Research on Ocean Path Loss Prediction Method Based on Diffusion Model

With the rapid development of the Internet of Underwater Things (IoUT) and underwater communication technologies, the accurate prediction of path loss in complex ocean environments has become a key challenge. Aiming at the problem of insufficient prediction accuracy of traditional empirical models in complex ocean channels, this paper proposes, for the first time, a fast ocean path loss prediction method based on a diffusion model with physical constraints. A path loss dataset that couples multiple physical fields such as temperature, salinity, depth, and wave height is generated through the BELLHOP model. A conditional diffusion model framework with UNet as the core is constructed. The cross-attention mechanism is used to dynamically integrate environmental parameters, and the sound speed equation constraint is introduced to optimize the loss function, enhancing physical consistency. Experiments show that the mean squared error (4.3 dB²) of this method on the test set is significantly better than that of the two-ray model (12.5 dB²) and deep learning models (6.8 dB² for ResNet18). The coefficient of determination reaches 0.92, and it can accurately capture the sound field attenuation in the thermocline and the characteristics of seabed reflection. The research reveals the advantages of diffusion models in quantifying the uncertainty of ocean channels and high-fidelity modeling, providing theoretical support for the design of underwater communication systems. However, it is necessary to further optimize the generalization ability under extreme sea conditions and the edge deployment efficiency. This achievement opens up a new path for ocean Internet of Things channel modeling.