变异萤火虫优化的粗糙K-均值聚类算法

A rough K-means clustering algorithm optimized by mutation firefly algorithm

粗糙K-means聚类算法存在如下不足:随机选取初始聚类中心会导致算法过早收敛容易陷入局部最优,质心更新公式中的权重和决定对象划分的阈值采用定值有时会导致聚类结果波动较大和精度下降。针对以上问题,引入一种变异策略和差分进化的萤火虫算法,从3个方面进行优化:构造新的目标函数,以目标函数值作为萤火虫光亮强度进行初始聚类中心点的搜索,把萤火虫算法求得的最优解作为算法的聚类中心进行聚类迭代;以下近似集和边界集中对象数量的变化以及对象分布的差异性动态调整质心权重;给出一种通过迭代次数自动获取阈值的方法。试验结果表明,改进后的算法减少了迭代次数,聚类结果稳定性好,准确率更高,改善了算法对随机初始中心点的敏感和稳定性不足等问题。

The rough K-means clustering algorithm had the following shortcomings:Selecting the initial clustering center randomly would lead to premature convergence of the algorithm and it was easy to fall into local optimal,the weight in centroid updating for-mula and the threshold that determined the object division were fixed,which sometimes lead to the large fluctuation of clustering re-sults and the decrease of accuracy.Aiming at the above problems,a mutation strategy and differential evolution algorithm of firefly were introduced to optimize the algorithm from three aspects:A new objective function was constructed,the value of the objective function was used as the brightness intensity of firefly to search the initial clustering center,and the optimal solution of firefly algo-rithm was used as the clustering center of the algorithm for clustering iteration;Changes in the number of objects in the following ap-proximate set and boundary set and differences in object distribution dynamically adjust the weight of the centroid;A method to au-tomatically obtain threshold value through the number of iterations was presented.The experimental results show that the improved algorithm reduced the number of iterations,the clustering results had good stability and higher accuracy,and this algorithm improved the problem that the random initial center point was sensitive and the stability was insufficient.