时间序列分类问题是时间序列数据挖掘中的一项重要任务,近些年受到了越来越广泛的关注.该问题的一个重要组成部分就是时间序列间的相似性度量.在众多相似性度量算法中,动态时间规整是一种非常有效的算法,目前已经被广泛应用到视频、音...时间序列分类问题是时间序列数据挖掘中的一项重要任务,近些年受到了越来越广泛的关注.该问题的一个重要组成部分就是时间序列间的相似性度量.在众多相似性度量算法中,动态时间规整是一种非常有效的算法,目前已经被广泛应用到视频、音频、手写体识别以及生物信息处理等众多领域.动态时间规整本质上是一种在边界及时间一致性约束下的点对点的匹配算法,能够获得两条序列间的全局最优匹配.但该算法存在一个明显的不足,即不一定能实现序列间的局部合理匹配.具体的讲,就是具有完全不同局部结构信息的时间点有可能被动态时间规整算法错误匹配.为了解决这个问题,提出了一种改进的基于局部梯度和二进制模式的动态时间规整算法LGBDTW(local gradient and binary pattern based dynamic time warping),通过考虑时间序列点的局部结构信息来强化传统动态时间规整算法.所提算法虽然实质上是一种动态时间规整算法,但它通过考虑序列点的局部梯度和二进制模式值来进行相似性加权度量,有效避免了具有相异局部结构的点匹配.为了进行全面比较,将所提出的算法应用到了最近邻分类算法的相似性度量中,并在多个UCR时间序列数据集上进行了测试.实验结果表明,所提出的方法能有效提高时间序列分类的准确率.此外,实例分析验证了所提出算法的可解释性.展开更多
A time-series similarity measurement method based on wavelet and matrix transform was proposed,and its anti-noise ability,sensitivity and accuracy were discussed. The time-series sequences were compressed into wavelet...A time-series similarity measurement method based on wavelet and matrix transform was proposed,and its anti-noise ability,sensitivity and accuracy were discussed. The time-series sequences were compressed into wavelet subspace,and sample feature vector and orthogonal basics of sample time-series sequences were obtained by K-L transform. Then the inner product transform was carried out to project analyzed time-series sequence into orthogonal basics to gain analyzed feature vectors. The similarity was calculated between sample feature vector and analyzed feature vector by the Euclid distance. Taking fault wave of power electronic devices for example,the experimental results show that the proposed method has low dimension of feature vector,the anti-noise ability of proposed method is 30 times as large as that of plain wavelet method,the sensitivity of proposed method is 1/3 as large as that of plain wavelet method,and the accuracy of proposed method is higher than that of the wavelet singular value decomposition method. The proposed method can be applied in similarity matching and indexing for lager time series databases.展开更多
To quickly find documents with high similarity in existing documentation sets, fingerprint group merging retrieval algorithm is proposed to address both sides of the problem:a given similarity threshold could not be t...To quickly find documents with high similarity in existing documentation sets, fingerprint group merging retrieval algorithm is proposed to address both sides of the problem:a given similarity threshold could not be too low and fewer fingerprints could lead to low accuracy. It can be proved that the efficiency of similarity retrieval is improved by fingerprint group merging retrieval algorithm with lower similarity threshold. Experiments with the lower similarity threshold r=0.7 and high fingerprint bits k=400 demonstrate that the CPU time-consuming cost decreases from 1 921 s to 273 s. Theoretical analysis and experimental results verify the effectiveness of this method.展开更多
文摘时间序列分类问题是时间序列数据挖掘中的一项重要任务,近些年受到了越来越广泛的关注.该问题的一个重要组成部分就是时间序列间的相似性度量.在众多相似性度量算法中,动态时间规整是一种非常有效的算法,目前已经被广泛应用到视频、音频、手写体识别以及生物信息处理等众多领域.动态时间规整本质上是一种在边界及时间一致性约束下的点对点的匹配算法,能够获得两条序列间的全局最优匹配.但该算法存在一个明显的不足,即不一定能实现序列间的局部合理匹配.具体的讲,就是具有完全不同局部结构信息的时间点有可能被动态时间规整算法错误匹配.为了解决这个问题,提出了一种改进的基于局部梯度和二进制模式的动态时间规整算法LGBDTW(local gradient and binary pattern based dynamic time warping),通过考虑时间序列点的局部结构信息来强化传统动态时间规整算法.所提算法虽然实质上是一种动态时间规整算法,但它通过考虑序列点的局部梯度和二进制模式值来进行相似性加权度量,有效避免了具有相异局部结构的点匹配.为了进行全面比较,将所提出的算法应用到了最近邻分类算法的相似性度量中,并在多个UCR时间序列数据集上进行了测试.实验结果表明,所提出的方法能有效提高时间序列分类的准确率.此外,实例分析验证了所提出算法的可解释性.
基金Projects(60634020, 60904077, 60874069) supported by the National Natural Science Foundation of ChinaProject(JC200903180555A) supported by the Foundation Project of Shenzhen City Science and Technology Plan of China
文摘A time-series similarity measurement method based on wavelet and matrix transform was proposed,and its anti-noise ability,sensitivity and accuracy were discussed. The time-series sequences were compressed into wavelet subspace,and sample feature vector and orthogonal basics of sample time-series sequences were obtained by K-L transform. Then the inner product transform was carried out to project analyzed time-series sequence into orthogonal basics to gain analyzed feature vectors. The similarity was calculated between sample feature vector and analyzed feature vector by the Euclid distance. Taking fault wave of power electronic devices for example,the experimental results show that the proposed method has low dimension of feature vector,the anti-noise ability of proposed method is 30 times as large as that of plain wavelet method,the sensitivity of proposed method is 1/3 as large as that of plain wavelet method,and the accuracy of proposed method is higher than that of the wavelet singular value decomposition method. The proposed method can be applied in similarity matching and indexing for lager time series databases.
基金Project(60873081) supported by the National Natural Science Foundation of ChinaProject(NCET-10-0787) supported by the Program for New Century Excellent Talents in University, ChinaProject(11JJ1012) supported by the Natural Science Foundation of Hunan Province, China
文摘To quickly find documents with high similarity in existing documentation sets, fingerprint group merging retrieval algorithm is proposed to address both sides of the problem:a given similarity threshold could not be too low and fewer fingerprints could lead to low accuracy. It can be proved that the efficiency of similarity retrieval is improved by fingerprint group merging retrieval algorithm with lower similarity threshold. Experiments with the lower similarity threshold r=0.7 and high fingerprint bits k=400 demonstrate that the CPU time-consuming cost decreases from 1 921 s to 273 s. Theoretical analysis and experimental results verify the effectiveness of this method.
