This paper presents a method for the online prediction of urban congestion patterns including their spatio-temporal propagation based on historic state data. Traffic state data for each link and time interval within the Berlin street network comes from a dynamic route choice and traffic assignment model. From extensive historic traffic state data congestion patterns are generated and classified in an appropriate manner. Based on this analysis, a method was developed to predict the propagation of congestion within the network based on pattern recognition. Significant parts of the network-wide prognosis are selected and sent as messages to the operator of the traffic management centre. A further step identifies actuators at in- and outflow areas of current and predicted congestion in order to increase the outflow from and decrease the inflow to the congested area. Messages for variable message signs are generated automatically and displayed to the operator with other appropriate measures. The work presented was carried out within the German research project IQ Mobility, which was funded by the initiative Verkehrsmanagement 2010 (Traffic Management 2010).
Abstract