Differential evolution algorithm with local and global parameter adaptation

Abstract

Differential Evolution (DE) is an effective meta-heuristic algorithm for numerical optimization. However, it suffers from persistent limitations such as sensitivity to parameter settings and premature convergence tendencies. This paper presents a novel Local and Global Parameter Adaptation (LGP) mechanism to mitigate these deficiencies through two key innovations. First, we develop a dual historical memory strategy that dynamically classifies successful control parameters into local or global historical record based on the Euclidean distance between parent-offspring vector pairs, the local and global historical memory are updated accordingly at each generation. Second, we introduce a parameter adaptation strategy that adaptively selects elements from appropriate historical memory for the generation of new control parameters to maintain exploitation-exploration balance. Extensive experimental validation demonstrates LGP’s effectiveness. When integrated with four DE variants, LGP consistently improves their performance, and the LGP-enhanced algorithm demonstrates remarkable performance compared with seven State-of-the-Art DE algorithms. Results confirm that LGP improves solution accuracy and prevents premature convergence simultaneously.