两种神经网络模型预测煤工尘肺发病工龄的性能比较
A Comparative Study on Two Neural Network Models for Predicting Working Age of Pneumoconiosis Occurrence in Coal Workers
-
摘要:
目的 探讨反向传播(back propagation, BP)神经网络模型与径向基函数(radial basis function, RBF)神经网络模型在煤工尘肺发病工龄预测中预测性能的优劣。 方法 采用SPSS 19.0中的BP和RBF神经网络模型对研究数据进行预测分析。采用均方根误差、平均相对误差和平均绝对误差对两种模型的预测结果进行比较分析,从而得到比较各模型预测性能的目的。 结果 由BP神经网络模型进行预测可以得出发病工龄预测值和真实值的散点分布图大致位于从原点起始的45度线上,符合理想状态下值的分布情况,而RBF神经网络模型的分布则较为混乱。RBF和BP神经网络模型真实值与预测值之间的差异均无统计学意义, t值分别为0.231、0.530, P值分别为0.817、0.596。RBF和BP神经网络模型的均方根误差分别为3.51、1.89;平均相对误差分别为12%、6%;平均绝对误差分别为2.76、1.42。 结论 实证表明,在煤工尘肺发病工龄的预测中, BP神经网络模型的预测性能优于RBF神经网络模型,有较高的拟合和预测精度。 Abstract:Objective To compare the pros and cons of back propagation (BP) neural network (BPNN) and radial basis function (RBF) neural network (RBFNN) in prediction performance on working age of pneumoconiosis occurrence in coal workers. Methods BPNN and RBFNN were constructed using SPSS 19.0 software. Root of mean square error, average relative error, and mean absolute error were applied to compare the predicting outcomes of the two models. Results Combined the predicted values and the true values in the same scatter diagram, the distribution of values predicted by the BPNN model was concentrated around a 45-degree line, indicating an approximate shape of the ideal state; but the distribution of values predicted by the RBFNN model was in disorder. There was no significant difference between the true value and the predicted values by the BPNN model or the RBFNN model (t=0.530 and 0.231 respectively, P=0.596 and 0.817 respectively). The root of mean square error, average relative error, and mean absolute error of the RBFNN model and the BPNN model were 3.51 and 1.89, 0.12 and 0.06, 2.76 and 1.42, respectively. Conclusion For the working age of pneumoconiosis occurrence in coal workers, the prediction performance of BPNN is superior to RBFNN, owing to its better fitting to true values and prediction accuracy.
下载:
