论文部分内容阅读
针对复杂工业过程中的多操作、非高斯等问题,提出了一种基于局部组标准化的多模型动态主元分析(LGS-MMDPCA)在线监测新方法。由于过程中的随机操作造成每个批次数据呈非高斯分布,因此不能用整个批次数据统一建模,而应根据操作,将同一操作下的数据单独建立模型,从而构成多模型结构。然而单一操作下数据量较少,往往不符合统计建模的数量要求,故将相似操作数据融合在一起,为使其符合高斯分布建立多元统计模型,提出了一种局部组标准化(LGS)方法。同时为提高模型精度,采用多模型动态主元分析(MMDPCA)建模。最后以精炼炉炼钢过程为例验证了新方法的有效性。
Aiming at the multi-operation and non-Gaussian problems in complex industrial processes, a new method based on local group normalization for multi-model dynamic principal component analysis (LGS-MMDPCA) online monitoring is proposed. Due to the non-Gaussian distribution of each lot data due to random operation in the process, the entire batch data can not be modeled uniformly. Instead, the data under the same operation should be separately modeled according to the operation to form a multi-model structure. However, with a small amount of data under a single operation, it often does not meet the quantitative requirements of statistical modeling. Therefore, similar operation data are merged together. In order to establish a multivariate statistical model according to the Gaussian distribution, a local group normalization (LGS) method . At the same time, in order to improve the accuracy of the model, multi-model dynamic principal component analysis (MMDPCA) modeling was used. Finally, the process of refining furnace steelmaking as an example to verify the effectiveness of the new method.