Microseismic event location using artificial neural networks

Recent advances in the field of machine learning coupled with available computational resources provide a great opportunity to address location of seismic events. Recently a new approach was proposed that uses feed-forward neural networks. First, the method utilizes only the P-wave arrival times instead of full waveforms. Second, instead of relying on historical training data, the neural network is trained on synthetically generated data. Once trained, the network can be deployed to locate real events by feeding their observed P-wave arrival times as input. The main challenge in the application of feed-forward neural networks to real datasets is in changing set or receivers where we pick time arrivals, leading to a non-regular input. We show how this problem can be overcome with a transfer learning approach and validate it by locating microseismic events that occurred during a hydraulic fracturing operation in Oklahoma.