A study of data-driven optimization models with application to freight transportation

Name: Thananut Phiboonbanakit

keyword: ABC analysis

; Artificial intelligence

; Behavior analysis

; Cost analysis

; Data mining

; Freight transport

; Logistics

; Machine learning

; Reinforcement learning

; Vehicle routing problem

Abstract: The vehicle routing problem (VRP) is a combinatorial optimization and integer programming problem that asks, “What is the optimal set of routes for a fleet of vehicles to traverse so that they can successfully deliver items to a given set of customers?” The VRP is solved to optimize vehicle routes by minimizing the traversal costs and without violating constraints. The performance can be assessed using the variables of capacity, energy consumption, and time-windows. In real-world applications, the VRP is applied by logistics agencies to optimize vehicle usage for delivering goods to customers. However, owing to the rapid growth in demand for logistics transportation, vehicle route optimization has become more crucial than ever. Vehicle route optimization tasks are becoming increasingly complicated because of the large number of requirement constraints and uncertainties in the environment. Thus, a solution that is calculated using daily dynamic information is required. Traditional route optimization methods cannot be applied directly, because most VRP models are static. This dissertation presents a proposal for a new methodology that was developed for vehicle route optimization in an attempt to address these issues. Essentially, the models in the proposed methodology comprises four general interconnected stages: data processing, detection, optimization, and analysis and validation. First, the data are collected, and the mathematics optimization problem is transformed into a reinforcement-learning problem using reinforcement problem formulation strategies. Second, behavior analysis is used to detect and evaluate the action of the agent responsible for reinforcement learning for vehicle route optimization. Third, new procedures are proposed to determine the utility and reward of the action of the reinforcement-learning agent using multi-attribute utility theory. The application of the procedure differs from that of general and hybrid models. To train the agent, two different types of algorithms comprising an actor- critic algorithm and asynchronous actor-critic algorithm are used. After all crucial steps are formulated and executed, vehicle route optimization is performed. Finally, case studies are used to evaluate and demonstrate the practicality of these methodologies. The result is analyzed and validated against the baseline using real operational data and well-known state-of-the-art approaches. The experimental results from the route optimization process demonstrate that the new hybrid model, which accounts for dynamic information, yields a profit of up to 27,381.35 THB with a 24.86% improvement on average for the non-incident case study and a profit of up to 48,895.68 THB with 57.91% improvement on average for the incident case. The results confirm that the new methodologies presented in this dissertation can replace individual experts without modification and monitoring. Further, the proposed approach handles the uncertainty in the environment more efficiently than current state-of-the-art approaches

Thammasat University. Thammasat University Library

Address: BANGKOK

Email: preserv@tu.ac.th

Name: Teerayut Horanont

Role: advisor

Name: Huynh, Van-Nam

Role: co-advisor

Created: 2020

Modified: 2022-11-07

Issued: 2022-11-07

วิทยานิพนธ์/Thesis

application/pdf

DOI: 10.14457/TU.the.2020.74

eng

DegreeName: Doctor of Philosophy

Level: Doctoral Degree

Descipline: Engineering and Technology

Grantor: Thammasat University

©copyrights Thammasat University

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	9282thananut.pdf	9.6 MB	3	2023-12-23 12:36:00

ใช้เวลา

0.034141 วินาที