基于DNPE-SVDD的化工过程监控  

Chemical Process Monitoring Based on DNPE-SVDD

作　　者：韩晓春 薄翠梅 易辉 

机构地区：[1]南京工业大学电气工程与控制科学学院,江苏南京211816

出　　处：《系统仿真学报》2018年第1期184-190,共7页Journal of System Simulation

基　　金：国家自然科学基金(61203020;61503181);江苏省自然科学基金(BK20140953);江苏省高校自然科学基金(13KJB510013)

摘　　要：针对化工过程中检测数据变量维数高、非线性与动态特性相结合的特点,而传统的线性降维算法不能提取局部结构信息和动态特性,提出了基于动态邻域保持嵌入–支持向量数据描述(DNPE-SVDD)算法的化工过程监控模型。结合DNPE在非线性降维和SVDD在异常点检测的优势,使用DNPE算法进行维数约减,对降维后的流形空间采用SVDD算法建立监控模型,通过Tennessee Eastman(TE)化工过程进行仿真研究,同时与DPCA、DNPE算法对比验证所提算法的性能,结果表明DNPE-SVDD能获得更高的故障检测准确率。Chemical processes test dataset are high dimensional, and have the combined feature of nonlinear and dynamic characteristics. However, traditional linear dimension reduction algorithm cannot extract the local structure information and dynamic characteristics. The monitoring model of chemical process based on dynamic neighborhood preserving embedding-support vector data description （DNPE-SVDD） algorithm is proposed. With the superiority of NPE in nonlinear dimensionality reduction and SVDD in the detection of outliers, dimension is reduced by DNPE algorithm and the monitoring model of the manifold space with reduced dimension is established by SVDD algorithm. The Tennessee Eastman （TE） process is simulated using the proposed model. Compared with DPCA and DNPE algorithms, the simulation results show that the DNPE-SVDD has a higher accuracy of fault detection.

关 键 词：邻域保持嵌入 支持向量数据描述 数据降维 过程监控 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]