基于RIME和1DCNN-LSTM-Attention的无创血糖预测模型研究

Research on non⁃invasive blood glucose prediction model based on RIME and 1DCNN⁃LSTM⁃Attention

实现无创血糖检测对于糖尿病患者来说具有重要意义,然而目前市面上的无创血糖仪存在检测精度不高的问题。为了提高无创血糖检测的准确度,基于近红外无创血糖检测仪,构建了1DCNN-LSTM-Attention混合预测模型,同时引入了霜冰优化算法(RIME)。该模型通过一维卷积神经网络(1DCNN)提取数据中的局部特征,将所提取的特征向量作为长短期记忆(LSTM)网络的输入,捕捉数据中的依赖关系;采用注意力机制(Attention)为LSTM的输出赋予不同的权重,增强关键信息提取;通过RIME算法优化模型参数,避免陷入局部最优解。结果表明,引入RIME的1DCNN-LSTM-Attention混合模型预测效果优于1DCNN、LSTM、1DCNN-LSTM、1DCNN-LSTM-Attention等模型,预测血糖值与有创血糖值的平均绝对误差为0.121 0,均方误差为0.018 6,相关系数达到了0.982 3。该模型有助于提高近红外无创血糖检测的精确度和可靠性。

The realization of non-invasive blood glucose detection is of great significance for diabetes patients.However,the current non-invasive blood glucose meters on the market have the problem of low detect ion accuracy.In order to improve the accuracy of non-invasive blood glucose detection,a 1DCNN-LSTM-Attention hybrid prediction model is constructed based on a near-infrared non-invasive blood glucose detector,and the rime ice optimization algorithm(RIME)is introduced.The model can extract local features from the data by means of 1-dimensional convolutional neural network(1DCNN),and the extracted feature vectors are used as the inputs of the long short term memory(LSTM)network,to capture dependencies in the data.The attention mechanism(Attention)can be used to assign different weights to the output of LSTM to enhance key information extraction.The RIME algorithm is used to optimize model parameters so as to avoid getting stuck in local optima.The results show that the prediction effect of the 1DCNN-LSTM Attention mixed model with RIME is better than that of the 1DCNN-LSTM single model and the 1DCNN-LSTM and 1DCNN-LSTM Attention models.The mean absolute error between the predicted blood glucose value and the invasive blood glucose value is 0.1210,the mean square error is 0.0186,and the correlation coefficient can reach 0.9823.The model is helpful for improving the accuracy and reliability of near-infrared non-invasive blood glucose detection.