An efficient ensemble of GA and PSO for real function optimization

Wolpert and Macready asserted that no single search algorithm is best on average for all problems, which is confirmed by most practical experiences. Therefore, optimization results are highly dependent on which optimization algorithm is selected and what values its parameters take. So, it is interesting to explore some more robust optimization ensembles to reduce this dependency. This paper proposed a simple and efficient ensemble model of genetic algorithm (GA) and particle swarm optimization (PSO). This ensemble holds one population called public population on which GA and PSO run. After running on the public population, each component optimization gets an offspring population. Then the next generation public population will be renewed by the combination of both offspring populations according to their best individuals´ fitness. In order to illustrate that the ensemble is superior to its component algorithms, we compared this ensemble with GA and PSO on a suit of 36 widely used benchmark problems. Results show that the ensemble is best on many more benchmarks than PSO or GA in terms of whether the average best or the best of 30 independent trials, especially in high dimensional spaces.