基于ARIMA模型在美国COVID-19累计确诊人数中的应用

Based on the Application of the ARIMA Model in the Cumulative Number of Confirmed Cases of COVID-19 in the United States

摘要: 本文针对新型冠状病毒肺炎给世界人民造成的不良影响，收集2020年1月20日~2020年6月1日内以美国为主的各个国家和地区每日COVID-19累计确诊人数，建立自回归求和滑动平均(auto regressive integrated moving average, ARIMA)模型对美国累计确诊人数进行分析与预测，用SPSS25.0和MATLAB2019a拟合，结合拟合优度R2和Q检验评价拟合效果，将后5日累计确诊人数预测值和真实值进行比较，评价该模型预测精度及预测美国未来10日累计确诊人数。结果表明，原始序列经2次差分后能较好拟合ARIMA (0,2,1)模型，R²在0.95以上，Q检验p值为0.19 > 0.05，认为残差为白噪声，且预测值与实际值动态趋势基本一致，预测值在真实值0.33%误差内波动，ARIMA (0,2,1)模型对美国COVID-19累计确诊人数预测精度很高，对疫情防控具有很强的指导意义。

Abstract: To address the adverse impact of COVID-19 on people around the world, this article collected the cumulative number of daily COVID-19 diagnosed in countries and territories, mainly the United States, from January 20 to June 1, 2020. Auto Regressive Integrated Moving Average Model (ARIMA) was established to analyze and predict the cumulative number of diagnosed cases in the United States. With SPSS25.0 and MATLAB2019a as fitting methods, combined with the R2 and Q test to evaluate the fitting effect, the predicted and real values for the cumulative number of confirmed cases for the last 5 days were compared to evaluate the prediction accuracy of the model and the cumulative number of confirmed cases for the next 10 days. The results show that the original sequence can fit the ARIMA (0,2,1) model well after two differences. The R² is above 0.95, and the p value of the Q test is 0.19 > 0.05 that is regarded as white noise, and the predicted value is basically consistent with the actual value dynamic trend. The forecast value fluctuated within 0.33% of the true value. The ARIMA (0,2,1) model is under a high accuracy in predicting the number of COVID-19 diagnosed in the United States, which has a strong guiding significance for epidemic prevention and control.