Sprint lends momentum to AI enabled weather routing services
The Naval Architect: May 2020
Launched in 2018, Sprint [SPace Research and Innovation Network for Technology] is a British government-backed scheme that brings together the UK’s best academic expertise in space technologies with SMEs. Although part of its remit involves getting things into space, it’s more focused upon taking advantage of satellite imagery and data that is being constantly generated by initiatives such as the European Union’s Copernicus programme, much of which is freely available.
In March, met-ocean forecast data specialists Theyr Ltd and the University of Southampton (UoS) announced that the project would be funded by a grant from the £4.8 million SPrint programme and will create an ‘industry leading’ marine vessel routing application. The project, which will run for one year, aims to bring together the latest in ocean forecasting data with leading edge ‘Genetic Algorithms’ developed by the UoS team to create a route optimisation module and thereby minimise GHG emissions.
“Our approach to the developing market requirements considers our routing module to be complimentary to Voyage Optimisation Software (VOS) E-Navigation and Maritime Internet of Things providers,” says David Young, Theyr’s managing director. “This module incorporates an operational met-ocean service and forms an essential element to an increasingly diverse and modular marketplace that provides a myriad of solutions.”
Young believes that the nascent aspect of the current routing market opens up significant commercial opportunities for innovative and progressive companies such as Theyr. However, its ability to compete with larger companies will be significantly expedited by the SPrint project given that both parties bring considerable commercial and technical experience to the project.
The project aims to exploit how two recently developed Genetic Algorithms methodologies can be used to take advantage of the increasingly higher fidelity data that is being made available.
UoS researchers have developed new methods that build on these approaches by creating mechanisms that replicate multi-level selection, the theory of natural selection that proposes that the fitness of an individual can be judged not only on their own fitness but also the collective of individuals with which they are associated (e.g. a wolf and the pack it belongs to). This has led to an algorithm that exhibits a high diversity and is particularly strong on large or constrained problems.
Meanwhile, another approach, currently under development with Quaid-i-Azam University in Islamabad, looks at the crossover mechanisms in the algorithm, changing the probability that determines how similar children in a new generation are to the parents. This method has been shown to increase performance on dynamic problems over the multi-objective evolutionary algorithm based on decomposition (MOEA/D), the strongest convergence-based Genetic Algorithm.
The research of UoS and Theyr aims to develop a path-routing algorithm, or set of algorithms, that can be used on different problems, from short paths with higher fidelity met-ocean data, to longer paths with lower fidelity data. “This will future-proof our software against these increases in fidelity and provide leading performance over competitor software, reducing the emissions at sea through more effective utilisation of high-fidelity data,” says Young.
To help them achieve this, the project partners will also have access to UOS’s Iridis 5 supercomputer to run verifications.
“The idea is to run some realistic routes from real ship data. We will be looking at a specific route and the met-ocean conditions at that same time period to see whether our predicted routes can improve on the performance that the vessel took in those conditions. We will replicate this process a number of times to see when the VOS provides the best performance and the types of conditions where the GHG savings can be highest.”