【摘要】 基于多元统计分析的过程监测需要大量的历史数据来进行离线建模，数据中存在的异常值会影响建模的准确性，从而降低过程监测的性能，甚至产生误报和漏报。针对数据中含有异常值的问题，提出一种基于人工神经网络(artificial neural network, ANN)和典型相关分析(canonical correlation analysis, CCA)的鲁棒过程监测方法。该方法分为离线训练和在线监测2个阶段。离线训练阶段：根据异常值服从亚高斯分布的原理，将扩展后的峭度函数作为ANN的损失函数进行训练，消除异常值的影响。在线监测阶段：首先根据异常值和故障连续发生的概率不同来剔除异常值，然后利用CCA进行过程监测。为了验证所提方法的有效性，使用数值算例和连续搅拌釜式反应器的仿真案例进行仿真验证。结果表明，所提方法可以有效消除异常值的影响，提高故障检测率，并降低误报率。更多还原

【Abstract】 Process monitoring based on multivariate statistical analysis requires large amounts of historical data for modeling. The presence of outliers in the data can affect the accuracy of the modeling, thus reducing process monitoring performance, even increasing the false alarm rate and the missing alarm rate. A robust process monitoring method based on artificial neural network(ANN) and canonical correlation analysis(CCA) are proposed to deal with the outlier issue in the data. The method is divided into two phases: offline training and online monitoring. In the offline training phase, the expanded Kurtosis is chosen as the loss function of ANN according to the principle that the outliers obey the sub-Gaussian distribution to eliminate the influence of outliers; in the online monitoring phase, outliers are first rejected based on the difference in probability of consecutive occurrence of outliers and faults, and then process monitoring is performed using CCA. To verify the effectiveness of the proposed method, simulation tests are performed on a numerical example and continuous stirred-tank reactor. The results show that the proposed method can effectively eliminate the effect of outliers, improve the fault detection rate, and reduce the false alarm rate.更多还原