Development and implementation of algorithms used in ground and internet traffic monitoring

Abstract

Two different problems have been analyzed, one involving Traffic Management Centers (TMCs), while the other involves Internet traffic monitoring at a network router. TMCs detect traffic incidents and manage traffic using Automatic Incident Detection (AID) and traffic control algorithms respectively. They use traffic data (speed, volume and occupancy), most of which is obtained from Inductive Loop Detectors (ILDs). The success of these algorithms heavily depends upon the accuracy of this data. The development and real-time implementation of a new algorithm that filters out erroneous data from the ILDs is the focus of the first half of this thesis. The developed algorithm is based solely on vehicle kinematics and is implemented in real-time. Traffic data is used to construct a parameter, namely average vehicle length, and calibrate its threshold range. If this parameter, computed from the data from ILDs, falls outside the calibrated range, an error is indicated. The developed algorithm was run with actual traffic data in real-time over a number of days, for average vehicle length threshold calibration and for test purposes. Tests were conducted to check the accuracy of the algorithm in identifying deliberately introduced errors, and also to estimate its false alarm rate. It was found to have negligible false alarm rate and high accuracy of identifying errors in the loop detector setup. The second part of this research involves Internet traffic monitoring at a router. Internet traffic is predominantly based on the TCP protocol; the TCP protocol is responsive to congestion, but recently various applications based on protocols different from TCP have been developed and are unresponsive to network congestion. Development of new controllers to manage this heterogeneous traffic and ensure fairness of bandwidth distribution between TCP and non-TCP flows at the router is necessary and requires estimating proportion of the two flows at the routers. A scheme for estimating proportion of TCP and non-TCP flow at a router is developed that utilizes queue length and packet drop rate history at the router. Previously developed fluid based model of TCP traffic at the router is extended to include non-TCP flows and MatLab simulations are carried out for algorithm validation. The scheme was unsuccessful due to the inadequate persistence of excitation in queue length and packet drop rate data obtained through MatLab simulations. However it performed well when the Ns Simulator was used to simulate the router traffic.