A self-learning traffic signal control method for CO2 reduction using prediction of vehicle arrivals

A large part of CO₂ emission comes from road transportation. The amount of CO₂ emission from vehicles can be reduced by controlling traffic flows using traffic signals, because it is strongly affected by vehicular behavior. Although there are a number of traffic signal control systems to reduce travel time, such systems might not be effective to reduce vehicular emission because of the trade-off between stop duration and stop frequency at signals. In this paper, we propose a decentralized signal control technique for reduction of vehicle stops using vehicle arriving information collected by inter-vehicle communication. The proposing technique controls signals based on an evaluation function to predict CO₂ emission amount. The function is gradually improved by an unsupervised learning technique. Through simulation based evaluation, we show that the proposing technique is efficient to reduce CO₂ emissions from vehicles without increasing the average travel time not so much.