Scholarlink

<p>This paper investigates a two-echelon vehicle routing problem involving a permanent fleet and occasional drivers who may reject delivery assignments. Customers are divided into two groups based on the type of driver assigned. When occasional drivers reject their tasks, an emergency vehicle is deployed to complete the deliveries. A stacking regressor, combining multiple machine learning models, predicts the acceptance probability for each driver-customer pair. These predictions inform the selection of drivers, customer assignments, and routing decisions across all transporters. The problem is formulated as a mixed-integer linear program to minimize total expected delivery costs while satisfying constraints on capacity, distance, and compensation. Solutions are obtained for test instances using both CPLEX and a custom metaheuristic, i.e., a variable neighborhood search algorithm enhanced with tabu search memory and adaptive local search strategies designed to efficiently handle large-scale problems. The metaheuristic consistently outperforms CPLEX within fixed time limits, achieving delivery cost reductions of up to 91.74% and computation time savings of up to 99.96%. Managerial insights highlight that selecting the optimal amount for transfer at the appropriate transshipment nodes, leveraging the permanent fleet strategically, using occasional drivers effectively, and reserving emergency resources for exceptions enhances reliability, flexibility, and cost-efficiency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-driven approach for cost-optimized last-mile delivery with transshipment nodes and occasional drivers

  • 期刊论文

  • 发表日期：

    2025年10月15日

查看全文
  • Dimitra G. Triantali,
  • Konstantinos E. Parsopoulos,
  • Konstantina Skouri

摘要

This paper investigates a two-echelon vehicle routing problem involving a permanent fleet and occasional drivers who may reject delivery assignments. Customers are divided into two groups based on the type of driver assigned. When occasional drivers reject their tasks, an emergency vehicle is deployed to complete the deliveries. A stacking regressor, combining multiple machine learning models, predicts the acceptance probability for each driver-customer pair. These predictions inform the selection of drivers, customer assignments, and routing decisions across all transporters. The problem is formulated as a mixed-integer linear program to minimize total expected delivery costs while satisfying constraints on capacity, distance, and compensation. Solutions are obtained for test instances using both CPLEX and a custom metaheuristic, i.e., a variable neighborhood search algorithm enhanced with tabu search memory and adaptive local search strategies designed to efficiently handle large-scale problems. The metaheuristic consistently outperforms CPLEX within fixed time limits, achieving delivery cost reductions of up to 91.74% and computation time savings of up to 99.96%. Managerial insights highlight that selecting the optimal amount for transfer at the appropriate transshipment nodes, leveraging the permanent fleet strategically, using occasional drivers effectively, and reserving emergency resources for exceptions enhances reliability, flexibility, and cost-efficiency.