The multi-agent system for constructing normative train schedules was developed within the Integrated Railway Transport Management System for Russian Railways. For five years, from 2015 to 2020, the system has been in trial operation at the Eastern Polygon of Russian Railways, and has now been implemented into commercial operation.

The Eastern Polygon is located within the boundaries of 4 railways — Krasnoyarsk, East Siberian, Transbaikal and Far Eastern. It serves the transport needs of 14 regions of the Russian Federation and provides transit for the entire country. The operational length of railways is more than 17 thousand kilometers. The polygon has 810 railway stations.

The normative train schedule is the organizational and technological basis for work of all divisions of railways and must ensure efficient use of throughput and carrying capacity of sections and processing capacity of stations, rational use of rolling stock, compliance with the established duration of continuous operation for engine crews, as well as the ability to

БАМ 2019.jpg

The main difficulty in constructing a normative train schedule is associated with planning the passage of the maximum possible number of trains. Two oncoming streams of trains move «through» each other along the same resources – railroad hauls and station tracks in a limited time window, taking into account a large number of different restrictions on speeds and time intervals for safe traffic, working hours of engine crews, etc. Train streams are so tightly intertwined with each other that many sections and stations are more than 90% loaded.

To build such a schedule, an experienced schedule engineer who knows his section in great detail can spend a work-week or even more, whereas our system can do the same in less than 1 hour.

In world practice, there are many attempts to solve the problem of automatically constructing a standard train schedule, however, our calculations propose a significant increase in dimension of the planned section while maintaining an acceptable calculation time.

Actually, this increase in productivity, due to the choice of multi-agent solution, made it possible to gradually move from small model problems to practical use in real conditions at the most complex and highly loaded ranges of Russian Railways.

The system uses a hybrid optimization algorithm that combines multi-agent models and decision-making methods with the Monte-Carlo Tree Search (stochastic search) method.

Further development of the system and expansion of its functions are planned for 2020.

The system is also used in planning the reconstruction and development of the Baikal-Amur Mainline (BAM), accelerating and facilitating analysis and comparison of various design solutions.

The main efficiency evaluation for the Eastern Polygon are its performance indicators. Thus, over the past 5 years, with an increase in freight turnover by 25.5%, the section speed has increased by 5%, and the average train weight by 3%.