三种时间序列模型在尘肺发病预测中的适用性研究

Applicability study on three time series models in incidence prediction of pneumoconiosis

关键词: 尘肺发病预测  时间序列  自回归滑动平均混合模型（ARIMA）  灰色模型(GM)  广义回归神经网络模型（GRNN）  模型比较

Keywords: incidence prediction of pneumoconiosis  time series  autoregressive integrated moving average model（ARIMA）,gray model（GM）,generalized regression neural network model（GRNN）,model comparison

摘 要: 摘要: 目的 对基于时间序列的三种预测模型即自回归滑动平均混合模型（ARIMA）、灰色模型（GM）、广义回归神经网络模型（GRNN）进行尘肺发病预测的适用性比较。方法 选用河北省1954—2015年62年的尘肺发病数据，前54年数据用来拟合预测，后8年数据来比较三种模型的预测效果；采用预测误差( Prediction Error,PE)、平均绝对误差(Mean Absolute Error,MAE)和平均相对误差(Mean Relative Error,MRE)评价拟合效果。结果 GM(1,1)的预测结果较差，ARIMA的MAE和MRE是三种模型中最小的，其短期预测的PE也最低；三种方法长期预测的PE都比较大，比较而言GRNN的长期预测结果最好。结论 ARIMA适用于尘肺发病的短期预测，GRNN适用于长期预测。

Abstract: Abstract：Objective The applicability of three prediction models based on time series namely autoregressive integrated moving average model（ARIMA）、gray model（GM）and generalized regression neural network model（GRNN）in incidence prediction of pneumoconiosis was compared. Methods The pneumoconiosis incidence data of Hebei Province from 1954 to 2015 were collected，the first 54 years of data was used for fitting predictions，the last 8 years of data was used for comparing the prediction effect of the three models，the prediction error（PE）、mean absolute error（MAE）and mean relative error（MRE）were used as well in the study to evaluate the fitting effect. Results The results showed that prediction effect of GM（1,1）was poor；the MAE and MRE of ARIMA were the smallest among three methods，its PE was also the lowest in short-term prediction；meanwhile，the PE of three methods were all larger in long-term prediction，but in comparison，GRNN's long-term prediction effect was the best in these models. Conclusion The results suggested that ARIMA is more suitable for short-term incidence prediction of pneumoconiosis，and GRNN seems more suitable for long-term incidence prediction of pneumoconiosis.