Time Series Forecasting of Significant Wave Height using GRU,
CNN-GRU, and LSTM

Time Series Forecasting of Significant Wave Height using GRU,
CNN-GRU, and LSTM

wave height, pelabuhan ratu, cnn-gru, long short-term memory, gated recurrent unit

Predicting wave height is essential to reduce significant risks for shipping or activities carried out at sea. Waves inherit a
stochastic nature, mainly generated by wind and propagated through the ocean, making them challenging to forecast. In this
paper, we design time series wave forecasting using a deep learning model, which is a hybrid Convolutional Neural Network
(CNN)-Gated Recurrent Unit (GRU) or CNN-GRU. We use two time series of wave data sets, i.e., reanalysis data from ERA5
by ECMWF and GFS from NOAA. As a study area, we choose Pelabuhan Ratu, located in the south of West Java which is
connected to the open Indian Ocean. Moreover, we also compare the results by using other deep learning models, i.e., the Long
Short-Term Memory (LSTM) and GRU. We evaluated these models to forecast 7, 14, and 30 days. Models' performance is
assessed using RMSE, MAPE, and Correlation Coefficient (CC). For predicting 30 days, using the ERA5 data, the CNN-GRU
model produces relatively accurate results with an RMSE value of 1.8844 and CC of 0.9938, whereas for the GFS data, results
in RMSE value of 1.8852 and CC of 0.9915.

Cornelius Stephanus Alfredo1
, Didit Adytia2

Telkom University

31-10-2022

Fajar bagus W

PDF

Indonesia

Text