ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES".
No. 3. Vol. 28. 2022

DOI: 10.17587/it.28.148-155

R. A. Gorbachev¹, Ph. D. in Technology, Leader Engineer, Head MIPT Laboratory, E. M. Zakharova¹, Ph. D. in Technology, Leader Engineer, I. S. Makarov², Postgraduate Student, I. S. Frolov¹, Student,
¹ Federal State Autonomous Educational Institution of Higher Education "Moscow Institute of Physics and Technology (National Research University)", Dolgoprudny, Moscow Region, 141701, Russian Federation
² Federal Research Center "Informatics and Control" RAS (FRC IU RAS), Moscow, 119333, Russian Federation

Neural Network Training in an Automated Dispatcher Problem

The application of artificial intelligence in the development of à decision support system for the implementation of transport traffic is presented. Such systems are designed to adjust the schedule of objects in cases of unforeseen situations. A fully connected artificial neural network with several hidden layers, trained using a genetic algorithm, is used. During training, the functionality that characterizes the deviation from the specified schedule is minimized. Railway traffic is one of the most important types of transport in Russia. Every year it becomes more and more intense, the density and volume of both cargo and passenger traffic increases. As a result, the requirements for the exact execution of the planned traffic schedule increase, since any deviation leads to significant penalties due, for example, to an increase in train delays, their cancellation, etc. The work of a dispatcher, a person who controls railway traffic, is quite time-consuming and becomes more difficult every day, so the development of dispatcher assistance systems is one of the most relevant areas in control automation in this area. At the same time, the existing high requirements for traffic safety, which impose additional restrictions, finally lead to the fact that in this kind of system, the final decision is left to the person, and computer development has a recommendatory character. This article describes the artificial intelligence apparatus in the form of training neural networks using a genetic algorithm to build an automated dispatcher that corrects movement.
Keywords: neural networks, genetic algorithm, railway traffic, optimization problem, game theory

P. 148–155

To the contents