Classification and Localization of Fracture-hit Events in Low-frequency DAS Strain Rate with Convolutional Neural Networks
Abstract
Distributed acoustic sensing (DAS) has been used in the oil and gas industry as an advanced technology for surveillance and diagnostics. Operators use DAS to monitor hydraulic fracturing activities, to examine well stimulation efficacy, and to estimate complex fracture system geometries. Particularly, low-frequency DAS can detect geomechanical events such as fracture-hits as hydraulic fractures propagate and create strain rate variations. Analysis of DAS data today is mostly done post-job and subject to interpretation methods. However, the continuous and dense data stream generated live by DAS offers the opportunity for more efficient and accurate real-time data-driven analysis.
The objective of this study is to develop a machine learning-based workflow that can identify and locate fracture-hit events in low-frequency DAS data. The study is conducted in two phases. In phase one, a fracture propagation model is used to produce strain rate patterns observed at a hypothetical monitoring well. Using this model, two sets of strain rate responses are generated with one set containing fracture-hit events. The simulated data are then used to train a custom convolutional neural network (CNN) model for identifying the presence of fracture-hit events. The same model is trained again for locating the event with the output layer of the model replaced with linear units. The models achieved near-perfect predictions for both event classification and localization. In phase two, the same workflow is applied to field data, which includes 8.4 days of DAS monitoring data while two offset wells are hydraulically stimulated. A more complex model (AlexNet) is used to train for classifying events and for localizing fracture-hits. Using AlexNet, we achieved f1 score of 0.9 for identifying fracture-hits and R2 of 0.93 for localizing fracture-hits.
Additionally, edge detection techniques are used for recognizing fracture-hit event patterns in the simulated strain rate images. The accuracy is also plausible, but edge detection is more dependent on image quality and shape, hence less robust compared to CNN models. It can only be applied to simulated data since field data often shows irregular fracture-hit patterns. This comparison further supports the need for CNN applications in image-based real-time fiber optic sensing event detection.
Citation
Chen, Mengyuan (2021). Classification and Localization of Fracture-hit Events in Low-frequency DAS Strain Rate with Convolutional Neural Networks. Master's thesis, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /195078.