global optimization, differential evolution, adaptation, experimental comparison

The influence of used adaptive approach on the performance of algorithm is addressed. Adaptive approaches applied in competitive differential evolution and in differential evolution using an ensemble of mutation strategies and parameter values are compared. The approaches used in these algorithms can be divided into two parts: adaptive mechanism and pool of strategies. Four variants of algorithm combining mutually these two parts are compared experimentally in six benchmark functions at two levels of dimension. It was found that the variants using the pool of ensemble of mutation strategies and parameter values need mostly less number of function evaluation to reach the stopping condition, while the algorithms with pool of competitive differential evolution are more reliable.

The problem of optimal partitioning by minimizing pooled-within-variance of groups is addressed. Three state-of-the-art adaptive differential evolution algorithms are compared on four real-world data sets. A~novel hybrid differential evolution algorithm, including k-means algorithm for local search is proposed. The experimental comparison is done with either the plain adaptive differential evolution variants or the hybrid algorithms. Experimental results showed that hybrid algorithms are substantially better preforming when compared with plain differential evolution variants. Among hybrid variants, the competitive differential evolution appeared to be the most efficient.