基于ARIMA和LSTM神经网络的乌鲁木齐市乙型肝炎发病预测研究

目的 了解乌鲁木齐市2012—2021年乙肝发病趋势，建立合适的发病预测模型，探讨ARIMA模型和LSTM神经网络在乙肝发病预测中的应用。方法 根据2012—2021年乙肝月报告病例数据，建立ARIMA模型和LSTM神经网络模型，对乌鲁木齐市乙肝发病数进行拟合及预测，通过比较RMSE的大小评价模型效果。结果 LSTM神经网络模型拟合和预测的RMSE分别为50.13、42.70，ARIMA(0,1,1)(0,0,2)12模型拟合和预测的RMSE分别为67.62、66.85。前者的拟合及预测效果显著优于后者。结论 乌鲁木齐市10年来乙肝发病呈逐年下降趋势，且存在一定季节性变化。LSTM神经网络模型可较好地拟合和预测乌鲁木齐市乙肝的发病数及趋势，且模型效果优于ARIMA(0,1,1)(0,0,2)12，能在一定程度上提高预测精确度。

Prediction study of hepatitis B incidence in Urumqi based on ARIMA model and the LSTM neural network model

Objective To observe the incident trend of hepatitis B in Urumqi from 2012 to 2021, and to establish a suitable model for predicting the number of the incidence and the epidemic trend of hepatitis B, so as to provide a scientific basis for hepatitis B prevention and control. Methods Based on the reported monthly hepatitis B incidence case data from 2012 to 2021, ARIMA model and the LSTM neural network model were established to fit and predict the incidence of hepatitis B in Urumqi and then to compare their effects by the size of root mean square errors (RMSEs). Results The LSTM neural network model was significantly better than the ARIMA (0,1,1) (0,0,2)12 model in fitting and predicting. The RMSE was 50.13 and 42.70 for the former and 67.62 and 66.85 for the latter, respectively. Conclusion The incidence of hepatitis B in Urumqi has been decreasing year by year over the last decade with seasonal changes. The LSTM neural network model can better fit and predict the incidence and the epidemic trend of hepatitis B in Urumqi city than ARIMA (0,1,1) (0,0,2)12, which can improve the predicting accuracy to certain extent.