Research on Improving Numerical Accuracy of Ocean Circulation Prediction Using Mixed-former Model
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
With the in-depth research on global climate change, accurately grasping ocean circulation patterns has become increasingly crucial for understanding the evolution laws of the climate system. However, traditional numerical models have limited prediction accuracy in ocean circulation prediction due to their insufficient characterization of multivariate spatiotemporal dependencies, especially in short-term prediction scenarios where error fluctuations are relatively large. This study aims to propose an ocean circulation prediction method based on the Mixed-Former model to improve the numerical accuracy of predictions. In the research process, we first constructed a Mixed-Former model that integrates multi-scale spatiotemporal information. This model fully retains temporal and dimensional information and effectively captures cross-variable dependencies through the innovative Dimension-Segment-Wise (DSW) embedding technology; it accurately captures temporal fluctuations using the Two-Stage Attention (TSA) layer; and it achieves efficient fusion of multi-scale information by adopting a Hierarchical Encoder-Decoder (HED) structure. Subsequently, we conducted comprehensive experiments on multiple sets of ocean circulation datasets covering different ocean regions and different spatiotemporal characteristics. These datasets include multi-time-scale prediction scenarios ranging from short-term (7 days) to long-term (six months). Meanwhile, to comprehensively and objectively evaluate the model performance, we selected several core evaluation indicators such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and Correlation Coefficient (R).In conclusion, the Mixed-Former model demonstrates significant advantages in ocean circulation prediction, whether for predicting short-term fluctuations or long-term trends. It provides a more reliable tool for marine dynamics research and long-term climate prediction, and is expected to promote the further development of research in related fields.
Fields of Study & Subjects
Oceanographic and Atmospheric Processes
Model Reduction and Neural Networks
Hydrological Forecasting Using AI
Publication details
- Journal / source
- Frontiers in Computing and Intelligent Systems
- Volume
- 15
- Issue
- 3
- Pages
- 140–146
- Publication date
- 27 March 2026
- Language
- EN