基于时间序列的混合神经网络数据融合算法

张巧灵; 高淑萍; 何迪; 程孟菲

doi:10.21656/1000-0887.410056

基于时间序列的混合神经网络数据融合算法

doi: 10.21656/1000-0887.410056

张巧灵(1992—), 女, 硕士生(E-mail: zql1048903596@163.com);高淑萍(1963—), 女, 教授, 博士, 硕士生导师(通讯作者. E-mail: gaosp@mail.xidian.edu.cn).

基金项目: 国家自然科学基金(91338115);111引智计划(B08038)

详细信息

中图分类号: TP391|TP181
计量
- 文章访问数: 1282
- HTML全文浏览量: 220
- PDF下载量: 270
- 被引次数: 0
出版历程
- 收稿日期: 2020-02-19
- 修回日期: 2020-06-27
- 刊出日期: 2021-01-01

A Hybrid Neural Network Data Fusion Algorithm Based on Time Series

¹School of Mathematics and Statistics, Xidian University,Xi’an 710126, P.R.China;²School of Telecommunications Engineering, Xidian University,Xi’an 710071, P.R.China

Funds: The National Natural Science Foundation of China（91338115）

摘要

摘要: 针对传统的数据融合算法对高噪声、大规模、数据结构复杂的时间序列数据融合性能较差的问题,该文提出了一种混合神经网络的数据融合算法(即SCLG算法).SCLG算法的思想是首先利用奇异谱分析算法对数据分解重构以达到去噪的目的;其次,通过深层卷积神经网络提取数据的空间特征和短期时间特征;然后,利用长短期记忆(LSTM)网络和门控循环单元(GRU)网络双层网络，进一步深度提取数据时间维度上的特征;最后,利用全连接网络综合主要信息输出最终的决策.通过SP&500和AQI数据集上的实验结果表明，该算法在融合性能及稳定性方面均优于DCNN、CNNLSTM、FDL数据融合算法.
- 数据融合 /
- 时间序列 /
- 奇异谱分析 /
- 混合神经网络 /
- 特征提取
Abstract: For traditional data fusion algorithms, the fusion performance of high-noise, large-scale and complex-structure time series data is poor. A hybrid neural network data fusion algorithm (i.e. the SCLG algorithm) was proposed to solve this problem. Firstly, the time series data were decomposed and reconstructed with the singular spectrum analysis algorithm to eliminate noise. Secondly, the spatial and short-term characteristics of the data were extracted by means of the deep convolutional neural network. Thirdly, the long short-term memory neural network and the gated recurrent unit neural network were introduced to extract data features in the time dimension. Finally, the fully connected layer was applied to integrate the main information and output the final decision. The experimental results from the SP&500 and AQI data sets show that, the proposed algorithm is superior to DCNN, CNN-LSTM and FDL in terms of fusion performance and stability.
- data fusion /
- time series /
- singular spectrum analysis /
- hybrid neural network /
- feature extraction

HTML全文

参考文献(24)

[1]	孙友强. 时间序列数据挖掘中的维数约简与预测方法研究[D]. 博士学位论文. 合肥: 中国科学技术大学, 2014.(SUN Youqiang. Research on dimensionality reduction and prediction methods in time series data ming[D]. PhD Thesis. Hefei: University of Science and Technology of China, 2014.(in Chinese))
[2]	王晓锋. 时间序列数据挖掘在医疗领域的应用[J]. 软件导刊, 2011,10(5): 123-124.(WANG Xiaofeng. Application of time series data mining in the medical field[J]. Software Guide,2011,10(5): 123-124.(in Chinese))
[3]	黄超, 龚惠群. 金融领域时间序列挖掘技术研究[J]. 东南大学学报(哲学社会科学版), 2007,9(5): 36-39.(HUANG Chao, GONG Huiqun. A study of time series mining technology in financial field[J]. Journal of Southeast University(Philosophy and Social Science),2007,9(5): 36-39.(in Chinese))
[4]	彭跃华, 于江龙. 非线性时间序列分析在气候中的应用研究进展[J]. 气象, 2009,32(10): 3-7.(PENG Yuehua, YU Jianglong. Research advances of nonlinear time series analysis applying in climatology[J]. Meteorological Monthly,2009,32(10): 3-7.(in Chinese))
[5]	孟小峰, 杜志娟. 大数据融合研究: 问题与挑战[J]. 计算机研究与发展, 2016,53(2): 231-246.(MENG Xiaofeng, DU Zhijuan. Research on the big data fusion: issues and challenges[J]. Journal of Computer Research and Development,2016,53(2): 231-246.(in Chinese))
[6]	何绪飞, 艾剑良, 宋智桃. 多元数据融合在无人机结构健康监测中的应用[J]. 应用数学和力学, 2018,39(4): 395-402.(HE Xufei, AI Jianliang, SONG Zhitao. Multi-Source data fusion for health monitoring of unmanned aerial vehicle structures[J]. Applied Mathematics and Mechanics,2018,39(4): 395-402.(in Chinese))
[7]	LECUN Y, BENGIO Y, HINTON G. Deeplearning[J]. Nature,2015,521: 436-444.
[8]	SHEN F, CHAO J, ZHAO J. Forecasting exchange rate using deep belief networks and conjugate gradient method[J]. Neurocomputing,2015,167: 243-253.
[9]	KIM T Y, CHO S B. Predicting residential energy consumption using CNN-LSTM neural networks[J]. Energy,2019,182: 72-81.
[10]	MA Q L, TIAN S, WEI J, et al. Attention-based spatio-temporal dependence learning network[J]. Information Sciences,2019,503: 92-108.
[11]	KARIM F, MAJUMDAR S, DARABI H, et al. Multivariate LSTM-FCNs for time series classification[J]. Neural Networks,2019,116: 237-245.
[12]	KARIM F, MAJUMDAR S, DARABI H, et al. LSTM fully convolutional networks for time series classification[J]. IEEE Access,2017,6(99): 1662-1669.
[13]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . 2018: 7132-7141.
[14]	JING L, WANG T, ZHAO M, et al. An adaptive multi-sensor data fusion method based on deep convolutional neural networks for fault diagnosis of planetary gearbox[J]. Sensors,2017,17(3): 414. DOI: 10.3390/s17020414.
[15]	GU Y L, LU W Q, QIN L Q, et al. Short-term prediction of lane-level traffic speeds: a fusion deep learning model[J]. Transportation Research Part C: Emerging Technologies,2019,106: 1-16.
[16]	HASSANI H, MAHMOUDVAND R, ZOKAEI M, et al. On the separability between signal and noise in singular spectrum analysis[J]. Fluctuation and Noise Letters,2012,11(2): 1250014.
[17]	ZHIGLJAVSKY A. Singular spectrum analysis for time series: introduction to this special issue[J]. Statistics and Its Interface,2010,3(3): 255-258.
[18]	俞颂华. 卷积神经网络的发展与应用综述[J]. 信息通信, 2019(2): 39-43.(YU Songhua. The development and application of convolutional neural networks[J]. Information & Communications,2019(2): 39-43.(in Chinese))
[19]	MALEK S, MELGANI F, BAZI Y. One-dimensional convolutional neural networks for spectroscopic signal regression[J]. Journal of Chemometrics,2018,32(5): e2977. DOI: 10.1002/cem.2977.
[20]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation,1997,9(8): 1735-1780.
[21]	HUANG C J, KUO P H. A deep CNN-LSTM model for particulate matter (PM2.5) forecasting in smart cities[J]. Sensors,2018,18(7): 2220.
[22]	SAGHEER A, KOTB M. Time series forecasting of petroleum production using deep LSTM recurrent networks[J]. Neurocomputing,2019,323: 203-213.
[23]	ABDEL-HAMID O, DENG L, YU D. Exploring convolutional neural network structures and optimization techniques for speech recognition[C]//Interspeech 2013. Lyon, France, 2013: 1173-1175.
[24]	WEI B, JUN Y, RAO Y L, et al. A deep learning framework for financial time series using stacked autoencoders and long-short term memory[J].Plos One,2017, 12(7): 1-24.