Method for determining trace metal elements in residual oil based on partial least square method

Abstract

The invention provides a method for determining trace metal elements in residual oil based on partial least squares. The method comprises the following steps of: acquiring spectral data of a plurality of residual oil by utilizing a laser-induced breakdown spectrometer, dividing the spectral data into a calibration set and a test set, firstly, optimizing the spectral data of the calibration set by a one-leaving cross validation method to obtain an optimal latent variable, and analyzing the influence of different preprocessing and variable selection methods on the prediction performance of a partial least square model; and establishing a partial least squares correction model for characteristic variable selection after optimal pretreatment, and predicting the content of the residual oil trace metal elements in the test set. According to the method, the spectral data of the calibration set is optimized through a one-leaving cross validation method, the obtained optimal latent variable is used for establishing the partial least square method calibration model based on the preprocessed feature interval, the content of the trace metal elements in the residual oil of the test set is predicted, the partial least square method calibration model is optimized, and the residual oil content of the test set is predicted. And the accuracy of the partial least square method correction model is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of spectral analysis, and in particular relates to a method for determining trace metal elements in residual oil based on partial least squares method. Background Art

[0002] Residue oil is a byproduct produced during the oil processing process, which contains heavy components that cannot be extracted from light oil products during the oil refining process. For example, trace metal impurities in residue oil will cause the quality of oil products (light oils such as diesel and gasoline) to deteriorate, which will have a significant impact on oil processing, safe transportation and the economic benefits of oil refining. In the process of catalytic cracking of residue oil, the deposition of Ni and V on the catalyst will block the micropores on the catalyst surface and reduce the activity of the catalyst. In addition, the nickel elements deposited on the catalyst will promote carbon deposition and dehydrogenation reactions, thereby mainly changing the selectivity of the catalyst, causing the selectivity of the target product to deteriorate and reducing the recovery rate of light oil. In addition, Ni and V will promote the acceleration of coking during thermal processing, deteriorate product distribution and product quality, form corrosive acidic compounds under high temperature and high pressure, and cause serious wear on some equipment. Therefore, the analysis of nickel and vanadium in residue oil is of great significance for fuel quality control, equipment maintenance and optimization of residue oil processing. At present, conventional detection methods for determining the content of trace metal elements in residual oil include X-ray fluorescence spectrometry, inductively coupled plasma spectrometry, flame atomic absorption spectrometry, etc. These technologies require sample pretreatment, have slow detection speeds, and cause damage to samples. In addition, the equipment is large and cannot achieve real-time, online analysis of trace metal elements in residual oil.

[0003] Laser Induced Breakdown Spectroscopy (LIBS) is an emerging technology used for quantitative analysis of metal elements. It has the technical characteristics of simple operation and no need for complex sample pretreatment. LIBS can quantitatively and qualitatively analyze the samples to be tested, and has the advantages of fast analysis speed, easy use and no need for complex sample pretreatment. Therefore, LIBS technology is considered to be one of the most promising analytical methods. In recent years, LIBS technology has been widely used in soil, metallurgy, chemical industry, medicine, geology and agriculture. At present, the application of LIBS technology mainly includes the identification of oil products, the content of asphalt in oil sands and the detection and analysis of element content in soybean oil. The above studies show the feasibility of LIBS combined with partial least squares method for quantitative analysis of trace metal elements in residual oil. How to combine LIBS with partial least squares method for quantitative analysis of trace metal elements in residual oil has not been reported. There is an urgent need for a fast, non-destructive and accurate method for quantitative analysis of trace metal elements in residual oil, which provides a theoretical basis and technical support for residual oil quality detection and the promotion of laser induced breakdown spectrometer. Summary of the invention

[0004] The purpose of the present invention is to provide a method for determining trace metal elements in residual oil based on partial least squares method.

[0005] The present invention is achieved through the following technical solutions:

[0006] The present invention relates to a method for determining trace metal elements in residual oil based on partial least squares method, comprising the following steps:

[0007] Step 1: collect 10 kinds of residual oil samples, calcine the residual oil samples in a muffle furnace at 400°C for 5 hours, weigh the calcined residual oil samples and grind them into powder in a mortar; mix them in a mass ratio of 1:1 and 1:2 until fully uniform, and obtain 20 residual oil samples; among them, 10 samples are 1:1 and 10 samples are 1:2;

[0008] Step 2: Using a laser induced breakdown spectroscopy instrument to collect spectral data of 20 residual oil samples;

[0009] Step 3: Divide the 20 residual oil sample sets into 14 calibration set samples and 6 prediction set samples based on the KS algorithm;

[0010] Step 4: construct a partial least squares model based on 14 calibration set samples, and optimize the latent variables of the partial least squares model based on leave-one-out cross-validation to determine the optimal value of the latent variable;

[0011] Step 5: Based on the normalization method of the maximum LIBS spectrum intensity of the residual oil sample, mean centering, multivariate scattering correction, standard normal transform, derivative method, and wavelet transform method, the above methods are combined in pairs to preprocess the laser induced breakdown spectrum, and the optimal preprocessing method is determined based on leave-one-out cross validation;

[0012] Step 6: Use the joint interval partial least squares method to further screen the characteristic variables of the residue oil LIBS spectral data, and optimize the number of intervals and the number of interval combinations of the joint interval partial least squares method based on the leave-one-out cross-validation method.

[0013] Preferably, in step 2, the spectral data collection conditions are: carried out under indoor lighting conditions, the indoor temperature is 22-26° C., and the spectral range of the laser induced breakdown spectroscopy instrument is 200-500 nm.

[0014] Preferably, in step 2, the specific method of spectral data collection is: randomly selecting 50 LIBS spectral data test points in each residual oil sample, and then taking the average value of the collected spectral data as the modeling analysis data of the sample.

[0015] Preferably, in step 4, the evaluation parameters for optimizing the latent variables of the partial least squares model based on leave-one-out cross-validation are the coefficient of determination and the root mean square error.

[0016] Preferably, in step five, the evaluation parameters for determining the optimal preprocessing method based on leave-one-out cross-validation are the coefficient of determination and the root mean square error.

[0017] Preferably, in step six, the evaluation parameters for optimizing the number of intervals and the number of interval combinations of the joint interval partial least squares method based on the leave-one-out cross-validation method are the determination coefficient and the root mean square error.

[0018] Preferably, in step six, the optimization result of the number of intervals is 20 to 40.

[0019] Preferably, in step six, the optimization result of the number of interval combinations is 2 to 5.

[0020] The principle of the method of the present invention is as follows: a laser induced breakdown spectrometer is used to collect spectral data of a plurality of residual oils, which are divided into a calibration set and a test set. Firstly, the spectral data of the calibration set is optimized by a leave-one-out cross-validation method to obtain the optimal latent variable, and the influence of different pretreatment and variable selection methods on the prediction performance of the partial least squares model is analyzed. A partial least squares correction model based on feature variable selection after optimal pretreatment is established to predict the content of trace metal elements in the residual oil of the test set.

[0021] The present invention has the following advantages:

[0022] The present invention optimizes the spectral data of the calibration set by a leave-one-out cross-validation method, establishes a partial least squares calibration model with the obtained optimal latent variable, and then explores the influence of different pretreatment and variable selection methods on the prediction performance of the partial least squares model, establishes a partial least squares calibration model based on the selection of characteristic variables after the optimal pretreatment, predicts the content of trace metal elements in the residual oil of the test set, optimizes the partial least squares calibration model, and improves the accuracy of the partial least squares calibration model. Thus, a method for rapid, non-destructive, and accurate quantitative analysis of trace metal elements in residual oil is established, which provides a theoretical basis and technical support for residual oil quality detection and the promotion of laser induced breakdown spectrometer. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is the LIBS spectrum of the residual oil sample in Example 1 of the present invention; wherein (a) is a wavelength range of 200 to 500 nm; (b) is a wavelength range of 200 to 300 nm; (c) is a wavelength range of 300 to 400 nm; (d) is a wavelength range of 400 to 500 nm;

[0024] Figure 2 This is a schematic diagram of the idea of ​​constructing a residual oil quantitative analysis model in Example 1 of the present invention;

[0025] Figure 3 This is a distribution diagram of characteristic variables screened based on the joint interval partial least squares method in Example 1 of the present invention; wherein, (a) is the characteristic variable screened by the nickel element; (b) is the characteristic variable screened by the vanadium element;

[0026] Figure 4 The figures are scatter plots of the prediction performance of the residual oil quantitative analysis model in Example 1 of the present invention; wherein (a) is a scatter plot of the nickel element prediction model; and (b) is a scatter plot of the vanadium element prediction model. DETAILED DESCRIPTION

[0027] The present invention is described in detail below in conjunction with specific embodiments. It should be noted that the following embodiments are only further descriptions of the present invention, but the protection scope of the present invention is not limited to the following embodiments.

[0028] Example 1

[0029] This embodiment relates to a method for determining the content of trace metal elements nickel and vanadium in residual oil based on partial least squares method. Figure 2 As shown, the following steps are included:

[0030] Step 1: Use a laser induced breakdown spectrometer to collect spectral data of 20 treated residual oil samples. 50 test points are randomly selected during the spectral data collection, and then the collected spectral data are averaged. The spectral data collection of the laser induced breakdown spectrometer is carried out under indoor lighting conditions, the indoor temperature is 22-26° C., and the spectral range of the laser induced breakdown spectrometer is 200-500 nm. Figure 1 As shown, (a) is the wavelength range of 200 to 500 nm; (b) is the wavelength range of 200 to 300 nm; (c) is the wavelength range of 300 to 400 nm; (d) is the wavelength range of 400 to 500 nm.

[0031] Step 2: Divide the residual oil samples in step 1 into a calibration set and a test set using the KS algorithm, wherein the number of samples in the calibration set is 14 and the number of samples in the test set is 6; wherein the KS algorithm repeatedly selects the samples with the farthest and closest distances according to the Euclidean distance to enter the training set until the preset number of training set sample requirements is met; the calculation formula of the residual oil sample data set division algorithm based on the KS algorithm is specifically as follows: In this embodiment, the sample numbers of the selected prediction set are #3, #5, #9, #11, #12, and #20, respectively.

[0032]

[0033] In the above formula, d x Represents two points x in N-dimensional Euclidean space p , x q The distance between.

[0034] Step 3: Optimize the spectral data of the calibration set in step 2 by the leave-one-out cross-validation method, and use the coefficient of determination and root mean square error as evaluation parameters to obtain the optimal latent variables of the partial least squares calibration model. When optimizing the latent variables of the residual oil trace metal element PLS quantitative analysis model based on leave-one-out cross-validation, first set the initial latent variables, and then randomly select a sample in the training set for model performance evaluation by enumeration method. The remaining samples are used to construct the PLS model, and the process is repeated 14 times (until all samples in the training set are left out once). The average root mean square error obtained from 14 modelings is used as an evaluation index to evaluate the correctness of the selection of the residual oil PLS model latent variables; see Figure 3 As shown, the results show that the number of latent variables optimized for the quantitative model of nickel and vanadium elements is 6 and 7 respectively;

[0035] Step 4: Preprocess the LIBS spectra of the residual oil samples based on different preprocessing methods, including maximum intensity normalization of the residual oil LIBS spectra, standard normal transformation, multivariate scattering correction, first-order derivative, second-order derivative and wavelet transform, as well as different combinations of the above single preprocessing methods, and determine the optimal preprocessing method based on leave-one-out cross-validation.

[0036] The results show that the optimized spectral preprocessing method for nickel is multivariate scattering correction-first-order derivative (smoothing point number n=11), and the optimized spectral preprocessing method for vanadium is multivariate scattering correction-second-order derivative (smoothing point number n=5). Among them, when the proposed residual oil sample LIBS spectral data multivariate scattering correction-first-order derivative and multivariate scattering correction-second-order derivative preprocessing methods are used for residual oil LIBS spectrum preprocessing, multivariate scattering correction preprocessing is first performed, and then first-order derivative preprocessing and second-order derivative preprocessing are performed respectively to obtain the optimized residual oil LIBS spectral data, see Figure 4 As shown. Among them, the calculation formula for the preprocessing of residual oil LIBS spectrum multivariate scattering correction is as follows:

[0037] x msc =x-b0 / m

[0038] In the above formula, x msc is the spectrum after multivariate scattering correction, x is the LIBS sample spectrum, b0 is the baseline shift, and m is the offset.

[0039] Step 5: Extract characteristic variables from the optimal preprocessing method obtained in step 4, and further optimize the input variables. The optimized range of the number of intervals of residual oil LIBS spectrum data is 20 to 40, and the optimized range of the number of interval combinations is 2 to 5. For nickel, the number of intervals of the optimized joint interval partial least squares method is 20, the number of combined intervals is 4, the extracted characteristic interval numbers are 1, 2, 11 and 18, and the corresponding residual oil LIBS spectrum characteristic number intervals are [1,289], [290,578], [2891,3179], [4914,5202]. For vanadium, the number of intervals of the optimized joint interval partial least squares method is 22, the number of combined intervals is 3, the extracted characteristic interval numbers are 6, 9 and 10, and the corresponding residual oil LIBS spectrum characteristic number intervals are [1361,1578], [2105,2367], [2368,2630].

[0040] Step 6. Under the optimized number of latent variables, preprocessing methods and characteristic variable extraction method parameters, the established partial least squares correction model predicts the content of trace metal elements nickel and vanadium in the residual oil of the test set in step 2. The model performance prediction scatter plot is attached. Figure 3 , the prediction model of the content of trace metal elements nickel and vanadium in residual oil was constructed for Ni element, R 2 p is 0.9864, and the RMSep is 1.4402 mg / kg; for element V, R 2 p It is 0.9812, and the RMSep is 0.5888 mg / kg.

[0041] The residual oil described in this embodiment is an actual residual oil sample collected from a certain oil refinery.

[0042] Comparative Example 1

[0043] The calibration set spectral data of this comparative example 1 is the LIBS spectral data of the original residual oil, and partial least squares models based on the original residual oil LIBS spectrum, based on the optimized latent variable number, and based on the pre-processed LIBS spectral data are respectively constructed to verify the prediction performance of the model constructed in Example 1 of the present invention. The specific steps of model construction of comparative example 1 are as follows.

[0044] Step 1 and step 2 are the same as step 1 and step 2 of Example 1;

[0045] Step 3: Establish a partial least squares correction model based on the original LIBS spectrum for the spectral data of the calibration set in step 2, using the coefficient of determination and root mean square error as evaluation parameters;

[0046] Step 4: Optimize the number of latent variables of the partial least squares model by using leave-one-out cross validation, and establish partial least squares correction models based on the optimized number of latent variables respectively;

[0047] Step 5. Use the optimal number of latent variables obtained in step 4 to establish preprocessing of residual oil LIBS spectra based on normalization (maximum and minimum normalization), standard normal transformation, multivariate scattering correction, first-order derivatives and second-order derivatives, as well as different combinations of the above methods, and determine the optimal preprocessing method based on leave-one-out cross-validation, and establish a partial least squares correction model for spectral data after optimizing spectral preprocessing.

[0048] Step 6: Use the partial least squares correction model established above to predict the content of trace metal elements nickel and vanadium in the residual oil of the test set in step 2.

[0049] The prediction performance of the partial least squares correction model constructed based on different input variables and parameters in Example 1 and Comparative Example 1 for the content of trace elements nickel and vanadium in the residual oil in the test set is compared, as shown in Table 1:

[0050] Table 1

[0051]

[0052] The prediction performance of the partial least squares correction model of Example 1 and Comparative Example 1 for the trace metal elements nickel and vanadium in the residual oil of the test set was compared. The results show that the partial least squares correction model for residual oil trace element prediction constructed based on Example 1 has better correlation and lower root mean square error for both nickel and vanadium. Therefore, the laser induced breakdown spectroscopy technology combined with the partial least squares correction model can realize the accurate determination of trace metal elements in residual oil, and provide a theoretical basis and technical support for the detection of residual oil and the promotion of the application of laser induced breakdown spectrometer in the field of oil product analysis.

[0053] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art may make various modifications or variations within the scope of the claims, which will not affect the essence of the present invention.

Claims

1. A method for determining trace metal elements in residual oil based on partial least squares method, characterized in that: The following steps are involved: Step 1: Collect 10 kinds of residual oil samples, calcine the residual oil samples in a muffle furnace at 400°C for 5 hours, weigh the calcined residual oil samples, grind them into powder in a mortar, and mix them in a mass ratio of 1:1 and 1:2 until they are fully uniform, to obtain 20 residual oil samples; among them, 10 samples are 1:1, and the other 10 samples are 1:2; Step 2: Using a laser induced breakdown spectroscopy instrument to collect spectral data of 20 residual oil samples; Step 3: Divide the 20 residual oil sample sets into 14 calibration set samples and 6 prediction set samples based on the KS algorithm; Step 4: construct a partial least squares model based on 14 calibration set samples, and optimize the latent variables of the partial least squares model based on leave-one-out cross-validation to determine the optimal value of the latent variable; Step 5: Based on the normalization method of the maximum LIBS spectrum intensity of the residual oil sample, mean centering, multivariate scattering correction, standard normal transformation, derivative method, and wavelet transform method, the above methods are combined in pairs to preprocess the laser induced breakdown spectrum, and the optimal preprocessing method is determined based on leave-one-out cross validation; Step 6: Use the joint interval partial least squares method to further screen the characteristic variables of the residue oil LIBS spectral data, and optimize the number of intervals and the number of interval combinations of the joint interval partial least squares method based on the leave-one-out cross-validation method.

2. A method for determining trace metal elements in residual oil based on partial least squares method according to claim 1, characterized in that: In step 2, the conditions for collecting the spectral data are: under indoor lighting conditions, the indoor temperature is 22-26° C., and the spectral range of the laser induced breakdown spectroscopy instrument is 200-500 nm.

3. The method for determining trace metal elements in residual oil based on partial least squares method according to claim 1, characterized in that: In step 2, the specific method of spectral data collection is: randomly selecting 50 LIBS spectral data test points in each residual oil sample, and then taking the average value of the collected spectral data as the modeling analysis data of the sample.

4. The method for determining trace metal elements in residual oil based on partial least squares method according to claim 1, characterized in that: In step 4, the evaluation parameters for optimizing the latent variables of the partial least squares model based on leave-one-out cross-validation are the determination coefficient and the root mean square error.

5. The method for determining trace metal elements in residual oil based on partial least squares method according to claim 1, characterized in that: In step 5, the evaluation parameters for determining the optimal preprocessing method based on leave-one-out cross-validation are the determination coefficient and the root mean square error.

6. The method for determining trace metal elements in residual oil based on partial least squares method according to claim 1, characterized in that: In step six, the evaluation parameters for optimizing the number of intervals and the number of interval combinations of the joint interval partial least squares method based on the leave-one-out cross-validation method are the determination coefficient and the root mean square error.

7. The method for determining trace metal elements in residual oil based on partial least squares method according to claim 1, characterized in that: In step six, the optimization result of the number of intervals is 20 to 40.

8. The method for determining trace metal elements in residual oil based on partial least squares method according to claim 1, characterized in that: In step six, the optimization result of the number of interval combinations is 2 to 5.

Images