基于Python的LSTM模型对流感预测的研究  

Study on LSTM Model based on Python in Prediction of Influenza

作　　者：翟梦梦 王旭春 任浩[1] 全帝臣 李美晨 陈利民 仇丽霞[1] Zai Mengmeng;Wang Xuchun;Ren Hao(Shanxi Medical University(030001),Taiyuan)

机构地区：[1]山西医科大学卫生统计教研室,030001 [2]山西省人民医院

出　　处：《中国卫生统计》2022年第2期162-166,171,共6页Chinese Journal of Health Statistics

基　　金：山西省重点研发计划项目(201803D31066)。

摘　　要：目的 探讨基于keras的LSTM模型和SARIMA模型预测我国北方省份流感样病例数的可行性,为流感防控工作提供合理的预测方法。方法 利用国家流感中心2013-2019年北方省份的周流感监测数据构建LSTM模型和SARIMA模型,并进行预测。采用平均绝对误差(MAE)、均方根误差(RMSE)评价两种模型的预测效果。结果 LSTM、SARIMA模型的MAE值分别为304.19、352.74,RMSE值分别为398.71、521.07;相比之下,LSTM模型的预测性能优于SARIMA,较SARIMA模型预测性能分别提高了13.76%、23.5%。结论 基于Keras的LSTM模型的预测效果较好,优于SARIMA模型,可为流感预测提供科学依据。Objective To explore the feasibility of the LSTM model based on Keras and SARIMA model in predicting the number of influenza-like cases in northern provinces of China, and to provide a reasonable prediction method for influenza prevention and control.Methods The weekly influenza surveillance data of northern provinces during from 2013 to 2019 were obtained from the National Influenza Center, and the LSTM model and SARIMA model were constructed respectively to predict the number of influenza-like cases, and the prediction effect of the models were compared.Mean absolute error(MAE)and root mean square error(RMSE)were used to evaluate the prediction effect of the two models.Results The MAE of LSTM and SARIMA model were 304.19 and 352.74,the RMSE were 398.71 and 521.07,respectively.Compared with SARIMA model, the prediction performance of LSTM model was better, which increased by 13.76% and 23.5%.Conclusion The prediction effect of the LSTM model based on eras was better than SARIMA model, which can provide scientific basis for influenza prediction.

关 键 词：LSTM模型 SARIMA模型 PYTHON Keras 流感 预测 

分 类 号：R195.1[医药卫生—卫生统计学]