Λογότυπο αποθετηρίου

Πλοήγηση ανά Συγγραφέα "Tzagkarakis, Georgios"

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω
Φιλτράρετε τα αποτελέσματα πληκτρολογώντας τα πρώτα γράμματα
Τώρα δείχνει 1 - 1 από 1
  • Μικρογραφία εικόνας
    Τεκμήριο
    Machine learning for routing problems: the case of vehicle routing problem (VRP)
    (2025-10-14) Tzagkarakis, Georgios; Τζαγκαράκης, Γεώργιος; Lorentziadis, Panagiotis; Bageri, Vasiliki; Kritikos, Emmanouil
    This thesis investigates the integration of machine learning (ML) techniques with classical optimization methodologies to address the Capacitated Vehicle Routing Problem (CVRP), a well-known NP-hard problem in logistics and operations research. Traditional approaches to solving VRPs rely on heuristics and metaheuristics, which, while effective, often lack adaptability and efficiency when faced with complex, large-scale datasets. Recent advancements in ML, and particularly deep learning, offer new opportunities to enhance these classical methods by leveraging data-driven insights. The research introduces a hybrid methodology that combines the Greedy Randomized Adaptive Search Procedure (GRASP) with Variable Neighborhood Descent (VND) and augments it with a Graph Isomorphism Network with edge features (GINE), a type of Graph Neural Network (GNN). The proposed model classifies initial solutions generated by the GRASP procedure as either promising or unpromising before applying local search optimization. This approach aims to reduce computational effort by selectively refining only those initial solutions likely to yield significant improvements. The methodology was evaluated using benchmark CVRP instances from the CVRPLIB dataset. Experimental results show that the hybrid GINE-GRASP-VND approach achieved moderate classification accuracy (60.12%) and demonstrated its ability to generalize to instances larger than those used in training. While the performance gains were limited in smaller instances, significant improvements were observed in larger-scale problems, validating the potential of ML-enhanced optimization methods and the feasibility of incorporating GNNs into metaheuristics frameworks for complex routing problems.