Real-time network-level signal timing optimization

Mehrzad Mehrabipour

Signal controllers are key components of transportation networks that manage traffic congestion by minimizing travelers' delay and travel time. Controllers also contribute to managing environmental pollutants and fuel consumption. This contribution has social and economic effects. Signal settings of controllers can be determined by optimizing signal timing variables of formulated problems based on real-world transportation networks. Approaches to solving signal timing optimization problems can be categorized into centralized, decentralized, hierarchical, and distributed-coordinated control approaches. These categories of approaches are neither tractable nor real-time for large-scale networks. Some of the approaches place restrictive assumptions on network topology and movements at intersections or they generate low-quality solutions. In this thesis, a Distributed Optimization and Coordination Algorithm for a Signal Timing Optimization (DOCA-STO) problem is proposed. DOCA-STO is real-time and finds near-optimal values of signal timing variables by solving a mixed integer linear problem formulated based on the Cell Transmission Model (CTM). DOCA-STO distributes a network-level problem to several intersection-level sub-problems. Each sub-problem is an autonomous agent that is optimized separately and in a parallel architecture. This distribution increases the computational efficiency of DOCA-STO. The sub-problems broadcast several pieces of information to their neighbors to create a coordination scheme. The information includes the available capacity of receiving cells and the number of vehicles dispatched from each intersection. This coordination scheme is created using outputs of a CTM simulation for an entire network. The rolling horizon technique is also adopted to provide the possibility of quick responses to network changes and events and increase the efficiency of DOCA-STO by discretizing a study period. The algorithm was tested on various test networks including those of 2, 6 and 20 intersections, and its results were compared with a centralized control system, a decentralized control system, and theoretical upper-bounds. The results are promising and show the capability of DOCA-STO for real-world case studies. The differences between the solutions of the proposed algorithm from globally optimal solutions were at most 1%, a result which validates the accurate performance of the proposed algorithm.

Real-time network-level signal timing optimization

Files and links (1)

Abstract

Metrics

Details