S.V. Ivanov - Ph. D., Senior Research Scientist, National Research University of Information Technologies, Mechanics and Optics. E-mail: svivanov@mail.ifmo.ru, sergei.v.ivanov@gmail.com
K.V. Knyazkov - Ph. D., Senior Research Scientist, National Research University of Information Technologies, Mechanics and Optics. E-mail: constantinvk@gmail.com br> T.N. Tchurov - Senior Research Scientist, National Research University of Information Technologies, Mechanics and Optics. E-mail: tchurovtim@gmail.com
A.V. Dukhanov - Ph.D., Senior Research Scientist, Associate Professor, National Research University of Information Technologies, Mechanics and Optics. E-mail: dukhanov@niuitmo.ru
A.V. Boukhanovsky - Dr. Sc. (Eng.), Professor, National Research University of Information Technologies, Mechanics and Optics. E-mail: boukhanovsky@mail.ifmo.ru

The development of transport infrastructure in big cities needs a new generation control systems with functions of finding optimal paths in a public transit network taking into account traffic variability obtained during the operational forecast made with the modern mathematical models and detailing up to individual vehicles. In such a formulation the problem of modeling and optimization of public transport routes is a resource-intensive task and requires high performance technologies. However, the using of dedicated supercomputers or distributed resources for those tasks is inefficient by reason of the traffic variability due to multi-scale (daily, weekly, annual) rhythm of the number of vehicles and its routes in the city. Therefore it requires the special computational technologies - Urgent Computing (UC). These technologies allows to build up dynamically configurable distributed computing environment with flexible changing the number of cloud resources according to computational load. In this paper a cloud high performance software package functioning under the concept of UC for the operational traffic forecast and public transport optimization in Saint-Petersburg is announced. For the decision support of public transport control the model approach is used. The system is based on the model of operational traffic forecast working in the 24x7 mode with the regular data assimilation of real traffic data measured at fixed points (for example in the most loaded crossroads). The building of such a model needs the solution of three related tasks: (а) determine the characteristics of the environment - the transport network, (b) the calculation of the characteristics of vehicles, pedestrians and passengers, (c) the optimization of the routes of some categories (or even units) of public transport. To solve the problem of environment identification virtual society and transportation models are used. The model of virtual society is based on statistical data about the use of the territory with taking into account population of the city, the residences of agents, private vehicles, locations of companies and car fleets, public transport depot etc. It allows creating the correspondence network, linking the individual objects in the city. High-performance software system provides simulation of up to 500 thousand agents on the transport network at one moment enabling reproducing specific traffic in megacities. In this case the key feature is the ability to scale computing resources depending on the number of transport agents and time limit for the forecast calculation. Software system provides calculation of correspondence model on the road network, distributed multi-agent simulation of traffic, and optimization of urban transport system. In total it facilitates the information and intelligent decision making support for the traffic control in big urban areas.