TIME SERIES PREDICTION OF WAVE HEIGHT BY LONG SHORT-TERM MEMORY (LSTM) NEURAL NETWORK

Abstract

 The time series prediction of wave height is important for the countermeasure of high wave disaster in the coastal area. Since time series prediction of wave heights over a long recurrence period by dynamical methods is computationally expensive, statistical methods such as neural networks (NN) are considered for prediction. Among the deep learning methods, Long Short Term Memory (LSTM) is considered to be suitable for time series forecasting of wave height because it is good at handling long-term time series data. However, no research has been published so far on time series forecasting of wave height using LSTM. In this study, time series forecasting of wave height is performed using LSTM. The effects of temporal factors such as the input of explanatory variables at multiple times in the past, spatial factors such as the input range of the meteorological field, and the combination of the parameters of the LSTM were varied, and the effects of these factors on the results were compared.