On Random Numbers and the Performance of Genetic Algorithms

Pseudo random number generators (PRNGs) are the basic input to the stochastic selection, recombination, and mutation operations of genetic algorithms (GAs). Although it does not seem like a crucial decision, recent studies suggest that the choice of PRNG can affect the performance of GAs. The objective of this paper is to study the effect of PRNGs on a simple GA, and to identify the components that are most affected by the PRNG. The paper presents ablation experiments using two PRNGs and true random numbers from an atmospheric noise source. The experiments show that the PRNG used to initialize the population is critical, but the PRNG used as input to other operations does not affect the performance significantly. We confirmed these results with additional experiments that isolated single components of the GA. In a few cases, we obtained improved results with a poor PRNG, but we were unable to obtain improvements consistently across the test functions used or with different seeds. The results suggest that, in accordance with common practice in other fields, it is preferable to use the best PRNG available to avoid muddling the interpretation of the results.