New research for improved performance characterisation of foamed bitumen stabilised granular bases

This presentation introduces our recent research focused on real-time traffic incident prediction, tailored for large-scale network-wide road operations and disruption management. Traditional methods for incident prediction at a network scale often involve building separate models for each road segment or using one large model for the entire network, both of which can be costly and inefficient. Our study proposes an innovative ‘sub-area’ level incident prediction model capable of forecasting incidents within any given sub-area across the network using a single model. By leveraging deep learning and graph neural network techniques, this model effectively captures the relationships between different road links within each sub-area and predicts whether an incident will occur within a given sub-area based on link traffic conditions, temporal factors, and other contextual information such as holidays. Through rigorous experimentation and evaluation using real-world data, we demonstrate the superior performance of our model compared to traditional benchmark models across different scenarios, including various prediction horizons and different study networks like Brisbane and Gold Coast. This presentation aims to illustrate the real-world application of machine learning in enhancing real-time traffic incident prediction for network operations, providing road operators and traffic engineers with insights into the current research landscape and state-of-the-art modelling techniques.
This research project was conducted in collaboration with the University of Queensland, the Department of Transport and Main Roads, and Transmax under the Australian Research Council (ARC) Linkage Project in Real-time Analytics for Road Traffic Management.