A Multiple Change Points Detection Method Based On Random And Adaptive Strategy

Since the term "big data" was proposed in 1980,the development of the Internet and the information industry has enabled data streams to exist as a new form of data in all walks of life in society.Data mining and data analysis technologies have been promoted by the development of huge amounts of data.As an important branch of data stream research,data stream change point detection technology has attracted more and more attention from experts and scholars at home and abroad.Data stream change point detection technology has broad application prospects,but it also faces many difficulties and challenges.As the data stream is fast and timely,complex and changeable,and the amount of data is large,how to quickly analyze the effective information in the data stream to avoid data redundancy has become a hot topic in current research.The traditional data stream change point detection technology can no longer meet the needs of data development,and the sliding window model has become an effective detection technology in the detection of multiple change points.The sliding window model divides massive data into sub-windows to be detected,which greatly improves the speed and accuracy of detection.However,the size of the sliding window and the step length of the window update are two important factors that affect the detection performance of the sliding window.How to determine the appropriate sliding window model according to the distribution of the data stream will affect the detection accuracy of the change point of the data stream.In this paper,starting from two factors that affect the performance of sliding window detection,combined with the TSTKS algorithm,an overlapping window model and a sliding window model under a random strategy are proposed.On this basis,two adaptive adjustment strategies are proposed,which can dynamically update the candidate set of the window size according to the data distribution,and finally apply to the multi-path pathological signal analysis of patients with epilepsy.It provides a new idea for the early warning and diagnosis of epilepsy disease.Firstly,this paper combined the TSTKS change point detection algorithm and the sliding window theory to build a multi-change points detection model,which better solves the problem of time-consuming and low precision detection in traditional methods.By analyzing the impact of window update step size on sliding window detection performance,and aiming at the problem of decreased detection accuracy when the change point is located at the window boundary or blind zone,an overlapping window model is proposed.After many comparison experiments,the detection results show that the introduction of the overlapping window model has achieved good detection results.Secondly,aiming at the local optimal problem of adaptive sliding window model in the change point detection process,a sliding window model under stochastic strategy is proposed.A candidate set selected by the sliding window size is established,and two adaptive adjustment strategies are given on this basis.The range of the candidate set can be dynamically updated according to the distribution of data.The adjustment strategy based on the window volatility slope is to expand or reduce the range of the window candidate set by analyzing the change trend of the slope of the line between the boundary value of each window and the maximum value of the window.The adjustment strategy based on the amount of window fluctuation difference is to first calculate the fluctuation amount of a single window,judge the characteristics of the data distribution by analyzing the fluctuation amount of several consecutive windows,and then dynamically update the range of the candidate set.Finally,the real pathological data of epilepsy patients are analyzed by combining two adaptive sliding window models under random strategy.The multiple pathological signals of the same patient are selected for change point detection,and the early warning of epilepsy and the end of onset diagnosis are performed according to the characteristics of two adaptive adjustment strategies.In order to further verify the experimental results,the same pathological signal of different patients was selected for change point detection.The experimental results show that the adaptive sliding window model under the two random strategies proposed in this paper has a good detection effect in the early warning of epilepsy and the diagnosis of the end of onset.