Research And Application Of A Wave Vector Classification Technique Based On Multiple Change Points Detection

With the rapid development of the national economy and the increasing advancement of science and technology,all walks of life are paying more and more attention to data.Mass production,transaction,medical and other data are collected and used for research in many aspects such as enterprise product upgrades or service transformation.In the face of massive data with high complexity,large volume,and variability,how to accurately detect and quickly analyze it,and to dig out the potential value from it,has become a hot research issue in today's digital age.As an important research branch in the field of big data,the anomaly detection and rapid analysis technology of time series data mainly uses the distribution status of the overall data to find the location of mutations in the time series data,analyze and study the mutation information,in the field of data mining has extensive research and application value.In order to achieve accurate analysis and rapid diagnosis of abnormal state of large-scale time series data,this paper uses the method of combining sliding window and TSTKS algorithm to give a multiple change points detection model;constructs the fluctuation vector corresponding to the multiple mutation points of the time series data by extracting the fluctuation characteristics of the data in multiple time series windows,and then proposes a multi-threshold segmentation algorithm for the fluctuation vector for data abnormal state detection;uses simulated and actual lesion data to experimentally verify the threshold segmentation algorithm to realize the detection and rapid analysis of different lesion states of pathological data.First,based on the sliding window and TSTKS algorithm,a multiple change point detection model is given.The sliding window model is used to divide the data to be detected into several data sub-segment windows,the TSTKS change points detection algorithm is used to detect the time series data,and the time series data fluctuation vector characterizing the degree of data fluctuation is constructed by extracting the fluctuation amount of continuous multiple windows.Secondly,using the multi-mutation point detection model,two multi-threshold segmentation strategies for volatility vectors are proposed.The multi-threshold segmentation method based on fluctuation amount can divide the data into abnormal state and normal state when the change point detection accuracy is good,and the multi-threshold segmentation strategy based on the mean value can complete the state of the data when the change point detection fails.Finally,combining the different characteristics of the two strategies,an improved volatility vector multi-threshold segmentation algorithm is proposed.Experiments show that the algorithm has a good segmentation effect for time series data with a single data distribution and data characteristics.In addition,in order to deal with the problem of complex and changeable data state division,an adaptive multi-threshold segmentation algorithm is presented.The multi-threshold segmentation algorithm based on volatility and the multi-threshold segmentation algorithm based on mean value combine the advantages of the two strategies,and an adaptive multi-threshold segmentation algorithm is proposed.The experimental results show that the adaptive multi-threshold segmentation algorithm can make up for the shortcomings of a single strategy,and can automatically select different thresholds according to the different state distributions and data characteristics of the data.Finally,this paper uses real EEG signal data to verify the adaptive multi-threshold segmentation algorithm.The algorithm can divide brain epilepsy and other lesion data into multiple different states,and realize the analysis of different lesion states such as seizure warning and severity in the epileptic lesion cycle.Furthermore,an in-depth analysis of different time series data signal sources of the same patient was carried out,and a preliminary analysis was carried out on the duration and severity of epileptic seizures based on the detection results of different lesion states.