Selection Based on Colony Fitness for Differential Evolution

Differential evolution (DE) is a competitive and reliable computing technique for continuous optimization. A diversity-based selection has been proved to be valid to improve the performance of DE. However, further study can be done. In this paper, we propose two versions of colony fitness, fitness with the consideration of diversity information. Selection based on the first version of the colony is embodied in DE/rand/1, a basic DE algorithm, while selection based on the second version is used in CoBiDE, a state-of the-art DE algorithm. Our experiments are based on the 2005 Congress on Evolutionary Computation and the 2014 Congress on Evolutionary Computation benchmark functions. Experimental results show that our modification on algorithms leads to significantly better solutions than before.