Detection Method Of Ion Concentration By UV-vis Spectroscopy Based On Information Lossless Feature Extraction And Its Application

Feature extraction is a key step in the quantitative analysis of UV-Vis spectroscopy.Due to the high overlap of UV-Vis spectral signals of heavy metal ions in solution,conventional feature extraction methods are prone to serious feature loss when projecting high-dimensional variables into low-dimensional variable space,which degrades the prediction performance of the model.Because of the large difference of spectral signal distribution among different instruments,the spectral detection model established in laboratory cannot be directly applied to the field micro-spectrometer.Aiming at the above problems,this paper studies the detection modeling and model transfer method of heavy metal ion concentration in zinc liquid based on UV-Vis spectroscopy based on information lossless feature extraction.The main work contents are as follows:(1)To solve the problems of overlapping spectral signals,serious noise interference,and easy information loss of feature in conventional spectral analysis,an ion concentration modeling method based on information lossless feature extraction is proposed.Firstly,the distribution of spectral absorbance is studied by nonlinear independent component estimation method.Then,the spectral data is nondestructively transformed into independent potential variables satisfying the Gaussian distribution.Finally,the partial least squares regression model of independent potential variables and heavy metal ion concentration is established.The results show that the proposed method can extract effective potential features with information lossless,which significantly improves the detection accuracy of copper and nickel ions in zinc liquid.(2)Aiming at the problem of losing feature information and difficulty in ensuring model accuracy during model transfer,a spectral model transfer method based on feature scale punishment and domain adaptation is proposed.Firstly,the maximum mean difference is used as the metrics to minimize the distance between the spectral features of the source domain and the target domain in the common feature space.Then,an unsupervised domain adaptive migration model is constructed based on the improved feature scale penalty.The results show that the proposed method improves the accuracy and stability of the spectral detection model after model transfer and solves the problem that the detection model is difficult to share among different instruments.(3)To solve the problem that the detection procedure of polymetallic ion concentration is complicated and it is difficult to meet the demand of real-time detection,a heavy metal ion concentration detection system is developed.Firstly,we selected the appropriate technical framework according to the performance requirements,and analyzed the detection requirements of on-site process and the user demand,designed the overall functional framework of the system,and implemented the corresponding technical modules.Finally,the performance of the designed system is verified by experiments.