摘 要: 以我国尘2000—2016年肺病发病统计数据为样本数据,针对我国尘肺病发病组合预测模型的性能展开研究。方法 从国家统计数据库中选取采矿业从业人数、不同性质采矿企业数量、规模以上企业数量、原煤产量等作为模型预测的相关辅助决策因素,使用不同组合方式的灰色神经网络和灰色-广义回归神经网络预测模型,对未来全国尘肺病发病人数进行预测。结果 灰色模型的预测结果较差,其他组合方式的预测模型精度不能达到理想的预测精度,而6维输入的灰色-广义回归神经网络模型的预测精度较高,均方根误差(RMSE)也是所有组合模型中最小的。结论 通过对比分析和模型评价验证了多维输入的灰色-广义回归神经网络预测模型在尘肺病发病人数预测上的可行性和有效性。 |
关键词: 尘肺病 发病人数 灰色神经网络 广义回归神经网络 预测 |
中图分类号:
文献标识码:
|
基金项目: 国家重点研发计划资助项目(2016YFC0801707,2017YFC0805208) |
|
Comparative study of multi-dimension input in pneumoconiosis incidence prediction models |
ZHENG Lin-jiang, ZHOU Long-hui, HUANG Jing, CHEN Yan-qiu, ZOU Zhe, LI Chun-hui
|
College of Computer Science, Chongqing University, Chongqing 400044 China
|
Abstract: Take the statistical number of pneumoconiosis incidence in China from 2000 to 2016 as sample data, to study the performance of the combination forecasting models of pneumoconiosis incidence. Methods The data such as number of mining industry workers, number of different types of mining companies, number of enterprises above designated size, raw coal output, etc. were obtained from national statistical database as the related auxiliary decision factors for model prediction; different combinations of gray neural network and gray-generalized regression neural network prediction model were used to predict the national morbidity number of pneumoconiosis in the future. Results The results showed that the results of gray model was poor, the prediction accuracy could not reach satisfied prediction accuracy; while the prediction accuracy of 6-dimensional input gray-generalized regression neural network model was quite high, and the RMSE (root-mean-square error) was the smallest in all the combined models. Conclusion Through the comparison analysis and model evaluation, the feasibility and effectiveness of multi-dimensional grey-generalized regression neural network prediction model for predicting the number of \pneumoconiosis patients were well verified. |
Keywords: pneumoconiosis morbidity number gray neural network generalized regression neural network, prediction |