Time Series Classification And Prediction Based On LSH-Shapelets Transformation

Time series is a series of real values observed and changed with time.It is a kind of important time data object.It is widely used in the field of scientific research and finance,and the research and development of time series has become an important branch in the field of data mining.The classification and prediction of time series are the two main research directions in this field.With the continuous expansion of data scale,the disadvantages of traditional time series analysis methods gradually appear.It is not only the algorithm's universality is not high,but also needs to pay a huge time cost in the analysis process.Time series classification method based on Shapelet is a popular time series classification method at present.Shapelets are the time subseries that can best represent the category they belong to,which can be used to distinguish the differences between different categories and make the classification results more interpretable.However,there is a problem of time-consuming extraction of shapelets.In this paper,a new shapelets conversion method is proposed to ensure the quality of shapelets conversion and to solve the serious defect of long training time of shapelets.The method based on the original sequence,the candidate set for filtering step by step,on the premise of without calculation sequence quality first selected representative part of the sequence according to their shape,greatly narrow the scope of the candidate set,then on a small scale to calculate the quality of each sequence,picked out the best several sequences as shapelets,need not calculate a lot but the sequence of quality,achieve the purpose of save time.Contributions to this article are as follows:1.Aiming at the problem that Shapelets Transform algorithm takes too long due to the existence of a large number of similar sequences in the candidate set of Shapelets,a Locality Sensitive Hashing Shapelets Transform(LSHST)based method for Shapelets is proposed.Then,the transformed shapelets were combined with different classifiers to carry out the time series classification experiment,and the classification accuracy was used to compare the advantages and disadvantages of the extracted shapelets.Experimental results show that the LSHST method can greatly improve the conversion speed and classification accuracy.2.LSH-SHAPELETS transform algorithm is applied to time series prediction,and a time series prediction algorithm based on LSH-SHAPELETS is proposed.In this paper,the LSH-based shapelets extraction method was applied to the residual life prediction of the turbine engine in the background of the residual life assessment of the turbine engine,and a variable weight calculation method considering both distance and position was proposed.In this method,the template Rul-shapelets with the remaining service life information of the equipment were extracted rapidly from the historical data of the equipment operation by LSH.The Rul-shapelets closest to the test sequence were selected to be predicted,and the remaining life of the test sequence was estimated according to the degradation information carried by them.