Differential Evolution Through Variable Neighborhood Search for Constrained Real-Parameter Optimization Problems

This paper presents a differential evolution algorithm with the variable neighborhood search (VNS), called DEVNS. In the VNS loop, two distinct mutation strategies are used to solve real-parameter constrained optimization problems. As well-known, the performances of DE algorithms depend on the chosen mutation strategies and their control parameters. For these reasons, the proposed DEVNS generates each trial individual through the use of VNS local search having two distinct mutation strategies as well as random crossover and mutation rates. The idea behind the DEVNS is to generate multiple candidate solutions among which the better one is chosen as the trial individual. Furthermore, the DEVNS is well equipped with constraint handling methods to end up with feasible solutions. The algorithm was tested using benchmark instances in Congress on Evolutionary Computation 2017 [1]. To the best of our knowledge, the DEVNS algorithm was able to generate new best-known solutions up to 8 to 10 benchmark functions for the first time in this paper in the literature.