基于LSTM-GM神经网络模型的风暴潮增水预报方法

Research on forecasting method of storm surge based on LSTM-GM neural network model

为充分挖掘风暴潮增水的时序关联特性,提高风暴潮增水的预报精度,综合考虑台风因素、气象要素和天文潮因素对风暴潮增水的影响,结合长短期记忆(LSTM)神经网络和灰色模型(GM)的优势,提出基于LSTM-GM神经网络模型的风暴潮增水预报方法。利用该方法采用12场历史台风数据对小清河入海口风暴潮增水进行模拟预报,并将预报结果与LSTM神经网络、BP神经网络的预报结果进行对比。结果表明:相较于LSTM神经网络和BP神经网络,LSTM-GM神经网络模型的纳什效率系数分别提高了6.5%和11.4%,均方根误差分别降低了70.6%和72.2%,平均相对误差分别降低了50%和69.2%;LSTM-GM神经网络模型可有效处理风暴潮增水与各影响因素间的非线性关系,提高风暴潮增水预报的精度。

In order to fully explore the time series correlation characteristics of storm surge and improve the forecast accuracy of storm surge,this paper comprehensively considered the influence of typhoon,meteorological and astronomical tide factors on storm surge,combined the advantages of long short term memory(LSTM)neural network and gray model(GM),and proposed a forecasting method of storm surge based on LSTM-GM neural network model.Using this method,12 historical typhoon data were used to simulate and forecast the storm surge in the Xiaoqing River Estuary,and the forecast results were compared with the prediction results of the LSTM neural network and the BP neural network.The results showed that compared to LSTM neural network and BP neural network,the NSE index of LSTM-GM neural network model increased by 6.5%and 11.4%,RMSE decreased by 70.6%and 72.2%,and MAPE decreased by 50%and 69.2%,respectively.The LSTM-GM neural network model can effectively handle the nonlinear relationship between storm surge and various influencing factors,and improve the accuracy of storm surge forecast.