Adaptive strategies for Evolutionary Algorithm monitoring

Parameter tuning in Evolutionary Algorithms (EA), is a great obstacle that can become the key to success. Good parameter settings can yield optimal solutions, while bad settings may trap the EA, thus removing the chances of finding the optimal solutions. Therefore, it is vital that an optimal set of parameters configuration is chosen. It is a common practice to have a human expert that analyzes such parameters and modifies them accordingly. Such process is inefficient and expensive, since it requires time and is subject to human fatigue; it even becomes impractical if the environment is dynamic. This work proposes 2 adaptive strategies to tune such parameters: One Step Variation and a Fuzzy Logic Controller. A ranking scheme and modeling is introduced to evaluate the adaptive strategies. Results show that it may be possible to tune the parameters in an EA for achieving better results, without the need of an expert.