Researchers from the University of Rijeka in Croatia have developed a novel approach to estimate sea state parameters using ship motion responses and attention-based neural networks (AT-NN). The team, led by Denis Selimović and Franko Hržić, along with Jasna Prpić-Oršić and Jonatan Lerga, has published their findings in a study that promises to enhance the accuracy and efficiency of ship navigation systems.
The study focuses on the critical task of estimating sea state parameters—such as wave height, zero-crossing period, and relative wave direction—directly from ship motion responses. Traditional model-based methods have long been the standard, but the researchers have turned to machine learning (ML) and deep learning (DL) techniques to improve accuracy. By leveraging raw time-series data of ship pitch, heave, and roll motions, the team has demonstrated that attention-based neural networks can significantly outperform conventional approaches.
The researchers applied state-of-the-art deep learning techniques, including convolutional neural networks (CNN) for regression, multivariate long short-term memory CNN, and sliding puzzle neural networks. However, they found that even these enhanced methods could be improved. By modifying these techniques, they achieved a 23% reduction in mean squared error (MSE) and a 16% reduction in mean absolute error (MAE) compared to the original methods. But the real breakthrough came with the implementation of attention-based neural networks (AT-NN). This novel approach reduced estimation MSE by up to 94% and MAE by up to 70%, setting a new benchmark for accuracy in sea state parameter estimation.
Beyond improving accuracy, the researchers also addressed the critical issue of model trustworthiness. They proposed a novel approach for interpreting the uncertainty of neural network outputs using the Monte-Carlo dropout method. This method enhances the reliability of the model by providing a measure of confidence in its predictions, which is essential for real-world applications in ship navigation and stability.
The practical implications of this research are substantial. Accurate estimation of sea state parameters is vital for ship navigation systems, as it directly impacts route planning, fuel efficiency, and safety. By integrating AT-NN into shipboard systems, mariners can make more informed decisions, reducing the risk of accidents and improving overall operational efficiency. The reduced uncertainty in predictions also means that shipping companies can optimize their routes and schedules with greater confidence, leading to cost savings and environmental benefits.
This study not only advances the field of maritime technology but also highlights the potential of attention-based neural networks in other areas of marine research. As the shipping industry continues to embrace digital transformation, such innovations will play a pivotal role in shaping the future of maritime operations. The researchers’ work serves as a testament to the power of machine learning in solving complex problems, paving the way for safer, more efficient, and more sustainable shipping practices. Read the original research paper here.