flow monitoring instrument were used as object; the low warning precision of logging sensors, down hole sensors and cutting flow monitor was analyzed from the aspect of data distortion, data loss, data transmitting difficulty and single system. Abnormal condition discriminating parameters were designed with random forest algorithm in four dimensions, including Euclidean distance, Mahaton distance, GMBR distance and Marsh distance, and with conditions of no abnormality, abnormality increase and abnormality decrease were used as its discriminating target space. A drilling engineering accident warning model was built by using random forest algorithm with Euclidean distance of each parameter as its dimensions and various drilling accident conditions as its target space. Finally, emulation is made with actual field data, and the results show that the warning accuracy for both abnormal engineering parameters and accidents is improved significantly.