Predictive Speed Harmonization Using Machine Learning in Traffic Flow with Connected and Automated Vehicles

Abstract

Speed harmonization is an active traffic management strategy used to delay traffic flow breakdown and mitigate congestion by changing speed limits throughout a road segment based on prevailing traffic, weather, and road conditions. Traditional implementations rely on fixed roadway sensors to collect traffic information and variable speed signs at fixed locations to display updated speeds. Moreover, most implementations use a reactive rule-based decision tree to activate the control strategy. This set-up faces three challenges: 1) fixed infrastructure sensors provide an incomplete picture of traffic flow dynamics throughout the segment which can reduce the effectiveness of the strategy, 2) the limited set of scenarios in which speed control can be applied given the reliance on fixed road sensors and signs, significantly affecting performance, and 3) the difficulty in predicting future traffic state using data from fixed traffic sensors. To overcome these challenges, this paper puts forward a predictive speed harmonization system that uses the detailed vehicle trajectories broadcast by connected autonomous vehicles (CAVs) and machine learning techniques to predict the location of traffic congestion. The system relies on a traffic management center (TMC) to collect data from CAVs within a road segment, predict traffic congestion, and broadcast updated speed limits to CAVs to mitigate congestion. Furthermore, the system introduces an optimization-based formulation for speed control to maximize performance. Case studies of multiple operational scenarios show that the proposed speed harmonization system can reduce the severity and length of traffic shockwaves, improve overall traffic stability, increase overall speed, and reduce travel time.