基于组合模型的河南省甲型H1N1流感流行特征分析与预测

Analysis and Prediction of Epidemic Characteristics of H1N1 Influenza in Henan Province Based on Combination Models

摘要: 探讨河南省甲流的流行特点和精准预测模型，为流感的预防控制提供参考依据。在河南省卫健委收集2013年~2024年流感病例数，对流感流行时空特点进行分析；然后，利用Python软件和2018年1月~2022年11月河南省流感发病数据构建了ARIMA(8,1,9)模型，并使用2022年12月~2024年2月的发病数进行了测试，得到了评估指标RMSE；接着，在ARIMA模型的基础上利用深度学习算法进行残差修正，建立了ARIMA-LSTM、ARIMA-BP和ARIMA-SVM的组合模型，比较几个模型的预测效果。结果发现河南省流感发病趋势逐年上升，且具有显著的季节特征，每年的12月~3月是发病高峰期。模型拟合结果显示，与深度学习相结合的组合模型能够提高传统ARIMA模型的预测精度，ARIMA-BP和ARIMA-SVM组合模型更适合于传染病的追踪预测。

Abstract: Explore the epidemic characteristics and precise prediction model of H1N1 in Henan Province to provide a reference for influenza prevention and control. Influenza case numbers in Henan Province from 2013 to 2024 were collected by the Health Commission of Henan Province to analyze the temporal and spatial characteristics of influenza. Then, using the influenza incidence data from January 2018 to November 2022 in Henan Province, an ARIMA (8,1,9) model was constructed, and the incidence numbers from December 2022 to February 2024 were tested to obtain the evaluation index RMSE. Subsequently, based on the ARIMA model, deep learning algorithms were used for residual correction, establishing ARIMA-LSTM, ARIMA-BP, and ARIMA-SVM combined models, and comparing the predictive effects of these models. Results showed that the influenza incidence trend in Henan Province is increasing year by year, with significant seasonal characteristics, and the peak incidence is from December to March each year. The model fitting results show that the combined model with deep learning can improve the prediction accuracy of traditional ARIMA models and has smaller residuals, and the combination of ARIMA-BP and ARIMA-SVM models is more suitable for predicting infectious diseases.