Learning Automata-Based Particle Swarm Optimizer

Particle Swarm Optimizer (PSO) is a population-based optimization technique applied to a wide range of problem s. However, its performance may suffer from getting trapped into local optima because of its fixed search pattern. Most variants of PSO use the same search strategy in their whole search process, which may hurt their performance for those cases that require different strategies at different search stages. In order to enable the swarm to adaptively choose an appropriate learning strategy according to its current search stage, this work proposes a Learning Automata-based Particle Swarm Optimizer. Its learning automaton learns the search stage of a swarm and selects its corresponding search strategy for particles. In an early search stage, the automaton selects a global search strategy with a large probability. While in a late search stage, it likely chooses a local search strategy. The update of the selection probabilities of candidate strategies is not predetermined but learned from the evolution process by the learning automaton. Experimental results performed on CEC2013 benchmark functions verify its outstanding performance in comparison with several representative PSO variants.