Quantitative Prediction Method and System for Mineral Resources Based on Knowledge Embedding

Abstract

The present invention provides a method and system for quantitative prediction of mineral resources based on knowledge embedding. Among them, the method includes: constructing a prospecting conceptual model according to the ore-forming geological bodies, ore-forming tectonic systems, and ore-forming characteristic markers related to the target mining area; obtaining multiple characteristic variables according to the prospecting conceptual model and the prospecting geological knowledge of the target mining area, and through spatial interpolation, interpolating the discrete and discontinuous characteristic variable sampling points in the target mining area into continuous data to generate a prediction characteristic layer; determining the weights of each characteristic variable; generating a prediction model according to the weights of each characteristic variable and a pre-set ore deposit information characteristic depth network; generating multiple samples based on the prediction characteristic layer, one sample corresponding to one unit grid, and the samples include positive samples, negative samples, and samples to be predicted; inputting the multiple samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area. By adopting the above technical solutions, the present application can improve the prediction accuracy of mineral production.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of mineral exploration, and in particular, to a method and system for quantitative prediction of mineral resources based on knowledge embedding. Background Art

[0002] With the development of science and technology and the advancement of mineral exploration work, geo-science data (such as geological, geophysical, geochemical, remote sensing data, etc.) has been continuously growing, providing a solid data foundation for prospecting exploration. At the same time, supported by advanced computer science and artificial intelligence technologies, digital earth science has developed rapidly, and various quantitative mathematical models and calculation methods have emerged in an endless stream. The intelligent mineral prediction system centered on machine learning and deep learning methods has advantages in processing large-scale data and complex pattern recognition, can support the extraction and expression of deep-level features of metallogenic information, establish a non-linear correlation from metallogenic elements to ore bodies, and has made some good progress in the prediction and evaluation of mineral resources.

[0003] With the deepening of the understanding of geological processes and the development and improvement of modern metallogenic theories and technical methods, the prediction and evaluation of mineral resources have developed from traditional qualitative prediction research to quantitative evaluation based on multiple models; from simple similarity analogy to the mining and synthesis of abnormal information of complex multi-source geo-science data, from the prediction and evaluation of single ore types to the comprehensive evaluation of multiple ore types; from the qualitative prediction and evaluation of the occurrence of ore deposits to the comprehensive evaluation and quantitative prediction of mineral resources (ore belts, ore fields, ore deposits, etc.) of different scales and different sizes. The prediction and evaluation of mineral resources also show a trend of systematization, scientification, visualization, and dynamization. Machine learning, especially deep learning models, has excellent processing capabilities for complex and non-linear problems, making the intelligent mineral prediction system a powerful tool for analyzing complex geological metallogenic processes and conducting metallogenic prediction reasoning in the process of metallogenic prediction.

[0004] However, the above-mentioned prediction methods in the prior art still have limitations, especially unable to accurately predict mineral production. Summary of the Invention

[0005] In view of this, the embodiments of the present invention provide a method and system for quantitative prediction of mineral resources based on knowledge embedding, which can improve the prediction accuracy of mineral production.

[0006] The embodiments of the present invention provide a method for quantitative prediction of mineral resources based on knowledge embedding, including:

[0007] Construct a prospecting conceptual model according to the ore-forming geological bodies, ore-forming tectonic systems, and characteristic signs of ore-forming processes related to the target mining area;

[0008] According to the prospecting conceptual model and the prospecting geological knowledge of the target mining area, obtain multiple characteristic variables, and through spatial interpolation, interpolate the discrete and discontinuous characteristic variable sampling points in the target mining area into continuous data to generate a predicted characteristic layer;

[0009] Determine the weights of each of the characteristic variables;

[0010] Generate a prediction model according to the weights of each of the characteristic variables and a pre-set deposit information feature depth network;

[0011] Based on the predicted characteristic layer, generate multiple samples, where one sample corresponds to one unit grid, and multiple of the unit grids form the predicted characteristic layer, and the samples include positive samples, negative samples, and samples to be predicted;

[0012] Input the multiple samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area.

[0013] Optionally, the step of determining the weights of each of the characteristic variables includes:

[0014] Determine the first score and the second score of each characteristic variable;

[0015] Compare the first eigenvector with the highest first score with other characteristic variables, and determine a first score vector based on the relative importance between the first eigenvector and other characteristic variables;

[0016] Compare the second eigenvector with the lowest second score with other characteristic variables, and determine a second score vector based on the relative importance between the second eigenvector and other characteristic variables;

[0017] According to each characteristic variable, the first score vector value of each characteristic variable in the first score vector, and the second score vector value of each characteristic variable in the second score vector, use the BWM knowledge embedding model to determine the weights of each characteristic variable.

[0018] Optionally, the step of determining the first score and the second score of each characteristic variable includes:

[0019] Extract the key information elements in each characteristic variable, and determine the key information expression of each characteristic variable;

[0020] Input the key information expressions of each characteristic variable into a pre-set knowledge graph model to obtain multiple key nodes corresponding to each key information expression;

[0021] Using the knowledge graph model, map multiple key nodes of the key information expression to obtain multiple vector representations corresponding to each key information expression;

[0022] Fuse the multiple vector representations corresponding to each key information expression and each key information expression to form a fusion variable corresponding to each feature variable;

[0023] Input the fusion variable corresponding to each feature variable into a pre-trained neural network, and respectively obtain the first confidence level and the second confidence level of each fusion variable under different confidence conditions, and use the first confidence level as the first score and the second confidence level as the second score.

[0024] Optionally, the step of generating a prediction model according to the weights in each of the feature variables and a pre-set ore deposit information feature deep network includes:

[0025] Input the weights of the feature variables into the loss function of the ore deposit information feature deep network to determine the loss value of the ore deposit information feature deep network, where the loss value is determined based on the following formula:

[0026]

[0027] Based on the loss value, perform backpropagation using the gradient descent method to determine the parameter gradient of the ore deposit information feature deep network;

[0028] Based on the parameter gradient, use a parameter optimization algorithm to update the weight parameters between different layers of the ore deposit information feature deep network to generate the prediction model;

[0029] where N is the number of samples, C is the number of categories; y ij is the true label of the j-th category in the sample X i ; is the predicted probability of the j-th category in the sample X i ; w j is the weight.

[0030] Optionally, the sample is a positive sample, and the positive sample is used to represent an ore point;

[0031] The step of generating multiple samples based on the prediction feature layer includes at least one of the following:

[0032] Perform grid processing on the predicted feature layer to generate multiple cells; according to the spatial correlation of mineralization, taking the ore points in the target mining area as the center, generate a buffer zone, and use all the samples within the buffer zone as initial samples, where the range of the buffer zone is determined based on the area of the target mining area and the number of ore points in the target mining area;

[0033] Taking the cell where the initial sample point is located as the base point, and selecting any one of multiple neighboring cells as an extended sample; perform difference processing on the extended sample and the initial sample to generate multiple samples;

[0034] Label the initial samples to generate initial samples with first label information and obtain sample data with second label information; use the initial samples with the first label information to train a preset supervised model to obtain a learning model; input the sample data with the second label information into the learning model to generate sample data with pseudo-label information; input the initial samples with the first label information and the sample data with the pseudo-label information into the learning model to generate multiple samples.

[0035] Optionally, the sample is a negative sample, and the negative sample is used to represent non-ore points;

[0036] The step of generating multiple samples based on the predicted feature layer includes at least one of the following:

[0037] Perform grid processing on the predicted feature layer to generate multiple cells, where one cell corresponds to one feature value; calculate the statistics of each feature value, and use the samples corresponding to the cells outside the preset statistics as negative samples according to the statistics of each feature value and the preset statistics;

[0038] Based on the non-linear dimensionality reduction method, map the high-dimensional data in the predicted feature layer to a low-dimensional space, determine the abnormal area, and use the samples corresponding to the cells in the abnormal area as negative samples.

[0039] Optionally, the step of inputting the predicted feature layer and multiple samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area includes:

[0040] Based on the positive samples, negative samples and samples to be predicted, as well as the first label corresponding to the positive samples, the second label corresponding to the negative samples and the third label corresponding to the samples to be predicted, determine the mineralization probability of each unit grid based on the prediction model;

[0041] Based on the mineralization probabilities of each unit grid, the cumulative frequency method is used to determine the target area grading situation, and the cumulative result is used as the prediction result of the mineral resources in the target mining area. Among them, the prediction result of the mineral resources in the target mining area includes the delineation of the predicted target area and the mineralization probability within the predicted target area.

[0042] Optionally, before inputting the predicted feature layer and the multiple samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area, the prediction method further includes:

[0043] Using permutation importance, partial dependence plots, and model-agnostic local interpretation algorithms to deconstruct the prediction model and adjust the parameters of the prediction model.

[0044] Correspondingly, an embodiment of the present invention further provides a quantitative prediction system for mineral resources based on knowledge embedding, including:

[0045] A construction unit configured to construct a prospecting concept model according to the ore-forming geological bodies, ore-forming tectonic systems, and ore-forming characteristic markers related to the target mining area;

[0046] A first generation unit configured to obtain a plurality of characteristic variables according to the prospecting concept model and the prospecting geological knowledge of the target mining area, and through spatial interpolation, interpolate the discrete and discontinuous characteristic variable sampling points in the target mining area into continuous data to generate a predicted feature layer;

[0047] A processing unit configured to determine the weights of the respective characteristic variables; and generate a plurality of samples based on the predicted feature layer, where one sample corresponds to one unit grid, and the plurality of unit grids form the predicted feature layer, and the samples include positive samples, negative samples, and samples to be predicted; and input the plurality of samples into a prediction model to obtain the prediction result of the mineral resources in the target mining area;

[0048] A second generation unit configured to generate the prediction model according to the weights of the respective characteristic variables and a pre-set ore deposit information feature depth network.

[0049] Correspondingly, an embodiment of the present invention further provides an electronic device, including: at least one memory and at least one processor, where the memory stores one or more computer instructions, and where the one or more computer instructions are executed by the processor to implement the quantitative prediction method for mineral resources based on knowledge embedding as described in any of the foregoing embodiments.

[0050] Compared with the prior art, the technical solution of the embodiment of the present invention has the following advantages:

[0051] In the method for quantitatively predicting mineral resources based on knowledge embedding provided by the embodiments of the present invention, by constructing a prospecting conceptual model related to the target mining area, a prediction feature layer can be generated based on the prospecting conceptual model and the prospecting geological knowledge of the target mining area. Furthermore, multiple samples can be generated based on the prediction feature layer. At the same time, a prediction model is generated based on the weights of each feature variable and a pre-set deep network of deposit information features. Thus, by inputting multiple samples into the prediction model, the prediction result of the mineral resources in the target mining area can be determined. Since the prospecting conceptual model takes into account the ore-forming geological bodies, ore-forming tectonic systems, and ore-forming characteristic markers, the entire prediction process can perform multi-dimensional and multi-level model deconstruction from the global and local aspects, features, and samples. At the same time, geological knowledge is added to the loss value calculation of the deep network of deposit information features in the form of weights in feature variables, obtaining an intelligent mineral prediction system driven by the combination of knowledge and data, effectively improving the accuracy, stability, and reliability of the prediction model, and further improving the prediction accuracy of mineral production. This helps to promote the exploration and development of hidden minerals and is of great significance for achieving a new round of breakthroughs in prospecting. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flowchart of a method for quantitatively predicting mineral resources based on knowledge embedding in an embodiment of the present invention;

[0053] Figure 2 is a flowchart of determining the weights in each of the feature variables in an embodiment of the present invention;

[0054] Figure 3 is a schematic diagram of the acquisition process of the weights of each feature variable in an embodiment of the present invention;

[0055] Figure 4 is a schematic diagram of the generation process of a prediction model in an embodiment of the present invention;

[0056] Figure 5 is a schematic diagram of the acquisition process principle of positive samples in the first embodiment in an embodiment of the present invention;

[0057] Figure 6 is a schematic diagram of the acquisition process principle of positive samples in the second embodiment in an embodiment of the present invention;

[0058] Figure 7 is a schematic diagram of the output of a prediction model in an embodiment of the present invention;

[0059] Figure 8 is a schematic diagram of the structure of a system for quantitatively predicting mineral resources based on knowledge embedding in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] As can be seen from the background art, the prediction accuracy of mineral resources obtained by existing machine (or deep) learning models is still limited because:

[0061] The construction of machine (or deep) learning models is mainly data-driven, relatively lacking the guidance of geological knowledge, and there is inevitably a "black box" effect in the intelligent prediction process, which makes the current data-driven artificial intelligence methods face two important challenges: The machine (or deep) learning models purely based on data-driven lack the guidance of geological knowledge, and it is difficult to fully correspond to the characteristics of the ore deposit genetic model only relying on spatial data features, lacking cognitive consistency; at the same time, the machine (or deep) learning algorithm can mine the deep features of ore-forming information, but due to its nested non-linear network structure and abstract expression, it has a highly opaque black box property, and the interpretability of the model is relatively weak, making each stage have different uncertainty sources and will be propagated to the next stage, resulting in uncertainty in the results of mineral prediction.

[0062] To solve the above technical problems, an embodiment of the present invention provides a quantitative prediction method for mineral resources based on knowledge embedding, including: constructing an ore prospecting concept model according to the ore-forming geological bodies, ore-forming tectonic systems and ore-forming characteristic markers related to the target mining area; obtaining a plurality of characteristic variables according to the ore prospecting concept model and the ore prospecting geological knowledge of the target mining area, and by means of spatial interpolation, interpolating the discrete and discontinuous characteristic variable sampling points in the target mining area into continuous data to generate a prediction characteristic layer; determining the weights of each of the characteristic variables; generating a prediction model according to the weights of each of the characteristic variables and a pre-set deep network of ore deposit information characteristics; generating a plurality of samples based on the prediction characteristic layer, wherein one sample corresponds to one unit grid, and a plurality of the unit grids form the prediction characteristic layer, and the samples include positive samples, negative samples and samples to be predicted; inputting the plurality of samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area.

[0063] In the method for quantitatively predicting mineral resources based on knowledge embedding provided by the embodiments of the present invention, by constructing a prospecting conceptual model related to the target mining area, a prediction feature layer can be generated based on the prospecting conceptual model and the prospecting geological knowledge of the target mining area, and then multiple samples can be generated based on the prediction feature layer. At the same time, a prediction model is generated based on the weights of each feature variable and the pre-set deposit information feature deep network. Therefore, by inputting the prediction feature layer and multiple samples into the prediction model, the prediction result of the mineral resources in the target mining area can be determined. Since the prospecting conceptual model takes into account the ore-forming geological bodies, ore-forming tectonic systems, and ore-forming process characteristic signs, the entire prediction process can perform multi-dimensional and multi-level model deconstruction from the global and local aspects, features and samples, obtaining an intelligent mineral prediction system driven by knowledge-data combination, effectively improving the accuracy, stability, and reliability of the prediction model, and further improving the prediction accuracy of mineral production. This helps to promote the exploration and development of hidden minerals and is of great significance for achieving a new round of breakthroughs in prospecting.

[0064] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be illustrated by examples with reference to the accompanying drawings.

[0065] See Figure 1 The flowchart of a method for quantitatively predicting mineral resources based on knowledge embedding in the embodiments of the present invention shown in Figure 1 As shown, the following prediction steps can be performed:

[0066] Step A: Construct a prospecting conceptual model according to the ore-forming geological bodies, ore-forming tectonic systems, and ore-forming process characteristic signs related to the target mining area.

[0067] Among them, the target mining area (as a non-limiting example, the target mining area can refer to the Ke'eryin ore field) can refer to a mineral resource area that has been explored and evaluated and determined to have economic mining value.

[0068] The ore-forming geological body refers to the aggregate of mineral resources formed through a series of geological processes under specific geological environments and conditions, and it usually includes ore bodies, ore veins, ore deposits, etc.

[0069] The ore-forming tectonic system refers to the geological tectonic system in which mineral resources are formed and concentrated in a specific geological environment.

[0070] The ore-forming process characteristic signs refer to the characteristics and signs shown during the formation process of mineral resources, which can help geologists and mineral exploration personnel identify and understand the ore-forming process.

[0071] In other words, the ore-forming geological bodies, ore-forming tectonic systems, and characteristic signs of ore-forming processes are all used to characterize the ore-forming degree of the target mining area. Based on the above factors, the prospecting conceptual model can include different exploration perspectives, improve the accuracy of the prospecting conceptual model, and provide a solid basis for intelligent prediction.

[0072] In this embodiment, according to the ore-forming geological bodies, ore-forming tectonic systems, and characteristic signs of ore-forming processes related to the target mining area, the steps of constructing a prospecting conceptual model may include:

[0073] Using a pre-trained object detection model, respectively identify the first structural elements in the ore-forming geological bodies, the second structural elements in the ore-forming tectonic systems, and the third structural elements in the characteristic signs of ore-forming processes; according to the first structural elements, construct the first triple relationship between the ore-forming geological bodies and the initial prospecting conceptual model, according to the second structural elements, construct the second triple relationship between the ore-forming tectonic systems and the initial prospecting conceptual model, and according to the third structural elements, construct the third triple relationship between the characteristic signs of ore-forming processes and the initial prospecting conceptual model; randomly initialize the first triple relationship to obtain the first embedding vector, randomly initialize the second triple relationship to obtain the second embedding vector, and randomly initialize the third triple relationship to obtain the third embedding vector; use the ComplEx model to train the first embedding vector, the second embedding vector, and the third embedding vector, and respectively determine the first knowledge graph corresponding to the ore-forming geological bodies, the second knowledge graph corresponding to the ore-forming tectonic systems, and the third knowledge graph corresponding to the characteristic signs of ore-forming processes; use the attention mechanism to fuse the first knowledge graph, the second knowledge graph, and the third knowledge graph to obtain the prospecting conceptual model.

[0074] By constructing the prospecting conceptual model based on the knowledge graph, the accuracy of the prospecting conceptual model can be improved, which is beneficial to enhancing the identification and analysis capabilities of mineral resources.

[0075] In this embodiment, as shown in Table 1, the prospecting conceptual model can adopt the following representation form:

[0076]

[0077] Table 1 Schematic Table of Prospecting Conceptual Model

[0078] Step B, according to the prospecting conceptual model and the prospecting geological knowledge of the target mining area, obtain multiple characteristic variables, and through spatial interpolation, interpolate the discrete and discontinuous sampling points of the characteristic variables in the target mining area into continuous data to generate a prediction characteristic layer.

[0079] Specifically, prospecting geological knowledge mainly involves the distribution, genesis, exploration, and development of mineral resources. The prospecting conceptual model contains relevant mineral information within the target mining area. Based on the prospecting conceptual model and prospecting geological knowledge, multiple characteristic variables can be selected from them. As shown in Table 2, the characteristic variables can include data such as geological, remote sensing, and geochemical data. To reflect the spatial correlation of the data, and then through spatial interpolation processing (as a non-limiting example, spatial interpolation processing can include the inverse distance weighted method, Kriging method), the discrete and discontinuous sampling points of the characteristic variables in the target mining area are interpolated into continuous data to generate a predicted characteristic layer, so that the predicted characteristic layer can more truly represent the actual mineral distribution in the target mining area.

[0080]

[0081] Table 2 Schematic Table of Characteristic Variables

[0082] Step C, determine the weights of each of the said characteristic variables.

[0083] Referring to Table 2, there are multiple characteristic variables, and the types of the multiple characteristic variables are not completely the same. Thus, the weights of each characteristic variable are also different. Therefore, when making predictions, the importance levels of different characteristic variables are different, and thus the emphasis points in the prediction of the finally formed prediction model also vary.

[0084] In this embodiment, referring to Figure 2 , the steps for determining the weights of each of the said characteristic variables may include:

[0085] Step C1, determine the first score and the second score of each characteristic variable.

[0086] Specifically, based on the characteristic variables included in Table 2, the first score and the second score of each characteristic variable can be determined in advance, where the first score and the second score can refer to the scores of the characteristic variables under different evaluation systems.

[0087] In this embodiment, the steps of determining the first score and the second score of each feature variable include: extracting the key information elements in each feature variable and determining the key information expression of each feature variable; inputting the key information expressions of each feature variable into a preset knowledge graph model to obtain multiple key nodes corresponding to each key information expression; using the knowledge graph model to map the multiple key nodes of the key information expression to obtain multiple vector representations corresponding to each key information expression; fusing the multiple vector representations corresponding to each key information expression and each key information expression to form a fusion variable corresponding to each feature variable; inputting the fusion variable corresponding to each feature variable into a pre-trained neural network, and respectively obtaining the first confidence level and the second confidence level of each fusion variable under different confidence conditions, and using the first confidence level as the first score and the second confidence level as the second score.

[0088] By determining the first score and the second score of each feature variable based on the knowledge graph model, the knowledge graph model can fully understand the importance of each feature variable to obtain more reliable and accurate first and second scores.

[0089] Step C2: Compare the first feature vector with the highest first score with other feature variables, and determine the first score vector based on the relative importance between the first feature vector and other feature variables.

[0090] Specifically, based on step C1, the feature variable with the highest first score can be obtained from multiple feature variables as the first feature vector (it can be understood that the number of first feature vectors is at least 1), so that the first score vector of each feature variable in this case can be determined according to the first scores of other feature variables with the first feature vector as the reference benchmark.

[0091] Step C3: Compare the second feature vector with the lowest second score with other feature variables, and determine the second score vector based on the relative importance between the second feature vector and other feature variables.

[0092] Specifically, based on step C1, the feature variable with the lowest second score can be obtained from multiple feature variables as the second feature vector (it can be understood that the number of second feature vectors is at least 1), so that the second score vector of each feature variable in this case can be determined according to the second scores of other feature variables with the second feature vector as the reference benchmark.

[0093] Step C4: Based on each characteristic variable, the first scoring vector values of each characteristic variable in the first scoring vector, and the second scoring vector values of each characteristic variable in the second scoring vector, use the BWM knowledge embedding model to determine the weights of each characteristic variable.

[0094] Specifically, the BWM (Best-Worst Method) knowledge embedding model integrates the knowledge embedding technology of the best-worst method. After determining each characteristic variable, the first scoring vector and the second scoring vector of each characteristic variable through Steps C1 to C3, the weights of each characteristic variable can be determined based on the processing logic of the BWM knowledge embedding model itself.

[0095] As an optional example, refer to Figure 3 the schematic diagram of the acquisition process of the weights of each characteristic variable in the present embodiment shown in Figure 3 where (a) represents the relative importance of the assignments between each characteristic variable. Among them, the smaller the value, the lower the relative importance of the characteristic variable (that is, 9 is the most important and 1 is the least important). Figure 3 where (b) is used to represent the weights of each characteristic variable after Steps C1 to C4, and it can be seen from (b) that the weights of different characteristic variables are different.

[0096] For example, the weight of the albite spectrum is 0.16, while the weight of B is 0.025.

[0097] It should be noted that Figure 3 the schematic diagram of the acquisition process of the weights of the characteristic variables shown is only for illustrative purposes and should not be construed as a limitation of the present invention.

[0098] Step D: Generate a prediction model based on the weights of each of the characteristic variables and the pre-set deposit information feature deep network.

[0099] Specifically, all characteristic variables can characterize the mineral distribution in the target mining area, and the weights of the characteristic variables reflect the importance of the characteristic variables. In this way, based on the weights of each characteristic variable, the pre-set deposit information feature deep network can be pre-trained, and thus the corresponding prediction model can be generated.

[0100] Refer to Figure 4 the schematic diagram of the generation process of the prediction model in the embodiment of the present invention shown in Figure 4 As shown, based on the prospecting geological knowledge and in cooperation with the BWM knowledge embedding model, the weights of the characteristic weight variables can be determined.

[0101] Next, input the weights of the feature variables into the loss function of the ore deposit information feature deep network to determine the loss value of the ore deposit information feature deep network. Then, based on the loss value, use the gradient descent method for backpropagation to determine the parameter gradients of the ore deposit information feature deep network. Finally, based on the parameter gradients, use the parameter optimization algorithm to update the weight parameters between different layers of the ore deposit information feature deep network to generate the prediction model (as a non-limiting example, by updating the weight parameters between the same layers, input all the feature variables into the ore deposit information feature deep network again and perform forward propagation. When one or more verification evaluation indicators such as the mean square error and absolute error of the prediction results meet the requirements, the prediction model can be determined).

[0102] Among them, based on the parameter gradients, using the parameter optimization algorithm to update the weight parameters between different layers of the ore deposit information feature deep network can refer to the description in the existing solutions. The focus of this solution is: how to determine the loss value based on the weights of the feature variables.

[0103] In this way, based on the weights of the feature variables and the loss function of the ore deposit information feature deep network, a data-driven intelligent prediction model can be obtained, enhancing the generalization ability and result reliability of the model

[0104] In this embodiment, the loss value is determined based on the following formula:

[0105]

[0106] Among them, N is the number of samples, C is the number of categories (as a non-limiting example, if it is a binary classification task, then C = 2); y ij is the true label of the j-th category in the sample X i (as a non-limiting example, if the sample belongs to category j, then y ij = 1, otherwise y ij = 0); is the predicted probability of the j-th category in the sample X i ; w j is the weight, that is, the weight of the feature variable obtained through step C.

[0107] In this embodiment, the ore deposit information feature deep network can be a network architecture mainly composed of fully connected layers, including: an input layer composed of multiple first neurons; an output layer composed of multiple second neurons; and multiple hidden layers for connecting the input layer and the output layer, and the hidden layers are interconnected with each other.

[0108] Step E, generate multiple samples based on the prediction feature layer. Among them, one sample corresponds to one unit grid, and multiple unit grids form the prediction feature layer.

[0109] Among them, the samples include positive samples, negative samples, and samples to be predicted.

[0110] Specifically, after the prediction feature layer is cropped and gridded, multiple layers are merged into multi-channel data to adapt to the input of the model. After determining the grids where the positive and negative sample labels are located according to the positive and negative sample label setting method, the grids at this time are data including label data and feature variables. The grid where the positive sample label is located is regarded as a positive sample, the grid where the negative sample label is located is a negative sample, and the remaining grids with unknown labels are samples to be predicted.

[0111] Among them, positive samples are used to represent ore points, negative samples are used to represent non-ore points, and the content represented by the samples to be measured is unknown and needs to be determined by this solution.

[0112] In other words, the correctness of label setting has a great impact on the performance and prediction accuracy of the intelligent prediction algorithm. The prediction feature layer can represent the geological sketch of the area. Therefore, by processing the prediction feature layer, positive and negative samples can be generated.

[0113] It should be noted that when generating multiple samples, the following conditions should be met:

[0114] Representativeness: The positive and negative samples should be representative and able to reflect the geological characteristics and mineral distribution in the study area.

[0115] Independence: The positive and negative samples should be independent of each other to avoid the influence of the correlation between samples on model training.

[0116] Balance: The number of positive and negative samples should be balanced to avoid the influence of the imbalance of the positive and negative sample ratio on model training.

[0117] Accuracy: The labels of the positive and negative samples should be accurate to avoid the influence of label errors on model training.

[0118] In this embodiment, the sample is a positive sample, and the positive sample is used to represent an ore point. In this case, at least one of the following methods can be used to generate multiple samples.

[0119] Method 1: See Figure 5Schematic diagram of the acquisition process of positive samples in the first embodiment shown. The prediction feature layer is gridded to generate multiple cells; according to the spatial correlation of mineralization, with the ore points in the target mining area as the center, a buffer zone is generated, and all samples within the buffer zone are used as initial samples, where the range of the buffer zone is determined based on the area of the target mining area and the number of ore points in the target mining area (as a non-limiting example, if the area of the target mining area is large and the number of ore points is large, the buffer distance can be appropriately reduced and the number of buffer zones can be appropriately increased); with the cell where the initial sample point is located as the base point, and any one cell is selected from multiple neighboring cells as an extended sample; the extended sample and the initial sample are processed by taking the difference to generate multiple samples.

[0120] By using Method 1, information loss can be avoided, the possibility of noise generation can be reduced, and at the same time, the quality of the generated positive samples can be guaranteed, which conforms to geological cognition.

[0121] Method 2: Refer to Figure 6 Schematic diagram of the acquisition process of positive samples in the second embodiment shown. The initial samples are labeled to generate initial samples with first label information and sample data for obtaining second label information; a preset supervised model is trained using the initial samples with the first label information to obtain a learning model; the sample data with the second label information is input into the learning model to generate sample data with pseudo-label information; the initial samples with the first label information and the sample data with the pseudo-label information are input into the learning model to generate multiple samples.

[0122] By using Method 2, a large number of positive samples can be generated, thereby improving the prediction accuracy of the prediction model.

[0123] In this embodiment, the samples are negative samples, and the negative samples are used to represent non-ore points. In this case, at least one of the following methods can be used to generate multiple samples:

[0124] Method 1: The prediction feature layer is gridded to generate multiple cells, where one cell corresponds to one feature value; the statistics of each feature value are calculated, and the samples corresponding to the cells outside the preset statistics are used as negative samples according to the statistics of each feature value and the preset statistics.

[0125] In short, by analyzing the distribution of each feature in the data, the thresholds of ore-induced anomalies and ore-forming anomalies are determined, and the areas within the thresholds can be used as negative samples.

[0126] Specifically, statistical analysis is performed on each eigenvalue to calculate statistical measures such as its mean and standard deviation, and a reasonable threshold range is set. For example, for lithological features, non-ore-bearing rock masses or rock masses far from ore-bearing rock masses must be selected; for the concentration of geochemical elements, the abnormal concentration threshold can be determined by analyzing the frequency distribution of the data.

[0127] Method 2: Based on the non-linear dimensionality reduction method, map the high-dimensional data in the predicted feature layer to a low-dimensional space, determine the abnormal area, and use the samples corresponding to the cells within the abnormal area as negative samples.

[0128] In short, the selection of negative samples based on the non-linear dimensionality reduction method, that is, using the non-linear dimensionality reduction method (such as PCA, t-SNE) to map the high-dimensional data to a low-dimensional space to identify non-abnormal areas.

[0129] Specifically, use PCA or t-SNE to reduce the data dimension to two or three dimensions. Visualize the points in these low-dimensional spaces to intuitively identify the distribution of the samples. Select the sample points far from the positive samples as negative samples.

[0130] Based on the above Methods 1 and 2, representative and accurate negative samples can be selected, thereby improving the prediction accuracy of the prediction model.

[0131] It should be noted that after determining the positive and negative samples, the positive and negative samples can also be labeled to better distinguish the positive and negative samples (as a non-limiting example, the label of the positive sample is +1, and the label of the negative sample is -1); then the positive and negative samples can be combined to form different data sets. By inputting all the data sets into the data-driven intelligent prediction model, evaluating the prediction results, and selecting the best setting method, an optimal input data set can be formed.

[0132] Step F: Input multiple of the said samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area.

[0133] Specifically, by adopting the above steps, a prediction model, a predicted feature layer, and multiple positive and negative sample labels are formed at the same time. The predicted feature layer can be used as a feature, and multiple positive and negative sample labels can be used as training sample labels. Thus, the mineralization probability of the samples with unknown labels is predicted based on each unit as the basic unit. By comprehensively studying the prediction results of the mineralization probability in the study area and using the cumulative frequency method to determine the target area grading situation, the prediction target area can be further delineated, and thus the mineral resources of the target mining area can be further determined, that is, the prediction result can be obtained.

[0134] Furthermore, refer to Figure 7Schematic diagram of the output of the prediction model shown. Based on the positive samples, negative samples, and samples to be predicted, as well as the first label corresponding to the positive samples, the second label corresponding to the negative samples, and the third label corresponding to the samples to be predicted, based on the prediction model, determine the mineralization probability of each unit grid; based on the mineralization probability of each unit grid, use the cumulative frequency method to determine the target area classification, and use the cumulative result as the prediction result of the mineral resources in the target mining area, where the prediction result of the mineral resources in the target mining area includes the delineation of the prediction target area and the mineralization probability within the prediction target area.

[0135] It should be noted that the reason for obtaining the prediction model based on weights and outputting the prediction result based on the samples obtained from the prediction feature layer in this solution is as follows: The existing deep learning loss function does not consider the weights of variables, resulting in a relatively low accuracy of the finally output prediction result. And as point data, ore deposits are difficult to reflect the spatial correlation of data.

[0136] Therefore, through spatial interpolation processing, the ore deposit is transformed into a prediction feature layer, and the subsequent input data is also a picture. And at the beginning, there are no training samples and samples to be predicted. To ensure the prediction effect of the model, it is very important to set correct training samples. It is necessary to interpolate the discrete sampling points of the existing target mining area (the sampling point data is the coordinates and the values of the corresponding characteristic variables) into a picture to obtain the characteristic variable layer. The prediction feature layer is equivalent to a feature, and there is no label at this time.

[0137] The optimal input data set obtained by setting the positive and negative sample labels is equivalent to finding the positions of the training samples. Then, find the positions of these training samples on the prediction feature layer. By image cropping, that is, gridification, the grid where the positive sample label is located is regarded as a positive sample, and the grid where the negative sample label is located is regarded as a negative sample. At this time, the grid contains the values of the features and the labels, and a grid can be regarded as a training sample. There are also grids without labels, which are samples to be predicted.

[0138] In this way, through the above method, on the one hand, the accuracy of the prediction model is improved; on the other hand, the accuracy of the samples is also improved, so that the prediction result is more accurate.

[0139] In this embodiment, to further improve the prediction accuracy of the prediction result of the mineral resources in the target mining area, before inputting the prediction feature layer and multiple samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area, the prediction method further includes:

[0140] Use the permutation importance, partial dependence plot, and model-agnostic local interpretation algorithm to deconstruct the prediction model and adjust the parameters of the prediction model.

[0141] Specifically, by deconstructing the prediction model, potential metallogenic knowledge can be discovered, forming an iterative process of re-knowledgeization to adjust the model parameters of the data-driven prediction model and obtain an optimized prediction result of mineral resources.

[0142] Moreover, since the unexplainable mineral prediction model lacks reliability because machine learning models are similar to black boxes and the internal operation of the model cannot be controlled, only attempts can be made among different parameters. Therefore, the research introduces algorithms such as Permutation Importance, Partial Dependence Plot (PDP), and Model-Agnostic Local Explanation (LIME).

[0143] Among them, Permutation Importance is an effective method for measuring feature importance that is independent of the prediction model. By changing the magnitude of the feature value of a certain feature, the importance of this value is demonstrated. The greater the change, the greater the importance.

[0144] The specific steps are as follows: 1) Select a feature after the prediction model is trained; 2) Set a random number for this feature and calculate the new prediction result; 3) Compare the new and old prediction results to obtain the impact of this feature on model prediction. The global feature weights obtained by applying to this solution can be compared with the prior feature weights calculated from geological knowledge in the model construction stage, thereby analyzing the similarities and differences in the relationship between features and mineralization in the data-driven model and the knowledge-driven model.

[0145] The basic idea of the Partial Dependence Plot is to change the feature of key concern for a machine learning model, control other features unchanged, and observe the impact of this feature on the prediction result. At the same time, PDP can illustrate the relationship between the feature variable and the prediction result.

[0146] When applying the Partial Dependence Plot to this solution, the influence trend of each feature on the mineralization prediction probability is different inside. Taking the two features of the geochemical element combination of Na2O + K2O and the circular structure as examples for analysis, it can reflect the fluctuation of the mineralization probability with the change of the feature value.

[0147] The Local Explanation algorithm is a model-agnostic local interpretability algorithm that can truly reflect the behavior of the classifier when predicting samples. Applying it to this solution, positive and negative samples are randomly selected from the labeled samples for single-sample feature analysis, and the intelligent prediction result is compared with the manually marked result to achieve cross-validation.

[0148] In summary, through the interpretability method, deconstruct the model prediction result, quantitatively evaluate the importance degree of prediction indicators, form a loss function integrating geological knowledge, adjust the model parameters, optimize the prediction effect, realize the update and iteration of deposit knowledge, and form a repeatable, quantifiable, and controllable closed loop of intelligent prospecting prediction research.

[0149] An embodiment of the present invention also provides a system corresponding to the method for quantitative prediction of mineral resources based on knowledge embedding, which can be abbreviated as "mineral resource quantitative prediction system 100". As Figure 8 shown in the structural schematic diagram of a mineral resource quantitative prediction system based on knowledge embedding in an embodiment of the present invention, as Figure 8 shown, the mineral resource quantitative prediction system 100 may include:

[0150] A construction unit 110, configured to construct a prospecting concept model according to the ore-forming geological bodies, ore-forming tectonic systems, and ore-forming characteristic markers related to the target mining area;

[0151] A first generation unit 120, configured to obtain a plurality of characteristic variables according to the prospecting concept model and the prospecting geological knowledge of the target mining area, and by means of spatial interpolation, interpolate the discrete and discontinuous characteristic variable sampling points in the target mining area into continuous data to generate a prediction characteristic layer;

[0152] A processing unit 130, configured to determine the weights of each of the characteristic variables; and generate a plurality of samples based on the prediction characteristic layer, wherein one sample corresponds to one unit grid, and a plurality of the unit grids form the prediction characteristic layer, and the sample includes a positive sample, a negative sample, and a sample to be predicted; and input the plurality of samples into a prediction model to obtain the mineral resource prediction result of the target mining area;

[0153] A second generation unit 140, configured to generate the prediction model according to the weights of each of the characteristic variables and a pre-set ore deposit information feature depth network.

[0154] Among them, the specific working processes and principles of the construction unit 110, the first generation unit 120, the processing unit 130, and the second generation unit 140 can refer to the relevant descriptions in the foregoing examples.

[0155] It can be understood that the above division of each unit is only a division of logical functions. In actual implementation, all or part of them can be integrated into a physical entity, or physically separated. In addition, the above units can be implemented in the form of a processor calling software.

[0156] This embodiment also provides a computer system suitable for implementing the method for extracting key information from legal documents. Among them, the computer system includes a Central Processing Unit (CPU), which can perform various appropriate actions and processes according to the program stored in the Read-Only Memory (ROM) or the program loaded from the storage section into the Random Access Memory (RAM), such as executing the method described in the above embodiment. In the RAM, various programs and data required for system operation are also stored. The CPU, ROM, and RAM are connected to each other via a bus. The Input / Output (I / O) interface is also connected to the bus.

[0157] The following components are connected to the I / O interface: an input section including a keyboard, a mouse, etc.; an output section including, for example, a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), etc. and a speaker, etc.; a storage section including a hard disk, etc.; and a communication section including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section performs communication processing via a network such as the Internet. The drive 310 is also connected to the I / O interface as needed. Removable media, such as magnetic disks, optical discs, magneto-optical discs, semiconductor memories, etc., are installed on the drive as needed so that the computer programs read from them can be installed into the storage section as needed.

[0158] Specifically, according to the embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments of the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication section, and / or installed from the removable media. When the computer program is executed by the Central Processing Unit (CPU), various functions defined in the system of the present application are executed.

[0159] It should be noted that the computer-readable medium shown in the embodiments of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system

[0160] An apparatus or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0161] In this application, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. And in this application, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable computer program is carried. Such a propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0162] The flowcharts and block diagrams in the figures illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of this application. Among them, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code, and the above module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the figures. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0163] The units involved in the embodiments described in this disclosure may be implemented in software or in hardware, and the described units may also be provided in a processor. Among them, the names of these units do not constitute a limitation to the unit itself in certain cases.

[0164] According to one aspect of the present application, there is provided a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the above various alternative implementation manners.

[0165] As another aspect, the present application further provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or may exist alone without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the one or more programs are executed by an electronic device, the electronic device implements the methods described in the above embodiments.

[0166] It should be noted that although several modules or units of a device for action execution are mentioned in the above detailed description, such a division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of the two or more modules or units described above may be embodied in one module or unit. Conversely, the features and functions of one module or unit described above may be further divided and embodied by multiple modules or units.

[0167] Through the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described here can be implemented by software or by a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (such as a personal computer, a server, a touch terminal, or a network device, etc.) to execute the methods according to the embodiments of the present application.

[0168] Although the present specification discloses as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention should be subject to the scope defined by the claims.

Claims

1. A quantitative prediction method for mineral resources based on knowledge embedding, characterized in that, Including: Construct a prospecting conceptual model according to the metallogenic geological bodies, metallogenic tectonic systems and metallogenic characteristic markers related to the target mining area; According to the prospecting conceptual model and the prospecting geological knowledge of the target mining area, obtain multiple characteristic variables, and through spatial interpolation, interpolate the discrete and discontinuous characteristic variable sampling points in the target mining area into continuous data to generate a predicted characteristic layer; Determine the weights of each of the characteristic variables, including: determining the first score and the second score of each characteristic variable; comparing the first eigenvector with the highest first score with other characteristic variables, and based on the relative importance between the first eigenvector and other characteristic variables, determine the first score vector; comparing the second eigenvector with the lowest second score with other characteristic variables, and based on the relative importance between the second eigenvector and other characteristic variables, determine the second score vector; according to each characteristic variable, and the first score vector value of each characteristic variable in the first score vector, and the second score vector value of each characteristic variable in the second score vector, use the BWM knowledge embedding model to determine the weights of each characteristic variable; Generate a prediction model according to the weights of each of the characteristic variables and the pre-set ore deposit information characteristic deep network, including: inputting the weights of the characteristic variables into the loss function of the ore deposit information characteristic deep network to determine the loss value of the ore deposit information characteristic deep network, where the loss value is determined based on the following formula; based on the loss value, use the gradient descent method for backpropagation to determine the parameter gradient of the ore deposit information characteristic deep network; based on the parameter gradient, use the parameter optimization algorithm to update the weight parameters between different layers of the ore deposit information characteristic deep network to generate the prediction model; based on the predicted characteristic layer, generate multiple samples, where one sample corresponds to one unit grid, and multiple of the unit grids form the predicted characteristic layer, and the sample includes positive samples, negative samples and samples to be predicted; Input the multiple samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area; Wherein, the loss value is determined based on the following formula: Among them, N is the number of samples and C is the number of categories; y ij is the true label of the j-th category in the sample X i where the sample X i is a sample in N; is the predicted probability of the j-th category in the sample X i ; w j is the weight.

2. The quantitative prediction method of mineral resources based on knowledge embedding according to claim 1, characterized in that The steps of determining the first score and the second score of each characteristic variable include: Extract the key information elements in each characteristic variable and determine the key information expression of each characteristic variable; Input the key information expressions of each characteristic variable into the pre-set knowledge graph model to obtain multiple key nodes corresponding to each key information expression; Use the knowledge graph model to map the multiple key nodes of the key information expression to obtain multiple vector representations corresponding to each key information expression; Fuse the multiple vector representations corresponding to each key information expression and each key information expression to form a fusion variable corresponding to each characteristic variable; Input the fusion variables corresponding to each feature variable into a pre-trained neural network, and respectively obtain the first confidence level and the second confidence level of each fusion variable under different confidence conditions, and use the first confidence level as the first score and the second confidence level as the second score.

3. The quantitative prediction method of mineral resources based on knowledge embedding according to claim 1, wherein The sample is a positive sample, and the positive sample is used to represent a mining point; The step of generating multiple samples based on the predicted feature layer includes at least one of the following: Perform grid processing on the predicted feature layer to generate multiple cells; according to the spatial correlation of mineralization, take the mining points in the target mining area as the center, generate a buffer zone, and use all the samples in the buffer zone as initial samples, where the range of the buffer zone is determined based on the area of the target mining area and the number of mining points in the target mining area; Take the cell where the initial sample point is located as the base point, and select any one of multiple neighboring cells as an extended sample; perform a difference process on the extended sample and the initial sample to generate multiple of the samples; Label the initial sample to generate an initial sample with first label information and obtain sample data with second label information; use the initial sample with the first label information to train a preset supervised model to obtain a learning model; input the sample data with the second label information into the learning model to generate sample data with pseudo-label information; input the initial sample with the first label information and the sample data with the pseudo-label information into the learning model to generate multiple of the samples.

4. The quantitative prediction method of mineral resources based on knowledge embedding according to claim 1, characterized in that The sample is a negative sample, and the negative sample is used to represent a non-mining point; The step of generating multiple samples based on the predicted feature layer includes at least one of the following: Perform grid processing on the predicted feature layer to generate multiple cells, where one cell corresponds to one feature value; calculate the statistics of each feature value, and use the samples corresponding to the cells outside the preset statistics as negative samples according to the statistics of each feature value and the preset statistics; Based on a non-linear dimensionality reduction method, map the high-dimensional data in the predicted feature layer to a low-dimensional space, determine the abnormal area, and use the samples corresponding to the cells in the abnormal area as negative samples.

5. The quantitative prediction method of mineral resources based on knowledge embedding according to claim 1, wherein Before inputting multiple of the samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area, the prediction method further includes: Based on the positive samples, negative samples and samples to be predicted, as well as the first label corresponding to the positive samples, the second label corresponding to the negative samples and the third label corresponding to the samples to be predicted, determine the mineralization probability of each unit grid based on the prediction model; Based on the mineralization probability of each unit grid, use the cumulative frequency method to determine the target area grading situation, and use the cumulative result as the prediction result of the mineral resources in the target mining area, where the prediction result of the mineral resources in the target mining area includes the delineation of the predicted target area and the mineralization probability in the predicted target area.

6. The quantitative prediction method of mineral resources based on knowledge embedding according to any one of claims 1 to 5, characterized in that Before inputting multiple of the samples into the prediction model to obtain the prediction result of the mineral resources in the target mining area, the prediction method further includes: The prediction model is deconstructed by using permutation importance, partial dependence plots, and a model-agnostic local interpretation algorithm, and the parameters of the prediction model are adjusted.

7. A quantitative prediction system for mineral resources based on knowledge embedding, characterized in that, It includes: A construction unit configured to construct a prospecting conceptual model according to ore-forming geological bodies, ore-forming tectonic systems, and ore-forming characteristic markers related to the target mining area; A first generation unit configured to obtain a plurality of characteristic variables according to the prospecting conceptual model and the prospecting geological knowledge of the target mining area, and by means of spatial interpolation, interpolate the discrete and discontinuous characteristic variable sampling points in the target mining area into continuous data to generate a prediction characteristic layer; A processing unit configured to determine the weights of each of the characteristic variables, including: determining a first score and a second score for each of the characteristic variables; comparing the first eigenvector with the highest first score with other characteristic variables, and determining a first score vector based on the relative importance between the first eigenvector and other characteristic variables; comparing the second eigenvector with the lowest second score with other characteristic variables, and determining a second score vector based on the relative importance between the second eigenvector and other characteristic variables; according to each of the characteristic variables, and the first score vector value of each characteristic variable in the first score vector, and the second score vector value of each characteristic variable in the second score vector, using a BWM knowledge embedding model to determine the weights of each of the characteristic variables; and generating a plurality of samples based on the prediction characteristic layer, where one sample corresponds to one unit grid, and a plurality of the unit grids form the prediction characteristic layer, and the sample includes a positive sample, a negative sample, and a sample to be predicted; and inputting the plurality of samples into a prediction model to obtain a mineral resource prediction result of the target mining area; A second generation unit configured to generate the prediction model according to the weights of each of the characteristic variables and a preset deposit information characteristic deep network, including: inputting the weights of the characteristic variables into the loss function of the deposit information characteristic deep network to determine the loss value of the deposit information characteristic deep network, where the loss value is determined based on the following formula; based on the loss value, using the gradient descent method for backpropagation to determine the parameter gradient of the deposit information characteristic deep network; based on the parameter gradient, using a parameter optimization algorithm to update the weight parameters between different layers of the deposit information characteristic deep network to generate the prediction model; Wherein, the loss value is determined based on the following formula: Where N is the number of samples, C is the number of classes; y ij is the true label of the j-th class in the sample X i where the sample X i is a sample in N; is the predicted probability of the j-th class in the sample X i and w j is the weight.

8. An electronic device, characterized in that, It includes: At least one memory and at least one processor, the memory stores one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the knowledge-embedding-based quantitative prediction method for mineral resources according to any one of claims 1 to 6.

Images