Function Management in Multi-population Methods with a Variable Number of Populations: A Variable Action Learning Automaton Approach

Abstract

The fitness evaluation management (FEM) has been successfully applied to improve the performance of multi-population (MP) methods with a fixed number of populations for dynamic optimization problems (DOPs). Along with the benefits they offered, the proper number of populations in MP with a fixed number of populations is difficult to determine. The number of populations in this approach is specified according to the number of local optima and before the commencement of the optimization process. However, the number of optima in real-world problems is mostly unknown. Therefore, it is valuable to study the usefulness of FEM for MP with a varying number of populations. In this chapter, the concept of FEM is extended to the MP approach with a variable number of populations. Since the number of sub-populations in this method is varied during the run, the FEM system should adapt to this new emerging challenge. To do so, we use a variable-action set learning automaton (VALA) to change the action-set of the learning automaton (LA) accordingly. As a starting point, we first need to use an MP method with a variable number of populations as the base method to develop our proposed FEM scheme. Although several frameworks exist to create the MP with a variable number of sub-populations, we have considered the framework for locating and tracking multiple optima in dynamic environments using clustering particle swarm optimizer (CPSO). CPSO has been extensively studied in the past, showing very promising results in challenging DOPs. Therefore, in the next, we review the CPSO in brief. Then we describe our proposal in detail. Finally, the effectiveness of the proposed FEM is evaluated through numerical experiments.