基于LSTM神经网络的畸形波预测

Rogue wave prediction based on LSTM neural network

采用长短时记忆(LSTM)神经网络预测方法对某岛礁地形模型的四个典型波浪试验数据进行预测分析,并建立了单步和多步预测模型.首先对波高时间序列数据进行归一化处理;然后建立了包括输入层、隐藏层和输出层的LSTM网络模型框架;最后对测试样本进行单步预测,将预测结果与支持向量机(SVM)模型和反向传播(BP)模型进行了对比.结果表明:LSTM神经网络预测精度有明显优势;多步预测中,提高预测时长其预测精度并无明显降低.

Long-short term memory(LSTM)neural network prediction method was used to predict and analyze the four typical wave test data of a certain island reef topography model,and the single-step and multi-step prediction models were established.First,the wave height time series were normalized.Second,LSTM network model framework including input layer,hidden layer and output layer was established.Finally,the test samples were predicted in a single step,and the predicted results were compared with the support vector machine(SVM)and back propagation(BP)models.Results show that the LSTM neural network has obvious advantages in predicting accuracy.In the multi-step forecasting,the prediction accuracy would not decrease obviously when the prediction time is increased.