MSGM: A Markov Model Based Similarity Guide Matrix for Optimising Ordered Problems by Balanced-Evolution Genetic Algorithms

Abstract

Where traditional genetic algorithms tend to prematurely converge on local optima, adaptive strategies aim to maintain a healthy level of population diversity by introducing randomness to the population. Often times this is done through adjusting control parameters according to diversity measurements. While these approaches introduce diversity, they do not aid in focusing or directing the search effort. Meanwhile, other works in the literature propose creating individuals designed to improve the population's health and quality but their effectiveness is limited outside of general problems. This article proposes novel sequence-wise approach to designing and editing genotypes for ordered problems. A Markov model based similarity guide matrix (MSGM) is used to determine the relationships between gene nodes in order to produce new genotypes that focus on improving fitness and increasing population diversity. The proposed MSGM based approach is implemented in a balanced-evolution genetic algorithm framework in order to investigate its characteristics with encouraging results demonstrating its effectiveness when solving combinatorial ordered optimisation problems.