Research On Application Of Chemometrics To Microwave Plasma Torch Atomic Emission Spectrometry

The Microwave Plasma Torch(MPT)is a new type of excitation source developed after the Inductively Coupled Plasma(ICP).As an original technology in China,it has attracted wide attention and recognition from scholars around the world since it was invented in 1985.Recent studies have shown that argon,helium,nitrogen,and even air plasma can be formed at kilowatt-scale microwave power,and the detection performance of some elements is close to that of ICP.Therefore,the kilowatt MPT spectrometer has practical application prospects in food safety,environmental protection and so on.Due to the complex background of MPT atomic emission spectroscopy,wavelength shift and background interference are inevitable,which seriously affect the accuracy of measurement results of MPT spectrometry.When analyzing complex samples,it is necessary to establish a standard calibration curve of various elements,and parameter optimization of the instrument is generally regulated by a single univariate,but the sample preparation is complicated and the measurement takes a long time,which greatly limit the application and development of MPT.Based on these problems,from the perspective of chemometrics,by a series of research,such as experimental design,data preprocessing,data analysis modeling etc.,the MPT technology is analyzed in depth and applied to solve practical problems.The main innovations in the dissertation are as follows:(1)In order to obtain reliable spectral data,offset correction,spectral smoothing,feature extraction and other algorithms are proposed to correct MPT spectral background and measurement errors.For the first time,the simplex method was proposed to optimize the experimental condition parameters for MPT spectrometer,which provided a theoretical basis for the experimental design optimization.The parameters were rapidly optimized and the detection performance of the instrument was improved to some extent.(2)Based on the in-depth study of MPT spectral characteristics,the chemometrics method was applied to the origin tracing of ginseng systematically for the first time.By comparing the classification results of two models,Support Vector Machine(SVM)and Gaussian Process Classification(GPC),the SVM model has better classification ability and detection results,and shows the application prospect of combining MPT spectroscopy with chemometrics to solve practical problems.(3)MPT combined with Support Vector Regression(SVR)model was proposed for ginseng concentration prediction and quality analysis.Since traditional SVM adopting grid search parameters leads to low efficiency of parameters optimization,genetic algorithm(GA)and particle swarm optimization(PSO)are applied to improve the parameter optimization of SVR model.Compared with regression methods such as partial least squares(PLS),SVR has excellent prediction performance,which further reflects the application advantages of combining MPT spectroscopy with chemometrics in biological sample analysis.(4)Principal Component Analysis(PCA)is used to process and analyze the original spectral data of MPT in order to fully understand the spectral characteristics.At the same time,the probability and statistical significance of the output of the predicted value is studied by Gaussian Process Regression(GPR).The distribution range of the predicted value is given by GPR,and the reliability of the predicted result is analyzed intuitively.The influence of training set sampling on the results is discussed and analyzed.Three approximation methods,Subset of Datapoints(SD),Projected Process(PP),and Subset of Regressors(SR),are used to reduce the computational load and improve the efficiency of modeling.