Applying Neural Network Methods for Acoustic Signals Processing in Earthquake Prediction
摘要
Intensive technogeneous noises provoke negative impact on seismological observations. However, a number of studies stated that the man-made emitter source can be used as some probing signal to create additional seismic “illumination” of the geo-environment and to identify heterogeneities with strong scattering and reflecting properties. In this paper, it was offered to use the registration of alteration in characteristics of response to vibration impact as the earthquake precursors. In terms of signal reception, it was proposed to use experience of the inside-drill hole geo-acoustic monitoring that was conducted at Petropavlovsk-Kamchatsky territory geodynamic polygon. Numerical simulation of acoustic wave propagation in the geo-medium was performed using Physics-Informed Neural Networks, as well using the Kolmogorov-Arnold Networks as neural network architecture. The offered method allowed to realize modeling based on the small amount of data in the complex irregular boundaries available. The method was suggested for processing of the recorded acoustic signal, aimed at alteration signs identification in the geo-medium spatial location of inhomogeneity. The inverse task was solved applying Inverse Physics-Informed Neural Networks, as well by the Kolmogorov-Arnold Networks architecture. The environment parameters were determined basing on the measured field characteristics those, measured in a set of points.