Abstract: Flower pollinate algorithm (FPA) is a novel swarm intelligence optimization algorithm which is proposed recently, and it has been widely researched and used because of its advantages of solving the balance problem of local search and global search, and having less parameters, being implemented easily and so on. However, there are less current researches on the parameter, and the speed of convergence in the later stage is slow, what is more, it is easy to fall into local optimizations, which incline to restrict the application of the FPA. In order to improve the overall performance of the FPA, firstly, this paper modifies the scaling factor of the control step size. Secondly, this paper proposes a hybrid algorithm GMPFPA (flower pollination algorithm combination with Gauss mutation and Powell search method). The Gauss mutation is utilized to perturb the global search of the GMPFPA, which enhances the diversity of population of the GMPFPA, and improves the global detection ability of the GMPFPA, and then the strong local search capability of the Powell search method is introduced to enhance the local development ability of the hybrid algorithm. Through the comparison experiment of 12 high dimensional classical test functions, the effectiveness and superiority of the improved algorithm are verified.

Key words: Gauss mutation, flower pollination algorithm (FPA), Powell method, optimum value, optimization ability

摘要： 花朵授粉算法（flower pollination algorithm，FPA）是最近提出的一种新型群智能优化算法，由于其较好地解决了全局搜索和局部搜索的平衡性问题，且具有参数少，易实现等特点，已得到广泛应用和研究，但现有研究对其参数的研究较少，同时该算法也存在演化后期收敛速度慢且易陷入局部极小等缺陷，使其应用范围受到制约。为了提升FPA算法的整体性能，对其控制步长的缩放因子的取值进行了修正；提出了把高斯变异和Powell法融入到花朵授粉算法中的混合算法GMPFPA（flower pollination algorithm combination with Gauss mutation and Powell search method）。改进算法首先利用高斯变异对全局搜索进行扰动，增强种群的多样性，提高全局探测能力，然后引入局部寻优能力强大的Powell法提升其局部开发能力。通过12个高维经典测试函数对比实验，验证了改进算法的有效性和优越性。

关键词: 高斯变异, 花朵授粉算法（FPA）, Powell法, 最优值, 寻优能力