The QUATRE structure: An efficient approach to tackling the structure bias in Differential Evolution
Zhenyu Meng, Jeng‐Shyang Pan, Fang Lin · 2019
Differential Evolution (DE) is a very simple but powerful population-based stochastic algorithm for complex optimization, and the evolution of a certain individual in DE actually implements the search of the solution space. Usually, the movement of a certain individual from a former location to a new one during evolution is conducted by crossover operation, and there are two commonly used crossover schemes in different DE variants implementing such movement: one is exponential crossover and the other is binomial crossover. Though the binomial crossover conquers the representational bias existing in exponential crossover by separating all the parameters and treating them independently, the binomial crossover still has a bias, called the exploration bias, from a higher dimensional perspective of view. Therefore, in this paper, we first present the reason why such an exploration bias is still existing in the binomial crossover, and then present a novel QUATRE structure to tackle this bias. The new QUATRE structure is different from the former proposed QUATRE algorithm as some novel standards and adaptation schemes are involved into the new structure. This structure is validated under Congress on Evolutionary Computation (CEC) benchmarks for real-parameter single objective optimization, and the experiment results show its superiority.