Joint Microseismic Event Detection and Location Based on a Detection Transformer

Abstract

Summary Microseismic event detection and location, as two primary components in microseismic monitoring, offer us invaluable insights into the subsurface conditions and activities during production, CO₂ injection, and fracturing. However, conventional methods for event detection and source location often suffer from the requirements of manual intervention and heavy computation, while other Machine Learning based methods address these two tasks separately, preventing the potential for real-time monitoring. By adopting a Detection Transformer, which is built on a CNN backbone and a Transformer encoder-decoder block with a set-based Hungarian loss, we realize the simultaneous microseismic event detection and location within a single framework, directly from waveforms in which multiple events may arrive. This network is trained on synthetic data corresponding to random source locations in the area of suspected microseismic activity using a smooth velocity model like those obtained from tomography or migration velocity analysis. A synthetic test on the 2D time-lapse SEAM model illustrates the feasibility and potential of this proposed approach.