Predicting LNG cargo loss to boil-off at voyage start
A model that predicts a voyage's cargo loss within about ±2% from ten facts known at loading, so hard cargoes are flagged before departure.
Challenge
An LNG carrier operator wanted to know at loading how much of a cargo a voyage would lose to boil-off. Whether a cargo would cause high losses was judged from experience and rules of thumb, crew effort and charterer expectations were set without a forecast, and onboard composition data was sparse.
Solution
Arundo reconstructed 51 voyages on six vessels from sensor and voyage records and trained a regression model on ten features known or estimated at voyage start: cargo volume and pressure, nitrogen content, speed and duration among them. A prototype web app returns a predicted cargo loss for the voyage assumptions entered.
Impact
Cargo loss predicted within about ±2% on held-out voyages; a basis to flag hard-to-handle cargoes to crew and set charterer expectations before departure.