Monte Carlo-based mineral assemblage analysis method

Abstract

This invention discloses a Monte Carlo-based mineral assemblage analysis method, belonging to the field of mineral analysis. The invention includes the following steps: multi-source data fusion and preprocessing; physical model construction and initial solution generation; multi-scale feature fusion and extraction; Monte Carlo random sampling and solution space exploration; physical-data dual-driven evaluation and screening; iterative update and model optimization; convergence judgment and result output. This invention effectively solves the problem of multiple solutions in mineral assemblage identification by combining multi-source data fusion, a physical-data dual-driven model, and the Monte Carlo method with multi-scale feature fusion and iterative optimization. This method can integrate geophysical, geochemical, remote sensing, borehole core, and mineral spectral data, and through iterative model updates, gradually converges to the most probable mineral assemblage, providing uncertainty assessment, thereby improving the accuracy and reliability of mineral identification.

Description

Technical Field

[0001] This invention relates to the field of mineral analysis, and more specifically to a Monte Carlo-based method for mineral assemblage analysis. Background Technology

[0002] In mineral analysis, different mineral assemblages can produce similar major element chemical analysis results, leading to the problem of multiple solutions. For example, high CaO levels may originate from calcite, dolomite, or wollastonite. Traditional methods rely on single chemical analysis data, making it difficult to uniquely determine the mineral assemblages. In recent years, multi-source information fusion and physical-data dual-drive approaches have been applied in intelligent mineral exploration, but they are not common in the field of mineral analysis. Therefore, an iterative, multi-dimensional data fusion method is needed to approximate a unique true solution. Summary of the Invention

[0003] This invention provides a Monte Carlo-based mineral assemblage analysis method, which solves the problems of existing technologies.

[0004] In a first aspect, the present invention provides a Monte Carlo-based mineral assemblage analysis method, comprising the following steps:

[0005] Step 1: Multi-source data fusion and preprocessing; Input geophysical, geochemical, remote sensing images, borehole core and mineral spectral data, and establish a multi-source data cube with unified coordinate system, unified grid resolution and standardized values;

[0006] Step 2: Physical model construction and initial solution generation; Input the preprocessed multi-source data cube, perform inversion based on mineral co-occurrence sequence and hydrothermal alteration zoning, generate possible mineral assemblage solution space, and organize it into an initial mineral assemblage candidate set;

[0007] Step 3: Multi-scale feature fusion and extraction; The multi-source data cube obtained in Step 1 and the initial mineral combination candidate set obtained in Step 2 are used to extract spatial and compositional features at different scales using a multi-branch CNN. Pixel-level fusion is used to integrate the information, enhance the ability to represent the spatial distribution and compositional coupling relationship of minerals, and obtain the fused multi-channel feature tensor.

[0008] Step 4: Monte Carlo random sampling and solution space exploration; Take the fused feature map and the initial mineral assemblage candidate set from Step 3, add random perturbations to parameters such as mineral composition ratio and spatial distribution, and generate a new candidate set based on the Monte Carlo method;

[0009] Step 5: Physical-Data Dual-Driven Evaluation and Screening; The Monte Carlo candidate solution set and multi-channel feature tensor obtained in Step 4 are scored based on physical constraints such as mineral thermal stability and coexistence rules, and data-driven evaluation based on factors such as matching degree with geochemical features and spectral similarity. Each candidate solution is evaluated together, and a weighted scoring mechanism is used to eliminate solutions that do not conform to geological laws.

[0010] Step 6: Iterative updates and model optimization;

[0011] Based on the inversion error of the current optimal solution, the candidate solutions after scoring are compared with historical iteration records. The convolution kernel weights in the CNN and the component thresholds in the physical model are adjusted in reverse. More prior knowledge is introduced as soft constraints to gradually converge to a stable solution, and the updated mineral combination model and adjusted feature extraction parameters are output.

[0012] Step 7: Convergence judgment and result output;

[0013] The solution set of mineral combinations after multiple iterations is considered convergent if the difference between the solutions in three consecutive iterations is less than a threshold, which is a 2% change in the mineral proportion. The final mineral combination and its spatial distribution are output, along with the uncertainty interval estimated by the Monte Carlo method.

[0014] Furthermore, the borehole core data includes: converting point data into data layer data using Kriging interpolation; standardizing the data using Z-score to ensure all features are within the same range; outliers are detected and denoised using Z-score, removing values ​​with a Z-score greater than 3 or less than -3; the text or graphical record of the borehole core data is a borehole lithology log; and known mineral assemblage information is extracted from the borehole lithology log as prior knowledge. These prior combinations are then used to initialize the candidate set in the subsequent initial solution generation.

[0015] Furthermore, the inversion based on mineral co-occurrence sequences and hydrothermal alteration zoning includes: establishing a rule base based on mineral co-occurrence sequences and hydrothermal alteration zoning;

[0016] Mineral symbiotic sequences are the order in which minerals are formed under specific geological conditions, providing a temporal order for mineral assemblages; based on hydrothermal alteration zoning, specific mineral assemblages are expected to be found in specific regions.

[0017] Furthermore, for multi-branch CNNs, including:

[0018] Each branch is initialized with a different kernel size, and the weights are initialized using He or Xavier. It contains 4 parallel convolutional paths with kernel sizes of 1×1, 3×3, 5×5 and 3×3 max pooling + 1×1 convolution. Each branch independently processes the input data cube and extracts features at different scales. The 1x1 convolution captures point features, the 3x3 convolution captures local features, and the 5x5 convolution captures region features. Region features are larger than local features.

[0019] Furthermore, the pixel-level fusion includes: during pixel-level fusion, the feature maps output by each branch are added at the pixel level, that is, corresponding pixels are added together, and the feature maps output by all branches have the same spatial size and number of channels.

[0020] Furthermore, the addition of random perturbations to parameters such as mineral composition ratio and spatial distribution includes: adding random noise to the composition ratio or spatial distribution of each solution in the current mineral combination solution set; sampling noise from a Gaussian distribution for the mineral ratio; and moving the position of the mineral region in terms of spatial distribution.

[0021] Furthermore, for the Monte Carlo method, the specific steps for generating new candidate solutions through random sampling include:

[0022] Step 4.1: Randomly select a solution from the current solution set;

[0023] Step 4.2: Apply perturbation to the mineral proportions;

[0024] New ratio = Old ratio + ε;

[0025] Where ε ~ N(0, σ), and σ is the standard deviation, which controls the magnitude of the disturbance;

[0026] Step 4.3: Ensure the new ratio is within the range [0,1] and the sum is 1;

[0027] Step 4.4: Repeat steps 4.1-4.3 multiple times to generate multiple new candidate solutions.

[0028] Furthermore, the scoring also includes: a scoring function that combines physical constraints and data-driven matching degree;

[0029] Physical constraint scoring: Based on mineral symbiosis rules and thermal stability rules, points are deducted when the combination contains non-symbiotic minerals and points are added when they are symbiotic. Based on thermal stability rules: calculate the Gibbs free energy of the mineral combination at the regional temperature, and deduct points if it deviates from the standard value.

[0030] Data-driven scoring: calculates the matching degree between candidate solutions and multi-source features; geochemical matching degree: calculates the mean square error between the major elements predicted by the candidate solutions and the measured values; spectral similarity: uses spectral angle mapping to compare the spectral features of candidate solutions with remote sensing data.

[0031] Weighted scoring: Total score = w1 * Physical score + w2 * Data score, where w1 and w2 are the weights;

[0032] When the physical score is below the threshold, it is considered a solution that does not conform to geological laws and is eliminated, or when a known non-symbiotic mineral pair appears, it is eliminated.

[0033] Furthermore, the convergence judgment and result output specifically include:

[0034] The solution set of mineral combinations after multiple iterations is considered converged if the difference between the solutions in three consecutive iterations is less than a threshold, which is a 2% change in the mineral proportion. The final mineral combination and its spatial distribution are output, along with the uncertainty interval estimated by the Monte Carlo method.

[0035] The specific calculation method is as follows:

[0036] Calculate the change in mineral proportion: For each mineral, calculate the absolute difference in proportion between consecutive iterations, and then average it across all minerals. That is, for iterations i, i+1, and i+2, calculate |proportion_i+1 - proportion_i| and |proportion_i+2 - proportion_i+1|, and then take the average. If the average is less than 2%, then the convergence is achieved.

[0037] This invention effectively addresses the ambiguity problem in mineral assemblage identification by employing multi-source data fusion, a physics-data dual-driven model, and the Monte Carlo method, combined with multi-scale feature fusion and iterative optimization. This method integrates geophysical, geochemical, remote sensing, borehole core, and mineral spectral data, iteratively updates the model to gradually converge to the most probable mineral assemblage, and provides uncertainty assessment, thereby improving the accuracy and reliability of mineral identification. Attached Figure Description

[0038] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention. In the drawings:

[0039] Figure 1 A flowchart of a Monte Carlo-based mineral assemblage analysis method provided as an exemplary embodiment of the present invention. Detailed Implementation

[0040] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention.

[0041] Technical concept of the present invention:

[0042] Multi-source information fusion and physical-data dual-drive approaches have applications in intelligent mineral exploration, but are not common in mineral analysis. Different mineral combinations may yield similar major element chemical analysis results. For example, high CaO levels may originate from calcite, dolomite, or wollastonite. To address the problem of multiple solutions, iterative multi-dimensional data fusion can be used to approximate a unique true solution. Therefore, the challenge becomes how to integrate different types of data, algorithms, and prior knowledge to construct a continuously self-correcting analysis system that approximates the truth for application in mineral analysis—a crucial technical problem that urgently needs to be solved.

[0043] The design framework of this invention is based on multi-source data fusion. First, it gathers multi-source data such as geophysical (e.g., gravity, magnetism, electricity), geochemical, remote sensing geology, and borehole data. Preprocessing and standardization are then performed, including noise removal, coordinate system unification, and dimensional normalization, to ensure data comparability. Next, an initial model or candidate solutions are established using geological mineralization theory and geophysical inversion theory. A multi-scale feature fusion algorithm is then used to fuse the data, extracting deeper mineralization-related features that are difficult for the human eye to directly identify. Finally, iterative inference and approximation are performed. This step is a cyclical process, including: generating a preliminary solution; introducing new validation data or geological prior knowledge; screening the preliminary solution; adding constraints to the system; and updating the model by adjusting the parameters of the physical model or optimizing the feature extraction method of the data model based on the validation results, thereby narrowing down the range of solutions until the most stable and reasonable solution is found.

[0044] The Monte Carlo-based mineral assemblage analysis method provided by this invention aims to solve the above-mentioned technical problems of the prior art.

[0045] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0046] Example 1:

[0047] like Figure 1 As shown, Embodiment 1 of the present invention is carried out in the following manner:

[0048] Step 1: Multi-source data fusion and preprocessing;

[0049] Input geophysical, geochemical, remote sensing imagery, borehole core, and mineral spectral data to establish a multi-source data cube with standardized values ​​and uniform grid resolution under a unified coordinate system; the software and equipment used are ArcGIS platform, ENVI remote sensing processing software, and NumPy;

[0050] The multi-source data cube is a four-dimensional data structure that provides multi-dimensional features for each spatial location. For example, geochemical data may show high CaO, geophysical data may show high density, and remote sensing data may show specific spectral features. Each pixel contains values ​​from multiple data layers, which come from different sources including geophysical, geochemical, and remote sensing images. Under a unified coordinate system and grid resolution, each data layer is aligned to the same spatial grid, forming a multi-dimensional array.

[0051] Among them, a multi-source data cube with a unified coordinate system and grid resolution (format: [X, Y, Z, C], where X / Y are spatial coordinates, Z is depth / elevation, and C is a data channel) is used. Geophysical data (such as gravity anomalies) is used as one channel, geochemical data (such as CaO content) is converted into grid data through Kriging interpolation, remote sensing images (such as multispectral bands) are resampled to the same grid, and borehole core data is filled into the grid through three-dimensional interpolation.

[0052] Among them, the text or graphic record of the borehole core data is the borehole lithology log, which describes the lithology and mineral composition at different depths in the borehole. Combined with the borehole lithology log, known mineral combination information can be extracted. For example, if the log shows that there is calcite and quartz at a certain depth, the corresponding combination can be used as prior knowledge. In the subsequent initial solution generation, these prior combinations are used to initialize the candidate set. At the same time, different prior knowledge is obtained according to the adaptability of different mineral layer conditions, which ensures the universality of the invention for different geological conditions in the application context.

[0053] Kriging interpolation is used to convert the borehole data of point data into data layer data. During the standardization process, Z-score standardization is used to ensure that all features are within the same range. Outliers are detected and denoised using Z-score, and values ​​with a Z-score greater than 3 or less than -3 are removed.

[0054] Step 2: Physical model construction and initial solution generation;

[0055] Input the preprocessed multi-source data cube, perform inversion based on mineral co-occurrence sequence and hydrothermal alteration zoning, generate possible mineral assemblage solution space, and organize it into an initial mineral assemblage candidate set;

[0056] A rule base is established based on mineral co-occurrence sequences and hydrothermal alteration zoning. The mineral co-occurrence sequence is the order in which minerals are formed under specific geological conditions, providing a temporal order for mineral assemblages. Hydrothermal alteration zoning reflects the spatial zoning of rock and mineral changes caused by hydrothermal activity. For example, the spatial zoning of rock and minerals from the inside out may be silicification zone, sericitization zone, and chloritization zone. Based on hydrothermal alteration zoning, specific mineral assemblages can be expected to be found in specific regions. For example, in hydrothermal systems, quartz forms earlier than calcite.

[0057] Methods for generating the solution space: Use rule-based reasoning or decision trees to predict possible mineral combinations based on multi-source data; input multi-source data features and output a set of possible mineral combinations and their probabilities; the obtained initial solution space is a list, where each element is a mineral combination, such as ["Quartz", "Calcite"] and its initial probability, [("Calcite","Quartz"), ("Dolomite", "Wollastonite")] and its initial probability;

[0058] Step 3: Multi-scale feature fusion and extraction;

[0059] The multi-source data cube obtained in step 1 and the initial mineral combination candidate set obtained in step 2 are used to extract spatial and compositional features at different scales using a multi-branch CNN. Pixel-level fusion is used to integrate the information, enhance the ability to represent the spatial distribution and compositional coupling relationship of minerals, and obtain the fused multi-channel feature tensor.

[0060] For multi-branch CNNs, each branch is initialized with a different kernel size, and the weights are initialized using He or Xavier. It contains four parallel convolutional paths with kernel sizes of 1×1, 3×3, 5×5, and 3×3 max pooling + 1×1 convolution. Each branch independently processes the input data cube, extracting features at different scales: 1x1 convolutions capture point features, 3x3 captures local features, and 5x5 captures features from larger regions. Each branch's convolutional layer is followed by a ReLU activation function and a pooling layer with a stride of 1 and uniform padding to maintain spatial dimensions. During pixel-level fusion, the feature maps output by each branch are added at the pixel level, i.e., corresponding pixels are added together. All branch output feature maps have the same spatial dimensions and number of channels. Pixel-level fusion preserves spatial consistency; the features of each pixel are a synthesis of multiple scales, facilitating more refined spatial analysis. The obtained multi-channel feature tensor is a three-dimensional array: [H, W, ... [C], where H and W are the spatial height and width, and C is the number of channels. Each channel represents a feature map containing texture, composition and other features extracted from different branches. The multi-channel feature tensor can be further convolved or pooled to reduce dimensionality. The multi-channel feature tensor contains deep features extracted from multi-source data, which may correspond to mineral distribution, structural information, etc.

[0061] Step 4: Monte Carlo random sampling and solution space exploration;

[0062] Random perturbations are added to parameters such as mineral composition ratio and spatial distribution in the fused feature map and the initial mineral combination candidate set in step 3, and a new candidate set is generated based on the Monte Carlo method.

[0063] Adding perturbation: For each solution in the current mineral combination solution set, add random noise to its component ratio. This can be done by sampling noise from a Gaussian distribution to adjust the mineral ratio; it can also be applied to spatial distribution by slightly moving the position of the mineral region.

[0064] The Monte Carlo method involves the following steps for generating new candidate solutions through random sampling:

[0065] Step 4.1: Randomly select a solution from the current solution set;

[0066] Step 4.2: Apply perturbation to the mineral proportions;

[0067] New ratio = Old ratio + ε;

[0068] Where ε ~ N(0, σ), and σ is the standard deviation, which controls the magnitude of the disturbance;

[0069] Step 4.3: Ensure the new ratio is within the range [0,1] and the sum is 1;

[0070] Step 4.4: Repeat steps 4.1-4.3 multiple times to generate multiple new candidate solutions;

[0071] Incorporating Monte Carlo sampling allows for the exploration of new regions in the solution space, avoiding local optima;

[0072] Step 5: Physical-Data Dual-Driven Evaluation and Screening;

[0073] The Monte Carlo candidate solution set and multi-channel feature tensor obtained in step 4 are scored based on physical constraints such as mineral thermal stability and coexistence rules, and data-driven evaluation based on factors such as matching degree with geochemical features and spectral similarity. Each candidate solution is evaluated together, and a weighted scoring mechanism is used to eliminate solutions that do not conform to geological laws.

[0074] The scoring function combines physical constraints and data-driven matching:

[0075] Physical constraint scoring: Based on mineral coexistence rules and thermal stability rules, for example, points are deducted if the assemblage contains non-coexisting minerals. Scores can be binary (0 or 1) or continuous (based on the degree of violation); for example, calcite coexisting with quartz receives 1 point, while coexisting with olivine receives 0 points; based on thermal stability rules: calculate the Gibbs free energy of the mineral assemblage at the regional temperature, and deduct points if it deviates from the standard value;

[0076] Data-driven scoring: Calculate the matching degree between candidate solutions and multi-source features, such as using mean squared error between predicted mineral spectra and actual spectra, or using classification probabilities; Geochemical matching degree: Calculate the mean squared error between the major elements predicted by the candidate solutions and the measured values; Spectral similarity: Use spectral angle mapping to compare the spectral features of candidate solutions with remote sensing data.

[0077] Weighted scoring: Total score = w1 * Physical score + w2 * Data score, where w1 and w2 are the weights;

[0078] If the physical score is below the threshold, it is a solution that does not conform to geological laws and is eliminated. For example, when w1 is 0.6 and w2 is 0.4, the physical score is <0.3, or a known non-symbiotic mineral pair appears (such as a combination of high-temperature minerals and low-temperature minerals).

[0079] Step 6: Iterative updates and model optimization;

[0080] Based on the inversion error of the current optimal solution, the candidate solutions after scoring are compared with historical iteration records. The convolution kernel weights in the CNN and the component thresholds in the physical model are adjusted in reverse. More prior knowledge is introduced as soft constraints to gradually converge to a stable solution, and the updated mineral combination model and adjusted feature extraction parameters are output.

[0081] Among them, the historical iteration record contains the best solution, score, feature extraction parameters including CNN weights, and component threshold for each iteration. In the first iteration, the historical record is empty, so the initial model parameters such as mineral ratio threshold = 0.05 are used.

[0082] Adjusting convolutional kernel weights: Using the backpropagation algorithm, based on a loss function that includes the rating error of candidate solutions (e.g., a low rating results in a large loss), the CNN weights are adjusted via gradient descent.

[0083] Composition threshold correlation: Composition thresholds are used to truncate the proportion of minerals in the physical model. Threshold adjustment based on error: If some minerals are incorrectly predicted, their thresholds are adjusted.

[0084] Prior knowledge is introduced, including regional geological maps and mineral deposit models, as soft constraints, and incorporated into the loss function as a regularization term.

[0085] The adjusted feature extraction parameters are obtained from CNN training, including convolutional kernel weights and biases;

[0086] Step 7: Convergence judgment and result output;

[0087] The solution set of mineral combinations after multiple iterations is considered convergent if the difference between the solutions in three consecutive iterations is less than a threshold, which is a 2% change in the mineral proportion. The final mineral combination and its spatial distribution are output, along with the uncertainty interval estimated by the Monte Carlo method.

[0088] Calculate the change in mineral proportion: For each mineral, calculate the absolute difference in proportion between consecutive iterations, and then average it across all minerals. That is, for iterations i, i+1, and i+2, calculate |proportion_i+1 - proportion_i| and |proportion_i+2 - proportion_i+1|, and then take the average. If the average is less than 2%, then the convergence is achieved.

[0089] By conducting multiple Monte Carlo samplings, the posterior distribution and confidence intervals of the mineral proportions are obtained. The uncertainty intervals help to improve the reliability of the predictions and aid in risk assessment. For example, if the uncertainty is high, more data is needed.

[0090] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0091] Furthermore, in the embodiments of the present invention, the functional modules can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods or systems. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0093] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0094] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

[0095] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the invention disclosed herein in the specification and examples. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the foregoing claims.

[0096] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A Monte Carlo-based mineral assemblage analysis method, characterized in that, Includes the following steps: Step 1: Multi-source data fusion and preprocessing; Input geophysical, geochemical, remote sensing images, borehole core and mineral spectral data, and establish a multi-source data cube with unified coordinate system, unified grid resolution and standardized values; Step 2: Physical model construction and initial solution generation; Input the preprocessed multi-source data cube, perform inversion based on mineral co-occurrence sequence and hydrothermal alteration zoning, generate possible mineral assemblage solution space, and organize it into an initial mineral assemblage candidate set; Step 3: Multi-scale feature fusion and extraction; The multi-source data cube obtained in Step 1 and the initial mineral combination candidate set obtained in Step 2 are used to extract spatial and compositional features at different scales using a multi-branch CNN. Pixel-level fusion is used to integrate the information, enhance the ability to represent the spatial distribution and compositional coupling relationship of minerals, and obtain the fused multi-channel feature tensor. Step 4: Monte Carlo random sampling and solution space exploration; The fused multi-channel feature tensor and the initial mineral combination candidate set from Step 3 are subjected to random perturbations in terms of mineral composition ratio and spatial distribution parameters, and a candidate solution set is generated based on the Monte Carlo method; Step 5: Physical-Data Dual-Driven Evaluation and Screening; The Monte Carlo method obtained in Step 4 generates a candidate solution set and a multi-channel feature tensor. Based on the matching degree of physical constraints of mineral thermal stability and coexistence rules with geochemical features and spectral similarity, a data-driven scoring is performed to jointly evaluate each candidate solution. A weighted scoring mechanism is adopted to eliminate solutions that do not conform to geological laws. Step 6: Iterative updates and model optimization; Based on the inversion error of the current optimal solution, the candidate solutions after scoring are compared with historical iteration records. The convolution kernel weights in the CNN and the component thresholds in the physical model are adjusted in reverse. More prior knowledge is introduced as soft constraints to gradually converge to a stable solution, and the updated mineral combination model and adjusted feature extraction parameters are output. Step 7: Convergence judgment and result output; The solution set of mineral combinations after multiple iterations is considered convergent if the difference between the solutions in three consecutive iterations is less than a threshold, which is a 2% change in the mineral proportion. The final mineral combination and its spatial distribution are output, along with the uncertainty interval estimated by the Monte Carlo method.

2. The Monte Carlo-based mineral assemblage analysis method according to claim 1, characterized in that, The borehole core data includes: converting point data into data layer data using Kriging interpolation; standardizing the data using Z-score to ensure all features are within the same range; removing outliers using Z-score detection; and recording the text or graphical data of the borehole core data as a borehole lithology log. Known mineral assemblage information is extracted from the borehole lithology log as prior knowledge, and these prior combinations are used to initialize the candidate set in subsequent initial solution generation.

3. The Monte Carlo-based mineral assemblage analysis method according to claim 2, characterized in that, The inversion based on mineral co-occurrence sequences and hydrothermal alteration zoning includes: establishing a rule base based on mineral co-occurrence sequences and hydrothermal alteration zoning; Mineral symbiotic sequences are the order in which minerals are formed under specific geological conditions, providing a temporal order for mineral assemblages; based on hydrothermal alteration zoning, specific mineral assemblages are expected to be found in specific regions.

4. The Monte Carlo-based mineral assemblage analysis method according to claim 3, characterized in that, For multi-branch CNNs, including: Each branch is initialized with a different kernel size, and the weights are initialized using He or Xavier. It contains 4 parallel convolutional paths with kernel sizes of 1×1, 3×3, 5×5 and 3×3 max pooling + 1×1 convolution. Each branch independently processes the input data cube and extracts features at different scales. The 1x1 convolution captures point features, the 3x3 convolution captures local features, and the 5x5 convolution captures region features. Region features are larger than local features.

5. The Monte Carlo-based mineral assemblage analysis method according to claim 4, characterized in that, The pixel-level fusion includes: during pixel-level fusion, the feature maps output by each branch are added at the pixel level, that is, corresponding pixels are added, and the feature maps output by all branches have the same spatial size and number of channels.

6. The Monte Carlo-based mineral assemblage analysis method according to claim 5, characterized in that, The addition of random perturbations to the mineral composition ratio and spatial distribution parameters includes: adding random noise to the composition ratio or spatial distribution of each solution in the initial mineral combination candidate set; sampling noise from the Gaussian distribution for the mineral ratio; and moving the position of the mineral region in terms of spatial distribution.

7. The Monte Carlo-based mineral assemblage analysis method according to claim 6, characterized in that, For the Monte Carlo method, the specific steps for generating new candidate solutions through random sampling include: Step 4.1: Randomly select a solution from the current solution set; Step 4.2: Apply perturbation to the mineral proportions; New ratio = Old ratio + ε; Where ε ~ N(0, σ), and σ is the standard deviation, which controls the magnitude of the disturbance; Step 4.3: Ensure the new ratio is within the range [0,1] and the sum is 1; Step 4.4: Repeat steps 4.1-4.3 multiple times to generate multiple new candidate solutions.

8. The Monte Carlo-based mineral assemblage analysis method according to claim 7, characterized in that, For scoring, it also includes: a scoring function that combines physical constraints and data-driven matching degree; Physical constraint scoring: Based on mineral symbiosis rules and thermal stability rules, points are deducted when the combination contains non-symbiotic minerals and points are added when they are symbiotic. Based on thermal stability rules: calculate the Gibbs free energy of the mineral combination at the regional temperature, and deduct points if it deviates from the standard value. Data-driven scoring: calculates the matching degree between candidate solutions and multi-source features; geochemical matching degree: calculates the mean square error between the major elements predicted by the candidate solutions and the measured values; spectral similarity: uses spectral angle mapping to compare the spectral features of candidate solutions with remote sensing data. Weighted scoring: Total score = w1 * Physical constraint score + w2 * Data-driven score, where w1 and w2 are weights; When the physical score is below the threshold, it is considered a solution that does not conform to geological laws and is eliminated, or when a known non-symbiotic mineral pair appears, it is eliminated.

9. The Monte Carlo-based mineral assemblage analysis method according to claim 8, characterized in that, The convergence judgment and result output specifically include: The solution set of mineral combinations after multiple iterations is considered convergent if the difference between the solutions in three consecutive iterations is less than a threshold, which is a 2% change in the mineral proportion. The final mineral combination and its spatial distribution are output, along with the uncertainty interval estimated by the Monte Carlo method. The specific calculation method is as follows: Calculate the change in mineral proportion: For each mineral, calculate the absolute difference in proportion between consecutive iterations, and then average it across all minerals. That is, for iterations i, i+1, and i+2, calculate |proportion_i+1 - proportion_i| and |proportion_i+2 - proportion_i+1|, and then take the average. If the average is less than 2%, then the convergence is achieved.

Images