WO2024011687A1 - Method and apparatus for establishing oil product physical property fast evaluation model - Google Patents

Abstract

A method and apparatus for establishing an oil product physical property fast evaluation model. The method comprises: acquiring sample data of an oil product to be modeled and dividing the sample data into a correction set and a verification set (step 101), the sample data of the oil product to be modeled comprising spectrogram data and viscosity-temperature curve data of each sample, and assay data corresponding to each sample; respectively preprocessing correction set data by means of a plurality of data processing methods, so as to acquire a plurality of corresponding preprocessed data sets (step 102); respectively merging the spectrogram data and viscosity-temperature curve data of the preprocessed data sets to form a plurality of corresponding associated data sets, using a plurality of association algorithms to respectively associate the associated data sets with an assay data set corresponding to each sample, and constructing a plurality of corresponding physical property analysis models (step 103); and using verification set data to respectively verify and evaluate each physical property analysis model, and selecting an optimal model as a physical property fast evaluation model of the oil product to be modeled (step 104). The model constructed according to the method has high accuracy and can accurately predict the material physical property indexes of the oil product.

Description

A method and device for establishing a quick oil property evaluation model Technical field

The invention relates to the field of petrochemical industry, and in particular to a method and device for establishing a quick evaluation model of oil physical properties.

Background technique

At present, most refineries at home and abroad mainly use traditional laboratory analysis methods to analyze various raw materials and intermediate products in the production process. In traditional laboratory analysis methods, each indicator requires relevant personnel to use separate instruments or methods for analysis. Therefore, labor costs and equipment maintenance costs are high, and it is difficult to meet safety and environmental protection requirements. There is a lack of effective rapid evaluation analysis methods, resulting in Laboratory analysts have high work intensity and low efficiency.

Some petrochemical companies have introduced some quick analysis technologies, such as near-infrared spectroscopy (NIR), mid-infrared spectroscopy (MIR) and nuclear magnetic resonance spectroscopy (NMR). The detection principles and application fields of these rapid evaluation analysis technologies are different, but the technical routes of the spectrum analysis process are basically the same. Spectrum analysis mainly includes two processes: (1) establishment of analysis model; (2) analysis of spectrum. The quality of each indicator analysis model directly determines the accuracy of the quick evaluation analysis data. However, in the actual application process, the current analysis technology has some general disadvantages: several spectra can currently only realize the analysis of the functional group level inside the sample, the amount of information is small, the correlation with some complex indicators is not strong, and the accuracy of the model It is not high, and it is difficult for the model to predict the changing trend when the material physical property indicators fluctuate.

The current quick evaluation analysis technology has at least the following shortcomings: the accuracy of the model is not high, and it is difficult to accurately predict the material physical properties of the sample.

Contents of the invention

The invention provides a method and device for establishing a quick oil physical property evaluation model to solve the technical problem of low model accuracy and difficulty in accurately predicting material physical property indicators of samples.

In order to solve the above technical problems, embodiments of the present invention provide a method for establishing a quick evaluation model of oil physical properties, which includes:

Obtain the sample data of the oil product to be modeled and divide it into a calibration set and a verification set; wherein the sample data of the oil product to be modeled includes spectrum data, viscosity-temperature curve data of each sample, and laboratory data corresponding to each sample. ; The spectral data is the spectral data or wave spectrum data of the oil to be modeled;

Through a variety of data processing methods, the calibration set data are preprocessed respectively to obtain multiple corresponding preprocessed data sets;

The spectrum data and viscosity-temperature curve data of each of the preprocessed data sets are merged to form multiple corresponding associated data sets, and are associated with the assay data sets corresponding to each sample through multiple association algorithms to construct multiple corresponding physical property analysis model;

Use the verification set data to verify and evaluate each physical property analysis model, and select the optimal model as the physical property quick evaluation model of the oil to be modeled.

On the basis of the existing technology, the present invention creatively proposes a quick evaluation model establishment method that combines spectral data and viscosity-temperature curve data, further improving the accuracy of the quick evaluation model. The oil spectrum or wave spectrum mainly reflects the group characteristics of each organic molecule in the sample, and the viscosity information at each temperature point is helpful for analyzing the physical and chemical properties of heavy oil. Therefore, by combining the spectrum data with the viscosity-temperature curve data, the reflected oil information will be more detailed, rich and accurate. The physical property analysis model built on this basis has high accuracy and can accurately predict the material physical properties of the oil. index.

Further, the sample data of the oil product to be modeled is obtained and divided into a correction set and a verification set, specifically: the Kolmogorov-Smirnov algorithm is used to divide all samples into two categories: a correction set and a verification set.

Further, the correction set data are preprocessed respectively through a variety of data processing methods to obtain multiple corresponding preprocessed data sets, specifically:

For the spectrum data of each sample in the calibration set, multiple first data processing methods are selected for preprocessing, and preprocessed spectrum data corresponding to each sample is obtained, wherein the first data processing method includes: None Processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing processing, first derivative At least two or more combinations of method processing, detrending method processing, and baseline correction method processing;

For the viscosity-temperature curve data of each sample in the calibration set, a preset algorithm is used to fit the empirical formula of the viscosity-temperature curve, and the corresponding viscosity value is calculated based on the preset multiple temperature points, and then processed according to a variety of second data This method preprocesses multiple temperature points and the viscosity values corresponding to each temperature point, and obtains the preprocessed viscosity-temperature curve data corresponding to each sample; where each viscosity-temperature curve data corresponds to multiple temperature points and The viscosity value corresponding to each temperature point; the second data processing method includes: at least one or more combinations of logarithmic processing, averaging processing, mean variance processing, maximum value and minimum value normalization processing.

The quality of each physical property analysis model directly determines the accuracy of the prediction of material physical property indicators. Preprocessing the spectrum data before modeling can reduce the unavoidable negative effects such as data noise and baseline drift during the spectrum scanning process, and improve the quality of the model; any temperature point can be calculated through the empirical formula of the viscosity-temperature curve The viscosity value makes up for the defect of incomplete structural information of the viscosity data response at a single temperature point; the present invention uses a variety of data processing methods to preprocess the correction set data to form multiple corresponding preprocessed data sets to construct Multiple corresponding physical property analysis models are selected optimally to ensure the quality of the model, so that the material physical property indicators of the sample can be accurately predicted.

Further, after obtaining the preprocessed spectrum data corresponding to each sample, perform dimensionality reduction processing on each of the preprocessed spectrum data according to a preset dimensionality reduction method; wherein, the Dimensionality reduction methods include simple model method or principal component analysis method.

Spectral data are often high-dimensional data, accompanied by a lot of unnecessary information. Therefore, before formal modeling, the spectrum data must be dimensionally reduced. The simple model method and principal component analysis method can effectively reduce the number of spectrum data points. , on the one hand, it improves the modeling efficiency, on the other hand, it can also reduce the interference of useless information to improve the accuracy of the model, thereby improving the accuracy of predicting the material physical properties of the sample.

Further, the spectrum data and viscosity-temperature curve data of each preprocessed data set are merged to form multiple corresponding associated data sets, and are associated with the assay data sets corresponding to each sample through a variety of association algorithms to construct Multiple corresponding physical property analysis models, and the multiple correlation algorithms include: at least one or more combinations of partial least squares, artificial neural network, and kernel partial least squares.

The present invention uses multiple correlation algorithms to construct multiple corresponding physical property analysis models and selects them optimally, eliminating the possibility of inaccurate prediction results caused by the low quality of a single physical property analysis model, ensuring the accuracy of the model, and improving the material physical properties of the sample. Accuracy of indicator forecasts.

Further, each physical property analysis model was separately verified and evaluated using the verification set data, and the optimal model was selected as the physical property quick evaluation model of the oil to be modeled, specifically as follows:

The spectrum data and viscosity-temperature curve data of each sample in the verification set are combined to form a data set to be predicted, and each physical property analysis model is used for prediction to obtain multiple corresponding prediction result sets;

Each prediction result set is compared with the laboratory data set corresponding to each sample in the verification set, and the model with the smallest error in the verification set is selected as the physical property quick evaluation model of the oil to be modeled.

The smaller the validation set error, the higher the accuracy of the model. Therefore, using the model with the smallest validation set error among all models can ensure the accuracy of predicting the material physical properties of the sample.

A device for establishing a model for quick evaluation of oil physical properties, including: an acquisition module, a preprocessing module, a modeling module, and a screening module.

The acquisition module is used to obtain sample data of the oil to be modeled and divide it into a calibration set and a verification set; wherein the sample data of the oil to be modeled includes spectrum data, viscosity-temperature curve data of each sample, and The laboratory data corresponding to each sample; the spectral data is the spectral data or wave spectrum data of the oil to be modeled;

The preprocessing module is used to preprocess the correction set data respectively through multiple data processing methods to obtain multiple corresponding preprocessed data sets;

The modeling module is used to merge the spectrum data and viscosity-temperature curve data of each preprocessed data set to form multiple corresponding associated data sets, and use multiple association algorithms to combine them with the laboratory data corresponding to each sample. Set association to build multiple corresponding physical property analysis models;

The screening module is used to verify and evaluate each physical property analysis model using the verification set data, and select the optimal model as the physical property quick evaluation model of the oil to be modeled.

Wherein, the preprocessing module includes: a spectrum data processing unit and a viscosity-temperature curve data processing unit;

The spectrum data processing unit is used to preprocess the spectrum data, specifically:

For the spectrum data of each sample in the calibration set, multiple first data processing methods are selected for preprocessing, and preprocessed spectrum data corresponding to each sample is obtained, wherein the first data processing method includes: None Processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing processing, first derivative At least two or more combinations of method processing, detrending method processing, and baseline correction method processing;

After obtaining the preprocessed spectrum data corresponding to each sample, perform dimensionality reduction processing on each of the preprocessed spectrum data according to a preset dimensionality reduction method; wherein, the dimensionality reduction method Including simple model method or principal component analysis method;

The viscosity-temperature curve data processing unit is used to preprocess the viscosity-temperature curve data, specifically:

For the viscosity-temperature curve data of each sample in the calibration set, a preset algorithm is used to fit the empirical formula of the viscosity-temperature curve, and the corresponding viscosity value is calculated based on the preset multiple temperature points, and then processed according to a variety of second data This method preprocesses multiple temperature points and the viscosity values corresponding to each temperature point, and obtains the preprocessed viscosity-temperature curve data corresponding to each sample; where each viscosity-temperature curve data corresponds to multiple temperature points and The viscosity value corresponding to each temperature point; the second data processing method includes: at least one or more combinations of logarithmic processing, averaging processing, mean variance processing, maximum value and minimum value normalization processing.

The modeling module presets multiple correlation algorithms for modeling, wherein the multiple correlation algorithms include: at least one or more combinations of partial least squares, artificial neural network, and kernel partial least squares.

The screening module is used to verify and evaluate each physical property analysis model using the verification set data, and select the optimal model as the physical property quick evaluation model of the oil to be modeled, specifically as follows:

The spectrum data and viscosity-temperature curve data of each sample in the verification set are combined to form a data set to be predicted, and each physical property analysis model is used for prediction to obtain multiple corresponding prediction result sets;

Each prediction result set is compared with the laboratory data set corresponding to each sample in the verification set, and the model with the smallest error in the verification set is selected as the physical property quick evaluation model of the oil to be modeled.

The device provided by the invention creatively uses a quick evaluation analysis mode that combines spectral data and viscosity-temperature curve data, further improving the accuracy of the quick evaluation model. The oil spectrum or wave spectrum mainly reflects the group characteristics of each organic molecule in the sample. The viscosity information at each temperature point is helpful for analyzing the physical and chemical properties of heavy oil. Therefore, the spectrum data and the viscosity-temperature curve data are combined to reflect the oil The product information will be more detailed, rich and accurate. The physical property analysis model built on this basis has high accuracy and can accurately predict the material and physical property indicators of oil products.

Description of drawings

Figure 1: A schematic flow chart of a method for establishing a quick evaluation model of oil physical properties provided by the present invention;

Figure 2: A graph showing the error change trend of the model verification set constructed by each modeling method under different solutions in Embodiment 1 of the present invention;

Figure 3: A graph showing the error change trend of the model verification set constructed by each modeling method under different solutions in Embodiment 2 of the present invention;

Figure 4: A schematic structural diagram of a device for establishing a model for quick evaluation of oil physical properties provided in Embodiment 3 of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present invention.

Embodiment 1

Please refer to Figure 1 which is a schematic flow chart of a method for establishing a quick evaluation model of oil physical properties provided by the present invention. In this embodiment, this method is used to evaluate and analyze the asphalt components of a certain refinery. The analysis index takes penetration as an example. , including step 101 to step 104, the specific methods of each step are as follows:

In this embodiment, a mid-infrared spectrometer is used to collect spectral data of the oil sample to be modeled, a viscometer is used to collect viscosity-temperature curve data of the oil sample to be modeled, and the viscosity-temperature curve is fitted according to a preset algorithm The empirical formula is used to calculate the viscosity data at six temperature points of 50°C, 60°C, 70°C, 80°C, 90°C, and 100°C, and the corresponding test data is the penetration test data.

Step 101: Obtain the sample data of the oil product to be modeled and divide it into a calibration set and a validation set.

Wherein, the sample data of the oil to be modeled includes spectrum data of each sample, viscosity-temperature curve data and laboratory data corresponding to each sample; the spectrum data is spectral data or spectrum data of the oil to be modeled. ;

First, obtain 100 residual oil samples used for asphalt blending, with the penetration value ranging from 40 to 100. Obtain the mid-infrared spectrum data, viscosity-temperature curve data of each residual oil sample, and the corresponding penetration test of each sample. data _, and form _{a spectrum data} set *1. The mid-infrared spectrum band range is 600~4000cm ^-1 , the number of cutting points is 4417, and the number of temperature points taken for the viscosity value is 6, namely 50℃, 60℃, 70℃, 80℃, 90℃, and 100℃ At the temperature point, the number of physical property indicators that need to be modeled for oil products is 1, which is the penetration;

Further, the sample data of the oil product to be modeled is obtained and divided into a correction set and a verification set, specifically: the Kolmogorov-Smirnov algorithm is used to divide all samples into two categories: a correction set and a verification set.

The data required for modeling are divided into two parts: calibration set and validation set according to the Kolmogorov-Smirnov algorithm. The calibration set data is used to build the model, and the validation set data is used to verify and evaluate the model. In this embodiment, the sample ratio of the calibration set is set to 80%, that is, the calibration set contains 80 samples, the verification set contains 20 samples, the calibration set data are recorded as X _1c , X _2c , Y _{c, respectively,} and the verification set data They _are recorded as X _{1v ,} X _2v , and Y _v respectively. Among them, _c is the laboratory data corresponding to each sample in the calibration set, its dimension is 80*1, X _1v is the spectrum data of the validation set, its dimension is 20*4417, X _2v is the viscosity-temperature curve data of the validation set, its dimension is 20*6, Y _v is the laboratory data corresponding to each sample in the validation set, and its dimension is 20*1.

Step 102: Preprocess the correction set data respectively through multiple data processing methods to obtain multiple corresponding preprocessed data sets.

For the spectrum data of each sample in the calibration set, multiple first data processing methods are selected for preprocessing, and preprocessed spectrum data corresponding to each sample is obtained, wherein the first data processing method includes: None Processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing processing, first derivative At least two or more combinations of method processing, detrending method processing, and baseline correction method processing;

In this embodiment, the _classified correction set data The processing methods include: no processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum value normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing Processing, first-order derivative method processing, detrending method processing, baseline correction method processing;

Further, after obtaining the preprocessed spectrum data corresponding to each sample, perform dimensionality reduction processing on each of the preprocessed spectrum data according to a preset dimensionality reduction method; wherein, the Dimensionality reduction methods include simple model method or principal component analysis method;

In this embodiment, 40 principal components of the spectrum data are extracted through principal component analysis, and the spectrum dimension is reduced from 4417 to 40;

For the viscosity-temperature curve data of each sample in the calibration set, a preset algorithm is used to fit the empirical formula of the viscosity-temperature curve, and the corresponding viscosity value is calculated based on the preset multiple temperature points, and then processed according to a variety of second data This method preprocesses multiple temperature points and the viscosity values corresponding to each temperature point, and obtains the preprocessed viscosity-temperature curve data corresponding to each sample; where each viscosity-temperature curve data corresponds to multiple temperature points and The viscosity value corresponding to each temperature point; the second data processing method includes: at least one or more combinations of logarithmic processing, averaging processing, mean variance processing, maximum value and minimum value normalization processing;

In this embodiment, the empirical formula of the viscosity-temperature curve fitted using the preset algorithm for the calibration set data X _2c is: log(μ)=log(μ ₀ )+L/(T/T ₀ -1), and According to the fitted viscosity-temperature curve empirical formula, the viscosity values corresponding to the preset six temperature points of 50°C, 60°C, 70°C, 80°C, 90°C, and 100°C are calculated, and a second data processing is used The method is to perform logarithmic preprocessing to obtain the preprocessed viscosity-temperature curve data corresponding to each sample.

Among them, μ represents the viscosity of the oil component, μ ₀ represents the ultimate viscosity of the oil component when the temperature is infinitely high, T represents the temperature, and T ₀ represents the temperature when the oil component solidifies into a solid (viscosity is infinite), L is a related parameter used to measure the degree of fit between the viscosity-temperature curve and the measured viscosity-temperature data.

Table 1 shows the processing methods of spectral data and viscosity-temperature curve data under different schemes of this embodiment.

Table 1 Processing methods of spectral data and viscosity-temperature curve data under different schemes

Plan serial number Spectral data processing methods Viscosity-temperature curve data processing method plan 1 No processing Logarithmic processing Scenario 2 mean centering Logarithmic processing Option 3 Mean variance processing Logarithmic processing Option 4 Vector normalization Logarithmic processing Option 5 Maximum and minimum value normalization processing Logarithmic processing Option 6 Standard normal variable transformation processing Logarithmic processing Option 7 Multivariate scattering correction processing Logarithmic processing Option 8 Savitzky-Golay convolution smoothing Logarithmic processing Option 9 First order derivative method processing Logarithmic processing Option 10 detrending Logarithmic processing Plan 11 Baseline correction method processing Logarithmic processing

Step 103: Merge the spectrum data and viscosity-temperature curve data of each preprocessed data set to form multiple corresponding associated data sets, and associate them with the assay data sets corresponding to each sample through multiple association algorithms to construct Multiple corresponding physical property analysis models.

The multiple correlation algorithms include: at least one or more combinations of partial least squares, artificial neural network, and kernel partial least squares.

In this embodiment, 11 groups of processed mid _- infrared spectrum data sets X _1c and viscosity-temperature _curve data sets X _2c are respectively merged to form 11 groups of corresponding associated data sets Three preset algorithms, Partial Least Squares (PLS), Artificial Neural Network (ANN), and Kernel Partial Least Squares (KPLS), are respectively associated with the laboratory data set Y _c corresponding to each sample, and a total of 33 penetrations are constructed. Analytical models;

At the same time, the ₁₁ processed mid-infrared spectrum data sets The laboratory data set Y _c was correlated, and 33 mid-infrared spectrum-penetration analysis models were established as basic models for reference.

Step 104: Use the verification set data to verify and evaluate each physical property analysis model, and select the optimal model as the physical property quick evaluation model of the oil to be modeled.

The spectrum data and viscosity-temperature curve data of each sample in the verification set are combined to form a data set to be predicted, and each physical property analysis model is used for prediction to obtain multiple corresponding prediction result sets;

Each prediction result set is compared with the laboratory data set corresponding to each sample in the verification set, and the model with the smallest error in the verification set is selected as the physical property quick evaluation model of the oil to be modeled.

In this embodiment, the verification set data _{X 1v and} is Y _p ; compare the prediction result set Y _p with the laboratory data set Y _v corresponding to each sample in the validation set, and calculate the average error of the 33 penetration analysis model validation set samples respectively, recorded as E _p1 ~ E _p33 ;

At the same time, 33 mid-infrared spectrum-penetration analysis models were used as basic models for prediction, and the prediction result set was recorded as Y _1p ; the prediction result set Y _1p was compared with the laboratory data set Y _v corresponding to each sample in the validation set. Calculate the average error of the 33 mid-infrared spectrum-penetration analysis model validation set samples, recorded as E _p34 ~ E _p66 , as basic data for reference;

Please refer to Figure 2, which is a graph showing the error change trend of the model verification set constructed by each modeling method under different schemes in this embodiment;

As can be seen from Figure 2, compared with the model established using mid-infrared spectra alone, when combined with viscosity-temperature curve data, the overall accuracy of the models built by each modeling method under different preprocessing schemes has been greatly improved. The model established by using mid-infrared spectra alone has the best effect when using maximum and minimum normalization processing, that is, Scheme 5, and using the KPLS algorithm for modeling. The average error of the validation set samples is 3.02, while when combined with viscosity and temperature After curve data, the average error of the prediction results of the model built under this scheme dropped to 0.95, and the accuracy was significantly improved. Therefore, the model built by the KPLS algorithm under the conditions of Scheme 5 can be selected as the optimal model as the injection of the oil to be modeled. Quick evaluation model.

The above embodiments greatly improve the accuracy of the quick evaluation analysis results of physical property indicators of various raw materials or intermediate products in the petrochemical industry by using the combined quick evaluation analysis method of spectrum data and viscosity-temperature curve data. The spectral data or wave spectrum data are combined with the viscosity-temperature curve data and jointly associated with various physical property indicators. The prediction effects of the models built using various modeling methods are far better than those built using spectral data alone. The prediction results The error has been greatly reduced and the accuracy has been significantly improved. Therefore, based on the existing technology, introducing viscosity-temperature curve data into the field of petrochemical material quick evaluation analysis can significantly improve the accuracy of the analysis model of various physical property indicators of oil products, and greatly make up for the current modeling. The method has the disadvantage of low accuracy in predicting complex physical properties, so it has extremely high application value.

On the basis of the existing technology, the present invention creatively proposes a quick evaluation model establishment method that combines spectral data and viscosity-temperature curve data, further improving the accuracy of the quick evaluation model. The oil spectrum or wave spectrum mainly reflects the group characteristics of each organic molecule in the sample. The viscosity information at each temperature point is helpful for analyzing the physical and chemical properties of heavy oil. Therefore, the spectrum data and the viscosity-temperature curve data are combined to reflect the oil The product information will be more detailed, rich and accurate. The physical property analysis model built on this basis has high accuracy and can accurately predict the material and physical property indicators of oil products.

Embodiment 2

This embodiment uses a method for establishing a quick oil physical property evaluation model provided by the present invention to evaluate and analyze the third-line material of a refinery (straight-run diesel from a normal and vacuum unit). The analysis index takes the 95% distillation temperature of diesel as an example. It includes steps 201 to 204. The specific methods of each step are as follows:

In this embodiment, a near-infrared spectrometer is used to collect spectral data of the oil sample to be modeled, a viscometer is used to collect viscosity-temperature curve data of the oil sample to be modeled, and the viscosity-temperature curve is fitted according to a preset algorithm The empirical formula calculates the viscosity data at five temperature points of 20°C, 30°C, 40°C, 50°C, and 60°C, and the corresponding test data is the 95% distillation temperature test data.

Step 201: Obtain sample data of the oil product to be modeled and divide it into a calibration set and a validation set.

Wherein, the sample data of the oil to be modeled includes spectral data, viscosity-temperature curve data and laboratory data corresponding to each sample of each sample; the spectral data is the spectral data or spectral data of the oil to be modeled. ;

First obtain 100 normal third-line diesel samples, the 95% distillation temperature range is about 340-380°C, and obtain the near-infrared spectrum data, viscosity-temperature curve data and corresponding 95% distillation of each normal third-line diesel sample. temperature assay data, and form _a spectrum data _{set X1 , a} viscosity-temperature curve data set is 100*1, the near-infrared spectrum band range is 4000~12000cm ^-1 , the number of cutting points is 4094, and the number of temperature points taken for the viscosity value is 5, namely 20℃, 30℃, 40℃, 50℃, and 60℃ point, the number of physical property indicators that need to be modeled for oil products is 1, which is the 95% distillation temperature.

Further, the sample data of the oil product to be modeled is obtained and divided into a correction set and a verification set, specifically: the Kolmogorov-Smirnov algorithm is used to divide all samples into two categories: a correction set and a verification set.

The data required for modeling are divided into two parts: calibration set and validation set according to the Kolmogorov-Smirnov algorithm. The calibration set data is used to build the model, and the validation set data is used to verify and evaluate the model. In this embodiment, the sample ratio of the calibration set is set to 80%, that is, the calibration set contains 80 samples, the verification set contains 20 samples, the calibration set data are recorded as X _1c , X _2c , Y _{c, respectively,} and the verification set data They _are recorded as X _{1v ,} X _2v , and Y _v respectively. Among them, _c is the laboratory data corresponding to _{each sample in the calibration set, and its dimension is 80*1. X 1v is} the spectrum data of the validation set, and its dimension is 20*4094. 20*5, Y _v is the laboratory data corresponding to each sample in the validation set, and its dimension is 20*1.

Step 202: Preprocess the correction set data respectively through multiple data processing methods to obtain multiple corresponding preprocessed data sets.

For the spectrum data of each sample in the calibration set, multiple first data processing methods are selected for preprocessing, and preprocessed spectrum data corresponding to each sample is obtained, wherein the first data processing method includes: None Processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing processing, first derivative At least two or more combinations of method processing, detrending method processing, and baseline correction method processing;

In this embodiment, the _classified correction set data The processing methods include: no processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum value normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing Processing, first-order derivative method processing, detrending method processing, baseline correction method processing;

Further, after obtaining the preprocessed spectrum data corresponding to each sample, perform dimensionality reduction processing on each of the preprocessed spectrum data according to a preset dimensionality reduction method; wherein, the Dimensionality reduction methods include simple model method or principal component analysis method.

In this embodiment, 40 principal components of the spectrum data are extracted through principal component analysis, and the spectrum dimension is reduced from 4094 to 40.

For the viscosity-temperature curve data of each sample in the calibration set, a preset algorithm is used to fit the empirical formula of the viscosity-temperature curve, and the corresponding viscosity value is calculated based on the preset multiple temperature points, and then processed according to a variety of second data This method preprocesses multiple temperature points and the viscosity values corresponding to each temperature point, and obtains the preprocessed viscosity-temperature curve data corresponding to each sample; where each viscosity-temperature curve data corresponds to multiple temperature points and The viscosity value corresponding to each temperature point; the second data processing method includes: at least one or more combinations of logarithmic processing, averaging processing, mean variance processing, maximum value and minimum value normalization processing;

In this embodiment, the empirical formula of the viscosity-temperature curve fitted using the preset algorithm for the calibration set data X _2c is: log(μ)=log(μ ₀ )+L/(T/T ₀ -1), and According to the empirical formula of the fitted viscosity-temperature curve, the viscosity values corresponding to the preset five temperature points of 20°C, 30°C, 40°C, 50°C, and 60°C are calculated, and a second data processing method is used, namely The maximum and minimum values are normalized for preprocessing to obtain the preprocessed viscosity-temperature curve data corresponding to each sample.

Among them, μ represents the viscosity of the oil component, μ ₀ represents the ultimate viscosity of the oil component when the temperature is infinitely high, T represents the temperature, and T ₀ represents the temperature when the oil component solidifies into a solid (viscosity is infinite), L is a related parameter used to measure the degree of fit between the viscosity-temperature curve and the measured viscosity-temperature data.

Table 2 shows the processing methods of spectral data and viscosity-temperature curve data under different schemes of this embodiment.

Table 2 Processing methods of spectral data and viscosity-temperature curve data under different schemes

Plan serial number Spectral data processing methods Viscosity-temperature curve data processing method plan 1 No processing Maximum and minimum value normalization processing Scenario 2 mean centering Maximum and minimum value normalization processing Option 3 Mean variance processing Maximum and minimum value normalization processing Option 4 Vector normalization Maximum and minimum value normalization processing Option 5 Maximum and minimum value normalization processing Maximum and minimum value normalization processing Option 6 Standard normal variable transformation processing Maximum and minimum value normalization processing

Option 7 Multivariate scattering correction processing Maximum and minimum value normalization processing Option 8 Savitzky-Golay convolution smoothing Maximum and minimum value normalization processing Option 9 First order derivative method processing Maximum and minimum value normalization processing Option 10 detrending Maximum and minimum value normalization processing Plan 11 Baseline correction method processing Maximum and minimum value normalization processing

Step 203: Merge the spectrum data and viscosity-temperature curve data of each preprocessed data set to form multiple corresponding associated data sets, and associate them with the assay data sets corresponding to each sample through multiple association algorithms to construct Multiple corresponding physical property analysis models.

The multiple correlation algorithms include: at least one or more combinations of partial least squares, artificial neural network, and kernel partial least squares.

In this embodiment, 11 groups of processed near _- infrared spectrum data sets X _1c and viscosity-temperature _curve data sets X _2c are merged to form 11 groups of corresponding associated data sets Three preset algorithms, Partial Least Squares (PLS), Artificial Neural Network (ANN), and Kernel Partial Least Squares (KPLS), were associated with the laboratory data set Y _c corresponding to each sample, and a total of 33 95% fractions were constructed. Output temperature analysis model;

At the same time, the processed ₁₁ sets of near-infrared spectrum data sets The laboratory data set Y _c was correlated, and 33 near-infrared spectrum-95% distillation temperature analysis models were established as basic models for reference.

Step 204: Use the verification set data to verify and evaluate each physical property analysis model, and select the optimal model as the physical property quick evaluation model of the oil to be modeled.

The spectrum data and viscosity-temperature curve data of each sample in the verification set are combined to form a data set to be predicted, and each physical property analysis model is used for prediction to obtain multiple corresponding prediction result sets;

Each prediction result set is compared with the laboratory data set corresponding to each sample in the verification set, and the model with the smallest error in the verification set is selected as the physical property quick evaluation model of the oil to be modeled.

In this embodiment, _the verification set data X _{1v and} , recorded as Y _p ; compare the prediction result set Y _p with the laboratory data set Y _v corresponding to each sample in the validation set, and calculate the average error of the 33 95% distillation temperature analysis model validation set samples respectively, recorded as E _p1 ~ E _p33 ;

At the same time, 33 near-infrared spectrum-95% distillation temperature analysis models were used as the basic model for prediction, and the prediction result set was recorded as Y _1p ; the prediction result set Y _1p was compared with the laboratory data set Y _v corresponding to each sample in the verification set. For comparison, the average error of 33 near-infrared spectrum-95% distillation temperature analysis model validation set samples was calculated, recorded as E _p34 ~ E _p66 , as basic data for reference;

Please refer to Figure 3, which is a graph showing the error change trend of the model verification set constructed by each modeling method under different solutions in this embodiment;

As can be seen from Figure 3, compared with the model built using near-infrared spectra alone, after combining the viscosity-temperature curve data, the overall accuracy of the models built by each modeling method under different preprocessing schemes has been greatly improved. The model established using near-infrared spectra has the best effect when using the baseline correction method, that is, Scheme 11, and using the KPLS algorithm for modeling. The average error of the validation set samples is 1.88. After combining the viscosity-temperature curve data, the The average error in predicting the validation set results of the model built under the scheme dropped to 0.68, and the accuracy was significantly improved. Therefore, the model built by the KPLS method under the conditions of Scheme 11 can be selected as the optimal model as the 95% distillation temperature of the oil to be modeled. Quick review model.

The above embodiments greatly improve the accuracy of the quick evaluation analysis results of physical property indicators of various raw materials or intermediate products in the petrochemical industry by using the combined quick evaluation analysis method of spectrum data and viscosity-temperature curve data. The spectral data or wave spectrum data are combined with the viscosity-temperature curve data and jointly associated with various physical property indicators. The prediction effects of the models built using various modeling methods are far better than those built using spectral data alone. The prediction results The error has been greatly reduced and the accuracy has been significantly improved. Therefore, based on the existing technology, introducing viscosity-temperature curve data into the field of petrochemical material quick evaluation analysis can significantly improve the accuracy of the analysis model of various physical property indicators of oil products, and greatly make up for the current modeling. The method has the disadvantage of low accuracy in predicting complex physical properties, so it has extremely high application value.

On the basis of the existing technology, the present invention creatively proposes a quick evaluation model establishment method that combines spectral data and viscosity-temperature curve data, further improving the accuracy of the quick evaluation model. The oil spectrum or wave spectrum mainly reflects the group characteristics of each organic molecule in the sample. The viscosity information at each temperature point is helpful for analyzing the physical and chemical properties of heavy oil. Therefore, the spectrum data and the viscosity-temperature curve data are combined to reflect the oil The product information will be more detailed, rich and accurate. The physical property analysis model built on this basis has high accuracy and can accurately predict the material and physical property indicators of oil products.

Embodiment 3

Please refer to FIG. 4 which is a schematic structural diagram of a device for establishing a model for rapid evaluation of oil physical properties provided by an embodiment of the present invention.

The embodiment of the present invention provides a device for establishing a model for quick evaluation of oil physical properties, including: an acquisition module 301, a preprocessing module 302, a modeling module 303, and a screening module 304.

In this embodiment, the acquisition module 301 is used to obtain sample data of the oil product to be modeled and divide it into a calibration set and a verification set; wherein the sample data of the oil product to be modeled includes spectrum data of each sample, Viscosity-temperature curve data and laboratory data corresponding to each sample; the spectral data are spectral data or spectral data of the oil to be modeled;

The acquisition module uses the Kolmogorov-Smirnov algorithm to divide all samples into two categories: correction set and verification set. Among them, the correction set data is used to build the model, and the verification set data is used to verify the evaluation model.

The preprocessing module 302 is used to preprocess the calibration set data respectively through a variety of data processing methods to obtain multiple corresponding preprocessed data sets; wherein the preprocessing module includes: a spectrum data processing unit and a viscosity-temperature curve Data processing unit.

The spectrum data processing unit is used to preprocess the spectrum data, specifically:

For the spectrum data of each sample in the calibration set, multiple first data processing methods are selected for preprocessing, and preprocessed spectrum data corresponding to each sample is obtained, wherein the first data processing method includes: None Processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing processing, first derivative At least two or more combinations of method processing, detrending method processing, and baseline correction method processing;

After obtaining the preprocessed spectrum data corresponding to each sample, perform dimensionality reduction processing on each of the preprocessed spectrum data according to a preset dimensionality reduction method; wherein, the dimensionality reduction method Including simple model method or principal component analysis method;

The viscosity-temperature curve data processing unit is used to preprocess the viscosity-temperature curve data, specifically:

For the viscosity-temperature curve data of each sample in the calibration set, a preset algorithm is used to fit the empirical formula of the viscosity-temperature curve, and the corresponding viscosity value is calculated based on the preset multiple temperature points, and then processed according to a variety of second data This method preprocesses multiple temperature points and the viscosity values corresponding to each temperature point, and obtains the preprocessed viscosity-temperature curve data corresponding to each sample; where each viscosity-temperature curve data corresponds to multiple temperature points and The viscosity value corresponding to each temperature point; the second data processing method includes: at least one or more combinations of logarithmic processing, averaging processing, mean variance processing, maximum value and minimum value normalization processing.

The modeling module 303 is used to merge the spectrum data and viscosity-temperature curve data of each of the preprocessed data sets to form multiple corresponding associated data sets, and use multiple association algorithms to combine them with the assay data sets corresponding to each sample. Association to build multiple corresponding physical property analysis models.

The modeling module presets multiple correlation algorithms for modeling, wherein the multiple correlation algorithms include: at least one or more combinations of partial least squares, artificial neural network, and kernel partial least squares.

The screening module 304 is used to verify and evaluate each physical property analysis model using the verification set data, and select the optimal model as the physical property quick evaluation model of the oil to be modeled.

The screening module is used to verify and evaluate each physical property analysis model using the verification set data, and select the optimal model as the physical property quick evaluation model of the oil to be modeled, specifically as follows:

The spectrum data and viscosity-temperature curve data of each sample in the verification set are combined to form a data set to be predicted, and each physical property analysis model is used for prediction to obtain multiple corresponding prediction result sets;

Each prediction result set is compared with the laboratory data set corresponding to each sample in the verification set, and the model with the smallest error in the verification set is selected as the physical property quick evaluation model of the oil to be modeled.

The device provided in this embodiment creatively uses a quick evaluation analysis mode that combines spectral data and viscosity-temperature curve data, further improving the accuracy of the quick evaluation model. The oil spectrum or wave spectrum mainly reflects the group characteristics of each organic molecule in the sample. The viscosity information at each temperature point is helpful for analyzing the physical and chemical properties of heavy oil. Therefore, the spectrum data and the viscosity-temperature curve data are combined to reflect the oil The product information will be more detailed, rich and accurate. The physical property analysis model built on this basis has high accuracy and can accurately predict the material and physical property indicators of oil products.

The above-mentioned specific embodiments further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above-mentioned are only specific embodiments of the present invention and are not intended to limit the scope of the present invention. . It is particularly pointed out that for those skilled in the art, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims (10)

A method for establishing a quick evaluation model of oil physical properties, which is characterized by including: Obtain the sample data of the oil product to be modeled and divide it into a calibration set and a verification set; wherein the sample data of the oil product to be modeled includes spectrum data, viscosity-temperature curve data of each sample, and laboratory data corresponding to each sample. ; The spectral data is the spectral data or wave spectrum data of the oil to be modeled; Through a variety of data processing methods, the calibration set data are preprocessed respectively to obtain multiple corresponding preprocessed data sets; The spectrum data and viscosity-temperature curve data of each of the preprocessed data sets are merged to form multiple corresponding associated data sets, and are associated with the assay data sets corresponding to each sample through multiple association algorithms to construct multiple corresponding physical property analysis model; Use the verification set data to verify and evaluate each physical property analysis model, and select the optimal model as the physical property quick evaluation model of the oil to be modeled. A method for establishing a quick evaluation model of oil physical properties according to claim 1, characterized in that the sample data of the oil to be modeled is obtained and divided into a correction set and a verification set, specifically: using the Kolmogorov-Smirnov algorithm All samples are divided into two categories: calibration set and validation set. A method for establishing a quick oil physical property evaluation model as claimed in claim 1, characterized in that the calibration set data are preprocessed respectively through a variety of data processing methods to obtain multiple corresponding preprocessed data sets, Specifically: For the spectrum data of each sample in the calibration set, multiple first data processing methods are selected for preprocessing, and preprocessed spectrum data corresponding to each sample is obtained, wherein the first data processing method includes: None Processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing processing, first derivative At least two or more combinations of method processing, detrending method processing, and baseline correction method processing; For the viscosity-temperature curve data of each sample in the calibration set, a preset algorithm is used to fit the empirical formula of the viscosity-temperature curve, and the corresponding viscosity value is calculated based on the preset multiple temperature points, and then processed according to a variety of second data This method preprocesses multiple temperature points and the viscosity values corresponding to each temperature point, and obtains the preprocessed viscosity-temperature curve data corresponding to each sample; where each viscosity-temperature curve data corresponds to multiple temperature points and The viscosity value corresponding to each temperature point; the second data processing method includes: at least one or more combinations of logarithmic processing, averaging processing, mean variance processing, maximum value and minimum value normalization processing. A method for establishing a quick oil physical property evaluation model as claimed in claim 3, characterized in that after obtaining the preprocessed spectrum data corresponding to each sample, it further includes: According to the preset dimensionality reduction method, dimensionality reduction processing is performed on each of the preprocessed spectral data; wherein the dimensionality reduction method includes a simple model method or a principal component analysis method. A method for establishing a quick evaluation model of oil physical properties according to claim 4, wherein the multiple correlation algorithms include: at least one of partial least squares, artificial neural network, kernel partial least squares, or Various combinations. A method for establishing a quick oil physical property evaluation model according to claim 5, characterized in that each physical property analysis model is separately verified and evaluated using the verification set data, and the optimal model is screened out as the to-be-modeled model. Quick evaluation model of physical properties of oil products, specifically: The spectrum data and viscosity-temperature curve data of each sample in the verification set are combined to form a data set to be predicted, and each physical property analysis model is used for prediction to obtain multiple corresponding prediction result sets; Each prediction result set is compared with the laboratory data set corresponding to each sample in the verification set, and the model with the smallest error in the verification set is selected as the physical property quick evaluation model of the oil to be modeled. A device for establishing a quick oil physical property evaluation model, which is characterized by including: an acquisition module, a preprocessing module, a modeling module, and a screening module; The acquisition module is used to obtain sample data of the oil to be modeled and divide it into a calibration set and a verification set; wherein the sample data of the oil to be modeled includes spectrum data, viscosity-temperature curve data of each sample, and The laboratory data corresponding to each sample; the spectral data is the spectral data or wave spectrum data of the oil to be modeled; The preprocessing module is used to preprocess the correction set data respectively through multiple data processing methods to obtain multiple corresponding preprocessed data sets; The modeling module is used to merge the spectrum data and viscosity-temperature curve data of each preprocessed data set to form multiple corresponding associated data sets, and use multiple association algorithms to combine them with the laboratory data corresponding to each sample. Set association to build multiple corresponding physical property analysis models; The screening module is used to verify and evaluate each physical property analysis model using the verification set data, and select the optimal model as the physical property quick evaluation model of the oil to be modeled. A device for establishing a quick oil physical property evaluation model according to claim 7, wherein the preprocessing module includes: a spectrum data processing unit and a viscosity-temperature curve data processing unit; The spectrum data processing unit is used to preprocess the spectrum data, specifically: For the spectrum data of each sample in the calibration set, multiple first data processing methods are selected for preprocessing, and preprocessed spectrum data corresponding to each sample is obtained, wherein the first data processing method includes: None Processing, mean centering processing, mean variance processing, vector normalization processing, maximum and minimum normalization processing, standard normal variable transformation processing, multivariate scattering correction processing, Savitzky-Golay convolution smoothing processing, first derivative At least two or more combinations of method processing, detrending method processing, and baseline correction method processing; After obtaining the preprocessed spectrum data corresponding to each sample, perform dimensionality reduction processing on each of the preprocessed spectrum data according to a preset dimensionality reduction method; wherein, the dimensionality reduction method Including simple model method or principal component analysis method; The viscosity-temperature curve data processing unit is used to preprocess the viscosity-temperature curve data, specifically: For the viscosity-temperature curve data of each sample in the calibration set, a preset algorithm is used to fit the empirical formula of the viscosity-temperature curve, and the corresponding viscosity value is calculated based on the preset multiple temperature points, and then processed according to a variety of second data This method preprocesses multiple temperature points and the viscosity values corresponding to each temperature point, and obtains the preprocessed viscosity-temperature curve data corresponding to each sample; where each viscosity-temperature curve data corresponds to multiple temperature points and The viscosity value corresponding to each temperature point; the second data processing method includes: at least one or more combinations of logarithmic processing, averaging processing, mean variance processing, maximum value and minimum value normalization processing. A device for establishing a quick oil physical property evaluation model as claimed in claim 7, characterized in that the modeling module presets a plurality of correlation algorithms for modeling, wherein the plurality of correlation algorithms include: partial least quadratic At least one or more combinations of multiplication, artificial neural network, and kernel partial least squares. A device for establishing a quick oil physical property evaluation model according to claim 9, characterized in that the screening module is used to verify and evaluate each physical property analysis model using the verification set data, and select the optimal model as the desired model. Describe the physical property quick evaluation model of the oil to be modeled, specifically: The spectrum data and viscosity-temperature curve data of each sample in the verification set are combined to form a data set to be predicted, and each physical property analysis model is used for prediction to obtain multiple corresponding prediction result sets; Each prediction result set is compared with the laboratory data set corresponding to each sample in the verification set, and the model with the smallest error in the verification set is selected as the physical property quick evaluation model of the oil to be modeled.

Images