Multinational GAs: Multimodal Optimization Techniques in Dynamic Environments

Fitness landscapes of real world problems are in general considered to be complex and often with both local and global peaks. In the static case the local peaks are interesting because they represent other potential solutions to the problem. In dynamic problems the fitness landscape changes over time, so a local optimum might rise and become the new global optimum. In this context it would be beneficial to search for both local and global optima, because the algorithm would then have performed most of the search when the global optimum change. This paper describes the multinational GA and its application to six dynamic problems. The multinational GA is a self-organized genetic algorithm for multimodal optimization, which structures the population into subpopulations based on a method for detecting valleys in the fitness landscape. The experiments showed that multimodal optimization techniques are useful when optimizing dynamic problems with multiple peaks. A non-adaptive version was tested against a self-adaptive algorithm with genetic encoded parameters. The self-adaptive version outperformed the non-adaptive on a simple dynamic problem. However, this was not the case on a slightly more complex problem.