改进变分自编码器的工业时序数据异常检测
Anomaly detection of industrial time series data based on variational autoencoder
张志昂 1廖光忠2
作者信息
- 1. 武汉科技大学计算机科学与技术学院,湖北武汉 430065
- 2. 武汉科技大学智能信息处理与实时工业系统湖北省重点实验室,湖北武汉 430065
- 折叠
摘要
为解决传统的异常检测模型对工业时序数据异常点检测方面误判率大和抗干扰性差的问题,提出一种改进的变分自编码器模型.考虑到工业时序数据的不规律性,使用变分自编码器模型作为基础架构;由于变分自编码器本身存在难以准确检测出异常时序数据的问题,在编码和解码过程中分别引入时间卷积网络和通道注意力机制,实现扩大感受野和增强特征权重;对数据时序数据使用随机森林进行特征排序,提高检测的准确性.通过进行对比测试实验,验证了该模型可以有效提高对异常工业时序数据点检测的准确性和可靠性.
Abstract
To solve the problems of high misjudgment rate and poor anti-interference of the traditional anomaly detection model for industrial time series data,an improved variational autoencoder model was proposed.The variational autoencoder model was used as the infrastructure considering the irregularity of industrial time series data.Because the variational autoencoder itself is difficult to accurately detect abnormal time series data,temporal convolution network and channel attention mechanism were respectively introduced in the encoding and decoding process to expand receptive field and enhance feature weight.The random forest was used for feature ranking of data time series data to improve the accuracy of detection.The model can effectively im-prove the accuracy and reliability of abnormal industrial time series data point detection through comparative test experiments.
关键词
异常检测/时间卷积网络/变分自编码器/通道注意力机制/时序数据/随机森林/感受野Key words
anomaly detection/temporal convolutional network/variational autoencoder/channel attentional mechanism/time series data/random forest/receptive field引用本文复制引用
出版年
2024
NETL
NSTL
万方数据