基于二维卷积神经网络高层数据特征学习的过程故障检测

Fault Detection Based on High Level Data Feature Learning of Two Dimensional Convolutional Neural Network

针对化工生产过程中高维数据故障特征难以学习和提取的缺点,提出一种基于二维卷积神经网络的化工过程故障检测方法.首先,采集化工过程不同故障的数据构成训练集和测试集;然后,对训练集和测试集中对应的正常样本和故障样本标注标签;最后,将训练集中的样本数据作为卷积神经网络的输入来训练、优化模型.方法应用于田纳西-伊斯曼化工过程,数据结果表明:二维卷积神经网络能够提取出原始数据中样本与样本、变量与变量之间更为抽象的高层数据特征,通过特征提取和学习后的重构特征数据输入到全连接层BP神经网络进行故障分类,比单独使用全连接BP神经网络的检测率提高了14.42%,误报率降低了2.55%.

In order to solve the problems of low accuracy,high false alarm rate and heavy computation in the traditional machine learning algorithm in chemical process fault detection,a chemical process fault detection method based on two-dimensional convolutional neural network. Firstly,the data of different faults in chemical process were collected to form the training set and test set. Then,the corresponding normal samples and fault samples in the training and test set were labeled. Finally,the sample data in the training set was used as the input of the convolutional neural network to train and optimize the model. The method applied to Tennessee-Eastman chemical process data shows that the two-dimensional convolutional neural network can extract more abstract high-level data features between samples and variables in the original data,through feature extraction and learning. The reconstructed feature data is input into the fully connected layer BP neural network for fault classification,which is 14. 42% higher than the fully connected BP neural network alone,and the false positive rate is reduced by 2. 55%.