Dynamic Behavior Analysis of Membrane-Inspired Evolutionary Algorithms

Abstract

A membrane-inspired evolutionary algorithm (MIEA) is a successful instance of a model linking membrane computing and evolutionary algorithms. This paper proposes the analysis of dynamic behaviors of MIEAs by introducing a set of population diversity and convergence measures. This is the first attempt to obtain additional insights into the search capabilities of MIEAs. The analysis is performed on the MIEA, QEPS (a quantum-inspired evolutionary algorithm based on membrane computing), and its counterpart algorithm, QIEA (a quantum-inspired evolutionary algorithm), using a comparative approach in an experimental context to better understand their characteristics and performances. Also the relationship between these measures and fitness is analyzed by presenting a tendency correlation coefficient to evaluate the importance of various population and convergence measures, which is beneficial to further improvements of MIEAs. Results show that QEPS can achieve better balance between convergence and diversity than QIEA, which indicates QEPS has a stronger capacity of balancing exploration and exploitation than QIEA in order to prevent premature convergence that might occur. Experiments utilizing knapsack problems support the above made statement.