Revista internacional de inteligencia de enjambre y computación evolutiva

Abstracto

Scheduling Functions for Position Updating in Population Based Optimization Algorithms

Jeremy Mange* and Sara Pace

In many population-based optimization algorithms (Evolutionary Algorithms, Particle Swarm Optimization, etc.), each iteration of the algorithm involves a procedure-specific set of operations for each population member, followed by a resulting update of the position of that member within the problem search space. However, for algorithms in which these operations involve only a single population member and not the population as a whole, there is no inherent need to update every member at every iteration. In this paper, we propose a generalization of this updating procedure wherein a “scheduling” function is defined to dictate the ordering of updates through the application of algorithm, thus considering the typical procedure of updating every population member at every iteration as a particular “round-robin" schedule. Using the standard Particle Swarm Optimization algorithm (SPSO-2011) as a basis for demonstrating the concept, we compare a number of different scheduling functions and show that several of these functions outperform the typical round-robin schedule for a set of benchmark optimization problems.