Automated Drilling and Production Event Detection Using Advanced Time-Series Pattern Recognition Techniques

Abstract The identification of well/rig states in time is crucial for accurate risk assessment, event/incident detection, and efficiency tracking in drilling and production operations. State identification has traditionally relied on rule-based systems which are often insufficient, resulting in inaccurate predictions and unreliable risk assessment, imprecise event detection, and a biased efficiency estimation. In this paper, well/rig state identification is handled as a time-series multi-class classification problem, using data collected at a relatively high frequency (typically 0.02–1 Hz) with sensors installed downhole, on the wellhead, and on the surface equipment of the rig or coiled-tubing unit operating on the well. Three advanced, scalable methods are introduced for automatically identifying the well/rig state, namely a first-order logic inference system, a recurrent neural network (RNN) classifier, and a transformer-based classifier. We compare these methods in two use cases involving drilling and production operations, including the detection of stuck pipe drilling incidents and production pressure trend abnormalities respectively. The methods were evaluated on a withheld, pre-labeled test dataset consisting of 75 hours of 1-Hz drilling data and 7,133 hours of 0.017-Hz production data. This evaluation showed that the transformer-based classifier outperformed the other two methods in both drilling and production applications. Additionally, we observed that the deep learning-based classifiers were only slightly more computationally expensive than inference systems, making all models suitable for real-time prediction. During method testing, we also investigated the value of accurate well/rig state identification in the assessment of drilling and well-integrity risks. This revealed that the lack of accuracy can severely bias the risk assessment, for instance, by overestimating the risk of pipe sticking and generating spurious predictions of abnormal pressure trends. This, in turn, can result in unnecessary and costly preventative measures. This is the first study to compare the advantages and shortcoming of different methods for automatically identifying well/rig state, including one not evaluated before—the transformer model. It is also the first to propose an automated engine for well state identification during well production operations. The significant difference between the drilling and production use cases shows the versatility and general applicability of the three methods discussed. Moreover, the applicability of the analyzed methods is not limited to the examples provided here; they are applicable to any task involving the classification of time-series data, and thus can be employed in other well delivery / operation stages and other event detection applications.