Research On Information Granulation Method Of Time Series For Clustering And Forecasting

With the development of information technology,a large amount of time series data is generated and stored in the field of economics and management.By using data mining algorithms,potential and valuable information can be mined to support management and decision-making activities.However,these time series data usually have significant high-dimensional characteristics.If data mining algorithms are directly applied to them,it will cause excessive computational complexity and the data mining results will also be affected.Granular computing is a new method of simulating human problem-solving thinking and solving complex tasks of big data.The main idea of this theory is to abstract and divide complex problems into several simpler problems,namely granulation,contributing to better analysis and solve problems.This paper introduces the granulation idea of granular computing into time series analysis.By performing information granulation operations on the time series,the original high-dimensional time series is transformed into a low-dimensional granular time series.The constructed information granules can describe and reflect the structural characteristics of the original time series data,so as to achieve efficient dimensionality reduction and lay the foundation for subsequent data mining work.Aiming at the problem of time series information granulation,this paper proposes three different time series information granulation methods from the two aspects of time axis and value axis:a time series information granulation method of time axis based on fluctuation points,a time series information granulation method of time axis based on cloud model,and a fuzzy time series forecasting method based on information granulation of value axis.And apply these methods to stock time series data for clustering and forecasting can provide decision-making suggestions for the stock's selection and the future stock market trends judgment of Chinese stock market.The main contents and innovations of this study are as follows:(1)For the time axis of time series,in view of the structural features of low-frequency time series,an information granulation method of time series based on fluctuation points and a similarity measurement method of granular time series are proposed.Firstly,the time series information granulation method based on fluctuation points is put forward.We identify the fluctuation points to segment the original time series into several information granules,and then use linear function to describe each information granule,which solves the hard division problem of traditional time series dimensionality reduction and can more effectively extract the structural features of time series while reducing the dimension.Secondly,a new similarity measurement method of granular time series is proposed.Finally,we perform clustering experiments on some UCR datasets,and the experimental results show that the proposed information granulation method and similarity measurement method can improve the accuracy of the clustering results.We also apply the proposed algorithm on the stock dataset of Science and Technology Innovation Board for practical application research,and the research results can provide investors with a reference when selecting stocks on Science and Technology Innovation Board stock market.(2)For the time axis of time series,in view of the structural features of high-frequency time series,an information granulation method of time series based on cloud model and a similarity measurement method of granular time series are proposed.Firstly,the time series information granulation method based on cloud model is put forward.This method can adaptively represent the original time series as several normal cloud models without specifying the number of information granules in advance.And it has the stronger dimensionality reduction capabilities,so are more suitable for the dimensionality reduction requirements of high-frequency time series.Secondly,a new similarity measurement method of cloud model sequence is proposed.Finally,we perform clustering experiments on some UCR datasets,and the experimental results show that the proposed information granulation method and similarity measurement method can improve the accuracy of the clustering results.We also apply the proposed algorithm on the stock dataset of Shanghai and Shenzhen A Shares for practical application research,and the research results can provide investors with a reference when selecting stocks on Shanghai and Shenzhen A Share stock market.(3)For the value axis of time series,a time series domain division method based on fuzzy C-means clustering and information granulation is proposed,and a time series prediction method is further given.Firstly,we apply fuzzy c-means clustering method to partition the discourse of time series,and then put forward the discourse partition optimization algorithm based on information granulation.This method improves the accuracy of sample data division,and has stronger interpretability.Secondly,according to the partition results of time series,a fuzzy time series forecasting method is proposed.This method converts accurate time series data into a time series composed of semantic values conforming to human cognitive morphology,and the prediction results are more understandable.Finally,we perform forecasting experiments on Taiwan Stock Exchange Capitalization Weighted Stock Index(TAIEX)dataset,and the experimental results show that using the proposed time series information granulation method for time series forecasting can improve the accuracy of the forecast results.We also apply the proposed algorithm on Shanghai Composite Index(SHCI)dataset for practical application research,and the research results can help investors understand the future trend of the stock market and provide a reference for them to adjust their investment strategies.