摘要
单一的支持向量机在建模时存在一定的局限性,对于复杂的实际数据,不能很好地提取其中的信息,导致模型泛化性能较差,为此提出基于K近邻的组合支持向量机方法。该方法首先采用简单距离分类方法对经过主元分析的样本数据进行分类,并采用K近邻算法得到支持向量机子模型的组合参数,进而建立起基于支持向量机的多模型。将该方法应用于双酚A生产过程中质量指标的软测量建模,仿真结果表明基于K近邻方法的支持向量机多模型建模可以有效提高模型的泛化性能,并验证了该算法的可行性和有效性。
For complicated practical data, the single Support Vector Machine has some limitation that is the poor generalization performance in soft sensor modeling. In the article, a simple vector distance classification is used to classify the data which was processed by the principle component analysis (PCA) and an improved K-Nearest Neighbor (KNN) algorithm is used to find the compositional parameters for Support Vector Machine sub-models. For the data picked from a petrochemical company, the simulation result shows that the approach developed here can effectively improve the generalization performance.
出处
《江南大学学报(自然科学版)》
CAS
2010年第1期7-10,共4页
Joural of Jiangnan University (Natural Science Edition)
基金
国家自然科学基金项目(60674092)
江苏省高技术研究(工业)项目(BG2006010)
关键词
K近邻
支持向量机
软测量
K-nearest neighbor, support vector machine, soft sensor