Using the current traffic state as the input, a faster than real-time simulation can be used to preempt congestion and traffic jams. Predictive simulations will be a valuable component of Advanced Traffic Information and Management systems (ATMS) and other Intelligent Transportation Services (ITS). For traffic-state estimation, real world floating car data could be used to initialize simulations and predict how things would unfold. A modern city such as Singapore has a large number of vehicles such as taxis, busses equipped with GPS receivers along with several smart phone users. These vehicles and mobile users act as probes providing valuable information pertaining to their location and speed at fixed intervals. Given the sheer number of vehicles on the entire road network of a city, the location of vehicles is obtained in the form of an unbounded, continuous GPS data-stream. Innovative ways of harnessing the high volume GPS data-stream will aid real-time reconstruction of traffic state on the road network and enable advanced preemptive control mechanisms for reducing congestion.
A Dynamic Data Driven Adaptive Simulation (DDDAS) incorporates real-time data from the physical system (in this case the road network of Singapore) to initialize or steer the simulation system. A Symbiotic Simulation is a special class of DDDAS involving a mutually beneficial relationship between the physical and simulation systems. The physical system provides continuous inputs to steer the simulation which in turn gives recommendations to the physical system. The interested reader can refer to (Darema 2004) for a detailed account on the challenges in terms of incorporating real-time data streams to steer executing simulations. (Schoenharl et al. 2006) and (Celik et al. 2013) are some examples of DDDAS being used for emergency response services using call records and anomaly detection to ensure reliable electricity flow respectively. We believe that DDDAS in ITS has significant potential for effective control mechanisms to optimize traffic flow and provide better time dependent navigational services to users based on predictive simulations.
Model based online traffic state recognition and short term prediction has several applications in ITS based services such as traffic dependent navigational services. Further predictive simulations can help plan and optimize traffic control strategies such as Variable Message Signs (VMS) and dynamic ramp metering which controls the flow of traffic into an expressway at on-ramps through traffic-lights depending upon real-time information. Real-time monitoring and control of traffic could also help implementing robust green waves for certain intersections.
The general workflow for DDDAS in ITS is shown in the figure below. The general work flow uses a large historical archive (typically the last 30 days) to build per road statistics pertaining to flow, density and average velocity at different time intervals of the day. By integrating the historical data and combining it with real-time in memory data for each road, congestion and incidents such as traffic breakdown can be identified. Once a traffic breakdown is detected, a symbiotic traffic simulation is initialized for predicting the time required for the congestion to clear and evaluating alternate scenarios caused by other vehicles rerouting due to the accident.
Real-time traffic-state estimation and short-term prognosis using data-driven simulations offers exiting opportunities for several ITS based services. The challenge is in developing and building robust models and algorithms for achieving a fair degree of accuracy by fusing real time data from several sources. If implemented, such a system could pave the way for better emergency services deployment, dynamic routing and better control strategies for optimizing traffic flow. In my next post I shall be talking about real-time traffic-state recognition from probe vehicles. The details of how the current traffic state can be used for initializing faster than real-time predictive simulations for DDDAS will be elaborated in future posts.