基于连续密度隐马尔可夫的时间序列分类算法  

Time Series Classification Algorithm Based on Continuous Density Hidden Markov

作　　者：李霞[1] LI Xia(City Colledge,University of Science and Technology of Wuhan,Hubei Wuhan 430083,China)

机构地区：[1]武汉科技大学城市学院,湖北武汉430083

出　　处：《计算机仿真》2021年第1期291-294,共4页Computer Simulation

摘　　要：针对数据挖掘过程中对异常数据检测的准确率较低、分类速度较慢,导致数据分类准确率较低、效率较差的问题,提出基于连续密度隐马尔可夫的时间序列分类算法。构建时间序列变化趋势分割点目标函数,利用贪婪搜索法求解时间序列分段值,提取序列变化趋势特征得到数据主要信息,提升数据分类的准确性;改进帧内特征表达准确性,使用因子分析矩阵高斯分布建立连续密度隐马尔可夫模型,提高时间序列分类速度;采用平稳子空间分析法把数据划分为平稳子空间和非平稳子空间,运用相对熵权衡平稳子空间分布相似度,实现时间序列精准分类。仿真结果表明,所提方法分类正确率较高、计算速度快且鲁棒性好,可以满足真实场景下数据分析需求。Aiming at low accuracy and low classification speed of abnormal data detection in the process of data mining,an algorithm of time series classification based on continuous density hidden Markov was proposed.At first,the objective function of segmentation point of time series trend was established,and then the greedy search method was used to solve the piecewise value of time series,and thus to extract the sequence trend features and the main information of data.In this way,the accuracy of data classification was improved.Moreover,the accuracy of intra-feature expression was improved,and the Gaussian distribution of factor analysis matrix was used to build a continuous density hidden Markov model,and thus to improve the classification speed of time series.In addition,the stationary subspace analysis method was used to divide the data into a stationary subspace and a non-stationary subspace.Finally,the relative entropy was used to weigh the similarity of stationary subspace distribution,so as to achieve the accurate classification of time series.Simulation results show that the proposed method has high classification accuracy,fast calculation and good robustness,so it can meet the needs of data analysis in real scenarios.

关 键 词：时间序列分类 隐马尔可夫模型 因子分析 相对熵 连续密度 

分 类 号：TP353[自动化与计算机技术—计算机系统结构]