Time sequence anomaly detection method based on normalized mutual-information estimation

Abstract

The invention relates to a time sequence anomaly detection method based on normalized mutual-information estimation, and belongs to the technical fields of time sequence anomaly detection, informationtheory and data mining. The method comprises: A, carrying out data preprocessing to obtain sample point sets corresponding to time sequence sampling segments; B, carrying out mutual-information estimation on sample point sets, which correspond to every two adjacent sampling segments, on the basis of an extreme learning machine; C, using maximum entropy to normalize obtained mutual information; and D, repeating the step B and the step C to obtain a normalized mutual-information sequence, and determining a location of sequence mutation through comparison with a threshold value. The method describes an algorithm which does not need parameter optimization and training, and the algorithm uses the extreme learning machine for estimation of the mutual information, uses randomly generated parameter setting, reduces execution time, and ensures execution efficiency of an algorithm model; and at the same time, the maximum entropy is used to normalize the mutual information obtained by estimation, and accuracy of anomaly detection is guaranteed.

Description

technical field [0001] The invention relates to a time series anomaly detection method based on normalized mutual information estimation, and belongs to the technical fields of time series anomaly detection, information theory and data mining. Background technique [0002] The research of time series has been paid more and more attention in recent years, and it is widely used in the fields of clinical medicine, military affairs, geological exploration and network security. The abnormal time segments of time series have the characteristics of low frequency of occurrence, significant differences in patterns or statistical characteristics compared with normal states, etc., often contain important information, and have research significance and value. [0003] The classic time series research methods are based on time-frequency signal detection methods, such as autocorrelation method and circular correlation method. These methods can describe the characteristics of continuous si...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to propose a time series based on normalized mutual information estimation for the technical defects of the existing time series anomaly detection algorithm, which adopts the cross-validation method to select the optimal parameters, which leads to time-consuming and high calculation costs. Anomaly Detection Algorithm

Images