Short-Term Wave Forecast Based On Recurrent Neural Network

Waves in the ocean have a profound impact on human activities in the ocean and offshore,but tsunamis caused by excessively large waves can cause losses to human safety and economics.Therefore,accurate short-term wave forecasting is of great reference value for people's related activities such as ocean activities,fish farming,offshore oil exploration,and the layout and construction of offshore platforms in the open and offshore seas.Because of the randomness of ocean waves,nonlinear resonance and other factors,many conventional problems in ocean engineering are still challenging.Nowadays,the study of waves can be divided into two methods,one is based on dynamics,and the other is based on statistics.The deep learning theory based on statistics is gradually being used by people.Because the neural network model itself has a good adaptive learning mechanism and nonlinear mapping capabilities.It is not necessary to have a very clear understanding of the physical mechanism contained in the thing itself,so it is suitable for solving the problem that the physical mechanism is difficult to understand.This paper mainly uses the recurrent neural network model to make short-term forecasts of waves.The main work of this paper is as follows:(1)Discretely generate training set and test set for wave spectra;for the cyclic neural network model of single-step forecasting,establish the relationship between the optimal input step size and the autocorrelation function of wave spectra to provide input for determining the single-step forecast Theoretical guidance.(2)Using the cyclic neural network to perform short-term single-step forecasting of wave time history for the PM spectrum,ITTC dual-parameter spectrum and JONSWAP spectrum under different sea conditions.(3)Try to make a short-term multi-step forecast of the wave duration by directly increasing the input step length and the output step length to obtain the change law of the optimal input step length under different sea conditions.This paper uses cyclic neural network to forecast the wave spectrum in time series,which can provide a certain reference for the work of real-time wave forecasting.