多变量数据驱动的化工过程质量相关故障监测
Multi-variate data-driven monitoring method for quality-related faults in complex chemical processes
以多变量数据驱动为导向,分别对卷积神经网络(CNN)及交叉熵损失函数(CEL)进行改进优化,构建适用于复杂化工过程质量相关故障的监测模型——二维卷积神经网络(2DCNN)及基于类别加权的交叉熵损失函数(WCEL)。该方法能够将多变量数据转化为若干样本矩阵,并以此作为2DCNN模型的输入,分别有效地捕捉矩阵行数和列数所表征的时空维度特征,从而实现高精准的质量相关故障监测;同时,嵌入损失函数——WCEL,自适应地动态调整2DCNN模型的学习率,从而解决故障类别分配不均衡问题。
Based on multi-variate data,a monitoring method for quality-related faults in complex chemical processes is proposed.The convolution neural network (CNN) and cross-entropy loss (CEL) function are optimized to establish the two-dimensional convolution neural network (2DCNN) and the weighted cross-entropy loss (WCEL) function,respectively.By this method,multi-variate data can be converted into several sample matrixes,which are further used as inputs for a 2DCNN model that can capture effectively the spatiotemporal characteristics represented by the number of rows and columns in the matrix,thereby monitoring precisely quality-related faults.In addition,WCEL function is embedded to adjust the learning rate of 2DCNN model adaptively and dynamically,thereby solving the uneven problem in allocation of fault categories.
质量相关故障 / 基于类别加权的交叉熵损失函数 / 二维卷积神经网络 / 多变量数据 / 化工过程
quality-related fault / weighted cross-entropy loss function / two-dimensional convolution neural network / multi-variate data / chemical process
| [1] |
秦艳, 徐一凡, 杨燕霞, |
| [2] |
崔劲松, 贾波, 李学盛, |
| [3] |
李鑫妮, 王亚君, 许晓婷. 基于AOA优化SVM的工业过程故障检测[J]. 现代化工, 2024, 44(S2):343-347. |
| [4] |
程福松, 侯磊, 周亚云. 石油化工生产区域多米诺效应定量风险分析[J]. 油气储运, 2016, 35(5):457-464. |
| [5] |
宋晓云, 田文德, 靳满满. 非稳态过程故障诊断方法研究[J]. 现代化工, 2016, 36(8):182-185. |
| [6] |
陈文静, 代长春, 党亚固, |
| [7] |
张祥, 崔哲, 董玉玺, |
| [8] |
刘子健, 贾旭清, 张士发, |
| [9] |
杨余, 杨鑫, 王英, |
| [10] |
肖飞扬, 顾幸生. 基于并行LSTM-CNN的化工过程故障检测[J]. 华东理工大学学报:自然科学版, 2023, 49(3):382-390. |
| [11] |
张继旺, 王雪莉, 谢海博, |
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
|
| [16] |
|
国家重点研发计划资助项目(2021YFB3801304)
/
| 〈 |
|
〉 |