Wave Prediction in the Sea of Japan by Deep Learning Using Meteorological Data
Tracey H. A. TOM, Ai IKEMOTO, Hajime MASE, Koji Kawasaki, Masahide Takeda, Sooyoul Kim · 土木学会論文集B2(海岸工学) · 2019
Numerical wave prediction models require a large amount of computational power to timely complete the required calculations. Artificial Neural Networks (NN) have been introduced to perform predictions at a lesser computational cost and increased processing speed. Deep learning and specifically Convolutional Neural Networks (CNNs) have become accepted for various image recognition applications. Motivation for the examination of wave prediction by deep learning came from the success of CNNs in vision applications and the similarity of meteorological weather grid data to visual images. This study investigates a deep learning technique using the Japan Meteorological Agency’s Grid Point Value Mesoscale Model to predict wave height and period along Japanese coasts of the Sea of Japan. In particular, this study uses the Xception deep learning architecture with depthwise separable convolution to obtain improved wave height and period prediction over artificial neural networks, and gets overall success results.