Time Series Clustering Based On Fuzzy Clustering And Self-organizing Competitive Neural Network

Time series clustering is an important content in the field of data analysis,and it is widely used in various fields of social production.By integrating the data information that change with time,we can explore the development trends,the development rules and researching on cluster things changes.In order to make full use of the time series data,and discover the hidden links from historical data,the improved fuzzy C-means clustering algorithm and Kohonen network clustering algorithm are proposed in this paper to study the clustering of time series.Firstly,the classical dynamic time warping distance algorithm and particle swarm optimization algorithm are combined,and Two improved algorithms based on the particle swarm optimization and the weighted dynamic time warping distance are proposed,which are used to improve the similarity measure function of the time series.Using the combination of fuzzy C-means clustering and fuzzy C-medoids clustering,the problem that the objective function value falls into the local optimum and cannot reach the global optimum in the clustering process due to the improper selection of the initial cluster center,is solved.Finally,the proposed clustering model is simulated by using UCR time series datasets.The obtained clustering results are compared with the existing methods to verify the effectiveness of the clustering model.In addition,the clustering analysis is conducted on the air quality status of 74 cities.The fuzzy clustering algorithm presented in this paper is compared with some existing methods to prove the validity of the clustering model.Secondly,an output layer is added on the basis of the existing two-layer self-organizing Kohonen network,and the unsupervised layer two self-organized network is changed into a semi-supervised three-layer self-organizing network,which improves the clustering accuracy to some extent.In terms of weight initialization,a fuzzy C-means clustering algorithm is used to initialize the weight between the input layer and the competition layer.In the process of network training,the network learning rate is refined from a linear change to a nonlinear change,and a threshold condition is added to the maximum number of iterations given as a condition for ending the network training.For finding neuron distances,the improved dynamic time warping distance proposed before hand is used.Finally,the proposed improved Kohonen network clustering model is simulated by using UCR and UCI public time series datasets.The obtained clustering results are compared with several existing methods to verify the effectiveness of the clustering model.In addition,simulation experiments are performed on a series medical datasets.