基于LSNPE算法的化工过程故障检测

复杂化工过程通常具有多个操作模态，而且采集的数据不服从单一的高斯或非高斯分布。针对化工过程的多模态和复杂数据分布问题，将局部标准化（local standardized，LS）策略应用于邻域保持嵌入（neighborhood preserving embedding，NPE）算法，提出了一种新的基于局部标准化邻域保持嵌入（local standardized neighborhood preserving embedding，LSNPE）算法的故障检测方法。首先，使用LSNPE算法提取高维数据的低维子流形，进行维数约减，同时保持邻域结构不变。其次，通过特征空间中样本的局部离群因子（local outlier factor，LOF）构造监控统计量并确定其控制限。相较于监控多模态化工过程的多模型策略，提出的LSNPE方法不需要过程先验知识的支持，只需建立一个全局的监控模型。最后，通过数值仿真及Tennessee Eastman（TE）过程仿真研究验证了本文提出方法的有效性。

Fault detection for chemical process based on LSNPE method

Complex chemical processes often have multiple operating modes and the within-mode process data do not follow Gaussian or non-Gaussian distributions. To handle the problem of multiple operating modes and complex data distribution, a novel fault detection method, local standardized neighborhood preserving embedding (LSNPE) was proposed by applying local standardization (LS) strategy to the neighborhood preserving embedding (NPE) algorithm. Firstly, LSNPE algorithm was performed for dimensionality reduction and thus the main features of the collected data were extracted. At the same time, it could keep the neighborhood structure unchanged. Next, a monitoring statistics was established using the local outlier factor (LOF) of each sample in feature space and its control limit was determined. Instead of building multiple monitoring models for complex chemical process with different operating modes, the proposed LSNPE method built only one global model to monitor a multi-mode process without the support of any prior process knowledge. Finally, the feasibility and efficiency of the proposed method were illustrated through a numerical example and the Tennessee Eastman process.