小波-ARIMA模型在肺结核发病趋势预测中的应用研究

Application of the Wavelet-ARIMA Model in Prediction of Incidence Trend of Tuberculosis

目的应用小波-ARIMA模型对肺结核的发病趋势进行预测,为肺结核防控提供更为有效、准确的理论支持。方法利用2010-2016年中国肺结核发病率数据,选择不同的小波函数及不同分解层数,建立小波-ARIMA模型,并进行ADF检验、Box-Ljung检验,检验水准为0.05;利用2017年数据验证小波-ARIMA、ARIMA模型的预测性能。结果在预测未来3个月发病趋势时,小波-ARIMA模型的MSE、MAE、MAPE分别为0.040、0.160、0.027;在预测未来6个月发病趋势时,小波-ARIMA模型的MSE、MAE、MAPE分别为0.027、0.136、0.024;在预测未来12个月发病趋势时,小波-ARIMA模型的MSE、MAE、MAPE分别为0.021、0.118、0.023;小波-ARIMA模型在未来6个月、12个月的预测性能均优于ARIMA模型。结论小波-ARIMA模型的中长期预测效果优于ARIMA模型,可为肺结核预测提供科学依据。

Objective To predict the incidence trend of tuberculosis by using the wavelet-ARIMA model in order to provide more effective theoretical support for TB prevention and control.Methods Tuberculosis incidence data from the year 2010 to 2016 in China are used to establish the waveletARIMA model with different wavelet function and decomposition level.ADF and Box-Ljung tests were performed,and the significance level was set as 0.05.Incidence data of tuberculosis in 2017 were used to verify performance of the wavelet-ARIMA and ARIMA models.Results The MSE、MAE and MAPE of wavelet-ARIMA model in predicting the incidence trend in the three month were 0.040、0.160、0.027.When those models are used the prediction in six months,the MSE、MAE and MAPE were 0.027,0.136,0.024.When those models are used for the prediction in one year,the MSE、MAE and MAPE were 0.021,0.118,0.023.The prediction performance of wavelet ARIMA model was better than ARIMA model in the six and twelve months.Conclusion The medium and long-term prediction effect of wavelet-ARIMA model is better than ARIMA model,providing scientific basis for tuberculosis prediction.