摘要
针对工业生产过程非高斯分布的特点,并为解决因子分析(FA)监控方法假设生产过程高斯分布以及独立元分析(ICA)监控方法假设模型为无噪模型等问题,本文结合FA、ICA等多元统计方法的优点,提出1种基于独立因子分析(IFA)的过程监控新方法。此监控方法运用EM算法求解参数,建立数据的非高斯分布模型,构造GI^2、GSPE 2种监控指标,通过非参数全局估计算法计算控制限,并将生产过程采集数据实时输入监控系统,以判断有无故障发生。将此方法应用于化工吸附分离过程,通过监控图可以发现,IFA可以及时对故障发生予以报警,而ICA的I^2监控指标甚至无法给出故障报警,同时IFA的漏检率、误报率分别较FA降低了40%～60%,以上试验结果均验证了本方法的有效性及优越性。
Aimed at the non-Gaussian distributions characteristic of the process,this paper proposes a new process monitoring method based on Independent Factor Analysis（IFA）which integrates the advantage of all kinds of Multivariate Statistical Analysis such as Factor Analysis（FA）and Independent Component Analysis（ICA）to solve the problem that FA supposes the process is Gaussian distributions and ICA supposes the process data is non-noisy.The new process monitoring method computes the parameters by the expectation maximization algorithm,establishes a mixture model of adaptive Gaussian,constructs the monitoring indices including GI^2 and GSPE and presents the non-parametric algorithm for confidence bounds and inputs the process data into the monitoring system to determine whether the faults happen or not on time.The monitoring method based on Independent Factor Analysis is applied in the chemical separation process and the faults are detected on time by monitoring chart,however the monitoring indices of I^2 the ICA is failed.At the same time,false alarm rate and missed detection rate of the IFA decreases 40%-60%compared with the FA,the validity and superiority are proved.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2010年第10期1353-1356,共4页
Computers and Applied Chemistry
基金
国家自然科学基金资助项目(60904045)
江苏省基础研究计划(自然科学基金)(BK2009068)
江苏省六大人才高峰项目.
关键词
独立因子分析
混合高斯模型
EM算法
非参数控制限
过程监控
independent factor analysis
mixture gaussian models
expectation maximization algorithm
non-parametric confidence bounds
process monitoring
