A multi-dimensional fusion prospecting prediction method based on multi-source data

Abstract

The present invention discloses a multi-dimensional fusion prospecting prediction method based on multi-source data, and relates to the field of prospecting prediction. It includes: obtaining multi-dimensional data of the study area; constructing a multi-dimensional prediction model; inputting pixel-level geological data, space-level geological data and structure-level geological data into the multi-dimensional prediction model respectively; predicting the geological data of three dimensions respectively through the multi-dimensional prediction model to obtain prediction results of three dimensions; introducing corresponding weights into the multi-dimensional prediction model, assigning different weights to the prediction results of the three dimensions respectively and performing weighted fusion to obtain the final prospecting prediction results, which are used to characterize potential mineral resource areas. The present invention predicts the multi-dimensional geological data of multi-source data by constructing a multi-dimensional prediction model, and weightedly fuses the prediction results, thereby realizing the fusion of multi-dimensional features of multi-source data and improving the prediction accuracy of mineral prediction and the reliability of geological interpretation.

Description

Technical Field

[0001] The present invention relates to the field of geological prospecting technology, and in particular to a multi-dimensional fusion prospecting prediction method based on multi-source data. Background Art

[0002] At present, with the development of geological exploration technology, the integration and analysis of geophysical, geochemical, remote sensing data and traditional geological surveys has become a trend in modern prospecting. In particular, deep learning has shown great potential in processing geological data and mining geological information. These technologies can process multi-source geological data, including geophysical, geochemical and remote sensing data, as well as traditional geological survey results, and reveal the complex network relationship between geological data through intelligent analysis. Machine learning and deep learning can improve the prediction effect by simulating human learning activities and extracting key features through deep neural networks. In the field of geological exploration, machine learning technology has not only improved the accuracy and efficiency of prospecting predictions, but also provided a new research direction for prospecting prediction methods.

[0003] In the existing technology, mineral exploration prediction models based on machine learning and deep learning have achieved certain results in practical applications. For multi-source geological data, models such as Random Forest are used to process pixel-level data, Convolutional Neural Network (CNN) is used to analyze spatial-level features, and Graph Convolutional Network (GCN) is used to process the correlation of mining area structures.

[0004] However, machine learning prospecting and prediction is to find the nonlinear relationship between multi-source data and mineralization targets. Due to the spatial diversity and complexity of geological data, the existing deep network model cannot effectively integrate geological data information of various dimensions, resulting in low prediction accuracy. At the same time, the lack of expert knowledge often makes the results unconvincing to geological experts and lacks the reliability of geological interpretation.

[0005] Therefore, how to better utilize geological data information of different dimensions from multi-source data and perform multi-dimensional information fusion and prediction to improve prediction accuracy and the reliability of geological interpretation has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] Based on this, it is necessary to provide a multi-dimensional fusion prospecting and prediction method based on multi-source data to address the above technical problems.

[0007] The present invention adopts the following technical solutions:

[0008] The present invention provides a multi-dimensional fusion prospecting prediction method based on multi-source data, comprising:

[0009] Extract multi-source data from geophysical, geochemical and remote sensing data; and construct the multi-source data into pixel-level data sets, spatial-level data sets and structural-level data sets respectively;

[0010] Construct a multi-dimensional prediction model including a random forest network model, a convolutional neural network model, and a graph neural network model and train the multi-dimensional prediction model;

[0011] The pixel-level geological data, space-level geological data and structure-level geological data are respectively input into the trained multi-dimensional prediction model; the pixel characteristics of the pixel-level geological data are extracted through the random forest model, and the distribution area information of the mineral resources is classified and predicted to obtain the first prediction result; the high-level spatial distribution characteristics of the space-level geological data are extracted through the convolutional neural network model, and the spatial information of the mineral resources is predicted to obtain the second prediction result; the structural relationship characteristics between the mineral points in the structural-level geological data are extracted through the graph neural network model, and the geological structure information of the mineral resource distribution area is predicted to obtain the third prediction result;

[0012] Different weights are assigned to the first prediction result, the second prediction result and the third prediction result, and weighted fusion is performed to obtain the final prospecting prediction result, which is used to characterize the potential mineral resource area.

[0013] Preferably, the multi-source data are constructed into a pixel-level dataset, a space-level dataset and a structure-level dataset, respectively, including:

[0014] A regular grid is constructed based on multi-source data, and the multi-source data in the regular grid is interpolated using the Kriging interpolation method to obtain a multi-source data concentration map;

[0015] Overlay the multi-element feature layers in the multi-source data concentration map and perform data fusion to extract the multi-channel information of target label pixels to obtain a pixel-level data set;

[0016] Using a fixed window size, the multi-source data concentration map is cropped by translation, rotation, and scaling to obtain a spatial-level data set;

[0017] The structural threshold is determined by using fractal theory, and the node feature matrix and adjacency matrix are constructed through the structural threshold to obtain the structural level data set.

[0018] Preferably, the determination of the trained multi-dimensional prediction model comprises the following steps:

[0019] Select part of the data set as training set samples;

[0020] The random forest network model, the convolutional neural network model and the graph neural network model are integrated by voting to obtain a multi-dimensional network model; wherein the voting method is specifically: averaging the output results of the random forest network model, the convolutional neural network model and the graph neural network model;

[0021] The training set samples are input into the multidimensional network model, and the multidimensional network model is trained to obtain a trained multidimensional prediction model.

[0022] Preferably, the random forest model is used to extract pixel features of pixel-level geological data and classify and predict the distribution area information of mineral resources, including:

[0023] The bootstrap sampling method was used to randomly extract some pixel-level geological data without replacement to obtain multiple sub-datasets;

[0024] For each sub-dataset, a decision tree model is constructed, and a portion of features are randomly selected at the nodes of each decision tree for splitting;

[0025] For a single decision tree model, set the sub-dataset to x = [x1, x2, …, x p ], the output of a single decision tree model is h(x), expressed as:

[0026]

[0027] Where M is the number of sub-datasets, m is any sub-dataset among all sub-datasets, I(·) is the indicator function, and R m is the leaf node area of ​​the decision tree, c m It is region R m The output value of

[0028] The random forest network model is obtained by integrating N decision tree models, and the first prediction result is obtained by averaging the outputs of the N decision tree models. The formula is:

[0029]

[0030] in, is the first prediction result, h i (x) is the output of the i-th decision tree.

[0031] Preferably, a convolutional neural network model is used to extract high-level spatial distribution characteristics of spatial-level geological data and predict spatial information of mineral resources, including:

[0032] Use spatial-level geological data as input to the convolutional neural network model;

[0033] The convolution layer is used to perform convolution operations on the spatial-level geological data, and the high-level distribution characteristics of the geological data are extracted through nonlinear activation functions. (l) , the formula is:

[0034] f (l) =σ(W (l) *x (l-1) +b (l) );

[0035] Among them, x (l-1) is the input of the l-1th layer, W (l) is the convolution kernel of layer l, b (l) is the bias term, * represents the convolution operation, and σ is the nonlinear activation function;

[0036] The high-level distribution characteristics of the geological data are averaged by the pooling layer to obtain the pooling characteristics, and the formula is:

[0037] p (l) =Pool(f (l) );

[0038] Among them, p (l) Represents the pooling feature, Pool represents the average pooling operation;

[0039] A fully connected layer is used to transform the extracted pooling features into the output of the convolutional neural network model to obtain the second prediction result.

[0040] Preferably, the graph neural network model is used to extract the structural relationship characteristics between the mineral points in the structural geological data and predict the geological structure information of the mineral resource distribution area, including:

[0041] Extracting neighbor node information and edge information of structural-level geological data;

[0042] The embedding representation of the node is updated iteratively, and the formula is:

[0043]

[0044] in, is the embedding representation of the node after the l-th layer update, v represents the node, is the embedding representation of the node after l-1 layer update, is the set of neighbor nodes of node v, u is the neighbor node of node v, W (l) and are the weight matrices of the current layer, σ is the nonlinear activation function, is the embedding representation of the neighbor nodes after the l-1 layer update;

[0045] Through the global aggregation function, the embedded representations of all nodes are aggregated into the representation of the structure graph, the output of the graph neural network model is obtained, and the third prediction result is obtained. The formula is:

[0046]

[0047] Among them, h G is the representation of the structure, Aggregate(·) is the aggregation function, and V is the node set.

[0048] Preferably, obtaining the final prospecting prediction result includes the following steps:

[0049] Obtaining a pairwise comparison matrix of the first prediction result, the second prediction result, and the third prediction result;

[0050] Calculate the weight vectors corresponding to the first prediction result, the second prediction result and the third prediction result through the pairwise comparison matrix;

[0051] The rationality of the weight vectors corresponding to the first prediction result, the second prediction result and the third prediction result is judged by a pairwise comparison matrix;

[0052] The final prospecting prediction result is obtained by weighted fusion of the first prediction result, the second prediction result and the third prediction result and their corresponding weight vectors.

[0053] Preferably, the rationality of the weight vectors corresponding to the first prediction result, the second prediction result and the third prediction result is judged by using a pairwise comparison matrix, including:

[0054] By formula |A-λ max I|=0 Calculate the maximum eigenvalue of the pairwise comparison matrix; where A is the pairwise comparison matrix, I is the identity matrix, and λ max is the maximum eigenvalue;

[0055] Calculate the consistency index by the maximum eigenvalue Where n is the order of the pairwise comparison matrix A;

[0056] Look up the random consistency index in the random consistency index reference table by pairwise comparison matrix order;

[0057] The consistency ratio is obtained by the ratio of the consistency index to the consistency index, and the consistency of the corresponding weights is judged according to the consistency ratio;

[0058] If the consistency ratio is less than or equal to 0.1, it is considered that the weight judgment of the first prediction result, the second prediction result and the third prediction result is reasonable;

[0059] If the consistency ratio is greater than 0.1, it is considered that the weight judgment of each prediction result is unreasonable, and the weight vector is automatically corrected through the yaahp software until the consistency ratio is less than or equal to 0.1.

[0060] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0061] The present invention predicts mineral resources by integrating multi-dimensional prediction models including random forest network models, convolutional neural network models and graph neural network models, including multi-source geological data of different dimensions at pixel level, spatial level and structural level, effectively integrating multi-source geological data information of different dimensions, and more comprehensively and accurately capturing the diversity and complexity of multi-source geological data. At the same time, by weighted fusion of multi-dimensional prediction results and making rational judgments, the prediction results have better interpretability and traceability, improving the prediction accuracy of prospecting prediction and the reliability of geological interpretation, thereby providing powerful scientific guidance for further mineral evaluation and exploration decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0063] Figure 1 A schematic diagram of a flow chart of a multi-dimensional fusion prospecting prediction method based on multi-source data provided by the present invention;

[0064] Figure 2 A diagram of the architecture of a multi-dimensional prediction model of a multi-dimensional fusion prospecting prediction method based on multi-source data provided by the present invention;

[0065] Figure 3 A schematic diagram of a first prediction result of a multi-dimensional fusion prospecting prediction method based on multi-source data provided by the present invention;

[0066] Figure 4 A schematic diagram of a second prediction result of a multi-dimensional fusion prospecting prediction method based on multi-source data provided by the present invention;

[0067] Figure 5 A schematic diagram of a third prediction result of a multi-dimensional fusion prospecting prediction method based on multi-source data provided by the present invention;

[0068] Figure 6 A schematic diagram of the final prediction results of a multi-dimensional fusion prospecting prediction method based on multi-source data provided by the present invention. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0070] The technical solutions provided by various embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0071] Figure 1 The present invention is a schematic flow chart of a multi-dimensional fusion prospecting prediction method based on multi-source data, which specifically includes the following steps:

[0072] S101: Extract multi-source data from geophysical, geochemical, and remote sensing data; and construct the multi-source data into pixel-level data sets, spatial-level data sets, and structural-level data sets.

[0073] Optionally, the multi-source data are constructed into a pixel-level dataset, a space-level dataset and a structure-level dataset, respectively, including: constructing a regular grid based on the multi-source data, and using the Kriging interpolation method to interpolate the multi-source data in the regular grid to obtain a multi-source data concentration map; superimposing the multi-element feature layers in the multi-source data concentration map and performing data fusion to extract multi-channel information of target label pixels to obtain a pixel-level dataset; using a fixed window size, cropping the multi-source data concentration map by translation, rotation and scaling to obtain a space-level dataset; determining the structural threshold using fractal theory, and constructing a node feature matrix and an adjacency matrix respectively through the structural threshold to obtain a structure-level dataset.

[0074] Specifically, the Kriging interpolation method is used to interpolate the multi-source data within the grid, including:

[0075] For the sample point Z(s) of known multi-source data i ) and Z(s j ), calculate Z(s i ) and Z(s j ) ij =abs(Z(s i )-Z(s j )); where i, j = 1, 2, 3, …, abs(·) represents the absolute value.

[0076] Calculate the semivariogram function using the formula:

[0077]

[0078] Among them, γ(h) is the semivariogram function, N(h) is the distance hij The number of sample point pairs, z(s i ) and Z(s j ) is a pair of sample points of multi-source data; through the semivariogram function, the unknown sample point s is calculated. o With all known sample points s i The semivariance γ(h oi ).

[0079] Calculate the optimal weight coefficient through the constraints; the constraints are:

[0080]

[0081] In the formula, ω i is the optimal weight coefficient, ω j is a known weight coefficient, γ(h ij ) is the semivariogram function, σ 2 is the variance Var[R(i,j)] of the sample point pair, and R(i,j) is the deviation of the sample point pair.

[0082] Compute estimates for unknown points:

[0083]

[0084] In the formula, Z'(s o ) is the estimated value of the unknown point, ω i is the optimal weight coefficient, Z(s i ) is the sample point of multi-source data, and n is the number of unknown points.

[0085] Repeat the above process to interpolate all multi-source data in the regular grid and obtain the multi-source data concentration map.

[0086] S102: Constructing a multi-dimensional prediction model including a random forest network model, a convolutional neural network model, and a graph neural network model and training the multi-dimensional prediction model, including the following steps:

[0087] Select part of the multi-dimensional data as training set samples;

[0088] The random forest network model, convolutional neural network model and graph neural network model are integrated by voting to obtain a multi-dimensional network model;

[0089] The training set samples are input into the multidimensional network model, and the multidimensional prediction model is obtained by training the multidimensional network model.

[0090] S103: See Figure 2, respectively input pixel-level geological data, space-level geological data and structure-level geological data into the trained multi-dimensional prediction model; through the random forest model, extract the pixel characteristics of the pixel-level geological data and classify and predict the distribution area information of mineral resources to obtain the first prediction result; through the convolutional neural network model, extract the high-level spatial distribution characteristics of the space-level geological data and predict the spatial information of mineral resources to obtain the second prediction result; through the graph neural network model, extract the structural relationship characteristics between the mineral points in the structural-level geological data and predict the geological structure information of the mineral resource distribution area to obtain the third prediction result, which specifically includes:

[0091] Through the random forest model, the pixel features of pixel-level geological data are extracted and the distribution area information of mineral resources is classified and predicted to obtain the first prediction result, see Figure 3 ,include:

[0092] The bootstrap sampling method was used to randomly extract some pixel-level geological data without replacement to obtain multiple sub-datasets;

[0093] For each sub-dataset, a decision tree model is constructed, and a portion of features are randomly selected at the nodes of each decision tree for splitting;

[0094] For a single decision tree model, set the sub-dataset to x = [x1, x2, …, x p ], the output of a single decision tree model is h(x), expressed as:

[0095]

[0096] Where M is the number of sub-datasets, m is any sub-dataset among all sub-datasets, I(·) is the indicator function, and R m is the leaf node area of ​​the decision tree, c m It is region R m The output value of

[0097] The random forest network model is obtained by integrating N decision tree models, and the first prediction result is obtained by averaging the outputs of the N decision tree models. The formula is:

[0098]

[0099] in, is the first prediction result, h i (x) is the output of the i-th decision tree.

[0100] Furthermore, through the convolutional neural network model, the high-level spatial distribution characteristics of the spatial-level geological data are extracted and the spatial information of the mineral resources is predicted to obtain the second prediction result, including:

[0101] Use spatial-level geological data as input to the convolutional neural network model;

[0102] The convolution layer is used to perform convolution operations on the spatial-level geological data, and the high-level distribution characteristics of the geological data are extracted through nonlinear activation functions. (l) , the formula is:

[0103] f (l) =σ(W (l) *x (l-1) +b (l) );

[0104] Among them, x (l-1) is the input of the l-1th layer, W (l) is the convolution kernel of layer l, b (l) is the bias term, * represents the convolution operation, and σ is the nonlinear activation function;

[0105] The high-level distribution characteristics of the geological data are averaged by the pooling layer to obtain the pooling characteristics, and the formula is:

[0106] p (l) =Pool(f (l) );

[0107] Among them, p (l) Represents the pooling feature, Pool represents the average pooling operation;

[0108] The fully connected layer is used to transform the extracted pooling features into the output of the convolutional neural network model to obtain the second prediction result, see Figure 4 ;

[0109] Furthermore, the graph neural network model is used to extract the structural relationship characteristics between the mineral points in the structural-level geological data and predict the geological structure information of the mineral resource distribution area to obtain the third prediction result, including:

[0110] Extracting neighbor node information and edge information of structural-level geological data;

[0111] The embedding representation of the node is updated iteratively, and the formula is:

[0112]

[0113] Among them, among them, is the embedding representation of the node after l layers of updating, v represents the node, is the embedding representation of the node after l-1 layer update, is the set of neighbor nodes of node v, u is the neighbor node of node v, W (l) and are the weight matrices of the current layer, σ is the nonlinear activation function, is the embedding representation of the neighbor nodes after the l-1 layer update;

[0114] Through the global aggregation function, the embedding representations of all nodes are aggregated into the representation of the structure graph, and the output of the graph neural network model is obtained to obtain the third prediction result, see Figure 5 , the formula is:

[0115]

[0116] Among them, h G is the representation of the structure, Aggregate(·) is the aggregation function, and V is the node set.

[0117] S104: assigning different weights to the first prediction result, the second prediction result and the third prediction result respectively and performing weighted fusion to obtain a final prospecting prediction result to characterize the potential mineral resource area, specifically including:

[0118] Obtaining a pairwise comparison matrix of the first prediction result, the second prediction result, and the third prediction result;

[0119] Calculate the weight vectors corresponding to the first prediction result, the second prediction result and the third prediction result through the pairwise comparison matrix;

[0120] The rationality of the corresponding weight vector is judged through the pairwise comparison matrix;

[0121] The final prospecting prediction result is obtained by weighted fusion of the first prediction result, the second prediction result and the third prediction result and their corresponding weight vectors. Figure 6 ;

[0122] The rationality of the corresponding weights is judged by using a pairwise comparison matrix, including the following steps:

[0123] By formula |A-λ max I|=0 Calculate the maximum eigenvalue of the pairwise comparison matrix; where A is the pairwise comparison matrix, I is the identity matrix, and λ max is the maximum eigenvalue;

[0124] Calculate the consistency index by the maximum eigenvalue Where n is the order of the pairwise comparison matrix A;

[0125] Look up the random consistency index in the random consistency index reference table by pairwise comparison matrix order;

[0126] The consistency ratio is obtained by the ratio of the consistency index to the consistency index, and the consistency of the corresponding weights is determined according to the consistency ratio;

[0127] If the consistency ratio is less than 0.1, the corresponding weights are considered reasonable for the judgment of the first prediction result, the second prediction result, and the third prediction result;

[0128] If the consistency ratio is greater than 0.1, it is considered that the weight judgment of each prediction result is unreasonable, and the weight vector is automatically corrected through the yaahp software until the consistency ratio is less than or equal to 0.1.

[0129] This embodiment uses multi-source data such as land, physical, chemical, remote sensing, and drilling in a certain study area. Through the above steps, pixel-level geological data, space-level structural data, and structure-level geological data are obtained. The geological data of the above three dimensions are input into the multi-dimensional prediction model. The multi-dimensional prediction model uses a random forest network model to predict the pixel-level geological data and obtain the first prediction result (91.3%). The convolutional neural network model is used to predict the space-level geological data and obtain the second prediction result (94.7%). The graph neural network model is used to predict the structure and data and obtain the third prediction result (96.1%). A pairwise comparison matrix of each prediction result is established, and the corresponding weight vector of each prediction result is calculated. The weight judgment is introduced. The consistency ratio of the corresponding weight of each prediction result calculated by the ya ahp software is 0.056<0.1. Therefore, the weight judgment is reasonable;

[0130] The multi-dimensional predictions are weighted and fused to obtain the multi-dimensional prediction results. Through comparative studies, it is found that the prediction accuracy of the multi-dimensional fusion method reaches 97.6%, which is higher than the prediction accuracy of using only pixel-level, space-level and structure-level models. At the same time, the prediction results and the delineated target areas are consistent with the geological interpretation.

[0131] In summary, this embodiment predicts multi-source geological data of different dimensions through a multi-dimensional prediction model, effectively integrates geological data information of different dimensions, and can more comprehensively and accurately capture the diversity and complexity of multi-source geological data. In addition, by introducing the judgment of corresponding weights and integrating the prediction results of multiple dimensions, the prediction results have better interpretability and traceability, and the robustness and accuracy of prospecting prediction are improved, thereby providing powerful scientific guidance for further mineral evaluation and exploration decisions.

[0132] The technical solution of the present invention is not limited to the above-mentioned specific embodiments. All technical variations made according to the technical solution of the present invention fall within the protection scope of the present invention.

Claims

1. A multi-dimensional fusion prospecting prediction method based on multi-source data, characterized in that: The following steps are involved: Extract multi-source data from geophysical, geochemical and remote sensing data; and construct the multi-source data into pixel-level data sets, spatial-level data sets and structural-level data sets respectively; Construct a multi-dimensional prediction model including a random forest network model, a convolutional neural network model, and a graph neural network model and train the multi-dimensional prediction model; The pixel-level geological data, space-level geological data and structure-level geological data are respectively input into the trained multi-dimensional prediction model; the pixel characteristics of the pixel-level geological data are extracted through the random forest model, and the distribution area information of the mineral resources is classified and predicted to obtain the first prediction result; the high-level spatial distribution characteristics of the space-level geological data are extracted through the convolutional neural network model, and the spatial information of the mineral resources is predicted to obtain the second prediction result; the structural relationship characteristics between the mineral points in the structural-level geological data are extracted through the graph neural network model, and the geological structure information of the mineral resource distribution area is predicted to obtain the third prediction result; Different weights are assigned to the first prediction result, the second prediction result and the third prediction result, respectively, and weighted fusion is performed to obtain the final prospecting prediction result, which is used to characterize the potential mineral resource area.

2. A multi-dimensional fusion prospecting prediction method based on multi-source data as claimed in claim 1, characterized in that: The multi-source data are constructed into a pixel-level dataset, a space-level dataset and a structure-level dataset respectively, specifically including: A regular grid is constructed based on multi-source data, and the multi-source data in the regular grid is interpolated using the Kriging interpolation method to obtain a multi-source data concentration map; Overlay the multi-element feature layers in the multi-source data concentration map and perform data fusion to extract the multi-channel information of target label pixels to obtain a pixel-level data set; Using a fixed window size, the multi-source data concentration map is cropped by translation, rotation, and scaling to obtain a spatial-level data set; The structural threshold is determined by using fractal theory, and the node feature matrix and adjacency matrix are constructed through the structural threshold to obtain the structural level data set.

3. A multi-dimensional fusion prospecting prediction method based on multi-source data as claimed in claim 1, characterized in that: The determination of the trained multi-dimensional prediction model comprises the following steps: Select part of the data set as training set samples; The random forest network model, the convolutional neural network model and the graph neural network model are integrated by voting to obtain a multi-dimensional network model; wherein the voting method is specifically: averaging the output results of the random forest network model, the convolutional neural network model and the graph neural network model; The training set samples are input into the multidimensional network model, and the multidimensional network model is trained to obtain a trained multidimensional prediction model.

4. A multi-dimensional fusion prospecting prediction method based on multi-source data as claimed in claim 1, characterized in that: The random forest model is used to extract pixel features of pixel-level geological data and classify and predict the distribution area information of mineral resources, including: The bootstrap sampling method was used to randomly extract some pixel-level geological data without replacement to obtain multiple sub-datasets; For each sub-dataset, a decision tree model is constructed, and a portion of features are randomly selected at the nodes of each decision tree for splitting; For a single decision tree model, set the sub-dataset to x = [x1, x2, …, x p ], the output of a single decision tree model is h(x), expressed as: Where M is the number of sub-datasets, m is any sub-dataset among all sub-datasets, I(·) is the indicator function, and R m is the leaf node area of ​​the decision tree, c m It is region R m The output value of The random forest network model is obtained by integrating N decision tree models, and the first prediction result is obtained by averaging the outputs of the N decision tree models. The formula is: in, is the first prediction result, h i (x) is the output of the i-th decision tree.

5. A multi-dimensional fusion prospecting prediction method based on multi-source data as claimed in claim 1, characterized in that: The convolutional neural network model is used to extract high-level spatial distribution characteristics of spatial-level geological data and predict the spatial information of mineral resources, including: Use spatial-level geological data as input to the convolutional neural network model; The convolution layer is used to perform convolution operations on the spatial-level geological data, and the high-level distribution characteristics of the geological data are extracted through nonlinear activation functions. (l) , the formula is: f (l) =σ(W (l) *x (l-1) +b (l) ); Among them, x (l-1) is the input of the l-1th layer, W (l) is the convolution kernel of layer l, b (l) is the bias term, * represents the convolution operation, and σ is the nonlinear activation function; The high-level distribution characteristics of the geological data are averaged by the pooling layer to obtain the pooling characteristics, and the formula is: p (l) =Pool(f (l) ); Among them, p (l) Represents the pooling feature, Pool represents the average pooling operation; A fully connected layer is used to transform the extracted pooling features into the output of the convolutional neural network model to obtain the second prediction result.

6. A multi-dimensional fusion prospecting prediction method based on multi-source data as claimed in claim 1, characterized in that: The graph neural network model is used to extract the structural relationship characteristics between the mineral points in the structural geological data and predict the geological structure information of the mineral resource distribution area, including: Extracting neighbor node information and edge information of structural-level geological data; The embedding representation of the node is updated iteratively, and the formula is: in, is the embedding representation of the node after the l-th layer update, v represents the node, is the embedding representation of the node after l-1 layer update, is the set of neighbor nodes of node v, u is the neighbor node of node v, W (l) and are the weight matrices of the current layer, σ is the nonlinear activation function, is the embedding representation of the neighbor nodes after the l-1 layer update; Through the global aggregation function, the embedded representations of all nodes are aggregated into the representation of the structure graph, the output of the graph neural network model is obtained, and the third prediction result is obtained. The formula is: Among them, h G is the representation of the structure, Aggregate(·) is the aggregation function, and V is the node set.

7. A multi-dimensional fusion prospecting prediction method based on multi-source data as claimed in claim 1, characterized in that: The final prospecting prediction result is obtained by: Obtaining a pairwise comparison matrix of the first prediction result, the second prediction result, and the third prediction result; Calculate the weight vectors corresponding to the first prediction result, the second prediction result and the third prediction result through the pairwise comparison matrix; The rationality of the weight vectors corresponding to the first prediction result, the second prediction result and the third prediction result is judged by a pairwise comparison matrix; The final prospecting prediction result is obtained by weighted fusion of the first prediction result, the second prediction result and the third prediction result and their corresponding weight vectors.

8. A multi-dimensional fusion prospecting prediction method based on multi-source data as claimed in claim 7, characterized in that: The rationality judgment of the weight vectors corresponding to the first prediction result, the second prediction result and the third prediction result by using the pairwise comparison matrix includes: By formula |A-λ max I|=0 Calculate the maximum eigenvalue of the pairwise comparison matrix; where A is the pairwise comparison matrix, I is the identity matrix, and λ max is the maximum eigenvalue; Calculate the consistency index by the maximum eigenvalue Where n is the order of the pairwise comparison matrix A; Look up the random consistency index in the random consistency index reference table by pairwise comparison matrix order; The consistency ratio is obtained by the ratio of the consistency index to the consistency index, and the consistency of the corresponding weights is judged according to the consistency ratio; If the consistency ratio is less than or equal to 0.1, it is considered that the weight judgment of the first prediction result, the second prediction result and the third prediction result is reasonable; If the consistency ratio is greater than 0.1, it is considered that the weight judgment of each prediction result is unreasonable, and the weight vector is automatically corrected through the yaahp software until the consistency ratio is less than or equal to 0.1.

Images