Improved Artificial Bee Colony Algorithms Based On Multi-elitist Nelder-mead Simplex Method And Double-populations Comprehensive Learning

The Artificial Bee Colony(ABC) algorithm is a swarm intelligence random search optimization method, which has the characteristics of simple structure, few parameters, easy to combine with other al-gorithms. However, ABC has the weaknesses of mature easily, local search ability is weak, search for the best precision lowly when solv-ing unconstrained optimization problems. Here, we mainly focuses on the weak search capability and how to balance the exploitation and detection ability. Two improved artificial bee colony algorithms are proposed in this paper.Algorithm 1 is the improved artificial bee colony algorithm based on multi-elitist Nelder-Mead simplex method and directional strate-gy( MENM-DS-ABC) . Multi-elitist artificial bee colony (PS-MEABC)algorithm for real-parameter optimization use the global best solu-tion and an elitist randomly selected from the elitist set to enhance the exploitation. Algorithm 1 introduces directional select strategy,and modify the formula of selected probability of onlooker bees which based on objective function value ranking. As the same time, a sim-plex method is used on elitist solution set to balance the exploration and exploitation ability of the algorithm. The numerical experiment results show that the proposed algorithm has higher searching preci-sion and faster convergence speed. Moreover, the accuracy of solution is more stable when solving high-dimensional space optimization.Algorithm 2 is an improved double-populations artificial bee colony algorithm based on heterogeneous comprehensive learning(DPCLABC).In this algorithm, the swarm is divided into exploration-subpopulation named group 1 and exploitation-subpopulation named group 2. Illu-minated by particle swarm optimization(PSO), the food source will be updated on all dimensions rather than on a randomly selected di-mension. Meanwhile, in order to enhance the search ability of food sources, comprehensive learning strategy is used to generate the ex-emplars for two sub-populations. In addition, opposition based learn-ing is used to improve the quality of initial swarm, and multiplicative weight update method is used to update the selection probability of onlooker bees in each sub-population. To evaluate the performance of the improved multi-populations artificial bee colony algorithm,we implement numerical experiments based on 18 unimodal, mul-timodal, and rotated benchmark functions. Computational results demonstrate that the new algorithm can balance the exploration and the exploitation ability, it can prevent premature convergence and produce competitive optimization precision and convergence speed.