DeepSea models vessel behaviour with 99% accuracy using Tidetech metocean data

“Obviously metocean data is key. We need good reliable data, good coverage, and good granularity, to understand vessel behaviour under different conditions. If you don’t have that, the AI models are going to learn an average overall state which is not useful enough to meet the needs of the industry today.”

– Dr. Stavros Paschalakis, Chief Technology Officer at DeepSea Technologies

The challenge: Every vessel, every voyage is unique

DeepSea Technologies set out to use the industry’s most advanced AI to create ship and voyage optimisation tools based on unique behaviour models for each ship. The company’s founders formed a first-of-its-kind AI research team, bringing together marine and AI engineers to develop the solutions.

The company maintains its culture of research, contributing academic papers to conferences globally, and bringing these advancements to their products.

“Vessel performance monitoring and optimisation is no longer a secondary operational concern for shipping companies,” said Dr. Stavros Paschalakis, Chief Technology Officer at DeepSea.

“In the past, when weather and ocean conditions became significant, a generic vessel model that was 85% accurate would generally have been considered good. Optimisation, on the other hand, is a marginal activity relying on making small and precise changes, and adding many small gains together for major impact. Trying to do this with a model that doesn’t accurately understand the effect of weather and oceanic phenomena on vessel performance just won’t cut it.”

A perfect illustration of this can be seen when examining modern weather routing: the more advanced providers will run their weather suggestions through a relatively basic, non-specialised vessel model, with the aim of providing a voyage speed plan that saves fuel. However such a model doesn’t capture the vessel’s real performance in enough detail to actually save fuel. Incorrect speed suggestions during periods of strong or changing currents can result in higher fuel consumption than if the vessel had just travelled at a constant speed.

DeepSea put AI to task to overcome these limitations.

“Our models are data-driven deep neural networks. They take in vessel parameters about operating conditions such as how heavily loaded the vessel was, how fast the vessel was going, and so on. Based on that, we build a digital twin of the physical system,” Dr. Paschalakis said.

Ship operator priorities, such as minimising fuel consumption and emissions while meeting schedules, are accounted for in voyage planning, along with the impact of weather and oceanic conditions. These parameters are different for every vessel and every voyage. The neural networks need to be able to predict vessel performance in varied conditions such as unusually strong currents or at very specific vessel drafts.

“Metocean data is key. We need good reliable data, good coverage, and good granularity, to understand vessel behaviour under different conditions. If you don’t have that, the AI models are going to learn an average overall weather-agnostic state which is not useful enough to meet the needs of the industry today.”

The solution: Partnering with a metocean data specialist

DeepSea set out to find a long-term partner that could provide the necessary data.

“We are not metocean specialists, and we don’t want to become a metocean company,” Dr. Paschalakis said.

“We were looking for an agile partner that would listen to our needs, and we found Tidetech had the right mindset, the right synergies, for a long-term partnership.”

Like DeepSea, Tidetech maintains a strong scientific research presence. Co‑founders Penny Haire, a professional mariner and navigation specialist, and Roger Proctor, one of the world’s foremost coastal oceanographers, brought together a team that blends oceanography, engineering, and software expertise. Tidetech’s core capabilities include hydrodynamic and tidal modelling, data analysis and transformation, cloud‑based data infrastructure, and data delivery services.

DeepSea’s team use Tidetech services to train their deep neural networks. For example, Tidetech’s current and wave data is used to better estimate engine load and vessel speed.

The results are incorporated into DeepSea’s end-to-end optimisation platforms:

• DeepSea’s Vessel Intelligence suite is an AI-powered data and performance platform that maintains digital twins of a vessel for all-weather fuel consumption predictions as well as functions such as fouling estimation, tracking of emissions reporting metrics, and active machinery monitoring – all backed by a team of experts that turn insight into action.
• Their Voyage Automation suite is a world-first AI-powered optimisation platform that dynamically calculates optimal speeds, routes, and arrival times – and then automatically executes the suggestions by connecting directly to the ship’s main engine.

Metocean data is used to help make decisions for the future, including vessel speed and arrival time suggestions.

“We have found it very useful to have regular forecasts for the next week or 10 days, but the extended forecasts Tidetech provides to us, while more approximate, still give reasonable forecast up to 30 days from now,” said Mikhail Krechetov, Voyage Optimisation Director at DeepSea.

“Many vessels have voyage plans that take them on quite long trips, from Europe to Asia around the Cape of Good Hope, so we need this extended forecast data.”

Post-voyage, hindcast data is used to calculate actual fuel savings made from following DeepSea recommendations, and sometimes hindcast data from over a year ago is required to satisfy customer requests.

Tidetech’s data is also used to determine safety constraints. DeepSea ensures its models only provide safe speed suggestions that don’t risk crew, vessel, or cargo.

“We don’t want our vessels to suddenly sail into six-metre waves or into some dangerous combination of swell and wind that will cause effects such as parametric rolling, so we need metocean data not only for models, but also for the constraints of the algorithm,” Krechetov said.

Customers have occasionally queried the results.

“We have had some interesting cases in the past where customers have said: ‘Why did you tell me to go at this speed? It’s very uneconomical.’ And then we plotted the voyage for them and showed them that we were navigating the ship so that it would not catch up to a bad weather front ahead, or let bad weather from behind catch up to them. For this particular voyage, it was the optimal plan.”

The results: DeepSea’s models achieve 99% accuracy

DeepSea now performs optimisations for hundreds of ships every day.

The models can determine what the impact of wind and currents (to name a few factors) will be on fuel consumption at every point in the ocean, at any point in time, for any vessel, with 99% accuracy. The models accurately establish the impact of all the factors that influence voyage efficiency.

Crews can take DeepSea’s automatically generated speed and route instructions and input them into their ECDIS to follow the plan manually, or they can connect the speed instructions to the vessel’s main engine directly. This autonomous process ensures the vessel is always sailing at optimal speed.

Customers’ fuel costs are typically cut by between 4% and 10%, and even higher in some circumstances. For example, Wallenius Wilhelmsen (WW) had already digitised its fleet of over 130 vessels when it undertook a trial to determine if DeepSea could build on that foundation to make a real impact on optimising fleet operations. During the trial, voyage efficiency increased by 6.9%, saving $284,387 and 912 tonnes in CO2 equivalent emissions. Applied across a year, these results raise the average vessel’s carbon intensity indicator (CII) grade by at least one level. Wallenius Wilhelmsen is now partnered with DeepSea to help operate its entire fleet of ships more efficiently.

“With improvements in the algorithms and increased granularity in metocean data, we are raising the bar all the time to achieve optimal vessel operation, and our customers are coming on that journey with us.” Dr. Paschalakis said.

DeepSea is continuing to expand its use of Tidetech data by incorporating more sophisticated modelling of ice conditions. Warmer average temperatures in polar regions and the corresponding loss of sea ice means more icebergs at sea. Year-to-year variability in ice conditions is also greater than in the past, so there is a growing need for better voyage planning in polar regions.

DeepSea’s neural networks are a frontier technology – and enduring partnerships will ensure they remain transformational.