Research On Behavior Matching And Evaluation Technologies Of Time Series

Behavior matching is similarity comparison of the current behavior with the existing patterns of behavior, and behavioral evaluation is the completion and risk coefficient of the behavior, they are all important behavior research. Behavior’s definition is very broad, Such as biological and non-biological behavior, conscious and unconscious behavior etc. In this Thesis, system behaviors are treated as the study objects and time series data was study data, and conduct behavior matching on Time series.Time series is recording data which is sorted by time and changes over time. It is widely used in various fields of natural: economic, financial, scientific observation, humanities surveys and engineering. How to effectively manages and utilize these historical time series data and discovery implicit rule and knowledge in it, has become an issue of concern. Time series analysis techniques have developed rapidly in recent years, which include model representation, similarity search, classification, clustering, and anomaly detection.This thesis achieves behavior matching and evaluation by analyzing time series, It models behaviors by time series and presents a framework for behavior matching and evaluation on the basis of time series analysis techniques, Framework with the purpose of the behavior matching and evaluation integrates the major technical of time series analysis. Two examples are used to illustrate behavior matching and of the time series and cited a number of innovative methods and technologies proposed in this thesis.Contribution and innovation are as follows:·Proposed the behavior matching and evaluation framework of time series, achieves behavior matching and evaluation on univariate and multivariate time series. Take flight parameters and system calls for example to elaborate the framework’s applicability to real multivariate time series and symbolic space time series.·Proposed multi-dimensional piecewise linear fitting modeling method which based on vector space distance for multivariate time series. Achieved segmentation and modeling of multivariate time series accurately, and solved the inconsistency of cut-off points in multivariate time series segmentation.·Proposed Multivariate Dynamic Time Warping(MDTW) for similarity measurement of multivariate time series, and achieved unification measure of variables in multivariate time series.·Proposed Short sequence Matching and Markov Estimating(SM&ME) for modeling and anomaly detection of symbolic space time series, the model can effectively extract the probability characteristic of sequences. Short sequence Inquiry and probability estimation allows the combination of efficiency and accuracy and given good performance.The behavior matching and evaluation framework of time series and related technologies have been adopted in Flight Quality Monitoring System and Configurable Secure Computing Platform, it verifies the correctness and validity of this framework and innovative technologies.