Parameter Selection of Differential Evolution by another Differential Evolution Algorithm
Yen-Ching Chang · 2019
The performance of differential evolution (DE) highly depends on its control parameters, especially for the first proposed simple or standard DE. Control parameters suitable for one objective function are generally not beneficial to another. To automatically find out optimal control parameter settings for different objective functions, we propose a novel technique, parameter selection of DE by another DE algorithm. The conventional DE is one-level DE with various mutation and crossover schemes, and even with different restart mechanisms, and thus our proposed DE can be called a two-level DE algorithm or simply called a two-level algorithm. Experimental results show that our proposed two-level DE can easily find out the true minimum values of four benchmark functions even under different hyperparameters.