ATL-Net: Ocean Current Prediction Based on Self-Attention Mechanism and Feature Enhancement

The prediction of ocean currents is crucial to the development and utilization of marine resources. However, traditional prediction methods often fail to accurately capture the irregular behavior of ocean currents, which results in low accuracy. To solve this problem, this paper introduces a novel network for ocean current prediction, termed ATL-Net, which integrates the self-attention mechanism, temporal convolutional network (TCN), and long short-term memory (LSTM) network. ATL-Net enhances feature extraction through the inflated convolutions of TCN, captures ocean current dynamics more effectively with the self-attention mechanism, and uses LSTM for its ability to memorize long sequences of current information. Experimental results on WaMoSII data demonstrate the effectiveness of ATL-Net, which achieves the current speed prediction accuracy of 92.9% and the current direction accuracy of 90.4%. The results outperform those of existing methods. The method proposed in this paper is not only vital for the development and utilization of marine resources but also presents an efficient method for accurate ocean current prediction.