多变量时间序列(multivariate time series,MTS)分类任务旨在确定多变量时间序列样本的标签。多变量时间序列数据存在时序关系和样本相似性关系等丰富的关系信息,然而现有的算法未能充分利用关系信息导致分类性能难以提升。基于此,文章...多变量时间序列(multivariate time series,MTS)分类任务旨在确定多变量时间序列样本的标签。多变量时间序列数据存在时序关系和样本相似性关系等丰富的关系信息,然而现有的算法未能充分利用关系信息导致分类性能难以提升。基于此,文章提出一种基于图卷积网络(graph convolutional network,GCN)的多变量时间序列分类方法,通过挖掘样本间的潜在关系来提高分类性能。为了有效表示样本关系,设计基于样本相似度的构图规则,对样本数据进行建模从而将样本的时序特征和潜在关系信息映射到图空间中,提出基于图卷积的分类模型,通过聚合样本特征来捕获有利于分类的潜在样本关系,更新到样本自身特征向量以提升分类精度。在11个公共数据集上的大量实验结果表明,该文所提算法优于12种对比算法,可见通过挖掘时间序列数据之间潜在的关系用于分类对分类结果具有重要影响,从而为处理时间序列分类问题提供一种新的途径。展开更多
By the generalized variational principle of two kinds of variables in general me-chanics,it was demonstrated that two Lagrangian classical relationships can be applied to both holonomic systems and nonholonomic system...By the generalized variational principle of two kinds of variables in general me-chanics,it was demonstrated that two Lagrangian classical relationships can be applied to both holonomic systems and nonholonomic systems. And the restriction that two Lagrangian classical relationships cannot be applied to nonholonomic systems for a long time was overcome. Then,one important formula of similar La-grangian classical relationship called the popularized Lagrangian classical rela-tionship was derived. From Vakonomic model,by two Lagrangian classical rela-tionships and the popularized Lagrangian classical relationship,the result is the same with Chetaev's model,and thus Chetaev's model and Vakonomic model were unified. Simultaneously,the Lagrangian theoretical framework of dynamics of nonholonomic system was established. By some representative examples,it was validated that the Lagrangian theoretical framework of dynamics of nonholonomic systems is right.展开更多
文摘多变量时间序列(multivariate time series,MTS)分类任务旨在确定多变量时间序列样本的标签。多变量时间序列数据存在时序关系和样本相似性关系等丰富的关系信息,然而现有的算法未能充分利用关系信息导致分类性能难以提升。基于此,文章提出一种基于图卷积网络(graph convolutional network,GCN)的多变量时间序列分类方法,通过挖掘样本间的潜在关系来提高分类性能。为了有效表示样本关系,设计基于样本相似度的构图规则,对样本数据进行建模从而将样本的时序特征和潜在关系信息映射到图空间中,提出基于图卷积的分类模型,通过聚合样本特征来捕获有利于分类的潜在样本关系,更新到样本自身特征向量以提升分类精度。在11个公共数据集上的大量实验结果表明,该文所提算法优于12种对比算法,可见通过挖掘时间序列数据之间潜在的关系用于分类对分类结果具有重要影响,从而为处理时间序列分类问题提供一种新的途径。
基金Supported by the National Natural Science Foundation of China (Grant No. 10272034)the Research Fund for the Doctoral Program of Higher Education of Chinathe Basic Research Foundation of Harbin Engineering University (Grant No. 20060217020)
文摘By the generalized variational principle of two kinds of variables in general me-chanics,it was demonstrated that two Lagrangian classical relationships can be applied to both holonomic systems and nonholonomic systems. And the restriction that two Lagrangian classical relationships cannot be applied to nonholonomic systems for a long time was overcome. Then,one important formula of similar La-grangian classical relationship called the popularized Lagrangian classical rela-tionship was derived. From Vakonomic model,by two Lagrangian classical rela-tionships and the popularized Lagrangian classical relationship,the result is the same with Chetaev's model,and thus Chetaev's model and Vakonomic model were unified. Simultaneously,the Lagrangian theoretical framework of dynamics of nonholonomic system was established. By some representative examples,it was validated that the Lagrangian theoretical framework of dynamics of nonholonomic systems is right.
