Wave Forecasting Of Taiwan Strait And Its Surrounding Waters Based On Deep Learning

The ocean wave has a huge impact on human maritime activities and near-shore activities,and even cause huge human and economic losses.Therefore,accurate prediction of wave height is essential for human activities such as navigation,fisheries,maritime military activities,offshore operations,marine sports,coastal and offshore engineering planning and design.The Taiwan Strait is China's largest strait.It is located between the East China Sea and the South China Sea.It connects the motherland and the island of Taiwan.It is not only an important water area on the trade route in the history of the ocean,but also a strategic point of modern geopolitics.The manual operations,production activities and shipping in the Taiwan Strait are becoming more frequent.Therefore,timely and accurate wave forecasting in the strait is needed to ensure the smooth progress of these activities and avoid the loss of personnel and economy.Wave forecasting is the main purpose of wave research.There are two modes of wave forecasting,single-point wave forecasting for forecasting the wave condition of points and regional wave forecasting for forecasting the wave condition of the large-area of the sea.Wave research generally uses two methods,one is the dynamic method and the other is the statistical method.Deep learning based on statistics is the hotspot of current people's attention.Especially with the rise and rapid development of artificial intelligence,deep learning is becoming more and more the focus of research and business applications.Many researchers or a commercial company applies it to their own projects.Because of its good adaptive learning and nonlinear mapping ability,the deep learning model does not need to be very clear about the physical mechanism of things.It is suitable for dealing with nonlinear problems with complex physical mechanisms,causal relationships and complex inferential rules.Therefore,this paper mainly studies the deep learning model and applies it to the wave forecasting in the Taiwan Strait and its surrounding waters.The details are as follows:(1)To introduced the classic models in the deep learning model,such as the convolutional neural network,recurrent neural network,long short-term memory network and other models of the network structure,basic principles and training process and application areas.(2)The application of neural networks to single-point wave forecasting is common,but the studies of using long short-term memory networks for wave forecasting and applied to the Taiwan Strait have not yet been reported.This paper attempts to apply the long short-term memory network to the single-point wave prediction of four buoy observation points in the Taiwan Strait,and verifies the effectiveness of the method.The experimental results show that the correlation coefficient of the long short-term memory network model can reach 0.962,and the average absolute error is mostly in the range of 0.1 to 0.2 meters.(3)In the regional wave forecasting,the third-generation ocean wave numerical prediction mode is generally adopted.Through the numerical model,the wave condition of the study area can be predicted with high accuracy in the case of given wind field data.However,the numerical model is very computationally intensive and requires large-scale parallel computing.It often takes more time to perform a wave prediction.Therefore,in this paper,a deep learning model with ConvLSTM(Convolutional LSTM Network)layer is proposed for regional wave prediction and applied to the Taiwan Strait and its surrounding waters.The model uses the model data generated by the third-generation ocean wave numerical model SWAN as the training data of the model.The experimental results show that the model can reduce the running time of 98.59% compared with the traditional numerical model in the wave forecast of the 12 th period.