Enabling data-driven vessel traffic services
A vessel traffic service (VTS) is a marine traffic monitoring system established by harbor or port authorities, similar to air traffic control for aircraft. Since more than 80% of world trade is transported by sea, safety and efficiency of maritime transport is essential to the world economy and a key factor in sustainability efforts, concerning both environmental protection as well as safety of life.
The International Association of Marine Aids to Navigation and Lighthouse Authorities (IALA) requires many vessel types to be fitted with the Automatic Identification System (AIS). AIS is an autonomous and continuous broadcast system for ship-to-ship as well as ship-to-shore and shore-to-ship information exchange. AIS data enables human VTS operators to monitor ship movements from their control rooms. However, monitoring ship movements 24 hours a day 7 days a week requires substantial human resources. Therefore, there is a need to support human operators with automated data-driven decision support systems that inform operators of incidents or anomalies, such as intrusion of ships in forbidden areas, violation of the traffic rules and suspicious ship movements. Benefits of data-driven VTS include:
- Increased situation awareness through automated anomaly detection
- Increased navigation safety through earlier risk awareness based on trajectory prediction
- Increased efficiency through faster in and outbound traffic in ports and VTS areas, reduced waiting times and fuel combustion
- Improved port operations (e.g. pilot and berth planning) and logistics through arrival time predictions using statistical models trained with historical AIS data
Automated anomaly detection is based on the idea that a machine learning model can learn the “normal” vessel traffic situation in a port or VTS area using historical ship movement (AIS) data [Graser et al., 2020a]. The model can then be used to detect deviations from the expected traffic situation. It consists of a large number of so-called "prototypes", each representing a characteristic local movement pattern. Each prototype contains statistics on speed, heading and other ship information (e.g. the navigational status). The prototypes differentiate between ship types (in particular: tankers, cargo and passengers) in order to enable type-specific predictions and anomaly reports. This enables the system to automatically raise warnings, for example, when a tanker vessel enters an area which is usually not frequented by tankers (e.g. a nature reserve).
Figure 1: Vessel traffic patterns near the port of Gothenburg as modeled using M³ prototypes and flows [Graser et al., 2020a]
Data-driven trajectory prediction uses information learned from historical ship movement data to predict which path a ship will take. The difficulty of this prediction task varies widely depending on ship types and geographical area. For example, large ships on the open ocean usually move on a straight course with constant speed. In coastal areas and dense traffic situations, however, movements become much harder to predict [Graser et al., 2019]. The data-driven trajectory prediction algorithm aims to identify how similar ships moved. Using the ship's most recent AIS record, the algorithm first determines the best matching prototype based on the type of ship, location, speed, and direction of movement. Then, a kinetic prediction based on the ship's current position and speed is combined with information from the prototype, including flow information to the most likely next prototype.
Figure 2: 20 minute predictions for a ferry entering Gothenburg: linear prediction (blue) and data-driven trajectory prediction (pink)
Estimated time of arrival (ETA) predictions are essential to optimize all logistics processes that are necessary to guide ships into port and to unload the transported goods. Historical ship movement data provides valuable information about past travel times that is more reliable than the estimates provided by ship agents [Parolas et al., 2017].
Data-driven ETA prediction uses previously observed vessel trips to determine expected travel times from any point in the analysis area to a target port. Specifically, each prototype models the relationship between the ship’s current speed and the remaining travel time. Using the most recent AIS record, the algorithm determines the best-matching prototype’s travel time information. Depending on use case requirements, predictions can provide the most likely arrival time or optimistic “best case” or pessimistic “worst case” arrival times. For example, the goal could be to determine if the predicted ETA still agrees with the reported ETA that has been provided to the VTS operator by the ship’s agent. If the probability that the ship will arrive in time falls below a certain threshold value, the ETA can be used to update berth planning and to optimize hinterland logistics operations.
References
[Graser et al., 2019] Graser, A., Schmidt, J., Dragaschnig, M., Widhalm, P. (2019). Data-driven Trajectory Prediction and Spatial Variability of Prediction Performance in Maritime Location Based Services, LBS 2019, 11-13 November 2019, Vienna, Austria.
[Graser et al., 2020a] Graser. A., Widhalm, P., & Dragaschnig, M. (2020). The M³ massive movement model: a distributed incrementally updatable solution for big movement data exploration. International Journal of Geographical Information Science, 34(12), 2517-2540. doi:10.1080/13658816.2020.1776293.
[Graser et al., 2020b] Graser, A., Widhalm, P., & Dragaschnig, M. (2020). Extracting Patterns from Large Movement Datasets. GI_Forum – Journal of Geographic Information Science, 1-2020, 153-163. doi:10.1553/giscience2020_01_s153.
[Parolas et al., 2017] Parolas, I., Tavasszy, L., and Kourounioti, I. (2017). Prediction of Vessel’s Estimated Time of Arrival (ETA) in Container Terminals - A Case Study in the Port of Rotterdam. In TRB 96th Annual Meeting Compendium of Papers, Washington DC, United States.
