Perceptually Important Points-based Futures Time Series Pattern Recognition And Its Application Research

Time series is an important class of temporal data objects, and it is easy to getfrom the technology and financial application. Every data of a time series isobserved in chronological order, then the time series is obtained by ordered thesedata. The nature of time series includes: the amount of data is very large, highdimensions and must be frequently updated. In order to effectively mining theimplicit knowledge of the historical data, academic and industrial circles use avariety of techniques to reduce the time sequence complexity, and a commonly usedmethod is to convert the time sequence into another kind of easy to understand theimportant points method based on visual expression. Meanwhile, in order toovercome the dimension disaster of time series data, a corresponding indexstructure that based on the expressing method is proposed. Aimed at identifying afutures time series data set with certain characteristics and more effectiveidentification for some kind of model, this dissertation studies its featureidentification and indexing methods. The main research work and achievements areas follows.(1) Propose the ZPIP (Zigzag based Perceptually Important Points) importancepoint recognition method based on Zigzag perceptually important points. Themethod is based on the PIP (Perceptually Important Points) method. The algorithmis optimized by the maximum local method and state balance method to graduallyoptimize the locally important points, and make it more effective identificationfutures time series feature points.(2) Present the FIS (Feature Index Structure) feature index mechanism basedon the binary search tree to solve the inefficient issue of recognition process inreal-time. The method is based on a binary search tree, in order to more suitable forfutures time series of feature recognition, we improved algorithms by introducingthe concept of important point and the principle of the important points that havelarger index will more nearer the root node and get a good performance.(3) Put forward the FZPIP (Feature Zigzag based Perceptually ImportantPoints) feature recognition method based on the feature Zigzag perceptuallyimportant points by introducing the definition of futures trend characteristics. During the process of feature recognition, we obtain the candidate time series withcharacteristics condition by the difference method. The proposed method is provedhave good performance by the experiments that use practical futures trading data.(4) Analysis futures time series analysis functions-Zigzag function, ultimatelythe ZPIP important point identification method based on the perception of theimportant points was introduced to this function. And through the analysis andverification of the data set, prove that the time series for some features, can achievevery good results based on the ZPIP of Zigzag function.(5) Implement the futures time series pattern recognition system based on theK-line and perceptually important points, and above the basis of the researchresults, it can achieve the identification of a specific pattern.