Enhancing Ramp Metering Algorithms with the Use of Probability of Breakdown Models

Abstract

Ramp-metering, which controls the onramp flow into the freeway, is successful in increasing freeway throughput and reducing overall travel-time. The maximum flow allowed from the onramp during ramp-metering is typically estimated so that the sum of flows from the onramp and mainline do not exceed a predetermined threshold (either the capacity of the downstream section or a threshold based on occupancy at capacity). Recent research has shown that this threshold is probabilistic and the transition from noncongested to congested conditions (i.e., breakdown) occurs stochastically. Also, research has shown that the contribution of the ramp and freeway demands on breakdown is different; 100 additional vehicles arriving from the ramp increase the probability of breakdown more than 100 additional vehicles from the freeway. This fluctuation has been studied through the development of breakdown probability models, which provide the probability of breakdown as a function of the combination of the mainline and ramp flows. The writers’ objective was to develop suitable site-specific probability of breakdown models and use them within existing ramp-metering algorithms to evaluate their ability to postpone the breakdown and reduce congestion at freeway facilities with recurring congestion. The writers first develop a process for obtaining breakdown-probability models for existing critical ramps (i.e., those where breakdown starts). Next, the writers propose specific enhancements to existing ramp-metering algorithms that incorporate probability-of-breakdown models. Proposed enhancements are presented for two algorithms, as follows: (1) the Minnesota stratified ramp-metering algorithm (SZM), and (2) the Ontario COMPASS algorithm. Simulation was used to replicate these algorithms and evaluate the proposed enhancements. The results of these experiments showed that the enhancements are effective in postponing congestion at the two sites by 17–35 min.