基于ICA-SVDD非高斯过程监控研究.pdfVIP

2008 年中国过程系统工程年会会议论文集 基于 ICA-SVDD 的非高斯过程监控研究 王树青，谢磊 （先进控制研究所，工业控制技术国家重点实验室，浙江大学杭州，310027 ） 摘要：提出基于 ICA-SVDD 的统计过程监控和故障检测方法，避免了传统统计过程监控方法假设数据服从正态分布 的不足。针对传统的FastICA 算法容易陷入局部极小值的不足，结合微粒群算法提出非高斯投影算法，保证提取的 独立成分非高斯性最大化，并给出了非高斯成分数目的选择准则。获得过程非高斯独立成分之后，利用 SVDD 来描 述其分布情况，构造新的统计量确定其统计控制限。通过对数值仿真及工业应用研究表明，该方法能够及时地发现 过程中出现的异常情况。 关键词： 独立成分分析；支持向量描述过程监控 中图分类号：TP13 文献标识码：A 文章编号：0438-1157 （2006 ）00-0000-00 Non-gaussian process monitoring based on ICA-SVDD Shuqing Wang, Lei Xie （Institute of Advanced Process Control, National Key Laboratory of Industrial Control Technology, Zhejiang University, Hangzhou 310027, China ） Abstract ：A novel ICA-SVDD statistical process monitoring approach is proposed to address non-gaussian multivariate process systems. Such systems present typical challenges to traditional multivariate process monitoring approaches. The PSO based ICA approach is proposed to extract the most non-guassian component out of the process records and overcomes the deficiency of traditional FASTICA algorithm, which is easily trapped in local minimum. Criteria to automatically select the number of non-gaussian components are also given. Transforming the exacted non-gaussian components into a feature space using a support vector data description then allows the application of a simple parametric statistical inference. Application in the numerical study and industrial melting processes shows the efficiency of the proposed method. Key words ：independent component analysis ； support vector data description ；process monitoring 引 言 随着现代工业及科学技术的迅速发展，现代 义独立成分变量的监控统计量并确定其控制 化的流程工业系统不断朝着大规模、复杂化、集 限。SVDD 由Tax 提出(Tax an