Abstract
<jats:p>В работе предложена технология раннего обнаружения и обработки параметров пожара, основанная на объединении данных нескольких датчиков. Для этого можно использовать мультисенсорное объединение данных на основе интернета вещей. Показано применение многослойного персептрона в качестве решающего уровня системы обнаружения пожара как мультисенсорной системы объединения информации для оповещения о пожаре. Для обучения нейронной сети используется метод обратного распространения ошибки для объединения данных от датчиков температуры, плотности дыма и CO, а также может повысить точность оповещения.</jats:p> <jats:p>Fires are among the most tragic urban events that cause loss of life and property. Unfortunately, the fire statistics is disappointing. About 7–8 million fires occur annually in the world, which kill approximately 85–90 thousand people. And this is despite the availability of modern fire protection systems, which are becoming more complex every year. In order to increase the reliability and efficiency of decisions, optimize processes and adapt to a dynamic market environment, modern artificial intelligence technologies should be more actively introduced into existing decision support systems, which make it possible to extract valuable information from data arrays that cannot be analyzed using traditional methods. The purpose of the study is to form proposals for the implementation of systems using the principles of artificial intelligence by creating neural networks of the appropriate level. To obtain the results, general scientific and special methods were used – analysis, generalization, which were based on the general provisions of the theory of information synthesis and analysis. The technology for monitoring, early detection and processing of fire parameters is proposed, based on combining data from various types of sensors, complexes, subsystems and systems, for which multisensory data integration based on the principles of machine learning and the Internet of Things is used. The results obtained in the course of the work will help in solving the problems of forming requirements for algorithms and architecture of a neural network for more effective detection of potentially dangerous situations than individual sensors, since a set of data and their dynamics are taken into account. This approach increases safety and reduces the number of false alarms, allowing timely response to emergencies.</jats:p>