Success of metaheuristic algorithms depends on the efficient balance between of exploration and exploitation phases. Any optimization algorithm requires a combination of diverse exploration and proper exploitation to avoid local optima. This paper proposes a new improved version of the Woodpecker Mating Algorithm (WMA), based on opposition-based learning, known as the OWMA aiming to develop exploration and exploitation capacities and establish a simultaneous balance between these two phases. This improvement consists of three major mechanisms, the first of which is the new Distance Opposition-based Learning (DOBL) mechanism for improving exploration, diversity, and convergence. The second mechanism is the allocation of local memory of personal experiences of search agents for developing the exploitation capacity. The third mechanism is the use of a self-regulatory and dynamic method for setting the Hα parameter to improve the Running Away function (RA) performance. The ability of the proposed algorithm to solve 23 benchmark mathematical functions was evaluated and compared to that of a series of the latest and most popular metaheuristic methods reviewed in the research literature. The proposed algorithm is also used as a Multi-Layer Perceptron (MLP) neural network trainer to solve the classification problem on four biomedical datasets and three function approximation datasets. In addition, the OWMA algorithm was evaluated in five optimization problems constrained by the real world. The simulation results proved the superior and promising performance of the proposed algorithm in the majority of evaluations. The results prove the superiority and promising performance of the proposed algorithm in solving very complicated optimization problems.

No abstract

Hybrid metaheuristic algorithms have recently become an interesting topic in solving optimization problems. The woodpecker mating algorithm (WMA) and the sine cosine algorithm (SCA) have been integrated in this paper to propose a hybrid metaheuristic algorithm for solving optimization problems called HSCWMA. Despite the high capacity of the WMA algorithm for exploration, this algorithm needs to augment exploitation especially in initial iterations. Also, the sine and cosine relations used in the SCA provide the good exploitation for this algorithm, but SCA suffers the lack of an efficient process for the implementation of effective exploration. In HSCWMA, the modified mathematical search functions of SCA by Levy flight mechanism is applied to update the female woodpeckers in WMA. Moreover, the local search memory is used for all search elements in the proposed hybrid algorithm. The goal of proposing the HSCWMA is to use exploration capability of WMA and Levy flight, utilize exploitation susceptibility of the SCA and the local search memory, for developing exploration and exploitation qualification, and providing the dynamic balance between these two phases. For efficiency evaluation, the proposed algorithm is tested on 28 mathematical benchmark functions. The HSCWMA algorithm has been compared with a series of the most recent and popular metaheuristic algorithms and it outperforms them for solving nonconvex, inseparable, and highly complex optimization problems. The proposed algorithm is also used as a Multi-Layer Perceptron (MLP) neural network trainer to solve the software development effort estimation (SDEE) problem on three real-world datasets. The simulation results proved the superior and promising performance of the HSCWMA algorithm in the majority of evaluations.

No abstract