A method for quantitatively analyzing a clastic rock source

Abstract

The present application relates to the technical field of geological information processing, and discloses a method for quantitatively analyzing detrital rock sources, which comprises the following steps: collecting detrital rock standard samples; determining the multi-scale source characteristics of each detrital rock standard sample, and performing normalization processing on the analysis results of the mineral composition and the element geochemical composition of each detrital rock standard sample; based on the multi-scale source characteristics of each detrital rock standard sample, constructing a source characteristic dynamic database with a three-layer structure of source labels, feature vectors and metadata, and using a multi-level quantitative model to match the multi-scale source characteristics of an unknown source sample to be tested with the feature vectors of the source characteristic dynamic database module by module, so as to preliminarily identify the matching source of the sample to be tested; and taking the background attribute information of the sample to be tested as a hard constraint condition to perform consistency checking, and outputting a unique optimal matching source after inspection. The present application solves the technical bottleneck that the traditional method relies on static charts and is difficult to cover multiple sample types.

Description

Technical Field

[0001] This invention relates to the field of geological information processing technology, specifically to a method for quantitative analysis of clastic rock provenance. Background Technology

[0002] In basin analysis and mineral exploration, accurately determining the source area of ​​clastic sediments (i.e., provenance) is a key prerequisite for reconstructing paleogeography and restoring the regional tectonic evolution history. The essence of this process is to trace back to the parent rock based on the identity information carried in the sedimentary products.

[0003] Traditional provenance identification relies on experts to visually compare the mineral composition, element content and other geochemical characteristics of the sample with standard maps. Although this method is based on profound geological principles, it is inefficient when faced with a large number of samples and is prone to subjective experience bias. It is especially difficult to give repeatable quantitative conclusions for samples in transitional areas.

[0004] Most automated quantitative methods that have emerged in recent years simplify provenance analysis into a purely pattern recognition problem detached from the geological context. Their core algorithms only pursue minimizing the distance between feature vectors in mathematical space, completely ignoring the natural laws governing sediment formation and transport. For example, a discriminant function based purely on chemical composition might link a sample from a basin to a source region spatially isolated by the deep sea or with a completely opposite paleocurrent direction. While this result may be numerically optimal, it is geologically impossible. Summary of the Invention

[0005] The purpose of this invention is to provide a quantitative analysis method for clastic rock provenance, in order to solve the technical problem that the existing technology only pursues the minimization of the distance between eigenvectors in mathematical space, but completely ignores the natural laws that sediment formation and transport must follow.

[0006] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution: A quantitative analysis method for clastic rock provenance includes the following steps: Step 100: Collect standard clastic rock samples from different geological backgrounds and sedimentary environments, with known provenance. Step 200: Determine the multi-scale provenance characteristics of each clastic rock standard sample. The multi-scale provenance characteristics include the mineral composition, elemental geochemical composition, age determination, morphology and structure, and grain size parameters of the clastic rock standard sample. Normalize the analysis results of the mineral composition and elemental geochemical composition of each clastic rock standard sample. Step 300: Based on the multi-scale provenance characteristics of each clastic rock standard sample, construct a dynamic database of provenance characteristics with a three-layer structure of provenance tag, feature vector, and metadata. The provenance tag stores the geological background and sedimentary environment information of the clastic rock standard sample. The feature vector stores the multi-scale provenance characteristics in a standardized numerical matrix. The metadata stores the background attribute information associated with the provenance tag. The background attribute information includes, but is not limited to, sedimentary facies markers, paleocurrent data, and regional tectonic background. Step 400: Obtain the multi-scale source characteristics of the test sample with unknown source. Using a multi-level quantitative model, match the multi-scale source characteristics of the test sample with the feature vectors of the source characteristic dynamic database module by module to preliminarily identify the matching source of the test sample. The multi-level quantitative model includes at least: a preliminary screening module based on mineral composition, a probability matching module based on elemental geochemistry, and a fusion decision module using the remaining multi-scale features. Step 500: Obtain the background attribute information of the area where the sample to be tested is located. Use the background attribute information as a hard constraint and verify its consistency with the metadata background attribute information corresponding to the initially identified matching source. After verification, output the unique optimal matching source.

[0007] Furthermore, in step 200, high-precision equipment is used to analyze the mineral composition and elemental geochemical composition; The mineral composition of each clastic rock sample was measured to include quartz, feldspar, rock fragments, and other minerals. The normalization method for the mineral composition of each clastic rock sample was as follows: The original contents of quartz, feldspar, and rock fragments were measured to be: Qr, Fr, Lr; The measured contents of quartz, feldspar, and rock fragments were normalized. The normalized contents of quartz, feldspar, and rock fragments are Q, F, and L, respectively. : ; ; ; The elemental geochemical composition of each clastic rock sample was measured, including element / oxide content data. The element / oxide content data of each sample was normalized so that the sum of all elements / oxides was 1 or 100%. Set the measured contents of n elements or oxides as: c1, c2, ..., cn; The normalized element / oxide content is as follows: ci represents the content of the i-th element or oxide.

[0008] Furthermore, in step 300, a trigger period is set, and online dynamic clustering analysis is periodically performed on the feature vectors of the entire source feature dynamic database to check whether new source clusters are naturally formed. The specific implementation method is as follows: Read the feature vector matrix of the current full sample from the dynamic database of material source characteristics; Using the CF tree saved from the previous clustering as the initial state, add new samples one by one to the CF tree according to the incremental insertion rule, and update all affected CF triples. Traverse the leaf nodes of the CF (Clustered Flow) and use its CF as the initial sub-cluster to perform a lightweight global clustering to obtain the current global cluster distribution; For each global cluster: if all its members come from the same known source tag, no action is required; If it contains one or more unlabeled samples and meets all of the following conditions, a new source subclass will be automatically generated: This unlabeled subcluster sample Where Nmin is the minimum threshold for the number of unlabeled sub-clusters; Unlabeled sub-cluster radius Where Rmax is the maximum threshold for the radius of unlabeled subclusters; The Mahalanobis distance from the center of the unlabeled sample to the center of all known source clusters is greater than the threshold Dmin; It exists stably for three or more consecutive trigger cycles.

[0009] Furthermore, in step 400, the method for initially identifying the matching provenance of clastic rocks using a multi-level quantitative model is as follows: The first module: Automatic quantitative discrimination is performed using the standard triangulation diagram of clastic rocks. The contents of quartz, feldspar and rock fragments of the sample to be tested are projected onto the corresponding source structure discrimination diagram in the dynamic database of source characteristics. The Euclidean distance between the sample to be tested and the centroid of each source region is calculated to form a subset of candidate sources. The second module: using chemical data, through robust statistics and probabilistic calculations, to find the optimal matching source from a subset of candidate sources; The third module: uses all remaining feature vectors to independently verify and enhance the conclusion, and finally outputs a comprehensive confidence score.

[0010] Furthermore, the implementation method of the first module is as follows: The content of quartz Q, feldspar F, and rock fragment L skeleton particles corresponding to samples from the same geological background in the dynamic database of provenance characteristics is used to delineate the range of quartz Q, feldspar F, and rock fragment L skeleton particle content of all samples from the same geological background to form a discrimination region representing the background, and this discrimination region is used as a provenance structure discrimination map. Calculate the average content of quartz Q, feldspar F, and rock fragment L skeletal particles in all standard samples within the discrimination region under the same geological background, and use this average content as the barycentric component (Qc, Fc, Lc). Obtain the skeletal particle content of the sample to be tested, namely quartz content Q, feldspar content F, and rock fragment content L, and normalize the three to make... ; The normalized components (Q, F, L) of the sample to be tested are projected onto the Cartesian coordinate system of the standard QFL discriminant triangulation using the following formula to obtain the projected coordinates (xs, ys) of the sample to be tested. The formula for the projected coordinates of the sample to be tested is: ; ; The centroid components (Qc, Fc, Lc) of each source geological background in the dynamic database of provenance characteristics are calculated into projected coordinates (xci, yc) using the same projection coordinate formula. The maximum effective radius Rmax of the sample area and its region are given, where i is the number of discriminant regions formed by each source geological background. The Euclidean distance di from the sample point to the centroid of each source region is calculated using the following formula: ; like This indicates that the sample falls outside all possible ranges of the discrimination area and is therefore directly excluded. Will satisfy The source categories are sorted in ascending order of distance, and the top N categories are taken to form a subset of candidate sources.

[0011] Furthermore, the implementation method of the second module is as follows: Geochemical data of standard samples from all known sources are extracted from the dynamic database of provenance characteristics. The robust mean vector and robust covariance matrix are calculated using the minimum covariance determinant estimation method. The robust covariance matrix is ​​then subjected to eigenvalue decomposition, and the first k eigenvectors are taken to form the load matrix W, thus forming a robust principal component space. The chemical element vector of the sample to be tested is centered, along with the chemical vectors of all candidate source standard samples within the candidate source subset. Multiplying the chemical element vector of the centered test sample and the chemical vectors of all candidate source standard samples within the centered candidate source subset with the loading matrix W yields the score vector of the test sample in the robust principal component space. And the score vectors of all candidate source standard samples in the robust principal component space. ; For each candidate source category c, its mean score and covariance matrix are calculated using the score distribution of its standard samples in the robust principal component space. Calculate the posterior probability of the sample to be tested for each class, select the candidate source with the highest posterior probability as the preliminary identification source of the sample to be tested, and output the probability value as the matching confidence.

[0012] Furthermore, the specific implementation method for calculating the posterior probability of the test sample for each class is as follows; ; Among them, the likelihood The prior probability is calculated from the probability density function of the multivariate normal distribution of source category c. Set it to a uniform distribution; Calculate likelihood The implementation method is as follows: Calculate the mean score within the class Take all Nc standard samples belonging to source category c from the dynamic database of source characteristics, project these samples onto the robust principal component space zci to obtain their respective score vectors, and calculate the arithmetic mean of these vectors, which is the center of the class. ; Calculate the within-class score covariance matrix Using the above Nc score vectors, calculate their values ​​around the mean. Degree of dispersion: ; Regularize the covariance matrix: ,in Let I be a positive constant and I be the identity matrix. Obtain the score vector of the sample to be tested. Then, its likelihood of belonging to class c is given by the probability density function of the multivariate normal distribution: ; Where k is the dimension of the robust principal component space; It is the determinant of the regularized covariance matrix; It is the inverse of the regularized covariance matrix; ; Where M is the total number of candidate source categories in the shortlist.

[0013] Furthermore, the implementation method of the third module is as follows: Isotope dating matching: retrieve the standard age spectrum corresponding to the best matching source category output by the second module, convert the detrital zircon U-Pb age data of the sample to be tested into a nuclear density estimation curve, calculate the similarity between the two age spectrum curves, and output an isotope age spectrum similarity score between 0 and 1. Geological logic verification of sedimentary features: The measured grain size parameters and quartz surface structure features of the sample to be tested are compared with the optimal matching source category output by the second module, and a geological logic consistency score Score2 is generated based on the comparison results. The chemical matching posterior probability output by the second module The overall confidence level C of the source material is obtained by weighting and fusing the scores from each independent component of the third module. ; Among them, weight .

[0014] Furthermore, in step 500, the preliminary identification results of the sample to be tested are subjected to constraint correction and final confirmation based on independent source geological background evidence. The specific implementation method is as follows: Obtain independent geological background evidence of the area where the sample to be tested is located, including but not limited to paleocurrent data or sedimentary facies markers; Using the independent geological background evidence as a hard constraint, the preliminary identification source tag output in step 400 is retrieved, and the metadata corresponding to the preliminary identification source tag is queried in the dynamic database of source characteristics to verify whether the preliminary identification source tag meets the hard constraint. If the preliminary identification source label meets the aforementioned hard constraints, then the preliminary identification result of step 400 is determined to pass the geological rationality verification, and the preliminary identification source label is directly output as the final source determination result. If the initially identified source label does not meet the hard constraints, the initially identified source labels are removed according to the sorting order of the candidate source subset. Then, the source labels and metadata corresponding to each candidate source in the source feature dynamic database are retrieved one by one to verify whether the candidate source meets the hard constraints. The first candidate source that passes the hard constraint condition verification is output as the final source determination result; If all candidate sources in the candidate source subset fail the hard constraint verification, they are marked as pending review.

[0015] Furthermore, after the geological consistency logic verification of the sample to be tested in step 500, when the comprehensive confidence level of the source identification of the sample to be tested is higher than the set value for entry into the database, the sample to be tested is incrementally entered into the database and added to the source feature dynamic database, and the source feature dynamic database is dynamically updated.

[0016] Compared with the prior art, the present invention has the following advantages: This invention, through a three-layer dynamic database design, achieves standardized storage and adaptive updating of multi-scale provenance features, providing a scalable standard reference system for quantitative discrimination and overcoming the technical bottleneck of traditional methods that rely on static charts and struggle to cover multiple sample types. By constructing a multi-module quantitative matching model, it progressively integrates petrological initial screening, geochemical probabilistic discrimination, and multimodal fingerprint verification, realizing a complete reasoning chain from coarse screening to fine discrimination and then to cross-validation. It not only provides unique matching results but also simultaneously outputs quantitative comprehensive confidence levels, overcoming the shortcomings of insufficient accuracy and lack of credibility representation in single-feature discrimination. Attached Figure Description

[0017] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the quantitative analysis method for clastic rock provenance provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] like Figure 1As shown, the present invention provides a quantitative analysis method for clastic rock provenance. This method is not a simple combination of algorithms and geological data, but an organic whole from data organization and hierarchical reasoning to final geological attribution. Its core technical concept lies in constructing a three-layer dynamic database containing provenance tags, feature vectors, and metadata. This database standardizes and stores the multi-scale provenance characteristics (mineral composition, elemental geochemical composition, clastic rock dating, morphological structure, and grain size parameters) of standard clastic rock samples. Then, it uses a multi-level quantitative model to achieve step-by-step matching from coarse screening to fine identification. Finally, it introduces independent geological background evidence as a hard constraint for consistency verification, thereby ensuring the rationality and reliability of the quantitative analysis results in a geological sense.

[0021] The quantitative analysis method for the source material in this embodiment includes the following steps: Step 100: Collect standard clastic rock samples from different geological backgrounds and sedimentary environments, with known provenance. The essence of provenance is the collective term for the parent rock, transport history, and sedimentary environment of all clastic particles in clastic rocks. The standard clastic rock samples provide basic data support for the subsequent construction of a standard provenance fingerprint database.

[0022] These standard samples cover a variety of geological backgrounds (such as magmatic arcs, passive continental margins, and recycle orogenic belts) and sedimentary environments (such as fluvial, deltaic, and deep-sea facies), ensuring the representativeness of the database in terms of source diversity.

[0023] Step 200: Using high-precision equipment (such as X-ray diffractometer, X-ray fluorescence spectrometer, laser ablation inductively coupled plasma mass spectrometer, scanning electron microscope, and laser particle size analyzer), determine the multi-scale provenance characteristics of each clastic rock standard sample. The multi-scale provenance characteristics include the mineral composition, elemental geochemical composition, age determination of the clastic rock (e.g., detrital zircon U-Pb age), morphological structure (e.g., mechanical impact pits and chemical dissolution textures on quartz surface), and grain size parameters (e.g., average grain size, sorting factor, skewness, and kurtosis).

[0024] It should be noted that the above-mentioned measuring equipment and its operation are all conventional technical means in this field, and the specific operating steps will not be described in detail in this invention.

[0025] Based on this, the mineral composition and elemental geochemical composition are normalized to eliminate the dimensional differences introduced by the inconsistency in the total amount of testing between different samples, so that the data of each component can participate in subsequent mathematical operations and similarity comparisons on a unified scale.

[0026] Specifically, the mineral composition of each clastic rock sample was measured to include: quartz, feldspar, rock fragments, and other minerals. The normalization method for the mineral composition measured for each clastic rock sample was as follows: The original contents of quartz, feldspar, and rock fragments were measured to be: Qr, Fr, Lr; The measured contents of quartz, feldspar, and rock fragments were normalized. The normalized contents of quartz, feldspar, and rock fragments are Q, F, and L, respectively. ; ; ; ; The elemental geochemical composition of each clastic rock sample was measured, including element / oxide content data. The element / oxide content data of each sample was normalized so that the sum of all elements / oxides was 1 or 100%. Set the measured contents of n elements or oxides as: c1, c2, ..., cn; The normalized element / oxide content is .

[0027] Through the above normalization process, the raw data obtained from different sample sources and different batches of tests are mapped into a unified standard space, laying a quantitative foundation for subsequent feature comparisons across samples and categories.

[0028] Step 300: Based on the multi-scale provenance characteristics of each clastic rock standard sample, construct a dynamic database of provenance characteristics with a three-layer structure of provenance label, feature vector, and metadata. The provenance label stores the geological background (e.g., island arc environment, active continental margin, passive continental margin) and sedimentary environment information of the clastic rock standard sample. This layer serves as a "semantic identifier" for sample attribution, providing a category basis for subsequent supervised learning and result interpretation.

[0029] The feature vector stores multi-scale provenance features in a standardized numerical matrix, including but not limited to normalized mineral component content (Q, F, L), normalized element / oxide content, detrital zircon U-Pb age data, grain size parameters, and quartz surface structure feature encoding. This layer structure is the core data source for quantitative calculation and similarity matching of the entire database.

[0030] Metadata stores background attribute information associated with provenance tags, including but not limited to sedimentary facies markers (such as cross-bedding, scour surfaces, bioturbation structures), paleocurrent data (such as measurements obtained through cross-bedding dip and gravel imbricate structures), and regional tectonic context (such as plate subduction zones, rift basins, and foreland basins).

[0031] The design of this three-layer structure realizes the separate storage and associated mapping of "semantic identifier - mathematical feature - geological background", which enables subsequent steps to call quantitative feature vectors and qualitative geological background information at the same time when performing mathematical matching, thereby upgrading the pure data-driven matching to a reasoning process constrained by geological laws.

[0032] Furthermore, in order to enable the database to adaptively update as new samples are continuously accumulated, the present invention sets a trigger period in step 300 to periodically perform online dynamic clustering analysis on the feature vectors of the entire source feature dynamic database to check whether new source clusters are naturally formed.

[0033] In step 300, a trigger period is set, and online dynamic clustering analysis is periodically performed on the feature vectors of the entire source feature dynamic database to check whether new source clusters are naturally formed. The specific implementation method is as follows: Read the feature vector matrix of the current full sample from the dynamic database of material source characteristics; Using the CF tree saved from the previous clustering as the initial state, new samples are added to the CF tree one by one according to the incremental insertion rule, and all affected CF triples are updated to include the number of samples N, the linear sum LS, and the sum of squares SS. Traverse the leaf nodes of the CF (Clustered Flow) and use its CF as the initial sub-cluster to perform a lightweight global clustering (e.g., using the subsequent clustering stage of the BIRCH algorithm) to obtain the current global cluster distribution; For each global cluster: if all its members come from the same known source tag, no action is required; If it contains one or more unlabeled samples and meets all of the following conditions, a new source subclass will be automatically generated: This unlabeled subcluster sample Where Nmin is the minimum threshold for the number of unlabeled sub-clusters; Unlabeled sub-cluster radius Where Rmax is the maximum threshold for the radius of unlabeled subclusters; The Mahalanobis distance from the center of the unlabeled sample to the center of all known source clusters is greater than the threshold Dmin; It exists stably for three or more consecutive trigger cycles.

[0034] This online dynamic clustering mechanism enables the database to automatically identify new source types as samples accumulate, maintaining the timeliness and integrity of the source classification system without manual intervention, thus providing a continuously updated standard reference system for subsequent matching.

[0035] Through a three-layer dynamic database design, standardized storage and adaptive updating of multi-scale source characteristics are achieved, providing a scalable standard reference system for quantitative discrimination and solving the technical bottleneck of traditional methods that rely on static charts and are difficult to cover multiple sample types.

[0036] Step 400: Obtain multi-scale provenance characteristics of the test sample with unknown provenance (the determination and normalization methods are the same as those for the standard sample in Step 200). Using a multi-level quantitative model, match the multi-scale provenance characteristics of the test sample with the feature vectors of the provenance characteristic dynamic database module by module to preliminarily identify the matching provenance of the test sample. This multi-level quantitative model includes at least three progressive modules: a preliminary screening module based on mineral composition, a probabilistic matching module based on elemental geochemistry, and a fusion decision module utilizing the remaining multi-scale features. The design logic of this hierarchical architecture is that it first uses the most basic petrological framework components (quartz-feldspar-rock fragment content) to quickly narrow down the candidate range, then uses a more refined elemental geochemical fingerprint for probabilistic discrimination, and finally uses geochronology, morphology, and grain size features for independent verification and enhancement, thereby realizing a complete identification chain from coarse to fine, from single-modal to multi-modal fusion of provenance.

[0037] By constructing a multi-module quantitative matching model, the petrological initial screening, geochemical probability discrimination, and multimodal fingerprint verification are progressively integrated, realizing a complete reasoning chain from coarse screening to fine discrimination and then to cross-validation. It not only provides a unique matching result, but also outputs a quantitative comprehensive confidence level at the same time, overcoming the shortcomings of insufficient accuracy and lack of credibility representation of single feature discrimination.

[0038] Specifically, the first module (preliminary screening based on composition-petrological index) refers to: using the standard triangulation diagram of clastic rocks for automatic quantitative discrimination, projecting the contents of quartz, feldspar and rock fragments of the sample to be tested onto the source structure discrimination diagram corresponding to the dynamic database of source characteristics, calculating the Euclidean distance between the sample to be tested and the centroid of each source region, and forming a subset of candidate sources; The implementation method of the first module is as follows: The contents of quartz Q, feldspar F, and rock fragment L framework particles corresponding to standard samples from the same geological background within the dynamic database of provenance characteristics are used to delineate the range of quartz Q, feldspar F, and rock fragment L framework particle contents of all standard samples from the same geological background to form a discrimination region representing that background. This discrimination region is used as a provenance structure discrimination map. The average contents of quartz Q, feldspar F, and rock fragment L framework particles of all standard samples within the discrimination region of the same geological background are calculated, and this average content is used as the centroidal component (Qc, Fc, Lc).

[0039] Obtain the skeletal particle content of the sample to be tested, namely quartz content Q, feldspar content F, and rock fragment content L, and normalize the three to make... The normalized components (Q, F, L) of the sample to be tested are projected onto the Cartesian coordinate system of the standard QFL discriminant triangulation using the following formula to obtain the projected coordinates (xs, ys) of the sample to be tested. The formula for the projected coordinates of the sample to be tested is: ; ; The centroid components (Qc, Fc, Lc) of each geological background source in the dynamic database of provenance characteristics are calculated into projected coordinates (xci, yc) using the same projection coordinate formula. The maximum effective radius Rmax of the region is determined by the maximum distance from the coordinates of all standard sample projection points to the centroid of the background, where i is the number of discrimination regions formed by each source geological background. The Euclidean distance di from the sample point to the centroid of each source region is calculated using the following formula: ; like This indicates that the sample falls outside all possible ranges of the discrimination area and is therefore directly excluded. Will satisfy The source categories are sorted in ascending order of distance, and the top N categories are taken to form a subset of candidate sources.

[0040] The direct technical effect of this module is that by transforming the QFL triangulation used in traditional geology for qualitative analysis into an automatically calculable quantitative discrimination criterion, and combining it with the actual distribution of standard samples in the database to determine the discrimination boundary of each source region, it achieves efficient initial screening of candidate sources while ensuring geological rationality, and significantly reduces the computational overhead of subsequent fine matching.

[0041] The second module utilizes chemical data to find the optimal matching source from a subset of candidate sources through robust statistics and probabilistic calculations. The specific implementation method of the second module is as follows: Geochemical data (including major and trace element / oxide contents) of standard samples from all known sources were extracted from the dynamic database of provenance characteristics.

[0042] The robust mean vector and robust covariance matrix are calculated using the minimum covariance determinant (MCD) estimation method. The robust covariance matrix is ​​then decomposed into eigenvalues, and the first k eigenvectors are used to form the load matrix W, thus creating a robust principal component space. The MCD method can effectively reduce the interference of outliers on covariance estimation, making the constructed principal component space more robust.

[0043] The chemical element vector of the test sample and the chemical vectors of all candidate source standard samples within the candidate source subset are centered. Then, the centered chemical element vector of the test sample and the centered chemical vectors of all candidate source standard samples within the candidate source subset are multiplied by the loading matrix W to obtain the score vector of the test sample in the robust principal component space. And the score vectors of all candidate source standard samples in the robust principal component space. .

[0044] For each candidate source category c, its mean score and covariance matrix are calculated using the score distribution of its standard samples in the robust principal component space.

[0045] Calculate the posterior probability of the test sample for each class. Among them, the likelihood The prior probability is calculated from the probability density function of the multivariate normal distribution of source category c. Assuming a uniform distribution, (M is the total number of candidate source categories in the shortlist).

[0046] The candidate source with the highest posterior probability is selected as the initial source for identification of the sample to be tested, and the probability value is output as the matching confidence level.

[0047] The direct technical effect of this module is that it eliminates outlier interference that may exist in the element data by using robust statistical methods, and provides a quantitative index of the uncertainty of the matching results through a Bayesian probability framework, so that the matching results not only have uniqueness but also a numerical representation of credibility.

[0048] More specifically, in the calculation of the posterior probability above, for category c, the mean score within the class is calculated. Take all Nc standard samples belonging to source category c from the dynamic database of source features, project these samples onto the robust principal component space zci to obtain their respective score vectors, and calculate the arithmetic mean of these vectors, which is the center of the class. ; Calculate the within-class score covariance matrix Using the above Nc score vectors, calculate their values ​​around the mean. Degree of dispersion: ; To ensure numerical stability, the covariance matrix is ​​regularized: , where λ is a small positive constant (e.g., 10). -6I is the identity matrix. The regularized covariance matrix ensures the positive definiteness of the matrix, which enables subsequent inversion operations and determinant calculations to be performed stably.

[0049] Obtain the score vector of the sample to be tested. Then, its likelihood of belonging to class c is given by the probability density function of the multivariate normal distribution: ; Where k is the dimension of the robust principal component space; It is the determinant of the regularized covariance matrix; It is the inverse of the regularized covariance matrix.

[0050] The third module (multimodal source fingerprint fusion decision) is implemented by independently verifying and enhancing the conclusions of the second module using all remaining feature vectors, and finally outputting a comprehensive confidence score. This module includes two parallel verification modules: Isotope chronology matching module: retrieves the standard age spectrum corresponding to the optimal matching source category output by the second module (i.e., the nuclear density estimation curve of detrital zircon U-Pb age of all standard samples under this category in the dynamic database of source characteristics), converts the detrital zircon U-Pb age data of the sample to be tested into a nuclear density estimation curve, calculates the similarity between the two age spectrum curves (e.g., through cross-correlation area or Kolmogorov-Smirnov statistic), and outputs an isotope age spectrum similarity score Score1 between 0 and 1.

[0051] Sedimentary Feature Geological Logic Verification Module: The measured grain size parameters (average grain size, sorting coefficient, skewness, kurtosis) and quartz surface structure characteristics (e.g., mechanical impact crater density, chemical dissolution degree encoding) of the sample to be tested are compared with the optimal matching source category output by the second module. Based on the comparison results (e.g., weighted scoring by the deviation of each parameter from the category mean), a geological logic consistency score Score2 is generated. The chemical matching posterior probability output by the second module The overall confidence level C of the source material is obtained by weighting and fusing the scores from each independent component of the third module. ; Among them, weight The specific values ​​of each weight can be determined through cross-validation based on the actual application scenario. For example, w1=0.5, w2=0.3, and w3=0.2.

[0052] The direct technical effect of this module is that by introducing independent evidence from the dimensions of chronology and sedimentology, it cross-validates the matching conclusions based on chemical composition from multiple perspectives, effectively overcoming the misjudgment that may be caused by a single data source, and significantly improving the reliability and geological rationality of the final conclusions.

[0053] After completing the quantitative matching of the above multi-modules, step 500 obtains the background attribute information of the area where the sample to be tested is located. This background attribute information is used as a hard constraint and its consistency is verified with the metadata background attribute information corresponding to the initially identified matching source. The unique optimal matching source is then output after verification. The essence of this step is to place the mathematical matching results, detached from the geological background, within the actual geological evolution framework for a rationality review.

[0054] By introducing independent geological background evidence as a hard constraint to verify the consistency of the matching results, the purely data-driven matching conclusions are placed within the framework of geological evolution for rationality review. This fundamentally avoids the risk of mathematical optimization results contradicting geological laws and ensures the scientific reliability of the output results.

[0055] The specific implementation method is as follows: Obtain independent geological background evidence of the region where the sample is located. Independent geological background evidence includes, but is not limited to, paleocurrent data (e.g., paleocurrent direction obtained by field measurement of cross-bedding) or sedimentary facies indicators (e.g., typical sedimentary structural assemblage reflecting a specific sedimentary environment). This evidence is independent of the multi-scale provenance characteristics of the sample itself and provides constrained information on provenance direction from different dimensions.

[0056] Using independent source geological background evidence as a hard constraint, the preliminary source identification label output in step 400 is retrieved, and the metadata corresponding to the preliminary source identification label (such as the spatial orientation of the known source area relative to the basin and the sedimentary facies configuration relationship) is queried in the source feature dynamic database to verify whether the preliminary source identification label meets the hard constraint.

[0057] If the preliminary identification source label meets the hard constraints, the preliminary identification result of step 400 is determined to pass the geological rationality verification, and the preliminary identification source label is directly output as the final source determination result.

[0058] If the initially identified source label does not meet the hard constraints, the initially identified source labels are removed according to the sorting order of the candidate source subset. Then, the source label and metadata corresponding to each candidate source in the source feature dynamic database are retrieved one by one to verify whether the candidate source meets the hard constraints. The first candidate source that passes the hard constraint verification is output as the final source determination result.

[0059] If all candidate sources in the candidate source subset fail the hard constraint verification, they are marked as pending review.

[0060] The direct technical effect of this step is that it transforms the mathematical "optimal matching" into the geological "reasonable matching", avoiding the problem that pure mathematical optimization may lead to geologically impossible conclusions, and making the final output source determination result both statistically optimal and geologically reasonable.

[0061] Based on this, when the overall confidence level C of the sample to be tested after the geological consistency logic verification in step 500 is higher than the preset entry value (e.g., C>0.85), the sample to be tested is triggered to be incrementally added to the source characteristic dynamic database as a new standard sample, thereby realizing the dynamic updating and self-enhancement of the database. As the number of samples added to the database continues to accumulate, the source types covered by the database become richer, the statistical distribution of feature vectors of each category becomes more stable, and the matching accuracy and confidence of subsequent samples will be further improved, forming a positive feedback loop of "accumulation-matching-verification-re-accumulation". Through the incremental entry mechanism, the database and analysis method form a self-enhancing closed-loop system. With the accumulation of samples, the discrimination accuracy and coverage continue to improve, exhibiting good scalability and adaptability.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. All such modifications or substitutions should be covered within the protection scope of this application, and should not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method for quantitative analysis of clastic rock provenance, characterized in that, Includes the following steps: Step 100: Collect standard clastic rock samples from different geological backgrounds and sedimentary environments, with known provenance. Step 200: Determine the multi-scale provenance characteristics of each clastic rock standard sample. The multi-scale provenance characteristics include the mineral composition, elemental geochemical composition, age determination, morphology and structure, and grain size parameters of the clastic rock standard sample. Normalize the analysis results of the mineral composition and elemental geochemical composition of each clastic rock standard sample. Step 300: Based on the multi-scale provenance characteristics of each clastic rock standard sample, construct a dynamic database of provenance characteristics with a three-layer structure of provenance tag, feature vector, and metadata. The provenance tag stores the geological background and sedimentary environment information of the clastic rock standard sample. The feature vector stores the multi-scale provenance characteristics in a standardized numerical matrix. The metadata stores the background attribute information associated with the provenance tag. The background attribute information includes, but is not limited to, sedimentary facies markers, paleocurrent data, and regional tectonic background. Step 400: Obtain the multi-scale source characteristics of the test sample with unknown source. Using a multi-level quantitative model, match the multi-scale source characteristics of the test sample with the feature vectors of the source characteristic dynamic database module by module to preliminarily identify the matching source of the test sample. The multi-level quantitative model includes at least: a preliminary screening module based on mineral composition, a probability matching module based on elemental geochemistry, and a fusion decision module using the remaining multi-scale features. Step 500: Obtain the background attribute information of the area where the sample to be tested is located. Use the background attribute information as a hard constraint and verify its consistency with the metadata background attribute information corresponding to the initially identified matching source. After verification, output the unique optimal matching source.

2. The method for quantitative analysis of clastic rock provenance according to claim 1, characterized in that, In step 200, high-precision equipment is used to analyze the mineral composition and elemental geochemical composition; The mineral composition of each clastic rock sample was measured to include quartz, feldspar, rock fragments, and other minerals. The normalization method for the mineral composition of each clastic rock sample was as follows: The original contents of quartz, feldspar, and rock fragments were measured to be: Qr, Fr, Lr; The measured contents of quartz, feldspar, and rock fragments were normalized. The normalized contents of quartz, feldspar, and rock fragments are Q, F, and L, respectively. ; ； ； ； The elemental geochemical composition of each clastic rock sample was measured, including element / oxide content data. The element / oxide content data of each sample was normalized so that the sum of all elements / oxides was 1 or 100%. Set the measured contents of n elements or oxides as: c1, c2, ..., cn; The normalized element / oxide content is as follows: ci represents the content of the i-th element or oxide.

3. The method for quantitative analysis of clastic rock provenance according to claim 1, characterized in that, In step 300, a trigger period is set, and online dynamic clustering analysis is periodically performed on the feature vectors of the entire source feature dynamic database to check whether new source clusters are naturally formed. The specific implementation method is as follows: Read the feature vector matrix of the current full sample from the dynamic database of material source characteristics; Using the CF tree saved from the previous clustering as the initial state, add new samples one by one to the CF tree according to the incremental insertion rule, and update all affected CF triples. Traverse the leaf nodes of the CF (Clustered Flow) and use its CF as the initial sub-cluster to perform a lightweight global clustering to obtain the current global cluster distribution; For each global cluster, if all its members come from the same known source tag, no action is required; If it contains one or more unlabeled samples and meets all of the following conditions, a new source subclass will be automatically generated: The unlabeled sample Where Nmin is the minimum threshold for the number of unlabeled samples; Unlabeled sample radius Where Rmax is the maximum threshold for the radius of unlabeled samples; The Mahalanobis distance from the center of the unlabeled sample to the center of all known source clusters is greater than the threshold Dmin; It exists stably for three or more consecutive trigger cycles.

4. The method for quantitative analysis of clastic rock provenance according to claim 1, characterized in that, In step 400, the method for initially identifying the matching provenance of clastic rocks using a multi-level quantitative model is as follows: The first module: Automatic quantitative discrimination is performed using the standard triangulation diagram of clastic rocks. The contents of quartz, feldspar and rock fragments of the sample to be tested are projected onto the corresponding source structure discrimination diagram in the dynamic database of source characteristics. The Euclidean distance between the sample to be tested and the centroid of each source region is calculated to form a subset of candidate sources. The second module: using chemical data to find the optimal matching source from the subset of candidate sources through robust statistics and probabilistic calculations; The third module: uses all remaining feature vectors to independently verify and enhance the conclusion, and finally outputs a comprehensive confidence score.

5. The method for quantitative analysis of clastic rock provenance according to claim 4, characterized in that, The implementation method of the first module is as follows: The content of quartz Q, feldspar F, and rock fragment L skeleton particles corresponding to samples from the same geological background in the dynamic database of provenance characteristics is used to delineate the range of quartz Q, feldspar F, and rock fragment L skeleton particle content of all samples from the same geological background to form a discrimination region representing the background, and this discrimination region is used as a provenance structure discrimination map. Calculate the average content of quartz Q, feldspar F, and rock fragment L skeletal particles in all standard samples within the discrimination region under the same geological background, and use this average content as the barycentric component (Qc, Fc, Lc). Obtain the skeletal particle content of the sample to be tested, namely quartz content Q, feldspar content F, and rock fragment content L, and normalize the three to make... ; The normalized components (Q, F, L) of the sample to be tested are projected onto the Cartesian coordinate system of the standard QFL discriminant triangulation using the following formula to obtain the projected coordinates (xs, ys) of the sample to be tested. The formula for the projected coordinates of the sample to be tested is: ； ； The centroid components (Qc, Fc, Lc) of each source geological background in the dynamic database of provenance characteristics are calculated into projected coordinates (xci, yc) using the same projection coordinate formula. The maximum effective radius Rmax of the sample area and its region are given, where i is the number of discriminant regions formed by each source geological background. The Euclidean distance di from the sample point to the centroid of each source region is calculated using the following formula: ； like This indicates that the sample falls outside all possible ranges of the discrimination area and is therefore directly excluded. Will satisfy The source categories are sorted in ascending order of distance, and the top N categories are taken to form a subset of candidate sources.

6. The method for quantitative analysis of clastic rock provenance according to claim 4, characterized in that, The implementation method of the second module is as follows: Geochemical data of standard samples from all known sources are extracted from the dynamic database of provenance characteristics. The robust mean vector and robust covariance matrix are calculated using the minimum covariance determinant estimation method. The robust covariance matrix is ​​then subjected to eigenvalue decomposition, and the first k eigenvectors are taken to form the load matrix W, thus forming a robust principal component space. The chemical element vector of the sample to be tested is centered, along with the chemical vectors of all candidate source standard samples within the candidate source subset. Multiplying the chemical element vector of the centered test sample and the chemical vectors of all candidate source standard samples within the centered candidate source subset with the loading matrix W yields the score vector of the test sample in the robust principal component space. And the score vectors of all candidate source standard samples in the robust principal component space. ; For each candidate source category c, its mean score and covariance matrix are calculated using the score distribution of its standard samples in the robust principal component space. Calculate the posterior probability of the sample to be tested for each class, select the candidate source with the highest posterior probability as the preliminary identification source of the sample to be tested, and output the probability value as the matching confidence.

7. The method for quantitative analysis of clastic rock provenance according to claim 6, characterized in that, The specific implementation method for calculating the posterior probability of the sample to be tested for each class is as follows: ； Among them, the likelihood The prior probability is calculated from the probability density function of the multivariate normal distribution of source category c. Set it to a uniform distribution; Calculate the likelihood The implementation method is as follows: Calculate the mean score within the class Take all Nc standard samples belonging to source category c from the dynamic database of source characteristics, project these samples onto the robust principal component space zci to obtain their respective score vectors, and calculate the arithmetic mean of these vectors, which is the center of the class. ； Calculate the within-class score covariance matrix Using the above Nc score vectors, calculate their values ​​around the mean. Degree of dispersion: ； Regularize the covariance matrix: , Let I be a positive constant and I be the identity matrix. Obtain the score vector of the sample to be tested. Then, its likelihood of belonging to class c is given by the probability density function of the multivariate normal distribution: ； Where k is the dimension of the robust principal component space; It is the determinant of the regularized covariance matrix; It is the inverse of the regularized covariance matrix; ； Where M is the total number of candidate source categories in the shortlist.

8. The method for quantitative analysis of clastic rock provenance according to claim 7, characterized in that, The implementation method of the third module is as follows: Isotope dating matching: retrieve the standard age spectrum corresponding to the best matching source category output by the second module, convert the detrital zircon U-Pb age data of the sample to be tested into a nuclear density estimation curve, calculate the similarity between the two age spectrum curves, and output an isotope age spectrum similarity score between 0 and 1. Geological logic verification of sedimentary features: The measured grain size parameters and quartz surface structure features of the sample to be tested are compared with the optimal matching source category output by the second module, and a geological logic consistency score Score2 is generated based on the comparison results. The chemical matching posterior probability output by the second module The overall confidence level C of the source material is obtained by weighting and fusing the scores from each independent component of the third module. ； Among them, weight .

9. The method for quantitative analysis of clastic rock provenance according to claim 4, characterized in that, In step 500, the preliminary identification results of the sample to be tested are constrained, corrected, and finally confirmed based on independent geological background evidence. The specific implementation method is as follows: Obtain independent geological background evidence of the area where the sample to be tested is located, including but not limited to paleocurrent data or sedimentary facies markers; Using the independent geological background evidence as a hard constraint, the preliminary identification source tag output in step 400 is retrieved, and the metadata corresponding to the preliminary identification source tag is queried in the dynamic database of source characteristics to verify whether the preliminary identification source tag meets the hard constraint. If the preliminary identification source label meets the aforementioned hard constraints, then the preliminary identification result of step 400 is determined to pass the geological rationality verification, and the preliminary identification source label is directly output as the final source determination result. If the initially identified source label does not meet the hard constraints, the initially identified source labels are removed according to the sorting order of the candidate source subset. Then, the source labels and metadata corresponding to each candidate source in the source feature dynamic database are retrieved one by one to verify whether the candidate source meets the hard constraints. The first candidate source that passes the hard constraint condition verification is output as the final source determination result; If all candidate sources in the candidate source subset fail the hard constraint verification, they are marked as pending review.

10. The method for quantitative analysis of clastic rock provenance according to claim 9, characterized in that, After the geological consistency logic verification of the sample to be tested in step 500, when the comprehensive confidence level of the source identification of the sample to be tested is higher than the set value for entry into the database, the sample to be tested is incrementally entered into the database and added to the source feature dynamic database, and the source feature dynamic database is dynamically updated.

Images