Page 72 - 2024F

UEC Int’l Mini-Conference No.53 65

An Adaptive Evolutionary Algorithm for Dynamic Multi-objective

Optimization

Tasin-Mobin Mohtadi ∗1 and Hiroyuki Sato 2

1 UEC Exchange Study Program (JUSST Program)
2 Information and Communications
The University of Electro-Communications, Tokyo, Japan

Keywords: Adaptive DNSGA-II, DMOPs, Evolutionary Algorithm, Reinitialization Factor, Adaptive
Mechanism, Solution Ranking Distance.

Abstract

This study proposes an adaptive evolutionary algorithm, named Adaptive DNSGA-II, for solving
dynamic multi-objective optimization problems (DMOPs), where objective functions change over time.
The conventional DNSGA-II reinitializes a zeta proportion of individuals in the population when a
change in the objective functions is detected. The parameter zeta is user-defined and significantly
impacts search performance. However, conventional DNSGA-II employs a fixed zeta throughout the
search. A small zeta is effective when the problems before and after the change are similar, whereas a
large zeta is effective when the problems are dissimilar. This study introduces an adaptive mechanism to
adjust zeta within the DNSGA-II framework dynamically. The proposed Adaptive DNSGA-II calculates
the distance between solution rankings based on objective values before and after a problem change and
uses this distance to adjust zeta. A short distance brings a small zeta, leading to a small reinitialization,
while a long distance brings a large zeta, leading to a large reinitialization. The effectiveness of Adaptive
DNSGA-II is evaluated on benchmark DMOPs by comparing its performance with that of conventional
DNSGA-II

∗
The author is supported by JASSO Scholarship.