USING A NEURAL NETWORK TO IDENTIFY TERRITORIES AT RISK OF NATURAL EMERGENCIES

Abstract

A digital model has been developed on the basis of a set of geological, engineering geological and digital terrain maps for machine learning of a neural network using the content of a map of natural processes for the Sochi territory. The description of natural processes in the preparation of input data for neural network training is based on the results of research on the Southern slope of the Greater Caucasus ridge. The neural network is based on the input and hidden matrices that distribute the input training signal at the output into the constituent elements of events with acceptable accuracy achieved during training. The author has developed his own library with functions that implement the development of a neural network, which has expanded the possibilities for solving target tasks. A set of programs has been created to work with the neural network. The territories of possible manifestations of landslide processes constructed by the neural network are well correlated with the available landslide maps and areas of identified deformations of engineering structures. The completed work can become a convenient tool for developers of territorial planning documents in the functional zoning of the territory.