A simultaneous denoising and event-picking approach using supervised machine learning

Microseismic monitoring entails denoising of continuous long-duration time series, followed by event picking. Typically, both processes require different set of algorithms and analysis. We propose a machine learning approach that simultaneously denoises seismic records and picks phase attributes of seismic events of interest. The novelty of the proposed approach is to address two different problems (i.e., denoising and event detection) using a single network. A convolutional neural network is used to capture the high frequency times series, which is combined with gated recurrent unit networks to learn low frequency patterns. The network can also be used as a preliminary tool in human-aided event interpretation, where long time series can be denoised with noticeable events highlighted, which aids in identifying major events for detailed analysis. The network is suitable for different polarized arrays and its application can be extended to P and S event detection.