基于SEAIQR模型与Dropout-LSTM模型的西安市COVID-19疫情趋势预测  

Prediction of COVID-19 Epidemic in Xi′an based on SEAIQR Model and Dropout-LSTM Model

作　　者：马艺菲 许书君 秦瑶[1] 李建涛 雷立健[1] 贺鹭[1] 余红梅[1,3] 解军 Ma Yifei;Xu Shujun;Qin Yao(School of Public Health,Shanxi Medical University(030001),Taiyuan)

机构地区：[1]山西医科大学公共卫生学院,030001 [2]山西医科大学管理学院 [3]煤炭环境致病与防治教育部重点实验室 [4]山西医科大学反向病原学中心

出　　处：《中国卫生统计》2024年第2期207-212,共6页Chinese Journal of Health Statistics

基　　金：国家重点研发计划(2021YFC2301603);山西省研究生科研创新项目(2023KY363)。

摘　　要：目的基于传染病动力学SEAIQR(susceptible-exposed-asymptomatic-infected-quarantined-removed)模型和Dropout-LSTM(Dropout long short term memory network)模型预测西安市新型冠状病毒肺炎(COVID-19)疫情的发展趋势,为评估“动态清零”策略防控效果提供科学依据。方法考虑到西安市本轮疫情存在大量的无症状感染者、依时变化的参数以及采取的管控举措等特点,构建具有阶段性防控措施的时变SEAIQR模型。考虑到COVID-19疫情数据的时序性特征及它们之间的非线性关系,构建深度学习Dropout-LSTM模型。选用2021年12月9日-2022年1月31日西安市新增确诊病例数据进行拟合,用2022年2月1日-2022年2月7日数据评估预测效果,计算有效再生数(R_(t))并评价不同参数对疫情发展的影响。结果SEAIQR模型预测的新增确诊病例拐点预计在2021年12月26日出现,约为176例,疫情将于2022年1月24日实现“动态清零”,模型R^(2)=0.849。Dropout-LSTM模型能够体现数据的时序性与非线性特征,预测出的新增确诊病例数与实际情况高度吻合,R^(2)=0.937。Dropout-LSTM模型的MAE和RMSE均较SEAIQR模型低,说明预测结果更为理想。疫情暴发初期,R 0为5.63,自实施全面管控后,R_(t)呈逐渐下降趋势,直到2021年12月27日降至1.0以下。随着有效接触率不断缩小、管控措施的提早实施及免疫阈值的提高,新增确诊病例在到达拐点时的人数将会持续降低。结论建立的Dropout-LSTM模型实现了较准确的疫情预测,可为COVID-19疫情“动态清零”防控决策提供借鉴。Objective This study aims to predict the coronavirus disease 2019(COVID-19)epidemic in Xi′an based on SEAIQR model and Dropout-LSTM model,and to provide a scientific basis for evaluating the effectiveness of the“dynamic zero-COVID policy”.Methods Considering a large number of asymptomatic infections,the changing parameters,and control procedures,we developed a time-dependent susceptible-exposed-asymptomatic-infected-quarantined-removed(SEAIQR)model with stage-specific interventions.Considering the time-series characteristics of COVID-19 data and the nonlinear relationship between them,we constructed a deep learning Dropout-LSTM model.The data of newly confirmed cases in Xi′an from December 9th,2021 to January 31st,2022 were used to fit the model,and the data from February 1st,2022 to February 7th,2022 were used to evaluate the model performance of forecasting.We then calculated the effective reproduction number(R_(t))and analyzed the sensitivity of the different measurement scenarios.Results The peak of newly confirmed cases predicted by the SEAIQR model would appear on December 26th,2021,with 176 cases,and the“dynamic zero-COVID policy”may be achieved in January 24th,2022,with R^(2)=0.849.The Dropout-LSTM model can reflect the time-series and nonlinear characteristics of the data,and the predicted newly confirmed cases were highly consistent with the actual situation,with R^(2)=0.937.The MAE and RMSE of the Dropout-LSTM model were lower than those of the SEAIQR model,indicating that the predicted results were more ideal.At the beginning of the outbreak,R 0 was 5.63.Since the implementation of comprehensive control,R_(t) has shown a gradual downward trend,dropping to below 1.0 on December 27th,2021.With the reduction of effective contact rate,the early implementation of control measures and the improvement of immunity threshold,the peak of newly confirmed cases will continue to decrease.Conclusion The proposed Dropout-LSTM model forecasts the epidemic well,which can provide a reference for decision-ma

关 键 词：新型冠状病毒肺炎 SEAIQR模型 Dropout-LSTM模型 动态清零 预测 对比 

分 类 号：R181.3[医药卫生—流行病学]