LNG Diversions Detection Methodology
Overview
LNG diversions occur when a vessel deviates from its originally expected next destination. Kpler provides an automated detection system to identify these diversions and present them in a structured format for clients. This helps enhance transparency and decision-making for shipping professionals.
How Diversions Are Detected
Kpler's LNG diversions detection methodology relies on a combination of automated scripts and data science models. The system works by:
Identifying changes in a vessel’s reported destination via AIS signals.
Analyzing vessel trajectory and heading changes.
Filtering out false positives based on predefined rules.
Insights are provided through API, Excel Add-in, and web app front-end visualization.
Key Indicators of Diversions
A diversion is flagged when:
Port-Call-Based Detection: A vessel loads at a port and signals a discharge destination, but mid-journey updates its destination to a different port.
Example: A vessel loads in the U.S. with Spain as its expected discharge port, but halfway across the Atlantic, it signals for the UK instead.
Heading-Based Detection: A vessel, without an initially stated final destination, makes a significant course change.
Example: A vessel in transit suddenly alters its route in a direction inconsistent with its prior trajectory.
False Positives
The model excludes the following cases from being classified as diversions:
Estimated Model Ports: Initial model estimates (e.g., "U.S.") that later receive a more precise AIS update (e.g., "Qatar").
Fixture-Based Estimates: Discrepancies between estimated discharge zones and actual port arrivals.
Geographical Constraints: Route deviations due to landmasses or congestion do not qualify as diversions.
Holding Patterns Near Ports: LNG vessels often circle outside ports due to boil-off gas consumption requirements. These do not indicate diversions.
Parent-Child Zone Changes: A shift from "Europe" to "Spain" (a sub-zone) is not considered a diversion.
Implementation Stages
1. Semi-Automation (Initial Phase)
Automated scripts flag potential diversions daily.
LNG analysts validate or reject flagged cases manually.
Analysts can manually add diversions via the front end or API.
2. Full Automation (Final Phase)
If the majority of detected diversions prove accurate over time, the manual validation step is phased out.
The system automatically logs and reports detected diversions.
Data Science Approach
The model:
Captures vessel trajectories at three timeframes (48h, 24h, and present).
Uses computed features such as vessel type, last visited port, and heading change.
Trains using historical LNG diversion cases, augmented with synthetic examples.
Applies the LNG-trained model to Liquids and Dry commodities through iterative training.
This methodology ensures accurate LNG diversion detection, supporting Kpler’s mission of delivering actionable insights to the shipping industry.