Signal timing

Non-recurrent events, such as lane blockage incidents or demand surge due to traffic diversion or rerouting, can increase the congestion on signalized arterial streets, resulting in long queues, queue spillbacks, and significant vehicle delays. This paper compares two methods for the selection of signal timing plans for activation during these events. First, the paper introduces a multi-objective optimization model to determine the signal timing plans considering the performance on the impacted arterial intersection approaches as well as the whole intersection performance measures during the non-recurrent events. The multi-objective optimization problem is solved via a simulation-based optimization utilizing the Non-Dominated Sorting Genetic Algorithm (NSGA-III) algorithm to find a set of Pareto optimal fronts. The Pareto optimal fronts allow trade-offs among various objectives of the optimization. Microscopic simulation models are developed for use as part of the optimization and calibrated using high-resolution controller data to better replicate real-world conditions. The performance of the resulting signal plans is compared to the performance selected using a second approach previously developed by the authors that use machine learning to emulate signal timing expert's decisions during non-recurrent events. The evaluation results show that, although both approaches can improve the performance during non-recurrent congestion, the special signal timing plans obtained from the optimization method can produce better results in improving system performance during non-recurrent events in terms of travel time, intersection delay, throughputs, and queue lengths.