Date of Award

December 2017

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Automotive Engineering

Committee Member

Beshah Ayalew

Committee Member

Andrej Ivanco

Committee Member

Ardalan Vahidi

Committee Member

Pierluigi Pisu

Abstract

Mimicking the autonomous behaviors of animals and their adaptability to changing or foreign environments lead to the development of swarm intelligence techniques such as ant colony optimization (ACO) and particle swarm optimization (PSO) now widely used to tackle a variety of optimization problems. The aim of this dissertation is to develop an alternative swarm intelligence model geared toward decentralized congestion avoidance and to determine qualities of the model suitable for use in a transportation network.

A microscopic multi-agent interaction network inspired by insect foraging behaviors, especially ants, was developed and consequently adapted to prioritize the avoidance of congestion, evaluated as perceived density of other agents in the immediate environment extrapolated from the occurrence of direct interactions between agents, while foraging for food outside the base/nest. The agents eschew pheromone trails or other forms of stigmergic communication in favor of these direct interactions whose rate is the primary motivator for the agents' decision making process.

The decision making process at the core of the multi-agent interaction network is consequently transferred to transportation networks utilizing vehicular ad-hoc networks (VANETs) for communication between vehicles. Direct interactions are replaced by dedicated short range communications for wireless access in vehicular environments (DSRC/WAVE) messages used for a variety of applications like left turn assist, intersection collision avoidance, or cooperative adaptive cruise control. Each vehicle correlates the traffic on the wireless network with congestion in the transportation network and consequently decides whether to reroute and, if so, what alternate route to take in a decentralized, non-deterministic manner. The algorithm has been shown to increase throughput and decrease mean travel times significantly while not requiring access to centralized infrastructure or up-to-date traffic information.

Share

COinS