卞子龙, 汤佳琪, 倪春辉, 朱宝立, 张恒东, 丁帮梅, 沈涵, 韩磊. 应用ARIMA-GRNN组合模型分析江苏尘肺病发病情况[J]. 环境与职业医学, 2019, 36(8): 755-760. DOI: 10.13213/j.cnki.jeom.2019.19046
引用本文: 卞子龙, 汤佳琪, 倪春辉, 朱宝立, 张恒东, 丁帮梅, 沈涵, 韩磊. 应用ARIMA-GRNN组合模型分析江苏尘肺病发病情况[J]. 环境与职业医学, 2019, 36(8): 755-760. DOI: 10.13213/j.cnki.jeom.2019.19046
BIAN Zi-long, TANG Jia-qi, NI Chun-hui, ZHU Bao-li, ZHANG Heng-dong, DING Bang-mei, SHEN Han, HAN Lei. Analysis on prevalence of pneumoconiosis in Jiangsu Province using ARIMA-GRNN combined model[J]. Journal of Environmental and Occupational Medicine, 2019, 36(8): 755-760. DOI: 10.13213/j.cnki.jeom.2019.19046
Citation: BIAN Zi-long, TANG Jia-qi, NI Chun-hui, ZHU Bao-li, ZHANG Heng-dong, DING Bang-mei, SHEN Han, HAN Lei. Analysis on prevalence of pneumoconiosis in Jiangsu Province using ARIMA-GRNN combined model[J]. Journal of Environmental and Occupational Medicine, 2019, 36(8): 755-760. DOI: 10.13213/j.cnki.jeom.2019.19046

应用ARIMA-GRNN组合模型分析江苏尘肺病发病情况

Analysis on prevalence of pneumoconiosis in Jiangsu Province using ARIMA-GRNN combined model

  • 摘要: 背景 尘肺病是中国职业人群中最常见的、危害最严重、影响面最广的职业性疾病,江苏省尘肺病发病情况十分严峻。

    目的 基于求和自回归移动平均模型(ARIMA)与灰色模型GM(1,1)、广义神经回归网络模型(GRNN)的分别组合,构建适合江苏省尘肺病预测的组合模型。

    方法 利用2006年1月—2017年12月江苏省尘肺病逐月新诊断病例资料构建尘肺病ARIMA预测模型,以2018年1—8月的数据作为模型测试值。将拟合尘肺病ARIMA模型时产生的残差值加上阈值3后,再用GM(1,1)模型拟合构建ARIMA-GM组合模型,拟合并预测尘肺发病数。将尘肺病ARIMA模型拟合值作为GRNN模型的输入值,尘肺病的真实发病数作为输出值,构建ARIMA-GRNN组合模型,拟合并预测。评价模型拟合效果的指标分别采用均方误差(MSE)、平均绝对误差(MAE)和平均相对误差(MRE)。

    结果 三种模型拟合江苏省尘肺病新病例情况的MSE依次为ARIMA-GRNN(0.3214) < ARIMAGM(0.7046) < ARIMA(0.8079),MAE分别为ARIMA-GRNN(0.3986) < ARIMA-GM(0.6324) < ARIMA(0.659 1),MRE分别为ARIMA-GRNN(0.161 2) < ARIMA-GM(0.183 8) < ARIMA(0.187 9)。三种模型预测2018年1—8月江苏省尘肺病新诊断情况的MSE依次为ARIMA-GRNN(0.084 3) < ARIMA(0.243 5) < ARIMA-GM(0.263 4),MAE分别为ARIMA-GRNN(0.234 5) < ARIMA(0.388 7) < ARIMA-GM(0.4161),MRE分别为ARIMA-GRNN(0.0981) < ARIMA(0.1086) < ARIMA-GM(0.1149)。各项指标都表明针对研究期间江苏省尘肺病发病情况建立的ARIMA-GRNN模型的拟合及预测误差最小。

    结论 在江苏省尘肺病新病例预测中ARIMA-GRNN模型优于ARIMA-GM模型和单纯的ARIMA模型。

     

    Abstract: Background Pneumoconiosis is the most common, hazardous, and extensive occupational disease in China, and the prevalence in Jiangsu Province is severe.

    Objective The study aims to establish a combined model for pneumoconiosis prediction based on the combinaton of ARIMA with grey model GM (1, 1) or generalized neural regression network model (GRNN).

    Methods The newly diagnosed cases of pneumoconiosis in Jiangsu Province from January 2006 to December 2017 were used to construct ARIMA model, and the data from January to August 2018 were used as the test values of the model. The residual values generated during the ftng of the ARIMA model were further fitted with GM (1, 1) model after adding a threshold of 3, and then the ARIMA-GM combined model was constructed to ft and predict pneumoconiosis incidence. The fted value of the ARIMA model was taken as the input value of GRNN model and the real value of pneumoconiosis was taken as the output value, and then the ARIMA-GRNN combined model was constructed to ft and predict pneumoconiosis incidence. Mean square error (MSE), mean absolute error (MAE), and mean relatve error (MRE) were used to evaluate the ftng effect.

    Results The MSEs of the three models fitting the new cases of pneumoconiosis in Jiangsu Province were ARIMA-GRNN (0.321 4) < ARIMA-GM (0.704 6) < ARIMA (0.807 9), the MAEs were ARIMA-GRNN (0.398 6) < ARIMA-GM (0.632 4) < ARIMA (0.659 1), and the MREs were ARIMA-GRNN (0.161 2) < ARIMA-GM (0.183 8) < ARIMA (0.187 9), respectvely. The MSEs of the three models predictng the pneumoconiosis incidences in Jiangsu Province from January to August 2018 were ARIMA-GRNN (0.084 3) < ARIMA (0.243 5) < ARIMA-GM (0.263 4), the MAEs were ARIMA-GRNN (0.234 5) < ARIMA (0.388 7) < ARIMA-GM (0.416 1), and the MREs were ARIMA-GRNN (0.098 1) < ARIMA (0.108 6) < ARIMA-GM (0.114 9), respectvely. All indicators showed that ARIMA-GRNN combined model had the smallest ftng and predicton error.

    Conclusion ARIMA-GRNN model is superior to ARIMA-GM model and ARIMA model in predictng pneumoconiosis incidences in Jiangsu Province.

     

/

返回文章
返回