Utilizing real-time networking techniques to optimize urban traffic signals can significantly improve the transportation system performance. The objective of this project is to apply communications technologies and communications networking techniques to control both traffic signals and vehicles. This is an interdisciplinary proposal that will combine the optimization techniques based on uncertainties in the measured data that are used in traffic engineering, with distributed control strategies, based on real-time measurement and data dissemination that are used in communication networks. The research effort is organized into an orderly progression that uses increasing amounts of information and processing complexity to determine the incremental value of the procedures. We will start by optimizing the flows at isolated traffic signals, then progress to flows on arterials, and finally to flows in the entire traffic network. Initially we will control the traffic signals based on real-time flow information, then progress to using the signals to control the paths of vehicles, using deflection routing techniques, and finally perform route planning for individual vehicles. We will use clustering techniques and information reduction techniques, such as fish-eye routing, that are being developed in ad hoc networks, to scale techniques that are applicable to small networks to the traffic networks in large urban areas. The challenges of understanding and influencing traffic control open up new research issues in network flows, communication, optimization, and statistical modeling. All of the procedures will be tested using real data from a selected area of Manhattan.
The broader impacts of this project includes: (1) It will reduce fuel consumption and commute time by reducing the time spent at traffic signals; (2) It will establish collaboration between two complementary areas with the similar goals of increasing throughput and optimizing flows in networks; (3) It will actively engage graduate and undergraduate students by developing learning modules and encouraging minority students to be involved in this interdisciplinary effort. The work will be widely disseminated to the transportation and networking communities.
1. An Analytical Framework for Traffic Signal Control The research team developed a unified framework for optimizing traffic lights accounting for traffic dynamics and real-time traffic flow. In the problem, the decision variable is binary determining which phase to be activated at any simulation interval. The output from the proposed model provides optimal signal timing plans that minimizes the intersection delay and total system travel time. The objective function explicitly considers intersection delay and loss time due to phase switches in addition to traditional travel time objective. Moreover, the formulation increases its applicability by means of considering flexible cycle lengths. Two test networks are used to demonstrate the applicability of the proposed model. Results show better performance of the models when compared to pre-timed optimal signal plans. Figure 1 presents one sample result on the cycle length variation. 2. Adaptive signal control algorithms using real-time data The research team developed and implemented two adaptive signal control algorithms: Enhanced Longest Queue First (ELQF) and Maximal Weight Matching (MWM) within the agent based traffic simulator. The research is motivated to address limitations of the original Longest Queue First (LQF) algorithm. In LQF, one approach with high traffic volume will get green successively. This creates unreasonable delay for vehicles on other approaches. In the ELQF algorithm, we introduce the concept of second best queue and maximum green. The key idea is related to service provisioning for special type of vehicles (e.g. transit, EMS, fire trucks). The algorithm assigns more weight on certain vehicle types (emergency vehicles, buses etc) as compared to other vehicles. The MWM algorithm uses the cumulative value of weights of vehicles in queue to make control decisions. 3. Network traffic control accounting for fairness across users Traditional signal optimization algorithms focus on local information and aim to minimize average delay, number of stops, queue size, etc. for a particular intersection. Instead of looking at the delay incurred in one intersection in the route, our proposed algorithm accounts for the historical delay information of individual vehicles and seeks to minimize the stopped delay for the entire trip. This research proposes trip delay based control algorithms that accounts for the fairness across road users with respect to intersection delay. The fairness objective is to minimize the variation of trip intersection delay for a population of road users within the defined temporal and spatial boundaries. Figure 2 provides one sample result for the ELQF algorithm. 4. A Bi-level Formulation for the Dynamic Equilibrium based Traffic Signal Control The signal settings influence the route choices and departure time choice made by road users and the resulting flow redistribution influences the signal settings. Therefore, it is necessary to consider the interaction between signal settings and route choice behavior of the road users. In this study, we formulate and solve the combined dynamic user equilibrium and signal control (DUESC) as a bi-level optimization problem. The bi-level problem has two basic components: the dynamic user equilibrium (DUE) where road users manage to minimize their travel cost and the dynamic signal control (DSC) problem where the signal timing is adjusted to minimize the system travel time. Further, we propose a heuristic algorithm based on the iterative optimization and assignment (IOA) procedure to obtain the mutual consistency between two consecutive solutions (Figure 3 shows the solutions for one sample intersection). 5. Signal Control with Partial Information from Connected Vehicles In the deployment stage of connected vehicles, traffic flows are mixed with non-connected vehicles and the connected vehicles. The signal controller only accesses the traversing information of connected vehicles and then estimates the traffic state. This study examines the performance of signal control under the mixed connected vehicle environment. Firstly a microscopic traffic flow simulator is developed using a decentralized multi-agent system. Vehicles and signal controllers are modeled as intelligent agents that are capable of making decisions on their own. For the vehicle agent, we have defined car following, lane changing, and the adaptive routing behavioral rules. Connected vehicle agent has an add-in feature that allows it to communicate with other agents. Further, the signal control agent follows the reinforcement learning algorithm to activate different phases of the intersection. In the experiment design, we have selected the Seattle downtown network (shown in Figure 4) as the test network and analyzed the performance of the algorithm (one sample result is shown in Figure 5) under varied demand scenarios. Results show that we can reduce delays by almost 20 percent by optimizing traffic signals.
