This article is devoted to the development of multi-agent systems for predicting the time of earthquakes based on
seismic signals. The work uses a dataset from laboratory signals, which used to calculate the time predicted before the
next earthquake. The MadKIT platform and the Python programming language are used to build multi-agent systems.
The "tsfresh" package is used to calculate a large number of time series characteristics, so-called features, from seismic
signals for further use in regression. The article considers one of the regression models - LightGBM. Using it, a set of
data was processed and the predicted time of the earthquake was obtained. This article shows the relevance and prospects
of the research area, describes the functionality of the created agents