小样本条件下的血糖浓度预测算法研究

Study on Prediction Algorithm of Blood Glucose Concentration under Small Sample Condition

鉴于现有的血糖浓度预测模型在小样本情形下仍有不足,为更好地精确预测糖尿病患者血糖浓度在未来一段时间的变化情况,提出一种基于小样本条件的血糖浓度预测算法。算法可依照t分布检验准则剔除待分析血糖序列的异常数值,利用三次样条函数插值方法扩充血糖样本,最终基于广义回归神经网络实现血糖序列的浓度预测。实验结果表明,算法在小样本的条件下获得较好的预测性能,具有一定的实际应用价值,可在保证预测精度的同时,使血糖序列采样时间间隔大大延长,为保持患者血糖数值稳定在正常生理范围内提供有力保障。

In view of the shortcomings of existing blood glucose concentration prediction models in small samples,in order to accurately predict the change of blood glucose concentration of diabetic patients in the future,a prediction algorithm of blood glucose concentration based on small sample condition is proposed.The algorithm can eliminate the abnormal values of the blood glucose sequence to be analyzed according to the t-distribution test criterion,expand the blood glucose samples by cubic spline interpolation method,and finally realize the concentration prediction of the blood glucose sequence based on generalized regression neural network.The experimental results show that the algorithm has good prediction performance under the condition of small samples,and has certain practical application value,which can ensure the prediction accuracy while greatly extending the sampling time interval of blood glucose sequence,and provide a strong guarantee for keeping the blood glucose value of patients stable in the normal physiological range.