Research And Application Of Time Series Anomaly Detection Algorithm

Time series refers to data sequences that are chronological,time-varying,and interrelated and widely exist in various fields.Anomaly detection has always been an important issue in different areas of research and applications.Therefore,with the development of the Internet and the rise of artificial intelligence,the anomalous detection of time series has gradually attracted scholars at home and abroad.The anomaly detection of time series is generally divided into point abnormalities and pattern abnormalities.If a single data instance can be seen as an exception related to other data,then this instance is called a point exception.This is the simplest type of anomaly and is also the focus of most studies on anomaly detection.Sometimes the time series is more concerned with whether the process that occurred within a certain period of time is abnormal,then this is called a pattern exception.The research and concern of this paper is the problem of abnormal patterns.This article has carried on the thorough discussion and the research to the sequence anomaly of the time sequence,mainly revolves around the time series representation method and the anomaly detection algorithm to carry on the correlation research.The research content of this article is as follows.1)Propose the multi-dimensional symbolization of time series to represent the mSAX method.The method first uses a fixed length to divide the time series into inseparable sub-sequence fragments;then,according to the feature description of the subsequence fragments,it transforms the feature vectors;and then uses the discretization method to symbolize each dimension of each sub-sequence feature vector.,Generate multidimensional symbol vectors.Eventually the time series is transformed into an atomic sequence represented by a multidimensional symbol vector.2)An anomaly detection algorithm based on a single segment pattern is proposed.The algorithm first expresses the original time series PLR piecewise linearly and treats each single line segment as a pattern;then it maps the two end points of each single line segment pattern to the applied mSAX symbolic representation method to divide the subsequence fragments.Segmentation point,which takes the subsequence between the segmentation points in the symbolized time sequence as the object of excavation;finally calculates the nearest-neighbor non-self-matching distance of each mining object,and takes this distance as the degree of abnormality,and also proposes an The method of automatically selecting the abnormality threshold value is used to determine the abnormal sequence by comparing the size of the abnormality and the abnormality threshold.3)An anomaly detection algorithm based on adjacent segment patterns is proposed.The algorithm is based on a single segment pattern anomaly detection algorithm and extends a single segment to adjacent segments.Using the overlapped sliding window traversal method,the combination of adjacent line segments is regarded as a complex pattern,and the split points at both ends of the pattern are mapped into atomic sequences.Each segmentation of the atomic sequence is used as a mining object.Other processes are similar to the anomaly detection algorithm based on the single line segment mode.Finally,the effectiveness of the time-series anomaly detection algorithm based on the single line segment pattern and the adjacent line segment pattern is verified on the ECG data set.