A differential evolution algorithm with dual preferred learning mutation

Abstract

Differential evolution (DE) algorithm is widely used for solving real-parameter optimization problems due to its simplicity and efficiency. However the canonical DE is easy to suffer from the premature convergence. To further improve the performance of the DE, a differential evolution algorithm with dual preferred learning mutation (DPLDE) is proposed. Dual preferred learning mutation simultaneously learns behaviors from the individual with better fitness(BFI) and individual with better diversity(BDI). The learning factor of BFI is self-adaptively and independently adjusted for each individual. The learning factor of BDI is adaptively adjusted at each generation. A total of 26 Benchmark test functions with different characteristics are used for performance comparative experiments. The results show that DPLDE is superior to the eight state-of-the-art improved algorithms in terms of the convergence precision, convergence speed and stability. For the high-dimensional functions, with the same-scale population and maximum number of evolution generations, DPLDE can still get the excellent global optimization performance and has a more prominent advantage.