中文摘要：

在传统的K-means聚类算法中，初始聚类是随机选取的，其聚类结果易随着不同的初始聚类中心波动。针对这一问题，首先采用最大距离积法对传统K-means聚类算法的初始聚类中心进行了优化。同时定义了一种新的目标函数并将其引用到传统的K-means聚类算法中，以实现对聚类结构类别数K的优化选择。将训练集样本数据经上述方法聚类后，再将各个子类分别建立基于支持向量机的子模型，通过开关切换的方式连接各子模型得到组合的支持向量机模型。将该方法应用于双酚A生产过程的缩合反应单元溶解罐出口苯酚含量的软测量建模。工业实例仿真结果表明：该算法能较好地跟踪苯酚含量的变化趋势，有效地改善了数据分类效果，提高了软测量模型的估计精度，显示了它在工业领域的应用潜力。

英文摘要：

Clustering results of traditional K-means clustering algorithm easily fluctuates with random initializing cluster centers. The maximum distances product algorithm is used to optimize the initial clustering centers and a new target function is also defined so as to be introduced into the K-means clustering algorithm. The category number of cluster structural is optimally selected by this new func tion. The clustering accuracy is improved with the method and then a combination model based on support vector machine is estab lished. The method is applied to a soft sensor modeling for the quality index in a Bisphenol A production process. The simulation result shows that the the change trendof phenol content is tracked effectively and data classification result is improved by the algorithm. It also shows that the estimation precision of the soft sensor model is improved which demonstrates the potential application in industry field.