Microseismic source imaging using physics-informed neural networks with hard constraints: a field application

Microseismic source imaging is crucial for event location both on the exploration and the seismological scales due to its high accuracy and high resolution. There exists a challenge to source imaging in the case of common sparse observation and irregular geometry. Our recently proposed direct imaging method via physics-informed neural networks with hard constraints has already shown great potential in solving such a problem on synthetic data. Here, we further show the effectiveness of this method by means of the application to the Real hydraulic fracturing data. Specially, we have slightly modified the workflow by adding preprocessing and using the reference frequency loss function with causality implementation to obtain reasonable and reliable source locations. The field examples show that our method can correctly locate the source with physics-guided training signals in a label-free manner.