In this paper, it is proposed to expand the set of earthquake precursors by detecting fluctuations in the time of the sounding signal, which is used as electromagnetic waves. Probing is performed directly on the Earth's mass, preferably in earthquake-prone areas, and the probe signal emitter is located below the Earth's surface. This reduces the number of external factors that distort the nature of the information signal. By changing the resulting field, you can learn about changes in the parameters of the medium along the wave propagation path. The detected anomalies in the distribution of characteristics of the Earth's environment can indicate foci of accumulating deformations, and a quantitative assessment of the anomalies that have occurred can be used to assess the probability of an earthquake. Numerical simulation of electromagnetic wave propagation in the geomedium is performed using Physics-Informed Neural Networks, as well as the Kolmogorov-Arnold Networks architecture. The method used allows you to implement modeling based on a small amount of data in the presence of complex irregular boundaries. The inverse problem is solved using Inverse Physics-Informed Neural Networks, as well as the Kolmogorov-Arnold Networks neural network architecture, and the environment parameters are determined based on the measured field characteristics measured in a set of points. This method can be used for long-term and short-term earthquake forecasts with high efficiency in obtaining measurement results.