Using K-Means Clustering to Improve Traffic Signal Efficacy in an IntelliDrive SM Environment Jay A. Datesh; William T. Scherer Department of Systems and Information Engineering University of Virginia Charlottesville, USA jad3wp; wts@virginia.edu Brian L. Smith Department of Civil and Environmental Engineering University of Virginia Charlottesville, USA briansmith@virginia.edu Abstract—This paper presents an innovative traffic signal control algorithm, the IntelliGreen Algorithm (IGA), that utilizes IntelliDrive technologies to improve the efficacy of traffic signals. The IGA is fully decentralized and takes a novel approach to traffic signal control using k-means clustering. A VISSIM model of a real-world arterial was used to evaluate the IGA and its performance was compared to that of an actuated timing plan. The IGA consistently improved traffic mobility, and sustainability as volumes increased, even at lower IntelliDrive market penetration levels. The results demonstrate the power of IntelliDrive data and that decentralized traffic signal control can achieve system-wide benefits at lower computational costs. Keywords-intelligent transportation systems; traffic signal control; IntelliDrive; k-means clustering; I. INTRODUCTION Commuters in the United States spent 4.8 billion hours stuck in traffic, or approximately one workweek per traveler, in 2009. Factoring in this wasted time and the 3.9 billion gallons of wasted fuel, the overall cost of traffic congestion was $115 billion [1]. Continued urbanization and population growth will exacerbate the traffic congestion problems that exist in cities worldwide. Implementation of Intelligent Transportation Systems (ITS) is regarded as the future of cost-effective means to mitigate traffic congestion caused by poorly timed traffic signals [2]. Recent advances in computing capability and the emergence of artificial intelligence (AI) has shifted research away from centralized methods of traffic signal control towards developing decentralized algorithms (meaning that each intersection acts autonomously). However, most traffic signal researchers have assumed that static traffic detectors, e.g. in-ground loop detectors, were the source of the information about the traffic conditions. These supply a limited amount of information compared to the data that the U.S. DOT’s IntelliDrive SM initiative affords traffic signal control algorithms. Formerly known as Vehicle Infrastructure Integration (VII), IntelliDrive represents the state-of-the-art in ITS. The primary goal of this initiative is to integrate vehicles with the traffic network using emerging wireless technologies, such as dedicated short range communications (DSRC), to produce safer, greener, and more efficient traffic systems [3]. Vehicle-to-Vehicle (V2V) communications provide added safety and Vehicle-to-infrastructure (V2I) communications enable the intersections to leverage data about the instantaneous speed, location, etc. of every vehicle in the network (not just vehicles in the vicinity of a traffic detector), producing more traffic-adaptive signal control. This paper presents the IntelliGreen Algorithm (IGA), a decentralized adaptive traffic signal control algorithm that improves the efficacy of traffic signals utilizing the unprecedented data that IntelliDrive technologies provide. The distinguishing feature of the IGA is its use of k-means clustering to determine the optimal point to end the green phase based only on the traffic pattern present on the arterial. The use of k-means in a traffic signal control algorithm appears to be a novel application. A software interface between MATLAB and VISSIM was used in order to implement the algorithm. The IGA was tested on a model of a 4-intersection segment of US Route 50 in the Washington, D.C. metro area and its performance was compared to that of an actuated signal timing plan. The effects of changes in traffic volume and reduced IntelliDrive market penetration, i.e. the proportion of vehicles that possess IntelliDrive technology, were assessed. The next section will present relevant traffic signal optimization and control research. Section 3 describes the methodology. This includes details on the test network, specifics about the IntelliGreen Algorithm, the performance metrics used, and also the sensitivity analysis that was conducted. The simulation results are presented and discussed in Section 4. The conclusion and future work are contained in Section 5. II. LITERATURE REVIEW Early offline methods for timing traffic signals, such as TRANSYT [4], employed combinatorial optimization algorithms to optimize timing parameters using observed traffic properties, such as average vehicle flow. Besides their computational complexity, the major drawback of such techniques was that the traffic signal parameters were optimized only for specific traffic volumes and slight variations led to large delays – the signals could not adapt their timing plans to the traffic. Instead, the signals would need to be retimed at considerable time and expense. Online This research was supported by the Transportation Pooled Fund Program, study number TPF-5(206).