Best-worst individuals driven multiple-layered differential evolution

Conventional differential evolution (DE) algorithms have been widely used for optimisation problems but suffer from low performance and premature convergence. Hence, researchers have proposed advanced variants to enhance performance using information and strategies. However, the performance of the variants remains limited because they only utilise limited information of individuals. A more suitable search orientation for the algorithm is required to effectively leverage individual information and enhance the processing of mid-population data. This study presents a novel best-worst individual-driven multiple-layered differential evolution (BWDE) algorithm. A best-worst individual-driven mechanism is designed that leverages various pieces of individual information to overcome local optima or stagnation, facilitating escape from the current search space and maintaining group fitness levels. In addition, the five-layer structure of the BWDE algorithm allows for the adequate use of multiple layers of information to determine the evolutionary direction of a population. Consequently, a balance is achieved between population development and exploration at distinct evolutionary stages. Extensive experiments are conducted using the Congress on Evolutionary Computation (CEC) 2017 and 2011 standard benchmark functions to evaluate the effectiveness of the proposed algorithm. The results are compared with those of classical algorithms, a winning algorithm at a CEC competition, and state-of-the-art DE variants. The experimental results demonstrate that the proposed BWDE algorithm outperforms its competitors and achieves more competitive results.