基于改进HTM模型的时间序列异常检测

ANOMALY DETECTION OF TIME SERIES BASED ON IMPROVED HTM MODEL

基于时间序列异常检测在航天试验和航天测控领域应用广泛的现实背景,提出一种改进的层级时间记忆(HTM)实时异常检测模型。在层级时间记忆模型的基础上,通过引入滑动窗口和β分布,对该模型输出的预测偏差进行相关处理,实现对原异常检测模型的优化改进。经过改进,该模型可以对异常下降进行有效的识别判断,也消除了在前期学习过程中出现的误警现象,性能上有一定的提升。

Based on the practical background of the wide application of time series anomaly detection in space test and space TT&C,we propose an improved real-time anomaly detection model with hierarchical temporal memory.Based on the hierarchical temporal memory model,the sliding window and beta distribution were introduced to deal with the prediction deviation of the output of the model,thus the optimization and improvement of the original anomaly detection model was realized.After improvement,our model can effectively identify and judge the abnormal decline,and also eliminates the false alarm in the early learning process of the model,which improves the performance of the improved real-time anomaly detection model to a certain extent.