Considering the fact that in a multi-model soft sensing modeling, traditional clustering method can' t reflect real category properties of the new data and outlier because of their none-fully-considered characteristics, and this leads to a model with low precision. A clustering algorithm based on minimum loop energy was proposed to translate the clustering into looking for a minimum energy ring with the simulated annealing algorithm, and to have the clustering realized by the minimum energy ring through all sample points. According to their power and their own properties, the category of outlier and new test data were determined and the classification accuracy was improved; and through making use of SVM, a regression sub-model for each subclass was established and the soft sensor which combined model was obtained. Both applying this method to soft sensing modeling of the quality performance in Bisphenol A production and having it simulated proves its sound effectiveness.%在多模型软测量建模中,对于新的数据以及异常样本点,传统的聚类方法没有充分考虑它们的特性,因而所属类别往往不能反映其真实属性,最终导致模型精度不高.为此,提出一种基于最小环路能量聚类的算法,该方法将样本聚类转化为寻找一个最小能量环问题,通过模拟退火算法搜索一条经过所有样本点的最小能量环实现样本集的聚类；对侦破出的异常样本点和新的测试数据根据其能量值确定其所属属性,从而提高聚类和分类精度；然后利用支持向量机为各个子类建立回归子模型,得到软测量组合模型.将该方法应用于双酚A生产过程质量指标的软测量建模中,仿真结果验证了该方法的有效性.
展开▼
