基于群密度的改进果蝇优化算法及在异常检测中的应用

发布时间：2018-01-16 08:06

本文关键词：基于群密度的改进果蝇优化算法及在异常检测中的应用　出处：《工程科学与技术》2017年05期 　论文类型：期刊论文

【摘要】：针对传统果蝇算法面临的收敛稳定性差、难以协调全局搜索及局部搜索能力等缺点,提出一种基于群密度的改进果蝇优化算法。首先,借鉴现有算法的优势,将果蝇种群分为搜索果蝇和跟随果蝇,并分别使用两类果蝇进行全局化搜索与局部精细化搜索。然后,为提高算法全局搜索的稳定性,在每次迭代过程中使用基于最优区间回避的分区采样策略更新搜索果蝇的位置;该策略在每次迭代过程中获得表现最优的若干只果蝇以构造最优果蝇组,根据最优果蝇组中果蝇个体在每个维度上的取值范围确定最优区间,并通过对最优区间外的其他区间分区采样以确定搜索果蝇的新位置。最后,为协调算法的全局搜索能力与局部搜索能力,引入群密度的概念,通过计算果蝇群密度并结合相关阈值实现不同种群规模的动态调整。针对典型测试函数的实验结果表明,基于最优区间回避的分区采样策略相对于传统随机函数具有更强的全局优化性能。与传统优化算法相比,本文算法在保证收敛速度的同时获得了较高的寻优精度及稳定性,在综合性能上得到明显提升。在KDDcup99数据集上的异常检测仿真实验结果表明,本文基于分区采样及群密度的果蝇优化算法能有效避免局部最优,在获取异常检测分类器的重要参数最佳取值方面起到一定作用。
[Abstract]:Aiming at the shortcomings of the traditional Drosophila algorithm, such as poor convergence stability, difficult to coordinate global search and local search ability, an improved Drosophila optimization algorithm based on group density is proposed. Firstly, the advantages of the existing algorithms are used for reference. The fruit fly population is divided into search fruit fly and follow fruit fly, and two kinds of fruit fly are used for global search and local refinement search respectively. Then, to improve the stability of the algorithm global search. During each iteration, the location of the search fruit fly is updated by using the partition sampling strategy based on optimal interval avoidance. In this strategy, several flies with the best performance were obtained during each iteration to construct the optimal Drosophila group, and the optimal interval was determined according to the value range of the individual in each dimension in the optimal Drosophila group. The new location of the search fruit fly is determined by sampling other regions outside the optimal interval. Finally, the concept of group density is introduced to coordinate the global search ability and the local search ability of the algorithm. By calculating the population density of Drosophila and combining with the correlation threshold, the dynamic adjustment of different population size is realized. The experimental results of typical test function show that. Compared with the traditional random function, the partition sampling strategy based on optimal interval avoidance has better global optimization performance than the traditional optimization algorithm. In this paper, the convergence speed is guaranteed and the optimization accuracy and stability are obtained, and the performance of the algorithm is improved obviously. The simulation results of anomaly detection on KDDcup99 data set show that the proposed algorithm has good performance. In this paper, the optimal algorithm based on partition sampling and group density can effectively avoid local optimization and play a certain role in obtaining the best value of important parameters of anomaly detection classifier.
【作者单位】：中央财经大学信息学院;天津财经大学理工学院;
【基金】：北京市自然科学基金资助项目(4174105) 中央财经大学学科建设基金资助项目(2016XX02) 国家自然科学基金重点支持项目 NSFC-浙江两化融合联合基金项目资助(U1509214)
【分类号】：TP18
【正文快照】： 启发式优化算法具有智能性、并行性、渐进性、随机性等基本特点,近年来引起了众多学者的广泛研究,目前已成功应用于科学、工程和金融等领域。然而,传统启发式优化算法(如粒子群算法[1]、遗传算法[2]、调和搜索算法[3]等)在参数依赖性、计算复杂性、收敛速度、优化精度等方面面

【相似文献】