Learning Dynamic Graph Structures for Sea State Estimation with Deep Neural Networks
Kexin Wang, Xu Cheng, Fan Shi · 2023
Accurate estimation of sea state is crucial for the advancement of autonomous ships, given the complexity and dynamics of the marine environment. However, traditional methods for classifying sea states suffer from drawbacks such as low accuracy, limited coverage, and high cost. The emergence of deep learning has provided a significant breakthrough in sea state estimation (SSE). Nonetheless, existing models lack the ability to dynamically interact with features from various sensors. In this study, we propose a deep neural network model, named DynamicSSE, which aims to autonomously learn the dynamic correlations among sensors located at different positions. This facilitates the interaction between sensors, enabling the generation of more valuable features for prediction. DynamicSSE comprises two primary modules: a feature extraction module and a dynamic graph structure construction module. By combining a series of convolutional neural networks (CNNs) and a long short-term memory network (LSTM), our model effectively captures graph structures and dependencies over both long and short time periods. Moreover, our proposed approach demonstrates its superiority through experimental results on two ship motion datasets, supported by a comparative analysis with the baseline approach. Additionally, an ablation study highlights the significance of each component in our model.