Abstract
Monitoring soil conditions is a critical task in geotechnical engineering, as it directly affects the stability of buildings, transport infrastructure, and engineering structures. One of the most effective approaches to monitoring soil conditions is geophysical sounding, which provides information about physical properties at significant depths. However, analyzing time series obtained through such measurements is challenging due to the large data volume, high variability, and the presence of noise. This study proposes a method for the automatic detection of early signs of soil integrity violations based on the analysis of geophysical time series using neural networks. The application of machine learning techniques improves the accuracy and efficiency of anomaly detection that may precede structural failures. The primary focus is on developing and training a model capable of recognizing deviations from the normal state based on reconstruction errors in time series. The research includes a review of existing geophysical data analysis methods, preprocessing of time series, feature selection, and the implementation and evaluation of a neural network model. The proposed method was validated experimentally, demonstrating its applicability for early diagnosis in soil monitoring systems.