Real-time data-driven automatic design of multi-objective evolutionary algorithm: A case study on production scheduling

Biao Zhang

doi:10.1016/j.asoc.2023.110187

Return

Real-time data-driven automatic design of multi-objective evolutionary algorithm: A case study on production scheduling

2023-05-01

PRE

DOI：10.1016/j.asoc.2023.110187

Original

Original request for help

Save

Abstract

En 中文

Multi-objective evolutionary algorithms (MOEAs) have become an important choice for solving multi -objective optimization problems. The performance of MOEAs is highly dependent on the algorithm configuration. Therefore, the algorithm configuration is an essential task in the development and application of MOEAs. In this paper, a real-time data-driven automatic design method for configuring an MOEA with minimal user interference is developed. Real-time data are driven in two ways. One lies in that the learning model is constructed based on the elite configurations selected by the Iterated F-Race (I/F-Race), which is used to bias the sampling toward the best configurations. Another is that the decision tree model is constructed by collecting the evaluated configurations in the process of I/F-Race as the data, and used to help identify the potential configurations to improve the sampling quality. In addition, a configurable MOEA (CMOEA) framework is summarized by integrating three general fitness assignment methods. In the experimental study, a case study on a multi-objective hybrid flowshop scheduling problem is conducted. By comparing with other variants of I/F-Race, the developed method is verified to have the ability of evaluating the promising configurations more fully and conceiving the best MOEA. Compared with the famous frameworks and state-of-the-art MOEAs, the proposed CMOEA framework and the automated algorithm show their superiorities based on different performance metrics.& COPY; 2023 Elsevier B.V. All rights reserved.

Keywords:

Multi-objective evolutionary algorithm

Automatic algorithm design

Hybrid flowshop scheduling problem

F-race

Decision tree