Improved Deep learning method and high-resolution reanalysis model based intelligent marine navigation

Zeguo Zhang Liang Cao Jianchuan Yin ^*

• Guangdong Ocean University, Zhanjiang, China

Large-scale weather forecasting is crucial for ensuring the seaworthiness and safety of ships, particularly during transoceanic voyages. Accurate, high-resolution weather predictions enable dynamic voyage planning and support the advancement of intelligent ship navigation. However, most observation systems and weather forecasts provide sparse or incomplete meteorological data, which negatively impacts path planning accuracy and efficiency. Additionally, due to uncertainties in maritime communication, ships underway often struggle to access high-resolution weather forecasts. To address these challenges, we developed a novel model, IPCA-MHA-DSRUnet, which integrates incremental principal component analysis (IPCA) with a spatial-temporal depthwise separable Unet architecture, enhanced by attention and residual learning mechanisms. This model aims to achieve fine-grid, large-scale wind system predictions, improving voyage planning and navigation safety. The use of depthwise-separable convolution (DSC) blocks significantly reduces model parameters and computational complexity. Furthermore, IPCA is applied for the first time to preprocess 2D wind field data, effectively reducing dimensionality and minimizing noise. The proposed model demonstrates promising potential for onboard weather prediction during large-scale ocean voyages, contributing to intelligent ship path design.