To main content

Unsupervised Learning of Vessel Operating Modes for Deployment-Compatible Local Models for Maritime Digital Twins

Abstract

Maritime Digital Twins must represent vessel behaviour across diverse operating and environmental conditions while supporting forward-looking applications such as weather routing and operational simulation. This presentation introduces a data-driven modelling workflow developed in the Horizon Europe TwinShip project to prepare high-dimensional vessel data for such applications. The workflow first applies Gaussian mixture modelling to identify localized and interpretable operating regions within historical vessel data. A key challenge is that some measurements available during model development are unavailable when simulating a future route or operating condition. Operating modes are therefore discovered using a rich historical feature set that includes propulsion-response measurements, then reconstructed in a reduced feature space containing only variables available in downstream applications. Singular value decomposition and regression are subsequently applied within the localized regions to identify dominant relationships and develop models of vessel behaviour. The methodology has been applied across datasets from three pilot-vessel types. The presentation discusses the common modelling pipeline, the interpretation of statistical clusters as operating modes, and the challenges of developing localized models under practical data and deployment constraints. The work demonstrates how unsupervised learning and localized modelling can connect operational vessel data with simulation-ready Digital Twins and AI-supported maritime decision systems.

Category

Conference lecture

Language

English

Author(s)

Affiliation

  • SINTEF Ocean / Energi og transport

Presented at

Open Simulation Platform (OSP) Conference 2026

Place

Norwegian Maritime Competence Center (NMCC) in Ålesund

Date

10.11.2026 - 11.11.2026

Organizer

GCE Blue Maritime Cluster, Norsk Katapult Digital, SINTEF Nordvest, NTNU and Kongsberg Maritime AS

Date

10.11.2026

Year

2026

View this publication at Norwegian Research Information Repository