Research On Modeling Methods Of In-situ Analysis Of Oil Yield In Oil Shale And Its Application Based On Near-infrared Spectroscopy Technology

Oil shale,an important supplementary energy(oil substitute),mainly consists of shale oil,moisture and minerals.Oil yield and moisture content are important indicators for evaluating the quality of oil shale.The conventional oil yield detection method is complicated and inefficient,which seriously restricts the efficiency of oil shale resource exploration and mining.Therefore,it is of great significance to carry out in-situ analysis modeling method and its application research for oil yield of oil shale by using high-efficiency and non-destructive near-infrared spectroscopy.The national potential project of oil and gas resources(the research on the method of on-site oil yield detection of oil shale basing on the near-infrared spectroscopy technology,OSR-02-04),and the major special project of provincial science and technology department of Jilin Province(the development of the specialized near-infrared spectroscopy field detector for oil yield of oil shale,20116014),conduct the research on modeling methods of in-situ analysis of oil yield in oil shale and its application of moisture basing on near-infrared spectroscopy technology.The project team has developed the portable near-infrared spectrometer for oil shale,and has carried out research on the analysis method of near-infrared spectroscopy.However,the model of the method is of low accuracy and poor stability.In this paper,the main component of oil shale(oil yield,moisture content)is taken as the research object.For complex in-situ oil shale samples,the optimizes contrast technology of the modeling method is carried out from spectral data preprocessing,abnormal sample rejection,wavelength selection and model optimization.In order to improve the accuracy of the detection for oil shale,the modeling method was carried out.It would apply to the moisture of oil shale,extend the parameters of oil shale component detection,provide technical parameters for oil shale in situ mining,promote the development of oil shale parameter detection in China by near-infrared spectroscopy and provide technology support for the application of oil shale based on near-infrared spectral analysis technique.The specific research contents are as follows:1 The imitated sample of oil shale was used in the research at the influence from data form of diffuse reflection spectral(reflectance,absorbance and K-M function)and the signature absorption band of the near-infrared spectrum(full spectrum,first order frequency multiplication,first order frequency combination and combination interval)to the partial least square(PLS)model precision of oil yield in oil shale.The experiment result shows that,as for the imitated sample,the accuracy of the model based on the reflectance and absorbance is higher in the three spectral data forms,and the accuracy of the model based on full spectrum and combination interval are higher than the other models.Therefore,appropriate spectral data forms and modeling intervals can improve the accuracy of the PLS model.2 For the near-infrared spectral data of in-situ samples of oil shale,the wavelet transform processing method basing on PLS modeling and optimization is proposed.In order to conduct the complex background information of in-situ oil shale samples,the different wavelet bases and decomposition scales of continuous wavelet transform and discrete wavelet transform were studied which combined with the comparative evaluation of the PLS model precision.Basing on two in-situ sample databases of oil shale,the result shows that the decisive factors of the PLS model which base on the near-infrared spectral data of oil shale in-situ samples processed the wavelet transform with optimization parameter the in-situ sample of is 0.80(the best of the others is 0.72),0.92(the best of the others is 0.89).It indicates that the processing method of spectral data enhance the effective information of the oil content spectrum data of the in situ oil shale samples and inhibit the complex background information.3 To optimize the parameters of the method and reject the abnormal samples,the Monte Carlo sampling(MCS)was applied to the near-infrared spectroscopy analysis of oil yield in in-situ oil shale samples.Basing on two in-situ oil shale sample databases,the result shows that the PLS model precision processed by the principal component analysis of Mahalanobis distance(PCA-MD)is higher than that the PLS model precision processed by the MCS.Besides,the root mean squared errors of prediction for the PLS model precision processed by the MCS are 0.34,0.37 lower than the PCA-MD one.Therefore,it is determined that the MCS method is an abnormal sample rejection method for the modeling part for the spectral data of oil yield of in-situ oil shale samples and it could effectively improve model accuracy.4 To screen the near-infrared spectroscopy modeling data of oil shale in-situ samples,wavelength selection methods that combine the PLS model with colony optimization(ACO)was put forward.Four wavelength selection methods of Monte Carlo non-information variable elimination method(MCUVE),genetic algorithm(GA),competitive adaptive reweighted sampling(CARS)and ant colony optimization(ACO)were optimized the parameters by the PLS modeling precision comparison.It was found that ACO was the most effective way to improve the analysis accuracy for oil yield of in-situ oil shale.A comparative experiment applying the four methods to 2 databases of in-situ oil shale sample was taken.The result shows that the decisive factor of the PLS model basing on the ACO wavelength selection method raised to 0.70,0.86.And the wavelength preference was 17% and 10%.Therefore,the ACO method not only simplifies the calibration model but also improves the model precision of the model.5 To improve the model precision of in-situ oil shale sample,the least squares support vector machine(LSSVM)was applied to the near-infrared spectroscopy of oil yield of oil shale.And the ‘optimized wavelet transform + optimized ant colony optimization + partial least support vector machines' method(WT+ACO+LSSVM)basing on the robust modeling method of optimizing contrast technology.Then the comparative experiment for model precision and model stability between the PLS modeling method and the back propagation neural network(BPANN)based on LSSVM was operated.The result shows that decision factors and mean decision factors of the three modeling methods are 0.53,0.51,0.32 and 0.86,0.78,0.77.And the determining coefficient variance for oil yield model of the LSSVM modeling method and the BPANN modeling method is 0.01,0.16 and 0.02,0.18.It is obviously that the LSSVM method not only improves the model precision but also ensures the stability of the model.Besides,a model comparative experiment for the data screening method of different data processing method with optimized ant colony optimization basing on LSSVM was taken.The result shows that decision factors of 2 databases of in-situ oil shale sample basing on the ‘WT + ACO + LSSVM' method is higher than the decision factors basing on other methods,which are 0.84 and 0.94.6 To expand the detection parameters of oil shale and provide more technical parameters for oil shale in-situ mining,the ‘WT+ACO+LSSVM' modeling method was applied to moisture detection of oil shale.The result of comparative experiment shows that the model precision basing on LSSVM is 0.49 and the moisture model precision basing on ‘WT +LSSVM' with optimized wavelet transform is 0.89.Besides,the moisture model precision basing on ‘ACO +LSSVM' with ACO is 0.56;the wavelength preference rate is 13%.And the moisture model precision basing on ‘WT+ACO+LSSVM' is 0.94;the wavelength preference rate is 18%.It shows the effectiveness of the modeling method for near-infrared spectroscopy of moisture in oil shale.7 For requiring the needs of in-situ and online detection for oil shale,a field-specific analysis software of oil yield in oil shale for the portable special near-infrared spectrometer of oil shale was designed.This software is developed based on Matlab language.The special software includes four modules: sample collection and analysis,database operation,model building and data transmission.It can realize on-site measurement and processing analysis of sample spectrum,processing and analysis of batch spectral data,establishment and modification of database,establishment and evaluation of model.