Towards realistic mimicking of grey wolves hunting process for bounded single objective optimization

Abstract

Canis lupus (grey) wolves hunt for their prey in a pack. There exist evolutionary algorithms based on the hunting pattern of Canis lupus wolves known as Grey wolf optimizer (GWO). It is a powerful optimizer and had produced competitive results for many difficult problems. There already exist several variants of GWO because of its simplicity and exploitative qualities. GWO has been proved to be a very good exploiter. However, the modeling of grey wolves concentrates more on exploitation rather than exploration and thus results in the problem of local stagnation. To enhance the exploration capabilities, dynamic behavior is added to prey. It was assumed that the prey was static in the original GWO wherein the prey always tries to run away from the predator. The proposed algorithm enhances diversity by varying the position of prey with the help of Levy distribution rather than considering it to be static. This paper aims at introducing a novel, realistic version of GWO wherein the new population is generated with the help of Levy flight distribution of prey as well as with the different class of wolves by a suitable modification in the hierarchy of wolves. Also, the three primary strides of chasing, looking, circling, and assaulting of prey, are executed. However, the new population which is created has a better exploration because of levy flight distribution of prey position. The algorithm is tested on a well-known suite of 23 benchmark problems. The outcomes show that the proposed algorithm, GWOLF can give either better or competitive results as contrasted with the original GWO and PSO metaheuristics.