Experimental Comparison of Replacement Strategies in Steady State Genetic Algorithms for the Dynamic MKP

Abstract

In the steady-state model for genetic algorithms (SSGA), the choice of a replacement strategy plays an important role in performance. Being able to handle changes is important for an optimization algorithm since many real-world problems are dynamic in nature. The main aim of this study is to experimentally compare different variations for basic replacement strategies in a dynamic environment. To cope with changes, a very simple mechanism of duplicate elimination is used. As an example of a dynamic problem, a dynamic version of the multi-dimensional knapsack problem is chosen. The results obtained here are in keeping with previous studies while some further interesting results are also obtained due to the special landscape features of the chosen problem.