论文部分内容阅读
多向主元分析(MPCA)的统计监控模型,因为易受建模数据中离群点的影响,还需预估新批次未反应完的数据,所以提出一种新的间歇过程鲁棒在线监控法。先利用改进尺度的CDC/MVT算法获取常规建模的批次数据;再用多模型非线性结构代替传统的MPCA单模型线性化结构,并提出确定时滞变量的算法。前者用于监控β-甘露聚糖酶发酵批过程,并与移动窗多向主元分析(MWMPCA)法相比,即使建模数据中存在离群点,前者仍能获得正确的监控结果,减少建模时对数据的要求;同时克服了MPCA不能处理实时性的问题,避免了MPCA在线应用时预测值的误差;更能精确描述过程的故障,准确性和实时性良好。
Due to the influence of outlier in multi-direction principal component analysis (MPCA) and the influence of outliers in the modeling data, the data of unreacted batch in new batch need to be estimated. Therefore, a new batch process robust online Monitoring method. Firstly, the CDC / MVT algorithm with improved scale was used to get the batch data of conventional modeling. The multi-model nonlinear structure was used instead of the traditional MPCA single-model linearization structure and the algorithm to determine the time-lag variables was proposed. The former is used to monitor β-mannanase fermentation batch process. Compared with the MWMPCA method, the former can obtain the correct monitoring result even if there is outlier in the modeling data, Time-to-date data requirements; at the same time to overcome the MPCA can not handle real-time issues, to avoid the MPCA online application of the predicted error; more accurately describe the process of failure, accuracy and real-time good.