Improving the selection of differential evolution through a quartile-based ranked operator

The Differential Evolution (DE) algorithm is a well-founded and widely popular technique in the field of evolutionary computation. The standard DE algorithm is composed of four stages: initialization, mutation, crossover, and selection. The efficiency of its structure and operators have led to numerous modifications since its inception in 1995. However, the majority of these modifications have focused on the mutation and crossover stages. This paper presents an improved selection mechanism based on the classification of the population using quartiles. This preserves the classical structure of the DE while optimizing its effectiveness and efficiency. The proposed approach is designated as Quartile-based Ranked Differential Evolution (QRDE). It introduces a selection that categorizes the population into four groups to facilitate a comparison of the trial vector against a randomly selected individual from the worst group. Also, a scaling factor defines the exploration and exploitation of the QRDE from the selection part. To demonstrate the efficacy of the QRDE, its performance was evaluated on the set of benchmark functions from the CEC-2017 in 30 and 50 dimensions. The tests were conducted by comparing the QRDE against 9 state-of-the-art algorithms, including DE variants and recent optimization methodologies. The results of statistical tests indicate that the proposed QRDE is more computationally complex than the original DE. However, it outperforms the majority of algorithms in terms of stability and accuracy. Similarly, the QRDE was tested in 5 engineering problems, where the results demonstrate that the QRDE is highly competitive, outperforming other techniques in these domains.