Distributed traffic control and optimization for a large-scale urban network

Abstract

Since urban areas are centers of population and economy, the traffic demands increase dramatically but the road infrastructures are rarely expanded. Effective coordination of traffic flows is a uniquely possible way to avoid traffic congestion. Taking the advancement in sensing and communication techniques, distributed model predictive control (MPC) schemes become available in controlling a large-scale urban network (UN). The scope of this thesis is to design distributed MPC traffic signal control methods to exploit the capacity of existing transportation infrastructures in alleviating the traffic congestion risks and minimizing travel time delay. Modifying a well-known store-and-forward model, we introduce a simple but effective traffic control model for an urban road network to employ the MPC fashion and assess performance indexes in urban traffic control. The considered UN is described as a directed graph of junction nodes connecting by road links. The traffic states are the numbers of vehicles in road links and the control inputs are downstream traffic flows. As the number of vehicles leaving one road toward its neighboring roads is directly proportional to its assigned green time duration, traffic signal splits are determined from computed downstream traffic flows. Safety constraints on roads occupancy ratios and reliability constraints on downstream traffic flows and signal splits are required to guarantee smooth operations for all roads and junctions. In addition, we also provide a distributed minimum-time method to determine the steady downstream traffic flows of road links when the exogenous traffic flows (traffic demands and disturbance flows) are given. In this thesis, we first assume the near future traffic model parameters (including demands, disturbance flows and turning ratios) are known to introduce a nominal MPC traffic signal control problem. The main objective is to optimize the operation of the UN in improving traffic conditions. Two types of uncertainties in the estimation of traffic model parameters are considered. If they are given in small intervals, the considered MPC traffic signal control problem has a robust approximated formulation. When these parameters are considered as random variables with known expectations and variances, we use the distributionally robust chance constraint concept to formulate the stochastic version of the nominal traffic signal control and optimization problem. Thanks for linearity of the traffic control model, all the interested MPC traffic control problems are quadratic programs with linear and second-order cone constraints. By applying popular distributed optimization algorithms, we design two MPC traffic signal control methods in distributed manners. In the first one, we use an accelerated gradient method and dual decomposition technique to design update laws for individual road links and internal junctions. In the second one, we considered the UN as the union of multiple subnetworks and use alternating direction method of multipliers (ADMM) to design a distributed algorithm for (controller of) each subnetwork. Although the optimal control decisions derived from the proposed MPC traffic signal control methods are corresponding to a simple linear traffic model, their effectiveness is verified by simulations in VISSIM software.