Spectral Analysis And Modeling Research Of Mixed Solution Based On Singular Perturbation Technique

As an important mode of online detection and control, near infrared spectroscopy technology can be rapid non-destructive testing for material composition. The spectral preprocessing techniques and modeling analysis have become an important means of detecting material composition. In this thesis, by combining the spectral preprocessing techniques with partial least squares regression analysis method, then establish calibration models for the obtained infrared spectrum. To evaluate the model through cross-validation root mean square error and root mean square error of calibration set. And then to validate the predictive ability of the model through residuals, the coefficient of determination and root mean square error of prediction set. On the basis of that we do preprocessing and modeling analysis to infrared spectra of the mixed solution.Thanks to Savitzky-Golay preprocessing algorithm has the disadvantages of that it can only deal with an odd number of data points, multi-parameter selection and there are some problems in handling the boundary points. This thesis presents a new spectral preprocessing algorithm based on singular perturbation techniques, includes the design of high gain integrator chain differentiator and the design of a new algorithm for smoothing and differentiation. Disturbances of the high gain differentiator exist only in the last equations. Noise signal can be sufficiently suppressed by integrating the role of each layer. When the value of the perturbation parameter is small enough, this design linear dynamic tracking-differential system, signal x1(t)can be fully tracked the input spectral datau(t). The system selected parameters can be effective optimization by using inverse Taylor series method. For the new design algorithm of spectral smoothing and to strike first derivative, when the perturbation parameter unlimited approaches zero, signal x1(v) can be fully tracked and smooth spectral data u(v). And x2 (v) is the first derivative spectrum of the spectral signal. This design preprocessing algorithm can obtain the infrared spectrum with a higher signal to noise ratio.Two datasets infrared spectrum of glucose solution and beer samples were collected by using Fourier transform infrared spectrometer. By using Savitzky-Golay algorithm and perturbed system algorithm respectively to do pretreatment for the obtained infrared spectrum. Using leave-one-out cross validation method and to determine the optimal size of principal components by using root mean square error of cross validation. KS algorithm is used to divide the sample sets. Establishing calibration models based on partial least squares regression analysis method and then to predict the models. Verifying the ability to fitting and predict of the model by using model evaluation parameters, to test the feasibility of the new algorithm.In summary, we designed the new algorithm that can track and smooth infrared spectral data simultaneously. And perturbation parameters are optimal selected.The algorithm is simple, and we can use the point-by--point method to meet the demands of smooth and obtaining the all-order derivatives by adjusting a perturbation parameter. The algorithm also has a good prospect and can be achieved further optimization and study.