METHOD FOR OBTAINING A MACHINE LEARNING MODEL FOR ESTIMATING TOTAL ORGANIC CARBON (TOC) FROM HYPERSPECTRAL ROCK SAMPLE DATA

Description

1 / 21 METHOD FOR OBTAINING A MACHINE LEARNING MODEL FOR ESTIMATING TOTAL ORGANIC CARBON (TOC) FROM DATA Hyperspectral images of rock samples FIELD OF THE INVENTION

[001] The present invention falls within the field of geochemistry. More specifically, the present invention relates to methods for estimating the Total Organic Carbon content of rock samples in an automated manner using Artificial Neural Networks. FUNDAMENTALS OF THE INVENTION

[002] The determination of Total Organic Carbon (TOC) in source rocks is a crucial analysis for the characterization of source rocks, often being the first step performed to assess the potential of these rocks. A practice that occurs in some geochemical analysis laboratories consists of the responsible professional carrying out a pre-selection of samples of interest for detailed characterization, basing this choice on the type of lithology of the rock or other characteristics. However, this process has its disadvantages because the subjectivity inherent in each professional during the pre-classification process can result in varied interpretations of the selection criteria.

[003] Prior to the proposed invention, the initial characterization of potentially carbon-generating rocks based on TOC determination depended on performing laboratory analyses on all samples. This analytical step is carried out by burning the previously ground sample in a carbon analyzer with an infrared detector, the result of which is obtained, for each sample, on average from 30 to 240 seconds, depending on the equipment used. Furthermore, the analysis must be performed by professionals qualified for this function, which requires initial training in sample preparation and equipment use.

[004] Aiming to reduce analysis time and costs Petition 870250016331, dated 27 / 02 / 2025, page 44 / 72 2 / 21 involved in the process, it is common for samples to be pre-selected according to their lithology or other characteristics of potential interest, limiting measurements to a smaller set of samples. However, this pre-classification can lead to varied selection criteria, as the process is influenced by the subjectivity of the professional performing this function. Furthermore, an incorrect selection could directly impact the results obtained, leading to inaccurate conclusions or even erroneous interpretations of the analyzed sample set.

[005] The state of the art of inferring TOC from reflectance spectroscopy, focusing on the shortwave infrared (SWIR) region, is covered by the scientific works of Mehmani et al. (Multiscale characterization of spatial heterogeneity of petroleum source rocks via near-infrared spectroscopy. In: UNCONVENTIONAL RESOURCES TECHNOLOGY CONFERENCE, AUSTIN, TEXAS, 24-26 JULY 2017, 2017. Proceedings. [Sl: sn], 2017. p. 2539-2543.), Rivard et al. (Inferring TOC and major element geochemical and mineralogical characteristics of shale core from hyperspectral imagery. AAPG Bulletin, [Sl], n. 20,180,604, 2018.) and Alnahwi et al. (High-resolution hyperspectral-based continuous mineralogical and total organic carbon analysis of the eagle ford group and associated formations in south texas. AAPG Bulletin, [Sl], v. 104, n. 7, p. 1439-1462, 2020).

[006] In Mehmani et al. (2017) the TOC content was modeled from the intensity of the absorption feature near 1700 nm, related to organic matter. However, the model proposed by the authors was not robust enough to be applicable to other locations with minimal need for recalibration.

[007] The study by Rivard et al. (2018) proposed an approach that used more than one spectral variable to Petition 870250016331, dated 27 / 02 / 2025, page 45 / 72 3 / 21 predict the COT, these values ​​being derived from the wavelet decomposition of the reflectance curves at wavelengths of 2174, 2236, and 2392 nm. Using multiple linear regression, the authors proposed a model whose performance for COT values ​​below 10% resulted in a coefficient of determination (R2) of 0.61.

[008] Finally, the method proposed by Alnahwi et al. (2020) explores a first stage of classifying rocks with similar minerals (or mineral associations) using a type of unsupervised artificial neural network called a self-organizing map (SOM). After this process, TOC data were used to calibrate multivariate linear regression models for each established class.

[009] However, these studies were limited to exploring only traditional data correlation techniques, such as linear regression, in modeling the aforementioned parameters. STATE OF THE ART

[010] Document US2022290553A1, entitled “REAL-TIME MULTIMODAL RADIOMETRY FOR SUBSURFACE CHARACTERIZATION DURING HIGH-POWER LASER OPERATIONS”, discloses a method that includes: irradiating a target surface with a process beam during a drilling process; in response to irradiation with the process beam, receiving a signal beam containing scattered light from the target surface as well as light radiating from the target surface; splitting the signal beam into a first portion in a polarization arm and a second portion in a non-polarization arm; performing, in the polarization arm, a first plurality of polarization-dependent intensity and spectrum measurements of the first portion; performing, in the non-polarization arm, a second plurality of intensity and spectrum measurements of the second portion; and based on the application of one or more machine learning techniques to at least portions of (i) the first plurality of measurements Petition 870250016331, dated 27 / 02 / 2025, page 46 / 72 4 / 21 of polarization-dependent intensity and spectrum and (ii) the second plurality of intensity and spectrum measurements, determine a classification of the target surface.

[011] Document CN116927771A, entitled “Method, device, equipment and medium for predicting total organic carbon data of shale reservoir”, discloses a method, device, equipment and means for predicting total organic carbon data of a shale reservoir. The method comprises the following steps: acquiring initial log data, actually measured total organic carbon data and sedimentary facies data; performing depth calibration on the log curve to obtain lithological features corresponding to the initial log data; based on the sampling depth of the total organic carbon data, acquiring log data corresponding to each part of the total organic carbon data; combining the lithological features with the log data to obtain amended log data;Standardize the amended log data and split the standardized amended log data and the corresponding total organic carbon data into training data and test data; and establish a bidirectional long-short-term memory convolutional neural network model based on the training data, test the model based on the test data to obtain a shale reservoir total organic carbon data prediction model, and predict total organic carbon data corresponding to the log data to be predicted based on the shale reservoir total organic carbon data prediction model.

[012] The document High-resolution hyperspectral-based continuous mineralogical and total organic carbon analysis of the Eagle Ford Group and associated formations in south Texas (Alnahwi et al, 2020) reveals three different cameras within the Petition 870250016331, dated 27 / 02 / 2025, page 47 / 72 5 / 21 hyperspectral core imaging systems were used to image a 99.1 m (325 ft) core: a (1) line-scanning camera, which produces a high-resolution natural-color red-green-blue (120 mm) photograph of the dry core from the visible light spectrum; (2) shortwave infrared spectrometer (SWIR) (300–500 mm resolution); and (3) longwave infrared spectrometer (LWIR) (300–500 mm resolution). A self-organizing map neural network was used to classify the samples, and conventionally obtained TOC data were used to train and calibrate the neural network. SUMMARY OF THE INVENTION

[013] The present invention discloses a method for estimating the total organic carbon (TOC) of rock samples automatically and without the use of destructive techniques, comprising the steps of: (1) Collecting rock samples; (2) Collecting at least one hyperspectral data point from each of the rock samples collected in step (1); (2.1) Spectral smoothing of the spectral data collected in step (2); optionally (2.2) Selection of the spectral band of interest; (2.3) Highlighting the absorption and removal features of the spectral data continuum; optionally (2.4) Calculation of the average of the spectral signatures per sample; (2.5) Selection of input variables for machine learning models; (3) Laboratory acquisition of TOC data from rock samples collected in step (1); (4) Association of hyperspectral data with the respective rock samples and with the TOC data obtained in step (3), obtaining a dataset; (5) Division of the dataset into training data and test data; (6) Normalization of the training data; (7) Choice of one or more machine learning algorithms and definition of their hyperparameters; (8) Adjustment of the hyperparameters to obtain a machine learning model. Petition 870250016331, dated 27 / 02 / 2025, page 48 / 72 6 / 21 machine for each of the tested algorithms; (9) Training and error estimation of the machine learning models obtained in step (8); (10) Normalization of the test data; (11) Testing of the machine learning models trained in step (9) using the test dataset; (12) Comparison between the results of the tests performed in step (11); and (13) Selecting the machine learning model with the best result based on step (12). BRIEF DESCRIPTION OF THE DRAWINGS

[014] The present invention will now be described with reference to typical embodiments thereof and also with reference to the accompanying drawings, in which: Figure 1A illustrates a comparative scatter plot between the actual and predicted COT values ​​by an RNA / MLP model in the validation step according to a first example of the present invention; Figure 1B illustrates a comparative scatter plot between the actual and predicted COT values ​​by an RNA / MLP model in the test step according to the first example of the present invention; Figure 1C illustrates the actual and predicted TOC profiles of the SF-02 core sample according to the depth collected according to the first example of the present invention; Figure 2A illustrates a comparative scatter plot between the actual and predicted COT values ​​by an RNA / MLP model in the validation step according to a second example of the present invention; Figure 2B illustrates a comparative scatter plot between the actual and predicted COT values ​​by an RNA / MLP model in the test step according to the second example of the present invention; Figure 2C illustrates the actual and predicted TOC profiles of core SF-02 according to the collected depth of Petition 870250016331, dated 27 / 02 / 2025, page 49 / 72 7 / 21 in accordance with the second example of the present invention; Figure 3 illustrates the flowchart of the methodology according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[015] Specific embodiments of this disclosure are described below. In an effort to provide a concise description of these embodiments, all features of an actual implementation may not be described in the descriptive report. It should be appreciated that in the development of any actual implementation, as in any engineering or design project, numerous specific implementation decisions must be made to achieve the specific objectives of the developers, such as compliance with system and business-related constraints, which may vary from one implementation to another. Furthermore, it should be appreciated that such a development effort may be complex and time-consuming, but would nevertheless be a routine design and manufacturing undertaking for those of ordinary skill having the benefit of this disclosure.

[016] The present invention relates to a method for estimating the Total Organic Carbon (TOC) content in rock samples by means of spectral analysis of said samples. Such samples may be hand samples or borehole cores. The spectral analysis is performed using a spectroradiometer capable of measuring absolute reflectance on the rock surface in the spectral range of 1200 to 2400 nm. Preferably, the spectroradiometer should have its own light source, have a spectral resolution of up to 8 nm and perform calibration using a Spectralon® reference plate. For example, the Spectral Evolution SR 3500 device is considered suitable for carrying out the present invention.

[017] The inventors realized that the spectral data obtained by the spectroradiometer could be correlated with Petition 870250016331, dated 27 / 02 / 2025, pp. 50 / 72 8 / 21 the TOC content of rock samples. Machine learning models were specifically trained to recognize the patterns in the spectra obtained from the rock samples and thus estimate the TOC of each sample. Several types of algorithms, including artificial neural networks (ANNs), were tested to establish which was the most suitable for the task. Then, the chosen algorithm was trained with real spectral data obtained by a spectroradiometer and TOC data obtained by a carbon analyzer with a conventional infrared detector, such as Leco SC-144DR. With this, a model was obtained for estimating TOC from spectral data obtained from any rock sample.

[018] The methodological flow steps for obtaining the model are described below, with reference to figure 3: 1. Sample collection at the study site

[019] Due to the complexity of the problem and the high dimensionality of the data, a minimum of 50 representative samples is advisable so that machine learning experiments can be carried out effectively and the results are reliable. 2. Hyperspectral data collection

[020] Acquisition of spectral signature of samples with a spectroradiometer, with reflectance measurement at more than one point on the sample recommended if possible. If samples are extracted from borehole cores, it is important that measurements are recorded according to the depth of collection so that these data can be correlated with geochemical data. This step is followed by five sub-steps, as seen below. It is understood that the data preprocessing techniques that will be discussed below are, in a broad sense, well known in the state of the art, so it is not necessary to dwell on details that are routine for the technician in the field. Petition 870250016331, dated 27 / 02 / 2025, pp. 51 / 72 9 / 21 2.1. Hyperspectral data preprocessing - Spectral smoothing

[021] To reduce the noise inherent in the collected data, it is necessary to apply a spectral smoothing algorithm. In this case, the initial stage of the Savitzky-Golay filter is used, preferably and without limitation, which consists of fitting the data to a second-order polynomial, this fitting being calculated point by point and considering three points: the one to be smoothed, the one before and the one after. 2.2. Pre-processing of hyperspectral data - Selection of the spectral band of interest

[022] If the spectroradiometer used for hyperspectral data collection reads in a spectral range greater than between 1200 and 2400 nm, cut the data set to include only this range. This step can be omitted if the spectroradiometer already reads in the aforementioned range. 2.3. Hyperspectral data preprocessing - Absorption feature highlighting and continuum removal

[023] This approach is widely used in regression problems involving spectral data, as it seeks to normalize reflectance spectra and highlight the absorption features of each spectrum, thus allowing a better comparison between the different spectral curves and their absorption bands. First, the convex hull algorithm is applied to the highest reflectance values ​​of the analyzed spectral range to extract the curve continuum. Then, the highlighting of the features occurs by division, where the reflectance values ​​are converted to a normalized value relative to the curve continuum calculated previously. 2.4. Pre-processing of hyperspectral data - Calculation of the average spectral signatures per sample Petition 870250016331, dated 27 / 02 / 2025, pp. 52 / 72 10 / 21

[024] If more than one spectral measurement has been performed on the same sample, the average value between these must be calculated so that only one spectral signature is correlated to the TOC geochemical data. For example, multiple measurements can be performed from a large rock core. This step is unnecessary in cases where exactly one spectral signature is obtained from each rock sample. 2.5. Selection of input variables (features) for machine learning models

[025] Select the reflectance values ​​processed after continuum removal in substep 2.3 relative to the wavelengths: 1412, 1729, 1750, 1768, 1909, 2207, 2271, 2310, 2333, 2350 nm. The wavelength selection considers a domain knowledge-based approach, including central positions of absorption features of the constituent materials of these rocks (organic fraction - 1729, 1768 and 2310 nm; clay minerals and molecular water - 1412, 1909 and 2207 nm; and carbonate minerals - 2350 nm) and, for those features with nearby locations, a wavelength between these features (1750, 2271 and 2333 nm). It is important to note that depending on the spectral resolution of the sensor used, these wavelengths may vary slightly (for example, not having a spectral reading specifically at wavelength 1412 nm but rather at 1410 or 1420 nm), making it necessary to select the spectral bands closest to the 10 wavelengths mentioned above.To support this adjustment, the spectral response behavior of the samples can be observed, and thus the wavelength that best corresponds to the characteristic absorption features mentioned can be selected. 3. Laboratory analysis for determination of TOC (Total Organic Carbon).

[026] This step can be carried out, for example, by Petition 870250016331, dated 27 / 02 / 2025, pp. 53 / 72 11 / 21 combustion in a carbon analyzer with an infrared detector (destructive analysis), such as the Leco SC-144DR. As with hyperspectral sampling, it is important that the geochemical data be attributed to the sample or depth where measurements were taken with the spectroradiometer. 4. Combining the collected datasets.

[027] Associate the hyperspectral data set, that is, the 10 spectral variables obtained in substep 2.5, with the respective geochemical data of TOC content of each sample. 5. Dividing the dataset into training / validation and testing sets.

[028] In this step, the spectral data and their associated COT and depth data will be divided between a training set and a validation set for training the machine learning models. Preferably, 80% of the set is randomly divided for the training and validation steps of the predictive models, with the remaining 20% ​​reserved for testing the trained models and evaluating the generalization ability of the models to a new dataset. The training and validation of machine learning models are well understood in the art, so that those skilled in the art will be able to make adaptations to meet specific needs or criteria without departing from the scope of the present invention. 6. Standardization / Normalization in training and validation data

[029] Data preparation process so that the dataset is within a common range for the machine learning stage. Therefore, the spectral data were set to a mean of zero and a standard deviation of one, and the COT data were rescaled to the range between zero and one. 7. Adoption of the machine learning algorithm and definition of its hyperparameters that will be adjusted. Petition 870250016331, dated 27 / 02 / 2025, pp. 54 / 72 12 / 21

[030] For the application of the methodological flow proposed in this invention, algorithms from different areas within machine learning must be explored, as they have their respective approaches to the problem. Therefore, the defined machine learning algorithms are: Elastic Net (EN), K-Nearest Neighbours (KNN), Random Forest (RF), Support Vector Machine (SVM), and MultiLayer Perceptron (MLP) neural network. Obviously, any other machine learning algorithm considered suitable, whether known or yet to be developed, can be used without departing from the scope of the present invention. The hyperparameter grid that must be explored for tuning the aforementioned algorithms is:

[031] EN: (a) alpha: 0 to 1.0; (b) l1_ratio: 0 to 1.0.

[032] KNN: (a) n_neighbors: 1 to 9; (b) metric: Manhattan, Euclidean, Chebyshev, Angular; (c) weights: uniform, distance.

[033] RF: (a) n_estimators: 10 to 500; (b) criterion: squared_error, absolute_error, poisson; (c) mean_samples_leaf: 1, 2 or 3; (d) ccp: 0, 1, 10 or 100.

[034] SVM: (a) C: 0.0005 to 100; (b) kernel: linear, rbf, poly; (c) degree: 2 to 6; (d) gamma: auto, scale.

[035] MLP: (a) hidden_layer_sizes: 2 or 3 layers, with 5 to 200 neurons in each; (b) alpha: 0.0001 to 0.1; (c) learning_rate: constant, adaptive; (d) learning_rate_init: 1e05 to 0.01; (e) batch_size: 1 to 10. 8. Hyperparameter tuning

[036] This stage consists of the sequence of training and validation stages of models with different model configurations. The Bayesian search algorithm is used as an approach for hyperparameter optimization, which, based on the searches performed, selects only the relevant search space and discards the intervals that are unlikely to provide the best solution. A maximum value of is adopted. Petition 870250016331, dated 27 / 02 / 2025, pp. 55 / 72 13 / 21 1000 executions / searches are performed, stopping the experiment when it reaches the limit for classifiers like MLP, while in the case of algorithms with fewer hyperparameter combinations, such as KNN, the search ends when all possibilities are exhausted. In this way, 1000 models are generated and the adjustment is performed so that the value of the Mean Squared Error (MSE) performance metric is as low as possible. As a validation strategy, 10-fold crossvalidation is used, and it is important that the rock sample data are shuffled before the start of the process so that the folds do not represent combined depth ranges. Training is carried out in such a way that a machine learning model for COT estimation is obtained for each of the machine learning algorithms chosen in step 7. 9. Training and error estimation of the models

[037] New stage of training and validation of machine learning models with improved hyperparameter configuration for each algorithm (one model for each of the five adopted machine learning algorithms). The same training / validation set defined in stage 5 and standardized in stage 6 should be used. The 10-fold cross-validation strategy is also adopted. To interpret how well the estimates of each trained model were compared to the reference values ​​(measured COT), the performance metrics should be computed in the validation stage: coefficient of determination (R2), MSE, and mean absolute error (MAE), as well described in the technique. 10. Standardization / Normalization of test data

[038] Step similar to step 6, which should consider the same normalization parameters that were applied to the training / validation set. 11. Testing the models on the test dataset

[039] The models trained and validated in step 9 are Petition 870250016331, dated 27 / 02 / 2025, pp. 56 / 72 14 / 21 applied to a new test dataset. This test dataset was obtained in step 5. By analyzing the behavior of the selected models when dealing with data different from that with which they were trained, it is possible to obtain information about the predictive and generalization capacity of these algorithms, as well as to perceive, for example, if there was overfitting (i.e., over-fitting of the model to the training data). If all models fail in this regard, it is necessary to return to step 8 to select a new set of hyperparameters. 12. Comparison between the five generated models

[040] Evaluation of the results of the R2, MSE, and MAE metrics calculated in the validation and testing stages of the models for selecting the algorithm that best modeled the proposed problem (COT prediction) with the selected data. For this selection of the ideal model, the following criteria should be considered: the chosen model should present a high R2 in all stages (training, validation, and testing), being greater than 0.9 for training and 0.7 for validation and testing, thus indicating a good explanation of the data variability; in addition, the error values ​​(MSE and MAE) should be lower to ensure more accurate and robust predictions, preferably being below 2 to 3% (as a percentage of COT). If these conditions occur in more than one of the five trained and tested machine learning models, the chosen model should ideally present the highest R2 value, especially in the validation and testing stages, and the lowest error values ​​(MSE and MAE).The threshold values ​​suggested above for R2, MSE, and MAE are preferential, but not limiting. Particular applications may have different threshold values ​​without departing from the scope of the present invention. As these conditions may not occur simultaneously for the same model, the operator may choose the model with the highest R2 value at the step. Petition 870250016331, dated 27 / 02 / 2025, pp. 57 / 72 15 / 21 of the test or, optionally, perform a subjective selection considering all calculated metrics. If none of the models meet the established conditions, return to step 8. 13. Proposed model

[041] Based on the comparison from step 12, a final model is chosen to be applied to new hyperspectral data for indirect and non-destructive COT estimation.

[042] The above method can be repeated as desired to find an even more appropriate model. However, it should be noted that the geochemical characteristics of rocks can vary greatly between different locations. In this sense, it is vital that the methodological flow illustrated above be repeated for each new region that one wishes to investigate, for example, new sedimentary basins. In other words, one should not apply a model generated with data from a given area to another distinct area. In this way, the set of geochemical and hyperspectral data used in training the machine learning models will be representative of these new areas. Examples of the invention

[043] For the purpose of demonstrating the results of applying the method proposed in this invention, two distinct examples are presented: Example 1 - derived from modeling TOC content considering samples from only one study area; Example 2 - includes samples collected in two sedimentary basins, therefore the generated model tends to be generalizable to both.

[044] The first example involves applying the method to generate a predictive model of TOC at the local level, since geochemical and hyperspectral data from a single study site, the Sociedade outcrop, were used. Petition 870250016331, dated 27 / 02 / 2025, pp. 58 / 72 16 / 21 The Santa Fé Extractive Pit (Tremembé / SP), representative of the Tremembé Formation, in the Taubaté Basin, was the focus of this study. The aim was to validate the method and its ability to generate robust models that can be extrapolated to different targets within the same geological context. Therefore, for training and validation of the machine learning models, information from 54 samples collected from a core extracted at the site, designated SF-01, was used. The generalization capacity of the trained models was evaluated by applying them to a new dataset with 21 samples from a second core collected at the same site (SF-02). In this dataset, the TOC content of the samples ranged from 0.34% to 22.8%, with an average of 4.71%.

[045] After applying the methodological flow of the present invention to the aforementioned dataset, among the five machine learning algorithms explored (EN, KNN, RF, SVM, and MLP), the MLP neural network obtained the best results, with a coefficient of determination (R2) of 0.99 in model training, 0.84 in its validation, and 0.76 in the test stage on the second collected sample. The calculated Mean Absolute Error (MAE) was 0.70% in the validation stage and 2.4% during the test (values ​​relative to the same unit of the variable, i.e., as a percentage of COT). Figures 1A, 1B, and 1C demonstrate these results, where Figures 1A and 1B illustrate comparative scatter plots between the actual and predicted TOC values ​​by the MLP model in the validation and testing phases, respectively, and Figure 1C illustrates the actual and predicted TOC profiles of the SF-02 core sample according to the depth collected.

[046] Example 2 is a demonstration of the application of the method of the present invention to samples from different study areas. The method was applied to a dataset of 110 samples. Petition 870250016331, dated 27 / 02 / 2025, pp. 59 / 72 17 / 21 originating from two distinct sedimentary basins: the Taubaté Basin and the Araripe Basin. The objective of this application was to generate predictive models that could be generalized across different study areas, in this case from the two aforementioned basins, and that could be used to increase geochemical information from the study areas, considering that hyperspectral data collection was carried out in greater quantity than geochemical data. These data were collected from six borehole cores: two extracted near the outcrop of the Santa Fé Extractive Society (Tremembé / SP), representative of the Tremembé Formation, from the Taubaté Basin, and the other four originating from the Crato, Ipubi, and Romualdo Formations, from the Araripe Basin, one collected at the Três Irmãos Quarry (Nova Olinda / CE) and three at the São Jorge Mining Company (Trindade / PE). The concentration of TOC in the samples analyzed in this application ranges from 0 to 26%, with an average of 4.12%.

[047] The machine learning algorithm that best modeled the TOC content in the analyzed dataset was the MLP neural network, with a coefficient of determination (R2) of 0.96 during model training, 0.76 during validation, and 0.93 during the testing phase on data extracted from the dataset for subsequent evaluation of the models' predictive capacity. The MAE error statistic during validation was 1.64% and during testing 1.09%, meaning that in this case the incorrect prediction by the model did not exceed, on average, 2% of TOC. The results are illustrated in Figures 2A, 2B, and 2C in Figure 2, where Figures 2A and 2B illustrate comparative scatter plots between the actual and predicted TOC values ​​by the MLP model during the validation and testing phases, respectively, and Figure 2C illustrates the actual and predicted TOC profiles of the MSJ-02 core sample according to the collected depth.

[048] It is important to highlight the demonstration of another remarkable Petition 870250016331, dated 27 / 02 / 2025, pages 60 / 72 18 / 21 Applicability and advantage of the proposed method: the filling in of information between discrete laboratory samplings, which can be observed in Figure 2C. In the TOC data profile presented in the figure, there are 10 discrete data points between depths of 4 and 12 m, where samples were collected and laboratory analyses were performed. With the application of the method, approximately 70 TOC estimates were obtained from the MLP model for the same interval, therefore providing a prediction of the behavior of this variable between laboratory samplings.

[049] From the results presented, it is possible to perceive the potential of the method of the present invention for preliminary indirect characterization of potentially generating rock samples. In the TOC profiles presented in figures 1A, 1B, 1C, 2A, 2B and 2C, it can be seen that the TOC behavior predicted by the neural network was similar to the real data, following the trends of reduction and increase along the core.

[050] The advantages of the present invention will be immediately apparent to those skilled in the art, among which we can highlight: the present invention, proposed in the form of a means for estimating TOC from hyperspectral data, minimizes the difficulties of the prior art in performing laboratory analyses for this purpose, since spectroscopy constitutes a non-destructive approach for preliminary characterization of source rocks. This means that the samples remain intact after analysis, allowing their preservation and the possibility of future, more detailed analyses. In this way, only samples with estimated TOC concentrations above a certain established limit, considering the uncertainties of the applied model, would be selected for more precise and detailed quantitative laboratory investigations, such as analyses. Petition 870250016331, dated 27 / 02 / 2025, pages 61 / 72 19 / 21 geochemical, petrophysical, among others. The uncertainties of the proposed model are directly related to the model's performance metrics (R2, MSE, and MAE), which indicate the margin of error associated with TOC estimates. These uncertainties can vary significantly, influenced by sample characteristics, dataset homogeneity, and the complexity and limitations of the machine learning algorithms used. To address these uncertainties in sample selection for more detailed quantitative analyses, it is necessary to consider an appropriate confidence interval that incorporates the model's margin of error to ensure that the estimated TOC concentrations are above the selection limit. This process avoids selecting samples with values ​​that may have been overestimated by the model and ensures greater reliability in the screening process for subsequent investigations.

[051] It is also noteworthy that the technique used (collection of hyperspectral data with a spectroradiometer) does not require sample preparation, with measurements being taken directly on the surface of the raw rock. Furthermore, the equipment used is easy to handle and readings are taken in less than 10 seconds, allowing for quick and efficient data collection. Therefore, once the machine learning model that best matches the dataset is trained, the time between spectral data collection, preprocessing, and subsequent insertion into the model for TOC prediction is significantly shorter than that required for laboratory analysis.

[052] The ability to apply the models generated from the proposed methodological flow to hyperspectral images also solves the spatial resolution limitation of geochemical data from point sampling of outcrops or cores. Therefore, the content Petition 870250016331, dated 27 / 02 / 2025, pages 62 / 72 20 / 21 of TOC could be analyzed for its spatial distribution in these targets, allowing for a broader understanding of this parameter, which is so important in the geochemical characterization of rocks rich in organic matter.

[053] The method described here facilitates the daily work of geologists and geoscientists, whether from industry or academia, both in the field and in the laboratory, as it allows for the rapid obtaining of inferences about the amount of organic matter in source rocks without the need for laboratory analysis.

[054] The predictive model generated from the proposed method can function as an initial filter to select representative samples for more detailed geochemical analyses, which saves time, reduces costs, and increases the efficiency of the rock characterization process. This feature can also be useful for characterizing cores that are tens to hundreds of meters long, serving to fill gaps between discrete and spaced geochemical data. In this way, in laboratories with limited resources for laboratory analyses, it would not be necessary for a professional to classify potential samples, since the collection of hyperspectral data and its subsequent insertion into the trained model would allow obtaining an estimate of TOC, which would serve as a decision criterion for the necessary selection.

[055] The ability to extrapolate these models to hyperspectral images further expands the potential of the method, since it would allow the collection of images of various samples or cores for their preliminary characterization, or even their application at the outcrop scale. The application of the models generated in hyperspectral images of outcrops allows the analysis of the spatial distribution of the TOC parameter in the outcrop and, therefore, can be a Petition 870250016331, dated 27 / 02 / 2025, pages 63 / 72 21 / 21 is an important tool for more precise sample guidance, thereby reducing resources and sampling and analysis time, as well as avoiding possible errors in the evaluation and characterization of source rocks caused by point sampling.

[056] Although aspects of the present disclosure may be susceptible to various modifications and alternative forms, specific embodiments have been shown as examples in the drawings and have been described in detail in this document. But it should be understood that the invention is not intended to be limited to the particular forms disclosed. Instead, the invention should cover all modifications, equivalents and alternatives that fall within the scope of the invention, as defined by the following appended claims. Petition 870250016331, dated 27 / 02 / 2025, pp. 64 / 72

Claims

1 / 3 CLAIMS 1. Method for obtaining a machine learning model for estimating Total Organic Carbon (TOC) from hyperspectral data of rock samples, characterized in that it comprises the steps of: (1) Collecting rock samples; (2) Collect at least one hyperspectral datum from each of the rock samples collected in step (1); (2.1) Spectral smoothing of the spectral data collected in step (2); (2.3) Highlight absorption and removal features from the spectral data continuum; (2.5) Selection of input variables for the machine learning models; (3) Obtaining laboratory data on TOC from rock samples collected in step (1) (4) Association of hyperspectral data with the respective rock samples and with the TOC data obtained in step (3), obtaining a dataset; (5) Division of the dataset into training data and test data; (6) Normalization of training data; (7) Choosing one or more machine learning algorithms and defining their hyperparameters; (8) Adjusting the hyperparameters to obtain a machine learning model for each of the adopted machine learning algorithms; (9) Training and error estimation of the machine learning models obtained in step (8); (10) Normalization of test data; (11) Testing the machine learning models trained in step (9) using the test dataset, the test comprising calculating R2, MSE and MAE of each of the machine learning models trained in step (9); (12) Analysis of the results of the experiments carried out in steps (9) and (11) to determine the values ​​of R2, MSE and MAE; (13) Select the machine learning model based on the analysis of R2, MSE and MAE values ​​performed in step (12).

2. Method, according to any of the preceding claims, characterized in that step (2) is performed by a Spectroradiometer configured for absolute reflectance measurement on the surface of rock samples by means of a contact probe with its own light source.

3. Method, according to claim 2, characterized in that the Spectroradiometer operates in the range of 1200-2400 nm.

4. Method according to claim 2, characterized in that if the Spectroradiometer operates in a range greater than 1200-2400 nm, the method further comprises, after substep (2.1): (2.2) Selection of the spectral range of interest, wherein substep (2.2) comprises clipping the operating range so that the spectral data obtained are only within the range between 1200-2400 nm. 5.Method, according to any of the preceding claims, characterized in that if two or more hyperspectral data are obtained per sample in step (2), the method further comprises after substep (2.3): (2.4) Calculation of the average of the spectral signatures per sample.

6. Method, according to any of the preceding claims, characterized in that substep (2.5) comprises selecting the processed reflectance values ​​relating to wavelengths equal to 1412, 1729, 1750, Petition 870250016331, dated 27 / 02 / 2025, page 66 / 72 3 / 3 1768, 1909, 2207, 2271, 2310, 2333, and 2350 nm.

7. Method, according to any of the preceding claims, characterized in that step (3) uses a destructive technique performed by a carbon analyzer with an infrared detector. 8.Method, according to any of the preceding claims, characterized in that the machine learning algorithms are Elastic Net, K-Nearest Neighbours, Random Forest, Support Vector Machine, and Multi-Layer Perceptron.

9. Method, according to any of the preceding claims, characterized in that selecting the machine learning model (13) comprises selecting the model whose R2 value was calculated as greater than 0.9 in step (9) and greater than 0.7 in step (11) and whose MSE and MAE values ​​were below 2% in COT value in steps (9) and (11).

10. Method, according to claim 9, characterized in that if more than one model meets the conditions, the model with the highest R2 value in step (11) is selected.

11. Method, according to any of claims 9 or 10, characterized in that the operator can perform manual model selection. Petition 870250016331, dated 27 / 02 / 2025, pp. 67 / 72.

Images