Evolutionary fuzzy rules for dynamic optimisation

Abstract

Dynamic optimisation problems (DOPs) have attracted a lot of studies from the genetic algorithms (GAs) community due to the importance in real-world applications. Many researchers have proposed algorithms to enhance the performance of GAs in DOPs. This paper proposes a number of remedies to improve the performance of GAs in DOPs. First, we use GAs with dynamic niche sharing (GADNS) to maintain diversity in the population and to find multiple optima. Second, an unsupervised fuzzy clustering algorithm is utilised to track multiple optima and to overcome some weaknesses of GADNS such as the use of fixed sharing outside the dynamic niches. Third, we use a fuzzy system to adjust the mutation and crossover rates, in order to diversify the population. A modified tournament selection is used to control the selection pressure. The effectiveness of our approach is demonstrated by using the generalised dynamic benchmark generator (GDBG) and the moving peaks benchmark.