新型冠状病毒感染2种时间序列分析模型预测效果比较

Efficacy of time series gray model and exponential smoothing model in predictions of daily outpatient visits,numbers of SARS-CoV-2 nucleic acid and antigen positive among the visits during COVID-19 epidemic:an empirical analysis in Yunnan province

目的比较灰色模型和指数平滑模型2种时间序列分析模型对新型冠状病毒感染的预测效果,为构建适宜新型冠状病毒感染“乙类乙管”时期的短期预测模型提供参考依据。方法收集云南省西双版纳傣族自治州2022年12月12—31日所辖景洪市、勐腊县和勐海县医疗机构发热门诊上报的54836例就诊者相关数据,针对发热门诊就诊数、核酸检测阳性数、抗原检测阳性数3条时间序列分别构建灰色模型和指数平滑模型,并采用预测值与实际值的平均相对误差评价模型的预测效果。结果发热门诊就诊数、核酸检测阳性数和抗原检测阳性数3条时间序列的灰色模型后验比C值分别为0.50、0.40和0.43,3条时间序列的指数平滑模型的R2值分别为0.36、0.56和0.53;构建的GM(1,1)灰色模型对3条时间序列预测结果的平均相对误差分别为103%、72%和84%,霍尔特线性趋势非季节性指数平滑模型对3条时间序列预测结果的平均相对误差分别为103%、66%和51%。结论灰色模型和指数平滑模型对发热门诊就诊数的预测效果相当,指数平滑模型对核酸检测阳性数和抗原检测阳性数的预测效果相对优于灰色模型。

Objective To examine the efficacy of time series gray model and exponential smoothing mode in predictions of daily fever outpatient visits,number of severe acute respiratory syndrome coronavirus-2(SARS-CoV-2)nucleic acid and antigen positive among the visits during conronavirus disease 2019(COVID-19)epidemic for developing short-term prediction model of COVID-19 incidence.Methods The data on 54836 individuals visiting fever clinics during a COVID-19 epidemic period(December 12–31,2022)were collected from medical facilities in a prefecture and two districts of Yunnan province and analyzed statistically.The series gray model GM(1,1)and exponential smoothing model were fitted to the data to construct models for predicting daily number of fever clinic visits,SARS-CoV-2 nucleic acid and antigen positivity in the study region.Average relative errors between the actual and the predicted values were used to evaluate the efficacy of the prediction models.Results The posterior ratio C values of the constructed time series gray model GM(1,1)were 0.50,0.40,and 0.43 and the average relative errors were103%,72%,and 84%in predicting the daily number of fever clinic visits,SARS-CoV-2 nucleic acid positivity,and antigen positivity.In predicting the three numbers mentioned above,the R2 values were 0.36,0.56,and 0.53 and the average relative errors were 103%,66%and 51%,respectively,for the constructed Holt linear trend non-seasonal exponential smoothing model.Conclusion The efficacy of gray model is comparable to that of exponential smoothing model in predicting the number of fever outpatient visits but better than that of the latter model in predicting the number of SARS-CoV-2 nucleic acid and antigen positivity.