Theses and Dissertations

ORCID

https://orcid.org/0009-0001-0247-8564

Issuing Body

Mississippi State University

Advisor

Marufuzzaman, Mohammad

Committee Member

Bian, Linkan

Committee Member

Ma, Junfeng

Committee Member

Sirinterlikci, Arif

Date of Degree

5-12-2023

Document Type

Dissertation - Open Access

Major

Industrial Engineering

Degree Name

Doctor of Philosophy (Ph.D)

College

James Worth Bagley College of Engineering

Department

Department of Industrial and Systems Engineering

Abstract

Adaptive Large Neighborhood Search (ALNS) is a fairly recent yet popular single-solution heuristic for solving discrete optimization problems. Even though the heuristic has been a popular choice for researchers in recent times, the parallelization of this algorithm is not widely studied in the literature compared to the other classical metaheuristics. To extend the existing literature, this study proposes several different parallel schemes to parallelize the basic/sequential ALNS algorithm. More specifically, seven different parallel schemes are employed to target different characteristics of the ALNS algorithm and the capability of the local computers. The schemes of this study are implemented in a master-slave architecture to manage and assign loads in processors of the local computers. The overall goal is to simultaneously explore different areas of the search space in an attempt to escape the local minima, taking effective steps toward the optimal solution and, to the end, accelerating the convergence of the ALNS algorithm. The performance of the schemes is tested by solving a capacitated vehicle routing problem (CVRP) with available wellknown test instances. Our computational results indicate that all the parallel schemes are capable of providing a competitive optimality gap in solving CVRP within our investigated test instances. However, the parallel scheme (scheme 1), which runs the ALNS algorithm independently within different slave processors (e.g., without sharing any information with other slave processors) until the synchronization occurs only when one of the processors meets its predefined termination criteria and reports the solution to the master processor, provides the best running time with solving the instances approximately 10.5 times faster than the basic/sequential ALNS algorithm. These findings are applied in a real-life fulfillment process using mixed-mode delivery with trucks and drones. Complex but optimized routes are generated in a short time that is applicable to perform last-mile delivery to customers.

Download

Share

COinS