Improved evolutionary programming algorithm based on heuristic mutation

The common evolutionary programming (EP) algorithms are of poor robustness because they perform the mutation based on a fixed probability distribution. In this paper, first, the influence of mutation operators on the computational efficiency of evolutionary programming algorithms is analyzed, and the essential drawback of Gauss, Cauchy and Lévy mutation operators, namely the lack of heuristic information, is pointed out. Then, a heuristic mutation operator based on the differential information among individuals is designed, which uses the difference between two individuals to update the mutated variables and to provide chances for an individual to maintain its status quo in some dimensions. With the help of the proposed heuristic mutation operator, evolutionary programming algorithms can adapt to different continuous optimization problems and the algorithm robustness improves. Numerical experiments of several Benchmark problems demonstrate that the improved evolutionary programming algorithm based on heuristic mutation is of higher convergence speed and better average performance than six other evolutionary algorithms based on probability distribution.