Learning-driven memetic algorithm for solving integrated distributed production and transportation scheduling problem

doi:10.1016/j.swevo.2025.101945

Return

Learning-driven memetic algorithm for solving integrated distributed production and transportation scheduling problem

2025-07-01

PRE

Shicun Zhao

Hong Zhou *

DOI：10.1016/j.swevo.2025.101945

Original

Original request for help

Save

Abstract

En 中文

Production and transportation scheduling are critical components in modern manufacturing. However, existing studies on their integrated optimization are still limited, and most of them focus on the integration of production and local logistics within the shop. Different from previous investigations, this paper considers the integration of production scheduling with transportation across enterprises, which is especially typical and significant for production management in the large-scale distributed manufacturing environment. Considering the energy-aware orientation and production performance, the problem is formulated as a bi-objective integrated production planning and transportation scheduling problem for distributed flexible job shops. A mixed-integer linear programming model is developed to describe the considered problem with the aim of optimizing customer satisfaction and total energy consumption. To address this problem, an enhanced memetic algorithm with a reinforcement learning-driven breeding mechanism (RDMA) is proposed. Unlike existing literature that uses reinforcement learning to adjust parameters or select operators, RDMA marks the initial use of reinforcement learning to recommend the most suitable parents for breeding offspring. Additionally, a knowledge-driven adaptive variable neighborhood search is designed to make incremental improvements to the best solutions and continuously enhance RDMA's local search performance. Comparative results highlight the benefit of the reinforcement learning-based breeding mechanism and demonstrate the superiority of RDMA over major existing state-of-the-art algorithms. Moreover, experimental analysis indicates that each component in RDMA positively affects search performance, and their collaboration yields the best results.

Keywords:

Reinforcement learning

Multiobjective optimization

Integrated scheduling

Distributed flexible job shop

Production and transportation