AR模型和GM(1,1)模型在2型糖尿病患者血糖预测中的准确性比较

Comparison of accuracy of AR model and GM(1, 1) Model to predict blood glucose in patients with type 2 diabetes

目的研究比较AR模型和GM(1,1)模型在2型糖尿病患者的短期血糖预测中的准确性。方法 112例2型糖尿病患者均佩戴动态血糖监测系统(continuous glucose monitoring system, CGMS)连续3天,每个糖尿病患者可获得连续72 h的864个数据,从中选取符合完整的、无断点的42例糖尿病患者血糖数据作为研究对象,分别建立AR血糖预测模型和GM(1,1)血糖预测模型,应用误差分析、相关性分析、误差网格分析方法比较2种模型在提前5 min、15 min、30 min血糖预测的准确性。结果 AR模型血糖预测值和实际血糖值在提前5 min、15 min、30 min的相关系数、误差网格分析结果均高于GM(1,1)模型,均方根误差结果均低于GM(1,1)模型;AR模型和GM(1,1)模型在提前5 min时预测效果最好,提前30 min时预测效果最差。结论与GM(1,1)预测模型相比,AR模型在不同预测时间点血糖预测的准确性更高,但随着预测时间的延长,预测准确性逐渐下降。

Objective To compare the accuracy of AR model and GM(1,1)model in short-term blood glucose prediction in patients with type 2 diabetes.Methods Totally 112 patients with type 2 diabetes constantly should wear CGMS(continuous glucose monitoring system)for 3 days,and each diabetic patients can obtain 864 blood glucose values within 72 hours.We choose the whole and continuous blood glucose data series of 42 patients from those above patients with 2 type diabetes as the research object,respectively establish AR forecasting model and GM(1,1)predictive model,and apply error analysis,correlation analysis,the error grid analysis method to compare the predictive accuracy of two models in 5 minutes,15 minutes,30 minutes early.Results The correlation coefficients and EGA of the predictive blood glucose values in the AR model and actual blood glucose values in 5 mintues,15 mintes and 30 minutes early were all higher than those of GM(1,1)model and actual blood glucose,the RMSE of the AR model were lower than the GM(1,1)model;The accuracy of prediction of two predictive model in the 5 minutes early was the best,and the accuracy of prediction in the 30 minutes early was the worst.Conclusion Compared with GM(1,1)prediction model,the accuracy of AR model in predicting blood glucose at different time points is higher,but with the extension of predictive time,the forecasting accuracy gradually declines.