基于Transformer的短期血糖预测方法研究

Research on short-term blood glucose prediction methods based on Transformer

目的:血糖预测在糖尿病患者的自动化治疗中起着关键作用,为了提高血糖预测精度,更好地满足临床使用需求。方法:本研究结合DirectNet糖尿病临床数据集,提出了一种基于Transformer的短期血糖预测模型,同时分析了在不同输入序列长度下的Transformer模型预测性能;最后,为确定模型的有效预测范围,采用克拉克误差网格分析工具对未来一定时间梯度内的模型预测结果进行了误差分析。结果:基于Transformer的短期血糖预测模型取得了较好的血糖预测效果,输入序列长度为20 min时预测效果最优,所得血糖预测结果的均方根误差(RMSE)和平均百分比误差(MAPE)均优于LSTM与GRU模型。结论:基于Transformer的短期血糖预测模型的预测精度较高,并且其预测结果能够满足临床精度要求。

Aims:Blood glucose prediction plays a key role in the automated treatment of diabetic patients.This research aims to improve the accuracy of blood glucose prediction and better meet the clinical needs.Methods:Based on the DirectNet diabetes clinical data set,a short-term blood glucose prediction model based on Transformer was proposed.At the same time,the prediction performance of the Transformer model under different input sequence lengths was analyzed.Finally,in order to further determine the effective prediction range of the model,the error analysis of the model prediction results within a certain time gradient in the future was carried out by using the Clark error grid analysis tool.Results:The short-term blood glucose prediction model based on Transformer achieved good blood glucose prediction results.The prediction was the optimal with the input sequence length of 20 min.The root mean square error(RMSE)and the mean percentage error(MAPE)of the obtained blood glucose prediction results were better than those of LSTM and GRU models.Conclusions:The short-term blood glucose prediction model based on Transformer has high prediction accuracy;and its prediction results can meet the clinical accuracy requirements.