Survival exploration strategies for Harris Hawks Optimizer

摘要

This paper proposes new versions of Harris Hawks Optimizer (HHO) incorporated the survival-of-the-fittest principle of evolutionary algorithms. HHO is the recent swarm-based optimization algorithm imitating the surprise pounce behaviour of Harris’ hawks chasing style. HHO can show different patterns of the exploration and exploitation. It has a simple and time-varying structure, which further assist a smooth transition between the core phases. It has two main phases to iterate toward the optimal solution: exploration and exploitation. In the exploration phase, the current solution is either randomly modified based on any solution selected randomly or rebuilt from scratch. In evolutionary algorithms, selecting any solution from swarm basically relies on the natural selection principle of the survival-of-the-fittest to accelerate convergence. To make use of such principle, three selection strategies (i.e., tournament, proportional and linear rank-based methods) are employed in the exploration phase of HHO and introduces three new versions, which are Tournament HHO (THHO), Proportional HHO (PHHO), and Linear-Rank HHO (LHHO). In order to evaluate the performance of the proposed HHO versions, 23 well-regarded benchmark functions with various sizes and complexities are utilized as well as three real-world engineering problems. The sensitivity of proposed HHO versions to their parameter settings are studied and analyzed. Thereafter, a scalability study is conducted to show the effect of the population dimensions on the proposed HHO versions. Comparative evaluation shows that THHO version has superiority over other proposed HHO versions. Furthermore, the proposed HHO versions show enhanced trade off between the exploratory and exploitative trends and a better local optima avoidance. They are able to produce viable results competitively comparable with other eleven state-of-the-art methods using the same benchmark functions. Interestingly, the proposed variants of HHO are able to yield new results for some benchmark functions. Furthermore, three real-world engineering optimization problem of IEEE CEC2011 are also used in the evaluation process. Again, the proposed variants of HHO are able to achieve the best results. The information, guides and supplementary accessible files for this research will be publicly available at https://aliasgharheidari.com.