Differential evolution mutation crossover control parameters experimental comparison efficiency of search

Influence of a selected search strategy on the efficiency of differential evolution algorithm is studied. Twelve different kinds of mutation in combination with three different mutation and three various levels of parameter controlling the mutation are compared experimentally. Six commonly used scalable test functions at two levels of problem dimension are used as a benchmark. Experimental results shows that there are significant differences among the efficiency of various strategy and the efficiency of a strategy is problem-dependent. After statistical comparison of the performance of 60 tested search strategies we can conclude that strategies with mixed competitive crossover and randrl/1/ mutation perform best in average on these benchmark problems.