A Better Way: ITMO Presents AI-Based Service for Public Transit Route Optimization

Cities are constantly expanding with new residential areas and attraction points. This changes people’s demands for public transport: city residents require new routes that will meet their needs and account for existing infrastructure and resources.

For any transportation system to stay relevant, it needs to be regularly updated. Today, the common solution relies on transportation models – specialized software that can describe and predict transportation and passenger flows with mathematical formulas and algorithms. Among such models are PTV Visum or its Russian analog, AnyLogic. Big transport models take into account a variety of factors, from population density and public preferences in separate urban areas to the flow of traffic and public transport schedules. For instance, such software can model the movement of residents of a particular area at different times of day and the way new transport routes will change their behavioral patterns.

To develop such a model, specialists need to conduct preliminary research and collect a large amount of information, analyzing data from mobile operators or statistics of transport pass use. Such work can cost millions of rubles and take several years.

A team from ITMO’s Institute for Artificial Intelligence has developed a tool that can help urban planners and transport engineers evaluate existing above-ground public transport routes, taking into account the costs for passengers and carriers, as well as transport demand – all in mere hours. Based on this analysis, the system provides recommendations for improving routes. At the same time, the solution is not demanding in terms of input data and can predict routes based on information about urban development and the topology of the street-road network. This approach will allow specialists to reduce the time and funding needed for data collection and work on the transportation system, as well as protect the city from poorly planned projects.

The new open-source service is called ConnectPT and is based on an evolutionary algorithm that uses graph neural networks as a mutation operator. The operator is used to introduce random changes into the population of solutions so that the evolutionary algorithm selects optimal routes. A distinctive feature of the tool are graph neural networks, adapted to improve transport connectivity weighted by demand. Usually, route optimization involves two opposing metrics, reflecting the costs for carriers and passengers. Introducing a transport connectivity metric helps improve indicators such as urban area accessibility and the percentage of trips without transfers. Routes are generated on a graph where stops are vertices and paths between them are edges. Graphs based on city blocks are analyzed for demand using measures of population, as well as diversity and density of urban services. This demand is then distributed across stops within walking accessibility radius; then, a correspondence matrix between these stops is constructed.

“One advantage of ConnectPT is that it’s a light and inexpensive tool that can form and explain its recommendations for improving public transport routes based on minimal data available to urban planners and transport engineers. For instance, we use open-source data from OpenStreetMap about residential and non-residential buildings, their capacity and expected population density, location of business centers, production sites, recreation spaces, roads, and stops. ConnectPT generates sets of routes for different types of transport, taking into account transport demand and connectivity to optimize them into the most balanced option,” says Sergey Mityagin, the head of the project and ITMO’s Institute of Design and Urban Studies.

With ConnectPT, specialists can also generate routes by prioritizing various categories of users and tasks. For example, for residents, it may aim to reduce the number of transfers and travel time; for carriers, to eliminate underused routes; and for the city as a whole, to connect existing routes and reduce the number of enclaves without public transport. The tool can analyze data for any populated area or region, so in the future it may be used by urban planners, transport engineers, and authorities of any city. 

At the moment, ConnectPT is an open-source software library. In the future, the team is planning to develop a web interface for the service and implement it into the AI platform Prosto.R, also presented at ITMO in 2025. Prosto.R helps governmental bodies, property developers, and investors to quickly and affordably evaluate the development of a particular area, city, or region – to understand the attractiveness of a particular location and lower the risks of urban planning decisions. The AI platform evaluates multiple parameters: social, engineering, and transport infrastructure, environmental conditions, and the potential use of territory for different functions; it also models urban planning and predicts transport congestion, provision of social infrastructure facilities, and quality of life. Combining these two solutions will make it possible for urban planners to conduct comprehensive territory analysis and receive recommendations for their development.