4种模型对我国某地区肺结核发病率的预测

目的]对我国某地区的肺结核年发病率进行预测。方法]采用2000～2005年的年发病率建立GM(1,1)模型。采用2000～2005年月发病率建立ARIMA(p,d,q)(P,D,Q)s模型,并结合同期气象因素,建立多元线性回归模型和BP人工神经网络模型。以2006年的实际年发病率验证4种模型的预测效果,评价指标为相对误差。选取相对误差最小的预测模型为最佳预测模型。结果]GM(1,1)模型、ARIMA模型、多元线性模型、BP人工神经网络模型对2006年肺结核年发病率的预测值分别为126.18/10万、126.84/10万、98.95/10万和111.19/10万。以上4个模型的相对误差依次为19.84%、20.49%、5.39%和4.86%。BP人工神经网络模型为最佳预测模型。结论]对于肺结核发病率的预测,应同时拟合几种模型,并选择其中拟合效果最好的一种模型。

FORECASTING INCIDENCE OF TUBERCULOSIS IN A CITY WITH FOUR TYPES OF MOD-ELS

Objective To forecast yearly incidence of tuberculosis in a city.Methods Grey System GM(1, 1) model was developed based on the yearly incidence from 2000 to 2005.The ARIMA(p, d, q)(P, D, Q) s model was developed according to the monthly incidence from January 2000 to December 2005.And the data of meteorological factors during the same period was used as independent variable or input neuron to estimate multiple linear regression and back-propagation artificial neural network.All of models were tested by the data in 2006.Relative error was used to validate the predictability of these models.The model with least relative error would be chosen as the best model to forecast the incidence of next years.Results The estimated yearly incidence of tuberculosis in 2006 predicted by GM(1, 1), ARIMA, multiple linear regression and back-propagation artificial neural network was 126.18 / 105, 126.84 / 105, 98.95 / 105 and 111.19 / 105, respectively.And their corresponding relative errors were 19.84% , 20.49%, 5.39% and 4.86% , respectively.Back-propagation artificial neural network was the preferable model for forecasting.Conclusion Several models should be estimated to forecast incidence of tuberculosis, and the best fitted model would be chosen for forecasting.