Comparison of signalized junction control strategies using individual vehicle position data

摘要

This paper is concerned with the development of control strategies for urban signalized junction that can make use of individual vehicle position data from localization probes on board the vehicles. Strategy development involves simulating the behaviour of vehicles as they negotiate junctions controlled by prototype strategies and evaluating performance. Two strategies are discussed in this paper, a simple auctioning agent strategy and an extended auctioning agent strategy where a machine learning approach is used to enable agents to be trained by a human expert to improve performance. The performance of these two strategies are compared with each other and with the MOVA algorithm in simulated tests. The results show that auctioning agents using individual vehicle position data can out perform MOVA, but that this performance can be improved further still by using learning auctioning agents trained by a human expert.