인공신경망모형의 항만물동량 예측 적용에 관한 연구

Abstract

In this paper, the neural network models are proposed and developed for forecasting port cargo throughput. We selected 10 items each for import and export and classified them into 4 categories based on the types of change pattern. And this study compares and analyzes the effectiveness of the regression analysis and the neural network model in respect of fitness, reliability, and robustness. In 'continuing growth type' where port cargo throughput keeps growing on a sustainable basis, the regression analysis turned out to be more effective than the neural network model in the aspects of fitness, reliability, and robustness. In contrast to this, the neural network model proved to be superior in all these three aspects in the 'sluggish growth type' where the throughput does rise in general but growth rates decrease year-on-year and in the 'static or irregular growth type' in which handling volumes hardly change, neither increasing nor decreasing. In the 'irregular growth type' where change pattern is irregular but each section's bottom points of the irregular graph get higher as the pattern goes, the regression analysis showed strong in terms of fitness but the neural network model performed better in terms of reliability and robustness. The results indicate that effectiveness can differ according to the characteristics of subjects.