A Multi-feature Fusion Detection Mechanism And Template Matching Analysis Method For Abnormal State Of Time Series Data

With the continuous development of network technology,both life and production are more and more closely connected with the network.The resulting data has the characteristics of strong immediacy and large amount.The mutation point detection technology is one of the anomaly detection technologies of time series data,which can effectively detect the abnormal part of the data segment,while the off-line mutation point detection technology can no longer meet the requirements of online detection.The mutation point detection technology combined with the sliding window model can detect multiple mutation points in the timing data.However,the traditional window model has a fixed window length and cannot adjust the size autonomously according to the characteristics of the data stream,so the performance requirements cannot be guaranteed.The ultimate purpose of data detection is to obtain the information hidden in the data.Classifying and matching the data can quickly identify the data categories,so as to obtain the potential value information in the data.Based on the above problems and shortcomings,an online detection model based on multi-feature fusion and template matching was proposed,which can receive data in real time and adjust buffer and window size autonomously according to the mutation density.At the same time,multi-feature fusion and template matching are performed on the data to obtain data category information.Firstly,a sliding window model based on mutation point density was proposed,which automatically adjusted the window length according to the number of mutation points in unit window length.On this basis,a buffer was added to construct an online detection model based on mutation point density.Simulation experiments were used to verify the rapidity and accuracy of online detection of timing data.Secondly,a multi-feature fusion template is added on the basis of the above model to classify the data in the window.The multi-feature fusion template is based on the known data classification to create the category feature vector.The corresponding window feature vector is calculated for the data in the window,and the template is matched and classified according to the equal grading strategy.Based on the simulation data and epilepsy data,the classification effect and accuracy of the data are verified.Finally,by combining the online detection model based on mutation density with the multi-feature fusion and template matching model based on feature vector,the same EEG signal of different epileptic patients and different lesion signals of the same epileptic patient were tested to verify the occurrence rule of similar epileptic patients and the correlation between different lesion signals in the process of epileptic patients.It provides an auxiliary basis for the prevention and diagnosis of epilepsy diseases.