基于水流分割的石油钻井水流异常检测

Anomaly detection of oil drilling water flow based on water flow segmentation

摘要:

通过对钻井管道水流的智能监控技术实现，可以解决石油钻井污染气体的自动监测问题，最大程度的减少人工监测成本。但是依然有以下几个难点需要攻克：（1）传统的特征提取方式不能描述水流形态的变化过程；（2）因为异常情况发生的概率很低，所以异常样本稀少全监督的方法不在适用。为解决特征提取问题，提出了一种基于图像分割的新特征特提取方式——形态流，形态流可以从时序上描述水流形态的变化；另一方面，为克服异常样本稀少的问题，通过无监督的方式——多元高斯建模，来判别水流数据是否正常。实验表明在水流异常数据检测任务中算法检测精度达到了93.6%，在使用GPU并行加速处理时可达到28帧每秒的处理速度，能够准确地检测出水流数据中的异常数据帧。

Abstract:

Through the realization of intelligent monitoring technology for the water flow of the drilling pipeline can solve the problem of automatic monitoring of polluted gas from oil drilling and minimize the cost of manual monitoring. However, there are still several difficulties that need to be overcome: (1) The traditional feature extraction method cannot describe the change process of the water flow pattern; (2) Because the probability of abnormal situations is very low, the method of full supervision with rare abnormal samples is not applicable. In order to solve the problem of feature extraction, proposes a new feature extraction method based on image segmentation-morphological flow, which can describe the change of water flow morphology in time series; on the other hand, in order to overcome the problem of rare abnormal samples, an unsupervised method-multivariate Gaussian modeling is used to determine whether the water flow data is normal. Experiments show that the detection accuracy of the algorithm in the water flow abnormal data detection task reaches 93.6%, and the processing speed of 28 frames per second can be reached when using GPU parallel acceleration, and it can accurately detect abnormal data frames in the water flow data.