This is a post by Maritime Data made from a recent Vortexa white paper, Piecing together the puzzle of the Global Energy Markets that supported the recent release of their new Destination Model.
The Challenge
Predicting a tanker's destination is not merely about data points but serves as the foundation for understanding oil and gas flows, vessel availability, and the global energy market dynamics. These predictions are vital for clients, offering them invaluable insights that help manage the complexities of the energy industry with confidence and foresight.
Volatility in Energy Markets
The energy market is subject to constant changes due to demand fluctuations, geopolitical events, regulatory updates, and even natural disasters. This volatility adds an additional layer of complexity to predicting vessel routes and schedules.
Confidentiality and Competitive Advantage
Shipping routes and schedules are usually confidential, aimed at providing energy companies with a competitive edge. This makes the task of predicting a vessel's destination even more challenging.
Dynamic Changes in Vessel Routes
Vessels often change their declared destinations mid-voyage due to new orders, usually spurred by shifts in cargo buyers. This practice is prevalent in the industry and is driven by the desire to maximize profits.
Arbitrage Opportunities
Price fluctuations in the international scope of the energy market can lead to arbitrage opportunities. A cargo initially bound for one location could be rerouted to another if a significant price differential emerges, complicating predictions further.
Technological Limitations
Technologies like the Automatic Identification System (AIS) can be unreliable, providing incomplete or noisy data. In some cases, information may be deliberately falsified, or tracking systems tampered with, to maintain secrecy or create confusion.
Geopolitical and Regulatory Considerations
Countries under international sanctions may take roundabout routes or deactivate their AIS transponders to evade scrutiny and detection, further complicating the task of prediction.
Multifaceted Logistics
The process of transporting energy can include multiple stops, transfers, and possibly interim storage, adding another layer of complexity.
The Solution
1. Data Ingestion & Augmentation
Vortexa validates and cleans the incoming data to create a 'super-ping', a consolidated data point that captures a vessel's movement.
2. Data Enrichment and Feature Generation
'Super-pings' are enriched with additional details such as vessel history and fixtures (agreements between charterers and shipowners).
Manual adjustments, known as 'overrides', are incorporated into the models, adding a layer of human expertise.
The data is transformed into a normalized numerical vector, using techniques such as Natural Language Processing (NLP) and one-hot encoding.
3. The Destination Model Pipelines
The data undergoes further enrichment and then serves as input for predictive modeling.
The pipelines for training the model and making inferences are closely aligned to ensure consistency.
4. Neural Network Modelling
Neural Networks (NNs) form the core of the predictive model, designed to simulate human brain functions.
The NNs employ layers of interconnected nodes to process information, and their weights are adjusted through a process called backpropagation.
5. Evaluation
The model is evaluated based on three main criteria: Accuracy, Coverage, and Volatility.
Once evaluated, models are versioned and stored for reproducibility.
6. Final Output
The model produces a probability distribution that represents the likelihood of a vessel heading to specific ports or STS zones.
To make these predictions more comprehensible, the probabilities are aggregated on a per-vector basis.
7. ETA Prediction
Once a prediction is generated, it is enriched with an Estimated Time of Arrival (ETA) using a separate, battle-tested service.
For fine-tuning the ETA prediction, another model that predicts port congestion and potential delays is leveraged.
Through this comprehensive approach, Vortexa aims to provide a solution that is technologically advanced, accurate, and reliable, while also being deeply rooted in industry expertise.
For more details + segments we didn’t cover in our summary such as:
How the industry currently approaches this challenge
The results they have seen from the upgrade
A look to future work
Comments