Op-Ed: Is vessel modeling a solved problem?

Angus Whiston. (Credit: DeepSea Technologies)

By Angus Whiston, Communications Director, DeepSea Technologies

In the world of commercial shipping, accurate vessel modeling has always been an essential yet elusive goal. Whether for operational efficiency, technical optimization, or commercial decision-making, the ability to reliably predict vessel performance has always been understood to be a game-changer. A high-fidelity model allows operators to cut fuel costs, reduce emissions, and comply with tightening regulations—benefits that translate directly to both profitability and sustainability. But achieving this level of accuracy has historically been fraught with challenges.

The Challenge of Vessel Modeling

For decades, vessel modelling has relied on a variety of “traditional” techniques to create fuel tables, static performance curves, and empirical formulas—all of which, to a greater or lesser degree, fail to capture the full complexity of a ship’s behavior at sea. These methods struggle because vessel performance is shaped by an intricate web of co-dependent variables, from hull fouling and engine efficiency to wind resistance and wave patterns.

Traditional approaches have been unable to make sense of this vast, interrelated dataset, leading to models that provide only a rough approximation of reality. The consequence? Sub-optimal decision-making, wasted fuel, unnecessary emissions, and a lack of trust in data-driven insights. The industry has long needed a breakthrough.

A Major Step Forward: EPS and DeepSea’s 2024 Breakthrough

In late 2024, DeepSea and Eastern Pacific Shipping (EPS), delivered one of the most compelling demonstrations yet of what is possible with artificial intelligence in vessel performance modelling. EPS, a long-time proponent of data-driven decision-making, partnered with DeepSea to build AI-powered vessel models with over 99% accuracy—setting a new benchmark for the industry.

Using DeepSea’s Cassandra vessel performance platform, the project created highly detailed, individualized digital twins of EPS vessels. These models enabled fleet-wide performance enhancements, with the objective of driving significant efficiency and decision-making gains. The published case study showcased the enormous potential of AI to transform vessel performance analysis. But does this mean vessel modelling is now a solved problem?

The Reality Behind the 99% Figure

While the results from the EPS project were impressive, they also highlighted a crucial point: a model is only as good as the data fed into it. In many cases, the true barrier to vessel modelling accuracy isn’t the modelling technology itself—it’s the quality of the underlying data.

EPS had an advantage, in that it already had a well-established data pipeline with numerous verification processes to maximize data accuracy. Working with DeepSea, this data was further cleansed and validated, until it was capable of high-fidelity AI modeling. This clean, structured data was a critical factor in achieving the remarkable 99% accuracy metric.

The fact is that most of the industry still struggles with low-resolution, noisy, or inconsistent data. Even when assessing model accuracy, the challenge remains. The case study acknowledged that maritime data is inherently volatile, with measurement noise and inconsistencies that create an upper limit on how well a model’s predictions can be validated. Even a theoretically perfect model will show deviations when tested against real-world sensor data, simply because that data itself contains errors.

The Industry-Wide Implications

So, is vessel modeling a solved problem?

Well, the technology certainly now exists to build highly accurate, AI-powered vessel models that vastly outperform traditional methods. DeepSea’s work with EPS proves that, with a good quality data pipeline and the right sort of cleansing and validation, AI can create models with unprecedented predictive accuracy.

But the challenge remains around the industry’s level of data availability. Addressing this (in a high-frequency or even in a low-frequency setting) is the key to unlocking a widespread revolution in vessel understanding.