Differential Evolution with a Dimensional Mutation Strategy for Global Optimization

Differential evolution (DE) is an efficient means of solving the global optimization problems. In the classical and adaptive DE algorithms, each individual has the same value of F, the amplification factor of the difference vector, in all dimensions. However, some researchers' works showed that population may have different characteristics of converging in different dimensions. Individuals may be very similar to each other in some dimensions, but they may have obvious difference in other dimensions. In this paper, a dimensional mutation strategy is proposed for DE. In this new mutation strategy, each individual has different values of F in different dimensions. This new mutation strategy was implied into jDE algorithm and tested on the CEC05 functions. The experimental results suggested that the dimensional mutation can make a better performance of the jDE algorithm.