Mineral resource potential prediction method and system based on artificial intelligence

Abstract

The invention discloses a mineral resource potential prediction method and system based on artificial intelligence, and relates to the technical field of mine field resource prediction.The method comprises the steps that multi-source mineral data is collected, standardized processing and vectorization conversion are conducted, and a basic data set is generated; based on the basic data set, extracting spatial features and associated features of the basic data set, constructing a mineral resource network diagram and identifying a resource center node; based on the mineral resource network diagram and the resource center node, dividing same-level resource units by adopting an intelligent algorithm, and calculating basic scores of the units; constructing a mineral resource prediction model, and performing potential scoring and grade judgment on a target area after training and cross validation; and generating a resource potential prediction map, and outputting a prediction result. According to the method, a full-process intelligent mineral resource prediction system is constructed, accurate and efficient prediction of mineral resource potential is realized through multi-dimensional data integration, feature extraction and model training, and the problems of high subjectivity, poor adaptability, insufficient precision and the like of a traditional method are solved.

Description

Technical Field

[0001] This invention relates to the field of mineral resource prediction technology, and more specifically to a method and system for predicting mineral resource potential based on artificial intelligence. Background Technology

[0002] With the steady progress of the global economy, the demand for mineral resources continues to rise, especially in key sectors such as energy, construction, and electronics, where reliance on mineral resources is deepening. The development and utilization of mineral resources is one of the core drivers of economic development, and scientific and rational mineral resource assessment and potential prediction are crucial foundations for the sustainable development of the mining industry. However, traditional methods for predicting mineral resource potential face many prominent problems in practical applications, specifically in the following aspects: Traditional mineral resource assessment methods largely rely on the experience and judgment of geological experts. This approach is not only highly susceptible to subjective interference, but also prone to bias in assessment results when dealing with complex and ever-changing geological environments, making it difficult to accurately reflect the actual potential of mineral resources. Geological conditions themselves are highly complex and uncertain. The distribution patterns of ore bodies are influenced by multiple factors, including tectonic movements, erosion, and sedimentary environments. Traditional methods often lack sufficient flexibility and adaptability when dealing with such complex scenarios. Although information technology has evolved rapidly in recent years, many enterprises and institutions in the field of mineral resource exploration still need to improve their informatization levels and have failed to fully utilize cutting-edge technologies such as big data and artificial intelligence for resource potential prediction. Currently, most mainstream mineral resource potential prediction models are based on linear or simplified models, which are unable to fully capture the nonlinear correlations and complex interactions between data, resulting in prediction accuracy that fails to meet the needs of practical applications.

[0003] Therefore, in view of the shortcomings of existing technologies, how to provide a method and system for predicting mineral resource potential based on artificial intelligence is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, the present invention provides a method and system for predicting mineral resource potential based on artificial intelligence, constructs an intelligent mineral resource prediction system for the entire process, and achieves accurate and efficient prediction of mineral resource potential through multi-dimensional data integration, feature extraction and model training, solving the problems of strong subjectivity, poor adaptability and insufficient accuracy of traditional methods.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting mineral resource potential based on artificial intelligence, comprising: Collect multi-source mineral data, perform standardization and vectorization transformation to generate a basic dataset; Based on the aforementioned basic dataset, spatial and correlation features of the basic dataset are extracted to construct a mineral resource network map and identify resource center nodes. Based on the mineral resource network diagram and resource center nodes, an intelligent algorithm is used to divide resource units at the same level and calculate the basic score of each unit. A mineral resource prediction model is constructed, and after training and cross-validation, the potential of the target area is scored and its level is determined. Generate a resource potential prediction map and output the prediction results.

[0006] Preferably, the spatial features and association features of the basic dataset are extracted, including: GIS spatial analysis technology is used to calculate the spatial topological relationships, distance correlations, and density distribution characteristics of geological units, and the spatial features of the basic dataset are extracted using the sliding window method. The correlation between different mineral indicators is calculated using mutual information entropy, and the associated features are obtained by combining the random forest algorithm.

[0007] Preferably, constructing a mineral resource network map and identifying resource center nodes includes: Using geological units in the basic dataset as nodes and feature correlation coefficients as edge weights, an undirected weighted mineral resource network graph is constructed using a graph neural network. The PageRank algorithm is used to calculate the importance score of each node, and the resource center node is obtained by combining degree centrality and betweenness centrality.

[0008] Preferably, an intelligent algorithm is used to divide resource units at the same level and calculate the basic score of the unit, including: Simulated annealing-genetic algorithm is used to optimize the initial cluster centers of quantum particle swarm optimization, and fuzzy C-means clustering is combined to divide the same level resource units; The weights of each feature are determined based on the entropy weight method. A scoring system is constructed from four dimensions: geological condition adaptability, element anomaly intensity, spatial correlation, and exploration degree. The basic score of the unit is calculated using a weighted summation formula.

[0009] Preferably, the simulated annealing-genetic algorithm optimization process uses real number encoding, takes minimizing intra-class scatter as the fitness function, and improves clustering stability through crossover and mutation operations and simulated annealing acceptance criteria.

[0010] Preferably, the simulated annealing-genetic algorithm optimization process uses real-number encoding, with the fitness function being the minimization of intra-class scatter. It improves clustering stability through crossover and mutation operations and the simulated annealing acceptance criterion, specifically including: For the feature dimension m and the preset number of cluster centers c of mineral resource data, chromosomes are constructed using real number encoding, with each chromosome corresponding to a set of initial cluster centers; Randomly generate the initial population, and simultaneously initialize the membership matrix U and the initial cluster center matrix V; For each chromosome in the population, the cluster center matrix V is decoded, the intra-cluster scatter of each cluster is calculated, and the individual fitness value is obtained by substituting it into the fitness function. Using the arithmetic crossover method, two parent individuals P1 and P2 are randomly selected to generate two offspring individuals; Perform mutation operations on each gene locus of each individual; Set the initial temperature T0, the cooling coefficient β, and the minimum temperature threshold T. min Calculate the fitness difference between offspring and parents; Output the individual with the highest fitness value, and decode to obtain the optimized initial cluster center.

[0011] Preferably, fuzzy C-means clustering is used to divide resource units at the same level, specifically including: Step 1: Based on the optimized QPSO initial cluster centers k=1,2,...,c, set the fuzzy coefficient h and the upper limit of the number of iterations K. max Convergence threshold ε; Step 2: Calculate the initial membership degree of each geological unit sample to the kth cluster according to the FCM membership formula. ; Step 3: Based on the initial membership matrix, calculate the cluster centers for the first iteration according to the FCM cluster center update formula. ; ; The cluster centers are the weighted average of the samples, with the weights being h raised to the power of the membership degree; Step 4: Substituting into the membership formula and performing iterative optimization, we obtain... And calculate the change in cluster centers; If the change in cluster centers is less than the convergence threshold, then the convergence condition has been met and the iteration stops. If it does not converge and the number of iterations has not reached K max Then let = , = Repeat steps 3-4; if the number of iterations reaches K... max If convergence is still not achieved, force a stop and output the current cluster centers; Step 5: Based on the membership matrix obtained through iteration, assign each sample to the cluster with the highest membership degree. Each cluster corresponds to a peer resource unit. Calculate the number of samples, spatial range, and feature mean of each unit to complete the peer resource unit division.

[0012] Preferably, an artificial intelligence-based mineral resource potential prediction system includes: The data acquisition and preprocessing module is used to collect multi-source mineral data, perform standardization and vectorization transformation, and generate a basic dataset. The feature extraction module is used to extract the spatial and correlation features of the basic dataset based on the basic dataset, construct a mineral resource network map and identify resource center nodes; The intelligent partitioning module is used to partition resource units of the same level based on the mineral resource network map and resource center nodes, and calculate the basic score of the unit. The mineral resource prediction module is used to build a mineral resource prediction model. After training and cross-validation, it scores the potential and determines the grade of the target area. The results output module is used to generate a resource potential prediction map and output the prediction results.

[0013] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for predicting mineral resource potential based on artificial intelligence, constructs an intelligent mineral resource prediction system for the entire process, and achieves accurate and efficient prediction of mineral resource potential through multi-dimensional data integration, feature extraction and model training, solving the problems of strong subjectivity, poor adaptability and insufficient accuracy of traditional methods. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of a mineral resource potential prediction method based on artificial intelligence provided by the present invention.

[0016] Figure 2 This is a schematic diagram of the structure of an artificial intelligence-based mineral resource potential prediction system provided by the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention discloses an artificial intelligence-based method for predicting mineral resource potential, such as... Figure 1 As shown, it includes: Collect multi-source mineral data, perform standardization and vectorization transformation to generate a basic dataset; Based on the aforementioned basic dataset, spatial and correlation features of the basic dataset are extracted to construct a mineral resource network map and identify resource center nodes. Based on the mineral resource network diagram and resource center nodes, an intelligent algorithm is used to divide resource units at the same level and calculate the basic score of each unit. A mineral resource prediction model is constructed, and after training and cross-validation, the potential of the target area is scored and its level is determined. Generate a resource potential prediction map and output the prediction results.

[0019] Specifically, multi-source mineral data is collected, standardized, and vectorized to generate a basic dataset. This dataset includes geological structural data, geochemical element content data, remote sensing image interpretation data, mineral exploration engineering data, and historical mineral development data. Standardization uses Z-score normalization to eliminate differences in data dimensions, interpolation based on the K-nearest neighbor algorithm to fill in missing values, and box plots to identify outliers and correct them using neighborhood means. Vectorization directly retains the original feature vectors for numerical data, uses one-hot encoding for categorical data, and generates coordinate embedding vectors through grid partitioning for spatial data. Finally, these are integrated to form a dimensionally unified basic dataset.

[0020] Specifically, the spatial and association features of the base dataset are extracted, including: GIS spatial analysis technology is used to calculate the spatial topological relationships, distance correlations, and density distribution characteristics of geological units, and the spatial features of the basic dataset are extracted using the sliding window method. The correlation between different mineral-related indicators is calculated using mutual information entropy, and key correlation features are obtained by combining the random forest algorithm to remove redundant information.

[0021] Specifically, constructing a mineral resource network map and identifying resource center nodes includes: Using geological units in the basic dataset as nodes and feature correlation coefficients as edge weights, an undirected weighted mineral resource network graph is constructed using a graph neural network (GNN). The PageRank algorithm is used to calculate the importance score of each node. The degree centrality and betweenness centrality are combined to make a comprehensive judgment, and the top 20% of the nodes with the highest scores are selected as the resource center nodes.

[0022] The edge construction rule is as follows: when the feature correlation coefficient between two geological units is greater than 0.5, an edge is established, and the weight is the correlation coefficient; the PageRank algorithm is used to calculate the importance score of each node, as shown in the following formula: ; Where d is the damping coefficient; B(u) is the set of nodes pointing to node u; L(v) is the set of outgoing edges of node v. The degree centrality (number of edges connected to a node) and betweenness centrality (the proportion of nodes located on all shortest paths) of each node are also calculated. A weighted summation method is used to calculate the overall score of the nodes, and the top 20% of nodes are selected as the resource center nodes, i.e., the core units with the highest probability of mineral resource enrichment.

[0023] Specifically, intelligent algorithms are used to divide resource units at the same level and calculate the basic score of each unit, including: Simulated Annealing-Genetic Algorithm (SA-GA) is used to optimize the initial cluster centers of Quantum Particle Swarm Optimization (QPSO), and Fuzzy C-means Clustering (FCM) is used to divide the same level resource units. The FCM algorithm clusters the geological units in the basic dataset to obtain the same level resource units with high feature similarity. The weights of each feature are determined based on the entropy weight method. A scoring system is constructed from four dimensions: geological condition adaptability, element anomaly intensity, spatial correlation, and exploration degree. The basic score of the unit is calculated using a weighted summation formula.

[0024] The weights of each feature are determined based on the entropy weight method, including: Step 1. Feature standardization processing; Step 2. Calculate the entropy value of the j-th feature: ; Where, p ij The proportion of the j-th feature in the i-th sample; Step 3. Calculate the weights: .

[0025] A scoring system is constructed based on four dimensions: geological condition suitability (e.g., the matching degree between stratigraphy and mineralization, and the degree of fault zone development), elemental anomaly intensity (e.g., the anomaly coefficient of ore-forming elements and the number of concentration centers), spatial correlation (e.g., the distance to resource center nodes and the correlation with known mineral deposits), and exploration degree (e.g., borehole density and geophysical exploration coverage). A weighted summation formula is used to calculate the basic score of each unit (score range [0,1]). ; Where, x j Let be the standardized score of the j-th indicator.

[0026] Specifically, the simulated annealing-genetic algorithm optimization process uses real number encoding, takes minimizing intra-class scatter as the fitness function, and improves clustering stability through crossover and mutation operations and simulated annealing acceptance criteria.

[0027] Wherein, the intra-class scatter is the sum of squared Euclidean distances from all samples within each resource unit to the cluster center of that unit, and the fitness function is the reciprocal of the intra-class scatter: ; Among them, C k For the k-th cluster, c k It is the center of the k-th cluster.

[0028] Specifically, the simulated annealing-genetic algorithm optimization process uses real-number encoding, with the fitness function being the minimization of intra-class scatter. It improves clustering stability through crossover and mutation operations and the simulated annealing acceptance criterion, specifically including: Step 1: Population Initialization and Encoding: Based on the feature dimension m of the mineral resource data and the preset number of cluster centers c, a chromosome is constructed using real-number encoding. Each chromosome corresponds to a set of initial QPSO cluster centers. The chromosome length is c×m, meaning the m feature values ​​of each cluster center are arranged sequentially. For example, cluster center c1=(x 11 ,x 12 ,...,x 1m c2=(x) 21 ,x 22 ,...,x 2m ), chromosome [x 11 ,x 12 ,...,x 1m ,x 21 ,x 22 ,...,x 2m ,...,x c1 ,x c2 ,...,x cm ]); The initial population size is set to 50-100. The initial population is randomly generated, with each gene value randomly selected within the range of its corresponding feature. Simultaneously, the membership matrix U and the initial cluster center matrix V are initialized; the elements u of the membership matrix U... ik ∈[0,1], representing the membership degree of the i-th sample to the k-th cluster, and the initial cluster center matrix V is obtained by chromosome decoding; Step 2: Fitness Calculation: For each chromosome in the population, decode to obtain the cluster center matrix V, calculate the intra-cluster scatter of each cluster, and substitute it into the fitness function to obtain the individual fitness value; the higher the fitness value, the better the cluster center corresponding to the individual and the higher the intra-cluster sample similarity. Step 3: Genetic manipulation: Crossover operation: The arithmetic crossover method is adopted. Two parent individuals P1 and P2 are randomly selected to generate two offspring individuals O1 = αP1+(1 - α)P2 and O2=(1 - α)P1+αP2, where α is a random number between [0, 1]; the crossover probability is set to 0.6 - 0.8 to ensure population diversity; Mutation operation: Mutation operation is performed on each gene locus of each individual; random perturbation is carried out with a mutation probability of 0.01 - 0.05. If the original value of the gene locus is x, the mutated value is x + δ (δ is a random number between [-0.1x, 0.1x]), and it is ensured that the mutated gene value is still within the range of characteristic values; Step 4: Simulated annealing operation: Set the initial temperature T0, the cooling coefficient β, and the lowest temperature threshold T min , calculate the fitness difference between the offspring individual and the parent individual; Δf = f 子代 -f 父代 ; If Δf>0 (the offspring is better), directly accept the offspring; if Δf≤0 (the offspring is worse), accept the offspring with the Metropolis probability P = exp(Δf / T) (randomly generate a number r between [0, 1], if r < P, accept it, otherwise retain the parent); Step 5: Repeat steps 6.2 - 6.4, and reduce the temperature (T = β×T) every iteration until the temperature is lower than T min or the number of iterations reaches the preset number; output the individual with the highest fitness value, and decode to obtain the optimized initial clustering center of QPSO.

[0029] Specifically, combine fuzzy C - means clustering (FCM) to divide the same - level resource units, which specifically includes: Step 1: Parameter initialization: Based on the optimized initial clustering center of QPSO , k = 1, 2,..., c, set the fuzzy coefficient h, the upper limit of the number of iterations K max , and the convergence threshold ε; Among them, h = 2, which is used to control the clustering fuzziness. Taking 2 in the mineral resource scenario can balance the fuzziness and discrimination; the upper limit of the number of iterations K max = 100, and the convergence threshold ε = 1e - 5. The convergence threshold is the threshold of the change amount of the clustering center; Step 2: Calculate the initial membership degree of each geological unit sample x i (i = 1, 2,..., n) belonging to the k - th clustering cluster, and the formula is: ; Among them, is the distance from the sample x i to the initial clustering center The Euclidean distance is satisfied, and the constraints are met. , indicating that the sum of the membership degrees of each sample to all clusters is 1; Step 3: Based on the initial membership matrix, calculate the cluster centers for the first iteration according to the FCM cluster center update formula. ; ; The cluster centers are the weighted average of the samples, with the weights being h raised to the power of the membership degree; Step 4: Calculate the new membership matrix: Substituting into the membership formula and performing iterative optimization, we obtain... The change in cluster centers is calculated, and the calculation process for the change in cluster centers is as follows: .

[0030] If the change in cluster centers is less than the convergence threshold, then the convergence condition has been met and the iteration stops. If it does not converge and the number of iterations has not reached K max Then let = , = Repeat steps 3-4; if the number of iterations reaches K... max If convergence is still not achieved, force a stop and output the current cluster centers; Step 5: Determine peer resource units: based on the membership matrix u obtained through iteration. ik , for each sample x i Assign to the cluster with the highest membership degree, i.e., k'=argmax k u ik Each cluster corresponds to a peer resource unit. The sample size, spatial range, and mean features of each unit are counted to complete the peer resource unit division.

[0031] In one specific embodiment of the present invention, an artificial intelligence-based mineral resource potential prediction system, such as... Figure 2 As shown, it includes: The data acquisition and preprocessing module is used to collect multi-source mineral data, perform standardization and vectorization transformation, and generate a basic dataset. The feature extraction module is used to extract the spatial and correlation features of the basic dataset based on the basic dataset, construct a mineral resource network map and identify resource center nodes; The intelligent partitioning module is used to partition resource units of the same level based on the mineral resource network map and resource center nodes, and calculate the basic score of the unit. The mineral resource prediction module is used to build a mineral resource prediction model. After training and cross-validation, it scores the potential and determines the grade of the target area. The results output module is used to generate a resource potential prediction map and output the prediction results.

[0032] In a specific embodiment of the present invention, applied to the prediction of mineral resource potential in a mining area, the specific steps are as follows: S100. Collect multi-source mineral data, perform standardization and vectorization transformation to generate a basic dataset. The multi-source mineral data includes geological structure data, geochemical data, remote sensing image data, and exploration engineering data. Historical data is also developed, including statistical data such as the mining years of known mineral deposits, ore grade, and resource reserves.

[0033] Z-score normalization was used to process all numerical data to eliminate dimensional differences in indicators such as element content and resistivity; missing values ​​were filled by interpolation using the K-nearest neighbor algorithm; outliers were identified by box plot method, and outliers in element content were screened out and corrected by the mean of a 3×3 window in the neighborhood.

[0034] Vectorization transformation includes: for numerical data (element content, resistivity, etc.), the original feature vector is directly retained with a dimension of 12; for categorical data (strata lithology, mineralization type, etc.), one-hot encoding transformation is used to generate an 8-dimensional binary feature vector; for spatial data, a 1km×1km grid is used to generate a 3-dimensional coordinate embedding vector containing latitude, longitude, and altitude, which is finally integrated into a 23-dimensional unified basic dataset.

[0035] S200. Based on the basic dataset, extract the spatial features and correlation features of the basic dataset, construct a mineral resource network map and identify resource center nodes.

[0036] S210. Spatial feature extraction includes using ArcGIS spatial analysis tools to calculate the spatial topological relationship (distance from the fault zone, whether it is located in the magmatic rock contact zone), distance correlation (Euclidean distance from known mineral deposits), and density distribution characteristics (fault line density, sampling point density) of each grid cell. Spatial neighborhood features are extracted using the 3×3 sliding window method to generate a 10-dimensional spatial feature set.

[0037] The correlation feature extraction process involves calculating the correlation between 15 elements using mutual information entropy, combining a random forest algorithm (100 decision trees) to screen key correlation features, removing redundant information, and retaining 20 core features such as stratigraphic lithology matching degree, fault development density, Pb / Zn anomaly coefficient, and hydroxyl anomaly intensity.

[0038] S220. Mineral resource network graph construction: Using 1km×1km grid cells as nodes (920 nodes in total), and the correlation coefficient of the core features as the edge weights, an undirected weighted graph is constructed using a graph convolutional neural network (GCN); when the feature correlation coefficient of two nodes is >0.5, an edge connection is established, generating a total of 1286 effective edges.

[0039] Central node identification: The PageRank algorithm is used to calculate the node importance score (damping coefficient d=0.85). The associativity centrality and betweenness centrality are weighted and summed in a 1:1:1 ratio. The top 20% of nodes (184) are selected as resource central nodes, mainly concentrated in the main fault intersection zone, the contact zone between igneous rocks and carbonate rocks, and the area around known mineral deposits.

[0040] S300: Based on the mineral resource network diagram and resource center nodes, an intelligent algorithm is used to divide resource units at the same level and calculate the basic score of each unit.

[0041] S310 and SA-GA optimize the initial cluster centers of QPSO.

[0042] The specific parameter configuration is as follows: the population size is set to 80, the number of preset cluster centers is c=8, the population is divided into 8 resource units of the same level, and the chromosome length is 8×20=160 (20 core features).

[0043] The genetic operations are as follows: crossover probability 0.7, using arithmetic crossover (α randomly takes values ​​of [0,1]); mutation probability 0.03, gene mutation perturbation range [-0.1x, 0.1x].

[0044] The simulated annealing parameters are as follows: initial temperature T0 = 100, cooling coefficient β = 0.95, and minimum temperature T... min =1, with an upper limit of 200 iterations, and the final output is the optimized cluster center with the highest fitness (smallest intra-class dispersion).

[0045] S320 and FCM clustering are used to divide resource units at the same level.

[0046] The parameter initialization is as follows: fuzzy coefficient h=2, upper limit of iteration number K. max =100, convergence threshold ε=1e-5, initialize membership matrix U based on optimized QPSO cluster centers.

[0047] The iterations were performed using the FCM membership formula and the cluster center update formula. In the 32nd iteration, the change in cluster centers was less than 1e-5, which met the convergence condition, and the iteration stopped.

[0048] Each grid cell is assigned to the cluster with the highest membership degree, resulting in 8 resource cells of the same level. The number of samples in each cell ranges from 95 to 120, and they are spatially concentrated and contiguous.

[0049] S330, Calculation of Basic Unit Score

[0050] After standardizing the 20 core features, the entropy value of each feature is calculated, and the weights of the four dimensions are finally obtained. The basic score of each unit is calculated using a weighted summation formula.

[0051] S400. Construct a mineral resource prediction model, and after training and cross-validation, score the potential and determine the grade of the target area.

[0052] Specifically, a CNN-LSTM hybrid model is adopted (convolutional layers extract local features, and LSTM layers capture spatial and temporal correlations). The 20 core features of the basic dataset are used as input, and the resource potential level (level 1-5) of the known mining sites is used as the label. The training set and test set are divided in a 7:3 ratio.

[0053] The model parameters were optimized using 5-fold cross-validation, and the final values ​​were determined to be 32 convolutional kernels, 64 LSTM hidden layer dimensions, and a learning rate of 0.001. The model achieved an accuracy of 91.2% on the test set.

[0054] The feature data of 8 resource units at the same level are input into the trained model to obtain the potential score of each unit, and the levels are divided according to the threshold: high potential, medium potential, and low potential.

[0055] S500 generates a resource potential prediction map and outputs the prediction results.

[0056] Specifically, ArcGIS was used to draw a potential level distribution map of the mining area in a 1km×1km grid, marking the spatial range of high-potential units, the location of resource center nodes, and the distribution of known mineral deposits.

[0057] The prediction results include a basic score table for each unit, a potential level statistics table, a list of high-potential target area coordinates, and a suggestion for exploration priority (high-potential units should prioritize borehole verification, and medium-potential units should carry out in-depth geochemical exploration).

[0058] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0059] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting mineral resource potential based on artificial intelligence, characterized in that, include: Collect multi-source mineral data, perform standardization and vectorization transformation to generate a basic dataset; Based on the aforementioned basic dataset, spatial and correlation features of the basic dataset are extracted to construct a mineral resource network map and identify resource center nodes. Based on the mineral resource network diagram and resource center nodes, an intelligent algorithm is used to divide resource units at the same level and calculate the basic score of each unit. A mineral resource prediction model is constructed, and after training and cross-validation, the potential of the target area is scored and its level is determined. Generate a resource potential prediction map and output the prediction results.

2. The method for predicting mineral resource potential based on artificial intelligence according to claim 1, characterized in that, Extracting the spatial and association features of the base dataset, including: GIS spatial analysis technology is used to calculate the spatial topological relationships, distance correlations, and density distribution characteristics of geological units, and the spatial features of the basic dataset are extracted using the sliding window method. The correlation between different mineral indicators is calculated using mutual information entropy, and the associated features are obtained by combining the random forest algorithm.

3. The method for predicting mineral resource potential based on artificial intelligence according to claim 2, characterized in that, Constructing a mineral resource network diagram and identifying resource center nodes includes: Using geological units in the basic dataset as nodes and feature correlation coefficients as edge weights, an undirected weighted mineral resource network graph is constructed using a graph neural network. The PageRank algorithm is used to calculate the importance score of each node, and the resource center node is obtained by combining degree centrality and betweenness centrality.

4. The method for predicting mineral resource potential based on artificial intelligence according to claim 1, characterized in that, Intelligent algorithms are used to divide resource units at the same level and calculate the basic score of each unit, including: Simulated annealing-genetic algorithm is used to optimize the initial cluster centers of quantum particle swarm optimization, and fuzzy C-means clustering is combined to divide the same level resource units; The weights of each feature are determined based on the entropy weight method. A scoring system is constructed from four dimensions: geological condition adaptability, element anomaly intensity, spatial correlation, and exploration degree. The basic score of the unit is calculated using a weighted summation formula.

5. The method for predicting mineral resource potential based on artificial intelligence according to claim 4, characterized in that, The simulated annealing-genetic algorithm optimization process uses real number encoding, takes minimizing intra-class scatter as the fitness function, and improves clustering stability through crossover and mutation operations and simulated annealing acceptance criteria.

6. The method for predicting mineral resource potential based on artificial intelligence according to claim 5, characterized in that, The simulated annealing-genetic algorithm optimization process uses real-number encoding, with the fitness function being the minimization of intra-class scatter. It improves clustering stability through crossover and mutation operations and the simulated annealing acceptance criterion, specifically including: For the feature dimension m and the preset number of cluster centers c of mineral resource data, chromosomes are constructed using real number encoding, with each chromosome corresponding to a set of initial cluster centers; Randomly generate the initial population, and simultaneously initialize the membership matrix U and the initial cluster center matrix V; For each chromosome in the population, the cluster center matrix V is decoded, the intra-cluster scatter of each cluster is calculated, and the individual fitness value is obtained by substituting it into the fitness function. Using the arithmetic crossover method, two parent individuals P1 and P2 are randomly selected to generate two offspring individuals; Perform mutation operations on each gene locus of each individual; Set the initial temperature T0, the cooling coefficient β, and the minimum temperature threshold T. min Calculate the fitness difference between offspring and parents; Output the individual with the highest fitness value, and decode to obtain the optimized initial cluster center.

7. The method for predicting mineral resource potential based on artificial intelligence according to claim 4, characterized in that, The method of dividing resource units at the same level using fuzzy C-means clustering specifically includes: Step 1: Based on the optimized QPSO initial cluster centers k=1,2,...,c, set the fuzzy coefficient h and the upper limit of the number of iterations K. max Convergence threshold ε; Step 2: Calculate the initial membership degree of each geological unit sample to the kth cluster according to the FCM membership formula. ; Step 3: Based on the initial membership matrix, calculate the cluster centers for the first iteration according to the FCM cluster center update formula. ; Step 4: Substituting into the membership formula and performing iterative optimization, we obtain... And calculate the change in cluster centers; If the change in cluster centers is less than the convergence threshold, then the convergence condition has been met and the iteration stops. If it does not converge and the number of iterations has not reached K max Then let = , = Repeat steps 3-4; if the number of iterations reaches K... max If convergence is still not achieved, force a stop and output the current cluster centers; Step 5: Based on the membership matrix obtained through iteration, assign each sample to the cluster with the highest membership degree. Each cluster corresponds to a peer resource unit. Calculate the number of samples, spatial range, and feature mean of each unit to complete the peer resource unit division.

8. A mineral resource potential prediction system based on artificial intelligence, characterized in that, include: The data acquisition and preprocessing module is used to collect multi-source mineral data, perform standardization and vectorization transformation, and generate a basic dataset. The feature extraction module is used to extract the spatial and correlation features of the basic dataset based on the basic dataset, construct a mineral resource network map and identify resource center nodes; The intelligent partitioning module is used to partition resource units of the same level based on the mineral resource network map and resource center nodes, and calculate the basic score of the unit. The mineral resource prediction module is used to build a mineral resource prediction model. After training and cross-validation, it scores the potential and determines the grade of the target area. The results output module is used to generate a resource potential prediction map and output the prediction results.

Images