改进的堆优化算法及其宫颈细胞数据聚类优化

常见的聚类方法存在对初始点敏感和易陷入局部最优的不足，为此提出了一种改进HBO的聚类方法。首先，提出一种改进的HBO，即扰动替换的HBO（disturbance and replacement HBO，DRHBO）克服其不足，即采用一种随机维度值替换策略和高斯扰动机制用于HBO中最优个体的状态更新，解决HBO搜索效率低的问题，提出一种正弦差分扰动策略，以突破当前个体仅与直接领导和同事进行交流的限制，从而增强搜索能力。将随机维度值替换和随机差分扰动策略融合，用于HBO中前期个体状态更新以避免其产生无效解。其次，提出一种DRHBO聚类方法，并运用到宫颈细胞数据集上以获得更好的聚类效果。大量、不同类别和不同样本的宫颈细胞数据集实验结果表明，与HBO及其改进算法和其他最先进算法相比，DRHBO的优化性能更好、稳定性更强且效率更高。DRHBO聚类方法更适应于宫颈细胞数据集。

Improved heap-based optimizer and its application to cervical cell data clustering optimization

In view of the easy entrapment into local optima and sensitiveness to initial point of conventional clustering methods, this paper proposed an improved heap based optimizer(HBO) clustering method. Firstly, this paper presented an improved HBO, namely DRHBO. It used a random dimensional value replacement and Gaussian disturbance strategy to update the state of the best agent to solve the defects such as low efficiency of HBO. It utilized a sine differential disturbance to update a random agent'' state and that breaks through the shackle of the individual''s communication only with its direct leader and colleagues, to improve the search ability. It integrated the random dimensional value replacement and differential disturbance strategies to update the states of the agents in the initial stage of HBO to avoid generating inefficient solutions. Secondly, this paper presented a DRHBO clustering method and applied it to cervical cell data to get better effects. Lots of experimental results on cervical cell data sets with diverse types and different sample numbers show that compared with HBO, its variants and other state-of-the-art algorithms, DRHBO can get better performance, stronger stability and higher efficiency. DRHBO clustering is more suitable to cervical cell data.