Real-Time Traffic Congestion Prediction

ions to meet these challenges. Data Intensive Network Organization Traffic prediction networks require a combination of big data and large number of devices to combine semi-global historic traffic information and on-line vehicle updates. This requires the convergence of approaches used by the datacenter community and real-time community. Such dynamical systems with real-time constraints require a new approach to system infrastructure design and machineto-machine and machine-to-grid network protocols. As the underlying physical network is unreliable, network protocols must not associate routes with nodes but with primitives such time, position, driving direction and average speed. For example, a spatio-temporal grid structure may be overlaid on the street map to determine communication activity such that messages are delay-bounded along highways. By tightly synchronizing vehicles, we can derive reliable logical abstractions of the network through wireless interference control and end-to-end spatio-temporal schedules across a range of vehicle densities and street topologies. Scalable Real-Time Prediction Algorithms RealTime prediction algorithms are required in a wide array of CPS applications from financial markets and supply chain management to robot path planning and transportation. As stream processing is applied to larger scale problems, the need for real-time guarantees to data-intensive computation has recently surfaced. New approaches to real-time parallel computation and communication are necessary. For example, we used the Nvidia CUDA computation platform to build the AutoMatrix traffic congestion simulator (see Fig. 3). AutoMatrix is capable of simulating traffic on any street map in the US with first-order mobility, communication and traffic Fig. 3. AutoMatrix traffic simulator with 800K vehicles queuing models for analysis of recurring and nonrecurring congestion. By processing vehicle trajectories in parallel on an Nvidia graphics processing unit we are able to compute the fastest path origin-destinations of over 5 million vehicles. While this operation does not have real-time capabilities, the algorithms required have opened a new class of real-time parallel computation. Distributed Trust Management Architectures A framework for decentralization of security policies and policy enforcement is required to facilitate distributed authorization management and re-delegation of authorization in vehicular wireless networks. Key establishment mechanisms based on time, space and neighboring vehicles are necessary to ensure the security and privacy of drivers is guaranteed. Finally, proof-based and theorem-proving data validation algorithms must be used to ensure the integrity of traffic updates and to prevent denial of service