Time series and deep learning prediction study Using container Throughput at Busan Port

In recent years, technologies forecasting demand based on deep learning and big data have accelerated the smartification of the field of e-commerce, logistics and distribution areas. In particular, ports, which are the center of global transportation networks and modern intelligent logistics, are responding quickly to changes in the world economy and shipping port environment because of the 4th Industrial Revolution. Forecasting the port volume will have important effects in various fields, including the construction of a new port, port expansion and terminal operation. Therefore, the purpose of this study is to compare the demand forecasting model of ARIMA and SARIMA through deep learning and to derive the forecasting model suitable for future container forecasting at Busan Port. In addition, new factors related to the change of logistics volume were selected by correlation and applied to the multivariate deep learning prediction model. The results showed that ARIMA and SARIMA errors were low in the single-variable prediction model using only Busan Port container volume, and LSTM errors were low in the multivariable prediction model using external variables. This study provides important implications for comparing various physical volume prediction models and selecting appropriate prediction models.제 1 장 서 론 1 1.1. 연구의 배경 1 1.2. 연구목적 및 절차 3 제 2 장 선행연구 검토 5 2.1. 물동량 예측 선행연구 5 2.2. 물동량 예측모델 비교 선행연구 7 2.3. 연구의 차별성 및 시사점 12 제 3 장 분석방법론 13 3.1. Hurst 지수 산정 13 3.2. 상관관계 분석 14 3.3. 시계열 예측모델 15 3.4. 딥러닝 예측모델 19 제 4 장 실증분석 27 4.1. 부산항 컨테이너 물동량 데이터 분석 27 4.2. 외부변수 탐색 및 상관관계 분석 31 4.3. 시계열 예측모델 분석 39 4.4. 딥러닝 예측모델 분석 41 4.5. 최종 예측모델별 비교 44 제 5 장 결론 및 향후 연구방향 46 5.1. 결론 및 시사점 46 5.2. 연구의 한계 및 향후 연구방향 47 참고문헌 49Maste