The Importance of Proper Diversity Management in Evolutionary Algorithms for Combinatorial Optimization

Premature convergence is one of the most important recurrent drawbacks of Evolutionary Algorithms and other metaheuristics. As a result, several methods to alleviate this problem have been devised. One alternative is to explicitly control the diversity of the population. In this chapter, a recently proposed survivor selection strategy is incorporated into a memetic algorithm and analyzed using three different combinatorial optimization problems. This strategy is based on adopting multi-objective concepts for solving single-objective problems by considering the contribution to diversity as an explicit objective. Additionally, it incorporates the principle of adapting the balance between exploration and exploitation to the different stages of the optimization by taking into account the stopping criterion and elapsed time. These new methods provide important benefits when compared to more mature methods that rely on different principles to delay convergence of the population. Additionally, new best-known solutions are generated for several instances of the problems, thus providing proofs of the considerable benefits and robustness yielded by the schemes that incorporate this novel replacement strategy.