| 摘 要: 针对工业物联网(IIoT)系统中,状态量和模拟量时间序列之间存在的时延异常问题,提出了一种基于时间序列相关性和长短期记忆(LSTM)的时延异常检测方法,即 CLDAD。使用 LSTM 来学习状态量和模拟量时间序列之间存在相关性,把时间序列中的非正常时间延迟看成 LSTM 异常情况。通过 LSTM 支持的相关性,结合异常分数来设置动态阈值,构建状态量和模拟量时间序列的时延异常检测方法。实验结果表明,该方法在时延异常检测中的平均精确度为95.43%,F1值高达0.94,优于目前常用的工业异常检测模型,并验证了所提方法的有效性。 |
| 关键词: 工业物联网 状态量时间序列 非正常时间延迟 长短期记忆 异常检测 |
|
中图分类号: TP391.4
文献标识码: A
|
| 基金项目: 国家自然科学基金项目资助(62192751,61425027) |
|
| Time Series Delay Anomaly Detection Based on Correlation and LongShort-Term Memory |
|
FENG Jiahao1, PANG Aiping1, YANG Wen2
|
(1.College of Electrical Engineering, Guizhou University, Guiyang 550025, China;
2.Xichang Satellite Launch Center, Xichang 615000, China)
fengjh2000@126.com; appang@gzu.edu.cn; whutyw@126.com
|
| Abstract: To address the problem of delay anomalies between state quantity and analog quantity time series in Industrial Internet of Things (IIoT) systems, a delay anomaly detection method based on time series correlation and Long Shor-t Term Memory (LSTM), named CLDAD, is proposed. This method utilizes LSTM to learn the correlation that exist between state quantity and analog quantity time series, and regards abnormal time delays in time series as LSTM detection anomalies. By leveraging the correlations supported by LSTM and combining them with anomaly scores to set dynamic thresholds, a delay anomaly detection approach for state and analog quantity time series is constructed. Experimental results show that the average precision of this method in delay anomaly detection is 95.43% ,with an F1-score as high as 0.94, outperforming commonly used industrial anomaly detection models and verifying the effectiveness of the proposed method. |
| Keywords: Industrial Internet of Things state quantity time series abnormal time delay Long Shor-t Term Memory anomaly detection |
