Research On Modeatment Method Of Non-invasive Blood Glucose Concentration Measurement

In recent years,the incidence of diabetes is rising rapidly,the effective solution is to control blood sugar in a safe range,mainly using invasive medical measurement methods at present,which have no continuous measurement,so the domestic and international efforts in the monitoring system of non-invasive blood glucose detection technology of the research are among the popular microwave detection,photoacoustic spectroscopy and Raman spectroscopy etc..The above methods have great difficulties in application,and can only stay in laboratory research,and now generally begin to study near infrared spectroscopy detection method.In application,near infrared spectroscopy technology is a more advantageous detection technology.The key is that the technology can collect enough high signal-to-noise ratio spectral information to identify the absorption components of weak blood sugar,and thus can use statistical methods to establish the training and prediction model of glucose concentration.The level of human blood glucose is mainly determined by the concentration of glucose in the blood.In the research of non-invasive blood glucose detection,the concentration of glucose solution is the foundation of non-invasive blood glucose detection.Near infrared spectrum characteristics so this paper will focus on the analysis of aqueous glucose solutions of different concentrations,a detailed study of near infrared absorbance data of glucose solution corresponding to the full wavelength range,extracted by band near infrared absorbance change obviously,and from the viewpoint of physical chemistry explains why the band has the characteristic peaks.Through the baseline correction and normalization as pretreatment of glucose solution spectral data at the same time,this paper studies the chemical statistics discriminant method invalid information wavelength carrying point and the sample data matrix singular samples,mainly using the discriminant score plots.The sample data set of glucose water solution with a concentration range of 1.8%-18% and a concentration interval of 1.8% was trained and predicted.At the same time,this paper studies the method of chemical statistics to distinguish the wavelengths of invalid information in singular sample points and sample data matrix,mainly using the score graph to judge.The glucose water sample data set with concentration range of 1.8%-18% and concentration interval of 1.8% was trained and predicted.Partial least squares fitting method is used to establish prediction models for absorbance data and concentration of glucose solution,and verifies the generalization ability of the model.The absolute error and correlation coefficient of the actual value of glucose solution and the predicted value of the model are used as the evaluation indexes of the model prediction results.In order to compensate for the shortcomings of partial least squares nonlinear fitting,we chose BP neural network model,but we found that the training speed of BP network was slow,and the average time was 1 minutes.In order to improve training speed and change the family of output function family as closed set and improve nonlinear approximation ability,RBF radial basis function network is selected.Then,3 samples which are close to the frequency of absorbance signal of glucose molecules are selected as the independent variables to verify the feasibility of the model.Finally,the genetic algorithm is used to optimize the training parameters of BP and RBF networks.In order to compare the results before the genetic algorithm optimization,the same data set is used here.The model of the absorbance value and the concentration of glucose water solution in the prediction set sample is modeled.The correlation coefficient R2 is 0.99741 and 0.9984,respectively.The results show that the use of statistical anomaly processing method and genetic algorithm to optimize the RBF neural network model can be used to a certain degree of optimization model,verified the quantitative relationship between the NIR spectra and the concentration of glucose,and provides a theoretical basis for the non-invasive blood glucose detection equipment development and the improvement of measurement accuracy.