This thesis focuses on methods based on the integration of operations research and machine learning for the definition of the best consignment strategy for storing containers in an export yard of a maritime container terminal. Adopting the consignment strategy identifies better rules for grouping homogeneous containers characterized by different combinations of characteristics (size, type, weight, and destination), which are defined simultaneously with the assignment of each group of containers to the available positions (bay-locations) in the yard. This problem is called the Storage Strategy Definition problem in the current thesis. To support tackling this problem, a systematic literature review studying 303 contributions is conducted. In this systematic literature review, a new classification schema is proposed, and a PRISMA model is used for the paper-investigation. This overview supports the identification of current trends in the integration of machine learning and operations research in the context of container terminals. By analyzing the related works concerning storage strategies, a data-driven approach has been suggested for defining the best consignment strategy. A novel aspect of this work is the integration of unsupervised learning and optimization models, which allows the solution of large instances within a few seconds. Data distribution based on the containers' characteristics is analyzed. Five unsupervised learning algorithms are tested to obtain data-driven storage rules considering only the weight of containers and all features of containers in two separate campaigns. As Spectral Clustering outperforms others using a benchmark model, clustering results obtained by Spectral Clustering are then treated as input datasets for two optimization models. These machine learning-based integer programming models are described and compared: the main difference lies in how containers are assigned to bay-locations, shifting from a time-consuming individual container assignment to the assignment of groups of containers, which offers significant advantages in computational efficiency. Experimental results using randomly generated instances and real instances show good results. Some methodological and managerial insights are discussed in the conclusions.
Unsupervised Learning-based Optimization Models for Defining Storage Rules in Maritime Container Yards.
XIE, HAOQI
2025
Abstract
This thesis focuses on methods based on the integration of operations research and machine learning for the definition of the best consignment strategy for storing containers in an export yard of a maritime container terminal. Adopting the consignment strategy identifies better rules for grouping homogeneous containers characterized by different combinations of characteristics (size, type, weight, and destination), which are defined simultaneously with the assignment of each group of containers to the available positions (bay-locations) in the yard. This problem is called the Storage Strategy Definition problem in the current thesis. To support tackling this problem, a systematic literature review studying 303 contributions is conducted. In this systematic literature review, a new classification schema is proposed, and a PRISMA model is used for the paper-investigation. This overview supports the identification of current trends in the integration of machine learning and operations research in the context of container terminals. By analyzing the related works concerning storage strategies, a data-driven approach has been suggested for defining the best consignment strategy. A novel aspect of this work is the integration of unsupervised learning and optimization models, which allows the solution of large instances within a few seconds. Data distribution based on the containers' characteristics is analyzed. Five unsupervised learning algorithms are tested to obtain data-driven storage rules considering only the weight of containers and all features of containers in two separate campaigns. As Spectral Clustering outperforms others using a benchmark model, clustering results obtained by Spectral Clustering are then treated as input datasets for two optimization models. These machine learning-based integer programming models are described and compared: the main difference lies in how containers are assigned to bay-locations, shifting from a time-consuming individual container assignment to the assignment of groups of containers, which offers significant advantages in computational efficiency. Experimental results using randomly generated instances and real instances show good results. Some methodological and managerial insights are discussed in the conclusions.File | Dimensione | Formato | |
---|---|---|---|
phdunige_4408187.pdf
accesso aperto
Dimensione
9.34 MB
Formato
Adobe PDF
|
9.34 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/218829
URN:NBN:IT:UNIGE-218829