Abstract:

Intermittent process monitoring is of great importance to ensure the stable operation during the process running. However, it is difficult for traditional process monitoring methods to extract the structure-related and nonlinear dynamic time-varying features of batch process data. To solve this problem, this paper proposes a canonical correlation analysis method combining graph sample aggregate network and long-short term memory network (GSA-LSTM) for batch process monitoring. First, the K-nearest neighbor method is utilized to convert batch data into graph-structured form, and graph sample aggregate network (GraphSAGE) is used to extract the structure features of process data. Then, long-short term memory (LSTM) is used to extract the nonlinear time-varying feature of data at the same time. By integrating them with structure features through weight coefficients, the more representative batch process data features are obtained. After that, canonical correlation analysis method is utilized to carry out process monitoring. Finally, the proposed method is applied to numerical examples and injection molding process monitoring, and the results are analyzed to verify the effectiveness of the proposed method.

Key words: process monitoring, intermittent, graph neural network, long-short term memory, canonical correlation analysis, algorithm, process systems

摘要：

间歇过程监测对于保证批次生产过程的稳定运行具有重要意义。传统过程监测方法难以提取间歇过程数据特有的非线性结构和动态时变特征。为此，提出了一种融合图采样聚合网络和长短期记忆网络(GSA-LSTM)的典型相关分析方法用于间歇过程在线监测。首先，利用K近邻方法将批次过程数据转化为图结构形式，利用图采样聚合网络(GraphSAGE)提取数据内部的结构特征，然后利用长短期记忆网络(LSTM)提取数据的非线性动态特征，通过权重系数将结构特征和动态特征融合得到更具有代表性的间歇过程数据特征。进一步地，利用典型相关分析方法对残差建立监测模型。最后将所提方法应用于数值例子和注塑过程监测，结果分析验证了所提方法的有效性。

关键词: 过程监测, 间歇性, 图神经网络, 长短期记忆网络, 典型相关分析, 算法, 过程系统