Elite Representative Based Individual Adaptive Regeneration Framework for Differential Evolution

Differential evolution (DE) is a popular optimization algorithm which attracts numerous research studies and is applied widely. To address the problem of population stagnation in DE, this paper proposes an elite representative based individual adaptive regeneration framework (EIR) that can be incorporated into any DE variant easily. EIR is able to renovate stagnated individuals to regions centering around the weighted value of the individuals themselves and the elite representatives based on the Gaussian distribution with distinct crossover rates. Elite representatives are sampled from dynamically varied elite swarms for different individuals. A less greedy substitution mechanism considering the proximity of fitness values is introduced to avoid exclusion of some valuable information from inferior solutions. The whole regeneration and substitution mechanism is dynamically adapted according to the evolutionary process and the fitness ranking status of current individuals to balance between exploitation and exploration capability. Experimental results on CEC 2014 benchmark functions of $30D$ , $50D$ , $100D$ and two real world optimization problems show that the proposed EIR framework can significantly improve the performance of two standard DEs and six state-of-the-art DE algorithms.