几种预测模型对中国梅毒发病率预测效果的比较*
Efficacy of different prediction models in forecasting incidence rate of syphilis in China

目的：比较3种预测模型在中国梅毒疫情预测中的效果，筛选最优预测模型。 方法：收集2004年至2012年中国梅毒发病率数据，构建灰色模型［GM(1,1)］、趋势外推模型和求和自回归滑动平均(ARIMA)模型，比较预测值和实际值的吻合程度;用2013年发病率数据回代验证，选择相对误差最小的模型预测2014年至2016年的梅毒发病率。 结果：中国梅毒发病率呈整体上升趋势，年平均发展速度为1.173，但环比增长速度逐年降低。趋势外推模型中Cubic函数的拟合效果优于GM(1,1)，二者对历史数据拟合的平均相对误差分别为1.431%和7.560%。梅毒年发病率序列为白噪声序列(χ2=7.990，P=0.239），不适合用ARIMA模型来预测。采用Cubic函数预测2014年至2016年中国梅毒的发病率，分别为29.553/10万、26.293/10万和20.831/10万。 结论：Cubic函数对中国梅毒发病率的预测效果最好。
Aim: To compare the effect of different prediction models in forecasting the incidence rate of syphilis in China and screen the optimal model. Methods:GM(1,1), trend extrapolation model and ARIMA model were developed based on the yearly incidence rate of syphilis in China from 2004 to 2012. Relative error was used to evaluate the forecasting effect of the three models. The model with the least relative error was adopted to predict the incidence rate of syphilis from 2014 to 2016. Results: The incidence rate of syphilis was on the trend of increasing and the average development speed was 1.173. Link relative ratio growth speed was on the trend of decreasing. The predicted value by Cubic model matched with the observed value better than GM(1, 1). The average relative error of Cubic model and GM(1, 1) to historical data was 1.431% and 7.560%, which showed Cubic model was better than GM(1, 1). The yearly incidence rate of syphilis in China was a white noise series(χ2=7.990，P=0.239), which showed ARIMA model was not suitable for predicating incidence rate of syphilis. The predicated incidence rate of syphilis in China from 2014 to 2016 by Cubic model was 29.553/100 000，26.293/100 000 and 20.831/100 000, respectively. Conclusion: Cubic model is better in predicting incidence rate of syphilis in China