인공신경망 모델을 이용한 항만 다운타임 예측

Abstract

In this study, a downtime forecasting technique for Pohang New Harbour, which is located near the inmost part of pocket-type bay in the east coast of Korea and where downtime due to swell-like waves frequently occur, was developed using the artificial neural networks model as the non-parametric estimation model. For this purpose, a neural networks model was constructed using wide-area wave observational data from ocean wave buoys and observational data from bay entrance and several points in harbour from September 2008 onward. The back-propagation neural networks model, having excellent generalization capability and being widely employed in various engineering problems including weather forecasting, was used as the neural networks model. Wide-area observational data was used as input data, while in-harbour observational data was used as output data. Currently, results are being compared using various neural network structures, and it is expected that the optimum neural networks model would be constructed through this comparative study. Moreover, using such an optimally constructed artificial neural networks model, it is also expected that the forecasting system can be constructed and operated more efficiently. artificial neural networks model as the non-parametric estimation model. For this purpose, a neural networks model was constructed using wide-area wave observational data from ocean wave buoys and observational data from bay entrance and several points in harbour from September 2008 onward. The back-propagation neural networks model, having excellent generalization capability and being widely employed in various engineering problems including weather forecasting, was used as the neural networks model. Wide-area observational data was used as input data, while in-harbour observational data was used as output data. Currently, results are being compared using various neural network structures, and it is expected that the optimum neural networks model would be constructed through this comparative study. Moreover, using such an optimally constructed artificial neural networks model, it is also expected that the forecasting system can be constructed and operated more efficiently.