α-Sutte模型在疫情预测中的应用：基于R软件

刘天; 谢聪; 杨雯雯; 姚梦雷; 侯清波; 黄淑琼

doi:10.3784/jbjc.202109090486

摘要:

目的介绍α-Sutte模型的原理、方法，并利用R软件建立α-Sutte模型。比较α-Sutte模型与乘积季节自回归移动平均模型（SARIMA）拟合及预测效果，为α-Sutte模型在疫情预测中的应用提供参考。

方法收集2020年1月1日至2021年7月16日印度、美国、意大利、巴西、俄罗斯、南非各国新型冠状病毒肺炎（COVID-19）逐日累计报告病例数。以首例报告病例时间作为起点，起始日期至2021年6月16日数据作为训练数据，2021年6月17日至2021年7月16日作为测试数据。利用R语言根据α-Sutte模型计算公式自行编写拟合及预测函数α-Sutte。训练数据被用于训练α-Sutte模型和SARIMA模型。建立2个模型预测2021年6月17日至2021年7月16日COVID-19逐日报告病例数。拟合值与训练数据比较、预测值与测试数据比较评价模型拟合及预测效果。采用评价指标为平均绝对误差百分比（MAPE）。

结果印度、美国、意大利、巴西、俄罗斯和南非建立的最优SARIMA模型为SARIMA（5，2，2）、SARIMA（0，2，2）、SARIMA（2，2，2）、SARIMA（3，2，2）、SARIMA（0，2，1）和SARIMA（4，2，3）。 α-Sutte和SARIMA模型在印度、美国、意大利、巴西、俄罗斯、南非6个国家拟合的MAPE分别为1.32%、1.34%、0.89%、1.65%、0.99%、0.99%，以及1.51%、1.59%、0.89%、1.67%、1.03%、1.13%。 α-Sutte和SARIMA模型在6个国家预测的MAPE 分别为0.81%、0.09%、0.13%、1.58%、1.73%、3.77%，以及0.09%、0.09%、0.18%、1.13%、1.83%、3.43%。

结论 α-Sutte模型的原理、建模简单，利用R语言建立的模型拟合及预测精度高，值得在疾病监测领域推广使用。

Abstract:

Objective To introduce the principle and method ofα-Sutte model, establish a α-Sutte model by using software R, compare the fitting and prediction effects of theα-Sutte model and multiple seasonal autoregressive integrated moving average model, SARIMA model and provides reference for the application of theα-Sutte model in epidemic prediction.

Methods The daily cumulative number of reported cases of COVID-19 from India, the United States, Italy, Brazil, Russia, and South Africa from January 1, 2020 to July 16, 2021 were collected. Based on the time of the first reported case, the data reported by June 16, 2021 were used as training data, and the data reported from June 17, 2021 to July 16, 2021 were used as test data. According to the calculation formula of theα-Sutte model, the fitting and prediction functionα-Sutte() was written by software R. The training data was used to train theα-Sutte model and the SARIMA model. Two models were established to predict the number of daily reported cases of COVID-19 from June 17, 2021 to July 16, 2021. The fitted value was compared with the training data, the predicted value was compared with the test data, and the Mean Absolute Percentage Error (MAPE) was used to evaluate the model fitting and prediction effect.

Results The optimal SARIMA models established by India, the United States, Italy, Brazil, Russia and South Africa were SARIMA(5, 2, 2) SARIMA(0, 2, 2), SARIMA(2, 2, 2), SARIMA(3, 2, 2), SARIMA(0, 2, 1) and SARIMA(4, 2, 3) respectively. The MAPE fitted by theα-Sutte and SARIMA models in India, the United States, Italy, Brazil, Russia, and South Africa were 1.32%, 1.34%, 0.89%, 1.65%, 0.99%, 0.99% and 1.51%, 1.59%, 0.89%, 1.67%, 1.03%, 1.13% respectively. The MAPE predicted by theα-Sutte and SARIMA models in 6 countries were 0.81%, 0.09%, 0.13%, 1.58%, 1.73%, 3.77% and 0.09%, 0.09%, 0.18%, 1.13%, 1.83%, 3.43% respectively.

Conclusion The principle and modeling of theα-Sutte model are simple. Theα-Sutte model established by software R has high fitting and prediction accuracy, and it is worth to promote in disease surveillance.

α-Sutte模型在疫情预测中的应用：基于R软件

Application of α-Sutte model in epidemic prediction—based on software R