Application of Siamese Neural Networks for the Type of Emergency Determination

Abstract

For reliable operation of systems for early detection and prevention of emergencies, their algorithms should use machine learning methods. The use of machine learning methods is associated with the replacement of detectors, usually used in such systems, with sensors that transmit the measurement results to the computing unit of the system. Measurement of the main factors, along with their threshold processing, allowed the use of machine learning methods to quickly detect the fact of ignition, determine the type of ignition source and its localization. The study is devoted to the development of a neural network algorithm for determining the type of fire in the early stages of an emergency. The results of emergency detection refresh the information of human-machine interface immediately. We proposed to use a complex neural network consisting of five Siamese networks based on distance and a Bayesian network. The proposed neural networks have a simple architecture and a small number of layers. To train the neural network, a computer model has been developed. It simulates the ignition process and inertia of the system's sensors.