Three-Dimensional Spatio-Temporal Slim Weighted Generative Adversarial Imputation Network: Spatio-Temporal Silm Weighted Generative Adversarial Imputation Net to Repair Missing Ocean Current Data DOI Creative Commons

Ya Jie Yue,

Juan Li, Y. Zhang

et al.

Journal of Marine Science and Engineering, Journal Year: 2025, Volume and Issue: 13(5), P. 911 - 911

Published: May 4, 2025

Three-dimensional ocean observation is the foundation for accurately predicting information. Although sensor arrays can obtain internal data, their deployment difficult, costly, and prone to component failures environmental noise, resulting in discontinuous data. To address severe missing data problem three-dimensional flow fields, this paper proposes an unsupervised model: Spatio-Temporal Slim Weighted Generative Adversarial Imputation Network (3D-STA-SWGAIN). This method integrates spatio-temporal attention mechanisms Wasserstein constraints. The generator captures spatial distribution vertical profile dynamic patterns through module, while discriminator introduces gradient penalty constraints prevent vanishing. strives generate that conforms real field, attempts identify pseudo-ocean current samples. Through adversarial training of discriminator, high-quality completed are generated. Additionally, a continuity loss function designed ensure physical rationality Experiments show on field dataset South China Sea, compared with methods such as GAIN, under 50% random rate, reduces error by 37.2%. It effectively solves traditional interpolation have difficulty handling non-uniform correlations maintains field’s structure.

Language: Английский

Three-Dimensional Spatio-Temporal Slim Weighted Generative Adversarial Imputation Network: Spatio-Temporal Silm Weighted Generative Adversarial Imputation Net to Repair Missing Ocean Current Data DOI Creative Commons

Ya Jie Yue,

Juan Li, Y. Zhang

et al.

Journal of Marine Science and Engineering, Journal Year: 2025, Volume and Issue: 13(5), P. 911 - 911

Published: May 4, 2025

Three-dimensional ocean observation is the foundation for accurately predicting information. Although sensor arrays can obtain internal data, their deployment difficult, costly, and prone to component failures environmental noise, resulting in discontinuous data. To address severe missing data problem three-dimensional flow fields, this paper proposes an unsupervised model: Spatio-Temporal Slim Weighted Generative Adversarial Imputation Network (3D-STA-SWGAIN). This method integrates spatio-temporal attention mechanisms Wasserstein constraints. The generator captures spatial distribution vertical profile dynamic patterns through module, while discriminator introduces gradient penalty constraints prevent vanishing. strives generate that conforms real field, attempts identify pseudo-ocean current samples. Through adversarial training of discriminator, high-quality completed are generated. Additionally, a continuity loss function designed ensure physical rationality Experiments show on field dataset South China Sea, compared with methods such as GAIN, under 50% random rate, reduces error by 37.2%. It effectively solves traditional interpolation have difficulty handling non-uniform correlations maintains field’s structure.

Language: Английский

Citations

0