This work has been submitted to the IEEE for possible publication.
Copyright may be transferred without notice, after which this version
may no longer be accessible.
A major revision is done, compared to the previous version.
Abstract: Neural networks (NNs) have been applied to predict complex
ship motions for stable ship maneuvering and operations at sea, but the
coupling effects of different DoFs are still not effectively modeled in
black-box NNs. To this end, we propose a novel deep NN framework that
explicitly models the coupling effects through factorization machines,
and significantly improves the prediction accuracy without introducing
much extra complexity. The proposed framework is also lightweight and
frequency-aware, which serves real-time and high-frequency prediction
with high accuracy. We experimentally demonstrated that the framework is
not only friendly to real-time prediction and online learning, but also
interpretable and with great potential to be complemented by common
machine learning tricks and go deeper.