The Research And Application Of Time Sequence Similarity Query

In this paper, in-depth study about the method of multi-dimensional time series similarity search, focusing on multi-dimensional analysis of time series pretreatment and similarity measurement, an improved method of multidimensional time series piecewise linear, and on the basis of the segmentation method, this paper proposes a new dynamic time warping distance based on morphological characteristics and multidimensional time series similarity measure method, and the algorithm has been proved by the concrete methods as a result, the final application using the improved algorithm was designed and implemented.The main content of this paper is as follows1)The multivariate time series pattern representation. It is the basis of multivariate time series studies. Firstly, we proposed PAA_ERR algorithm based on PAA method. PAA ERR method can reduce the dimension on the time dimension. It calculates fitting error on all segments to ensure the size of sliding window. Finally we extract the inclination angle and the shape characteristics of the segmented time series values as the mode of time series. Experiments show that the algorithm can be a good fit for time series’ fitting, and the calculation is simple and easy to implement.2)The similarity measurement of multivariate time series. According to the advantages and disadvantages of multi-dimensional time series data, we proposed SA_DTW method based on dynamic time warping distance. Firstly, we used PAA_ERR algorithm to get the pattern representation,that is the shape characteristics of each segment of the sequence and the tilt angle. Finally, we carry out seeking the same pattern through a dynamic time warping distance between the different dimensions. Experiments show that this method can effectively improve the accuracy of similarity query.3)In this paper, designed and implemented a platform to show the practical application on the basis of he above algorithm. The platform consists of the pattern representaion of time series module, similarity query of time series module, and the application based on improved algorithm module.The first two module is mainly used to show actual operation effect of PAA_ERR and SA_DTW algorithm, and the modual of the application can provide real-time data through the screen gestures sliding track point data for time series similarity comparison. The results show that SA_DTW method have good match result for these sequences. Finally, Desktop Assistant based on the improved algorithm is simple to use, flexible and diverse mode of operation characteristics.