戴皓云, 周楠, 任香, 罗飘异, 易尚辉, 全梅芳, 查文婷, 吕媛. 基于自回归移动平均模型各亚型流行性感冒流行特征与趋势预测[J]. 疾病监测, 2022, 37(10): 1338-1345. DOI: 10.3784/jbjc.202204290185
引用本文: 戴皓云, 周楠, 任香, 罗飘异, 易尚辉, 全梅芳, 查文婷, 吕媛. 基于自回归移动平均模型各亚型流行性感冒流行特征与趋势预测[J]. 疾病监测, 2022, 37(10): 1338-1345. DOI: 10.3784/jbjc.202204290185
Dai Haoyun, Zhou Nan, Ren Xiang, Luo Piaoyi, Yi Shanghui, Quan Meifang, Zha Wenting, Lyu Yuan. Epidemiologic characteristics and prediction of incidence trend of all types of influenza based on ARIMA model[J]. Disease Surveillance, 2022, 37(10): 1338-1345. DOI: 10.3784/jbjc.202204290185
Citation: Dai Haoyun, Zhou Nan, Ren Xiang, Luo Piaoyi, Yi Shanghui, Quan Meifang, Zha Wenting, Lyu Yuan. Epidemiologic characteristics and prediction of incidence trend of all types of influenza based on ARIMA model[J]. Disease Surveillance, 2022, 37(10): 1338-1345. DOI: 10.3784/jbjc.202204290185

基于自回归移动平均模型各亚型流行性感冒流行特征与趋势预测

Epidemiologic characteristics and prediction of incidence trend of all types of influenza based on ARIMA model

  • 摘要:
      目的  分析我国2010—2019年流行性感冒的流行特征和分布规律,预测各亚型流感发病趋势。
      方法  采用ARIMA乘积季节模型,对流感数据进行原始序列预处理、模型识别、参数估计和统计建模,预测流感发病趋势。
      结果  构建流感自回归移动平均模型(ARIMA)乘积季节模型,预测模型为ARIMA(1,2,1)(0,1,1)12,数据信息提取充分(Q=14.257,P>0.05),相对误差约10%;甲型流感预测模型为ARIMA(2,1,1)(0,2,2)12,数据信息提取充分(Q=13.236,P>0.05),预测2018年12月至2019年3月的甲型流感发病率较高,4月开始,发病率迅速下降,与实际情况相似,相对误差控制在10%以内;乙型流感预测模型为ARIMA(1,2,1)(1,0,1)12,数据信息提取充分(Q=9.841,P>0.05),但模型预测2019年乙型流感发病率较低,相对误差较高。
      结论  流感、甲型流感ARIMA乘积季节模型预测效果较好;乙型流感预测模型数据信息提取充分,但相对误差较高,可能与乙型流感发病无明显的长期趋势有关。

     

    Abstract:
      Objective  To analyze the epidemiologic characteristics and distribution of influenza in China from 2010 to 2019, and predict the incidence trends of all types of influenza.
      Methods  Seasonal ARIMA model was used for original series pre-process, model identification, parameters estimation and statistical modeling to predict the incidence trend of influenza.
      Results  The influenza time series model constructed was ARIMA (1,2,1) (0,1,1)12, and the data information was fully extracted (Q=14.257, P>0.05), the relative error was about 10%. Influenza A prediction model was ARIMA (2,1,1) (0,2,2)12, the data information was fully extracted (Q=13.236, P>0.05). The predicted incidence of influenza A was high from December 2018 to March 2019, and the incidence decreased rapidly from April, similar to the actual situation. Relative error was controlled within 10%; The influenza B prediction model was ARIMA (1,2,1) (1,0,1)12, and the data information was fully extracted (Q=9.841, P>0.05), but the incidence of influenza B in 2019 predicted by the model was low and the relative error was high.
      Conclusion  Influenza and influenza A seasonal ARIMA models had better prediction effects. The data information of influenza B prediction model was fully extracted, but the relative error was high, which might be related to the absence of obvious long-term trend of influenza B incidence.

     

/

返回文章
返回