Abstract

Dynamic optimization in which global optima and local optima change over time is always a hot research topic. It has been shown that particle swarm optimization works well when facing dynamic environments. On the other hand, a learning automaton can be considered as an intelligent tool (agent) which can learn what action is the best interacting with its environment. The great deluge algorithm is also a search algorithm applied to optimization problems. All these algorithms have their drawbacks and advantages. This paper explores how one can combine these algorithms to reach better performance in dynamic spaces. Indeed a learning automaton is employed per particle in the swarm to decide whether its particle updates its velocity (and consequently its position) considering the best global particle position, local particle position or a combined position extracted out of global and local particle position. Water level in the deluge algorithm is used in the progress of the algorithm. Experimental results on different dynamic environments modeled by moving peaks benchmark show that the combination of these algorithms outperforms PSO algorithm, fast multi-swarm method (FMSO), a similar particle swarm algorithm for dynamic environments, for all tested environments.

Key words: Particle swarm optimization, great deluge, learning automaton, moving peaks, dynamic environments.