A method for determining the components of unconventional gasoline additives

Abstract

The present invention discloses a method for determining the components of unconventional additives in gasoline, and relates to the field of testing and analyzing unconventional additives in gasoline, including: extracting a plurality of gasoline samples containing unconventional additives in batches using a pipette; placing one sample extracted each time on a separate window piece for infrared spectrometer experiment to obtain corresponding mid-infrared spectrum data; substituting the mid-infrared spectrum data into a multiple linear regression method to construct a quantitative analysis model; establishing a wavelength selection network for wavelength selection based on a neural network and an attention mechanism, and using the wavelength selection network to perform model tuning on the quantitative analysis model; and determining the unconventional additives in gasoline based on the tuned model. The present invention can quickly identify new unconventional additives and determine their concentrations.

Description

Technical Field

[0001] The invention relates to the field of testing and analyzing unconventional additives in gasoline, and in particular to a method for determining components of unconventional additives in gasoline. Background Art

[0002] Infrared and visible light are both electromagnetic waves. Infrared is a section of electromagnetic waves between visible light and microwaves. According to the wavelength range, infrared light can be divided into three regions: near infrared, mid-infrared and far infrared. Among them, the mid-infrared region is 2.5~25um, 4000~400cm -1 The range of can effectively reflect the various physical processes inside the molecule and the characteristics of the molecular structure, and plays an important role in solving problems related to molecular structure and chemical composition. The five unconventional additives in gasoline show obvious characteristic absorption peak positions in the mid-infrared spectrum, and there is a good correlation between the position, intensity and area of ​​these absorption peaks and the concentration of the analyte.

[0003] In the prior art, the development of infrared spectrometers can be traced back to the early 20th century. Since Coblentz began to use sodium chloride crystals as prisms for infrared spectroscopy in 1908, relevant researchers and testers have been exploring the analysis method of mid-infrared spectroscopy in order to better utilize mid-infrared to quickly and efficiently complete sample component analysis.

[0004] However, there is currently no relevant application of mid-infrared spectroscopy in the detection of unconventional additives in gasoline. Summary of the invention

[0005] The embodiment of the present invention provides a method for determining the components of unconventional gasoline additives, which can solve the problem in the prior art that mid-infrared spectroscopy is not applied in the detection of unconventional gasoline additives.

[0006] The embodiment of the present invention provides a method for determining the components of unconventional additives in gasoline, comprising: extracting a plurality of gasoline samples to be tested containing unconventional additives in batches using a pipette; placing one sample to be tested extracted each time on a separate window plate for an infrared spectrometer experiment to obtain corresponding mid-infrared spectrum data; substituting the mid-infrared spectrum data into a multivariate linear regression method to construct a quantitative analysis model; establishing a wavelength selection network for wavelength selection based on a neural network and an attention mechanism, and using the wavelength selection network to perform model tuning on the quantitative analysis model; wherein the wavelength selection network uses the characteristics of the attention mechanism to obtain the contribution weight of each wavelength point in the spectrum to the wavelength selection network; the use of the wavelength selection network to perform model tuning on the quantitative analysis model is to add sparse constraints to the contribution weights; and determining the unconventional additives in gasoline according to the tuned model.

[0007] Furthermore, when using a pipette to extract multiple gasoline samples containing unconventional additives in batches, sampling tests are required to obtain the standard adopted amount; the standard adopted amount means: fill the space between the KBr window and the gasket in the KBr separable liquid pool with the collected samples, allowing a liquid film to form a gap between the two, and then cover the other window to ensure that there is no overflow, which is the standard adopted amount.

[0008] Furthermore, before substituting the mid-infrared spectral data into the multivariate linear regression method to construct a quantitative analysis model, the mid-infrared spectral data needs to be denoised using a sliding average method, and the sliding average method calculation formula is:

[0009]

[0010] Among them, xs i,k represents the value of the i-th wavelength point of the k-th spectrum after sliding average, x k,j It represents the value of the j-th wavelength point of the k-th spectrum in the original spectrum, and w represents the sliding window radius.

[0011] Furthermore, the mid-infrared spectrometer should comply with GB / T21186 standard, and the effective wave number range of the spectrum should be no less than 4000cm -1 ~1000cm -1 , spectral resolution better than 2cm -1 , signal-to-noise ratio is better than 10000:1, wavelength accuracy is better than 0.01cm -1 , data sampling interval is better than 0.5cm -1 .

[0012] Furthermore, the separate window has a transmittance of >92% and a transmission wavelength of 50,000 to 400 cm -1 Potassium bromide KBr window, and the quantitative gasket thickness is greater than 0.1mm.

[0013] Furthermore, the mid-infrared spectrum data needs to be made into a mid-infrared spectrum graph, which contains characteristic absorption peaks, and the peak area, peak intensity and peak position offset can directly reflect its concentration.

[0014] Furthermore, the construction of the quantitative analysis model is specifically to construct an MLR model combining a spectral matrix and a component matrix:

[0015] The calculation formula of the spectral matrix is:

[0016]

[0017] The calculation formula of the component matrix is:

[0018]

[0019] The mathematical expression of the MLR model is:

[0020] Y m×1 =X m×n β n×1 +e m×1 ;

[0021] Among them, m and n are the wavelength points of the mid-infrared spectral data, β is the regression coefficient of the model, and e represents the residual vector;

[0022] Finally, the mathematical expression for solving for β is:

[0023] β=[(X T X) -1 X T Y] T ;

[0024] When X T When X is full rank, the regression coefficient β can be obtained.

[0025] Furthermore, the specific operation of using the wavelength selection network to perform model tuning on the quantitative analysis model is:

[0026] A data processing method for adding sparse constraints to contribution weights, wherein the calculation formula of the sparse constraints is as follows:

[0027]

[0028] Among them, λ is the parameter that controls the sparsity, s is the model coefficient vector, represents the difference between the measured mid-infrared spectral data and the data predicted by the model, Represents the mid-infrared spectrum data from i to the Nth.

[0029] The embodiment of the present invention provides a method for determining the components of unconventional additives in gasoline. Compared with the prior art, the method has the following beneficial effects:

[0030] Use a pipette to extract the sample and place it on a separate window to conduct an infrared spectrometer experiment. Only an infrared spectrometer is used for the experiment. The test method is simple. At the same time, the amount of sample sampled by the pipette is small. Only a few samples are needed to complete the modeling, and the cost is very low.

[0031] After mid-infrared spectroscopy experiments were conducted on a variety of gasoline samples containing unconventional additives, the mid-infrared spectral data images obtained showed that the characteristic areas of the mid-infrared band were obvious and there was almost no interference. Therefore, it is possible to conduct qualitative and quantitative analysis of the concentrations of multiple substances in one test.

[0032] The quantitative analysis adopts the multivariate linear regression analysis modeling method to establish the quantitative analysis model. The modeling method is simple. Subsequently, the neural network and attention mechanism are used to establish a wavelength selection network to tune the model. This enables the model to screen a large amount of data and select important data, thereby optimizing the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A sample test result of a method for determining the components of unconventional additives in gasoline provided by an embodiment of the present invention;

[0034] Figure 2 A method for determining the components of unconventional additives in gasoline provided by an embodiment of the present invention provides test data results after gasoline is mixed with unconventional amine additives of different concentrations;

[0035] Figure 3 Comparison results of predicted values ​​and measured values ​​of a gasoline unconventional additive component determination method provided by an embodiment of the present invention;

[0036] Figure 4 The present invention provides a method for determining the components of unconventional gasoline additives according to an embodiment of the present invention, and provides the infrared absorption region of common groups. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0038] See also Figures 1 to 4 , an embodiment of the present invention provides a method for determining the components of unconventional additives in gasoline, and the method for determining the components of unconventional additives in gasoline includes: Step 1: Collect a group of samples with a reasonable concentration gradient of the substances to be tested due to database modeling data collection. Step 2: Adjust the pipette range to find a suitable liquid sampling volume, select the range between the KBr window and the gasket to fill and form a liquid film, and cover it with another window without overflowing as the most suitable sample usage, that is, the standard usage amount. Step 3: Use the most suitable sampling volume in step 2 to perform mid-infrared spectral sampling on the sample in step 1. If multiple substances to be tested need to be tested, each substance to be tested needs to find the best usage according to the steps in step 2. Step 4: After the test is completed, wash the window with a cleaning agent and dry it. Step 5: Preprocess the measured mid-infrared spectral data, establish a quantitative analysis model, clarify the model evaluation parameters, and use the wavelength selection method based on neural network to tune the model.

[0039] The optimal sample volume, or the standard volume, means that the collected sample fills the space between the KBr window and the gasket in the KBr separable liquid pool, forming a liquid film between the two, and then covers the other window to ensure that there is no overflow. This is the optimal sample volume, or the standard volume. The preprocessing uses the sliding average method to denoise the mid-infrared spectral data, as well as baseline correction and wavelength selection.

[0040] The infrared spectrometer should comply with GB / T21186 standard, and the effective wave number range of the spectrum should be no less than 4000cm -1 ~1000cm -1 , spectral resolution better than 2cm -1 (inclusive), signal-to-noise ratio better than 10000:1, wavelength accuracy better than 0.01cm -1 , data sampling interval is better than 0.5cm -1 (inclusive); the window piece adopts light transmittance>92%, and the transmission wavelength is 50000~400cm -1 (0.2~2.5um) potassium bromide (KBr) window; quantitative gasket thickness>0.1mm.

[0041] Mid-infrared region (2.5~25um; 4000~400cm -1 ) can effectively reflect the various physical processes inside the molecules and the characteristics of the molecular structure. For the five common unconventional additives in gasoline, there are characteristic absorption peaks in the mid-infrared spectrum, and the peak area, peak intensity and peak position offset can directly reflect their concentration. When the gradient of the unconventional additives to be tested in the sample is reasonable and the test data is accurate, a model that is quite accurate to the test results obtained using statistical methods can be achieved by using small-scale samples and adopting the Multiple Linear Regression (MLR) analysis modeling method. This method is conducive to the rapid identification and quantitative analysis of unconventional additives in gasoline, and greatly shortens the sampling time, and quickly completes the determination of the possible content of unconventional additives.

[0042] The test results are as follows Figure 2 As shown in Figure 2, there are obvious differences in the amine recognition and quantitative regions of the mid-infrared absorption spectra of amine-based unconventional additives with different concentrations in the solvent.

[0043] A network for wavelength selection is constructed using neural networks and attention mechanisms, and the contribution weights of each wavelength point in the spectrum to the model are obtained using the characteristics of the attention mechanism. The initial value of the weight is the spectral data, which is then compressed to prevent the gradient update of the wavelength selection module from being too large during error back propagation. At the same time, some redundant information can be removed during the dimensionality reduction process to retain useful feature information.

[0044] In order to overcome the problem that the number of weights learned by the wavelength selection module is large, the differences are small, and it is not easy to screen, we add sparse constraints to the weights. The difference between the measured value and the predicted value obtained by the model prediction is compared, and λ is a parameter that controls the sparsity. This makes the learned wavelength point weights sparse, reducing a large number of wavelength points for screening, thereby optimizing the prediction accuracy of the model, such as Figure 3 shown.

[0045] The gasoline component rapid determination method based on mid-infrared spectrometer used in the present invention can use small-scale samples for database modeling, can greatly shorten the test cycle and sample collection time, and realize the rapid identification and judgment of amine unconventional additives in gasoline.

[0046] Advantages of this test method: 1) Separate windows are used, the test method is simple, and the operation threshold is low. 2) The sample requirement is small, and only a few samples are needed to complete database modeling. 3) The modeling method is simple. The obtained data is first preprocessed, including denoising, baseline correction and wavelength selection. Multi-distance linear regression is used to quantitatively analyze the spectrum to determine the model evaluation parameters. During this period, the wavelength selection method based on neural networks is used to tune the model. 4) The sample amount is small, and the detection and judgment costs are low. Generally, the single test amount is less than 0.2ml, and a single test can determine the content of substances that do not overlap at multiple absorption peak positions. 5) The infrared absorption peak characteristics of the five unconventional additives in gasoline are obvious and almost interference-free, so qualitative and quantitative analysis of substances can be achieved in a single test.

[0047] A specific embodiment is as follows:

[0048] Example 1: Using this method to complete the test of aniline unconventional additives in gasoline

[0049] (1) Select gasoline-aniline mixed samples with aniline concentrations of (5 g / L, 10 g / L, 15 g / L, 25 g / L, 35 g / L), let the samples stand at room temperature until they reach room temperature, and then shake them well.

[0050] (2) Use an infrared spectrometer that complies with GB / T21186 standard, with a scanning wavenumber range of 4000cm -1 ~1000cm -1 , resolution 0.5cm -1 , sampling interval 0.5cm -1 , the window transmittance is >92%, and a gasket with a thickness of 0.15 mm is used for quantitative background sampling.

[0051] (3) Use a pipette with a maximum range of 1 ml and drop the sample on the window surface to test the optimal sample volume.

[0052] (4) Clean the windows and let them dry.

[0053] (5) Use the medium range in (3) to sample the sample and drop it on the surface of the window. Slowly cover it with another window at an angle to prevent the sample from spilling.

[0054] (6) Use an infrared spectrometer to take samples, and the parameters are the same as those described in (2).

[0055] (7) Repeat (4) to (6) to complete sampling of all samples.

[0056] (8) Modeling analysis: The sliding average method is used to denoise the collected spectral data. The calculation formula is:

[0057] where xs i,k represents the value of the i-th wavelength point of the k-th spectrum after sliding average, x k,j Indicates the value of the jth wavelength point of the kth spectrum in the original spectrum, and w represents the sliding window radius. In order to eliminate the spectral baseline drift caused by environmental, instrumental, background and other factors, the collected data is processed by the iterative polynomial baseline correction method to remove the baseline of the spectrum without destroying the characteristic peak of the spectrum. The quantitative analysis model is mainly based on the multivariate linear regression method to establish the spectral matrix

[0058] With ingredient matrix The MLR model, the mathematical expression of the model is: m×1 =X m×n β n×1 +e m×1 , where β is the regression coefficient of the model and e represents the residual vector. Then β is solved and can be expressed as: β=[(X T X) -1 X T Y] T , when X T When X is full rank, the regression coefficient β can be obtained. Then the components of the unknown spectrum can be predicted: in is the component content predicted by the MLR model, x unknown For unknown spectrum. 2 The model is evaluated to measure the linear correlation between the two groups of variables, and the model is improved by observing the deviation between the predicted value and the measured value. At the same time, a wavelength selection network is established based on the neural network and attention mechanism, and the contribution weight of each wavelength point to the model is adaptively obtained. Then, the wavelength points are screened by thresholds, and the wavelength points with large characteristic contributions to the model are selected to establish a regression model with stronger interpretability and better performance.

[0059] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for determining the components of unconventional gasoline additives, characterized in that: The following steps are involved: Use a pipette to extract multiple gasoline samples containing unconventional additives in batches; A sample to be tested extracted each time is placed on a separate window to conduct an infrared spectrometer experiment to obtain corresponding mid-infrared spectrum data; Substitute the mid-infrared spectral data into the multivariate linear regression method to construct a quantitative analysis model; Based on the neural network and attention mechanism, a wavelength selection network for wavelength selection is established, and the wavelength selection network is used to optimize the quantitative analysis model; The wavelength selection network uses the characteristics of the attention mechanism to obtain the contribution weight of each wavelength point in the spectrum to the quantitative analysis model; the use of the wavelength selection network to perform model tuning on the quantitative analysis model is to add sparse constraints to the contribution weight; The model tuning includes: A data processing method for adding sparse constraints to contribution weights, wherein the calculation formula of the sparse constraints is as follows: Among them, λ is the parameter that controls the sparsity, s is the model sparse vector, represents the difference between the measured mid-infrared spectrum data and the predicted value of the wavelength selection network, i is the i-th mid-infrared spectrum data, and N is the N-th mid-infrared spectrum data; The unconventional additives in gasoline were determined based on the tuned model.

2. A method for determining the components of unconventional gasoline additives as claimed in claim 1, characterized in that: When the pipette is used to extract multiple gasoline samples containing unconventional additives in batches, sampling tests are required to obtain the standard amount; The standard amount used means that the collected sample is filled between the KBr window and the gasket in the KBr separable liquid pool, a liquid film is formed in the gap between the two, and then another window is covered to ensure that there is no overflow, which is the standard amount used.

3. A method for determining the components of unconventional gasoline additives as claimed in claim 1, characterized in that: Before substituting the mid-infrared spectral data into the multivariate linear regression method to construct a quantitative analysis model, the mid-infrared spectral data needs to be denoised using a sliding average method. The sliding average method calculation formula is: Among them, xs i,k represents the value of the i-th wavelength point of the k-th spectrum after sliding average, x k,j It represents the value of the j-th wavelength point of the k-th spectrum in the original spectrum, and w represents the sliding window radius.

4. A method for determining the components of unconventional gasoline additives as claimed in claim 1, characterized in that: The infrared spectrometer should comply with GB / T21186 standard, and the effective wave number range of the spectrum should be no less than 4000cm -1 ~1000cm -1 , spectral resolution better than 2cm -1 , signal-to-noise ratio is better than 10000:1, wavelength accuracy is better than 0.01cm -1 , data sampling interval is better than 0.5cm -1 .

5. A method for determining the components of unconventional gasoline additives as claimed in claim 1, characterized in that: The separate window has a transmittance of >92% and a transmission wavelength of 50,000 to 400 cm -1 Potassium bromide KBr window, and the quantitative gasket thickness is greater than 0.1mm.

6. A method for determining the components of unconventional gasoline additives as claimed in claim 1, characterized in that: The mid-infrared spectrum data needs to be made into a mid-infrared spectrum graph, which contains characteristic absorption peaks, and the peak area, peak intensity and peak position offset can directly reflect its concentration.

7. A method for determining the components of unconventional gasoline additives as claimed in claim 1, characterized in that: The construction of the quantitative analysis model is specifically to construct an MLR model combining a spectral matrix and a component matrix: The calculation formula of the spectral matrix is: The calculation formula of the component matrix is: The mathematical expression of the MLR model is: Y m×1 =X m×n b n×1 +e m×1 ; Among them, m and n are the wavelength points of the mid-infrared spectral data, β is the regression coefficient of the model, and e represents the residual vector; Finally, the mathematical expression for solving for β is: β=[(X T X) -1 X T Y] T ; When X T When X is full rank, the regression coefficient β can be obtained.

Images