Résumé: Metaheuristics are intelligent optimization techniques that lead the searching procedure through utilizing exploitation and exploration. Increasing the number of hard problems with big data sets has encouraged researchers to implement novel metaheuristics and hybrid the existing ones to improve their performance. Hence, in this work, a novel metaheuristic called group learning algorithm is proposed. The main inspiration of the algorithm emerged from the way individuals inside a group affect each other, and the effect of group leader on group members. The two main steps of optimization, exploration and exploitation are outlined through integrating the behaviors of group members and the group leader to complete the assigned task. The proposed work is evaluated against a number of benchmarks. The produced results of classical benchmarks are compared against PSO, GWO, TLBO, BA, ALO, and SSA. In general, compared to other participated algorithms, out of 19 classical benchmarks, the proposed work showed better results in 11. However, the second best algorithm which is SSA performed better in 4 out of 19 benchmarks. To further evaluate the ability of the algorithm to optimize large scale optimization problems CEC-C06 2019 benchmarks are utilized. In comparison to other participated algorithms, the proposed work produced better results in most of the cases. Additionally, the statistical tests confirmed the significance of the produced results. The results are evidence that the proposed algorithm has the ability to optimize various types of problems including large scale optimization problems.

Langue: Anglais