[Paper Review] Determining offshore wind installation times using machine learning and open data

Bo Tranberg, Kasper Koops Kratmann+1 more|arXiv (Cornell University)|Sep 25, 2019

Maritime Navigation and Safety15 references4 citations

TL;DR

This paper proposes a machine learning approach to automatically determine offshore wind turbine installation times using open-access AIS data from jackup vessels, without prior knowledge of turbine locations. By clustering vessel trajectories, the method identifies installation, transit, and harbor phases, revealing that 57.2% of installation time is spent installing turbines, with significant variability linked to wind speed.

ABSTRACT

The installation process of offshore wind turbines requires the use of expensive jack-up vessels. These vessels regularly report their position via the Automatic Identification System (AIS). This paper introduces a novel approach of applying machine learning to AIS data from jack-up vessels. We apply the new method to 13 offshore wind farms in Danish, German and British waters. For each of the wind farms we identify individual turbine locations, individual installation times, time in transit and time in harbor for the respective vessel. This is done in an automated way exclusively using AIS data with no prior knowledge of turbine locations, thus enabling a detailed description of the entire installation process.

Motivation & Objective

• To develop an automated method for estimating offshore wind installation times using publicly available AIS data.
• To identify turbine installation times, transit durations, and harbor stays without prior knowledge of turbine locations.
• To reduce uncertainty in cost estimation for future offshore wind projects by providing data-driven time breakdowns.
• To enable benchmarking of jackup vessel performance across different wind farms and projects.
• To explore the impact of weather conditions, particularly wind speed, on installation efficiency.

Proposed method

• Collecting high-resolution AIS data (position, speed, course) from jackup vessels operating in Danish, German, and British waters.
• Applying a clustering algorithm to group vessel positions into distinct operational phases: installation, transit, and harbor.
• Using spatial and temporal clustering to infer turbine locations and installation events from vessel movement patterns.
• Validating cluster assignments by analyzing temporal sequences and spatial proximity to known wind farm layouts.
• Integrating wind speed data from internal Siemens Gamesa databases to assess weather impacts on installation duration.
• Automating the entire pipeline to eliminate reliance on site-specific scripting and reduce maintenance overhead.

Experimental results

Research questions

• RQ1 Can machine learning techniques accurately identify turbine installation events from AIS data without prior knowledge of turbine locations?
• RQ2 What proportion of total installation time is spent on actual turbine installation versus transit and harbor operations?
• RQ3 How does wind speed influence the duration and variability of installation times across different offshore wind farms?
• RQ4 To what extent can this automated method reduce uncertainty in cost estimation for future offshore wind projects?
• RQ5 How do installation times and vessel performance vary across different wind farms and jackup vessels?

Key findings

• The method successfully identified 13 offshore wind farms' installation, transit, and harbor phases using only AIS data, with no prior knowledge of turbine locations.
• For Horns Rev 3, 57.2% of the total 4,932.5 hours of installation time was spent on actual turbine installation, while 35.1% was spent in harbor.
• Average installation time per turbine at Horns Rev 3 was 100.6 hours, significantly higher than previous estimates due to inclusion of transit and harbor times.
• Wind speed showed a nonlinear relationship with installation time, with increased variability in installation duration at higher wind speeds.
• The machine learning approach reduced maintenance overhead compared to traditional scripting methods while maintaining comparable accuracy.
• The method enables performance benchmarking across wind farms and vessels, supporting iterative improvement of installation time estimates.