Rapid Analysis Method for Wax Content in Crude Oil Based on Infrared Spectrum Fusion

Abstract

The present invention provides a rapid analysis method for wax content in crude oil based on infrared spectrum fusion. This method uses Fourier transform infrared (FTIR) spectroscopy and near-infrared (NIR) spectrometers to collect mid-infrared (MIR) and NIR spectral data of crude oil samples, which are divided into a calibration set and a test set. First, different spectral combination pretreatment methods are used for the calibration set. Then, variable importance projection is used to screen characteristic variables from the pretreated spectral data, and a fusion model is constructed based on XGBoost. The hyperparameters of the XGBoost model are optimized by grid search combined with cross-validation to obtain the best model parameters. An XGBoost calibration model based on optimal pretreatment and optimal model parameters after intermediate fusion is established to predict the wax content in the crude oil of the test set, thereby establishing a method for rapid and accurate quantitative analysis of wax content in crude oil, and improving the accuracy and reliability of the analysis results of wax content in crude oil.

Description

Technical Field

[0001] The present invention belongs to the technical field of spectral analysis, and particularly relates to a rapid analysis method for wax content in crude oil based on infrared spectral fusion. Background Art

[0002] In reality, more than half of the crude oil production belongs to the category of waxy crude oil. The high wax content in waxy crude oil makes it prone to wax crystallization and precipitation during transportation, which not only blocks pipelines and equipment, but also poses major technical problems in various aspects such as the detection, extraction, transportation, storage, and refining of crude oil. Therefore, controlling and managing the petroleum wax content has become a key technical challenge faced by the petroleum industry.

[0003] Existing wax content detection techniques for crude oil, including differential scanning calorimetry (DSC), gas chromatography (GC), and thermal desorption method, each have their own advantages and disadvantages: Differential scanning calorimetry (DSC) is based on the linear correlation between the wax precipitation heat enthalpy and the wax content, and the wax content of the sample is inferred through the thermogram. This method can provide detailed information about the thermal properties of the sample, but the equipment cost is high and the operation is complex; Gas chromatography (GC) uses a gas chromatograph to analyze the sample to determine the wax content in the sample. This method can provide detailed information about the chemical composition of the sample, but the equipment cost is relatively high and professional operation skills are required; The thermal desorption method calculates the wax content by dissolving the crude oil sample in petroleum ether and separating the wax. This method has high technical requirements for sample processing and is prone to introducing human errors.

[0004] In the invention patent with the publication (announcement) number CN107966420B, a method for predicting crude oil properties by near-infrared spectroscopy is disclosed. By identifying a group of virtual library spectra that are most similar to the spectrum of the crude oil to be measured, and then according to the similarity degree between the spectrum of the crude oil to be measured and each spectrum in this group of virtual library spectra, the properties of the crude oil to be measured are calculated by similarity weighting based on the crude oil properties corresponding to the virtual library spectra. This method requires establishing a near-infrared spectral database of crude oil samples, and detects all the conventional physical properties of crude oil samples, such as density, acid value, carbon residue, sulfur content, nitrogen content, wax content, gum content, asphaltene content, and true boiling point distillation data, and does not specifically target the detection and analysis of wax content in crude oil.

[0005] The above methods generally have problems such as complex operation, high cost, lack of pertinence, and the use of a large amount of reagents may have an adverse impact on the environment. Therefore, it is particularly important to find a more simple, rapid, and environmentally friendly technique for detecting wax content in crude oil. Summary of the Invention

[0006] The object of the present invention is to provide a rapid analysis method for wax content in crude oil based on infrared spectrum fusion. Based on the complementarity of Fourier transform infrared (FTIR) and near-infrared (NIR) spectroscopy technologies, the spectral data obtained by the two technologies are fused to obtain more comprehensive spectral information, thereby improving the accuracy and reliability of the analysis results of wax content in crude oil.

[0007] A rapid analysis method for wax content in crude oil based on infrared spectrum fusion provided by the present invention includes the following steps:

[0008] Step 1: Use Fourier transform infrared (FTIR) spectroscopy and near-infrared (NIR) spectroscopy instruments to collect MIR and NIR spectral data for a number of crude oil samples.

[0009] Step 2: Divide the number of the crude oil sample sets in Step 1 into calibration set samples and test set samples according to the ratio of (3 - 4):1.

[0010] Step 3: For the calibration set samples in Step 2, construct an XGBoost calibration model, preprocess the MIR and NIR spectral data collected in Step 1 based on combinations of different preprocessing methods, and determine the optimal preprocessing method based on leave-one-out cross-validation. The function definition of leave-one-out cross-validation is as follows: Wherein, is the number of crude oil samples, is the loss function, is the true value of the wax content of the crude oil sample, is the value predicted on the training set excluding the th crude oil sample;

[0011] Step 4: Extract and screen characteristic variables on the basis of preprocessing the crude oil MIR and NIR spectral data with the optimal preprocessing method based on leave-one-out cross-validation in Step 3.

[0012] Step 5: Optimize the XGBoost calibration model in Step 4 by means of leave-one-out cross-validation and grid search to obtain the optimal hyperparameters of the XGBoost calibration model. The function definition of grid search is as follows:

[0013] Wherein, is the hyperparameter space of the XGBoost calibration model;

[0014] is the performance index of the prediction model for wax content in crude oil calculated using leave-one-out cross-validation;

[0015] Use the XGBoost calibration model with the optimal preprocessing method and optimal hyperparameters established in Step 3 and Step 5 to predict the wax content in the test set samples of crude oil in Step 2.

[0016] In the above technical solution, further, several crude oil samples in Step 1 are prepared from 1 crude oil sample through the following steps: collect 1 crude oil sample, place the crude oil sample in a constant temperature water bath and heat it at 60 °C for 1.5 h, then mix it evenly with an ultrasonic oscillator for 1 h, and then fully mix it with a vortex mixer for 0.5 h, add paraffin, and adjust the gradient ratio to 50% to obtain several crude oil samples.

[0017] In the above technical solution, further, the number of the crude oil samples is not less than 15.

[0018] In the above technical solution, further, the preprocessing of the MIR and NIR spectral data in Step 3 based on the combination of different preprocessing methods includes: based on the maximum intensity normalization method, standard normal variate transformation method, multiplicative scatter correction method, derivative method and wavelet transform method of crude oil MIR and NIR spectra, preprocess the MIR and NIR spectral data based on the combination of any two of the above methods.

[0019] In the above technical solution, further, in Step 4, variable importance projection is used to extract and screen characteristic variables from the preprocessed crude oil MIR and NIR spectral data respectively. After screening, among them, the wax characteristic peaks screened from the crude oil MIR spectral data are 3248 nm, 3383 nm, 3429 nm, 3496 nm, 6086 nm, 6813 nm, 7252 nm, 8468 nm, 10081 nm, 10121 nm, 10656 nm, 11274 nm and 13870 nm; in the NIR spectral data, the combined bands of C-H screened are 2222 - 2500 nm and 1250 - 1450 nm, and the first and second overtone intervals of C-H screened are 1600 - 1850 nm and 1100 - 1250 nm respectively.

[0020] In the above technical solution, further, when the XGBoost correction model is optimized by the leave-one-out cross-validation and grid search methods in Step 5, two indexes, the coefficient of determination and the root mean square error, are used as the model performance evaluation parameters.

[0021] In the above technical solution, further, the coefficient of determination function and the root mean square error in Step 5 are defined as follows: coefficient of determination function: , root mean square error: where, is the actual value of the wax content in the crude oil; is the predicted value of the wax content in the crude oil; is the mean value of the actual values of the wax content in the crude oil; is the number of samples of the crude oil.

[0022] In the above technical solution, further, the spectral range of the Fourier transform infrared (FTIR) spectrometer used in Step 1 is 2500 - 20000 nm, and the spectral range of the near-infrared (NIR) spectrometer is 800 - 2500 nm.

[0023] In the above technical solution, further, the optimal hyperparameters of the XGBoost calibration model obtained by the leave-one-out cross-validation and grid search methods in Step 5 include three hyperparameters: n_estimators, subsample, and colsample_bytree.

[0024] In the above technical solution, further, the parameter range of n_estimators is 50 - 300, the parameter range of subsample is 0.1 - 0.9, and the parameter range of colsample_bytree is 0.1 - 0.9.

[0025] Compared with the prior art, the present invention has the following advantages: Based on the complementarity of Fourier transform infrared (FTIR) and near-infrared (NIR) spectroscopy technologies, the spectral data obtained by the two technologies are fused to obtain more comprehensive spectral information, thereby improving the accuracy and reliability of the analysis results of the wax content in crude oil. Through the leave-one-out cross-validation method, the spectral data of the calibration set are optimized, and the optimal model hyperparameters are obtained to establish the XGBoost calibration model. An XGBoost calibration model based on feature variable selection after optimal preprocessing is established to predict the wax content in the test set of crude oil, improving the accuracy of the XGBoost calibration model, and thus establishing a method for rapid, non-destructive, and accurate quantitative analysis of the wax content in crude oil. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is the interval spectrum of the MIR and NIR of the crude oil sample in the embodiment of the present invention with a wavelength range of 2500 - 20000 nm;

[0027] Figure 2 It is the interval spectrum of the MIR and NIR of the crude oil sample in the embodiment of the present invention with a wavelength range of 800 - 2500 nm;

[0028] Figure 3 It is the MIR and NIR feature selection spectrum of the crude oil sample in the embodiment of the present invention;

[0029] Figure 4 It is the schematic diagram of the construction idea of the crude oil quantitative analysis model in the embodiment of the present invention;

[0030] Figure 5 It is the scatter plot of the prediction performance of the crude oil quantitative analysis model in the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0031] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0032] In this embodiment, the crude oil is an actual crude oil sample collected by an oil production plant.

[0033] As Figure 4 shown, a rapid analysis method for wax content in crude oil based on infrared spectrum fusion provided by the present invention includes the following steps:

[0034] Step 1: Use Fourier transform infrared (FTIR) spectroscopy and near-infrared (NIR) spectroscopy instruments to collect MIR and NIR spectral data for several crude oil samples.

[0035] Step 2: Divide the number of several crude oil sample sets in Step 1 into calibration set samples and test set samples according to a ratio of (3-4):1.

[0036] Step 3: For the calibration set samples in Step 2, construct an XGBoost calibration model, preprocess the MIR and NIR spectral data collected in Step 1 based on a combination of different preprocessing methods, and determine the optimal preprocessing method based on leave-one-out cross-validation. The function definition of leave-one-out cross-validation is as follows:

[0037] Among them, is the number of crude oil samples, is the loss function, is the true value of the wax content of the crude oil sample, is the value predicted on the training set excluding the th crude oil sample;

[0038] Step 4: Based on the optimal preprocessing method for crude oil MIR and NIR spectral data by leave-one-out cross-validation in Step 3, perform feature variable extraction and screening.

[0039] Step 5: Optimize the XGBoost calibration model in Step 4 through the methods of leave-one-out cross-validation and grid search to obtain the optimal hyperparameters of the XGBoost calibration model. The function definition of grid search is as follows:

[0040] Among them, is the hyperparameter space of the XGBoost calibration model;

[0041] It is the performance index of the wax content prediction model in crude oil calculated using leave-one-out cross-validation;

[0042] Use the optimal preprocessing method and the XGBoost calibration model with optimal hyperparameters established in Step 3 and Step 5 to predict the wax content in the crude oil samples of the test set in Step 2.

[0043] In this embodiment, the details of each step will be introduced in detail according to the step sequence of the above method.

[0044] The several crude oil samples in Step 1 are prepared from 1 type of crude oil sample through the following steps: Collect 1 type of crude oil sample, place the crude oil sample in a constant temperature water bath and heat it at 60 °C for 1.5 h, then mix it evenly with an ultrasonic oscillator for 1 h, then use a vortex mixer to fully mix it for 0.5 h, add paraffin, and gradiently mix it to 50% to obtain several crude oil samples.

[0045] As Figure 1-2 shown, use a Fourier transform infrared (FTIR) spectrometer and a near-infrared (NIR) spectrometer to analyze the calibration set samples respectively to obtain the corresponding spectral data.

[0046] The transform infrared spectrometer used in this example is an FTIR spectrometer (VERTEX 70, Bruker, Germany) equipped with a DTGS detector, and the spectral acquisition range is 2500 - 20000 nm. During the measurement process, the sample is directly placed on a germanium crystal ATR, and the background spectrum is collected with air as the blank before measuring each sample. Each sample is measured 16 times, and the average spectrum of the 16 measurements is used as the analysis spectrum.

[0047] The near-infrared spectrometer used in this example is Shimadzu UV-3600OPlus, and the spectral acquisition range is 800 - 2500 nm. During the measurement process, the sample is evenly smeared on a sample stage with barium sulfate pressed as the bottom, the background spectrum is collected with barium sulfate as the blank before measuring each sample, and then the spectral acquisition is carried out.

[0048] The spectral data acquisition by the MIR and NIR spectral instruments is carried out under indoor lighting conditions, and the indoor temperature is 22 - 26 °C.

[0049] In Step 2, taking the determination of wax content in crude oil as an example, this example uses 17 crude oil samples, and the reference values of their wax content are shown in Table 1. When building the model, randomly select 13 samples from the 17 samples as the calibration set, and the remaining 4 samples as the test set; in this embodiment, the sample numbers of the selected prediction set are #1, #2, #6, and #16 respectively.

[0050] Table 1 Reference values of crude oil wax content

[0051]

[0052] In Step 3, for the calibration set samples in Step 2, an XGBoost calibration model is constructed. Here, XGBoost is an existing machine learning algorithm that integrates multiple decision trees through the gradient boosting method to construct a prediction model. It has strong prediction ability and efficient computational performance and is widely used in regression and classification tasks. In the present invention, the NIR and MIR spectral intensities of crude oil samples are corresponded to the corresponding wax contents for constructing the XGBoost calibration model. The optimal hyperparameters of the XGBoost calibration model include three hyperparameters: n_estimators, subsample, and colsample_bytree. During the model optimization process, the coefficient of determination and the root mean square error are used as evaluation indicators for model performance. For the calibration set data in Step 2, based on different preprocessing methods, including the maximum intensity normalization method for crude oil MIR and NIR spectra, the standard normal variate method, the multiplicative scatter correction method, the derivative method, and the wavelet transform method. The above single preprocessing methods are all existing technologies. For example, the maximum intensity normalization method is a method for eliminating the differences in the dimensions of spectral data. The standard normal variate method is a method for eliminating the spectral scattering effect caused by the differences in the sizes of sample microscopic particles. The multiplicative scatter correction method is a method for eliminating the spectral scattering effect caused by the differences in the uniformity of sample microscopic particles. The derivative method includes the first-order derivative method and the second-order derivative method, which are used to correct the baseline drift phenomenon of spectral data. The wavelet transform method is a method for processing spectral data signals simultaneously from two dimensions of time domain - frequency domain. The spectral data collected from crude oil samples are preprocessed by combining any two of these methods, and the optimal preprocessing method is determined based on leave-one-out cross-validation. The results show that a relatively good combination of preprocessing methods for the spectral data collected from crude oil samples is the combination of the wavelet transform method and the standard normal variate method.

[0053] The preprocessed spectral data is optimized by the methods of grid search and leave-one-out cross-validation. Two indicators, the coefficient of determination and the root mean square error, are used as evaluation parameters to obtain the optimal hyperparameters of the XGBoost calibration model. The prediction accuracy of the model is improved by adjusting n_estimators, subsample, and colsample_bytree. The range of n_estimators is from 50 to 300. Selecting this range can balance performance and complexity. The range of subsample is from 0.1 to 0.9, which helps to prevent overfitting and ensure that the model obtains enough data for training. The range of colsample_bytree is from 0.1 to 0.9. Lower values (such as 0.1 and 0.3) help to reduce the correlation between features, thus preventing overfitting, while larger values retain more information and avoid underfitting.

[0054] In Step 4, as Figure 3 shown, feature variable extraction is performed on the obtained optimal pretreatment method, and the input variables are further optimized. The variable importance projection is used to screen the feature variables of the pretreated spectral data. Among them, the wax characteristic peaks screened from the crude oil MIR spectral data are 3248 nm, 3383 nm, 3429 nm, 3496 nm, 6086 nm, 6813 nm, 7252 nm, 8468 nm, 10081 nm, 10121 nm, 10656 nm, 11274 nm, and 13870 nm; in the NIR spectral data, the combined bands of C-H screened are 2222 - 2500 nm and 1250 - 1450 nm, and the first and second overtone intervals of C-H screened are 1600 - 1850 nm and 1100 - 1250 nm respectively.

[0055] In Step 5, under the optimized pretreatment method, model hyperparameters, and variable importance projection method, the XGBoost calibration model is established to predict the wax content in the crude oil of the test set samples in Step 2. When the XGBoost calibration model in Step 5 is optimized by the leave-one-out cross-validation and grid search methods, the coefficient of determination and root mean square error are used as two indicators for model performance evaluation parameters.

[0056] The coefficient of determination function and root mean square error in Step 5 are defined as follows:

[0057] Coefficient of determination function: , Root mean square error: where, is the actual value of the wax content in the crude oil; is the predicted value of the wax content in the crude oil; is the mean value of the actual values of the wax content in the crude oil; is the number of samples of the crude oil.

[0058] The scatter plot of model performance prediction is shown in Figure 5 , For the wax content, the prediction model of the wax content in the crude oil constructed, through optimal hyperparameter adjustment and leave-one-out cross-validation, the R 2 CV value of the XGBoost model is increased to 0.9844, and the RMSE CV is decreased to 1.6793%, and the R 2 P and RMSEP are 0.9957 and 3.3249% respectively. The above data shows that the quantitative analysis model constructed based on the infrared spectral fusion data shows very good prediction performance for the wax content in the crude oil, and from the coefficient of determination function value R 2From the cv and root mean square error value RMSEcv, it can be seen that the model has very good stability. In summary, this method has the advantages of simple operation, little manual intervention, rapid analysis, low cost, accurate prediction, and non-destructive analysis of samples, and is suitable for the on-site rapid detection of wax content in the complex matrix of crude oil.

Claims

1. A rapid analysis method for wax content in crude oil based on infrared spectrum fusion, characterized in that, It includes the following steps: Step 1: Use Fourier transform infrared (FTIR) spectroscopy and near-infrared (NIR) spectroscopy instruments to collect MIR and NIR spectral data for several crude oil samples; Step 2: Divide the number of several crude oil sample sets in Step 1 into calibration set samples and test set samples according to the ratio of (3 - 4):1; Step 3: For the calibration set samples in Step 2, construct an XGBoost calibration model, preprocess the MIR and NIR spectral data collected in Step 1 based on combinations of different preprocessing methods, and determine the optimal preprocessing method based on leave-one-out cross-validation. The function definition of leave-one-out cross-validation is as follows: Among them, is the number of crude oil samples, is the loss function, is the true value of the wax content of the crude oil sample, is the value predicted on the training set excluding the th crude oil sample; Step 4: On the basis of the optimal preprocessing method for crude oil MIR and NIR spectral data determined by leave-one-out cross-validation in Step 3, perform feature variable extraction and screening; Step 5: Optimize the XGBoost calibration model in Step 4 through leave-one-out cross-validation and grid search methods to obtain the optimal hyperparameters of the XGBoost calibration model. The function definition of grid search is as follows: Among them, is the hyperparameter space of the XGBoost calibration model; are performance indicators of the wax content prediction model in crude oil calculated using leave-one-out cross-validation; Use the XGBoost calibration model with the optimal preprocessing method and optimal hyperparameters established in Step 3 and Step 5 to predict the wax content in the test set sample crude oil in Step 2; The preprocessing of MIR and NIR spectral data based on combinations of different preprocessing methods in Step 3 includes: based on the maximum intensity normalization method, standard normal variate transformation method, multiplicative scatter correction method, derivative method, and wavelet transform method for crude oil MIR and NIR spectra, preprocess the MIR and NIR spectral data based on the combination of any two of the above methods; in Step 4, use variable importance projection to extract and screen feature variables for the preprocessed crude oil MIR and NIR spectral data respectively. After screening, among them, the wax characteristic peaks screened from the crude oil MIR spectral data are 3248nm, 3383nm, 3429nm, 3496nm, 6086nm, 6813nm, 7252nm, 8468nm, 10081nm, 10121nm, 10656nm, 11274nm, and 13870nm; in the NIR spectral data, the combined bands of C-H screened are 2222 - 2500nm and 1250 - 1450nm, and the first and second overtones of C-H screened are in the ranges of 1600 - 1850nm and 1100 - 1250nm respectively.

2. The rapid analysis method for wax content in crude oil based on infrared spectrum fusion according to claim 1 is characterized in that, The several crude oil samples in Step 1 are prepared from 1 type of crude oil sample through the following steps: collect 1 type of crude oil sample, place the crude oil sample in a constant temperature water bath and heat it at 60°C for 1.5h, then mix it evenly with an ultrasonic oscillator for 1h, and then fully mix it with a vortex mixer for 0.5h, add paraffin, and gradient ratio it to 50% to obtain several crude oil samples.

3. A rapid analysis method for wax content in crude oil based on infrared spectrum fusion according to claim 2, characterized in that, The number of the crude oil samples is not less than 15.

4. A rapid analysis method for wax content in crude oil based on infrared spectrum fusion according to claim 1, characterized in that When optimizing the XGBoost calibration model in Step 5 through leave-one-out cross-validation and grid search methods, two indicators, the coefficient of determination and the root mean square error, are used as model performance evaluation parameters.

5. A rapid analysis method for wax content in crude oil based on infrared spectrum fusion according to claim 4, characterized in that, The definitions of the coefficient of determination function and the root mean square error in Step 5 are as follows: Coefficient of determination function: , Root mean square error: where is the actual value of the wax content in crude oil; is the predicted value of the wax content in crude oil; is the mean value of the actual values of the wax content in crude oil; is the number of samples of crude oil.

6. A rapid analysis method for wax content in crude oil based on infrared spectrum fusion according to claim 1, characterized in that, The spectral range of the Fourier transform infrared (FTIR) spectrometer used in the first step is 2500 - 20000 nm, and the spectral range of the near-infrared (NIR) spectrometer is 800 - 2500 nm.

7. A rapid analysis method for wax content in crude oil based on infrared spectrum fusion according to claim 1, characterized in that, The optimal hyperparameters of the XGBoost calibration model obtained by the leave-one-out cross-validation and grid search methods in the fifth step include three hyperparameters: n_estimators, subsample, and colsample_bytree.

8. A rapid analysis method for wax content in crude oil based on infrared spectrum fusion according to claim 7, characterized in that, The parameter range of n_estimators is 50 - 300, the parameter range of subsample is 0.1 - 0.9, and the parameter range of colsample_bytree is 0.1 - 0.9.

Images