基于正交方程近似的多元轨迹同步化过程监控

The Process Monitoring Based on Orthonormal Function Approximation for Synchronization of Multivariate Traiectories

在间歇过程中，多向独立元分析方法（MICA）能够对不同批次的非正态分布的各变量三维数据列出多元统计模型，从建立的统计模型中提取出互为独立的信息元，经过计算发现其中的隐含的故障信息，从而改善过程监控能力。然而，事实上，在某些反应时间比较长的间歇过程中，由于无法保证各批次的反应速率都一致，这样的异步导致无法准确建立MICA模型。正交函数近似（OFA）方法，将各批次轨迹变量经过正交变换为按每一变量都相同个数的一系列系数的二维模型，并将其应用于监控中。用青霉素的发酵模拟软件（Pensimv2．0）进行仿真，通过某些批次设置故障与计算两种指标SPE和I^2比对，结果证明，与未同步的简单MICA建模相比，基于OFA的MICA过程监控能相对准确地发现过程中的异常。

On batch process, Multiway Independent Component Analysis can construct multivariate statistical model from three-dimensional non-gaussian distribution data of different batches to extract the independent information components from the model to find out latent faulty information after projection calculation so as to improve the ability of process monitoring. However, in fact, in most cases, some asynchronous batches have no same duration so that the MICA model cannot be built correctly as imagined. The approach of Orthonormal Function Approximation （OFA） changes all the variables of reference trajectories of batches in turn into a series of coefficients with same numbers to form two-dimensional model for monitoring. Alter having setting several faulty batches, the Penicillin fermentation process simulator （Pensim V2.0） is experimented and calculate indices of SPE and I^2 to show that the MICA monitoring based on OFA could detect the abnormal of process easier than the one of the MICA monitoring on the original unsynchronized model.