Construction Of Pathological Data Expert System Model Based On Dynamic Sliding Window And Template Matching

With the development of information technology and the advancement of data storage technology,the amount of data generated by various industries is increasing,and human society has entered the era of big data.As an important branch in the field of big data analysis,time series data anomaly detection has vast application prospects in many industries such as industry,medical care and education.Time series data change-point detection is an anomaly detection technology,which starts from the overall distribution of the data,searches for the location of the mutation in the data stream,and analyzes the causes of the change points.This technology can extract valuable information from large-scale data effectively,so it has become a research hotspot in the field of big data analysis.Sliding window model is one of the key technologies required for the change-point detection in time series data.It can divide the data stream into multiple pieces of data,and use change-point detection algorithm to analyze each segment of data,which can effectively locate the location of the change point.The traditional fixed-width sliding window cannot automatically adjust the window width,so it is ineffective in detecting pathological time series data.To solve this problem,this thesis proposes a dynamic change strategy of sliding window,and applies it to abnormal detection of pathological data.At present,the automatic detection and analysis of pathological data have become a hot issue in the field of anomaly detection.Based on change-point detection technology,this thesis constructs a set of expert system models with epilepsy disease screening and pathological state analysis,and applies it to the analysis of epilepsy.Firstly,this thesis proposes a dynamic sliding window model that can adaptively adjust the window width according to the intensity of abnormal fluctuations of data in the current window,and combines it with the change-point detection method based on multi-channel search tree to form a set of multiple change points detection model,which is applied to multiple change points detection of time series data.Experiments show that compared with the fixed-width sliding window,this model can optimize the effect of change point detection to a certain extent.Moreover,the detection effect of the model can be further improved by adding the mechanism of truncation and random overlapping of windows.Secondly,based on the multiple change point detection model,this thesis proposes the distribution density feature of change points to measure the intensity of abnormal fluctuations of data.This thesis also uses the distribution density of mutation points,combined with the K-means and random forest model to analyze the abnormal state of EEG signals of large-scale epilepsy.Experiments show that this method basically realizes the distinction between healthy EEG and epileptic EEG,which lays a foundation for solving the problem of epileptic disease screening.Thirdly,this thesis proposes an abnormal state recognition method based on symbolic template matching to solve the problem of epileptic state recognition.Firstly,the symbolic approximate representation based on peak valley difference is used to reduce the dimension of the EEG signal and obtain the symbolic representation of EEG data.Then,through the study of the EEG of typical epileptic seizures,the symbolic templates of three types of epileptic states are formed.Finally,the improved Sunday algorithm with a fault-tolerant mechanism is used to match the measured data and the lesion data template.Experiments show that the above method has a good effect in identifying epileptic state,which lays a foundation for solving the problem of epileptic state identification.Finally,this thesis constructs an expert system model based on disease screening and disease state analysis,which is applied to the rapid analysis and preliminary diagnosis of the disease state of EEG data.In this model,firstly,the data preprocessing stage uses the multiple change points detection technology to extract the EEG data features.Secondly,the expert system uses the random forest model to further distinguish the abnormal state of EEG into negative and positive.Thirdly,in the pathological state identification module,the EEG data diagnosed as epilepsy-positive is symbolized to locate the period of epilepsy.Finally,the standard pathological state template in the expert template library is used to match the corresponding pathological state.The analysis results of simulation experiments and actual epilepsy EEG data show that the expert system model proposed in this thesis can preliminarily complete the rapid analysis of epilepsy lesions and the diagnosis of epileptic seizures.This expert system model is designed to play the role of rapid analysis of lesion data,in order to provide an analysis tool for disease diagnosis and treatment.