A machine learning-based adaptive heuristic for vessel scheduling problem under uncertainty via chance-constrained programming

Efficient vessel scheduling in port is critical for enhancing navigational efficiency. However, it faces substantial challenges due to unforeseeable events. In this context, this paper addresses the vessel scheduling problem with stochastic sailing times in port. The problem is formulated into a cha...

Full description

Saved in:
Bibliographic Details
Published in:Computers & electrical engineering Vol. 119; p. 109523
Main Authors: Li, Runfo, Zhang, Xinyu, Wang, Chengbo, Cui, Jinlong, Mu, Mengfeng
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.10.2024
Subjects:
Adaptive differential evolution algorithm
Chance-constrained programming
Machine learning
Stochastic sailing times
Vessel scheduling
Adaptive differential evolution algorithm
Chance-constrained programming
Vessel scheduling
Stochastic sailing times
Machine learning
ISSN:0045-7906
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Efficient vessel scheduling in port is critical for enhancing navigational efficiency. However, it faces substantial challenges due to unforeseeable events. In this context, this paper addresses the vessel scheduling problem with stochastic sailing times in port. The problem is formulated into a chance-constrained programming (CCP) model and then transformed into an equivalent deterministic programming problem. A novel approach utilizing a machine learning-based adaptive differential evolution algorithm (MLDE) is proposed to address this model. In MLDE, a K-means clustering method is employed to generate initial population, aiming to enhance the population's quality and diversity while mitigating the impact of random interference. Throughout the mutation and crossover stages, we introduce a parameter adaption strategy based on Q-learning, which is established as a Markov decision process (MDP) model. The model effectively defines the state, action, and reward functions to guide the population toward selecting the optimal scaling factor and crossover probability parameters. Numerical experiments based on different instance sizes are conducted at the Comprehensive port. The obtained results reveal the superior performance of the MLDE algorithm in comparison to existing metaheuristic algorithms and traditional differential evolution (DE) variants. A statistical analysis experiment is also conducted to further confirm the superiority of MLDE.
ISSN:0045-7906
DOI:10.1016/j.compeleceng.2024.109523