Abstract: Dynamic search fireworks algorithm (dynFWA) which adopts a dynamic explosion amplitude for core fireworks (CF) has been proved to be a great algorithm for solving optimization problems. However, dynFWA has the disadvantages that it is easy to fall into local optimal solutions prematurely and has slow convergence rate. In order to improve the above mentioned problems, this paper improves conventional dynFWA by embedding two different learning factors which make use of the history successful information, referred as improved dynamic search firework algorithm (IdynFWA). The learning factors take advantages of the information of the best firework in each generation, which makes the fireworks have the ability to learn from the excellent search. Moreover, two different generations of learning factor are beneficial to balance the local search and global search ability. The improved algorithm has been tested on  28 benchmark functions of CEC2013. And the experimental results show that IdynFWA significantly outperforms dynFWA, and achieves better performance than both SPSO2011 and DE/rand-to-best/1.

Key words: dynamic search fireworks algorithm, explosion amplitude, mutation operator, learning factor

摘要： 采用核心烟花动态爆炸半径策略的动态搜索烟花算法（dynamic search fireworks algorithm，dynFWA）已被证明是解决优化问题的一个重要算法。然而，dynFWA的寻优精度低且容易过早地陷入局部最优解。为了改善上述的缺陷，通过嵌入一种利用历史成功信息生成两种不同的学习因子来改进传统的动态搜索烟花算法，称为改进的动态搜索烟花算法（improved dynFWA，IdynFWA）。算法中的学习因子充分利用搜索过程中每一代最好的烟花个体信息，使得烟花具有向群体的优良搜索信息学习的能力，并且它的两种不同产生方式有助于平衡算法的局部搜索和全局搜索能力。改进后的算法在CEC2013的28个Benchmark函数上进行测试，实验结果表明IdynFWA的寻优效果明显优于dynFWA，并且比粒子群算法SPSO2011和差分演化算法DE/rand-to-best/1能达到更好的寻优性能。

关键词: 动态搜索烟花算法, 爆炸半径, 变异算子, 学习因子