Pilot 1: Ship Modelling for Global Vessel Traffic Monitoring and Management
Accurate vessel tracking and route forecasting is of high importance for the stakeholders in the maritime industry. Most applied short-term route prediction approaches are based on simplistic kinematic models and historical information, decoupled from the actual characteristics or dynamic conditions of the trip and the capabilities of the vessel (e.g., vessel type and characteristics, loading and weather conditions, etc.). Pilot 1 improves the accuracy of predicting the location and movement of vessels and fleets by replacing the simplistic linear models, with improved data-driven models, for short-term and long-term route forecasting and for enhanced vessel traffic monitoring and increased maritime safety. Finally, these solutions are delivered to the end users through the VesselAI Visualization Platform for Maritime Situational Awareness.
Use Cases
During the first evaluation period of the project, the focus of the pilot was on the following use cases
Live monitoring of traffic in an extended area of interest for multiple vessels
Deliver an open and trusted platform capable of distributed data analysis workflows combining data-intensive capabilities, fuelling shore-side decision support and next-generation maritime services
Accurate short- and long-term estimation of routes of single or multiple vessels
Design and develop the appropriate software and hardware components, tools and libraries that will enable the efficient ingestion, curation and querying of extreme-scale data sets coming from different sources
Short- and long-term estimation of traffic in an area of interest
Deliver a set of trained, high-quality interpretable ML / DL models by exploiting extreme-scale data sets for advanced classifications, analysis and forecasts related to maritime entities
Pilot-participating Partners
Pilot Technological Highlights
The technological highlights of this pilot were the following
For the evaluation of the pilot application during the first evaluation phase of the project, real streaming data from the Peiraeus port were used coming from the MT’s AIS receivers. For the evaluation of the short-term prediction model, several open AIS datasets were used by UPRC, while MT also evaluated the performance of the model with AIS datasets from the Aegean Sea. Finally, the evaluation of the long-term route prediction model was conducted using an AIS dataset of 4 years with all container vessels of the global fleet.
Data
Several technologies produced in VesselAI were tested by MT during the first evaluation phase. MT’s major interest was to identify if and how the produced services could benefit the company facilitating several model creation related activities. To this end, the following components were evaluated (details of the evaluation will be published in D6.4): (a) Data Storage and Querying Service, (b), Data ingestion and Harmonization Service, (c) Advanced Visualization and Reporting Engine, (d), Data Exploration Service, (e) VesselAI Launcher, (f) Service Execution & Orchestration, (g) AI Model Evaluation and Serving Framework, and (h) the End-to-End Security Framework. Furthermore, the short-term prediction model developed using TensorFlow was integrated into the Akka.io distributed framework, which is widely used for the MT’s model serving activities according to the software stack of the company.
Components
- VesselAI develops components that can help data scientists working on the maritime industry in different aspects of their everyday work.
- Both developed short-term and long-term route prediction models outperform the state-of-the-art approaches (from 23% to 42%)
- MT using the Akka-based architecture was able to accurately predict in the short-term the position of more than 100K vessels with a latency of less than 0.5 seconds on average. In the next phases of the project, the pilot intends to test this solution targeting the global fleet
- The external evaluators of the pilot application were in general satisfied with its benefits on monitoring and predicting the traffic of the fleet. They also provided meaningful recommendations for the refinement of the application in the next development phases of the project.
Highlights
Pilot Results
The solutions provided by Pilot 1 cater to both data experts and non-technical maritime industry users, provide adaptable deployment options for overcoming IPR and legal constraints, ensure seamless integration of major VesselAI Launcher services with existing technology stacks, and reduce the gap of knowledge for maritime SMEs through integration and demonstration via AI4EU and the VesselAI Launcher.