SARIMA模型和LSTM神经网络在预测深圳市宝安区手足口病疫情中的应用

Application of SARIMA model and LSTM model in predicting incidence of hand-foot-mouth disease in Bao’an district of Shenzhen city

目的探讨季节性差分自回归移动平均模型(seasonal autoregressive integrated moving average model,SARIMA)和长短时记忆神经网络(long short term memory,LSTM)预测深圳市宝安区手足口病发病趋势的可行性。方法选取2009—2018年深圳市宝安区的手足口病月发病率作为训练集分别构建SARIMA模型和LSTM神经网络,预测2019年1—12月的手足口病月发病率,并与真实值比较。结果相对最优模型SARIMA(0,0,2)(0,1,2)_(12)和LSTM神经网络对2010—2018年的深圳市宝安区手足口病月发病率进行拟合,拟合性能中LSTM神经网络的均方误差(mean squared error,MSE)和均方根误差(root mean squared error,RMSE)均高于SARIMA模型,而平均绝对误差(mean absolute error,MAE)和平均绝对百分比误差(mean absolute percentage error,MAPE)均低于SARIMA模型,表明两种模型的拟合性能基本一致。使用两种模型预测2019年1—12月手足口病月发病率,SARIMA模型预测性能的MSE、RMSE、MAE、MAPE分别为2310.199,48.065,31.990和1.002,LSTM模型预测性能的MSE、RMSE、MAE、MAPE分别为1078.899,32.847,22.046和0.958,表明LSTM神经网络的预测性能高于SARIMA模型。结论LSTM神经网络能更好地预测深圳市宝安区手足口病发病趋势,可为相关部门制定手足口病防控策略提供依据。

Objective To explore the practicability of seasonal autoregressive integrated moving average model(SARIMA)and long short term memory(LSTM)for predicting the incidence trend of hand-foot-mouth disease(HFMD)in Bao’an district of Shenzhen city.Methods SARIMA model and LSTM model were respectively established based on the monthly incidence of HFMD in Bao’an district of Shenzhen city from 2009 to 2018.The models were used to forecast the monthly incidence of HFMD from January to December in 2019,and the predictive value was compared with the observed value.Results The optimal SARIMA(0,0,2)(0,1,2)_(12) model and LSTM model were fitted to predict the monthly incidence of HFMD from 2010 to 2018.The mean squared error(MSE)and the root mean squared error(RMSE)of LSTM were higher than those of SARIMA,but the mean absolute error(MAE)and the mean absolute percentage error(MAPE)of LSTM were lower than those of SARIMA,which showed the fitting performance of two models was basically the same.MSE,RMSE,MAE and MAPE of SARIMA model were 2310.199,48.065,31.990 and 1.002 for predicting the monthly incidence of HFMD from January to December in 2019,and MSE,RMSE,MAE and MAPE of LSTM model were 1078.899,32.847,22.046 and 0.958,which showed the predictive performance of LSTM was better than SARIMA.Conclusion LSTM model can better predict the incidence trend of HFMD in Bao’an district of Shenzhen city,and it can be used to provide scientific evidence for relevant organizations to develop HFMD prevention and control strategies.