Automatic microseismic event detection in downhole DAS data through convolutional neural networks: A comparison of events during and post-stimulation of the well

Downhole Distributed Acoustic Sensing (DAS) fibers lie within close proximity to microseismic events, and thus are able to detect events with very high resolution and accuracy. With this high resolution comes a substantial amount of data - several terabytes (TB) in size over a few days worth of recording. We created a convolutional neural network that can automatically detect microseismic events in DAS data. Prior work (Lellouch et al., 2021; Huot et al., 2021b), displayed the results of our network when tested on DAS data taken during well-stimulation phases. Here, we run our algorithm through 2-hours worth of data collected approximately one week after stimulation of the well has ceased, and compare the results to previous single-staged results. Our network detected events in approximately 2.04% of the data windows in the 2-hour post-stimulation phase, where previously it had predicted 3.67% of the data to hold events when tested on a single collection stage during well-stimulation. By analyzing the microseismic events in our data post-stimulation, we can analyze fracture evolution in time.