A methanol raman quantitative detection method for complex oil background environment

Abstract

This invention discloses a Raman quantitative detection method for methanol in complex oil background environments. The method first acquires Raman spectral data from multiple sets of pure insulating oil samples, constructs a characteristic spectral matrix of the oil matrix, and extracts characteristic vector sets of the oil matrix through principal vibrational mode decomposition, establishing a characteristic spectral space of the oil matrix. Spatial projection is performed on the Raman spectrum of the oil sample to obtain background projection coefficients and reconstruct the oil matrix background spectrum. The residual signal is obtained through difference. The residual signal is integrated within the methanol characteristic peak region to obtain the residual characteristic integral intensity. Simultaneously, a residual correction term is calculated, and a methanol concentration prediction model including the residual correction factor is constructed, achieving accurate quantification of methanol under complex oil background conditions. This method eliminates the need for extraction and separation of oil samples, enabling in-situ and rapid detection, and is applicable to background interference suppression and quantitative analysis under different oil sample types and operating conditions.

Description

Technical Field

[0001] This invention relates to the fields of spectral analysis and power equipment condition monitoring technology, and in particular to a quantitative Raman detection method for methanol in complex oil background environments. Background Technology

[0002] Transformers are critical equipment in power systems, and the reliability of their insulation systems directly affects the safe and stable operation of the power grid. The internal insulation structure of a transformer typically consists of insulating paper and insulating oil. Under long-term thermal stress, electrical stress, and oxidative environments, the insulating paper degrades, producing various low-molecular-weight decomposition products. Studies have shown that methanol is one of the important characteristic products generated during the early degradation of insulating paper, and its content changes in insulating oil can reflect the decrease in the degree of polymerization of the insulating paper. Therefore, it is widely used to assess the aging degree and remaining life of transformer insulation. Currently, the main method for detecting methanol in oil is chromatography. This method usually requires pretreatment operations such as extraction, concentration, or separation of oil samples, making the detection process relatively complex, the detection cycle long, and requiring high standards for experimental conditions and operators, making it difficult to meet the needs of rapid on-site detection and online monitoring. Raman spectroscopy, with its advantages of requiring no complex sample preparation, direct detection of liquid samples, and fast response speed, has potential application value in the analysis of trace components in oil.

[0003] However, in actual transformer insulating oil systems, the oil matrix itself has a strong and complex Raman background signal. Different oil sources, additive types, operating years, and operating conditions can all lead to changes in the background spectral morphology. When the methanol content is in a low concentration range, its characteristic peak intensity is relatively weak and easily affected by the coverage or overlap of the oil matrix background signal, thereby reducing the accuracy and stability of quantitative analysis.

[0004] Existing quantitative methods based on Raman spectroscopy often employ simple baseline correction, polynomial fitting, or fixed background subtraction for spectral preprocessing. These methods typically assume relatively stable background variations, making it difficult to fully characterize background differences between different oil samples. When the oil matrix background changes, insufficient or excessive background subtraction can lead to systematic biases in the quantitative results. Furthermore, under complex background conditions, relying solely on the area of ​​a single characteristic peak to establish a concentration model often fails to account for background interference factors, affecting the reliability of low-concentration methanol detection.

[0005] Therefore, in complex oil background environments, there is an urgent need for a Raman quantitative analysis method that can characterize the background variation law of the oil matrix and achieve adaptive background subtraction, so as to improve the accuracy and stability of methanol detection in oil and meet the practical application needs of in-situ rapid detection and aging assessment of transformers. Summary of the Invention

[0006] Objective of the Invention: The objective of the present invention is to provide a Raman quantitative detection method for methanol in a complex oil background environment.

[0007] Technical Solution: The Raman quantitative detection method for methanol in a complex oil background environment described in the present invention includes the following steps: (1) Collect Raman spectral data of multiple groups of pure insulating oil samples, and construct an oil matrix characteristic spectral matrix to characterize the background spectral change characteristics of insulating oil under different states; (2) Perform principal vibration mode decomposition on the oil matrix characteristic spectral matrix, extract a characteristic vector group representing the main background change law of the oil matrix, and construct an oil matrix characteristic spectral space based on the characteristic vector group; (3) Collect the Raman spectrum of the oil sample to be detected, project the Raman spectrum of the oil sample to be detected into the oil matrix characteristic spectral space, obtain the background projection coefficient, and reconstruct the oil matrix background spectrum based on the projection coefficient; (4) Perform differential processing on the original Raman spectrum of the oil sample to be detected and the reconstructed oil matrix background spectrum to obtain a residual signal, which is used to characterize the effective Raman response of the non-oil matrix components in the oil sample; (5) Integrate the residual signal within the methanol characteristic peak interval to obtain the residual characteristic integral intensity I res ; (6) Based on the residual characteristic integral intensity I res and the fitting parameters of the pre-established standard curve, and combined with the residual correction term Δ, construct a methanol concentration prediction model: ; In the formula, C is the predicted methanol concentration in the oil, k1, k2, and b are fitting parameters, Ires is the residual integral intensity within the methanol characteristic peak interval, and Δ is the residual correction term.

[0008] Furthermore, the method for constructing the oil matrix characteristic spectral matrix in step (1) includes: respectively collecting n groups of pure insulating oil Raman spectra: , construct a characteristic matrix: , where is the Raman shift variable.

[0009] Furthermore, the method for principal vibration mode decomposition in step (2) includes: performing eigenvalue decomposition on the characteristic matrix M to obtain the oil matrix principal vibration mode vector group: , where k < n; the vector group constitutes the oil matrix characteristic spectral space.

[0010] Furthermore, the method for projecting the Raman spectrum of the oil sample to be detected into the space and reconstructing the oil matrix background spectrum in step (3) includes: performing projection calculation on the Raman spectrum of the oil sample to be detected: Using background projection coefficient Reconstructed oil matrix background spectrum: .

[0011] Further, the residual signal mentioned in step (4) is: ,in Characterize the Raman response of methanol molecules after the oil matrix background is removed.

[0012] Furthermore, the residual characteristic integral intensity I mentioned in step (5) res The calculation formula is: ,in and This represents the Raman characteristic peak range of methanol.

[0013] The characteristic Raman peak range of the methanol is 1025-1055 cm⁻¹. -1 .

[0014] Furthermore, the formula for calculating the residual correction term Δ in step (6) is as follows: , used to reflect the degree of influence of oil-based background modeling error on the quantitative results of methanol; where N is the total number of spectral sampling points.

[0015] Furthermore, the method for establishing the standard curve fitting parameters k1, k2, and b in step (6) includes: preparing various methanol standard oil samples of known concentrations; performing steps (3) to (6) on the standard oil samples respectively; and using the residual integral intensity I... res Using the residual correction term Δ as the independent variable and the actual methanol concentration as the dependent variable, regression fitting is performed to obtain the values ​​of k1, k2, and b.

[0016] Furthermore, the characteristic spectral space of the oil matrix described in step (2) supports dynamic updates, supplementing new pure oil spectral data according to different oil sample types or operating conditions, and reconstructing the characteristic matrix M and the principal vibration mode vector group, thereby improving the Raman quantitative accuracy in complex oil background environments.

[0017] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) Improve the quantitative accuracy under complex oil background conditions By constructing the characteristic spectral matrix of the oil matrix and establishing the characteristic spectral space of the oil matrix through principal vibration mode decomposition, the background variation law of the oil matrix can be structurally characterized. By projecting the spectrum to be measured onto the characteristic space and reconstructing the background spectrum, adaptive separation of the background signal is achieved, thereby reducing the interference of different oil sample types and operating state changes on the methanol characteristic signal and improving the quantitative accuracy under complex oil background conditions.

[0018] (2) Enhance the signal-to-noise ratio of low-concentration methanol detection By performing differential processing on the original spectrum and the reconstructed background spectrum, the residual signal is extracted as an effective response of the non-oil matrix component, which separates the methanol characteristic peak from the strong background of the oil matrix, enhances the relative response intensity of the target component, and is beneficial for the stable detection of low concentration methanol.

[0019] (3) Introduce a residual correction mechanism to improve model stability By constructing a concentration prediction model that includes a residual correction term, the influence of background modeling error on the quantitative results is incorporated into the regression model for compensation, thereby reducing systematic errors caused by background matching bias and improving the adaptability and stability of the quantitative model under different oil samples and operating conditions.

[0020] (4) Enable rapid in-situ detection without extraction or separation This method is based on raw Raman spectroscopy data for modeling and quantitative analysis. It eliminates the need for extraction or separation of oil samples, allowing for direct detection of oil samples, thus shortening the detection process. It is suitable for rapid in-situ detection and online condition assessment.

[0021] (5) Supports dynamic updates and adapts to various oil sample environments. The characteristic spectral space of the oil matrix can be expanded and updated according to different oil sample types or operating conditions, so that the background modeling capability can be continuously optimized with data accumulation, thereby improving the detection reliability under long-term operating conditions. Attached Figure Description

[0022] Figure 1 These are Raman spectra of different oil sample matrices in Example 1 of the present invention.

[0023] Figure 2 This is a comparison diagram of the Raman spectrum of methanol in high-concentration oil in Example 2 of the present invention with that of crude oil and pure methanol.

[0024] Figure 3 The images show the Raman spectra of methanol in oils of different concentrations in Example 3 of this invention. Detailed Implementation

[0025] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and embodiments.

[0026] Example 1: Constructing the characteristic spectral space of oil-based matrix To construct a characteristic spectral model of the oil matrix under complex oil background conditions, Raman spectroscopy was first performed on a pure insulating oil sample. The test conditions were: excitation wavelength of 638 nm, laser power of 30 mW, integration time of 10 s, and scanning range of 600-1800 cm⁻¹. -1 Spectra of some pure oils, such as... Figure 1As shown in the figure. It can be observed from the figure that the oil matrix is ​​located at 800-1200 cm⁻¹. -1 The interval exhibits a relatively wide and gentle vibrational band structure, at approximately 1300-1500 cm. -1 Strong characteristic peaks of CH bending vibration exist in the region, especially at approximately 1450 cm⁻¹. -1 A distinct strong peak appears nearby, with an intensity much higher than other vibrational peaks, belonging to a typical vibrational mode of the long-chain alkane structure of oil molecules. Simultaneously, at 600-900 cm⁻¹... -1 Multiple sets of superimposed weak peaks exist in the interval, forming a continuously fluctuating background signal. It can be seen that the oil matrix spectrum as a whole exhibits a broad peak superposition structure with high overall intensity, belonging to a typical complex background environment.

[0027] Given that the oil matrix spectrum is not a simple constant baseline, but rather a structured signal formed by the superposition of multiple molecular vibrational modes, this invention employs a multi-sample modeling approach to express its overall vibrational characteristics. n sets of pure insulating oil Raman spectra were collected respectively:

[0028] in, The Raman shift variable represents the intensity distribution of the spectrum at different wavenumber positions. Each spectrum... All of these can be considered as function vectors defined at discrete wavenumber points. Arranging the above n spectra column-wise, we construct the characteristic spectral matrix of the oil matrix:

[0029] In matrix M, each column corresponds to the spectral data of a pure oil sample, and each row corresponds to the intensity values ​​of different samples at the same Raman shift. This matrix form enables a unified representation of oil-based spectral data.

[0030] The singular value decomposition of the constructed oil-matrix characteristic spectral matrix M is expressed as follows:

[0031] Where U is the left singular vector matrix, and its column vectors are orthogonal basis functions defined on the Raman shift space; Σ is the singular value diagonal matrix, and its diagonal elements represent the contribution of the corresponding vibrational mode to the overall spectral variation; V T It is a right singular vector matrix that reflects the weight distribution of different samples under each vibration mode.

[0032] This decomposition essentially decomposes the original spectral matrix M into a linear combination of several orthogonal vibrational modes, where the first k columns of the left singular vector U correspond to the main vibrational modes of the oil matrix; singular values This characterizes the contribution of the corresponding vibrational mode to the overall spectral variation. It is achieved by selecting the left singular vectors corresponding to the top k largest singular values. This allows the construction of the main vibration mode vector set of the oil-based matrix.

[0033] Thus, the vector set constitutes the characteristic spectral space of the oil matrix:

[0034] This space enables high-precision linear reconstruction of the spectra of oil samples from different sources, indicating that the vibrational structure of the oil matrix possesses low-dimensional structural characteristics mathematically. Since k is significantly smaller than n, it suggests that although the oil background is complex, its main variations can be expressed by a finite number of principal vibrational modes, thus providing a stable basis for subsequent background reconstruction of the samples to be tested.

[0035] Through the above modeling process, a structured representation of the complex vibration background of the oil matrix is ​​achieved, avoiding the error problem caused by subtracting only a single background spectrum in traditional methods, and providing mathematical support for the subsequent separation of methanol characteristic signals.

[0036] Example 2: Determining the position of the characteristic peak of methanol in oil After determining the background structure of the oil matrix, a high-concentration methanol oil solution (100 mL / L) was prepared to pinpoint the characteristic peak positions of methanol in the oil environment. Raman spectroscopy was performed using a 638 nm laser as the excitation source, with a laser power of 30 mW and a single-point integration time of 10 s. Raman signals were acquired, and the results are as follows: Figure 2 As shown in the figure. It can be seen from the figure that, compared to the spectrum of pure oil, at approximately 1040 cm⁻¹... -1 A distinct new peak appears at approximately 1040 cm⁻¹, and this peak does not show a significant intensity change in the pure oil spectrum, indicating that the peak originates from methanol molecule vibration. Comparison with the pure methanol spectrum confirms that this peak corresponds to the methanol CO stretching vibration mode. Therefore, the characteristic peak position of methanol in the oil is determined to be approximately 1040 cm⁻¹. -1 This step solves the problem of identifying methanol characteristic peaks in complex oil backgrounds, providing an accurate window range for subsequent residual signal integration.

[0037] Example 3: Spectral background reconstruction and quantitative model establishment of methanol in oils of different concentrations This invention enables the detection of methanol in oil at concentrations down to ppm levels, with some concentrations as low as 50 μL / L. In Example 1, a characteristic spectral space of the oil matrix was constructed, and the principal vibrational mode vector set was obtained. To verify the quantitative detection capability of the method of this invention for methanol in complex oil background environments, methanol standard oil samples with different volume fractions were prepared at concentrations of 50 μL / L, 100 μL / L, 500 μL / L, 1000 μL / L, and 4000 μL / L. Each sample was subjected to Raman spectral acquisition under the same conditions as in Example 2: excitation wavelength of 638 nm, laser power of 30 mW, integration time of 10 s, and scanning range of 600-1800 cm⁻¹. -1The obtained spectrum is as follows Figure 3 As shown.

[0038] from Figure 3 It can be observed that as the methanol concentration increases, at approximately 1040 cm⁻¹... -1 The characteristic peak intensity at this location shows a monotonically increasing trend, while the oil matrix shows an intensity at approximately 1450 cm⁻¹. -1 The morphology of the strong peak of the CH bending vibration at 1040 cm⁻¹ remained basically stable, with only a slight increase in overall intensity. Meanwhile, at 1040 cm⁻¹... -1 The presence of a broad, gently undulating background in the nearby crude oil indicates that the measured spectrum is essentially a superposition of the vibrational signals of the oil matrix and methanol. Therefore, to eliminate interference from the vibrational structure of the oil matrix, this embodiment uses the Raman spectrum of the oil sample to be tested. Perform subspace projection calculations.

[0039] Specifically, Projected onto the characteristic spectral space of the oil matrix, the projection coefficients in each principal vibrational mode direction are calculated according to the following formula:

[0040] Where i = 1, 2, ..., k, This represents the i-th principal vibrational mode function of the oil-based matrix. The projection coefficients... Characterizes the contribution of the measured spectrum to the vibrational structure direction of the corresponding oil matrix.

[0041] After obtaining all projection coefficients, according to the following formula:

[0042] The reconstructed spectrum of the oil matrix background was obtained. This reconstructed spectrum represents the optimal expression of the oil background of the current sample within the oil matrix feature space. Its physical meaning is: to extract the portion of the spectrum to be measured that can be explained by the principal vibrational modes of the oil matrix, thereby realizing the modeling and expression of the vibrational structure of the oil background.

[0043]

[0044] Subsequently, according to the above formula, the original Raman spectrum of the oil sample to be tested is differentially processed with the reconstructed background spectrum of the oil matrix to obtain the residual signal. The residual spectral signal after removing the vibrational background of the oil matrix is ​​characterized. Since the methanol vibrational mode does not belong to the principal vibrational mode space of the oil matrix, it is retained in the residual signal after differencing. This results in the signal at 1040 cm⁻¹. -1 The nearby residual peaks are more prominent, and the background fluctuations are significantly reduced, proving that the interference from the oil matrix has been effectively weakened.

[0045] The characteristic peak range of methanol was determined to be 1025-1055 cm⁻¹.-1 Then, the residual signal is integrated over this interval to obtain the characteristic integral strength of the residual:

[0046] This integral value represents the net area of ​​the characteristic vibrational peaks of methanol, reflecting the correlation between the vibrational intensity of methanol molecules and their concentration. Experimental results show that as the methanol concentration increases, It exhibits a healthy monotonic linear growth trend.

[0047] To further reduce the impact of system background modeling errors on quantitative results, a residual correction term Δ is introduced, and its calculation formula is as follows:

[0048] Where N is the total number of spectral sampling points, the numerator is the L2 norm of the residual signal, and the denominator is the L2 norm of the original spectrum. This correction term reflects the proportion of the residual signal in the overall spectral energy. When the oil matrix modeling accuracy is high, Δ remains stable; when there are background disturbances or modeling errors, Δ will change, thereby compensating for and correcting the quantitative model.

[0049] The residual integral intensity of samples at each concentration was obtained. After adding the residual correction term Δ, with the actual methanol concentration C as the dependent variable, and using... With Δ as the independent variable, establish a multiple linear regression model:

[0050] Where k1, k2, and b are regression fitting parameters. The model parameter values ​​are obtained by performing least-squares fitting on the above concentration gradient samples. The fitting results show that there is a good linear correlation between the predicted concentration and the actual concentration, with a correlation coefficient R0. 2 The result of 0.9 indicates a stable correlation between the residual integral strength and the methanol concentration.

[0051] Compared with the traditional method of directly integrating the original spectrum, this embodiment effectively eliminates the interference of complex oil background vibrations by combining oil matrix feature space modeling, spatial projection reconstruction, residual extraction and correction term compensation, improves the signal-to-noise ratio in the low concentration range, reduces quantitative error, and can still maintain a small prediction deviation at low methanol levels in oil, thus verifying the stability and applicability of the method of the present invention in complex oil background environments.

Claims

1. A Raman quantitative detection method for methanol in complex oil background environments, characterized in that, Includes the following steps: Raman spectral data of multiple pure insulating oil samples were collected, and an oil matrix characteristic spectral matrix was constructed to characterize the background spectral variation of insulating oil under different conditions. The principal vibrational mode decomposition is performed on the characteristic spectral matrix of the oil matrix to extract a set of feature vectors characterizing the main background variation law of the oil matrix, and the characteristic spectral space of the oil matrix is ​​constructed based on the set of feature vectors. Raman spectra of the oil sample to be tested are collected, the Raman spectra of the oil sample to be tested are projected onto the characteristic spectral space of the oil matrix to obtain background projection coefficients, and the background spectrum of the oil matrix is ​​reconstructed based on the projection coefficients. The original Raman spectrum of the oil sample to be tested is compared with the reconstructed oil matrix background spectrum to obtain the residual signal, which is used to characterize the effective Raman response of non-oil matrix components in the oil sample. The residual signal is integrated within the characteristic peak range of methanol to obtain the residual characteristic integral intensity I. res ; Based on the residual characteristic integral intensity I res The parameters are fitted to a pre-established standard curve, and a methanol concentration prediction model is constructed by combining the residual correction term Δ: ； In the formula, C is the predicted methanol concentration in the oil, k1, k2 and b are fitting parameters, Ires is the residual integral intensity within the methanol characteristic peak range, and Δ is the residual correction term.

2. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 1, characterized in that, The method for constructing the characteristic spectral matrix of the oil matrix includes: acquiring n sets of Raman spectra of pure insulating oil respectively. Construct the feature matrix: ,in For Raman displacement variables.

3. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 1, characterized in that, The method for decomposing the principal vibration modes includes: performing eigenvalue decomposition on the feature matrix M to obtain the oil-based principal vibration mode vector set. , where k < n; the vector group constitutes the characteristic spectral space of the oil matrix.

4. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 1, characterized in that, Methods for spatial projection of the Raman spectrum of the oil sample to be tested and reconstructing the background spectrum of the oil matrix include: Raman spectrum of the oil sample to be tested Perform projection calculations: Using background projection coefficient Reconstructed oil matrix background spectrum: .

5. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 1, characterized in that, The residual signal is: ,in Characterize the Raman response of methanol molecules after the oil matrix background is removed.

6. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 1, characterized in that, The residual characteristic integral intensity I res The calculation formula is: ,in and This represents the Raman characteristic peak range of methanol.

7. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 6, characterized in that, The characteristic Raman peak range of the methanol is 1025-1055 cm⁻¹. -1 .

8. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 1, characterized in that, The formula for calculating the residual correction term Δ is: , used to reflect the degree of influence of oil-based background modeling error on the quantitative results of methanol; where N is the total number of spectral sampling points.

9. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 1, characterized in that, The method for establishing the standard curve fitting parameters k1, k2, and b includes: preparing methanol standard oil samples of various known concentrations; acquiring Raman spectra of the methanol standard oil samples respectively, and performing spatial projection and background reconstruction to obtain the residual signal, and calculating the residual integral intensity I. res Calculate the residual correction term Δ; use the residual integral strength I res Using the residual correction term Δ as the independent variable and the actual methanol concentration as the dependent variable, regression fitting is performed to obtain the values ​​of k1, k2, and b.

10. The method for quantitative detection of methanol using Raman spectroscopy in complex oil background environments according to claim 1, characterized in that, The oil matrix characteristic spectral space supports dynamic updates, supplementing new pure oil spectral data according to different oil sample types or operating conditions, and reconstructing the characteristic matrix M and the principal vibration mode vector group, thereby improving the Raman quantitative accuracy in complex oil background environments.

Images