Smart traffic signal control for multi-junctions
Date of Issue2016-03-22
School of Electrical and Electronic Engineering
Singapore-MIT Alliance Programme
Singapore-MIT Alliance for Research and Technology (SMART)
Along with economic growth and more affluence of population, traffic congestion has become a severe and growing problem in modern cities since more and more vehicles enter urban public road network with limited capacities. Traffic signal management is recognized as one of the most efficient ways to improve urban network congestion. Various signal control strategies have been proposed and applied all over the world in the past decades, which greatly improve traffic conditions. Whereas most of these strategies may not be capable for saturated conditions. Once road is congested, the throughput capacity decreases and in turn causes further deterioration of network congestion. Therefore, intelligent urban traffic control is an urgent need for regulating urban traffic systems to satisfy large fluctuating traffic demands. Furthermore, new motivations for traffic signal control research are the coming of new technologies such as fast communication systems, accurate sensing and network systems etc. \\ This thesis investigates the phase-based repetitive characteristic of historical traffic flow patterns and proposes Iterative Tuning (IT) strategy with anticipation of traffic demand. In general, traffic signals comprise four parameters: phase split, offset, cycle time and phase specifications. For phase split, IT strategy balances the degree of saturations for all phases of every junction in the urban network iteratively and automatically by taking evolution of traffic patterns into consideration. Rigorous analysis provides the sufficient conditions for guaranteeing the convergence of IT strategy globally over repetitions with traffic variations. With respect to offset, traffic signals synchronizes neighboring junctions to ensure as few vehicles stopped by red signal as possible. Repetitive histograms of traffic demand are proposed to estimate the number of queued vehicles off-line and seek the offset to minimize total queued vehicles on two-way links between junctions. The IT strategy smartly integrates phase split and offset tuning to provide suitable signal schedules for the entire day. \\ To ensure the robustness against non-repetitive day-to-day variations, Junctionbased Model Predictive Control (JMPC) strategy is proposed. Traffic dynamics is modeled as lane-group-based Frequently and Regularly Initialized Difference Equation (FRIDE) to make short-term predictions, which decomposes an urban network into decentralized junction-based subsystems. Every subsystem has its own controller to work cooperatively towards system-wide control objectives by coordinating traffic signals of neighboring junctions. IT and JMPC are supplementary to each other for repetitive traffic flow patterns and traffic pattern variations, which effectively and efficiently control phase split in a pre-emptive and predictive manner.\\ In this thesis, aforementioned strategies are tested and validated based on real urban areas in Singapore. A test-bed is setup in simulation software named Vissim, to model an actual urban road network. According to real traffic demand, timesliced origin-destination pairs are estimated to simulate traffic conditions during morning peak hours, afternoon non-peak hours and evening peak hours. Simulation results reveal the efficiency and convergence of proposed signal control strategies on scenarios with small and large day-to-day variations, respectively.