April 2025
AITPM Academic Partner | Data-driven Digital-twin Model (DDM) for large-scale traffic network management
Current AI-based traffic prediction approaches, while scalable, are typically trained on well-organised, stationary datasets collected over large geographic areas. However, they often fail to capture rapidly changing traffic patterns during local disruptions, leading to biased traffic management decisions. Retraining these models in response to disruptions is impractical due to the limited availability of real-time data. A potential solution lies in traffic flow models capable of accurately predicting environmental changes to mitigate the impact of disruptions on traffic and drivers. However, applying such models at scale presents challenges, as they are computationally intensive and difficult to scale across large networks.
To address this, a team from Monash University including Drs Dong Ngoduy, Hai Vu and Takao Dantsuji has developed an adaptive model-based framework that integrates machine learning techniques to enhance urban traffic operations during disruptions. Unlike traditional black-box AI models that predict future traffic patterns based on past and present data, this approach explicitly incorporates traffic flow models to provide an alternative to black-box models (DDM) as shown in the Figure below. These models function as a digital twin of the traffic network, enabling real-time evaluation of disruptions' impact. The spatiotemporal traffic conditions generated by the traffic model provides rich additional synthetic real-time data (together with the actual data from traffic sensors) as input to the continuous learning of the DDM, which consequently enhances its output accuracy.
This innovative method was first implemented in Ho Chi Minh City (HCMC) to optimize real-time traffic management. As part of a $3 million project, the Monash team collaborated with LK Engineering Ltd. (LKE) to deploy microwave radars and an IoT network across key 36-km corridors in HCMC. The collected radar data was integrated with AI-powered simulation software to monitor and predict traffic conditions in real time. The system is now managed and operated by HCMC Department of Transport’s (DoT) traffic control centre. Following its successful implementation, the HCMC DoT has decided to expand the system to a 188 signalised intersections CBD network, further enhancing its impact on road users.