Success Rate Based Scaling Factor Adaptation in Dual-Population Differential Evolution

The paper proposes a simple adaptation technique for the scaling factor parameter in differential evolution algorithm. The adaptation is based on the success rate, i.e. the ration of the number of successfully generated new solutions to the total population size. This value is used to sample scaling factors for mutation using Cauchy distribution. The experiments are performed on two sets of test problems, taken from the benchmark suites of the Congress on Evolutionary Computation from 2017 and 2022. The experiments with different parameters of the proposed heuristic show that the it allows outperforming most state-of-the-art differential evolution variants, and it also works better than the success-history adaptation in most scenarios. The proposed method can be easily applied to any other differential evolution algorithm, as it only relies on the success rate and does not require storage of values in memory cells and averaging steps.