The Influence Analysis of Pseudorandom Number Generators and Low Discrepancy Sequences for the Family of Compact Genetic Algorithms: Search Behavior Research from Outside Causes to Internal Causes

摘要

According to probability vectors, compact genetic algorithms (CGAs) generate new individuals by using pseudorandom number generators (PRNGs) or low discrepancy sequences (LDSs). These new generated individuals are the only factors which determine the search directions in CGAs. Therefore, we experimentally study the relationship between probability vectors and PRNGs (or LDSs). Moreover, we primarily investigate the influence analysis of PRNGs and LDSs for the effectiveness and efficiency of the family of CGAs by using analysis of variance (ANOVA), success rate and success performance. According to experimental results, we provide conclusive evidence to suggest using PRNGs (or LDSs) for CGAs. In essence, the frameworks of CGAs and the update method of probability vectors of CGAs are the internal causes that determine the performance of CGAs for different PRNGs and LDSs.