基于改进的OFA-MPCA的监控方法

多向主元分析(MPCA)是利用多变量统计方法从纷杂的海量数据信息中提取出能够准确表征数据信息的几个主元,并通过投影法来降低数据的维数,主要应用于间歇生产过程中. 在实际的间歇生产过程中,由于各种原因导致各批次异步造成它们运行时间的不一致,而无法直接建立有效的统计模型,正交函数近似(OFA)是一种基于正交基的投影变换技术,通过对原始数据进行OFA处理后,可以用投影系数来描述原始数据所具有的特征,并且可以达到轨迹同步化和压缩数据量的目的. 对OFA法进行了部分改进,并结合MPCA法对典型的间歇过程——青霉素发酵过程进行了仿真研究. 结果表明,改进的OFA计算速度有了极大的提高,且改进的OFA-MPCA法能完好地对各批次进行同步、建模并得出准确的监视结果.

Monitoring based on improved OFA-MPCA

Multiway Principal Component Analysis (MPCA) is a multivariable statistical approach, which can extract several principal components from the numerous of data to express the data information well, and is mainly used in batch process. In practice, for many reasons, the runtime of each batch is different from others so that the effective statistical model can not be built directly. Orthonormal Function Approximation (OFA) is a technique of project transformation based on orthonormal base, after OFA we can use the projection coefficient to express the characteristics of the original data and synchronize the trajectories of each historical batch and reduce the dimension. This paper presents some improvement on the OFA and combined the MPCA to model and monitor the typical batch process——Penicillin fermentation process. The simulation results show that the improved OFA can deal with data more quickly and the improved OFA-MPCA is able to synchronize the trajectories of all the batches, and monitor the batches perfectly.