基于SARIMAX-AttLSTM和BEAST模型对重庆市秀山县流感特征的预测分析

Analysis and Prediction of Influenza Characteristics in Xiushan County, Chongqing Based on the SARIMAX-AttLSTM Model and BEAST Algorithm

摘要: 背景：流感是一种急性呼吸道传染病，对公共卫生系统构成严重威胁。明确区域流感流行特征并准确预测发病趋势，对制定防控策略具有重要意义。目的：分析重庆市秀山土家族苗族自治县(秀山县)流感流行特征及发病率变化趋势，为区域防控提供依据。方法：利用2018年第1周至2024年第26周的流感报告卡数据，结合同期气象数据分析流行特征。采用BEAST算法识别趋势与季节变化点，并构建SARIMAX-AttLSTM混合模型预测发病趋势。结果：共报告流感病例35,057例，男性52.7%、女性47.3% (χ² = 53.8, P < 0.001)。主要发病年龄为1~9岁 (52.9%, χ² = 2957.0, P < 0.001)，学生和幼托儿童为主要受影响人群。发病呈明显季节性，冬春为高发期。BEAST算法识别出4个季节性与5个趋势性变化点。SARIMAX-AttLSTM模型表现最佳(MAE = 8.064, RMSE = 11.724, R² = 0.922)，引入温度和降雨量后预测精度进一步提升。模型预测2024年第27周后发病率将上升。结论：秀山县流感呈冬春高发、儿童和学生易感的特征。SARIMAX-AttLSTM模型可有效捕捉流感季节与短期波动，具有较高的预警与决策支持应用价值。

Abstract: Background: Influenza is an acute respiratory infectious disease that poses a serious threat to public health systems. Understanding regional influenza transmission patterns and accurately predicting incidence trends are essential for developing effective prevention and control strategies. Objective: To analyze the epidemiological characteristics and temporal trends of influenza incidence in Xiushan Tujia and Miao Autonomous County (Xiushan County), Chongqing, and to provide evidence-based support for regional influenza control. Methods: Weekly influenza case report data from the 1st week of 2018 to the 26th week of 2024 were analyzed in conjunction with meteorological data to characterize influenza patterns. The BEAST algorithm was applied to identify trend and seasonal change points, and a hybrid SARIMAX-AttLSTM model was developed to predict future influenza incidence. Results: A total of 35,057 influenza cases were reported, with males and females accounting for 52.7% and 47.3%, respectively (χ² = 53.8, P < 0.001). The 1~9-year age group had the highest incidence (52.9%, χ² = 2957.0, P < 0.001), with students and preschool children being the most affected populations. Influenza incidence showed clear seasonality, peaking in winter and spring. The BEAST algorithm identified four seasonal and five trend change points. The SARIMAX-AttLSTM hybrid model achieved the best performance (MAE = 8.064, RMSE = 11.724, R² = 0.922), and its accuracy improved further after incorporating temperature and rainfall as exogenous variables. The model predicted an upward trend in influenza incidence after the 27th week of 2024. Conclusions: Influenza in Xiushan County exhibits strong winter-spring seasonality, with children and students as the main susceptible groups. The SARIMAX-AttLSTM hybrid model effectively captures both seasonal cycles and short-term fluctuations, providing a valuable tool for early warning and public health decision-making.