基于时间窗长短期记忆模型分析新型冠状病毒肺炎疫情

程宁; 丁长松; 高婉卿; 刘佳俊

doi:10.16462/j.cnki.zhjbkz.2021.05.015

基于时间窗长短期记忆模型分析新型冠状病毒肺炎疫情

doi: 10.16462/j.cnki.zhjbkz.2021.05.015

程宁¹,
丁长松^{1, 2, ,},
高婉卿¹,
刘佳俊¹

1.
410208 长沙，湖南中医药大学信息科学与工程学院
2.
410208 长沙，湖南省中医药大数据分析实验室

基金项目:

国家重点研发计划 2017YFC1703306

湖南省重点研发计划 2017SK2111

湖南省自然基金 2018JJ2301

湖南省教育厅重点项目 18A227

湖南省中医药科研计划重点课题 2020002

详细信息

通讯作者:
丁长松，E-mail: dingcs1975@hnucm.edu.cn

中图分类号: R563.1;R181;TP183
计量
- 文章访问数: 706
- HTML全文浏览量: 225
- PDF下载量: 63
- 被引次数: 0
出版历程
- 收稿日期: 2020-08-18
- 修回日期: 2020-12-16
- 网络出版日期: 2021-06-16
- 刊出日期: 2021-05-10

Analysis of COVID-19 outbreak based on Window-Time-LSTM Model

1.
School of Informatics, Hunan University of Chinese Medicine, Changsha 410208, China
2.
Big Data Analysis Laboratory of Traditional Chinese Medicine in Hunan Province, Changsha 410208, China

Funds:

National Key Research and Development Plan 2017YFC1703306

Key R & D Plan of Hunan Province 2017SK2111

Hunan Provincial Natural Resources Foundation 2018JJ2301

Key Project of Hunan Education Department 18A227

Key Subject of Hunan TCM Research Program 2020002

More Information

Corresponding author: DING Chang-song, E-mail: dingcs1975@hnucm.edu.cn

摘要

摘要: 目的提出一种新的网络模型以解决COVID-19出现初期认知不够、检测能力受限以及潜伏期长等因素导致每天检测出的感染人数与真实感染人数存在差异的问题，并预测COVID-19疫情发展趋势。方法以历史时间窗检测出的数据为依据，将时间窗策略结合长短期记忆(long short term memory, LSTM)网络模型对世界各个地区的确诊人数进行预测分析。结果基于时间窗的LSTM网络模型与其他模型相比准确度较高。对时间窗宽度进行分析发现，当宽度取5时预测结果最接近真实数据，这与COVID-19的潜伏期普遍为3~7 d相吻合。结论该方法为COVID-19疫情发展以及其潜伏期的分析提供了参考。
- COVID-19 /
- 时间窗 /
- 长短期记忆网络
Abstract: Objective To solve the data difference between COVID-19 confirmed cases and actual number of COVID-19 infections, a new model is proposed to predict the spread of the disease. The data difference has been mainly caused by insufficient understanding in the early stage of transmission, limited detection capabilities and the long incubation period. Methods The historical data of the number of confirmed cases are analyzed based on Window-Time. A Long Short-Term Memory (LSTM) network model is combined with the Window-Time strategy to analyze and predict the actual number of infections according to data published of various regions in the world. Results The LSTM network model with Window-Time strategy has higher accuracy than other models. Tuning the width of the Window-Time to the width of 5, the prediction result shows that it is closest to the real actual number of infections, which is consistent with the incubation period of COVID-19 generally known as 3-7 days. Conclusion This method provides a reference for the analysis of the transmission rate of COVID-19 and the incubation period of the epidemic.
- COVID-19 /
- Window-Time /
- LSTM network

HTML全文

图 1 WT-LSTM网络模型

Figure 1. Window-Time-LSTM model

下载: 全尺寸图片幻灯片

图 2 模型预测结果

Figure 2. Prediction results of models

下载: 全尺寸图片幻灯片

表 1 时间窗宽度分析结果

Table 1. Window-Time width analysis results

宽度参数(n)	时间窗双向关联权重系数(a)	模型平均(RMSE)	模型平均(MAE)	模型平均(MAPE)
1	0.576 116	8 834.264	7 368.893	1.676
2	0.498 397	10 038.900	7 654.034	1.223
3	0.474 832	7 589.277	5 705.968	1.242
4	0.466 715	6 258.321	4 588.270	1.235
5	0.463 798	5 595.714	3 549.351	1.132
6	0.462 734	6 304.506	4 992.106	1.278
7	0.462 343	6 407.598	4 274.729	1.330
8	0.462 201	6 553.170	4 122.248	1.344
9	0.462 147	7 103.133	4 366.778	3.371
10	0.462 128	7 503.113	4 666.238	1.346
11	0.462 121	6 901.131	4 432.232	1.226
12	0.462 118	6 812.243	5 889.700	2.279
13	0.462 117	6 193.711	3 773.097	2.274
14	0.462 116	6 070.431	3 927.931	2.013

下载: 导出CSV

表 2 模型对比结果

Table 2. Comparison results of different models

模型	RMSE	MAE	MAPE
LSTM	475.839	308.708	0.244
Logistic	1 663.310	1 343.311	0.357
MLP	16 664.960	3 232.017	17.914
WT-LSTM	418.427	180.829	0.059
WT-Logistic	1 406.375	721.579	0.267
WT-MLP	14 541.510	2 243.632	15.321

下载: 导出CSV

表 3 WT-LSTM网络模型预测实验评价指标

Table 3. Evaluation index of experiments of WT-LSTM model

模型	RMSE	MAE	MAPE
中国湖南省	6.271	2.160	0.001
美国	16 935.450	15 415.163	2.424
中国广东预测浙江	11.602	6.361	0.023
意大利预测浙江	93.185	89.672	0.283
加拿大预测美国	21 470.885	17 144.790	1.346
中国北京预测美国	644 572.900	567 618.440	271.889

下载: 导出CSV

参考文献(15)

[1]	World Health Organization. Coronavirus Disease (COVID-19)[EB/OL]. (2020-11-20)[2020-11-20]. https://www.who.int/emergencies/diseases/novel-coronavirus-2019
[2]	Lauer SA, Grantz KH, Bi Q, et al. The incubation period of coronavirus disease 2019 (COVID-19) from publicly reported confirmed cases: estimation and application[J]. Ann Intern Med, 2020, 172(9): 577-582. DOI: 10.7326/M20-0504.
[3]	Sohrabi C, Alsafi Z, O'Neill N, et al. World Health Organization declares global emergency: A review of the 2019 novel coronavirus (COVID-19)[J]. Int J Surg, 2020, 76: 71-76. DOI: 10.1016/j.ijsu.2020.02.034.
[4]	Huang C, Wang Y, Li X, et al. Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China[J]. Lancet, 2020, 395(10223): 497-506. DOI: 10.1016/S0140-6736(20)30183-5.
[5]	Ying S, Li F, Geng X, et al. Spread and control of COVID-19 in China and their associations with population movement, public health emergency measures, and medical resources[J]. medRxiv, 2020, 2020.02.24.20027623. DOI: 10.1101/2020.02.24.20027623.
[6]	Li MY, Muldowney JS. Global stability for the SEIR model in epidemiology[J]. Math Biosci, 1995, 125(2): 155-164. DOI: 10.1016/0025-5564(95)92756-5.
[7]	梅文娟, 刘震, 朱静怡, 等. 新冠肺炎疫情极限IR实时预测模型[J]. 电子科技大学学报, 2020, 49(3): 362-368. DOI: 10.12178/1001-0548.2020063. Mei WJ, Liu Z, Zhu JY, et al. Extreme IR model for COVID-19 real-time forecasting[J]. Journal of University of Electronic Science and Technology of China, 2020, 49(3): 362-368. DOI: 10.12178/1001-0548.2020063.
[8]	Chintalapudi N, Battineni G, Amenta F. COVID-19 virus outbreak forecasting of registered and recovered cases after sixty day lockdown in Italy: A data driven model approach[J]. J Microbiol Immunol Infect, 2020, 53(3): 396-403. DOI: 10.1016/j.jmii.2020.04.004.
[9]	Yin R, Zhang Y, Zhou X, et al. Time series computational prediction of vaccines for influenza A H3N2 with recurrent neural networks[J]. J Bioinform Comput Biol, 2020, 18(1): 2040002. DOI: 10.1142/S0219720020400028.
[10]	王淑平, 杜敏, 罗建伟, 等. GM(1, 1)模型与LSTM神经网络在肺结核发病数预测中的应用[J]. 公共卫生与预防医学, 2019, 30(5): 11-14. doi: 10.3969/j.issn.1006-2483.2019.05.003 Wang SP, Du M, Luo JW, et al. Application of GM(1, 1) model and LSTM neural network in predicting the incidence of pulmonary tuberculosis[J]. J of Pub Health and Prev Med, 2019, 30(5): 11-14. DOI: 10.3969/j.issn.1006-2483.2019.05.003.
[11]	高秋菊, 周宇畅, 赵树青, 等. ARIMA乘积季节模型和LSTM深度神经网络对石家庄市手足口病疫情预测效果的比较[J]. 中华疾病控制杂志, 2020, 24(1): 73-78. DOI: 10.16462/j.cnki.zhjbkz.2020.01.015. Gao QJ, Zhou YC, Zhao SQ, et al. Comparison on predictive capacity of ARIMA model and LSTM model for incidence of hand, foot and mouth disease in Shijiazhuang[J]. Chin J Dis Control Prev, 2020, 24(1): 73-78. DOI: 10.16462/j.cnki.zhjbkz.2020.01.015.
[12]	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]. J Thorac Dis, 2020, 12(3): 165-174. DOI: 10.21037/jtd.2020.02.64.
[13]	Tomar A, Gupta N. Prediction for the spread of COVID-19 in India and effectiveness of preventive measures[J]. Sci Total Environ, 2020, 728: 138762. DOI: 10.1016/j.scitotenv.2020.138762.
[14]	Chimmula VKR, Zhang L. Time series forecasting of COVID-19 transmission in Canada using LSTM networks[J]. Chaos Solitons Fractals, 2020, 135: 109864. DOI: 10.1016/j.chaos.2020.109864.
[15]	Chatterjee A, Gerdes MW, Martinez SG. Statistical explorations and univariate timeseries analysis on COVID-19 datasets to understand the trend of disease spreading and death[J]. Sensors (Basel), 2020, 20(11): 3089. DOI: 10.3390/s20113089.