基于SARIMA模型与LSTM神经网络模型的中国肺结核发病预测比较研究  

Comparative Study of SARIMA and LSTM Neural Network Models for Pulmonary Tuberculosis Incidence Prediction in China

作　　者：武长礼 徐锋顺 段萌 李海川 罗银波 WU Chang-Li;XU Feng-Shun;DUAN Meng;LI Hai-Chuan;LUO Yin-Bo(Hubei International Travel Health Care Center(Outpatient Department of Wuhan Customs Port),Wuhan 430070;Wuhan Customs,Wuhan 430048;Hubei Provincial Center for Disease Control and Prevention,Wuhan 430070)

机构地区：[1]湖北国际旅行卫生保健中心(武汉海关口岸门诊部),武汉430070 [2]武汉海关,武汉430048 [3]湖北省疾病预防与控制中心,武汉430070

出　　处：《中国口岸科学技术》2025年第3期4-12,共9页China Port Science and Technology

基　　金：武汉海关科研项目(2024WK004)。

摘　　要：本研究通过评估季节性自回归差分移动平均模型(Seasonal Autoregressive Integrated Moving Average,SARIMA)与长短期记忆(Long Short-Term Memory,LSTM)神经网络在预测中国肺结核发病的效能差异,为传染病预警系统构建提供模型选择依据。基于2011—2023年全国肺结核监测数据,通过划分训练集与测试集,分别构建SARIMA模型和LSTM神经网络模型,采用绝对百分比误差(Absolute Percentage Error,APE)、平均绝对百分比误差(Mean Absolute Percentage Error,MAPE)和均方根误差(Root Mean Square Error,RMSE)等指标来评价外推预测精度,最后用上述最优参数模型来进行预测分析。结果显示,SARIMA最优模型(1,1,0)(1,1,0)_(12)在预测后期APE较LSTM高,SARIMA模型和LSTM神经网络模型的MAPE分别为11.26%和9.45%,RMSE分别为9791.58和8337.80。结果表明,SARIMA模型和LSTM神经网络模型均能对中国肺结核发病情况进行较好的预测,LSTM神经网络模型在捕捉非线性时序特征方面具有显著优势,尤其适用于中长期预测场景。This study evaluated the performance differences between the SARIMA and LSTM models in predicting monthly pulmonary tuberculosis incidence in China,providing evidence for infectious disease surveillance system development.Based on the national pulmonary tuberculosis surveillance data from 2011 to 2023,we trained and tested both SARIMA and LSTM models.Predictive accuracy was assessed by Absolute Percentage Error(APE),Root Mean Absolute Percentage Error(MAPE) and Root Mean Square Error(RMSE).The optimal SARIMA model(1,1,0)(1,1,0)_(12) has a higher APE than the LSTM model in the later stage of prediction.The SARIMA model demonstrated higher MAPE(11.26% vs 9.45%) and RMSE(9791.58 vs 8337.80) compared to the LSTM model.Both models demonstrated satisfactory predictive capabilities,with LSTM exhibiting superior performance in capturing nonlinear temporal patterns,particularly in later prediction phases.

关 键 词：时间序列分析 肺结核监测 模型比较 预测 

分 类 号：R521[医药卫生—内科学]