Detecting microseismic events in downhole distributed acoustic sensing data using convolutional neural networks

Abstract

Distributed Acosutic Sensing (DAS) is a growing technology that enables affordable downhole recording of strain wavefields from microseismic events with spatial sampling down to ~1 m. Leveraging this information calls for different event detection approaches than conventionally used in downhole geophone processing. Here a convolutional neural network method is described that operates on the full DAS strain wavefield. The method is developed and tested using data recorded in a horizontal observation well during a hydraulic fracturing operation in the Eagle Ford Shale, Texas where a surface geophone array simultaneously recorded microseismic events. The neural network was trained using synthetic microseismic events injected into recorded ambient noise, and it was applied to detect events in the remaining DAS data. The signal-tonoise ratio of the DAS data was high compared to surface geophones, and 327 of 959 cataloged events from the surface array were detected by the neural network in the DAS data, more than detected by a standard STA/LTA detection algorithm. As DAS data quality continues to improve in the future, neural networks offer many advantages for automated, real-time microseismic event detection, including low computational cost, minimal preprocessing, and continuous performance improvement as more training data is acquired. Presentation Date: Tuesday, September 17, 2019 Session Start Time: 8:30 AM Presentation Start Time: 9:20 AM Location: 303B Presentation Type: Oral