RESEARCH

We reformulated the signal timing optimization problem from a central architecture to a decentralized approach, where a mathematical program controls the timing of only a single intersection. We developed a mixed integer linear programming (MILP) formulation and developed solution technique for efficient coordination among intersections' decisions. This scalable solution technique relies on collecting connected vehicle information at each time step, optimize signal timing parameters over a prediction period, and decide the signal state at an intersection.

A stochastic gradient-based optimization model is proposed to offer a functional and feasible way to accommodate nonconvex energy consumption bounds within a signal control optimization model to achieve maximal mobility with minimal energy consumption. The signal control problem is formulated as a mixed-integer linear program, which incorporates nonconvex constraints to limit the total energy consumption in the network. A modified Stochastic Perturbation Simultaneous Approximation (SPSA) is proposed by transforming constrained optimization to unconstrained formulation with quadratic penalty terms. At each iteration, SPSA approximates gradient and first-order derivative of decision variables by cell transmission model simulation to search the direction of the optimal decision. Our novel solution technique allows the traffic control system to achieve certain level of reduction in energy consumption while improving mobility performance in the network through signal control.

Real-time adaptive traffic signal control systems that utilize Connected Vehicle (CV) data have shown a great potential to enhance network-level traffic operations. However, the improvement depends in part on data availability, which is a function of the CV market penetration rate and traffic volume. Currently, a high CV market penetration rate (20% - 50%) is required to outperform the existing point detector-based signal control systems. However, the market penetration rate of CVs is not anticipated to get to the required level in the foreseeable future. Therefore, there is a need for methodologies that can work with low CV market penetration rates and match or outperform the existing traffic control systems. To achieve this goal, accurate traffic state estimation algorithms are required. We proposed two algorithms to account for the partial information to estimate the traffic state in the network under. Both methods rely on integrating high-resolution CV data with low-resolution point detector data in congruity. Both network state estimation algorithms are incorporated in adaptive traffic signal control methodology. The results show that both traffic state estimation algorithms are effective under all CV market penetration rates

A common way to facilitate the movement of transit vehicles at signalized intersections is to grant them a priority over passenger cars. Existing transit signal priority-based methods accommodate transit vehicles on a movement by either extending the associated green time or reducing the duration of the red signal on other movements. However, the existing methods become less efficient when several transit vehicles arrive at an intersection on conflicting movements that compete for the green time. Rule-based priority control methods grant priority to transit vehicle regardless of the condition of non-transit vehicles; therefore, they may negatively impact the performance of other vehicle classes in the network. This research presents a traffic signal control system that prioritize the movement of transit vehicles in an intersection.

We developed a two-stage regression model that predicts crash severity and rate by integrating linear and logistic regression model. The proposed model:

• Provides an unbiased crash rate estimation without using complex data structure assumptions.

• Works with datasets with excessive number of segments with zero observed crashes.

• Estimates the causal effects of time-varying roadway surface conditions in a regression model more accurately.

Objective:

• • • predict crash frequency based on roadway, environmental and traffic characteristics.

Contribution:

• • • developed a BN to predict total crash rate based on geometric roadway characteristics and AADT;

    • proposed a forward selection approach to reduce the number of variables in BN, and

    • estimated the causal effects of contributing factors to traffic crashes.

Summary:

• • • influential variables: segment length, AADT, number of lanes, lane and shoulder width, grade

    • selected BN significantly outperformed the benchmark negative binomial regression model

    • safer road attributes: wider shoulders, and three 12-foot lanes per direction.