基于物理关联深度学习的海浪浪高预测方法

Wave height prediction method based on physical correlation constrained deep learning

精确的海浪有效波高(简称浪高)预测对于海上生产生活具有重要意义。针对现有海浪浪高预测模型对不同海洋要素间关联信息考虑不足,以及长时序浪高数据本身存在非平稳性的问题,本文设计了一种考虑物理约束与差值约束的海浪浪高时间序列预测方法。该方法基于风速与浪高之间的物理关联,设计物理约束,并通过提取差分信息设计差值约束,结合现有基于深度学习的时间序列预测模型,实现浪高预测。采用黄海和东海的6个不同站点浮标数据进行了大量实验。实验结果表明,本文提出的方法可以利用海洋要素间的物理关联,有效提高浪高预测精度,并避免因不同要素间融合造成的信息间干扰;同时,利用差值约束,限制时间序列预测结果的变动范围。本文方法可以与不同类型的时间序列预测模型相结合,显著提升原有模型的性能,并在长时间序列的预测中体现出很好的鲁棒性,为海洋要素预测中物理与数据驱动模型的有效结合提供了思路和验证。

Accurate prediction of significant wave height(wave height in short)is of great significance to marine operations and coastal human life.In response to the insufficient attention given to the physical correlations between oceanic elements in existing prediction models,as well as the non-stationarity issue with wave height time series itself,this paper proposes a wave height prediction method based on a physical constraint and a difference constraint.This method designs a physical constraint loss function by taking into account the physical correlation between wind speed and wave height,as well as a difference constraint loss function by extracting the time difference information of wave height.The loss functions are then embedded into existing deep learning-based time series prediction models to achieve accurate wave height prediction.Comprehensive experiments were conducted using buoy data from six different observation stations in the Yellow Sea and East China Sea.The results show that the proposed method can effectively improve the prediction accuracy of wave height by utilizing the physical constraint between oceanic elements while avoiding information interference caused by covariate fusion,and using difference constraints to limit the variation range of time series prediction.This method can be combined with various types of time series prediction models to significantly improve the performance of the original model and demonstrate good robustness in long-term sequence prediction.The proposed method provides ideas and validation for the effective combination of physical and data-driven models in oceanic element prediction.