Study On Similarity Search Method Of Uncertain Time Series Based On Spatial Index

Time series is a special series of time series data.It is widely used in applications such as predictive analysis,pattern matching,and information search.It is an important research object in academic research and industrial production.As the operation of data inevitably introduces noise,companies are increasingly concerned about the impact of uncertainty caused by noise on the results.In this paper,the index construction is studied based on the continuous time series model,the traditional index structure is improved,and the index structure suitable for the uncertainty time series is constructed.Firstly,this paper compares the effects of preprocessing techniques such as MA,ARMA and UMA on the data in the preprocessing method of traditional time series,and summarizes the continuity model based on probability distribution and the discrete type based on the set of uncertain time series.The representation of the model.At the same time,based on the research of traditional spatial index,the role of preprocessing technique DFT on timing type data is discussed.Next,the similarity metrics that can be used to calculate the uncertainty time series are discussed,as well as the computational complexity of the different metrics.After that,the characteristics of traditional spatial index structure are discussed.The characteristics of different indexes on sequence data are summarized.A more targeted index structure is constructed for offline data and sequence data in real-time environment.Then,this paper improves the traditional R-tree index structure,based on the continuous model of uncertainty time series,further deduces the screening formula based on the error function,and quantifies the similarity difference using the Euclidean distance based on the mean,and is the mean value.A continuous sequence model consisting of variance and variance is used to construct VR-tree.In order to improve the efficiency of the search,this paper proposes a DP prune strategy for fast screening/filtering by using the quantifiability of metrics and the monotonicity of thresholds.Different from the traditional index structure,this paper calculates the minimum value of the distance threshold by calculating the extreme value of the variance at each tree node in advance,and then compares the two possibilities of similarity at the maximum value of the threshold.The situation thus quickly filters non-candidate sets and acquires candidate sets,ensuring efficiency and accuracy of the lookup.In order to deal with the situation that each time stamp corresponds to different variances,this paper firstly sets the heteroskedasticity sequence to the same extreme value variance and then constructs theindex structure and then searches.At the same time,in order to avoid excessive deviation caused by the protocol process,this paper proposes a variance based on variance.Weight preprocessing algorithm.Finally,for the streaming data in the online environment,this paper considers the characteristics of real-time data requiring fast response,low latency,high throughput,and targeted optimization of update and lookup algorithms.This paper takes advantage of KD-tree's update advantages and R-tree's search advantages,and proposes KDR-tree as an index for dynamic construction.The KDR-tree reduces the number of splits to the leaf nodes by adjusting the maximum number of points K that can be accommodated in the corresponding nodes,thereby improving the efficiency of data update.At the same time,the K value is more efficient for KDR-tree search for nodes because it reduces the depth of the tree during dynamic construction and increases the intensity of point distribution within the space.