数据驱动的肺结核传播过程的建模与分析

Modeling and analysis of data-driven tuberculosis transmission process

为研究肺结核传播过程,通过数据解析和机理建模相结合的方法,建立了三种能够较精确描述肺结核传播过程的模型。运用中国防控中心提供的近六年肺结核患病数据,分别基于SIS近似解析解的改进模型、BP神经网络(BPNN)模型和局部加权线性回归模型进行建模与分析。通过对所得预测数据运用方差对比的方法可知,SIS近似解析解的改进模型和局部加权线性回归模型预测值和实际值之间的方差较小,可以更加精确地预测肺结核传播过程。仿真结果表明,这三种建模方法都可对肺结核的防控工作提供科学依据,且在所采用的样本范围内,与BP神经网络模型相比局部加权线性回归模型预测的准确度更高,而基于SIS近似解析解改进的模型可以在一定范围内稳定,预测值波动较小。

To study the process of tuberculosis transmission, by the combination of data analysis and mechanism modeling, three models that were able to accurately describe the process of tuberculosis transmission were established. Using the data of tuberculosis for near six years provided by China National Defense Control Center, modeling and analysis were carried out for the improved model of SIS approximate analytical solution, Back Propagation Neural Network (BPNN) model and locally weighted linear regression model. By applying variance comparison method to the obtained prediction data, it can be known that, the variance between the predicted and actual values of the improved model of SIS approximate analytical solution and locally weighted linear regression model are small, which means the two models can predict the process of tuberculosis transmission more accurately. Simulation results show that all three modeling methods can provide a scientific basis for the prevention and control of tuberculosis, and within the adopted sample range, compared to BPNN model, locally weighted linear regression model has higher predicton accuracy. However, the improved model based on SIS approximate analytical solution can be stable within a certain range, and the predicted value fluctuates less.