| Shale gas is a newly developed and important clean energy source.However,a large amount of drilling fluid waste water produced during the mining process seriously pollutes the environment.The waste liquid treatment needs fast on-line analysis to ensure real-time monitoring and control.Near infrared(NIR)is a non-destructive,environmentally friendly and rapid detection method.There is a complex indirect relationship between the NIR simple and the values to be measured.Through the model calibration,the corresponding relationship between the measured data and the content in the analysis system should be established.However,the spectral acquisition process is affected by the changes in the test objects,aging of the instrument and disturbance of the environment.When the test process evolves slowly,the effect of the calibration model based on the early data is not guaranteed.The traditional method involves supplementing the sample database and reestablishing the calibration model at a fixed period to ensure its adaption,but the updated calibration model contains a large number of early state information of the spectrometer and cannot accurately represent the latest measurement state of the instrument.It manifests itself in that the performance of the multivariate calibration model established in a certain instrument or a certain measurement conditions,abnormal work in another instrument or under different measurement environment conditions.Therefore,how to establish a long-term,stable and effective calibration model is the key for the qualitative and quantitative on-line analysis of near-infrared spectroscopy at present,and directly determines the availability of the technology.Based on this,the dissertation focuses on the near-infrared spectroscopy of waste liquid and studies the main aspects affecting the dynamic multivariate model calibration in near-infrared spectroscopy.The main contents include:(1)Improvement of the method of selecting wavelength of near infrared spectrumAiming at the serious collinearity problem of near-infrared spectroscopy data,a method of feature wavelength selection based on the combination of partial least squares(PLS)and false nearest neighbors(FNN)was proposed.Firstly,partial least square method is used to extract the principal components of spectral information which are closely related to the concentration of chemical oxygen demand(COD)in the waste liquid to meet the independent nature required for establishing the similarity measure of the false nearest neighbors and keep the maximum correlation between wavelength selection and concentration information.Then,the correlation between the distance of each original wavelength point before and after the selection of the original wavelength points is calculated by means of the FNN method.After the original wavelength characteristics are obtained,the explanatory power of the COD concentration variables is sorted,and then the weakly explanatory features are eliminated in order to search for the best prediction accurate and contains the minimum wave number of the combination of characteristic wavelengths.Compared with the full-wavelength PLS calibration method,the proposed method not only simplifies the model,but also enhances the predictive ability of the calibration model.Therefore,this study provides a new method for the selection of characteristic wavelength of near infrared spectroscopy.(2)Improvement of Calibration Method of Near Infrared Spectroscopy ModelThe establishment of multivariate calibration model is the basis for qualitative and quantitative analysis of near infrared spectroscopy.New modeling methods have been one of the hot topics in this field.In this paper,a model calibration method(KF-PLS)combining Kalman Filter(KF)with Partial Least Squares(PLS)is proposed to make the calibration model adaptive.First,the principal factor of partial least-squares regression is taken as the state variable of KF algorithm.The newest standard sample to be measured is taken as the observed variable of KF,a new standard sample is added,the PLS model is modified for one time and a forgetting factor is introduced to gradually forget the older samples.In fact,the problem has been translated into the state variable's estimation process.The method is applied to the determination of COD in waste liquid by near-infrared spectroscopy measured by different spectrometers.The results show that the proposed method can effectively track the state and the external sample information of the dynamically evolving NIR spectrometer.Compared with the PLS algorithm,this method can improve the adaptability of the model.(3)Improvement of stability of near infrared spectral calibration modelWhile the number of samples to be tested increases,the calibration model established by the KF-PLS method may be unstable.This is caused by near-infrared spectroscopy measurement noise measurement,in order to obtain accurate KF-PLS model,the need to calculate accurate measurement noise statistics.Based on the idea of Gamma Test(GT),this paper first estimates the input and output data in real time using attenuated memory GT to get accurate observed noise estimates and then uses KF-PLS to model.The method was validated by a total of 600 near infrared spectroscopy samples of waste drilling fluid measured by different instruments and compared with KF-PLS algorithm.The results show that the proposed algorithm improves the stability of the KF-PLS calibration model of near-infrared spectroscopy and COD of waste liquid.In summary,the study proposes: 1)a wavelength selection method combining PLS and FNN;2)a KF-PLS-based dynamic evolution model calibration method;3)an enhanced stability based on GT-KF-PLS dynamic evolution model calibration method.The correlative method of multivariate dynamic evolution model for on-line analysis of near-infrared spectroscopy studied in this paper provides an effective way for on-line analysis,real-time monitoring and real-time control of wastewater treatment. |
PDF Full Text Request