摘要
针对目前卷烟厂制丝生产设备异常均是事后维修或基于时间定期维修的问题,提出了基于MSCRED(Multi-Scale Convolutional Recurrent Encoder-Decoder,多尺度卷积递归编码解码器)算法的卷烟厂制丝设备异常检测方法。通过设备5个控制回路的传感器数据构建向量相似度对角分块特征矩阵,使用MSCRED方法训练异常得分,通过给定阈值进行比较来判断此时是否异常,还可通过重构特征矩阵将异常分解,精准定位异常部位。实验结果表明,该方法的短中长三个时间序列均可较好地检测出异常点,但只有短时序列可以更精准地定位到异常部位。
To address the issue of post-maintenance or time-based regular maintenance for abnormalities detection in cigarette manufacturing equipment,this paper proposes an abnormal detection method based on the MSCRED(Multi-Scale Convolutional Recurrent Encoder-Decoder)algorithm for cigarette factories.The sensor data from five control loops in the equipment is utilized to construct a vector similarity diagonal block feature matrix,and the MSCRED method is employed to train an abnormal score.By comparing this score with a predefined threshold,it can be determined whether there is any abnormality at a given time.Mean while,anomalies can also be decomposed by reconstructing the feature matrix to precisely locate the ab normal parts.The experiment results show that this method can detect anomalies well in all three time series,including short,medium,and long,but only short-term sequences can more accurately locate abnormal parts.
作者
刘彦辰
朱静
曹海涛
LIU Yan-chen;ZHU Jing;CAO Hai-tao(Computer and Information Engineering College,Xinjiang Agricultural University,Urumqi 830052,China)
出处
《信息技术》
2025年第12期94-101,107,共9页
Information Technology
