Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/399A82B2-CF6E-49E1-86A7-9EA88EA4046F | ||
| Year | 2016 | ||
| Type of Item | Peer-Reviewed Journal Publication | ||
| License |
|
||
| Bibliographic Citation | U. Aliubavicius, J. Obermaier, W. Fourati, D. Manolis, I. T. Michailidis, C. Diakaki, E. B. Kosmatopoulos and M. Krause, "Use of system of systems and decentralized optimization concepts for integrated traffic control via dynamic signalization and embedded speed recommendation," Transp. Res. Proc., vol. 14, pp. 3416-3425, 2016. doi: 10.1016/j.trpro.2016.05.300 https://doi.org/10.1016/j.trpro.2016.05.300 | ||
| Appears in Collections |
In the frame of the European research project Local4Global, urban traffic control is one of the demonstrative use cases of a developed decentralized control method based on the Technical System of Systems (TSoS) concept and using machine learning capabilities. TSoS concept consists of dividing the system to semiautonomous elementary systems, called constituent systems, which shall enjoy to a major extent a local decision possibility. A remaining part of the decision shall be made after exchanging information between all participating systems to learn from each other and improve the overall performance. In the traffic context, two basic classes of constituent systems are suggested: dynamically signalized traffic junctions and connected vehicles with speed control capabilities. Both traffic signals and vehicle speed controls receive a correction from the L4GCAO global optimizer in a bigger and common control cycle, namely each day. This paper describes the methodology and the results of a VISSIM microscopic traffic simulation of a road section situated near Munich. For the strategy evaluation, the results in terms of the performance index, waiting time per link, coordination proportion, mean network speed and travel time are compared to a baseline. This is during off peak demand a currently running fixed green wave signalization and during rush hour demand on the evening time of day signalization, having additional both demands combined with a speed recommendation with corrections. First results show that during rush hour the overall performance is improved compared to the initial scenario, nevertheless in low demands opposite situation is observed. A general advantage of such method is that it is easily scalable and transposable to other portions of the network. Since machine learning capabilities are introduced, algorithms are self-adaptive to yearly and seasonally varying demand and no important human involvement is needed. An outlook is given, how to transfer the strategy to the real road and test it in a field test.
Copyright © DIAS 2013
