An end-to-end data driven model for coordinated ramp metering systems

Abstract

Model based ramp metering systems have long been in use to alleviate congestion in freeways. However, data driven ramp metering systems potentially offer improved performance and greater flexibility in usage. An end-to-end data driven model for ramp metering using multi-layer feedforward neural networks is presented. The proposed model is trained using traffic and metering data from the M3 freeway in Brisbane, Australia, where a coordinated ramp metering algorithm is currently used to generate metering rates. The results indicate that the proposed model has the capacity to learn the underlying metering scheme from the data, thereby enabling a shift from a model based system to a data driven one. Further work such as training models to learn more complex metering systems and using feedback mechanisms to create continually improving models are also discussed.