United States Department of Transportation
i
Prediction of Port Recovery Time after a Severe Storm Project
-
2023-08-01
-
[PDF-938.75 KB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Research Report
-
Corporate Publisher:
-
Abstract:Predicting the impact of incoming tropical cyclones on ports in terms of the number of days underperforming is crucial for the effective management of the ports. However, existing methods perform undesirably due to the limited data and the inherent uncertainty associated with cyclone trajectory forecasting. This study applies a recommendation algorithm to address these challenges by focusing on predicting port impact rankings instead of predicting the duration of port impacts, which is often inaccurate and unreliable. First, we have collected comprehensive features of ports and hurricanes in the Gulf of Mexican and employed a modular time-series regression model to determine the duration of port impacts due to tropical cyclones, leveraging vessel count data extracted from the Automatic Identification System (AIS). Inspired by the recommendation algorithm, we recast tropical cyclones and ports as “user” and “items,” respectively, while the duration of port impacts represents their “interaction,” offering an innovative approach to model and analyze cyclone effects on ports. The Factorization Machine (FM) is adopted to learn the relationship between features (i.e., ports and cyclones) and subsequently conduct the port impact ranking. Finally, utilizing the hurricanes Alex, Ian, and Nicole that happened in 2022 as testing cases, the FM-based model excels in prediction performance and robustness against uncertainties compared to the widely used distance-based method. This study aims to provide port authorities and other stakeholders with a trustworthy tool for informed disaster management decisions, thereby enhancing port resilience.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
