Collaborative Research: Stochastic Sensing Control Models for Safe and Efficient Traffic Signal Strategies

Purdue University, West Lafayette IN

Investigators

Abstract

Abstract for CMS-0528225 and CMS-0528143 Over the past decade, the ability to obtain detailed, reliable and time-dependent traffic flow data has tended to mostly emphasize strategic level control algorithms for large systems of controllers in the real-time traffic control context. These algorithms are typically targeted at large urban areas and depend on rather extensive and costly detector placements. They do not cost-effectively address isolated intersections in semi-urban or rural areas, where either extensive system-wide detector installation cannot be economically justified, or signalized intersections are spatially sparse precluding a systems perspective. The investigator and his colleagues develop a range of tactical stochastic sensing control models that enhance the safety and efficiency at signalized intersections. They use a probabilistic paradigm that is consistent with state-of-the-art technological capabilities of existing signals, is sensitive to user requirements, and captures the time-dependency and randomness. This new paradigm is incorporated in a probabilistic control layer that works within the existing traffic control framework. It explicitly acknowledges that the arrival process of vehicles to a traffic intersection is stochastic in several respects, representing a shift from the current standard practice of signal control logic where this randomness is implicitly incorporated to some extent through actuated control. However, the current actuated logic is methodologically limited in capturing the vagaries of vehicle headways under multiple lanes, the influence of weather, the effect of vehicle arrivals on the competing phases, as well as in exploiting the rich array of readily-available historical data. It also lacks a robust mechanism to safely terminate the green by ensuring dilemma zone protection. The proposed methodology for signal control logic: (i) is probabilistic to reflect the multiple facets of randomness associated with the operation of signalized intersections, (ii) can exploit valuable historical data in addition to the current conditions reflected by real-time data, (iii) can enhance operational robustness through strategic sensor placements, (iv) can incorporate a holistic view of the intersection rather than focusing on just the intersection approaches corresponding to the current phase, (v) is technology-neutral to accommodate a variety of sensing technologies, and (vi) can react to inclement weather conditions and special events. The study leverages existing state-of-the-art traffic signal laboratory facilities (Purdue University instrumented intersections in West Lafayette and Noblesville, IN as well as the Traffic Operations Laboratory at the University of Tennessee). Rather than just use sensor data passively, the study incorporates data from the instrumented intersections and prototype models and algorithms into actual laboratory-based closed loop signal control systems. This represents a significant technological paradigm for the next generation of sensor-based methodologies that are less tolerant of performance inefficiencies and safety drawbacks for traffic systems. Our society is continuing to demand safer and more efficient roadways. Traditional traffic signal control uses deterministic algorithms which are reliable, but frequently inefficient in their allocation of green time. As traffic congestion increases, it is necessary to obtain more efficiency out of existing systems. This study reduces delay and improves safety at the widely prevalent rural and suburban isolated intersections by developing new probabilistic paradigms that exploit advances in information and sensor technologies. The proposed approaches incorporate a new control layer that is compatible with control infrastructure, thereby allowing direct implementation of this research without large and expensive upgrades to signal system infrastructure. The study reduces dilemma zone exposure and increases operational efficiency. Reduced dilemma zone exposure reduces human factors related crashes. More efficient operations reduce fossil fuel consumption and vehicle emissions. The proposed solutions are a paradigm shift compared to methodological constructs adopted for the past four decades, and are synergistically enabled by the rich array of data afforded by advances in sensor and information technologies. The study also contributes to current efforts on developing a national network of traffic signal control laboratories that leverage capabilities at geographically distributed universities.