Using fitness comparison disagreements as a metric for promoting diversity in Dynamic Optimisation Problems

In Evolutionary Algorithms (EAs), it is well-known that the adoption of diversity is highly beneficial for evolutionary search. This has also been explored and confirmed in Dynamic Optimisation Problems (DOPs) using EAs. Multiple works have been proposed to encourage diversity in EAs in the face of a change, where the most common form to promote diversity is to replace a number of individuals by new genetic material. A common element when adopting this form of diversity is the fact that, frequently, the number of individuals to be replaced is picked rather arbitrarily. In this work, we propose the adoption of the Kendall tau distance that quantifies pairwise dissimilarities among two lists (of fitness values) with the hope to make a better informed decision in terms of the number of individuals that need to be replaced in a population by new individuals. Results on continuous fitness-valued cases indicate that the adopted distance is beneficial in DOPs.