基于经验模态分解的我国布鲁菌病月发病率预测研究

摘要

Objective To develop a model for the prediction of monthly incidence of brucellosis in China based on empirical mode decomposition (EMD) and time series analysis results of the volatility characteristics of the incidence of brucellosis in China during 2004—2016, and use it to predict the monthly incidence of brucellosis in China in 2017. Methods From the websites of Data-center of China Public Health Science and Bureau of Disease Prevention and Control, National Health and Family Planning Commission of the PRC, we collected the data about the incidence of brucellosis in China from January 2004 to December 2016 and calculated the monthly incidence of brucellosis during this period. We used the data from 2004 to 2015 as training data to develop the model for the prediction of monthly incidence of brucellosis in China and adopted the data between January and December 2016 as test data to verify the model. Then, the intrinsic mode functions IMF1-IMF4 and trend term r were generated by EMD. Support vector machine was used to model IMF1-IMF4 components, and ARIMA to trend term r. The monthly incidence of brucellosis was finally obtained by weighting the model predicted values linearly. Results The range of penalty parameter c and kernel function parameter g of SVM model were 0.088 4-100.000 0 and 0.010 0-128.000 0, respectively. In the ARIMA((1,12,24),1,0) model, constant term and the autoregressive coefficient of first-order lag, twelfth-order lag and twenty-fourth-order lag were 0.002 003, 1.087 788, -0.145 494 and 0.028 783, respectively. For predicting the monthly incidence of brucellosis in China between January and November 2016, the root mean square error (RMSE),mean absolute error (MAE) and mean absolute percentage error (MAPE) of proposed method were 0.020 1, 0.016 9, 0.066 5, respectively; the errors of single SVM model based on undecomposed sequence were 0.072 2, 0.056 0, 0.197 5, respectively; the errors of single ARIMA model based on undecomposed sequence were 0.165 0, 0.156 2, 0.610 0, respectively. In addition, the monthly incidence of brucellosis in 2017 was predicted to be 0.287 0/100 thousand people-0.372 6/100 thousand people according to the proposed method. Conclusion The model for the prediction of monthly incidence of brucellosis in China based on the related incidence data using EMD and time series analysis has high prediction accuracy with minor prediction error. The mean monthly incidence of brucellosis in 2017 was predicted to be about 0.35/100 thousand people.%目的 根据我国布鲁菌病(简称布病)月发病率的波动特征,采用经验模态分解(EMD)和时间序列分析,构建布病月发病率预测模型,并预测2017年我国布病月发病率.方法 从公共卫生科学数据中心和国家卫生计生委疾病预防控制局网站,收集并计算2004年1月—2016年12月我国布病月发病率.选取2004年1月—2015年12月的数据作为训练集建模,2016年1—12月的数据作为测试集验证模型.通过EMD算法将发病率序列分解为本征模态函数(IMF)1~IMF4和趋势项,对IMF1~IMF4建立支持向量机(SVM)模型,对趋势项建立自回归移动平均模型(ARIMA)疏系数模型,最后将5个模型的输出值进行线性加权求和,得出布病月发病率预测值.结果 SVM模型的惩罚参数c的取值范围是0.0884~100.0000,核函数参数g的取值范围是0.0100~128.0000;ARIMA((1,12,24),1,0)模型中,常数项及滞后1、12、24阶的自回归系数分别为0.002003、1.087788、-0.145494、0.028783.本文方法预测2016年1—11月布病发病率的均方根误差(RMSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPE)分别为0.0201、0.0169、0.0665,序列未分解单一SVM模型预测2016年1—11月布病发病率的RMSE、MAE、MAPE分别为0.0722、0.0560、0.1975,序列未分解单一ARIMA模型预测2016年1—11月布病发病率的RMSE、MAE、MAPE分别为0.1650、0.1562、0.6100.根据本文方法计算得出2017年1—12月布病发病率预测值为0.2870/10万人~0.3726/10万人.结论 本研究根据相关发病率数据构建了基+于EMD和时间序列分析的我国布病月发病率预测模型,其预测误差较小,预测准确度较高;2017年我国布病月发病率预测值约为0.35/10万人.