基于规则迭代的时间序列特征提取模型
|
摘要
针对时间序列分类特征提取困难、提取的特征不明显等问题,提出一种基于普通规则嵌套迭代的时间序列特征提取模型。以常规的序列特征提取规则为基本单元,构建多级迭代规则对时间序列数据进行空间变换产生特征序列集,采用基本的特征提取方法对序列集进行特征提取得到最终的特征集合,模型的迭代次数和规则可由用户自己设定。以用户实际用电序列数据为例,对原始用电序列数据作为特征和该模型提取的特征进行分类比较,实验结果表明,在规则迭代模型下提取的特征,分类评价指标MAP最高可达到0.5389,对分类性能提升的增幅最高可达48.83%。
A feature extraction model of time series based on regular nested iteration was proposed,which deal with the difficulty of feature extraction and the lack of obvious features in time series classification.Conventional sequence feature extraction rules were regarded as the basic unit,a multi-level iterative rule was constructed to transform the time series data to generate the feature sequence set,and the basic feature extraction method was used to extract the feature from the sequence set to get the feature set.The number of iterations and the rules of the model could be set by the user.Taking the user actual electricity consumption sequence data as an example,the original time series was compared as a feature with the feature extracted from this model.Experimental results show that the features extracted using the rule iteration model,the classification evaluation index MAP can reach up to 0.5389,and the increase in classification performance is up to 48.83%.
A feature extraction model of time series based on regular nested iteration was proposed,which deal with the difficulty of feature extraction and the lack of obvious features in time series classification.Conventional sequence feature extraction rules were regarded as the basic unit,a multi-level iterative rule was constructed to transform the time series data to generate the feature sequence set,and the basic feature extraction method was used to extract the feature from the sequence set to get the feature set.The number of iterations and the rules of the model could be set by the user.Taking the user actual electricity consumption sequence data as an example,the original time series was compared as a feature with the feature extracted from this model.Experimental results show that the features extracted using the rule iteration model,the classification evaluation index MAP can reach up to 0.5389,and the increase in classification performance is up to 48.83%.
引文
[1]YUAN Jidong,WANG Zhihai.Review of time series representation and classification techniques[J].Computer Science,2015,42(3):1-7(in Chinese).[原继东,王志海.时间序列的表示与分类算法综述[J].计算机科学,2015,42(3):1-7.]
[2]Harvey AC,Shephard N.10 structural time series models[J].Handbook of Statistics,2015,11(5):261-302.
[3]Chernick MR.Wavelet methods for time series analysis[J].Cambridge University Press Cambridge,2016(43):491.
[4]Ho CC,Ting CY.Time series analysis and forecasting of dengue using open data[C]//4th International Visual Informatics Conference,2015:17-19.
[5]Kulkarni C,Mengshoel OJ,Basak A,et al.Optimizing the decomposition of soil moisture time-series data using genetic algorithms[C]//AGU Fall Meeting,2015.
[6]Zhu K.Bootstrapping the portmanteau tests in weak auto-regressive moving average models[J].Journal of the Royal Statistical Society,2016,78(2):463-485.
[7]Shao Q,Yang L.Oracally efficient estimation and consistent model selection for auto-regressive moving average time series with trend[J].Journal of the Royal Statistical Society,2016,79(2):507-524.
[8]Greff K,Srivastava RK,Koutnik J,et al.LSTM:A search space odyssey[J].IEEE Transactions on Neural Networks&Learning Systems,2016,28(10):2222-2232.
[9]ZHANG Yuhang,QIU Caiming,HE Xing,et al.A shortterm load forecasting based on LSTM neural network[J].Electric Power Information&Communication Technology,2017,15(9):19-25(in Chinese).[张宇航,邱才明,贺兴,等.一种基于LSTM神经网络的短期用电负荷预测方法[J].电力信息与通信技术,2017,15(9):19-25.]
[10]Jia G,Lu Y,Lu W,et al.Verification method for Chinese aviation radiotelephony readbacks based on LSTM-RNN[J].Electronics Letters,2017,53(6):401-403.
[11]Xiang Y.Identifying high quality document-summary pairs through text matching[J].Information,2017,8(2):64.
[12]Zhao F,Feng J,Jian Z,et al.Robust LSTM-autoencoders for face de-occlusion in the wild[J].IEEE Transactions on Image Processing,2018,27(2):778-790.
[2]Harvey AC,Shephard N.10 structural time series models[J].Handbook of Statistics,2015,11(5):261-302.
[3]Chernick MR.Wavelet methods for time series analysis[J].Cambridge University Press Cambridge,2016(43):491.
[4]Ho CC,Ting CY.Time series analysis and forecasting of dengue using open data[C]//4th International Visual Informatics Conference,2015:17-19.
[5]Kulkarni C,Mengshoel OJ,Basak A,et al.Optimizing the decomposition of soil moisture time-series data using genetic algorithms[C]//AGU Fall Meeting,2015.
[6]Zhu K.Bootstrapping the portmanteau tests in weak auto-regressive moving average models[J].Journal of the Royal Statistical Society,2016,78(2):463-485.
[7]Shao Q,Yang L.Oracally efficient estimation and consistent model selection for auto-regressive moving average time series with trend[J].Journal of the Royal Statistical Society,2016,79(2):507-524.
[8]Greff K,Srivastava RK,Koutnik J,et al.LSTM:A search space odyssey[J].IEEE Transactions on Neural Networks&Learning Systems,2016,28(10):2222-2232.
[9]ZHANG Yuhang,QIU Caiming,HE Xing,et al.A shortterm load forecasting based on LSTM neural network[J].Electric Power Information&Communication Technology,2017,15(9):19-25(in Chinese).[张宇航,邱才明,贺兴,等.一种基于LSTM神经网络的短期用电负荷预测方法[J].电力信息与通信技术,2017,15(9):19-25.]
[10]Jia G,Lu Y,Lu W,et al.Verification method for Chinese aviation radiotelephony readbacks based on LSTM-RNN[J].Electronics Letters,2017,53(6):401-403.
[11]Xiang Y.Identifying high quality document-summary pairs through text matching[J].Information,2017,8(2):64.
[12]Zhao F,Feng J,Jian Z,et al.Robust LSTM-autoencoders for face de-occlusion in the wild[J].IEEE Transactions on Image Processing,2018,27(2):778-790.