Adaptive Crossover Selection for Differential Evolution to Solve Global Optimization Problems

In response to the escalating demand for optimal solutions to intricate optimization problems in real world applications, differential evolution (DE) algorithms have shown very promising performance. However, most of existing DE studies leverage only one crossover strategy to generate the offspring. Since different crossover strategies preserve different advantages in generating the offspring with different qualities, this paper develops four hybridization ways to assemble two popular crossover schemes, namely the binomial crossover and the exponential crossover. In this way, it is expected that the advantages of the two crossover schemes are integrated to generate more promising offspring and thus help DE achieve more promising performance. Experiments have been extensively conducted on the CEC2017 benchmark functions by embedding the four hybridizations of the two crossover mechanisms into DE with totally five different mutation schemes. Experimental results have demonstrated the effectiveness and efficacy of the proposed four hybridizations of the two crossover strategies.