GM(1,1)模型与LSTM神经网络在肺结核发病数预测中的应用  

Application of GM(1,1) model and LSTM neural network in predicting the incidence of pulmonary tuberculosis

作　　者：王淑平 杜敏[1] 罗建伟[1] 胡佩佩[1,2] 程冬 WANG Shuping;DU Min;LUO Jianwei;HU Peipei;CHENG Dong(Hubei Cancer Hospital,Wuhan 430079,China;Tongji Medical College,Huazhong University of Science & Technology,Wuhan 430030,China)

机构地区：[1]湖北省肿瘤医院,武汉430079 [2]华中科技大学同济医学院

出　　处：《公共卫生与预防医学》2019年第5期11-14,共4页Journal of Public Health and Preventive Medicine

摘　　要：目的运用GM(1,1)灰色预测模型和LSTM神经网络模型对全国肺结核发病数进行拟合,对拟合结果进行比较,为肺结核病防控工作提供科学依据。方法利用2008—2018年全国肺结核发病数分别构建GM(1,1)灰色预测模型和LSTM神经网络模型,对建立的模型进行拟合,同时运用得到的模型对2019年全国肺结核发病数进行预测。两种模型拟合评价指标为均方根误差(RMSE)和平均绝对百分比误差(MAPE)。结果全国肺结核发病数GM(1,1)灰色预测模型和LSTM神经网络模型的RMSE和MAPE分别为16108.39、15169.81和1.57%、1.15%。两种模型预测2019年全国肺结核发病数分别为777004和822125。结论LSTM神经网络模型拟合效果优于GM(1,1)模型,预测结果表明2019年全国肺结核年发病数将呈现出减少的趋势,但是肺结核发病数仍然较多。Objective To fit the incidence of pulmonary tuberculosis in China with GM(1,1)model and LSTM neural network model,and to provide a scientific basis for the prevention and treatment of pulmonary tuberculosis by comparing the fitting results.Methods GM(1,1)model and LSTM neural network model were constructed by using the incidence of pulmonary tuberculosis in China from 2008 to 2018,and the established models were fitted.At the same time,the obtained models were used to predict the incidence of pulmonary tuberculosis in China in 2019.The RMSE and MAPE were used to evaluate the fitting degree of the two models.Results The RMSE of GM(1,1)model and LSTM neural network model was 16 108.39 and 15 169.81,respectively,while the MAPE was 1.57%and 1.15%,respectively.The incidence of pulmonary tuberculosis in China in 2019 predicted by the two models were 777 004 and 822 125,respectively.Conclusion In this study,the fitting result of LSTM neural network model was better than that of GM(1,1)model.The predicted results showed that the incidence of pulmonary tuberculosis in China will decrease in 2019 but it is still at a high level.

关 键 词：GM(1 1)模型 LSTM神经网络 肺结核 

分 类 号：R183.3[医药卫生—流行病学]