Wave height prediction method based on physical correlation constrained deep learning
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.
wave height predictionphysical constraindifference constraintime series
NETL
NSTL
万方数据
