Novel optimization methods and model for improving sustainability and efficiency in last-mile logistics
- Autores: Erick Rodríguez Esparza
- Directores de la Tesis: Antonio David Masegosa Arredondo (dir. tes.), Enrique Onieva Caracuel (dir. tes.)
- Lectura: En la Universidad de Deusto ( España ) en 2024
- Idioma: inglés
- Programa de doctorado: Programa de Doctorado en Ingeniería para la Sociedad de la Información y Desarrollo Sostenible por la Universidad de Deusto
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
This thesis delves into the challenges in Last-Mile Logistics (LML) optimization in urban environments, intensified by the rapid growth of e-commerce and the consequential demands on logistics systems, particularly highlighted during the rapid increase in online shopping and the transformative effects of the COVID-19 pandemic. It emphasizes the role of LML in connecting businesses with consumers, which is significantly impacting urban infrastructure and logistics efficiency.
This remarkable growth has posed challenges for logistics service providers (LSPs) in the LML sector, where efficiency and timely delivery are crucial. This shift also has significant social implications, particularly evident in urban environments.
Specifically, LSPs face the fact that LML is the most resource-intensive and costly segment of the delivery process, accounting for approximately 28% to 41% of the total delivery costs. Moreover, factors such as failed deliveries, lower delivery success rates in certain areas, and consumer environmental concerns further exacerbate these costs.
Furthermore, delivering goods in urban environments presents additional obstacles, such as complex traffic management, limited parking availability, and convoluted delivery routes. The increasing number of delivery vehicles increases urban congestion, dominates infrastructure use, and disrupts traffic flow, thereby affecting vehicle circulation and compromising the safety of pedestrians and cyclists, ultimately impacting overall urban mobility. As the number of freight vehicles in cities increases, LML operations significantly influence urban sustainability. Addressing these challenges requires innovative solutions that streamline the delivery process, reduce travel distances, and mitigate economic, environmental, and social impacts.
Among the proposed solutions to address the challenges and impacts of LML, three strategic policies are gaining considerable attention and yielding positive results: Vehicle Fleet Electrification, Demand Aggregation Infrastructure, and Enhanced Traffic Management. These policies aim to address various aspects of LML optimization. However, each of these policies presents unique challenges, necessitating the development of new tools and technologies to assist LSPs, public authorities, and infrastructure managers. This thesis aims to contribute to these advanced optimization methods and models, which will improve efficiency, sustainability, and resilience in LML.
The use of optimization methods in this thesis to address the challenges associated with the three policies is explained. To overcome the challenges posed by the vehicle fleet electrification policy, this thesis employs advanced optimization techniques to develop intelligent route planning solutions that minimize distances and reduce delivery times.
Specifically, the focus is on the Capacitated Electric Vehicle Routing Problem (CEVRP), which entails strategic planning of Electric Vehicle (EV) routes regarding their limited load and range. A significant gap in current CEVRP methodologies is their lack of scalability and robustness, especially when handling scenarios with high customer density. Therefore, a key contribution is the development of the Hyper-heuristic Adaptive Simulated Annealing with Reinforcement Learning (HHASARL) algorithm for solving the CEVRP, which demonstrates significant progress over existing methods in the state-of-the-art by identifying optimal solutions for high-dimensional instances on the IEEE WCCI2020 competition benchmark.
To overcome the challenges presented by the demand aggregation infrastructure policy, this thesis delves into the intricacies of the Stationary Parcel Locker (SPL) Location Problem. SPLs are physical facilities consisting of automated storage units placed in accessible locations, and the problem lies in determining the optimal locations to install the lockers and their setup in terms of size and volume. However, a significant gap in current SPL models is their lack of consideration for dynamic customer behavior and realistic constraints, including limited locker capacity, stochastic demand distributions, and variable costs and collection periods. Therefore, another key contribution is the development of the new Constrained Locker Location Problem under the Threshold Luce Model (CLLPTLM), which incorporates these mentioned factors and the probability of customer acceptance, offering realistic solutions. It was solved using an adapted Genetic Algorithm (GA) using data from synthetic and real datasets.
To overcome the challenges inherent in the enhanced traffic management policy, this thesis focuses on vehicle segmentation in image-based vehicle traffic monitoring. This monitoring process involves analyzing traffic camera video to identify and detect vehicles, aiming to estimate the state of roads. It enables real-time assessment of traffic flow, density, and patterns, providing valuable data to inform and optimize traffic control measures.
However, current methodologies have gaps in the accuracy and adaptability of vehicle segmentation, especially in adjusting to various environmental conditions and incorporating real-time data for traffic management. Therefore, the final contribution of this thesis is the Hyper-heuristic Genetic Algorithm based on Thompson Sampling with Diversity (HHGATSD), which demonstrates robustness and adaptability in vehicle segmentation within vehicle monitoring. The proposed method incorporates sophisticated optimization techniques to dynamically segment vehicles in traffic camera video frames, allowing for precise and adaptive estimation of road occupancy.
This dissertation underscores the potential for leveraging Metaheuristics (MHs) and Hyper-heuristics (HHs) in crafting solutions for the LML to improve sustainability in operational challenges for LSPs and society, mitigating economic, environmental, and social impacts. Future research directions are suggested, particularly in refining these algorithms and models to adapt to more complex and varied urban logistics scenarios, further integrating technological advancements to build smarter and more responsive urban logistics systems.
