A Hierarchical Kolmogorov–Arnold Network Model for Underwater Sound Speed Prediction With Ocean Experimental Results

The distribution of underwater sound speed is crucial for determining the propagation mode of underwater acoustic signals. Traditional methods for obtaining sound speed profiles (SSPs) typically rely on instrument measurements or sonar data inversion, resulting in high equipment costs and insufficient real-time performance. To eliminate the need for on-site data measurement and improve the real-time acquisition of sound velocity, methods for predicting sound velocity distribution from a time series prediction perspective based on recurrent neural networks, long short-term memory networks, Transformer, etc. have emerged in recent years. However, these methods face problems of gradient vanishing and explosion, and are prone to overfitting in cases of sample scarcity. To address these issues, we propose an hierarchical kolmogorov-arnold network (HKAN) model for predicting sound speed distribution in this paper leveraging historical data to achieve real-time estimation SSPs. Real data experiments show that the KAN network has stronger generalization ability and higher accuracy in predicting sound velocity in few-shot learning situations.