Research On Dynamic Optimization And Classification Method Of Multi-index Time Series Data

The comprehensive evaluation technology based on multi-index time series data analysis is widely used in enterprise decision,fault diagnosis,smart medical care and weather forecasting.Multi-index time series data consists of multiple single time-series data.There are specific heterogeneity characteristics between each index series data,such as corporate financial data and human resources data,in terms of time-series data collection cycle,data valid range,etc.There are significant differences.At the same time,there are also differences in the integrity and validity of multi-index data among different comprehensive evaluation subjects.For example,in the forecast analysis of the enterprise market,listed companies and unlisted companies have different meanings in terms of stock indicators such as total share capital,so stock indicators are not suitable for analyzing the financial status of unlisted companies,and should be excluded to reduce the difficulty of implementing the evaluation method.According to the survey,the current research lacks effective and objective methods for multi-index optimization and screening individual objects under the whole index system.Most research on time series data classification is also aimed at a single index.Therefore,effectively optimizing,screening,and classifying multi-index time series data is a crucial technology that needs to be solved urgently in the current comprehensive evaluation application.This thesis first defines the multi-index optimization problem based on the correlation of indexes and designs the corresponding solution algorithm to solve the above problems.Secondly,a classification method based on multi-index shapelets time series data is proposed based on the optimized index system.The main work of this thesis is as follows:1)A Multi-Index Optimization Based on Relevance Constraints(MIO-RC)problem is proposed.The theoretical proof and time complexity analysis of this problem are carried out,and it is proved that the MIO-RC problem is an NP-hard problem.According to the different application scenarios of whether to consider the correlation constraints between indexes,a dynamic programming algorithm and an IR-GA algorithm are designed to solve the problem.2)Based on the optimized multi-index time series,this thesis fully considers the inaccurate classification of objects with single-index time series data,and the classification results are not interpretable.The method firstly performs data preprocessing on the index data with missing data,so that the processed data can meet the operation of the model;secondly,based on the learning shapelets technology,multiple groups of shapelets of each index are extracted from the index data of multiple indexes,that is,to capture each different features of time series data for each index;finally,the final classification of individual objects is completed by combining multi-index shapelets and weighted voting.Finally,this thesis uses the financial data of companies to conduct experimental analysis on the research content from the perspectives of the optimization performance,convergence and classification accuracy of multi-indexes time series data.The experimental results prove that the algorithm proposed in this thesis not only can effectively solve the problem of multi-index time series data analysis,and the convergence is faster,and the accuracy can be improved by about 3% compared with the traditional method.