论文部分内容阅读
[目的]探讨灰色-广义回归神经网络组合模型[GM(1,1)-GRNN]在我国尘肺病发病人数预测中的应用,并比较其与灰色模型(GM)和反向传播网络(BPNN)模型的预测效果。[方法]收集2003—2012年我国尘肺病发病资料,用SAS9.3建立GM(1,1)模型,用Matlab 8.0建立BPNN模型和GM(1,1)-GRNN组合模型,并用2013年的数据评价模型的预测效果。[结果]GM(1,1)模型拟合及预测的平均相对误差(MRE),平均误差率(MER),均方误差(MSE)和平均绝对误差(MAE)分别为12.041%,0.122,4 999 319.100,1 781.100和20.033%,0.200,2 151 104.000,4 638.000;BPNN模型拟合及预测的MRE,MER,MSE和MAE分别为9.891%,0.124,3 615 099.600,1 802.000和6.932%,0.069,2 576 025.000,1 605.000;GM(1,1)-GRNN组合模型拟合及预测的MRD,MER,MSE和MAE分别为6.490%,0.069,1 900 198.400,1 001.200和3.939%,0.039,831 744.000,912.000。GM(1,1)-GRNN组合模型预测的2014—2015年我国尘肺病的发病人数分别为23 768和23 434。[结论]GM(1,1)-GRNN组合模型的拟合及预测效果优于GM(1,1)模型和BPNN模型。
[Objective] The purpose of this study was to explore the application of gray-generalized regression neural network combined model [GM (1,1) -GRNN] in forecasting the number of pneumoconiosis patients in China and to compare it with gray model (GM) and back propagation network (BPNN) Predictive effect of the model. [Methods] The data of pneumoconiosis in our country from 2003 to 2012 were collected. The GM (1,1) model was established with SAS9.3. BPNN model and GM (1,1) -GRNN combination model were established with Matlab 8.0. Estimate the predictive effect of the model. [Results] The average relative error (MRE), mean error rate (MER), mean square error (MSE) and mean absolute error (MAE) of GM (1,1) model fitting and prediction were 12.041%, 0.122,4 999, 319.100, 1781.100 and 20.033%, 0.200, 2151 104.000, 4638.000; the MRE, MER, MSE and MAE of BPNN model fitting and prediction were 9.891%, 0.124, 36159999, 1 802.000 and 6.932%, 0.069 , 2 576 025.000 and 1 605.000 respectively. The fitted and predicted MRD, MER, MSE and MAE of GM (1,1) -GRNN model were 6.490%, 0.069,1 900 198.400,1 001.200 and 3.939%, 0.039,831 744.000,912.000. The number of pneumoconiosis cases predicted by GM (1,1) -GRNN combination model in 2014-2015 were 23 768 and 23 434, respectively. [Conclusion] The fitting and forecasting effect of GM (1,1) -GRNN combination model is superior to GM (1,1) model and BPNN model.