基于SARIMA-LSTM模型的肾综合征出血热发病率预测研究

Study on the Prediction of Incidence of Hemorrhagic Fever with Renal Syndrome Based on the SARIMA-LSTM Model

目的/意义探究前沿技术在肾综合征出血热发病率预测中的应用,梳理、组合多种时序分析方法,评价并筛选最佳模型。方法/过程利用2004—2020年全国肾综合征出血热发病率数据,基于统计学方法的SARIMA、STL-ARIMA、TBATS模型,基于神经网络的NNAR、LSTM模型,基于3种加权方式的SARIMA-LSTM组合模型进行预测,运用RMSE、MAE、MAPE综合评价模型效果。结果/结论SARIMA、LSTM在单一模型中较优;SARIMA-LSTM组合模型效果相较单一模型均有提升;基于误差倒数法的SARIMA-LSTM组合模型为最优模型。本研究有望为肾综合征出血热发病预警系统模型设计提供技术支持与参考。

Purpose/Significance To investigate the application of cutting-edge technologies in predicting the incidence of hemorrhagic fever with renal syndrome(HFRS),to compile and integrate various time-series analysis methods,evaluate and select the optimal model.Method/Process By utilizing national HFRS incidence data from 2004 to 2020,the effectiveness of models is predicted based on statistical methods:SARIMA,STL-ARIMA and TBATS,neural network approaches:NNAR,LSTM and combined models of SARIMA-LSTM with 3 different weighting schemes.The performance of these models is comprehensively assessed using RMSE,MAE and MAPE.Result/Conclusion The SARIMA and LSTM models are identified as the superior individual models.The combined SARIMA-LSTM model demonstrates enhanced performance compared to individual models.The SARIMA-LSTM model optimized using the reciprocal of error method is deemed the optimal model.The optimal model is expected to provide technical support and references for the early warning system model design of HFRS.