AI giant wave predictor – a force for good
Getting your Trinity Audio player ready...
AI may be threatening to wipe out jobs like a ten-pin bowler chasing a perfect score, but there are applications that even union leaders would agree are a force for good. And one of those is the ability of deep learning models to steer container ships away from the perils of giant waves.
If you’ve ever seen a shipping container up close, it’s hard to imagine how such a thing – weighing over two tons empty and capable of carrying a maximum payload of more than 28 tons – could tumble from its bay into the ocean. However, 2,301 containers were lost each year on average (from 2020 to 2022), according to World Shipping Council records [PDF].
One of the reasons for these losses is structural failure, but the biggest culprits are so-called rogue waves arising from natural ocean phenomena. What’s more, shipping captains can have little warning that conditions are about to become dangerous.
The master of the Queen Elizabeth 2 reportedly said that an almost 100 ft high wave ‘came out of nowhere and looked like the White Cliffs of Dover’, while sailing across the North Atlantic in 1995.
Freakishly large waves form when wave systems cross each other and trigger a process dubbed linear superposition. “If two wave systems meet at sea in a way that increases the chance to generate high crests followed by deep troughs, the risk of extremely large waves arises,” explains Dion Häfner – a Senior Research Engineer at Pasteur Labs in New York, US.
AI as a force for good
Häfner is first author in a study titled ‘Machine-Guided Discovery of a Real-World Rogue Wave Model’ submitted to arXiv in November 2023 and published in PNAS. And together with colleagues at the Institute for Simulation Intelligence –which is staffed by ‘industry-hardened experts in AI and computational sciences’ from organziations such as Deepmind, Cerebras, CERN, NASA, and other centers of excellence – they wondered whether AI could be used as a force for good to enable safer shipping.
At any given time, there can be as many as 50,000 cargo vessels navigating the waters around our planet and the goods that they carry are essential to global supply chains. The team’s goal was to provide shipping operators, and anyone with an interest in conditions out at sea, with an accurate prediction of the likelihood of giant waves.
Underpinning the group’s work, is a framework known as a causal directed acyclic graph, which takes a series of environmental conditions and relates them to sea state parameters. In turn, those sea state parameters give rise to physical effects, which can be converted into observations.
“The probability to measure a rogue wave based on the sea state can be modelled as a sum of nonlinear functions, each of which only depends on a subset of the sea state parameters representing a different causal path,” write the researchers in their paper.
Neural networks are able to model these nonlinear functions beautifully. And the group was able to feed its many-layered architecture with training data to adjust all of the various parameters. There’s a wealth of measurements and historical records that can be used to teach the model what inputs are likely to result in dangerous conditions.
The team made use of its Free Ocean Wave Dataset, which had been prepared previously for exactly this task and involved processing a buoy data catalog containing information on 4 billion waves. Of those, around 100,000 fall into the category of being potentially deadly rogue waves, which translates to around one per day occurring somewhere out at sea.
AI’s ability to spot patterns and encode the underlying causality into hidden layers that form deep neural networks becomes a force for good when directed at problems such as the prediction of supply chain disruption.
“As shipping companies plan their routes well in advance, they can use our algorithm to get a risk assessment of whether there is a chance of encountering dangerous rogue waves along the way. Based on this, they can choose alternative routes,” Häfner points out.
28 February 2024
28 February 2024
27 February 2024