基于自编码的长流程造纸过程断纸故障识别

Paper Break Fault Recognition in Long Process Papermaking Process Based on Autoencoder

关键词:  造纸  特征提取  堆栈自编码器  故障识别

Key Words:papermaking  feature extraction  stacked autoencoder  fault recognition

摘要:断纸故障造成的非预期停机对生产效率影响较大，预测和识别断纸故障对造纸工业具有重要意义。本研究提出了一种基于堆栈自编码器（stacked autoencoder，SAE）和Softmax分类器的断纸故障分类和识别方法，通过相关性分析与机理分析相结合的方式筛选出关键过程变量，并采用聚类方法对断纸故障进行类型标记。建立并训练SAE模型以提取数据中的深度特征，并基于提取后的特征，可通过Softmax分类器识别断纸故障。以造纸厂历史数据进行验证，得到该模型的断纸故障分类准确率为96.2%，精准率为93.1%，召回率为91.9%，结果优于其他模型，说明基于SAE特征提取的分类模型可以较好地实现断纸故障识别。

Abstract:Paper breaks leading to unanticipated downtime can significantly impact the production efficiency， the prediction and diagnosis of paper break is rather important for the papermaking industry. In this study， a method of paper break fault classification and identification based on stacked autoencoder （SAE） and Softmax classifier was proposed to select the key process variables by combination of correlation analysis and mechanism analysis， and adopt clustering method to label the type of paper breaks. SAE model was established and trained to extract deep features from the data， and paper break faults were identified based on the extracted features using a Softmax classifier. Validated using historical data from a paper mill， and the model results were better than those of other machine learning models， with achieving the paper break classification accuracy of 96.2%， the precision of 93.1%， and the recall rate of 91.9%， which indicated that the classification model based on SAE feature extraction could effectively identify paper break faults.