Research On Prediction Model Of Antimony Sulfide Grade Based On Raman Spectroscopy And Transfer Learning

Antimony sulfide grade is an important index of the production process in the antimony flotation process.The complicated operation steps of traditional chemical detection methods make it difficult to meet the needs for efficient detection in the flotation process.Therefore,by studying the problems of strong fluorescence background baseline interference,difficulty in extracting feature information,and lack of sample data in Raman spectroscopy for mineral quality detection in antimony flotation,a new quantitative analysis and model transfer method is realized to solve the challenging problem of online prediction of mineral grade based on Raman spectroscopy.The main work and innovations of this thesis are as follows:(1)Focusing on the problem of strong fluorescence background baseline signal interference in antimony sulfide Raman spectra,the spectral baseline correction method based on DSTAPLS is proposed.Under the discrete state transition strategy,the global optimization process in the background baseline of the Raman spectrum is realized.The penalized least squares method is used to automatically remove the fluorescence background baseline interference signal in the Raman spectrum by weighted smoothness and baseline fidelity.The effectiveness of the baseline correction method is verified by simulations and public data.(2)Focusing on the problem of extracting the characteristic information of antimony sulfide Raman spectrum,which leads to the difficulty in obtaining antimony sulfide grade online,the quantitative analysis and modeling method based on AP-DE-PLTS is proposed.The characteristic wavelength combination with high correlation is selected under the mutation,crossover,and selection strategy of AP-DE.A quantitative analysis and prediction method for all wavelength combinations is established by PLTS.The experimental results show that the method can accurately predict antimony sulfide grade.(3)Focusing on the problem that the Raman spectrum data of antimony sulfide raw ore with complex composition is small,which leads to the problem of low prediction accuracy of its grade quantitative analysis model,a model migration method based on PCA-ELMTr Ada Boost combined with transfer learning is proposed.After PCA extracts low-dimensional features of spectral data.Tr Ada Boost is selected to continuously update each sample’s weights in the target and source domains in the iterative process.The predicted value of grade was calculated by the weighted average strategy using the ELM model.The experimental results show that the proposed method improves the prediction performance of antimony sulfide raw ore grade greatly.Figures 40,Tables 5,References 85...