Two-level parameter cooperation-based population regeneration framework for differential evolution

Abstract

Differential evolution (DE) has shown excellent performance in dealing with optimization problems in continuous space, but it still faces the problems of population stagnation and premature convergence. To address these difficulties and enhance the performance of existing DE algorithms, this paper proposes a two-level parameter cooperation-based population regeneration (TPPR) framework that can easily be integrated with various DE algorithms without increasing the time consumption. In the TPPR framework, whether an individual executes the regeneration operation depends on a specific cooperation strategy for a macroparameter and a microparameter, and the regeneration operation updates the positions of the selected individuals using a dynamic coordination strategy for the Cauchy and Gaussian distributions, that are constructed using those two parameters. Comparative experiments of ten representative DE algorithms with and without the TPPR framework on thirty functions with different dimensions from the IEEE CEC 2014 test platform and four real-life problems are conducted to verify the effectiveness of the TPPR framework. The comparative results based on the improvement rate and deterioration rate indicate that using the TPPR framework, the performance of the adopted DE algorithms can be significantly improved with a very high probability, and the possibility of deterioration is very low.