딥러닝 기법을 이용한 태풍시 이상조위 예측 연구

Abstract

In general, typhoons are serious marine disasters that cause enormous human and economic damage every year. In particular, unpredictable high and abnormal tides caused by typhoons cause great damage to ships, coastal structures, and residential facilities in coastal areas. Recently, many studies have been conducted that apply deep learning techniques to predict the tide changes due to typhoons. However, most of these studies have limitations in that they cannot reflect the characteristics of time series data by using the classic MLP (Multi-Layer Perceptron). However, recently, RNN (Recurrent Neural Network) implemented as LSTM (Long Short-Term Memory) is being applied to various fields to analyze sequential data and its excellence is being recognized. In this study, we tried to develop a tidal prediction model using RNN considering the rapid drop in atmospheric pressure spatial change during typhoon. The input data to the model are the time series of harmonic predictions, the atmospheric pressure at the station and the atmospheric pressure at the reference station. The data used were obtained from the Korea Meteorological Administration (KMA) and the Korea Hydrographic and Oceanographic Administration (KHOA). The output of the model is the high tide after 3 hours. 47.4%, 5.3%, and 47.4% of the total data were used for training, validation, and testing of the built model, respectively. Sensitivity analysis was performed on the input sequence length, LSTM layer and units for hyperparameter tuning. As a result, in the case of the 2020 data, which is the validation set, the best model performance was obtained in the case of one layer consisting of 9 sequence lengths and 40 units of LSTM. Finally, in this study, it was confirmed that the results of the tide prediction model using the deep learning method for typhoons affecting the Korean Peninsula from 2011 to 2019 had an RMSE of 7.90 cm and a calculation performance of 0.974 with a correlation coefficient.