Voyage Fuel Consumption Prediction
Executive Summary
A leading liner operator replaced simple speed × distance heuristics with machine learning that predicts per-voyage fuel consumption using vessel, weather, and cargo signals. The model feeds a planning simulator that helps chartering, network, and bunker teams test scenarios before committing. Early analysis indicated more than €10M annual savings potential at current fleet scale, with measurable KPI improvement against actual bunker consumption.
The program runs as a joint initiative between Fleet, Network Design, and Treasury to tighten planning accuracy during peak season volatility and reduce unplanned bunkering.
Business Challenge
Improve Fuel Prediction KPI
Enable planners to trust the numbers and act on them with confidence, replacing unreliable heuristic estimates.
Reduce Waste and Over-Purchasing
Limit unplanned bunker calls and avoid costly safety buffers that tie up working capital.
Provide Simulation Workspace
Quantify trade-offs of speed, draft, trim, and routing before a voyage is fixed.
Industry Context
- Fuel costs can represent 40-60% of voyage OPEX depending on market conditions
- Consumption is nonlinear in speed, strongly affected by weather, hull fouling, engine load curves, cargo mix, and trim
- Regulatory pressure around intensity metrics increases the need for auditable predictions
What We Built
Data and Signals
Vessel Data
- • Class, age, engine specs
- • Historical consumption curves
- • Maintenance windows
- • Hull cleaning schedules
Voyage Data
- • Leg geometry
- • Speed profile & RPM bands
- • Trim and ballast state
- • Port times
Weather & Sea State
- • Wind patterns
- • Swell and current data
- • Wave height/direction
- • Along-track conditions
Cargo Information
- • Mass and stowage
- • Reefer share
- • Deck load
- • Container stack height
External Factors
- • Traffic density
- • Congestion indices
- • Seasonal effects
- • Port conditions
Modeling Approach
Core ML Model
Fuses vessel metadata, weather, and cargo features to predict fuel consumption per leg and voyage. Feature engineering tailored to maritime telemetry with continuous tracking against actuals in production.
Calibration & Guardrails
Outlier filtering for bad AIS points, storm flags, and engine offline events to avoid spurious recommendations and ensure model reliability.
Planning and Simulation Tool
The prediction output powers a scenario simulator used by planners and bunker buyers. Users can adjust speed, sequence, trim assumptions, and alternative routings to quantify fuel and time outcomes before issuing final instructions.
Daily Updates
Data pipelines refresh vessel performance baselines continuously
Role-Based UX
Lane templates for fast what-if experiments by different teams
API Integration
Network team embeds predictions in rolling schedule optimizer
Change Management
Started with 6 reference vessels across 3 lanes to prove generalization
Side-by-side comparisons with legacy estimates for 8 weeks before planners relied on model outputs
Short enablement sessions with bunker buyers and port captains to align on thresholds and exceptions
Results and Impact
Operational Outcomes
- Reduced average bunker uplift variance by 20-30% on pilot lanes
- 12% fewer safety buffer top-ups without increasing low-fuel risk
- Earlier bunker nominations improved supplier terms on several calls
- Ongoing tracking in production to ensure gains persist
Financial View
- Savings from right-sized bunkering and route-speed optimization
- Reduced deviation hours through better planning
- Tighter cash forecasting through predictable fuel outlay
- Secondary benefits in supplier negotiations