Research On Dimensionality Reduction And Storage Methods Of Uncertain Time Series

Time series is an important data type of big data,and it is also an important research direction in the field of data mining.The existing research results on time series mainly focus on the definite time series,and there are still many deficiencies in the research on the uncertain time series.This paper focuses on the dimensionality reduction and storage of uncertain time series.The main work is divided into three parts.First of all,the existing methods of representation of time series are analyzed.Aiming at uncertain time series,a new method of integrated dimensionality reduction is proposed.The method reduces dimensionality of the original time series from time dimension and probability dimension respectively.In the time dimension,the key-point based linear dimensionality reduction method is adopted.Double traversal and selection of data points are carried out,and a good balance is achieved between data reduction and over-removal.In the probability dimension,use high probability points to replace small probability points.Additionally,a repeat cleaning strategy is proposed to avoid the drawbacks of data screening caused by abnormal points.Finally,by means of experiments,the superiority of the proposed integrated dimensionality reduction algorithm is verified.Then,aiming at the dynamicality of the probability distribution of the continuous uncertain time series at each moment,a new method of dynamically inferring the probability distribution of the uncertain time series is proposed.The method relies on the existing ARMA model and GARCH model,and proposes a new estimation model I-GARCH.The model takes into account the variation of time series and dynamically deduces the probability distribution at each moment.At the same time,in order to further enhance the fault tolerance of the model,a corresponding error value elimination algorithm is proposed.Finally,the experiment verifies the accuracy of the proposed method for the representation of continuous uncertain time series.Finally,the disadvantages of the existing time-series storage framework are analyzed,and the basic principles of the design of the new storage framework for the uncertain time series are proposed.On this basis,a new storage framework of uncertain time series is proposed based on the above existing work on discrete and continuous uncertain time series.And through specific structural analysis shows its superior performance in processing and storage of uncertain time series data.