Abstract: A new adaptive differential evolution algorithm was put forward to improve search speed and avoid local optimal value. Sufficiently analyzing the characteristics of classic/adaptive mutation operators and the solution state, individuals were divided into three subgroups according to individual fitness values, thereby optimizing based on multiple populations, and different mutation operators were placed in different subpopulations. In addition, self-adaptive adjustment was introduced to adjust control parameters. Performance of the new approach was superior to other algorithms when tested on eight standard test functions.