An ethanol gasoline near-infrared detection method and system based on double-drive feature screening

Abstract

The application discloses an ethanol gasoline near-infrared detection method and system based on double-drive feature screening, and the method comprises the following steps: obtaining an ethanol content reference value and a near-infrared spectrum of an ethanol gasoline sample; performing pretreatment on the near-infrared spectrum to obtain a pretreated spectrum matrix; performing statistical significant spectrum segment screening according to the pretreated spectrum matrix and the ethanol content reference value to obtain statistical significant spectrum segments; determining effective chemical information spectrum segments according to the absorption mechanism of O-H bonds and C-H bonds in ethanol molecules in the near-infrared region; performing set intersection operation on the statistical significant spectrum segments and the effective chemical information spectrum segments to obtain associated effective spectrum segments; setting joint weights of variables in the associated effective spectrum segments, and establishing a partial least squares quantitative model according to the weighted associated effective spectrum segments; performing pretreatment on an ethanol gasoline sample to be detected, and performing prediction by using the partial least squares quantitative model to obtain an ethanol content prediction value, and determining whether the sample meets the requirements according to a preset threshold value.

Description

Technical Field

[0001] This invention belongs to the fields of rapid near-infrared spectroscopy analysis, chemometric modeling, and refined oil quality testing technology, specifically relating to a near-infrared detection method and system for ethanol gasoline based on dual-drive feature screening. Background Technology

[0002] Near-infrared spectroscopy utilizes the overtone and combination frequency absorption information of chemical bonds such as OH, CH, and CO in molecules to reflect the composition of samples. It offers advantages such as being non-destructive, rapid, requiring no complex pretreatment, and suitable for portable detection, and has been widely used for rapid testing of petroleum products. However, in ethanol gasoline systems, the CH-related absorptions of alkanes and aromatics in the gasoline matrix overlap with the OH and CH absorptions of ethanol molecules. Spectral acquisition is also affected by instrument drift, environmental noise, and baseline shift. Directly using full-spectrum modeling easily introduces redundant variables and non-target interference.

[0003] Existing methods for selecting spectral features typically rely on single statistical screening or single chemical prior screening. When relying solely on statistical methods such as VIP and UVE, although some bands may have statistical correlations with the target content, these correlations may originate from gasoline matrix, environmental noise, or chance, easily leading to "spurious correlations" and reducing model interpretability. When defining characteristic spectral bands solely based on chemical structure or effective chemical information, although the physical meaning is relatively clear, the statistical stability of each band under actual sample and instrument conditions is not fully considered, and it may be sensitive to instrument drift or differences in sample matrix.

[0004] Therefore, existing technologies still lack a method for screening and quantitatively modeling near-infrared features of ethanol gasoline that can simultaneously take into account the chemical absorption mechanism of the target components, statistical correlation, model prediction accuracy, computational efficiency, and the deployment requirements of portable devices. Summary of the Invention

[0005] This invention aims to address the shortcomings of existing technologies by proposing a rapid near-infrared detection method for ethanol gasoline based on chemo-statistical dual-drive characteristic spectral band selection. This method uses the near-infrared spectrum of an ethanol gasoline sample as input and ethanol volume fraction or ethanol content as a reference value for modeling. Through spectral preprocessing, determination of effective chemical information spectral bands, screening of statistically significant spectral bands, extraction of intersection spectral bands, joint weight allocation, and partial least squares quantitative modeling, a quantitative prediction model suitable for the rapid detection of unknown ethanol gasoline samples is obtained.

[0006] To achieve the above objectives, the present invention provides the following solution: A near-infrared detection method for ethanol gasoline based on dual-drive feature screening includes the following steps: Obtain reference values ​​for ethanol content and near-infrared spectra of ethanol gasoline samples; The near-infrared spectrum is subjected to baseline correction, noise suppression, and scale correction to obtain a preprocessed spectral matrix. Based on the preprocessed spectral matrix and the ethanol content reference value, statistically significant spectral bands are screened to obtain statistically significant spectral bands. ; Based on the absorption mechanism of OH and CH bonds in the near-infrared region of the ethanol molecule, the effective chemical information spectral bands were determined. ; The statistically significant spectral segment With the aforementioned effective chemical information spectral segment Perform set intersection operations to obtain the associated valid spectral bands. ; For the associated valid spectral bands The variables are set with joint weights, and a partial least squares quantitative model is established based on the weighted associated effective spectral segments. The ethanol gasoline sample to be tested is preprocessed and predicted using the partial least squares quantitative model to obtain the predicted ethanol content. The sample is then judged to meet the requirements based on a preset threshold.

[0007] Preferably, the method for obtaining the ethanol content reference value includes: collecting gasoline samples from different batches and different sources, adding a predetermined volume fraction of anhydrous ethanol to the gasoline samples, preparing ethanol gasoline samples covering the range of normal ethanol gasoline and abnormal blending, and recording the ethanol content reference value of each sample.

[0008] Preferably, the method for obtaining the near-infrared spectrum includes: using a Fourier transform near-infrared spectrometer or an equivalent near-infrared spectral acquisition device, using air or a suitable background as a reference, and employing transmission or diffuse reflection methods to acquire multiple parallel samples of each ethanol gasoline sample and take the average spectrum to obtain the near-infrared spectrum.

[0009] Preferably, the statistically significant spectral band is obtained. The methods include: Based on the preprocessed spectral matrix and the ethanol content reference value, an initial PLS model is established, and the statistical contribution of each wavenumber variable to the prediction target is calculated. Statistical importance is derived based on the aforementioned statistical contributions. And set a threshold τ to filter the statistically significant spectral segments. .

[0010] Preferably, the effective chemical information spectral segment for: in, This indicates that the combined frequency of the first-order stretching and bending vibrations of the OH bond in ethanol is located between 4661 cm⁻¹ and 5000 cm⁻¹. [8300,8500] indicates that the first harmonic and combination frequency of the CH bond are located in the range of 5660 cm⁻¹ to 7407 cm⁻¹, and [8300,8500] indicates that the second harmonic of the CH bond is located in the range of 8300 cm⁻¹ to 8500 cm⁻¹.

[0011] Preferably, the joint weights are: in, Indicates the joint weight. Represents the chemical prior weights. α represents the contribution ratio of the chemical prior, and β represents the contribution ratio of the statistical significance.

[0012] The present invention also provides a near-infrared detection system for ethanol gasoline based on dual-drive feature screening. The system applies the above-mentioned method and includes: a data acquisition module, a preprocessing module, a statistical spectral band determination module, a chemical spectral band determination module, a correlation spectral band determination module, a model construction module, and a sample detection module. The data acquisition module is used to obtain the reference value of ethanol content and near-infrared spectrum of ethanol gasoline samples; The preprocessing module is used to perform baseline correction, noise suppression, and scale correction on the near-infrared spectrum to obtain a preprocessed spectral matrix. The statistical spectral band determination module performs statistically significant spectral band screening based on the preprocessed spectral matrix and the ethanol content reference value to obtain statistically significant spectral bands. ; The chemical spectrum determination module determines the effective chemical information spectrum based on the absorption mechanism of OH and CH bonds in the near-infrared region of the ethanol molecule. ; The associated spectral segment determination module is used to determine the statistically significant spectral segments. With the aforementioned effective chemical information spectral segment Perform set intersection operations to obtain the associated valid spectral bands. ; The model building module is used to analyze the associated valid spectral bands. The variables are set with joint weights, and a partial least squares quantitative model is established based on the weighted associated effective spectral segments. The sample detection module is used to process the ethanol gasoline sample to be tested, and to predict the ethanol content using the partial least squares quantitative model. The module then determines whether the sample meets the requirements based on a preset threshold.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention breaks through the limitations of traditional methods where chemical priors and statistical screening are independent or simply spliced ​​together. For the first time, it uses set intersection operation to integrate the effective chemical information of the target molecule (such as the OH and CH characteristic absorption regions of ethanol) with the statistical significance index SIj of the variable, ensuring that the selected spectral segments have both clear physical origins and high data-driven reliability. This effectively suppresses the "false correlation" interference that is easily generated by pure statistical methods (such as the CH peak of aromatic hydrocarbons being mistakenly selected), and also makes up for the defect of pure chemical methods being sensitive to instrument drift. Under five different preprocessing conditions, the RMSEP fluctuation range of the model is only 0.4568~0.4876, and the anti-interference ability is significantly better than that of traditional methods. (2) The dynamic joint weight allocation mechanism of this invention breaks away from the conventional approach of equal weight fusion (α=β=0.5) or empirical assignment, and introduces adjustable coefficients α and β. The chemical and statistical weights were dynamically synthesized, and through system optimization, it was found that the model performance was optimal when β=0.6 (statistical contribution is slightly higher than chemical contribution). The RMSEP was reduced to 0.4469 and the RPD was increased to 41.7521. This counterintuitive optimal ratio quantifies the contribution ratio of chemical and statistical information in the real system for the first time, and provides a transferable weight optimization strategy for similar spectral modeling. (3) The technical effect of the simultaneous "dimensionality reduction and precision improvement" of this invention subverts the traditional understanding of the trade-off relationship that "dimensionality reduction inevitably leads to a loss of precision"—ICSBS reduces the feature dimension by 84.52% (from 3201 dimensions to 495 dimensions) while improving the prediction accuracy by 17.2% (RMSEP from 0.5397 to 0.4469). Its internal mechanism is that it eliminates the harmful variables and noise sensitive areas of "false correlation" in the full wavelength model through chemical-statistical dual verification, improves the effective information density and suppresses overfitting. In addition, the model's computational performance is improved by orders of magnitude: the single sample prediction time is reduced by 72.45%, memory usage is reduced by 81.45%, and batch throughput is increased by 6.20 times. The model file is only 209KB and can be easily deployed on portable near-infrared devices, providing a lightweight and high-precision solution for the engineering application of rapid on-site oil testing. Attached Figure Description

[0014] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 The images show the near-infrared spectra of 44 ethanol gasoline samples after different preprocessing methods in this embodiment of the invention. Among them, (a) is before preprocessing, (b) is after processing with the first derivative of SG, (c) is after processing with standard normal transformation, (d) is after processing with vector normalization, (e) is after processing with multivariate scattering correction, and (f) is after processing with SG smoothing. Figure 3 The following are band distribution diagrams of different feature selection methods in embodiments of the present invention, wherein (a) is after pure statistical screening, (b) is after statistical screening according to effective chemical information ECI / H, and (c) is after statistical screening according to the method of the present invention; Figure 4 The above is a performance comparison chart of different modeling methods in the embodiments of the present invention, wherein (a) is a radar chart of prediction performance and (b) is a radar chart of computational performance. Figure 5 This is a schematic diagram of the model parameter optimization results in an embodiment of the present invention, where (a) shows the effect of the number of latent variables on the explained variance, and (b) shows the effect of the β value in the joint weights on RMSEP and RPD. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] Example 1 In this embodiment, as Figure 1 As shown, a near-infrared detection method for ethanol gasoline based on dual-drive feature screening includes the following steps: S1. Obtain the reference value of ethanol content and near-infrared spectrum of ethanol gasoline samples.

[0019] The method for obtaining reference values ​​for ethanol content includes: collecting gasoline samples from different batches and sources, adding a predetermined volume fraction of anhydrous ethanol to the gasoline samples, preparing ethanol gasoline samples covering normal ethanol gasoline and abnormal blending ranges, and recording the reference values ​​for ethanol content of each sample.

[0020] The method for obtaining near-infrared spectra includes: using a Fourier transform near-infrared spectrometer or an equivalent near-infrared spectral acquisition device, with air or a suitable background as a reference, employing transmission or diffuse reflectance methods, acquiring multiple parallel spectra for each ethanol gasoline sample, and averaging the spectra to obtain the near-infrared spectrum. In this embodiment, the preferred conditions are: 5mm optical path cuvette, transmission mode, scanning range 4000cm⁻¹ to 10000cm⁻¹, 32 scans, and a resolution of 4cm⁻¹; acquiring multiple parallel spectra for each sample and averaging the spectra.

[0021] This embodiment is not limited to a specific model of near-infrared spectrometer, nor is it limited to transmission mode; as long as a stable spectral response of an ethanol gasoline sample in the near-infrared region can be obtained, this embodiment can be used for modeling and detection.

[0022] S2. Baseline correction, noise suppression, and scale correction are performed on the near-infrared spectrum to obtain the preprocessed spectral matrix.

[0023] In this embodiment, the preprocessing method may be one or more of Savitzky-Golay first derivative (SG first derivative), Savitzky-Golay smoothing (SG smoothing), standard normal variable transformation (SNV), vector normalization (VN), and multivariate scattering correction (MSC); preferably, SG first derivative preprocessing is used to reduce the influence of baseline offset and instrument fluctuation.

[0024] S3. Based on the preprocessed spectral matrix and the reference value of ethanol content, statistically significant spectral bands are screened to obtain statistically significant spectral bands. .

[0025] Statistically significant spectral bands were obtained The method includes: establishing an initial PLS model based on the preprocessed spectral matrix and ethanol content reference values, and calculating the statistical contribution of each wavenumber variable to the prediction target; obtaining statistical importance based on the statistical contribution. And set a threshold τ to filter statistically significant spectral segments. Statistical significance screening can employ principal component correlation weighting, pure statistical screening (VIP), uninformative variable elimination (UVE), continuous projection algorithm (SPA), competitive adaptive reweighted sampling (CARS), or other feature selection methods that can characterize the contribution of variables; preferably, the final spectral segment is formed by cross-constraint of statistical significance and effective chemical information.

[0026] In this embodiment, let the preprocessed spectral matrix be X, the ethanol content reference value vector be y, and the wavenumber value corresponding to the k-th wavenumber variable be y. Statistical importance is obtained through principal component correlation weights, VIP values, stability indices, or equivalent statistical indices. ,when When this wavenumber variable is classified into a statistically significant spectral segment... .

[0027] S4. Determine the effective chemical information spectral band based on the absorption mechanism of OH and CH bonds in the near-infrared region of the ethanol molecule. .

[0028] In this embodiment, a set of chemical spectral segments is determined based on the effective chemical information of the ethanol target molecule, thus obtaining the effective chemical information spectral segments. for: in, This indicates that the combined frequency of the first-order stretching and bending vibrations of the OH bond in ethanol is located between 4661 cm⁻¹ and 5000 cm⁻¹. The range [5660 cm⁻¹ to 7407 cm⁻¹] indicates that the first-order harmonics and combination frequencies related to the CH bond are located in the range of 5660 cm⁻¹ to 7407 cm⁻¹, and [8300, 8500] indicates that the second-order harmonics of the CH bond are located in the range of 8300 cm⁻¹ to 8500 cm⁻¹. Without changing the essence of the present invention, the above ranges can be appropriately expanded or contracted according to the instrument resolution, wavenumber calibration error, or differences in the sample system.

[0029] S5. Statistically significant spectral segments With effective chemical information spectrum Perform set intersection operations to obtain the associated valid spectral bands. .

[0030] In this embodiment, by Obtain the associated valid spectral bands Compared to full-wavelength modeling, this step can eliminate non-target interference absorption such as aromatics and alkanes in the gasoline matrix, as well as noise-sensitive regions; compared to single statistical screening, this step can reduce random correlation variables; compared to single chemical screening, this step can introduce statistical stability constraints on sample data.

[0031] S6. For the associated valid spectral segments The variables are set with joint weights, and a partial least squares quantitative model is established based on the weighted associated effective spectral segments.

[0032] In this embodiment, for The internal variables are assigned joint weights, and the joint weights are: in, Indicates the joint weight. Represents the chemical prior weights. α represents the normalized statistical weight, β represents the contribution ratio of the chemical prior, and β represents the contribution ratio of statistical significance. It can be set to 1 or allocated according to the strength of chemical absorption. can be Normalization yields the result.

[0033] A partial least squares quantitative model was established using weighted correlation effective spectral segments, and the number of latent variables and weight ratios were optimized using leave-one-out cross-validation, prediction set validation, or equivalent validation. The preferred model was the SG first derivative-ICSBS-PLS model, with an optimal number of latent variables of 2.

[0034] S7. Preprocess the ethanol gasoline sample to be tested, and use the partial least squares quantitative model to predict the ethanol content. Then, determine whether the sample meets the requirements based on the preset threshold.

[0035] In this embodiment, near-infrared spectra of unknown ethanol gasoline samples are collected, and the samples are input into a quantitative model using the same preprocessing, spectral band screening, and weighting methods as described above. The predicted ethanol content is output, and the sample is judged to meet the requirements based on a preset threshold or quality standard.

[0036] Example 2 In this embodiment, the chemostatistical dual-drive characteristic spectral band selection method (ICSBS) was systematically validated based on 44 ethanol gasoline samples (ethanol concentration 0.5%~80%, v / v, acquisition range 4000~10000 cm⁻¹). The specific steps were as follows: First, preprocessing methods such as the first derivative of SG (SG 1stD) or standard normal transformation (SNV) were used to correct the spectral baseline. Figure 2 As shown; then, based on the effective chemical information of the ethanol molecule, a chemical constraint set Scheme was defined (including OH combination frequency 4661~5000cm-1, CH first overtone frequency 5660~6001cm-1, OH first overtone frequency 6060~7141cm-1, CH second overtone frequency 8300~8500cm-1). A partial least squares quantitative model was then used to extract principal components highly correlated with the dependent variable. And based on the correlation coefficient between the principal components and the dependent variable and the weight of wavelength on principal components Calculate the statistical importance index for each wavelength and select the set of statistically significant spectral bands that meet the criteria (threshold is 1): in, The threshold for statistically significant spectral band importance is represented by k, where k represents the total number of wavelengths, m represents the number of principal components, and h represents the principal component index number. denoted as the set of statistically significant spectral bands, and j represents the wavelength index number selected for chemical characteristics.

[0037] Based on the effective chemical information (ECI) of the target molecule, its characteristic absorption region in the near-infrared spectrum is identified. By analyzing the relationship between the vibrational modes of molecular groups and the spectral response, a set of candidate spectral bands directly related to the target properties is delineated. And exclude interference signals from non-target components.

[0038] in, This indicates the specific wavelength value corresponding to index j. Represents full-spectrum information By using set intersection operations to extract overlapping regions from chemical and statistical screening, a set of correlated and effective spectral bands is formed: .

[0039] This step ensures that the selected spectral bands have both clear physical meaning and data-driven reliability, and further... Wavelength allocation joint weight To obtain the input spectral information used for modeling Its chemical weight (Default setting is 1) and statistical weights (and Positive correlation) Dynamic synthesis: Where α and β are adjustable coefficients ( To determine the optimal weighting ratio, the model performance was examined under conditions where α:β were 0.9:0.1, 0.5:0.5, and 0.4:0.6, respectively. Simultaneously, control methods such as full-wavelength modeling, purely statistical screening (VIP), and purely chemical screening (ECI) were compared. Figure 3 As shown, there are significant differences in the distribution of bands selected by different feature selection methods, and the bands selected by different feature selection methods differ significantly. Pure statistical screening (VIP) mistakenly selected some spurious variables related to aromatic CH bonds, such as... Figure 3 As shown in (a); although pure chemical screening (ECI) has a clear physical meaning, it retains a noise-sensitive region, such as... Figure 3 As shown in (b). The ICSBS method of this invention, as... Figure 3 As shown in (c), through chemo-statistical intersection operation, the selected bands are all located within the effective absorption range of ethanol OH and CH, and statistical significance verification has eliminated spurious correlations and noise interference from the gasoline matrix, providing reliable input for high-precision modeling. The performance of different modeling methods was comprehensively compared using radar charts. From the prediction performance radar chart, as shown... Figure 4As shown in Figure (a), the ICSBS-PLS model of this invention outperforms full-wavelength modeling, pure statistical screening (VIP), and pure chemical screening (ECI) methods in terms of RMSEP, RPD, R², and MAE. Specifically, the RMSEP is reduced to 0.4469, and the RPD reaches 41.75, demonstrating a significant improvement in prediction accuracy. This is illustrated by the computational performance radar chart, as shown in Figure (a). Figure 4 As shown in (b), ICSBS-PLS exhibits significant advantages in prediction time (27 μs), batch throughput (37,037 / s), and memory usage (0.209 MB). The final model was then optimized by adjusting the number of principal factors and weight ratios. Figure 5 Figure (a) shows the effect of the number of latent variables on the cumulative explained variance of the X and Y matrices. When the number of latent variables is 2, the explained variance of both X and Y has reached a high level. Further increasing the number of latent variables has very limited effect on improving the explained variance. Therefore, the present invention prefers the number of latent variables to be 2, so as to avoid overfitting while ensuring the prediction accuracy. Figure 5 (b) examined the impact of the β value (statistical contribution ratio) in the joint weights on RMSEP and RPD. The model performance was optimal when β=0.6 (RMSEP=0.4469, RPD=41.75), indicating that the statistical contribution was slightly higher than the chemical contribution. The optimal allocation ratio was found to be 1, verifying the effectiveness of the dynamic joint weight allocation mechanism. The prediction performance under different schemes is summarized in Table 1.

[0040] Table 1 ; As shown in Table 1, the model performed best when using SG 1stD preprocessing, ICSBS screening, and a weighting of α:β = 0.4:0.6: the root mean square error of prediction (RMSEP) was 0.4469, the residual prediction ratio (RPD) was 41.75, the number of features was only 495 (a reduction of 84.52% compared to the full wavelength model), the single-sample prediction time was 27 μs, and the memory usage was 209 KB. If the weights were changed to equal weight (0.5:0.5) or chemically dominant (0.9:0.1), the RMSEP increased to 0.4568 and 0.4597 respectively, indicating that a slightly higher statistical contribution than a chemical contribution (β = 0.6) was optimal. Pure statistical screening (VIP) deteriorated to 0.4888 due to the misselection of a spurious correlation variable involving aromatic CH bonds; pure chemical screening (ECI) significantly reduced the RMSEP to 0.6179 by retaining the noise-sensitive region; although the full wavelength model had an RMSEP of 0.4703, its computational load was high. Further investigation was conducted on the robustness of ICSBS (α:β=0.4:0.6) under different preprocessing methods. The results are shown in Table 2. The RMSEP fluctuation range was only 0.4469~0.4876, and the RPD was higher than 38, which proves that the method of the present invention has strong anti-interference ability and universality.

[0041] Table 2 In summary, the chemostatistical dual-drive characteristic spectral band selection method (ICSBS) proposed in this invention achieves the technical effect of "dimensionality reduction and refinement" under the optimal dynamic weight (α:β=0.4:0.6), and is insensitive to the preprocessing method, providing a high-precision, lightweight, and robust solution for rapid on-site detection of near-infrared spectroscopy.

[0042] Example 3 In this embodiment, an ethanol gasoline near-infrared detection system based on dual-drive feature screening includes: a data acquisition module, a preprocessing module, a statistical spectral band determination module, a chemical spectral band determination module, a correlation spectral band determination module, a model construction module, and a sample detection module.

[0043] The data acquisition module obtains the ethanol content reference value and near-infrared spectrum of the ethanol gasoline sample. The preprocessing module performs baseline correction, noise suppression, and scale correction on the near-infrared spectrum to obtain the preprocessed spectral matrix. The statistical spectral band determination module screens statistically significant spectral bands based on the preprocessed spectral matrix and the ethanol content reference value to obtain the statistically significant spectral bands. The chemical spectrum determination module identifies effective chemical information spectrum bands based on the absorption mechanism of OH and CH bonds in the near-infrared region of the ethanol molecule. The associated spectral band determination module is used to identify statistically significant spectral bands. With effective chemical information spectrum Perform set intersection operations to obtain the associated valid spectral bands. The model building module is used to correlate valid spectral bands. The variables are weighted together, and a partial least squares quantitative model is established based on the weighted associated effective spectral segments. The sample detection module processes the ethanol gasoline samples to be tested, uses the partial least squares quantitative model to predict the ethanol content, and determines whether the sample meets the requirements based on a preset threshold.

[0044] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A near-infrared detection method for ethanol gasoline based on dual-drive feature screening, characterized in that, Includes the following steps: Obtain reference values ​​for ethanol content and near-infrared spectra of ethanol gasoline samples; The near-infrared spectrum is subjected to baseline correction, noise suppression, and scale correction to obtain a preprocessed spectral matrix. Based on the preprocessed spectral matrix and the ethanol content reference value, statistically significant spectral bands are screened to obtain statistically significant spectral bands. ; Based on the absorption mechanism of OH and CH bonds in the near-infrared region of the ethanol molecule, the effective chemical information spectral bands were determined. ; The statistically significant spectral segment With the aforementioned effective chemical information spectral segment Perform set intersection operations to obtain the associated valid spectral bands. ; For the associated valid spectral bands The variables are set with joint weights, and a partial least squares quantitative model is established based on the weighted associated effective spectral segments. The ethanol gasoline sample to be tested is preprocessed and predicted using the partial least squares quantitative model to obtain the predicted ethanol content. The sample is then judged to meet the requirements based on a preset threshold.

2. The near-infrared detection method for ethanol gasoline based on dual-drive feature screening according to claim 1, characterized in that, The method for obtaining the ethanol content reference value includes: collecting gasoline samples from different batches and sources, adding a predetermined volume fraction of anhydrous ethanol to the gasoline samples, preparing ethanol gasoline samples covering the range of normal ethanol gasoline and abnormal blending, and recording the ethanol content reference value of each sample.

3. The near-infrared detection method for ethanol gasoline based on dual-drive feature screening according to claim 1, characterized in that, The method for obtaining the near-infrared spectrum includes: using a Fourier transform near-infrared spectrometer or an equivalent near-infrared spectral acquisition device, with air or a suitable background as a reference, and employing transmission or diffuse reflection methods to acquire the near-infrared spectrum of each ethanol gasoline sample multiple times in parallel and taking the average spectrum.

4. The near-infrared detection method for ethanol gasoline based on dual-drive feature screening according to claim 1, characterized in that, The statistically significant spectral bands were obtained. The methods include: Based on the preprocessed spectral matrix and the ethanol content reference value, an initial PLS model is established, and the statistical contribution of each wavenumber variable to the prediction target is calculated. Statistical importance is derived based on the aforementioned statistical contributions. And set a threshold τ to filter the statistically significant spectral segments. .

5. The near-infrared detection method for ethanol gasoline based on dual-drive feature screening according to claim 1, characterized in that, The effective chemical information spectral segment for: in, This indicates that the combined frequency of the first-order stretching and bending vibrations of the OH bond in ethanol is located between 4661 cm⁻¹ and 5000 cm⁻¹. [8300,8500] indicates that the first harmonic and combination frequency of the CH bond are located in the range of 5660 cm⁻¹ to 7407 cm⁻¹, and [8300,8500] indicates that the second harmonic of the CH bond is located in the range of 8300 cm⁻¹ to 8500 cm⁻¹.

6. The near-infrared detection method for ethanol gasoline based on dual-drive feature screening according to claim 1, characterized in that, The joint weight is: in, Indicates the joint weight. Represents the chemical prior weights. α represents the contribution ratio of the chemical prior, and β represents the contribution ratio of the statistical significance.

7. A near-infrared detection system for ethanol gasoline based on dual-drive feature screening, wherein the system applies the method described in any one of claims 1-6, characterized in that, include: The system includes a data acquisition module, a preprocessing module, a statistical spectral band determination module, a chemical spectral band determination module, a correlation spectral band determination module, a model building module, and a sample detection module. The data acquisition module is used to obtain the reference value of ethanol content and near-infrared spectrum of ethanol gasoline samples; The preprocessing module is used to perform baseline correction, noise suppression, and scale correction on the near-infrared spectrum to obtain a preprocessed spectral matrix. The statistical spectral band determination module performs statistically significant spectral band screening based on the preprocessed spectral matrix and the ethanol content reference value to obtain statistically significant spectral bands. ; The chemical spectrum determination module determines the effective chemical information spectrum based on the absorption mechanism of OH and CH bonds in the near-infrared region of the ethanol molecule. ; The associated spectral segment determination module is used to determine the statistically significant spectral segments. With the aforementioned effective chemical information spectral segment Perform set intersection operations to obtain the associated valid spectral bands. ; The model building module is used to analyze the associated valid spectral bands. The variables are set with joint weights, and a partial least squares quantitative model is established based on the weighted associated effective spectral segments. The sample detection module is used to process the ethanol gasoline sample to be tested, and to predict the ethanol content using the partial least squares quantitative model. The module then determines whether the sample meets the requirements based on a preset threshold.

Images