基于混合分块DMICA-PCA的全流程过程监控方法  

Plant-wide process monitoring based on mixed multiblock DMICA-PCA

作　　者：江伟[1] 王振雷[1] 王昕[2] 

机构地区：[1]化学工程联合国家重点实验室,华东理工大学化工过程先进控制和优化技术教育部重点实验室,上海200237 [2]上海交通大学电工与电子技术中心,上海200240

出　　处：《化工学报》2017年第2期759-766,共8页CIESC Journal

基　　金：国家自然科学基金重点项目(61134007);国家自然科学基金青年项目(61403141);上海市“科技创新行动计划”研发平台建设项目(13DZ2295300);上海市自然科学基金项目(14ZR1421800);流程工业综合自动化国家重点实验室开放课题基金资助项目(PAL-N201404)~~

摘　　要：分块策略被广泛运用于全流程过程监控领域,以解决全流程过程变量关系复杂性较高的问题,但传统的分块策略与子块建模方法都未考虑过程的动态性问题,并且传统的分块策略都片面依赖于过程知识或过程数据信息,影响了过程监控的效果,为此提出了一种基于混合分块DMICA-PCA的过程监控方法。在分析过程的动态性后,先利用已知的部分过程知识进行变量的初步分块,接着利用各分块变量之间改进的广义Dice’s系数(MGDC)进行进一步的分块。然后采用DMICA-PCA方法对每个子块进行建模得到子块的统计量,并通过加权方法得到总的联合指标进行故障检测。同时对每个子块采用改进的故障诊断方法,提高了诊断效果。最后将该方法应用在TE过程的过程监控中,证明了该方法的有效性。Multiblock strategy is widely used in plant-wide process monitoring to solve problems with complicated relationships between process variables. Traditional multiblock strategies and sub-block modeling methods are not effective in plant-wide process monitoring, because dynamic characteristics of the process have not been considered and knowledge or data information of the process is exclusively exploited. A mixed multiblock DMICA-PCA method was proposed to improve process monitoring performance. First, variables were sliced into initial sub-blocks by obtained process knowledge after analysis of process dynamics and further sliced into final sub-blocks by modified general Dice＇s coefficient（MGDC） between variables of initial sub-blocks. Then, the DMICA-PCA method was used to establish model and acquire statistical values of variables in final sub-blocks and a combined overall index from weighted sum was developed for fault detection, which improved performances by simultaneous diagnosis on each sub-block. Effectiveness of the proposed method was validated on monitoring the Tennessee-Eastman（TE） process.

关 键 词：主元分析 过程控制 过程系统 混合分块 全流程 改进的广义Dice’s系数 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]