A Case Study of Parameter Control in a Genetic Algorithm: Computer Network Performance

Genetic Algorithms use different parameters to control their evolutionary search for the solution to problems. However, there are no standard rules for choosing the best parameter values, being difficult to know whether the parameter values must be fixed during a run or must be modified dynamically. Besides, there are many theoretical results on parameter control, but however, very often real world problems call for shortcuts and/or some ad hoc solutions. This paper presents an effective approach for optimization of control parameters which is based on a meta-GA combined with an adaptation strategy to improve the GA performance. In order to validate the approach, it has been applied to verify the performance of a real system: a computer network. The results have been compared with the ones obtained for other methods: using fixed and adapted parameter values. A statistical analysis has been done to ascertain whether differences are significant.