Research On Wave Height Prediction For Different Time Spans Based On Machine Learnin

The prediction of wave height plays an important role in sea activities.The numerical model based on energy balance equation invented in the last century is still the mainstream method of wave height prediction research.However,the prediction process of numerical models is complex and requires a large amount of input data and processing time,which is a huge challenge for a high-frequency forecast object such as wave height.It is not easy to use numerical simulation,especially in many situations that need to quickly obtain wave height prediction results.With the massive increase in observational data,the study of marine environmental changes with the aid of artificial intelligence methods has become a topic of increasing interest to oceanographers.Using machine learning can achieve faster and more accurate wave height prediction.However,there are still many problems in the existing wave height prediction research based on machine learning.In order to give scholars an overall understanding of the current wave height prediction research,this paper firstly constructs a literature visualization analysis method for the relevant literature in the field of wave height prediction.This method displays the literature analysis results related to wave height prediction in the form of images.In short time interval prediction,in order to solve the following problems in existing research: the training and testing of model at the same station,and seldom consider whether the trained model can be used in other station;most of the existing studies focus on the open sea area,and there is a lack of wave height prediction research for the near-shore sea area.This paper firstly improves the long short-term memory(LSTM)network,and proposes a training-prediction stations separation wave height prediction model based on Bayesian optimization of LSTM.Different from previous studies,the training of this model is based on data from three non-test stations,and the optimal values of model hyperparameters are obtained by Bayesian optimization algorithm.In the 1-hour,6-hour and 12-hour wave height predictions,the prediction effect of the model is satisfactory,and it can be used for wave height prediction of non-training station.By comparing with the commonly used multilayer perceptron(MLP),support vector machine(SVM),and random forest(RF)algorithms,the superiority of LSTM based on Bayesian optimization is verified.In order to discuss the effect of machine learning in the prediction of wave heights in the offshore area,this paper uses the LSTM based on Bayesian optimization to achieve 1-hour,6-hour and 12-hour wave height predictions for station in China offshore.The results demonstrate that machine learning can achieve satisfactory accuracy in offshore area wave height prediction.Through the comparison of algorithms,the LSTM based on Bayesian optimization proposed in this paper has advantages in offshore area wave height prediction.In long time interval prediction,in order to solve the problem of low prediction accuracy due to hysteresis,this paper improves the temporal convolutional network(TCN)by incorporating the empirical mode decomposition(EMD)algorithm,and proposes a long time interval wave height prediction model based on EMD-TCN.In the 24-hour,36-hour, and 48-hour predictions,it is verified that the model helps to improve the prediction lag problem and can improve the accuracy of long time interval prediction.Finally,this paper proposes a vision of the future application of wave height prediction,which can optimize ship’s traveling route based on the combination of ocean Internet of Things and wave height prediction.