Development of a model for optimizing the management of fire and rescue units during a fire using neural networks
Abstract
Development of a model for optimizing the management of fire and rescue units during a fire using neural networks
Incoming article date: 12.08.2025The article is devoted to the development of an innovative neural network decision support system for firefighting in conditions of limited visibility. A comprehensive approach based on the integration of data from multispectral sensors (lidar, ultrasonic phased array, temperature and hygrometric sensors) is presented. The architecture includes a hybrid network that combines three-dimensional convolutional and bidirectional LSTM neurons. To improve the quality of processing, a cross-modal attention mechanism is used to evaluate the physical nature and reliability of incoming signals. A Bayesian approach is used to account for the uncertainty of forecasts using the Monte Carlo dropout method. Adaptive routing algorithms allow for quick response to changing situations. This solution significantly improves the efficiency of firefighting operations and reduces the risk to personnel.
Keywords: mathematical model, intelligence, organizational model, gas and smoke protection service, neural networks, limited visibility, fire department, management, intelligent systems, decision support