Research On Time Series Clustering Based On Manifold Learning

The real world is not static, but is changing with time. Time series is usually a high dimensional and according to time order of sequence.Its generating process is very easy to be influenced by the surrounding environment, and there is a certain amount of noise and time is usually continuous and uniform distribution. Clustering is of no class label samples were grouped according to the similarity and similarity of large samples into a group, the minimum similarity between different groups of samples. Time series clustering is widely used in industry, agriculture, transportation, gesture recognition and other fields. It has a lot of application examples, such as investment portfolio risk management, natural language understanding, traffic flow, and so on. In this paper, time series data for the study, to explore the use of different manifold learning algorithms for time series data clustering and cluster integration to improve clustering performance. The main research work is as follows:(1) Research on time series clustering based on manifold learning. Time series data is often not only a large amount of data, but also high-dimensional, directly on the raw data time series clustering performance is not satisfactory. How to effectively reduce the dimension of time series, and to maintain the essential characteristics of the original data set, is one of the research points in this paper. According to the characteristics of time series of 10 time series data from different fields set, we use three kinds of manifold learning method Locality Preserving Projection, Locally Linear Embedding, Neighborhood Preserving Embedding to its dimensionality reduction, and reduce the dimension of the data using K-means algorithm for clustering. The three manifold learning methods can not only reduce the dimensionality of high-dimensional data, but also try to find the low dimensional structures hidden in the high dimensional data. The three manifold learning algorithm experimental results were compared with the raw data directly K-means clustering, using Principal Component Analysis dimension reduction after clustering, using Piecewise Aggregate Approximation dimension reduction after clustering and clustering results were analyzed by paired samples t test,and the results show that the clustering performance of the three manifold learning algorithms is better than those of the method.(2) Research on clustering fusion of time series based on manifold learning. Using only one clustering algorithm may not be stable, and the result of the fusion of multiple clusters has become a trend. Cluster fusion is the fusion of multiple clustering algorithms or a clustering algorithm with different initialization or parameters, and the final clustering results are obtained by using the consensus function. It can improve the stability of clustering algorithm. From different fields of 10 time series data set dimensionality reduction using LPP, LLE, NPE and dimensionality reduction of data repeatedly use k-means clustering to get cluster members. The cluster members are fused by weighted voting method. The three manifold learning algorithm experimental results were compared with the raw data directly Kmeans clustering, using Principal Component Analysis dimension reduction after clustering, using Piecewise Aggregate Approximation dimension reduction after clustering and clustering results were analyzed by paired samples t test,and the results show that the clustering performance of the three manifold learning algorithms is significantly better than those of the method.