Hybrid Evolutionary Algorithms Based On Mean Shift

In recent years,evolutionary algorithm has been widely used in the field of complex optimization because of its simplicity,flexibility and robustness.Compared with traditional optimization methods,evolutionary algorithms have better performance in solving hard problems,such as complex function optimization,constrained optimization,dynamic optimization and multi-objective optimization.However,evolutionary algorithm has some shortcomings,like slow convergence,easy to fall into local optimum.Aiming at these problems,researchers put forward some substantive solutions and the hybrid evolutionary algorithms have been attracting more and more attention.In this paper,we propose three hybrid evolutionary algorithms based on multi-operators to improve the performance of evolutionary algorithms and explore the effect of mean shift operator and other operators.The main contributions are as follows:(1)We propose a generation operator based on mean shift to improve the algorithm performance by using the search ability of mean shift.(2)We propose a hybrid evolutionary algorithm based on mean shift operator and estimation of distribution algorithm operator for combining mean shift operator with probability model operator.Estimation of distribution algorithm operator creates solutions by global information and mean shift operator generates solutions by local information.Based on the idea of information fusion,we combine two kinds of solutions at the chromosome level,which make use of both information efficiently.(3)We propose a hybrid evolutionary algorithm based on mean shift operator and composite differential evolution operator for combining mean shift operator and crossover-mutation operator.The local search method in mean shift provide a clear direction for finding solutions and the composite differential evolution operator guarantees the diversity.(4)We propose a hybrid evolutionary algorithm based on mean shift operator,composite differential evolution operator and estimation of distribution algorithm operator.We use the different characteristics of the three operators respectively to improve the search ability and convergence of the evolutionary algorithm.The experimental results show that the three hybrid evolutionary algorithms based on mean shift operator can significantly improve the performance and accelerate the convergence speed of evolutionary algorithm.Also,each operator has different roles and not a single one can be omitted.