Seismic Moment Tensor Inversion in Anisotropic Media using Deep Neural Networks

We design a deep neural network (DNN) and train it to invert the focal mechanism of microseismic events that occur during a hydraulic fracture treatment of unconventional reservoirs. For the testing, we generate synthetic microseismic events in anisotropic 3D media and consider a realistic dual-well monitoring scenario. We show that for this geometry a trained DNN can successfully retrieve the six independent elements of the moment tensor. We statistically analyze the correlation coefficients and relative errors of the results and demonstrate that the moment tensor can be accurately estimated using the proposed DNN, providing a reliable alternative to other conventional inversion techniques.