摘要
随着物联网的进步,时间序列数据得以大量采集,对时间序列数据进行准确预测和可靠检测异常变得越来越重要。针对线性预测方法无法提取多维时间序列特征的缺点,基于自注意力机制能同时提取不同序列之间特征相关性的机制,提出了基于自注意力机制的时间序列线性预测方法。在线性预测模型中引入自注意力机制,可更准确地提取多维时间序列数据中的关键信息,提高预测的准确度,实现异常检测。从工程和算法的角度进行优化,对比了引入自注意力机制前后线性预测方法的性能。实验结果表明,该方法在SMD数据集和MSL/SMAP数据集上取得了更好的预测性能和异常检测准确度,明显提高了准确性和鲁棒性,有助于工控条件下的状态预测和异常检测。
With the progress of the Internet of Things(IoT),time series data can be collected in large quantities,and it is becoming more and more important to accurately predict time series data and reliably detect anomalies.This paper proposes a linear time series prediction method based on self-attention mechanism,which can simultaneously extract the inter-sequence feature correlations.The introduction of self-attention mechanism in linear prediction model facilitates accurate extraction of key information from multi-dimensional time series data,and hence,improving prediction accuracy and realizing anomaly detection.The performance of linear prediction method before and after the introduction of self-attention mechanism was experimentally compared from the perspective of engineering and algorithm,and the modified method achieved better prediction performance and anomaly detection accuracy on SMD data set and MSL/SMAP data set,exhibited significantly improved accuracy and robustness,and thereby was potentially promising in state prediction and anomaly detection under industrial control conditions.
作者
王汝桥
张谊
何玉鹏
周岱
WANG Ruqiao;ZHANG Yi;HE Yupeng;ZHOU Dai(Science and Technology on Reactor System Design Technology Laboratory Nuclear Power Institute of China,Chengdu 610213,China)
出处
《电工技术》
2024年第19期55-57,63,共4页
Electric Engineering
关键词
时间序列
异常检测
自注意力机制
线性预测
time series
anomaly detection
self-attention mechanism
linear prediction
