Enhancing the virtual loser genetic algorithm for dynamic environments

Abstract

In recent years, several memory schemes have been used in Evolutionary Algorithms (EAs) for dynamic optimization problems (DOPs). The Virtual Loser Genetic Algorithm (VLGA), recently proposed, uses a novel type of associative memory to deal with DOPs. This memory scheme memorizes past errors concerning the performed mutations and uses this information to create new individuals when a change in the environment occurs. In this paper the VLGA is further investigated in order to enhance its performance in different types of DOPs: the influence of an important parameter is analyzed, and the interaction between the memory scheme and the use of immigrants is also investigated. A novel immigrant scheme is proposed and compared with the random immigrants approach. The results show that the investigated methods significantly enhance the previous version of the VLGA for cyclic and random environments.