基于Python语言的ARIMA模型在天津市结核病发病率预测中的应用

 目的 探讨差分自回归移动平均（ARIMA）模型在结核病发病率预测中的可行性。方法 基于Python语言的statsmodels模块，以天津市2004年1月—2015年12月结核病月发病率数据作为训练集建立最优季节性差分自回归移动平均（SARIMA）模型，以2016年1—12月数据对SARIMA模型进行效果评价，并对2017年1月—2019年12月天津市结核病月发病率进行预测。结果 流行病学结果显示，2004年1月—2015年12月天津市结核病月发病率总体呈下降趋势。2005—2008年出现一个发病高峰，2009年后大幅度下降，随后趋于平稳。2017年1月—2019年12月天津市结核病月发病率与往年相比平稳下降。建立的最佳模型为SARIMA（1，1，1）×（3，1，1）₁₂，该模型残差BOX-Ljung统计量P值为0.493，提示残差为白噪声序列，模型拟合良好。预测结果实际值均在预测值的95%置信区间。结论 SARIMA（1，1，1）×（3，1，1）₁₂模型可对天津市结核病月发病率进行较准确的预测。

Application of ARIMA model in predicting the incidence of tuberculosis in Tianjin City based on Python language

Objective To evaluate feasibility of autoregressive integrated moving average (ARIMA) model in predicting the incidence of tuberculosis (TB). Methods Using statsmodels module-based Python language, incidence of TB in Tianjin City from January 2004 to December 2015 was as training set, the optimal seasonal ARIMA (SARIMA) model was established, data from January to December 2016 were used to evaluate the efficacy of SARIMA model, and monthly incidence of TB in Tianjin City from January 2017 to December 2019 was predicted. Results Epidemiological results showed that monthly incidence of TB in Tianjin showed a overall downward trend from January 2004 to December 2015. There was a of peak disease incidence in 2005-2008, which dropped sharply after 2009 and then stabilized. From January 2017 to December 2019, monthly incidence of TB in Tianjin City declined steadily compared with previous years. The established optimal model was SARIMA(1,1,1)×(3,1,1)₁₂, residual BOX-Ljung statistic of the model was P=0.493, which indicated that the residual was a white noise sequence and the model fitted well. The actual value of predicted results was within 95% confidence interval of predicted value. Conclusion SARIMA (1,1,1)×(3,1,1)₁₂ model can accurately predict the monthly incidence of tuberculosis in Tianjin City.