Differential evolution, exponential crossover, binomial crossover, real-world problems, experimental comparison

The influence of exponential crossover on the performance of differential evolution (DE) is studied experimentally. Nine DE variants known from literature are selected for experimental comparison. Four selected DE variants come from the state-of-the-art set, four algorithms are chosen from recently published well-performing DE variants and the last algorithm is standard classical DE. The experimental comparison was carried out in pairwise manner, the performance of two variants of each selected algorithm (one using binomial crossover and the latter variant using exponential crossover) was compared on each problem of CEC 2011 real-world optimization benchmark. The DE variants with exponential crossover outperformed signicantly the binomial-crossover counterparts in the algorithms, in four algorithms the performance was approximately equal, and only in jDE case the variant with binomial crossover outperformed signicantly the counterpart in nine test problems, while it was signicantly worse in four problems. The results show that the application of DE strategies with the exponential crossover is beneficial in the design of new adaptive DE variants.