101. Feature disentanglement learning model for ocean temperature field forecast.
- Author
-
Lei, Lei and Jianxing, Zhang
- Subjects
- *
MACHINE learning , *OCEANOGRAPHY , *DEEP learning , *FORECASTING - Abstract
• High-accuracy dynamic modeling is proposed for the ocean temperature field forecast. • Incremental data learning is developed for solving the continuous observation process. • The proposed method can disentanglement the dynamic and stable features. With the development of sensor technology, data-driven modeling methods for physical oceanography have boomed. Existing deep learning studies lack interpretability from the physical mechanisms. This paper proposes a novel feature disentanglement learning model for the ocean temperature field forecast. In this method, feature disentanglement is adopted to decompose the high-dimensional measured temperature information based on Tucker decomposition. The extracted dynamic feature is filtered and predicted by incremental dynamic learning, and the incremental data information is balanced by maximizing the Gaussian likelihood. Then, the dynamic feature can be entangled with the stable features to predict the ocean temperature field. Last, incremental residual learning can compensate for the feature disentanglement learning error. Experiments on the temperature dataset of the Western Pacific have verified that the proposed method is superior to the existing baselines. [ABSTRACT FROM AUTHOR]
- Published
- 2023
- Full Text
- View/download PDF