Fujitsu, SMU, and A*STAR collaborate on vessel traffic management technologies with Singapore; AI and big data
Fujitsu Limited, Singapore Management University (SMU), and A*STAR’s Institute of High Performance Computing (IHPC), are collaborating to develop innovative new technologies for vessel traffic management in the Port of Singapore, with the support of the Maritime and Port Authority of Singapore (MPA).
The Straits of Singapore and Malacca comprise one of the world’s busiest sea lanes. At any given moment there are about 1,000 vessels in the Singapore port, with a ship arriving to or leaving Singapore once every 2-3 minutes.
The predictive technologies will leverage artificial intelligence (AI) and big data analytics to optimize the management of Singapore’s port and surrounding waters. The technologies will also be validated using real-world data to improve the forecasting of congestion and identification of potential collisions and other risk hotspots before they occur at sea.
The research and development for these new maritime technologies has been conducted under the guidance of the Urban Computing and Engineering Center of Excellence (UCE CoE), a public-private partnership consisting of the Agency for Science, Technology and Research (A*STAR), SMU, and Fujitsu, that was established in 2014. This collaboration demonstrates the UCE CoE’s continued commitment to harnessing high-performance computing capabilities in the development of solutions for sustainable urban operations, offering another example of how researchers at UCE CoE are using Singapore to test-bed next generation solutions for real-world issues faced by industry and government.
The outcomes of this research and development phase, as well as the practical knowledge and experience gained through the project trials, will be integrated into Fujitsu’s future maritime solutions.
The UCE CoE initiated research and development into technologies for maritime vessel traffic management in 2015, employing the diverse strengths of Fujitsu, IHPC, and SMU. IHPC contributed its capabilities in modeling and simulation, as well as probabilistic modeling and machine learning techniques, while SMU provided its expertise in large-scale multi-agent optimization models. Fujitsu Laboratories leveraged its data analytics and artificial intelligence technologies to support the endeavor.
As a result of the collaboration between Fujitsu, IHPC, and SMU, several key technologies are being developed for improving the management of maritime vessel traffic. These include:
Prediction models, such as a short-term trajectory prediction model that accurately predicts the trajectory of a vessel using machine learning and motion physics and a long-term traffic model that can forecast the traffic situation based on the traffic patterns of a large number of vessel types, derived from historical data.
Risk and hotspot calculation models, such as a risk calculation model that can reliably quantify the near-miss risk of a pair of vessels, by integrating various risk models (ensemble risk model) and a hotspot model that dynamically reveals changing risk hotspots through spatio-temporal data analysis.
Intelligent coordination models, such as a spatial coordination model that seeks to re-route vessels to avoid near-miss and collision incidents and a temporal coordination model that coordinates the passage timing of vessels to reduce hotspots. These models will support real-time decision-making to mitigate predicted risks while minimizing disruptions and ensuring smooth navigation for the vessels.
These technologies will eventually be integrated and test-bedded for their potential to enhance navigational safety, such as the ability to detect and recognize a near-miss risk prior to the event (e.g. 10 minutes beforehand) by combining short-term trajectory prediction with risk calculation. Another target is to forecast and mitigate the dynamically changing hotspot before it is generated (e.g. 30 minutes beforehand) by integrating long-term traffic forecasts, hotspot calculation, and intelligent coordination models.