Research Of Time Series Approximate Representation And Clustering Algorithm

Time series is the change tendency described on the time axis.Time series exists in everywhere,existing in many fields.But Time series is complex and high dimension,it is very important to research data mining effectively.In this paper,we use time series as the research object.Because time series is very big,not conducive to the subsequent processing,so studying time series approximation method.In time series approximate representation,segmentation is simple,intuitive and supporting for the similarity search,so been wider application.In this paper,the time series segmentation representation is further studied.Since the time series data represented by the segmentation is different from other types of data when clustering,researching on how to effectively clustering the piecewise time series data is very important.In this paper,we firstly introduce the background and research s tatus of time series.Secondly,summarize the basic theory of time series approximation and clusters,then extract the further study-A high-order polynomial approximate representation method based on key points for time series(KPPR)is proposed according to the exis ting algorithm.The main idea of the algorithm is divided into two parts :(1)find time series’ s key points: The local max-min points and the extreme points which meet the conditions;(2)the theoretical foundation and ma thematical derivation of higher order polynomial representation.Finally,the simulation results show that the approximation method can guarantee the good trend extraction under different compression ratio.By comparing different algorithm,the proposed algorithm can effectively reduce the fitting error while ensuring the compression rate,and improve the time series approximation.Secondly,clustering is an important step in identifying the inner relations of data objects.However,clustering algorithms are mostly clustered on discrete data sets,few of studies about segmentation time series.Therefore,this paper studies the hierarchical clustering of the result set obtained by KPPR algorithm,and proposes an modified hierarchical clustering algorithm based on DTW distance measurement(DTWMHC).The overall idea of the algorithm is divided into two parts :(1)Improvement of distance measurement: Use DTW instead of Euclidean distance to measure the distance,which is more accurate.(2)Improvement of algorithm efficiency: Hierarchical clustering algorithm has high complexity and is not conductive to dealing with large-scale sequences.In this paper,improve the hierarchical clustering distance matrix update method to reduce the computational complexity.Finally,the simulation results show that the proposed algorithm is more efficient and reduce the running time of the algorithm.