一种多视图深度融合的连续性缺失补全方法

doi:10.19665/j.issn1001-2400.2019.02.011

西安电子科技大学学报 ›› 2019, Vol. 46 ›› Issue (2): 61-68.doi: 10.19665/j.issn1001-2400.2019.02.011

一种多视图深度融合的连续性缺失补全方法

毛莺池¹,张建华¹,陈豪^2,³

^1. 河海大学计算机与信息学院,江苏南京 211100
^2. 河海大学水利水电学院,江苏南京 210098
^3. 华能澜沧江水电股份有限公司,云南昆明 650214

收稿日期:2018-09-21 出版日期:2019-04-20 发布日期:2019-04-20
作者简介:毛莺池(1976-),女,教授,博士,E-mail:yingchimao@hhu.edu.cn.
基金资助:
"十三五"国家重点研发计划(2018YFC0407105);"十三五"国家重点研发计划(2018YFC0407905);"十三五"国家重点研发计划(2016YFC0400910);华能集团重点研发课题(HNKJ17-21);中央高校基本科研业务费专项资金(2017B16814);中央高校基本科研业务费专项资金(2017B16814)

Successive missing completion based on deep fusion from multiple views

MAO Yingchi¹,ZHANG Jianhua¹,CHEN Hao^2,³

^1. College of Computer and Information, Hohai University, Nanjing 211100, China
^2. College of Water Resources and Hydropower, Hohai University, Nanjing 210098, China
^3. Huaneng Lancang River Hydropower Co., Ltd., Kunming 650214, China

Received:2018-09-21 Online:2019-04-20 Published:2019-04-20

摘要/Abstract

摘要：

针对现有连续性缺失补全方法的不足,建立了一种多视图深度融合的连续性缺失补全方法。该方法采用反转距离加权插值、双向简单指数平滑、用户协同过滤、能量扩散协同过滤及文本嵌套的方法,分别得到时空和语义缺失数据补全中间结果;构造了神经网络模型融合跨时空和语义视图中的互补异构信息,完成连续性缺失补全。实验表明,该方法补全连续性缺失不但效率高,而且比时空多视图补全在平均绝对误差与平均相对误差上分别降低7%和22%,具备普适性且适用于相关时空连续性缺失序列补全领域。

关键词: 连续性缺失数据补全, 人工神经网络, 时空, 深度融合

Abstract:

Aiming at the shortcomings of existing successive missing complement methods, a successive missing data completion method for multi-view depth fusion is established. The method adopts inverse distance weighted interpolation, bidirectional simple exponential smoothing, user-based collaborative filtering, the collaborative filtering based on mass diffusion and structure embeddings, to obtain intermediate results of five missing data in spatiotemporal and semantic respectively; then, this method constructs a neural network model that combines complementary heterogeneous information across time and space and semantic views to achieve successive missing completion. Experimental results show that the method is universally applicable to the field of Spatial-Temporal successive missing sequence completion and, that it not only achieves a high efficiency, but also reduces the mean absolute error and the mean relative error by 7% and 22%, respectively, compared with the Spatial-Temporal Multi-view completion method.

Key words: successive missing completion, artificial neural network, spatial and temporal, deep integration

中图分类号:

TP311

毛莺池,张建华,陈豪. 一种多视图深度融合的连续性缺失补全方法[J]. 西安电子科技大学学报, 2019, 46(2): 61-68.

MAO Yingchi,ZHANG Jianhua,CHEN Hao. Successive missing completion based on deep fusion from multiple views[J]. Journal of Xidian University, 2019, 46(2): 61-68.

图/表 6

图1

表1

图2

表2

图3

图4

参考文献 14

[1]	TONG S, FU Y, LING H , et al. Gait Identification by Joint Spatial-temporal Feature[C]// Lecture Notes in Computer Science: 10568. Heidelberg: Springer Verlag, 2017: 457-465.
[2]	PICCOLI B, HAN K, FRIESZ T L , et al. Second-order Models and Traffic Data from Mobile Sensors[J]. Transportation Research Part C: Emerging Technologies, 2015,52:32-56. doi: 10.1016/j.trc.2014.12.013
[3]	YAO H X, WU F, KE J T , et al. Deep Multi-view Spatial-temporal Network for Taxi Demand Prediction[CP/OL]. [2018-09-03]. https://arxiv.org/abs/1802. 08714.
[4]	CHEN L, CONG G, CAO X , et al. Temporal Spatial-keyword Top-k Publish/Subscribe[C]// Proceedings of the 2015 IEEE 31st International Conference on Data Engineering.Piscataway: IEEE, 2015: 255-266.
[5]	YI X W, YU Z, ZHANG J B , et al. ST-MVL: Filling Missing Values in GEO-sensory Time Series Data[C]// Proceedings of the 2016 25th International Joint Conference on Artificial Intelligence. New York: International Joint Conferences on Artificial Intelligence, 2016: 2704-2710.
[6]	KARYPIS G . Recent Advances in Recommender Systems: Sets, Local Models, Coverage, and Errors[C]// Companion Proceedings of the The Web Conference 2018. New York: ACM, 2018: 1369.
[7]	VARATHARAJAN R, MANOGARAN G, PRIYAN M K , et al. Visual Analysis of Geospatial Habitat Suitability Model Based on Inverse Distance Weighting with Paired Comparison Analysis[J]. Multimedia Tools and Applications, 2018,77(14):17573-17593. doi: 10.1007/s11042-017-4768-9
[8]	HOU L, LIU K, LIU J , et al. Solving the Stability Accuracy Diversity Dilemma of Recommender Systems[J]. Physica A: Statistical Mechanics and Its Applications, 2017,468:415-424. doi: 10.1016/j.physa.2016.10.083
[9]	NIE D C, AN Y H, DONG Q , et al. Information Filtering via Balanced Diffusion on Bipartite Networks[J]. Physica A: Statistical Mechanics and Its Applications, 2015,421:44-53. doi: 10.1016/j.physa.2014.11.018
[10]	QIU X, MAO Q, TANG Y , et al. Reversed Graph Embedding Resolves Complex Single-cell Trajectories[J]. Nature Methods, 2017,14(10):979-982. doi: 10.1038/nmeth.4402 pmid: 28825705
[11]	TANG J, QU M, WANG M , et al. Line: Large-scale Information Network Embedding[C]// Proceedings of the 2015 24th International Conference on World Wide Web. New York: ACM, 2015: 1067-1077.
[12]	LIN G, SHEN W . Research on Convolutional Neural Network Based on Improved Relu Piecewise Activation Function[J]. Procedia Computer Science, 2018,131:977-984. doi: 10.1016/j.procs.2018.04.239
[13]	SZEGEDY C, IOFFE S, VANHOUCKE V , et al. Inception-v4, Inception-resnet and the Impact of Residual Connections on Learning[C]// Proceedings of the 2017 31st AAAI Conference on Artificial Intelligence. Palo Alto: AAAI press, 2017: 4278-4284.
[14]	张兴国, 刘晓磊, 李靖 , 等. BP神经网络下的限速交通标志实时检测识别[J]. 西安电子科技大学学报, 2018,45(5):136-142. doi: 10.3969/j.issn.1001-2400.2018.05.022
	ZHANG Xingguo, LIU Xiaolei, LI Jing , et al. Real-time Detection and Identification of Speed Limit Traffic Signs under the BP Neural Network[J]. Journal of Xidian University, 2018,45(5):136-142. doi: 10.3969/j.issn.1001-2400.2018.05.022

方法	连续性空间缺失			连续性时间缺失
方法	MAE	MSE	MRE	MAE	MSE	MRE
ARMA	25.34	29.11	0.35
SARIMA	21.36	27.79	0.32	25.42	30.24	0.56
Kriging				15.31	27.85	0.28
DEMS	16.890	25.140	0.2840	12.950	23.970	0.277
ST-KNN	17.55	28.87	0.31	12.32	22.59	0.27
CF	16.780	24.980	0.285	12.760	21.560	0.274
ST-MVL	14.620	21.620	0.256	11.510	19.470	0.230
DNN-MVL	13.540	19.450	0.190	9.750	18.380	0.156

方法	连续性空间缺失			连续性时间缺失
方法	MAE	MSE	MRE	MAE	MSE	MRE
全局	IDW			12.64	21.12	0.242
SES	16.31	23.68	0.32
IDW+SES	16.31	23.68	0.32	12.64	21.12	0.242
空间:IDW+UCF				12.730	21.340	0.245
时间:SES+ICF	15.980	22.140	0.294
时间:SES+MD-CF	15.12	21.92	0.27
ST-MVL	14.62	21.62	0.256	11.51	19.47	0.21
DNN-MVL	13.540	19.450	0.190	9.750	18.380	0.156

一种多视图深度融合的连续性缺失补全方法

Successive missing completion based on deep fusion from multiple views

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 6

参考文献 14

相关文章 7

Metrics

本文评价

推荐阅读 10

[1]	许缓缓,李洪梅,李富余,孙学梅. 一种时空卷积的步态识别方法[J]. 西安电子科技大学学报, 2021, 48(4): 144-150.
[2]	齐小刚, 马久龙, 刘立芳. 保证高可靠度和低传输开销的DTN拓扑控制[J]. 西安电子科技大学学报, 2018, 45(5): 7-12.
[3]	张兴国;刘晓磊;李靖;王环东. BP神经网络下的限速交通标志实时检测识别[J]. 西安电子科技大学学报, 2018, 45(5): 136-142.
[4]	秦国栋;陈伯孝;陈多芳. 大入射余角下海杂波仿真及其时空相关性分析 [J]. J4, 2008, 35(4): 600-604.
[5]	刘洋;李玉山. 二维时空熵运动检测中的自适应门限技术 [J]. J4, 2007, 34(4): 647-653.
[6]	郑建国;刘芳;焦李成. 自适应子波神经网络数据挖掘方法[J]. J4, 2002, 29(4): 474-477.
[7]	暂时无作者信息. 基于神经网络的高校科研管理系统[J]. J4, 1996, 23(2): 0-0.