一种新的带障碍约束的混合模糊聚类

Fuzzy Clustering with Obstacles Distance Based on Quantum-behaved Particle Swarm Optimization

为解决二维平面上存在障碍约束的聚类问题,在量子粒子群聚类算法的基础上,提出一种新的带障碍约束的模糊聚类算法,该算法引入隶属度的概念,提出了粒子逃逸原则以避免聚类中心点陷入障碍物中,采用绕过障碍物距离的新定义函数extdistance(),重新定义数据点绕过障碍的聚类目标函数,替代了模糊C-均值算法的基于梯度下降的迭代过程,在很大程度上克服了FCM算法易陷入局部极小值和对初值敏感的缺陷。

In order to solve the clustering problem of obstacles exist in two-dimensions ,a new fuzzy cluste-ring algorithm with obstructed constraints was proposed based on OPSO algorithm .It adopted the member-ship grade in the object function of QPSO , applied the Escaping Principle to avoid the updated cluster center particles sinking into the area of the obstacles ,redefined the Clustering objective function of data points bypassing obstacles instead of FCM iterative process based on gradient descent ,and it overcame the problems of FCM algorithm which is apt to fall into local extremum and be sensitive to initial parameters .