S-Space College of Engineering/Engineering Practice School (공과대학/대학원) Dept. of Civil & Environmental Engineering (건설환경공학부) Theses (Ph.D. / Sc.D._건설환경공학부)
A Data-based Downtime Forecasting Model in a Harbor : Application to Pohang New Port
하역중단 예보를 위한 자료기반 모델 : 포항신항에 적용
- 공과대학 건설환경공학부
- Issue Date
- 서울대학교 대학원
- 학위논문 (박사)-- 서울대학교 대학원 공과대학 건설환경공학부, 2017. 8. 서경덕.
- Unexpected downtime due to harbor agitation has been a serious problem in Pohang New Port. Such downtime occurrences have not well been predicted by field wave observation or numerical modeling around the port. Hence it is required to make an effort to enhance understanding intrinsic causes of downtime and to develop an efficient forecasting model that can be incorporated into daily port operation procedures.
Considering inefficiency of predicting downtime by conventional wave simulation models, a data-based model is developed in this study based on extensive wave observed data over years and corresponding downtime records available at multiple locations inside and outside the port. The main structure of the downtime forecasting model consists of Neural Network (NN) that predict wave parameters inside and outside the port from simulated wave data at outside the port and a classification model that predicts downtime occurrences based on information about wave field produced by the NN. The overall predictive performance of the model was good, showing more than 80% of correct identification of the downtime occurrences on average.
In addition, spatio-temporal changes in spectral energies of various wave components during downtime events were examined by using Hilbert-Huang Transform (HHT) analysis, the details of which have never been studied so far. By conducting this analysis, a better understanding was obtained about the influence of each of gravity wave (GW), infragravity wave (IGW), and natural oscillation period (NOP) on the downtime occurrences inside the port. It also significantly contributed to check and improve the quality of the manually-written downtime records, which consequently enhance the eventual performance of the forecasting model.