Spectral Information Mining Of The Complex Solution And The Modeling Strategy Research

At present,the infrared spectral analysis technology is widely used in food,medicine,environmental monitoring and the analysis of the medical diagnosis and other fields.Especially the analysis of complex solution sample is the hot issue of the research.The interference and noise of the measured spectra of the complex solution samples are the main obstacles in the analysis of actual application.For the complex solution sample,therefore,how to excavate the spectral information to construct an accurate and stable analysis model is the key to the quantitative analysis in the actual applications.Design suitable data processing methods,spectrum information mining and spectral data set partitioning strategy and modeling scheme is a major means of promoting model performance.The access of spectrum multidimensional space,data sets partition strategy and integrated modeling scheme based on multiple information space strategy are mainly researched in the thesis.First,based on different orders derivative spectrum information space,the method of based on the strategy of consensus SPXY data sets partition strategy is put forward,termed as CSPXY.This method can make full use of the characteristic information of the different information space to enhance the rationality and representative of the data sets partition.Secondly,combining with the high quality of the derivative spectrum,the method of based on the strategy of consensus SPXY data sets partition and fusion modeling method,multiple derivative spectrum space ensemble iPLS model and based on the fusion of heterogeneous spatial spectrum iPLS model are put forward,termed as DSE-iPLS and HSEiPLS,respectively.The core idea of DSE-iPLS model is that under the premise of getting high quality derivative spectrum,construct the fusion modeling of each order derivative spectra by using interval partial least squares.And then,the construction of different order derivative spectrum fusion modeling is the further integration,means forming the final integration model.The main idea of HSEiPLS model is that under the premise of obtaining the high quality derivative spectra,the derivative spectra of different order space is divided into the same number intervals,then optimizing the intervals according to the rules,in this way the integration model is established.In order to verify the effectiveness of the proposed algorithm,the samples of beer,wine,whole blood are used for experiments.Compared with the SPXY data sets partition strategy,beer data sets are separated by CSPXY data sets partition strategy,RMSECV and RMSEP of PLS model of the derivative spectrum are lower than using SPXY data sets partition strategy to the corresponding value of the model.Compared with PLS and DSE-iPLS,RMSECV and RMSEP of the DSE-iPLS model were increased by 93.54%,23.28%and 23.28%,93.54%.Compared with PLS and DSE-iPLS model,RMSECV and RMSEP of HSEiPLS model were increased by 71.52%,29.52%and 29.52%,71.52%.Experimental results show that the presented SPXY division strategy based on the consensus theory,the iPLS model based on multiple derivative spectrum space fusion method and fusion iPLS modeling method based on heterogeneous space,all have their own characteristics and application range,thus suitable data set partitioning strategy and modeling method should be used for the samples of different complexity to achieve the purpose of better quantitative analysis.