Method for processing data obtained through raman spectroscopy in order to determine the thermal maturity of organic matter in rock

Abstract

The present invention pertains to the field of petroleum, and more specifically to the areas of reservoir modelling, simulation and evaluation, and discloses a method for processing data obtained through Raman spectroscopy in order to determine the thermal maturity of organic matter in rock, based on calibration using vitrinite reflectance standards, using different excitation wavelengths.

Description

[0001] FIELD OF THE INVENTION

[0001] The present invention is within the field of Petroleum, more specifically in the areas of Modeling, Simulation and Evaluation of Reservoirs, and describes a method of processing data obtained by Raman spectroscopy to determine the thermal maturity of organic matter in rock from calibration with vitrinite standards, using different exciting radiations in a Raman spectrometer, dispersive or interferometric. BACKGROUND OF THE INVENTION

[0002] One of the areas of great interest to the petroleum industry is determining the thermal maturity of organic matter in rock. Vitrinite is a type of maceral widely used as an indicator of the thermal maturity of organic matter when analyzed using the percentage reflectance (% Ro) technique. Therefore, studying vitrinite reflectance is a key method for determining the temperature history of sedimentary basins. The alteration of kerogen to hydrocarbons can be measured using this technique by fixing, cutting, and polishing samples to a thickness of 30 micrometers, resulting in high-quality palynofacies sections suitable for experimental reflectance determination.

[0003] Além disso, a espectroscopia Raman tem sido amplamente usada para correlacionar perfis espectrais com dados de reflectância de vitrinita ou betume com consequente estimativa da maturidade térmica da matéria orgânica (Sauerer et al., 2017 “FAST AND ACCURATE SHALE MATURITY DETERMINATION BY RAMAN SPECTROSCOPY MEASUREMENT WITH MINIMAL SAMPLE PREPARATION” International Journal of Coal Geology, v. 173, p. 150-157; Schmidt et al., 2017 “MATURITY ESTIMATION OF PHYTOCLASTS IN STREW MOUNTS BY MICRO-RAMAN SPECTROSCOPY” International Journal of Coal Geology, v. 173, p. 1-8; Henry et al., 2018 “ASSESSING LOW-MATURITY ORGANIC MATTER IN SHALES USING RAMAN SPECTROSCOPY: EFFECTS OF SAMPLE PREPARATION AND OPERATING PROCEDURE” International Journal of Coal Geology, v. 191, p. 135-151; Henry et al., 2019a “A RAPID METHOD FOR DETERMINING ORGANIC MATTER MATURITY USING RAMAN SPECTROSCOPY: APPLICATION TO CARBONIFEROUS ORGANIC-RICH MUDSTONES AND COALS” International Journal of Coal Geology, v. 203, p.87-98; Henry et al., 2019b “RAMAN SPECTROSCOPY AS A TOOL TO DETERMINE THE THERMAL MATURITY OF ORGANIC MATTER: APPLICATION TO SEDIMENTARY, METAMORPHIC AND STRUCTURAL GEOLOGY” Earth-Science Reviews, v. 198, p. 102936; Khatibi et al., 2018 “RAMAN SPECTROSCOPY: AN ANALYTICAL TOOL FOR EVALUATING ORGANIC MATTER” J Oil Gas Petrochem Sci, v. 1, n. 1, p. 28-33; Schito & Corrado, 2020 “AN AUTOMATIC APPROACH FOR CHARACTERIZATION OF THE THERMAL MATURITY OF DISPERSED ORGANIC MATTER RAMAN SPECTRA AT LOW DIAGENETIC STAGES” Geological Society, London, Special Publications, v. 484, n. 1, p. 107-119; Wilkins et al., 2018 ”THERMAL MATURITY EVALUATION FROM INERTINITES BY RAMAN SPECTROSCOPY: THE ‘RAMM’ TECHNIQUE” International Journal of Coal Geology, v. 128, p. 143-152).Compared to traditional techniques for determining vitrinite or bitumen reflectance, Raman spectroscopy is a rapid, non-destructive technique that can be used to complement traditional methods or employed independently, offering a means of screening samples prior to more expensive and destructive analysis.

[0004] The parameters obtained in Raman spectra typically used to estimate reflectance are based on the area and position of the D and G bands, which characterize organic matter. However, there are several difficulties related to data acquisition, such as choosing the excitation line to use, sample fluorescence, laser type and power, among others. In previous studies (e.g., Henry et al., 2019b; Schmidt et al., 2020), factors related to the processing and mathematical treatment of spectral data, such as noise smoothing filters, baseline calculations, assignment, and deconvolutions, so that the result can be correlated with the spectra of vitrinite or bitumen, were performed manually and without standardization.Although this method is extremely powerful and has become the standard in organic matter maturation studies, it is a manual, slow, subjective, costly, and error-prone procedure that requires highly specialized training. Therefore, the search for an alternative that requires less analysis time and generates an equivalent reflectance value to be used as a reference in prospecting is desirable, as it can reduce the resource consumption in this process.

[0005] In this sense, the present invention allows the use of a methodology to process the Raman spectral result obtained from different excitation lines, from the near-infrared to the ultraviolet region, and generate an equivalent vitrinite reflectance value in a few seconds through software calibration and training. This solution can be applied to assess the maturity of organic matter present in rock samples from oil exploration wells. Since the solution does not require sample pretreatment, it can be applied directly in exploration areas with the aid of portable equipment.The benefits of this solution, compared to the study of vitrinite reflectance, are greater speed in obtaining results, greater wealth of information in each sample and greater standardization of the procedure, preventing subjectivity and susceptibility to operator errors from leading to different results for the same sample depending on the complexity of the analysis.

[0006] The development of the methodology includes the implementation of algorithms for processing spectroscopic data, eliminating manual and repetitive procedures for an analyst and standardizing a set of approaches for data processing and analysis. Thus, the main objective is to ensure the reliability of the results by training these algorithms so that the established and standardized mathematical procedures for processing Raman spectra reduce analysis time, subjectivity, and potential human error. STATE OF THE ART

[0007] The following paragraphs will describe seven (7) documents that describe the state of the art, in terms of what was proposed and the limitations of each proposal. After the description, the differentiation of the present invention proposal in relation to this prior art will be discussed.

[0008] Document US11029250B2 describes a method that takes into account the separation value of the G and D1 band for Raman spectra to estimate the vitrinite reflectance through the equation RBS = c1 ln(x) + c2, where x is the vitrinite reflectance and c1 and c2 are constants obtained for each equipment.

[0009] Document US2023118696A1 aims to obtain Raman spectra of organic matter in geological rock formation samples to determine kerogen properties. The obtained Raman spectra are quantitatively correlated with thermal maturation through the G- and D-band Raman parameters and the separation of G- and D-bands. It uses a generic description of rocks encompassing kerogen types 1, 2, and 3, asphaltene, and bitumen. It should be noted that the analysis of petroleum samples containing asphaltene requires a prior separation step, such as gas chromatography, and cannot be performed directly on rock samples, as mentioned in the document. Furthermore, they used a computational model but did not calibrate and validate it with vitrinite, which, it is worth noting, represents a more reliable indicator of organic matter maturity compared to other carbonized substances.

[0010] The document, presented by DeHan et al., titled “SAMPLE MATURATION CALCULATED USING RAMAN SPECTROSCOPIC PARAMETERS FOR SOLID ORGANICS: METHODOLOGY AND GEOLOGICAL APPLICATIONS,” refers to a methodology and geological applications for calculating the maturation of solid organic samples using Raman spectroscopic parameters. The carbonized substances (such as vitrinite, semi-vitrinite, semi-fusinite, fusinite, or inertinite in maceral coal) discussed in this document may present different polycondensations of the aromatic ring, leading to notable differences in the Raman spectral profile and reflectance values. Among these carbonized substances, vitrinite is the most reliable indicator of organic matter maturity (ASTM D7708-23 and ASTM D7708-14).

[0011] Therefore, it is extremely important that the analysis focuses specifically on vitrinite to infer the maturity of the organic matter. Unlike DeHan et al., who obtained Raman spectra of different carbonized substances with a single 532 nm laser line, they related only two Raman parameters: the separation between the G and D bands and the ratio of the height of the D to G Raman band to the reflectance. Furthermore, the aforementioned document does not perform data processing, such as baseline correction, smoothing, band deconvolution, or intensity normalization.

[0012] The paper, presented by Henry et al., entitled "A Rapid Method for Determining Organic Matter Maturity Using Raman Spectroscope: Application to Carboniferous Organic-Rich Mudstones and Coals," estimates vitrinite reflectance using parameters obtained by Raman spectroscopy. It uses standard %Ro values ​​obtained by optical microscopy to relate them to those estimated by Raman spectroscopy and apply this method to other wells with coal and shale samples. The Raman data were obtained using only laser excitation at 514.5 nm. Several variables were obtained from the Raman spectra, such as the half-width at half-height (FWHM) of the G and D bands, the separation between these two bands (RBS), the band ratio, R1 for intensities, and SSA for area. Through these parameters, curves were obtained to estimate %Ro with each of them individually, with the FWHM of the G band inferred as the best parameter.It does not estimate the values ​​of the spectral parameters after performing band deconvolution, allowing questions about the reproducibility of the methodology.

[0013] The paper, presented by Henry et al., titled "RAMAN SPECTROSCOPY AS A TOOL TO DETERMINE THE THERMAL MATURITY OF ORGANIC MATTER: APPLICATION TO SEDIMENTARY, METAMORPHIC, AND STRUCTURAL GEOLOGY," is a review of several studies in the literature that use Raman spectroscopy of standard vitrinite samples to obtain spectral parameters that allow estimating % Ro and determining the thermal maturity of organic matter. The paper cites the need to standardize the band nomenclature and the Raman parameters obtained, but the best correlations were obtained for % Ro ranges above 3.0, indicating that studies typically use these calibration methods for highly mature organic matter.

[0014] The paper, presented by Sauerer et al., titled "FAST AND ACCURATE SHALE MATURITY DETERMINATION BY RAMAN SPECTROSCOPY MEASUREMENT WITH MINIMAL SAMPLE PREPARATION," analyzes sedimentary rock samples containing organic matter classified as type II kerogen. The Raman spectral parameters were obtained by deconvolution and emphasize that RBS is the only parameter with an adequate correlation with % Ro. They do not calibrate with % Ro in accordance with the standard.

[0015] The paper by Wilkins et al., entitled "Thermal maturity evaluation from inertinites by Raman spectroscopy: The 'RaMM' technique," describes the possibility of determining the % Ro equivalent from inertinite and vitrinite samples. The reflectance range studied was 0.4 to 1.2 % Ro. Raman spectra were collected using a 488 nm excitation line, and spectral parameters were correlated with the % Ro of the vitrinite and inertinite macerals from deconvolution using only two bands. However, they do not use samples calibrated to the standard. The authors themselves suggest performing analyses on calibrated samples to make the method more reliable.

[0016] The previous paragraphs 008 to 015 show that the prior art uses Raman spectroscopy to characterize organic matter in rocks using a conventional method, with non-standardized data processing that differs in each of the cited documents, subject to the analyst's subjective approach. However, none of them demonstrates the possibility of establishing a procedure for processing Raman spectral data collected with different excitation lines, involving the use of machine learning to train the % Ro prediction based on calibration with standard vitrinite samples analyzed according to ASTM D7708-23 and ASTM D7708-14. However, it is also noted that some of the studies use matrices other than petroleum, such as coal. Therefore, a comprehensive calibration and training development would be required.Although the principle is similar for both, organic matter is at very different stages of maturation when comparing oil and coal.

[0017] Although both the present invention and the previous documents discussed in the paragraphs above, from 008 to 015, involve the use of Raman spectroscopic measurements, there are significant distinctions that make the method proposed in the present invention application innovative. Initially, in the present invention application, the Raman spectra of vitrinite were obtained on samples in the form of 30-micrometer-thick palynofacies slides, produced by an industry-known procedure, containing previously selected fragments and with reflectance values ​​measured by ASTM D7708-23 and ASTM D7708-14 in the range of 0.46 to 2.72, which were used for distinct method calibrations for different excitation wavelengths to obtain the Raman spectra. This allowed the prediction of vitrinite reflectance values ​​by Raman spectroscopy to be validated against values ​​obtained by an internationally recommended method.

[0018] The calibration and subsequent estimation of vitrinite reflectance from the Raman spectra of these standard samples were developed so that the collection of Raman spectra of organic matter present in rocks, performed with excitation lines of different wavelengths, allows estimating the maturity of the organic matter in the analyzed rock, after standardized data processing. To this end, the spectroscopic data were processed by baseline adjustments and deconvolution using six bands in the wavenumber range between 1100 and 1750 cm -1 were standardized for each excitation wavelength.

[0019] The methodology for developing a trained algorithm for data processing by deconvolution involved managing the data by imposing limits on band intensities and widths to standardize and implement the analyses. A mathematical prediction method was used through linear fitting, weighting the Raman spectral fit parameters, such as band center, intensity, width at half height (FWHM), and the exponential and logarithmic functions of all these variables, to establish a correlation with the reflectance values ​​of the vitrinite samples.

[0020] Therefore, a total of 72 variables were used to train the vitrinite reflectance prediction using machine learning. Subsequently, the LASSO (Least Absolute Shrinkage and Selection Operator) regression analysis method was used, which removed contributions that were not useful in improving the correlation, retaining only the variables that had the highest correlation with the % Ro value.

[0021] The variable with the greatest weight was the G-band FWHM; however, other variables also had significant weight, such as the exponential and log D1, among others. The model prevents different analysts from using different methods to process the data and obtaining different values, thus reducing the possibility of errors in the analysis caused by user subjectivity.

[0022] Thus, a robust and comprehensive relationship was obtained involving the spectral parameters of vitrinite and the maturity of organic matter than that reported in the aforementioned document.

[0023] Therefore, the present invention represents an innovative contribution to the use of vitrinite Raman spectra in the calibration for determining the maturity of organic matter in rock that goes far beyond those presented in the different previous documents that were discussed in paragraphs 008 to 015.

[0024] Furthermore, it is noteworthy that the use of Raman spectroscopy in the method proposed in this invention allows the direct use of the source rock, without prior treatment, except for fractionation into sizes appropriate for the irradiation site, or even petrographic or palynofacies slides. The invention developed for the use of Raman spectroscopy also allows data to be obtained using different excitation laser lines based on appropriate calibration, which allows the data to be extended to other types of macerals and carbon-based compounds. Finally, the data can be processed in a computational application that will allow any user to use it. Such an application can be implemented with different computational approaches, such as selecting excitation lines, baseline matching methods, band deconvolution, spectral regions of interest, etc.

[0025] Therefore, the details described in the proposed method, which involves the processing of spectral data, make the process robust, as it considers all the variables that are relevant to the prediction of vitrinite reflectance, but subject to parameterization, resulting in an objective analysis that is little subject to user subjectivity. SUMMARY OF THE INVENTION

[0026] The present invention aims to develop a method for processing Raman spectral results and generating an equivalent vitrinite reflectance value in a few seconds. This solution can be applied to assess the maturity of organic matter present in rock samples from oil exploration wells. Since the solution does not require sample pretreatment, it can be applied directly to exploration areas using portable equipment. The benefits of this solution, compared to vitrinite reflectance analysis, include faster results, greater information richness in each sample, and greater standardization of the procedure. BRIEF DESCRIPTION OF THE FIGURES

[0027] Figure 1 shows vitrinite reflectance values ​​that encompass the hydrocarbon generation range of interest to the petroleum industry, which includes the geological processes of diagenesis, catagenesis, and metagenesis. Element A, highlighted, shows the range encompassing the established calibration curve.

[0028] Figure 2 shows Raman spectra obtained with excitation lines of wavelength at 514.5, 532, 632.8 and 785 nm of the organic matter present in the same palynofacies layer, emphasizing the change in the band profile.

[0029] Figure 3 shows the analysis range, in wavenumbers, that allows the characterization of organic matter.

[0030] Figure 4 shows the application of the spectrum smoothing method so that the presence of noise arising from the initial analysis is minimized and does not interfere with the evaluation of the data.

[0031] Figure 5 shows how the two-point baseline adjustment method is applied so as not to take into account the background radiation of different origins that may be present in the Raman spectra.

[0032] Figure 6 shows the normalization of intensities by the most intense band so that relative intensities are the focus of the analysis and the calculated areas are comparable between different spectra.

[0033] Figure 7 shows the deconvolution of a Raman spectrum obtained with an excitation line at 514.5 nm from a palynofacies sheet in six bands, the sum of which perfectly reproduces the original profile.

[0034] Figure 8 shows a flowchart indicating the parameters obtained from the Raman spectra used in the model to predict the percentage of vitrinite reflectance (%Ro): for each band obtained in the deconvolution, the position of the band center, the band width at half-height (FWHM), and the height of each of them are used as variables; other variables are obtained by applying the logarithm and exponential functions to each of the initial variables; this set of variables is also the same used in the processing of the calibration data of the vitrinite standard samples, whose reflectance was previously measured and correlated with the thermal maturity of the organic matter; finally, a machine learning technique allows the selection and weighting of each variable to predict the thermal maturity of the problem sample based on the variables that most contribute to the final analysis. DETAILED DESCRIPTION OF THE INVENTION

[0035] The present invention relates to a method for processing data obtained by Raman spectroscopy to determine the thermal maturity of organic matter in rock from calibration with vitrinite standards, using different exciting radiations. The method comprises the following steps: (a) Selection of the wavenumber and collection of Raman spectra of standard vitrinite samples with known reflectance values; (b) Smoothing of the Raman spectra; (c) Correction / Baseline Subtraction of the spectra; (d) Normalization of the relative intensity of the Raman spectra; (e) Deconvolution of the Raman spectra to obtain the parameters for the model; (f) Calibration of the vitrinite reflectance prediction model; and (g) Collection of Raman spectra of organic matter present in rock and prediction of the equivalent reflectance values ​​from the previous calibration.

[0036] To begin calibration, samples containing reflectance data produced by another technique must be used. To perform the analyses, the Raman scattering excitation laser must be positioned in the region of the organic matter. Raman spectra are obtained from different points to ensure they accurately represent the material's diversity. A separate Raman spectrum is generated for each laser incidence.

[0037] The calibration can be extended to different Raman excitation lines, using the same standard vitrinite samples, which will allow encompassing problem samples that present better results in Raman spectroscopy in excitation lines in the ultraviolet, visible or near-infrared region.

[0038] The Raman spectrum is recorded in a text file in x, y format containing two columns (wavenumber and Raman intensity) and up to thousands of lines with values ​​for each variable in the column. The files are manually selected through the software's graphical interface and loaded into memory. The wavenumber segment between 1100 cm -1 and 1750 cm -1 is selected in each spectrum. The Savitzky-Golay filter is applied to the intensity of each spectrum, ordered by wavenumber, with a window size of 21 and a polynomial degree of 2. The baseline correction for the intensities of each spectrum is calculated. Each spectrum is normalized. Deconvolution is performed using a parameter estimation procedure using a composition model of Voigt and Gaussian functions to represent the G and D curves (D1, D2, D3, D4, and D5).

[0039] The estimated parameters are the centers, the FWHM, and the intensities of each curve. Combinations of these estimated parameters and applications of nonlinear functions are performed to generate new descriptors. Specifically, the width at half-height of each band and the function representing an exponential decay are applied to the parameters in this step. Finally, using these descriptor parameters and the reflectance data, the LASSO (Least Absolute Shrinkage and Selection Operator) regression technique is manually applied.

[0040] As a result, a mathematical model is generated, with linear terms in the parameters and variables. The model is manually recorded in a file, containing the parameter values ​​in the original sequence presented by the descriptors. To begin prediction, it is necessary to use rock samples containing organic matter not yet characterized by reflectance analysis.

[0041] These samples undergo the same spectral generation and mathematical procedures described above, except for the final step, which involves applying the LASSO technique. In this case, the file containing the mathematical model's parameter values ​​must be selected and loaded into memory. Finally, the expanded sets of descriptor parameters from the spectra of uncharacterized samples are used as inputs to the mathematical model to perform the vitrinite reflectance prediction calculation.

[0042] It is worth mentioning that the treatments of the Raman spectra were performed optimally and can be transferred to spectra obtained by any Raman spectrometer and in any exciting radiation used.

[0043] Furthermore, deconvolution generated a large number of variables that are actively evaluated by the LASSO regression method to obtain the variables with greatest relevance for determining % Ro.

[0044] The method is transferable to different samples, without the need for palynofacies slides or any special treatment, except for fractionation for introduction into the sampling compartment of the Raman equipment. Example of the Invention

[0045] To construct the calibration curve using the Raman parameters measured according to the known % Ro values, 9 palynofacies slides were used (with % Ro values ​​of 0.46, 0.64, 0.70, 0.76, 0.86, 0.85, 2.07, 2.61 and 2.72, respectively). The 9 slides mentioned have vitrinite reflectance values ​​that encompass the hydrocarbon generation range of interest to the petroleum industry, that is, between 0.46 and 2.72%, which covers the geological processes of diagenesis, catagenesis and metagenesis, as illustrated in Figure 1. Each slide studied has around 10-20 fragments of organic matter classified as phytoclasts, for which Raman spectra were obtained, and each of the slides has a specific % Ro value that was assigned by organic petrography according to standard D7708-23.In the acquisition of Raman spectra, the 9 slides were analyzed in the Horiba Scientific LabRAM HR Evolution equipment, with an excitation line with a wavelength of 514.5 nm, using a 50×, 3.2%, 600 (500nm) objective lens, with an accumulation time of 45s, in the range of 100 to 1900 cm. -1 .

[0046] The spectra obtained in the excitation line with a wavelength of 632.8 nm were obtained in a Bruker dispersive spectrometer, model SENTERRA, with a spectral resolution of 3-5 cm -1 and 50× ULWD magnification objective lens (NA = 0.51) and 50 μm confocal aperture.

[0047] Mathematical treatments were then performed on specific regions of the spectra of the palynofacies slides (1750-1100 cm -1 ), through and using noise correction filters, baseline correction and normalization, to obtain the Raman parameters (Bands D1, D3, D4, D5, D6 and G) using the Origin 9.0 and FityK 1.3.1 software.

[0048] A calibration curve was also constructed using exponential fits, considering the G-FWHM parameter, which showed the best correlation with % Ro, for the two different excitation lines, 514.5 nm and 632.8 nm. Both showed a good correlation between the measured and predicted values. These parameters were used as initial input data for method calibration.

[0049] This methodology, developed to obtain % Ro values ​​from Raman spectra, was designed to establish the best routine for data processing, standardization, and parameterization. The LASSO (Least Absolute Shrinkage and Selection Operator) regression analysis method was used for calibration because it has the advantage of reducing the number of model input variables, retaining only those most relevant to prediction. These variables are the estimated parameters calculated from the spectral curves, as well as linear and nonlinear applications and combinations of these parameters.

[0050] Although there is only estimation for the center, sigma, and amplitude parameters, it is possible to calculate several other parameters of the function, such as width at half height (FWHM), height, and area under the curve. As an example, Equation 1 presents the Gaussian function and its main parameters, ^ ^ ^ ^^^ ^^ ^ ^ ^ ^^ ^^^^

[0051] Where x is the wavenumber of the spectrum and A, ^, ^, FWHM and h are the amplitude, center, sigma, width at half height and height of the curve, respectively. Table 1: Reference used to estimate the parameters of the curves. Curves G D1 D2 D3 D4 D5 Voigt function Gaussian Gaussian Gaussian Gaussian Gaussian Centers 1596 1350 1557 1468 1249 1177 Sigmas 20 20 30 30 30 30 Note: Ik is the intensity corresponding to the center of the curve. Table 2: Search interval used to estimate the curve parameters. Curves G D1 D2 D3 D4 D5 1586- 1340- 1547- 1458- 1239- 1167- Centers 1606 1360 1567 1478 1259 1187 Sigmas 10-50 10-50 20-50 20-50 20-50 20-50 Note: Ik is the intensity corresponding to the center of the curve.

[0052] The presented scheme refers to the generation of a vitrinite reflectance prediction model or the calculation of its prediction. The algorithm used, as previously reported, was LASSO (Least Absolute Shrinkage and Selection Operator), which has the advantage of reducing the number of model input variables, keeping only those most relevant to the prediction. This algorithm has an input parameter that strikes a balance between the number of non-zero parameters (the fewer, the better) and prediction performance (the higher, the better). The value of this parameter was set at 0.05. The input variables for this model are the estimated and calculated parameters from the spectral curves, as well as linear and nonlinear applications and combinations of these parameters, as shown in Equation 2 below. ^^^ = ^ !⁄ ^"^$% = ^ ! − ^"

[0053] Thus, the algorithm will use dozens of input variables and one output variable (vitrinite reflectance value) in each spectrum-% Ro pair to perform the fit (calibration). It is worth mentioning that a certain number of points (spectrum-% Ro pairs) are required for the fit to obtain statistical significance. At the end of this fit, the algorithm provides a calibration model, which is ready to be used to predict % Ro from new spectra. This model contains a reduced set of non-zero parameters (less than 10, for example), which correspond to the input variables that contribute most to the % Ro prediction. Equation 3 presents an example of the generated prediction model, ^21 = 30 + 31 ∙ exp^− ^^^^"⁄ 5 ^ + 32 ∙ log 1 + ℎ !

[0054] Where ^ 21 represents the prediction of vitrinite reflectance (% Ro) and p0, p1, p2 and p3 are the parameters of this prediction model, while FWHMG, hD1 and ^D2 are the input variables of the model, which are obtained by processing the spectrum from which we want to know the maturity of the organic matter.

[0055] As mentioned above, this step also refers to using the model to obtain the % Ro prediction from a Raman spectrum, which must be processed by the same steps as the scheme shown in Figure 8.

[0056] The method can be easily installed on computers using an installer file (.exe) and a wizard, in which the installation location must be selected. Furthermore, it can be associated with any Raman device, different types of lasers, and can be used for calibration or direct estimation of vitrinite or bitumen reflectance.

[0057] Nevertheless, the method performs all steps of the spectral analysis in a few seconds. It reads files containing Raman spectra (".txt" and ".dpt"), which are easily viewed in text editors and may or may not have a header and two or more variables, the first representing the wavenumbers and the others representing the Raman intensities. Raman spectra of organic matter in rock can be recorded with the excitation lines of wavelengths 514.5 or 632.8 nm, mentioned in this example, or any other excitation lines for Raman scattering.

[0058] The method recognizes the indicated range that contains organic matter, which comprises wavenumbers between 1100 cm -1 and 1750 cm -1and can be interpreted as input parameters for this algorithm. It then applies the mathematical parameterization treatments: filtering, baseline correction, and band deconvolution, according to what was previously established during training with standard samples. This approach avoids human subjectivity and standardizes all mathematical treatments applied to the input data.

[0059] Furthermore, the method utilizes a greater number of variables in its mathematical processing, providing a novelty and demonstrating the development of something impossible to reproduce manually, as was achieved in previous studies. The results are presented in graph form for each processed data point, containing all files for a specific measurement selected for analysis. This allows you to view the analysis, deconvolutions, band assignments, and the final % Ro value for the organic matter present in the sample. Training these algorithms significantly reduced analysis time, resulting in greater accuracy and standardization.

[0060] Some general aspects of the method proposed in the present invention can be observed, which as a whole confer its novel and inventive character: • Collection of Raman spectra of organic matter present in rock samples in the form of petrographic slides and in standard samples containing vitrinite fragments, previously selected and with reflectance values ​​measured by ASTM D7708-23 and ASTM D7708-14 standards in the range of 0.46 to 2.72, which allows calibration of the method through values ​​with a high degree of reliability; • Raman spectra recorded using different excitation lines (Figure 1), which makes the method flexible and adaptable to different needs of characterization of organic matter; • Treatment of spectroscopic data by smoothing, baseline adjustments and deconvolution using six bands in the wavenumber range between 1100 and 1750 cm -1, ensuring robustness and ease of transfer to the results obtained; • Systematization methodology in the acquisition and processing of data obtained by deconvolution with management, by imposing limits on intensities and bandwidths, for standardization and implementation of a predictive model for determining reflectance (Figure 2); • Use of a mathematical prediction method by linear adjustment with parameter weighting to obtain the vitrinite reflectance value using the Raman spectral adjustment parameters, such as band center, band intensity, band area, half-width at half-height (FWHM), in addition to the exponential and log functions of all these variables. A total of 72 variables were used to train the prediction of vitrinite reflectance through the machine learning technique.Next, the LASSO (Least Absolute Shrinkage and Selection Operator) regression analysis method was used, which removes unnecessary contributions, keeping only variables that have some correlation with the % Ro value. It was found that the variable with the greatest weight is the FWHM of the G band; however, other variables also had a significant weight, such as the exponential and the log of D1, among others.The model works with multiple inputs, generating a single output, which is the inference of the % Ro value of problem samples (Figure 3); • Implementation of different calibrations, using vitrinite standard samples, which can be performed for any type of sample that presents a signal in Raman spectroscopy and with any Raman equipment, with excitation lines that can be in the ultraviolet, visible or near-infrared region, as long as the vitrinite standard samples are used for the initial calibration performed by machine learning. .

[0061] The method proposed in this invention has direct application in the study of the thermal maturity of organic matter. Currently, this procedure is performed destructively on samples and requires the preparation of specific slides for this estimation, as well as a highly trained professional. Therefore, because it is a rapid, non-destructive procedure, the present invention has an immediate impact. Furthermore, it can be used on any type of sample that presents a signal in Raman spectroscopy and with any Raman equipment, as long as vitrinite standard samples are used for the initial calibration using machine learning.

[0062] One of the areas of great interest in the petroleum industry is determining the thermal maturity of organic matter in rock. Vitrinite is a type of maceral widely used as an indicator of the thermal maturity of organic matter when analyzed using the percentage reflectance (% Ro) technique. Raman spectroscopy has been widely used to estimate the reflectance of vitrinite or bitumen, consequently determining the thermal maturity of organic matter.

[0063] Compared to traditional vitrinite or bitumen reflectance determination techniques, Raman spectroscopy is a fast, nondestructive technique that can be used to complement traditional methods or independently, offering a means of screening samples prior to more expensive and destructive analyses. In previous studies (e.g., Henry et al., 2019b; Schmidt et al., 2020), factors related to the processing and mathematical treatment of spectral data, such as noise smoothing filters, baseline calculations, assignment, and deconvolutions, so that the results can be correlated with vitrinite or bitumen spectra, are performed manually and without standardization.

[0064] As these procedures directly affect the reliability and reproducibility of determining the thermal maturity of organic matter more accurately, the aim of this patent was to establish an innovative method for treating Raman spectra.

Claims

CLAIMS 1. A method for processing data obtained by Raman spectroscopy to determine the thermal maturity of organic matter in rock, comprising the following steps: (a) Selection of the wavenumber and collection of Raman spectra of standard vitrinite samples with known reflectance values; (b) Smoothing of the Raman spectra; (c) Correction / baseline subtraction of the spectra; (d) Normalization of the relative intensity of the Raman spectra; (e) Deconvolution of the Raman spectra to obtain the parameters for the model; (f) Calibration of the vitrinite reflectance prediction model; and (g) Collection of Raman spectra of organic matter present in rock and prediction of the equivalent reflectance values ​​from the previous calibration.

2. A method according to claim 1, characterized in that the wavenumber stretch between 1100 cm -1 and 1750 cm -1is selected in each spectrum.

3. Method, according to claim 1, characterized by the fact that the Savitzky-Golay filter is applied to the intensity of each spectrum, ordered by wavenumber, with window size equal to 21 and polynomial degree equal to 2.

4. Method, according to claim 1, characterized by the fact that combinations are made between the following parameters: the centers, the FWHM and the intensities of each curve.

5. Method according to claim 1, characterized in that, having these descriptor parameters and the reflectance data, the LASSO (Least Absolute Shrinkage and Selection Operator) regression technique is manually applied.

6. Method according to claim 1, characterized in that the increased sets of descriptor parameters of the spectra of samples not yet characterized are used as inputs to the mathematical model for calculating the prediction of vitrinite reflectance.

7. Method according to claim 1, characterized in that the calibration is extended to different Raman excitation lines, using the same standard vitrinite samples, in excitation lines in the ultraviolet, visible or near infrared region.

Images