Improved Owl Search Algorithm

Owl search algorithm(OSA)is a population intelligence optimization algorithm proposed in recent years.Due to its simple structure,the advantages of easy to achieve are favored by many researchers,but it is still difficult to avoid the problems of early maturation convergence,easy to fall into local optimization,and imbalance between search and mining due to the optimization mechanism.Therefore,in order to improve the performance of OSA,improve its convergence performance,and balance the global search and local mining capabilities of OSA,this paper makes the following work:(1)In order to improve the performance of OSA,an owl search algorithm based on global optimal population perturbation is proposed(GPOSA),and a dynamic tail-end elimination strategy is first introduced to enhance the accuracy of the initial solution.Secondly,a group-based perturbation strategy is proposed,which improves the information exchange between OSA and other individuals and optimal individuals by constructing an optimal population containing 1/3 of the global optimal solution and 2/3 of the individuals in the population.The introduction of standard Brownian motion with choice into OSA balances the learning communication between the individual himself,the global optimal individual,and other individuals.Comparative experiments on CEC2017 test functions by GPOSA with 9 other well-known algorithms have verified its effectiveness.(2)In order to overcome the shortcomings of OSA convergence accuracy and poor local mining capacity,a hybridizing manta ray foraging optimization and society owl search algorithm is proposed(HMRSOSA).Firstly,a social learning strategy is proposed,which strengthens the information exchange between OSA and global optimal and neighborhood individuals,and in addition,random factors with pseudo-levy flight are introduced to adjust the learning pace.The OSA under this strategy is called the Social Learning Owl(SOSA).The feasibility of the strategy is verified by comparison with OSA on the CEC2017 test function.Second,the use of a half-uniform initialization strategy avoids the phenomenon that randomly generated individuals are too dense.The global search capability of the algorithm is improved by combining the control factors count and scount with the manta ray foraging algorithm(MRFO).Finally,the comparison experiment with HMARSOSA on 48 classical benchmark functions using 8 algorithms shows that HMRSOSA has higher convergence accuracy and convergence speed,and has better generalization ability than OSA.