Method and system for constructing deep learning model of mineral prediction coupled with data and knowledge

Abstract

The application discloses a kind of data and knowledge coupling's mineral prediction deep learning model construction method and system, method includes the following steps: selecting the convolutional neural network considering adjacent unit spatial structure and relationship;Buffer zone analysis is carried out to closely related ore-controlling factors with mineralization, and the buffer width and deposit distribution density are obtained;Ore-controlling factor layer is valued;The fully connected hidden layer considering the coupling relationship between ore-controlling factor and deposit space is constructed;Deep learning loss function based on the spatial distribution law of deposit is constructed;According to the valued ore-controlling factor layer, hidden layer structure and deep learning loss function, the data and knowledge coupling's mineral prediction deep learning model is constructed.The application constructs data and knowledge coupling's double-driven mineral prediction deep learning model, improves the reliability of mineralization information integration in study area and the accuracy of prediction result, so as to provide reliable key evidence for next step prospecting work.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mineral resources, in particular to a data and knowledge coupled mineral prediction deep learning model construction method and system. BACKGROUND

[0002] At present, the mineral exploration theory and method taking the genetic model and the combination of ore-prospecting indicators as the core has been difficult to meet the needs of the deep exploration and the exploration of the covered area characterized by weak information, mixed information and incomplete information.

[0003] As a hierarchical machine learning algorithm with multiple nonlinear transformations, deep learning can effectively mine complex and nonlinear geoscience spatial data and extract unknown patterns of geological processes, has strong classification and prediction accuracy, and is becoming an important tool in the field of mineral prediction.

[0004] However, the pure data-driven deep learning mainly focuses on the spatial correlation between the ore-prospecting information and the deposit, ignores the related ore-forming mechanism knowledge in the mineral prediction theory, so that the model can reach a high accuracy rate after multiple training, but it is difficult to correctly model from a physical point of view, leading to the inconsistency between the decision results and the field cognition, and the decision mechanism is also difficult to verify and understand, thereby restricting the credibility and interpretability of the deep learning model, and limiting the further application of deep learning in mineral prediction. At present, how to construct a deep learning model driven by data and knowledge to improve the reliability of the integration of mineralization information in the study area and the accuracy of the prediction results, so as to provide reliable key evidence for the next step of the prospecting work has not been reported. SUMMARY

[0005] The purpose of the present application is to overcome the shortcomings of the background art, and provide a data and knowledge coupled mineral prediction deep learning model construction method and system.

[0006] In a first aspect, the present application provides a data and knowledge coupled mineral prediction deep learning model construction method, comprising the following steps:

[0007] A convolutional neural network considering the spatial structure and relationship of adjacent units is selected;

[0008] A buffer zone analysis is performed on the ore-controlling elements closely related to mineralization to obtain the buffer zone width and the deposit distribution density;

[0009] According to the obtained buffer zone width and the deposit distribution density, a power law function of the deposit distribution density and the buffer zone width is constructed, and the constraint weight is obtained according to the constructed power law function and the ore-controlling element layer is valued;

[0010] A fully connected hidden layer considering the spatial coupling relationship between the ore-controlling elements and the deposit is constructed;

[0011] Construct a deep learning loss function based on the spatial distribution of the deposit;

[0012] According to the assigned ore-controlling factor layer, the hidden layer structure and the deep learning loss function, the training process of the selected convolutional neural network is provided with geological prior knowledge guidance, and a mineral prediction deep learning model coupled with data and knowledge is constructed.

[0013] According to the first aspect, in a first possible implementation manner of the first aspect, the step of selecting a convolutional neural network considering the spatial structure and relationship of adjacent units specifically includes the following steps:

[0014] According to the advantages of different deep learning model structures and the characteristics of geological prospecting big data, a convolutional neural network considering the spatial structure and relationship of adjacent units is selected;

[0015] Based on the supervised learning mode of the convolutional neural network, a mineral prediction technical process is established.

[0016] According to the first aspect, in a second possible implementation manner of the first aspect, the step of performing buffer zone analysis on the ore-controlling factors closely related to mineralization to obtain the buffer zone width, the cumulative number of deposits and the buffer zone deposit distribution density specifically includes the following steps:

[0017] The buffer zone analysis is performed on the ore-controlling factors closely related to mineralization, and the buffer zone width and the buffer interval are determined according to the geological characteristics of the area where the deposit is located;

[0018] The number of deposits falling in each buffer zone of each ore-controlling factor is counted to obtain the deposit distribution density of each buffer zone.

[0019] According to the first aspect, in a third possible implementation manner of the first aspect, the step of constructing a power law function of deposit distribution density and buffer zone width according to the obtained buffer zone width and deposit distribution density, and obtaining the weight according to the constructed power law function and assigning values to the ore-controlling factor layer specifically includes the following steps:

[0020] According to the obtained buffer zone width and deposit distribution density, a power law function of deposit distribution density and buffer zone width is constructed as follows:

[0021] ρ=cd a-2 ,

[0022] In the formula, c is a constant, and a is a singularity index;

[0023] The power law function of deposit distribution density and buffer zone width is normalized to obtain the constraint weight of different buffer zones on the formation of deposits;

[0024] Constrained weight assignment is performed on the ore-controlling element layer.

[0025] According to the first aspect, in a fourth possible implementation manner of the first aspect, the step of constructing the fully connected hidden layer considering the spatial coupling relationship between the ore-controlling element and the deposit specifically includes the following steps:

[0026] According to the training method of the convolutional neural network, the mathematical principle of the fully connected hidden layer and the geological connotation of the ore-controlling element, the fully connected hidden layer considering the spatial coupling relationship between the ore-controlling element and the deposit is constructed.

[0027] According to the first aspect, in a fifth possible implementation manner of the first aspect, the step of constructing the deep learning loss function based on the spatial distribution law of the deposit specifically includes the following steps:

[0028] The function relationship between the ore-controlling element and the convolutional neural network prediction of the mineralization probability is constructed, and is combined with the cross-entropy function of the convolutional neural network to obtain the deep learning loss function based on the spatial distribution law of the deposit.

[0029] According to the first aspect, in a sixth possible implementation manner of the first aspect, the step of constructing the data and knowledge coupled mineral prediction deep learning model specifically includes the following steps:

[0030] According to the assigned ore-controlling element layer, the hidden layer structure and the deep learning loss function, the ore-controlling element layer is taken as an input variable, and the convolutional neural network structure, the constructed fully connected hidden layer and the deep learning loss function are combined to comprehensively establish a mineral prediction model, so as to provide geological prior knowledge guidance to the model training process, thereby constructing the data and knowledge coupled mineral prediction deep learning model.

[0031] The second aspect is a data and knowledge coupled mineral prediction deep learning model construction system, which includes:

[0032] The convolutional neural network selection module is used to select the convolutional neural network considering the spatial structure and relationship of adjacent units.

[0033] The buffer information acquisition module is used to perform buffer analysis on the ore-controlling element closely related to mineralization, and to acquire the buffer width and the deposit distribution density.

[0034] The ore-controlling element layer assignment module is in communication connection with the buffer information acquisition module, and is used to construct a power law function of the deposit distribution density and the buffer width according to the acquired buffer width and the deposit distribution density, and to acquire a constraint weight according to the constructed power law function and to assign the ore-controlling element layer.

[0035] The full-connection hidden layer construction module is in communication connection with the convolutional neural network selection module, and is used for constructing a full-connection hidden layer considering the spatial coupling relationship of ore-controlling elements and ore deposits.

[0036] The deep learning loss function construction module is in communication connection with the convolutional neural network selection module, and is used for constructing a deep learning loss function based on the spatial distribution law of ore deposits.

[0037] The mineral prediction deep learning model construction module is in communication connection with the ore-controlling element layer assignment module, the full-connection hidden layer construction module and the deep learning loss function construction module, and is used for providing geological prior knowledge guidance to the training process of the selected convolutional neural network according to the assigned ore-controlling element layer, the hidden layer structure and the deep learning loss function, and constructing a data and knowledge coupled mineral prediction deep learning model.

[0038] According to a second aspect, in a first possible implementation manner of the second aspect, the convolutional neural network selection module comprises:

[0039] The convolutional neural network selection unit is used for selecting a convolutional neural network considering the spatial structure and relationship of adjacent units according to the advantages of different deep learning model structures and the characteristics of geological prospecting big data.

[0040] The mineral prediction process establishment unit is used for establishing a mineral prediction technical process thereof based on a supervised learning mode of the convolutional neural network.

[0041] According to the second aspect, in a second possible implementation manner of the second aspect, the buffer information acquisition module comprises:

[0042] The buffer information acquisition unit is used for performing buffer analysis on ore-controlling elements closely related to mineralization, and determining the buffer width and buffer interval according to the geological characteristics of the region where the ore deposit is located.

[0043] The deposit distribution density acquisition unit is used for statistically acquiring the number of ore deposits falling in each buffer of the ore-controlling element, and acquiring the deposit distribution density of each buffer.

[0044] Compared with the prior art, the application has the following advantages:

[0045] The data and knowledge coupled mineral prediction deep learning model construction method provided by the application provides geological prior knowledge guidance to the training process of the selected convolutional neural network according to the assigned ore-controlling element layer, the hidden layer structure and the deep learning loss function, constructs a data and knowledge coupled double-driven mineral prediction deep learning model, improves the reliability of mineralization information integration in the study area and the accuracy of the prediction result, and thus provides reliable key evidence for the next prospecting work. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is a method flowchart of a data and knowledge coupled mineral prediction deep learning model construction method of an embodiment of the present application;

[0047] Figure 2 is a buffer zone analysis example graph taking a fault as an example of an embodiment of the present application;

[0048] Figure 3 is a weighted function graph taking a fault as an example of an embodiment of the present application;

[0049] Figure 4 is a data and knowledge coupled mineral prediction deep learning model structure schematic diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0050] Reference will now be made in detail to the present application, examples of which are illustrated in the accompanying drawings. While the present application will be described in conjunction with the specific embodiments, it will be understood that the present application is not intended to be limited to the described embodiments. On the contrary, the present application is intended to cover modifications, alternatives and equivalents, which are within the scope of the present application as defined by the appended claims. It should be noted that the steps of the methods described herein can be implemented by any of the functional blocks or functional arrangements, and any of the functional blocks or functional arrangements can be implemented as physical entities or logical entities, or a combination of both.

[0051] In order for those skilled in the art to better understand the present application, the present application will be further described in detail below in conjunction with the drawings and specific embodiments.

[0052] Note: The examples to be introduced next are only one specific example, and are not as a limitation on the embodiments of the present application must be as follows specific steps, values, conditions, data, order, etc. Those skilled in the art can use the concept of the present application to construct more embodiments not mentioned in the present specification by reading the present specification.

[0053] The purpose of the present application is to overcome the deficiencies of the above background art, provide a data and knowledge coupled mineral prediction deep learning model construction method and system, to solve the technical problem that the pure data driven deep learning in the prior art is difficult to correctly model from a physical point of view, which restricts the credibility and explainability of the deep learning model, and limits the further application of deep learning in mineral prediction, so that it can consider the ore-forming theory knowledge to reduce the phenomenon that the delineation of the prospecting area does not match the geological cognition, and further improve the effectiveness and explainability of mineral prediction based on deep learning.

[0054] In a first aspect, with reference to Figure 1 The present application provides a data and knowledge coupled mineral prediction deep learning model construction method, comprising the following steps:

[0055] Step S1, selecting a convolutional neural network considering the spatial structure and relationship of adjacent units;

[0056] Step S2, performing buffer zone analysis on ore-controlling elements closely related to mineralization to obtain buffer zone width and deposit distribution density;

[0057] Step S3, constructing a power-law function of deposit distribution density and buffer zone width according to the obtained buffer zone width and deposit distribution density, obtaining constraint weights according to the constructed power-law function, and assigning values to the ore-controlling element layer;

[0058] Step S4, constructing a fully connected hidden layer considering the spatial coupling relationship between ore-controlling elements and deposits;

[0059] Step S5, constructing a deep learning loss function based on the spatial distribution regularity of deposits;

[0060] Step S6, providing geological prior knowledge guidance to the training process of the selected convolutional neural network according to the assigned ore-controlling element layer, hidden layer structure, and deep learning loss function, and constructing a data and knowledge coupled mineral prediction deep learning model.

[0061] The present application provides a data and knowledge coupled mineral prediction deep learning model construction method, and a data and knowledge coupled double-driven mineral prediction deep learning model is constructed, which improves the reliability of mineralization information integration in the study area and the accuracy of the prediction results, and provides reliable key evidence for the next step of prospecting work.

[0062] In an embodiment, the step of selecting a convolutional neural network considering the spatial structure and relationship of adjacent units specifically includes the following steps:

[0063] According to the advantages of different deep learning model structures and the characteristics of geological prospecting big data, a convolutional neural network considering the spatial structure and relationship of adjacent units is selected to capture the spatial coupling relationship of mineralization information, and then to delineate the prospecting prospective area;

[0064] Based on the supervised learning method of convolutional neural network, the establishment of its mineral prediction technical process includes training set making, model building and parameter optimization, model training and testing, and prospecting prospective area delineation.

[0065] In an embodiment, as shown in Figure 2 The step of selecting ore-controlling elements closely related to mineralization to perform buffer zone analysis to obtain buffer zone width and deposit distribution density specifically includes the following steps:

[0066] Selecting the ore-controlling factors related to mineralization (taking faults as an example) and using ArcGIS and other software to conduct buffer analysis, according to the geological characteristics of the area where the deposit is located, the buffer width and buffer interval are determined, and the buffer zone of different ore-controlling factors is made by using GIS;

[0067] The number of deposits falling in each buffer zone of each ore-controlling factor is counted respectively, and the relationship between the cumulative number of deposits n and the deposit distribution density p is calculated according to the formula p = n / d, wherein d represents the width of the buffer zone.

[0068] In an embodiment, the step of constructing a power-law function of deposit distribution density and buffer width according to the obtained buffer width and deposit distribution density, and assigning values to the ore-controlling factor layer according to the constructed power-law function, specifically includes the following steps:

[0069] According to the obtained buffer width and deposit distribution density, a double logarithmic scatter plot is made with the buffer width d as the abscissa and the deposit spatial distribution density p as the ordinate, a straight line is fitted by using the least square method, and the power-law function of deposit distribution density and buffer width is obtained as follows:

[0070] p = cd a-2 ,

[0071] In the formula, c is a constant, and a is a singularity index;

[0072] The power-law function of deposit distribution density and buffer width is normalized to obtain the constraint weight of different buffer zones on the formation of deposits as shown in the following formula:

[0073] e = cd a-2 / p max ,

[0074] Where p max is the maximum value of the power-law function of deposit distribution density and buffer distance, so that the closer the area to the ore-controlling factor, the greater the weight assigned, and the farther the area to the ore-controlling factor, the smaller the weight assigned. The weighting function of faults as an example is shown in Figure 3 The weight distribution formula is: In the formula, d is in kilometers;

[0075] Assigning constraint weights to the ore-controlling factor layer.

[0076] In an embodiment, the step of constructing a full-connection hidden layer considering the coupling relationship between ore-controlling factors and deposit space, specifically includes the following steps:

[0077] According to the training mode of the convolutional neural network, the mathematical principle of the fully connected hidden layer, and the geological connotation of the ore-controlling elements, a fully connected hidden layer considering the spatial coupling relationship between the ore-controlling elements and the ore deposit is constructed after the SoftMax layer of the convolutional neural network. Since a complete neuron is composed of a linear model and an activation function, the linear equation of the neuron in the hidden layer is where x represents the output probability of the SoftMax layer, k i is the value of each input sample and the grid cell corresponding to the i-th ore-controlling element layer in the convolutional neural network, ω is a trainable weight parameter, and b is a trainable bias term parameter. The output of the above linear equation is operated by using the Sigmoid activation function to map the neuron features, that is, where θ is a constant, so as to realize the construction of the fully connected hidden layer considering the spatial coupling relationship between the ore-controlling elements and the ore deposit by using a hard constraint method. Through continuous training of the model, the forward propagation process of the model can learn the spatial coupling relationship between the key ore-controlling elements and the sample; in the backward propagation, the difference between the predicted value and the target value of the sample is used to adjust the parameters in the reverse direction, so as to improve the prediction accuracy of the model. In addition, in the training process, the relationship between the sample and the spatial distribution of the key ore-controlling elements is adjusted by continuously updating the ω parameter in the forward propagation of the model, so as to improve the explainability and prediction accuracy of the deep learning model for mineral prediction.

[0078] In an embodiment, the step of constructing the deep learning loss function based on the spatial distribution law of the ore deposit specifically includes the following steps:

[0079] The cross-entropy loss function L CNN of the convolutional neural network is constructed as follows: Sigmoid (y))+(1-p)log(1-f Sigmoid (y))];

[0080] In the formula, p is the true label of the input sample, f Sigmoid (y) is the probability prediction value of the input sample.

[0081] The function relationship L geo between the ore-controlling elements and the predicted mineralization probability of the convolutional neural network is constructed as follows: sigmoid (y)-ε||2, L geo is a regularization term constructed based on the spatial distribution law of the ore deposit, and ε is the grid cell value of the ore-controlling element layer corresponding to the input sample, and L geo is obtained by using the L2 regularization formula.

[0082] The function relationship constructed by using the soft constraint method is combined with the cross-entropy loss function to obtain the deep learning loss function based on the spatial distribution law of the ore deposit as follows:

[0083] L total =L CNN +λL geo ,

[0084] In the formula, L geo L is a regularization term constructed based on the spatial distribution patterns of ore deposits. CNN Let L be the cross-entropy loss function of the convolutional neural network constructed using hard constraints. λ is a weighting coefficient that measures the relative importance of the cross-entropy loss function and the regularization term based on the spatial distribution patterns of ore deposits. Based on the constructed deep learning loss function L based on the spatial distribution patterns of ore deposits... total This loss function allows the model to apply a smaller error penalty term during training to areas with high mineralization probability that are closer to the ore-controlling element or areas with low mineralization probability that are farther away from the ore-controlling element; conversely, it applies a larger error penalty term to areas with low mineralization probability that are closer to the ore-controlling element or areas with high mineralization probability that are farther away from the ore-controlling element. The purpose of this loss function is to enable the deep learning model to learn the spatial distribution patterns of the deposits in the study area during backpropagation, thus achieving soft constraints on the deep learning model.

[0085] In one embodiment, a deep learning model for mineral prediction, coupling data and knowledge, is constructed using a combination of soft and hard constraints to delineate prospective mineral exploration areas, such as... Figure 4 As shown.

[0086] In a second aspect, based on the same inventive concept, the application provides a data and knowledge coupled mineral prediction deep learning model construction system, comprising a convolutional neural network selection module, a buffer information acquisition module, an ore-controlling element layer assignment module, a fully connected hidden layer construction module, a deep learning loss function construction module, and a mineral prediction deep learning model construction module. The convolutional neural network selection module is used to select a convolutional neural network that considers the spatial structure and relationship of adjacent units. The buffer information acquisition module is used to analyze the buffer of ore-controlling elements closely related to mineralization, and to obtain the buffer width and deposit distribution density. The ore-controlling element layer assignment module is in communication with the buffer information acquisition module, and is used to construct a power law function of deposit distribution density and buffer width according to the obtained buffer width and deposit distribution density, and to obtain a constraint weight according to the constructed power law function and assign the ore-controlling element layer. The fully connected hidden layer construction module is in communication with the convolutional neural network selection module, and is used to construct a fully connected hidden layer that considers the spatial coupling relationship between ore-controlling elements and deposits. The deep learning loss function construction module is in communication with the convolutional neural network selection module, and is used to construct a deep learning loss function based on the spatial distribution regularity of deposits. The mineral prediction deep learning model construction module is in communication with the ore-controlling element layer assignment module, the fully connected hidden layer construction module, and the deep learning loss function construction module, and is used to provide geological prior knowledge guidance to the training process of the selected convolutional neural network according to the assigned ore-controlling element layer, the hidden layer structure, and the deep learning loss function, and to construct a data and knowledge coupled mineral prediction deep learning model.

[0087] In an embodiment, the convolutional neural network selection module comprises a convolutional neural network selection unit and a mineral prediction process establishment unit. The convolutional neural network selection unit is used to select a convolutional neural network that considers the spatial structure and relationship of adjacent units according to the advantages of different deep learning model structures and the characteristics of geological prospecting big data. The mineral prediction process establishment unit is used to establish a mineral prediction technical process based on the supervised learning method of convolutional neural networks.

[0088] In an embodiment, the buffer information acquisition module comprises a buffer information acquisition unit and a deposit distribution density acquisition unit. The buffer information acquisition unit is used to select ore-controlling elements closely related to mineralization for buffer analysis, and to determine the buffer width and buffer interval according to the geological characteristics of the area where the deposit is located. The deposit distribution density acquisition unit is used to statistically obtain the number of deposits falling within each buffer of the ore-controlling elements, and to obtain the deposit distribution density of each buffer.

[0089] Although the present application has been described with reference to the embodiments above, various changes and modifications can be suggested to one skilled in the art, and it is intended that the present application encompass such changes and modifications as fall within the scope of the appended claims. Particularly, each feature of the disclosed embodiments can be combined with each other feature, provided that the combination does not result in structural absurdity. Therefore, the present application is not intended to be limited to the particular embodiments disclosed in the specification, but includes all the technical solutions falling within the scope of the claims.

Claims

1. A data and knowledge coupled mineral prediction deep learning model construction method, characterized in that, The method comprises the following steps: select a convolutional neural network considering the spatial structure and relationship of adjacent units; conduct buffer zone analysis on the ore-controlling elements closely related to mineralization to obtain buffer zone width and deposit distribution density; construct a power-law function of deposit distribution density and buffer zone width according to the obtained buffer zone width and deposit distribution density, obtain the weight according to the constructed power-law function, and assign values to the ore-controlling element layer; construct a fully connected hidden layer considering the spatial coupling relationship between the ore-controlling elements and the deposits; construct a deep learning loss function based on the spatial distribution law of the deposits; provide geological prior knowledge guidance to the training process of the selected convolutional neural network according to the assigned ore-controlling element layer, the hidden layer structure, and the deep learning loss function, and construct a data and knowledge coupled mineral prediction deep learning model. The step of constructing a power-law function of deposit distribution density and buffer zone width according to the obtained buffer zone width and deposit distribution density, and assigning values to the ore-controlling element layer according to the constructed power-law function, specifically comprises the following steps: The step of constructing a power-law function of deposit distribution density and buffer zone width according to the obtained buffer zone width and deposit distribution density, and assigning values to the ore-controlling element layer according to the constructed power-law function, specifically comprises the following steps: The step of constructing a power-law function of deposit distribution density and buffer zone width according to the obtained buffer zone width and deposit distribution density, and assigning values to the ore-controlling element layer according to the constructed power-law function, specifically comprises the following steps: The step of constructing a fully connected hidden layer considering the spatial coupling relationship between the ore-controlling elements and the deposits, specifically comprises the following steps: = According to the training method of the convolutional neural network, the mathematical principle of the fully connected hidden layer, and the geological connotation of the ore-controlling elements, a fully connected hidden layer considering the spatial coupling relationship between the ore-controlling elements and the deposits is constructed. a-2 ， wherein c is a constant, d is a buffer width, a is a singularity index; The step of constructing a deep learning loss function based on the spatial distribution law of the deposits, specifically comprises the following steps: = The step of constructing a deep learning loss function based on the spatial distribution law of the deposits, specifically comprises the following steps: a-2 / The step of constructing a deep learning loss function based on the spatial distribution law of the deposits, specifically comprises the following steps: max ， wherein, According to the advantages of different deep learning model structures and the characteristics of geological prospecting big data, a convolutional neural network considering the spatial structure and relationship of adjacent units is selected. max is the maximum value of the power-law function of the deposit distribution density and the buffer distance. Based on the supervised learning method of the convolutional neural network, a mineral prediction technical process thereof is established. The step of conducting buffer zone analysis on the ore-controlling elements closely related to mineralization to obtain buffer zone width and deposit distribution density, specifically comprises the following steps: Select the ore-controlling elements closely related to mineralization for buffer zone analysis, determine the buffer zone width and buffer interval according to the geological characteristics of the area where the deposits are located, and obtain the number of deposits falling in each buffer zone of each ore-controlling element. Let the linear equation of the neurons in the hidden layer be: x)+ b ,in, x k represents the output probability of the SoftMax layer. i It is the relationship between each input sample and the first input sample in the convolutional neural network. i Assign values ​​to the grid cells corresponding to each mineral control element layer. These are trainable weight parameters. b These are the trainable bias term parameters; the Sigmoid activation function is used to operate on the output of the above linear equation to map neuron features, i.e. (y) = ,in The constant is used to construct a fully connected hidden layer that takes into account the spatial coupling relationship between ore-controlling elements and ore deposits using hard constraint methods; The step of constructing a data and knowledge coupled mineral prediction deep learning model, specifically comprises the following steps: A function relationship between ore-controlling elements and convolutional neural network prediction of metallogenic probability is constructed, and combined with the cross-entropy function of the convolutional neural network to obtain a deep learning loss function based on the spatial distribution of the ore deposit: the cross-entropy loss function of the convolutional neural network is constructed = [ ]. In the formula, p is a true label of the input sample, is a mineralization probability prediction value of the input sample; Establishing the functional relationship between ore-controlling factors and convolutional neural network prediction of metallogenic probability a regularization term based on the spatial distribution of the deposit, Assign values to the grid cells of the ore-controlling factor layer corresponding to the input sample, and obtain the regularization formula through L 2Regularization formula term; ​ , In the formula, is a regularization term constructed based on the spatial distribution law of the deposit, is a cross-entropy loss function of the convolutional neural network constructed based on the hard constraint method, is a weight coefficient for measuring the proportion between the cross-entropy loss function and the regularization term based on the spatial distribution law of the deposit.

2. The data and knowledge coupled mineral prospectivity deep learning model building method of claim 1, wherein, ​ ​ ​ 3. The data and knowledge coupled mineral prospectivity deep learning model building method of claim 1, wherein, ​ ​ ​ 4. The data and knowledge coupled mineral prediction deep learning model building method of claim 1, wherein, ​ According to the assigned ore-controlling element layer, the implicit layer structure and the deep learning loss function, the ore-controlling element layer is taken as an input variable, a convolutional neural network structure is combined with the constructed full connection implicit layer and the deep learning loss function to comprehensively establish a mineral prediction model, so as to achieve the purpose of providing geological prior knowledge guidance to the model training process, and to construct a data and knowledge coupled mineral prediction deep learning model.

5. A data and knowledge coupled mineral prospectivity deep learning model building system characterized in that, Comprise: The convolutional neural network selection module is used for selecting the convolutional neural network considering the spatial structure and relationship of adjacent units; The buffer zone information acquisition module is used for buffer zone analysis on ore-controlling elements closely related to mineralization, and is used for acquiring buffer zone width and deposit distribution density; The ore-controlling element layer assignment module is in communication connection with the buffer zone information acquisition module, and is used for constructing a power law function of deposit distribution density and buffer zone width according to the acquired buffer zone width and deposit distribution density, and acquiring constraint weights and assigning values to the ore-controlling element layer according to the constructed power law function; The full connection implicit layer construction module is in communication connection with the convolutional neural network selection module, and is used for constructing a full connection implicit layer considering the spatial coupling relationship of ore-controlling elements and deposits; The deep learning loss function construction module is in communication connection with the convolutional neural network selection module, and is used for constructing a deep learning loss function based on the spatial distribution law of deposits; The mineral prediction deep learning model construction module is in communication connection with the ore-controlling element layer assignment module, the full connection implicit layer construction module and the deep learning loss function construction module, and is used for providing geological prior knowledge guidance to the training process of the selected convolutional neural network according to the assigned ore-controlling element layer, the implicit layer structure and the deep learning loss function, and constructing a data and knowledge coupled mineral prediction deep learning model; The step of constructing a power law function of deposit distribution density and buffer zone width according to the acquired buffer zone width and deposit distribution density, and acquiring weights and assigning values to the ore-controlling element layer according to the constructed power law function, specifically Comprise the following steps: The power law function of deposit distribution density and buffer zone width is constructed as follows according to the acquired buffer zone width and deposit distribution density: The power law function of deposit distribution density and buffer zone width is constructed as follows according to the acquired buffer zone width and deposit distribution density: = The power law function of deposit distribution density and buffer zone width is constructed as follows according to the acquired buffer zone width and deposit distribution density: a-2 ， wherein c is a constant, d is a buffer width, a is a singularity index; The ore-controlling element layer is assigned with constraint weights; = The step of constructing a full connection implicit layer considering the spatial coupling relationship of ore-controlling elements and deposits, specifically comprises the following steps: a-2 / A full connection implicit layer considering the spatial coupling relationship of ore-controlling elements and deposits is constructed according to the training mode of the convolutional neural network, the mathematical principle of the full connection implicit layer and the geological connotation of the ore-controlling elements: max ， wherein, The step of constructing a deep learning loss function based on the spatial distribution law of deposits, specifically comprises the following steps: max is the maximum value of the power-law function of the deposit distribution density and the buffer distance. The function relationship and the cross-entropy loss function are combined by using a soft constraint method to acquire a deep learning loss function based on the spatial distribution law of deposits as shown in the following formula: The convolutional neural network selection module comprises: ​ Let the linear equation of the neurons in the hidden layer be: x )+ b ,in, x Represents the output probability of the SoftMax layer. k i It is the relationship between each input sample and the first input sample in the convolutional neural network. i Assign values ​​to the grid cells corresponding to each mineral control element layer. These are trainable weight parameters. b These are the trainable bias term parameters; the Sigmoid activation function is used to operate on the output of the above linear equation to map neuron features, i.e. (y) = ,in The constant is used to construct a fully connected hidden layer that takes into account the spatial coupling relationship between ore-controlling elements and ore deposits using hard constraint methods; ​ A function relationship between ore-controlling elements and convolutional neural network prediction of metallogenic probability is constructed, and combined with the cross-entropy function of the convolutional neural network to obtain a deep learning loss function based on the spatial distribution of the ore deposit: the cross-entropy loss function of the convolutional neural network is constructed = [ ]. In the formula, p is a true label of the input sample, is a mineralization probability prediction value of the input sample; Establishing the functional relationship between ore-controlling factors and convolutional neural network prediction of metallogenic probability a regularization term based on the spatial distribution of the deposit, The grid cells of the ore-controlling factor layer corresponding to the input sample are valued, and the term is obtained through the L2 regularization formula; ​ , In the formula, is a regularization term based on the spatial distribution of the deposit, is a cross-entropy loss function of the convolutional neural network based on the hard constraint method, is a weight coefficient for measuring the proportion between the cross-entropy loss function and the regularization term based on the spatial distribution of the deposit.

6. The data and knowledge coupled mineral deposit predictive deep learning model building system of claim 5, wherein, ​ The convolutional neural network selection unit is used for selecting a convolutional neural network considering the spatial structure and relationship of adjacent units according to the advantages of different deep learning model structures and the characteristics of geological prospecting big data. The mineral prediction process establishment unit is used for establishing a mineral prediction technology process based on a supervised learning mode of the convolutional neural network.

7. The data and knowledge coupled mineral prediction deep learning model building system of claim 5, wherein, The buffer information acquisition module comprises: The buffer information acquisition unit is used for selecting ore-controlling elements closely related to mineralization for buffer analysis, determining the buffer width and buffer interval according to the geological characteristics of the area where the deposit is located; The deposit distribution density acquisition unit is used for statistically obtaining the number of deposits falling on each buffer of each ore-controlling element and obtaining the deposit distribution density of each buffer.

Images