Mu-RCATN:Multifactor Residual Channel Attention Temporal Net for Sea Surface Current Field Time Series Analysis

Accurate prediction of the sea surface current field (SSCF) is of vital importance for understanding ocean dynamic processes and has significant implications for environmental monitoring. Current time series prediction methods for SSCF struggle to achieve global optimal solutions and often fail to account for seasonal variations, while also neglecting the collaborative influence of multiple marine factors on SSCF changes. In this article, we propose a novel multifactor residual channel attention temporal net (Mu-RCATN), specifically designed to address the SSCF prediction challenges. Mu-RCATN employs differential operations to precisely capture trends in marine data changes and incorporates a channel attention mechanism to model the synergistic effects of various marine factors on SSCF. In addition, the introduction of residual modules to improve the TimesNet architecture, and concurrently capture and fuse intraperiod and interperiod variations, thus improving prediction accuracy. Comprehensive experiments conducted on a self-built South China Sea dataset validate the effectiveness of Mu-RCATN, with results showing a 62.84% increase in accuracy, a 64.29% reduction in mean absolute error, and a 95.71% decrease in mean squared error, outperforming the bidirectional long short-term memory model proposed in the past five years. This model demonstrates high prediction accuracy and can be effectively applied to SSCF prediction across various sea regions.