留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于时空注意力机制的新冠肺炎疫情预测模型

鲍昕 谭智一 鲍秉坤 徐常胜

鲍昕, 谭智一, 鲍秉坤, 等 . 基于时空注意力机制的新冠肺炎疫情预测模型[J]. 北京航空航天大学学报, 2022, 48(8): 1495-1504. doi: 10.13700/j.bh.1001-5965.2021.0535
引用本文: 鲍昕, 谭智一, 鲍秉坤, 等 . 基于时空注意力机制的新冠肺炎疫情预测模型[J]. 北京航空航天大学学报, 2022, 48(8): 1495-1504. doi: 10.13700/j.bh.1001-5965.2021.0535
BAO Xin, TAN Zhiyi, BAO Bingkun, et al. Prediction model of COVID-19 based on spatiotemporal attention mechanism[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(8): 1495-1504. doi: 10.13700/j.bh.1001-5965.2021.0535(in Chinese)
Citation: BAO Xin, TAN Zhiyi, BAO Bingkun, et al. Prediction model of COVID-19 based on spatiotemporal attention mechanism[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(8): 1495-1504. doi: 10.13700/j.bh.1001-5965.2021.0535(in Chinese)

基于时空注意力机制的新冠肺炎疫情预测模型

doi: 10.13700/j.bh.1001-5965.2021.0535
基金项目: 

国家重点研发计划 2020AAA0106200

国家自然科学基金 6193000388

国家自然科学基金 61872424

江苏省自然科学基金 BK20200037

江苏省自然科学基金 BK20210595

详细信息
    通讯作者:

    谭智一, E-mail: tzy@njupt.edu.cn

  • 中图分类号: TP183

Prediction model of COVID-19 based on spatiotemporal attention mechanism

Funds: 

National Key R & D Program of China 2020AAA0106200

National Natural Science Foundation of China 6193000388

National Natural Science Foundation of China 61872424

Natural Science Foundation of Jiangsu Province BK20200037

Natural Science Foundation of Jiangsu Province BK20210595

More Information
  • 摘要:

    新冠肺炎疫情持续蔓延给人类社会带来深远影响,准确预测各地区的病毒传播趋势对防控疫情而言至关重要。现有研究主要基于传统的时序预测模型和传染病模型,鲜有考虑疫情地区关联复杂和时序依赖性强的特点,限制了其疫情预测的性能。为此,针对新冠肺炎疫情的预测任务,提出了一种时空注意力驱动的自编码器框架。通过引入空间注意力机制捕捉病毒感染序列间的动态空间关联性,利用时间注意力机制挖掘病毒感染序列中复杂的时序依赖性,以此实现对不同地区的新冠肺炎病毒传播趋势的准确预测。在模型的编码器端,融合空间注意力机制的长短期记忆(LSTM)网络,关联目标地区与其他地区的病毒感染序列,提取该区域近期新冠肺炎疫情的时序特征。在模型的解码器端,将时间注意力机制引入基于LSTM网络的解码器中,通过捕捉病毒感染序列的时序依赖性推测未来的新冠肺炎疫情趋势变化。在多个公开的新冠肺炎疫情数据集上对所提模型进行验证,实验结果表明:所提模型的预测性能超越了LSTM等模型;在公开的欧洲部分国家新冠肺炎疫情数据集上,预测误差指标RMSE和MAE分别降低了22.3%和25.0%,在中国部分省级单位新冠肺炎疫情数据集上,RMSE和MAE分别降低了10.1%和10.4%。

     

  • 图 1  基于时空注意力机制的新冠肺炎疫情预测框架

    Figure 1.  COVID-19 prediction framework based on spatial temporal attention mechanism

    图 2  欧洲部分国家新增确诊人数曲线

    Figure 2.  Curves of new cases in some European countries

    图 3  输入序列长度对模型性能的影响曲线

    Figure 3.  Influence curves of input sequence length on model performance

    图 4  输出序列长度对模型性能的影响曲线

    Figure 4.  Influence curves of output sequence length on model performance

    表  1  数据集描述

    Table  1.   Dataset description

    数据集 预测目标 时间范围 训练集时间划分 测试集时间划分
    欧洲部分国家新冠肺炎疫情数据集 新增确诊人数 2020/1/24—2021/3/22 2020/1/24—2020/12/24 2020/12/25—2021/3/22
    中国部分省级单位新冠肺炎疫情数据集 累计确诊人数 2020/1/22—2020/10/15 2020/1/22—2020/8/22 2020/8/23—2020/10/15
    下载: 导出CSV

    表  2  不同模型在欧洲部分国家新冠肺炎疫情数据集上的预测性能比较(部分数据)

    Table  2.   Prediction performance comparison among different methods in COVID-19 epidemic dataset of some European countries (partial data)

    国家 LSTM GRU Seq2Seq
    RMSE MAE MAPE RMSE MAE MAPE RMSE MAE MAPE
    阿尔巴尼亚 703.92 691.41 0.86 701.24 688.64 0.86 662.86 650.38 0.81
    丹麦 938.59 903.71 0.86 929.71 894.01 0.87 901.43 860.9 0.81
    罗马尼亚 3 188.53 3 075.03 0.97 3 184.12 3 070.45 0.97 3 157.08 3 042.02 0.95
    奥地利 1 786.8 1 736.26 0.94 1 781.35 1 729.26 0.94 1 750.72 1 696.74 0.91
    希腊 1 036.59 980.39 0.88 1 032.73 975.95 0.88 1 002.07 943.65 0.83
    德国 13 128.95 12 358.32 0.99 13 125.13 12 354.14 0.99 13 102.8 12 325.16 0.98
    英国 26 716.02 26 339.62 1.68 26 711.37 26 335.18 1.62 26 705.26 26 295.7 0.99
    法国 20 001.07 3 911.05 0.99 19 997.16 18 644.59 0.99 19 951.54 18 591.85 0.98
    平均 4 112.48 3 911.05 1.68 4 105.65 3 903.92 1.62 4 075.76 3 868.85 1.98
    国家 T-GCN STAEP(本文模型)
    RMSE MAE MAPE RMSE MAE MAPE
    阿尔巴尼亚 798.227 3 764.024 9 0.96 565.97 552.767 0.65
    丹麦 1 013.16 819.04 0.96 754.97 679.68 0.43
    罗马尼亚 3 271.24 3 088.82 0.98 1 563.26 1 398.67 0.49
    奥地利 1 868.26 1 769.33 0.97 1 064.07 997.13 0.51
    希腊 1 252.66 1 063.47 0.96 740.09 675.71 0.44
    德国 13 385.03 11 823.22 0.99 10 243.3 9 217.24 0.62
    英国 27 062.34 25 433.32 0.99 23 676.71 23 183.08 0.75
    法国 20 541.78 19 158.85 0.99 17 127.35 15 549.89 0.72
    平均 4 383.86 3 922.16 1.12 3 166.93 2 902.36 0.92
    下载: 导出CSV

    表  3  不同模型在中国部分省级单位新冠肺炎疫情数据集上的预测性能比较(部分数据)

    Table  3.   Prediction performance comparison among different methods in COVID-19 epidemic dataset of some Chinese provinces (partial data)

    省(直辖市、特别行政区) LSTM GRU Seq2Seq
    RMSE MAE MAPE RMSE MAE MAPE RMSE MAE MAPE
    安徽省 910.56 911.04 0.91 909.37 909.06 0.91 829.33 829.72 0.83
    湖南省 938.96 938.78 0.92 937.39 937.31 0.92 857.86 858.03 0.84
    河南省 1 198.18 1 198.53 0.93 1 196.62 1 196.84 0.93 1 117.44 1 116.79 0.87
    湖北省 68 058.96 68 059.28 0.99 68 057.51 68 057.81 0.99 67 977.7 67 976.45 0.99
    江西省 854.42 855.05 0.91 853.39 853.39 0.91 773.9 773.36 0.82
    重庆 504.077 503.97 0.86 502.49 502.63 0.86 422.71 423.7 0.83
    香港 4 886.68 4 886.25 0.98 4 885.07 4 884.77 0.98 4 805.51 4 804.7 0.96
    江苏省 585.54 585.47 0.87 584.08 584.21 0.87 504.39 504.22 0.75
    平均 2 661.7 2 661.39 0.78 2 660.65 2 660.38 0.92 2 592.01 2 591.38 0.59
    省(直辖市、特别行政区) T-GCN STAEP(本文模型)
    RMSE MAE MAPE RMSE MAE MAPE
    安徽省 953.89 942.18 0.95 106.16 86.22 0.16
    湖南省 981.89 969.9 0.95 122.41 99.24 0.16
    河南省 1 241.01 1 226.61 0.96 212.95 186.52 0.1
    湖北省 68 101.24 68 086.82 0.99 54 061.26 54 059.26 0.8
    江西省 897.24 882.82 0.95 96.81 77.65 0.15
    重庆 547.06 535.36 0.92 72.21 58.87 0.14
    香港 4 981.62 4 970.01 0.99 7 666.55 7 536.1 1.51
    江苏省 628.56 616.86 0.93 79.7 64.47 0.23
    平均 2 231.44 2 222.01 1.82 2 006.32 1 990.52 0.51
    下载: 导出CSV
  • [1] LEI S, JIANG F, SU W, et al. Clinical characteristics and outcomes of patients undergoing surgeries during the incubation period of COVID-19 infection[J]. EClinicalMedicine, 2020, 21: 100331.
    [2] MASTAKOURI A A, SCHÖLKOPF B. Causal analysis of Covid-19 spread in Germany[EB/OL]. (2020-08-03)[2021-09-01]. https://arxiv.org/abs/2007.11896v1.
    [3] XIAO C, ZHOU J, HUANG J, et al. C-Watcher: A framework for early detection of high-risk neighborhoods ahead of COVID-19 outbreak[EB/OL]. (2021-03-01)[2021-09-01]. https://arxiv.org/abs/2012.12169v2.
    [4] YANG Z, ZENG Z, WANG K, et al. Modified SEIR and AI prediction of the epidemics trend of COVID-19 in China under public health interventions[J]. Journal of Thoracic Disease, 2020, 12(3): 165-174. doi: 10.21037/jtd.2020.02.64
    [5] ACHTERBERG M A, PRASSE B, MA L, et al. Comparing the accuracy of several network-based COVID-19 prediction algorithms[J]. International Journal of Forecasting, 2022, 38(2): 489-504. doi: 10.1016/j.ijforecast.2020.10.001
    [6] GUPTA R, PANDEY G, CHAUDHARY P, et al. Machine learning models for government to predict COVID-19 outbreak[J]. Digital Government: Research and Practice, 2020, 1(4): 26.
    [7] LIU X X, FONG S. Towards a realistic model for simulating spread of infectious COVID-19 disease[C]//Proceedings of the 2020 the 4th International Conference on Big Data and Internet of Things. New York: ACM, 2020: 96-101.
    [8] KUMAR M, GUPTA S, KUMAR K, et al. Spreading of COVID-19 in India, Italy, Japan, Spain, UK, US: A prediction using ARIMA and LSTM model[J]. Digital Government: Research and Practice, 2020, 1(4): 24.
    [9] WANG P, ZHENG X, LI J, et al. Prediction of epidemic trends in COVID-19 with logistic model and machine learning technics[J]. Chaos Solitons & Fractals, 2020, 139: 110058.
    [10] NGUYEN K L P, CHEN H W, YANG C T, et al. Implementation of a respiratory disease forecasting model using LSTM for central Taiwan[M]//KIM K J, KIM H Y. Information science and applications. Berlin: Springer, 2020: 441-450.
    [11] RAUF H T, LALI M I U, KHAN M A, et al. Time series forecasting of COVID-19 transmission in Asia pacific countries using deep neural networks[J/OL]. Personal and Ubiquitous Computing, 2021(2021-01-10)[2021-09-01]. https://doi.org/10.1007/s00779-020-01494-0.
    [12] WANG J, MCMICHAEL A J, MENG B, et al. Spatial dynamics of an epidemic of severe acute respiratory syndrome in an urban area[J]. Bulletin of the World Health Organization, 2006, 84(12): 965-968. doi: 10.2471/BLT.06.030247
    [13] BISWAS K, KHALEQUE A, SEN P. Covid-19 spread: Reproduction of data and prediction using a SIR model on Euclidean network[EB/OL]. (2020-03-16)[2021-09-01]. https://arxiv.org/abs/2003.07063.
    [14] GHAMIZI S, RWEMALIKA R, CORDY M, et al. Data-driven simulation and optimization for Covid-19 exit strategies[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2020: 3434-3442.
    [15] WAN H, CUI J A, YANG G J. Risk estimation and prediction of the transmission of coronavirus disease-2019 (COVID-19) in the mainland of China excluding Hubei province[J]. Infectious Diseases of Poverty, 2020, 9: 116. doi: 10.1186/s40249-020-00683-6
    [16] ZHENG N, DU S, WANG J, et al. Predicting COVID-19 in China using hybrid AI model[J]. IEEE Transactions on Cybernetics, 2020, 50(7): 2891-2904. doi: 10.1109/TCYB.2020.2990162
    [17] 黎维, 陶蔚, 周星宇, 等. 时空序列预测方法综述[J]. 计算机应用研究, 2020, 37(10): 2881-2888. https://www.cnki.com.cn/Article/CJFDTOTAL-JSYJ202010001.htm

    LI W, TAO W, ZHOU X Y, et al. Survey of spatio-temporal sequence prediction methods[J]. Application Research of Computers, 2020, 37(10): 2881-2888(in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-JSYJ202010001.htm
    [18] SHI X, CHEN Z, WANG H, et al. Convolutional LSTM network: A machine learning approach for precipitation nowcasting[M]. Cambridge: MIT Press, 2015.
    [19] LIANG Y, KE S, ZHANG J, et al. GeoMAN: Multi-level attention networks for geo-sensory time series prediction[C]//Proceedings of the 27th International Joint Conference on Artificial Intelligence. New York: ACM, 2018: 3428-3434.
    [20] YAO H, WU F, KE J, et al. Deep multi-view spatial-temporal network for taxi demand prediction[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2018: 2588-2595.
    [21] WANG X, MA Y, WANG Y, et al. Traffic flow prediction via spatial temporal graph neural network[C]//Proceedings of the Web Conference 2020. New York: ACM, 2020: 1082-1092.
    [22] SUTSKEVER I, VINYALS O, LE Q V. Sequence to sequence learning with neural networks[C]//Advances in Neural Information Processing Systems, 2014: 3104-3112.
    [23] ZHAO L, SONG Y, ZHANG C, et al. T-GCN: A temporal graph convolutional network for traffic prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 21(9): 3848-3858.
  • 加载中
图(4) / 表(3)
计量
  • 文章访问数:  86
  • HTML全文浏览量:  22
  • PDF下载量:  18
  • 被引次数: 0
出版历程
  • 收稿日期:  2021-09-07
  • 录用日期:  2021-09-17
  • 刊出日期:  2021-11-02

目录

    /

    返回文章
    返回
    常见问答