Mineral resource prediction method and device based on deep network proxy

Abstract

The invention discloses a deep network proxy-based mineral resource prediction method and device, and the method comprises the steps: carrying out the exploration of an ore deposit through an exploration assembly, obtaining the exploration data of the ore deposit, and building a metallogenic numerical calculation model of the ore deposit based on the exploration data; outputting a metallogenic condition parameter calculation result through the metallogenic numerical calculation model, and training the deep neural network model; after model parameters are resampled, forward modeling calculation is carried out on metallogenic condition parameters through the trained deep neural network agent model; generating a data set according to the forward modeling data in combination with the exploration data, and training a machine learning model based on the data set to obtain a trained machine learning model; and outputting a preliminary prediction result through the trained machine learning model, detecting the preliminary prediction result through the prediction index quantity, and outputting a prediction result. And the metallogenic data is obtained through forward calculation, and the most possible metallogenic conditions and metallogenic conditions are deduced through the metallogenic data, so that the prediction accuracy of the prediction model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mining area exploration, and in particular to a method and device for predicting mineral resources based on a deep network agent. Background Art

[0002] The existing exploration and prediction of mineral resources in mineral deposits mainly relies on real-time monitoring of the geophysical characteristics in the mineral deposit area, combined with the historical data of mineral deposit mineralization data, to predict the mineralization of the mineral deposit according to the changes in the measured geophysical characteristics, and then output mineral exploration suggestions. The existing prediction model mainly sets the mineralization simulation conditions and calculation conditions based on the physical property parameters. Due to the complexity of the geological physical property parameters in the mineral deposit area, the physical property parameters have a great influence on the prediction model, resulting in a decrease in the prediction accuracy of the prediction model, and there is a problem of uncertainty in the mineralization prediction of the mineral deposit area. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art. The present invention provides a method and device for predicting mineral resources based on a deep network agent, which calculates the mineralization data in the ore deposit area through forward calculation. The most likely mineralization conditions and mineralization conditions are reversely deduced from the mineralization data to improve the prediction accuracy of the prediction model, thereby achieving accurate prediction of the mineralization conditions in the ore deposit area.

[0004] The present invention provides a mineral resource prediction method based on deep network agent, the prediction method comprising:

[0005] S11: surveying the mineral deposit through the exploration component, acquiring exploration data of the mineral deposit, and extracting model parameters based on the exploration data;

[0006] Extracting a number of index parameters from the exploration data, and constructing a mineralization numerical calculation model of the ore deposit based on the number of parameter indicators;

[0007] S12: outputting the calculation results of the metallogenic conditions through the metallogenic numerical calculation model, and training the deep neural network model according to the calculation results of the metallogenic condition parameters;

[0008] After resampling the model parameters, forward modeling is performed on the model parameters through the trained deep neural network proxy model to obtain forward modeling data;

[0009] S13: generating a data set according to the forward modeling data combined with the exploration data, and training the machine learning model based on the data set to obtain a trained machine learning model;

[0010] S14: Output the prediction result through the trained machine learning model, and detect whether the prediction index of the prediction result reaches the maximum value. If not, return to step S12; if so, proceed to step S15;

[0011] S15: Output prediction results.

[0012] Furthermore, the extracting of several index parameters from the exploration data and constructing the mineralization numerical calculation model of the ore deposit based on the several parameter indicators includes:

[0013] Extracting physical property parameters of the ore deposit from the exploration data;

[0014] constructing a constitutive model of the ore deposit based on the exploration data;

[0015] Setting conditional parameters of the mineralization numerical calculation model based on the exploration data;

[0016] A numerical calculation model of the mineralization of the ore deposit is constructed by combining the physical property parameters, the constitutive model and the conditional parameters.

[0017] Furthermore, the conditional parameters for setting the mineralization numerical calculation model based on the exploration data include:

[0018] According to the numerical changes of various physical parameters in the deposit area, by obtaining the numerical change amplitude of each physical parameter and converting the amplitude change value of each physical parameter, the stress loading conditions and boundary constraint conditions of the mineralization simulation are set in combination with the relevant research data of the mineralization background and model.

[0019] Furthermore, the outputting of the calculation results of the mineralization condition parameters through the mineralization numerical calculation model, and training the deep neural network model according to the calculation results of the mineralization condition parameters include:

[0020] Simulation parameter sensitivity analysis to obtain sensitive simulation parameters;

[0021] Resample sensitive simulation parameters and output simulation results based on high-fidelity numerical simulation;

[0022] A deep neural network model is trained according to the simulation results.

[0023] Furthermore, the simulation parameter sensitivity analysis to obtain sensitive simulation parameters includes:

[0024] The sensitivity analysis of simulation parameters is performed according to the COMSOL Multiphysics finite element solver to extract sensitive parameters.

[0025] Furthermore, the output of simulation results based on high-fidelity numerical simulation includes:

[0026] Retain several sensitive parameters among several physical property parameters, and resample several sensitive parameters;

[0027] By operating the resampled sensitive parameter data samples on the mineralization numerical calculation model, the corresponding operation results are output in the mineralization numerical calculation model based on the sensitive parameters;

[0028] Perform high-fidelity numerical simulation operations on sensitive simulation parameters and output simulation results.

[0029] Furthermore, outputting the calculation results of the mineralization condition parameters through the mineralization numerical calculation model, and training the deep neural network model according to the calculation results of the mineralization condition parameters also includes:

[0030] A training set and a test set are generated according to the simulation results, and a deep neural network is trained based on the training set and the test set.

[0031] Furthermore, generating a data set based on the forward modeling data combined with the exploration data, training a machine learning model based on the data set, and obtaining a trained machine learning model includes:

[0032] It is planned to use the random forest algorithm to split the data for training machine learning into training sets and test sets, train the machine learning model and optimize the hyperparameters to obtain the trained machine learning model.

[0033] Furthermore, outputting the prediction result through the trained machine learning model and detecting whether the prediction index of the prediction result reaches the maximum value includes:

[0034] Through the trained machine learning model, the mineralization prediction variable information of unknown mineralization data points is integrated, the posterior probability of mineralization is calculated, and the preliminary prediction results are output;

[0035] The counting results are calculated based on the parameters of the machine learning model, and it is detected whether the prediction index quantity of the preliminary prediction result reaches the maximum value.

[0036] The present invention also provides a mineral resource prediction device based on a deep network agent, wherein the prediction device is used to execute the prediction method;

[0037] Exploration module: used to survey the mineral deposit through the exploration component, obtain the exploration data of the mineral deposit, and extract model parameters based on the exploration data;

[0038] Computational model building module: used to extract several index parameters from the exploration data, and build a mineralization numerical calculation model of the ore deposit based on the several parameter indicators;

[0039] Training module: used to output the calculation results of metallogenic condition parameters through the metallogenic numerical calculation model, and train the deep neural network model according to the calculation results of metallogenic condition parameters;

[0040] Forward calculation module: after resampling the model parameters, forward calculation is performed on the model parameters through the trained deep neural network proxy model to obtain forward data;

[0041] Machine learning module: generating a data set according to the forward modeling data combined with the exploration data, training a machine learning model based on the data set, and obtaining a trained machine learning model;

[0042] Judgment module: used to output the prediction result through the trained machine learning model and detect whether the prediction index of the prediction result reaches the maximum value;

[0043] Output module: used to output prediction results.

[0044] The present invention provides a method and device for predicting mineral resources based on a deep network agent. The physical parameters, constitutive model, calculation conditions and initial operating conditions of the ore deposit area are extracted through exploration data. Based on the combination of physics, data and model, the accuracy of exploration data processing in the ore deposit area is improved. According to the model parameters obtained by processing the exploration data, the model parameters are forward calculated by a deep neural network model, and the preliminary prediction results of the mineralization data in the ore deposit area are output. According to the preliminary prediction results, the inversion calculation is combined with the machine learning model, and the mineralization data in the ore deposit area is calculated by forward calculation. The most likely mineralization conditions and mineralization conditions are inverted through the mineralization data, so as to achieve accurate prediction of the mineralization conditions in the ore deposit area. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 It is a flow chart of a method for predicting mineral resources based on a deep network agent in an embodiment of the present invention;

[0047] Figure 2 It is a schematic diagram of the flow of the mineralization numerical calculation model in the embodiment of the present invention;

[0048] Figure 3 Schematic diagram of the training process of the deep neural network model in the embodiment of the present invention

[0049] Figure 4 It is a schematic diagram of a mineral resource prediction device based on a deep web agent in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Embodiment 1:

[0052] Figure 1 The flowchart of the mineral resource prediction method based on deep network agent in the embodiment of the present invention is shown, and the prediction method includes:

[0053] S11: surveying the mineral deposit through the exploration component, acquiring exploration data of the mineral deposit, and extracting model parameters based on the exploration data;

[0054] A number of index parameters are extracted from the exploration data, and a mineralization numerical calculation model of the ore deposit is constructed based on the number of parameter indicators.

[0055] The geology of the ore deposit is explored through exploration components, drilling operations are carried out at the corresponding positions in the mining area through drilling rigs, and sampling and analysis are carried out on the geology of the mining area at the drilling positions. Based on the differences in the density, magnetism, electricity, elasticity, radioactivity and other physical properties of various rocks and ores, different physical methods and geophysical instruments are used to detect the changes in geological physical properties at different positions and depths in the mining area.

[0056] The physical property parameters of the ore deposit area are extracted based on the exploration data, and a constitutive model is generated based on the exploration data of the ore deposit area. The boundary conditions, initial conditions, etc. of the mineralization calculation are analyzed based on the exploration data of the ore deposit area, so as to generate a numerical calculation model of the mineralization of the ore deposit area based on the exploration data.

[0057] S12: outputting calculation results of metallogenic condition parameters through the metallogenic numerical calculation model, and training the deep neural network model according to the calculation results of the metallogenic condition parameters;

[0058] After resampling the model parameters, the model parameters are forward calculated through the mineralization numerical simulation model, and various forms of parameters are simulated, such as stress field parameters (including the first principal stress, the second principal stress, the third principal stress, the first principal strain, the second principal strain, the third principal strain, etc.), temperature field parameters (including temperature, gradient, etc.), fluid field parameters (including Darcy velocity field, flow rate, fluid potential, etc.) and chemical field (lead ion concentration, lead sulfide concentration, etc.).

[0059] Under the constraints of the reasonable value range of model parameters, boundary conditions and initial conditions, the model parameters are resampled at high density and precision with equal intervals. By automatically reading the value of a certain resampled parameter combination and combining it with a deep neural network proxy model, the gravity field, thermal field, flow field and chemical field of the ore deposit area are quickly solved to obtain the mineralization condition parameter values ​​corresponding to the corresponding input parameters, complete a forward calculation of the mineralization condition parameters, and obtain forward modeling data.

[0060] S13: Generate a data set according to the forward modeling data combined with the exploration data, and train the machine learning model based on the data set to obtain a trained machine learning model.

[0061] The calculation results of each mineralization condition parameter are obtained through a deep neural network proxy model, and a data set is generated by combining the exploration data in the ore deposit area. The machine learning model is trained based on the data set, thereby optimizing the accuracy of the machine learning model in mineralization prediction of the ore deposit area.

[0062] Other data with unknown mineral properties are used as data to be predicted. The random forest algorithm is used to split the data for machine learning training into training sets and test sets, train the machine learning model and optimize the hyperparameters to obtain the trained machine learning model.

[0063] S14: Output a preliminary prediction result through the trained machine learning model, and detect whether the prediction index of the preliminary prediction result reaches the maximum value. If not, return to step S12; if so, go to step S15.

[0064] By constructing a model and training a machine learning model for each mineralization condition parameter in the deposit area, and combining the calculation results of the deep neural network proxy model to inversely calculate different mineralization condition parameters, a mineralization prediction model with high prediction accuracy is obtained.

[0065] S15: Output prediction results.

[0066] Through particle swarm intelligent optimization iterative inversion calculation, the particle swarm is initialized using the discrete resampled simulation parameters, and the training and construction of the machine learning model are repeated based on different simulation parameters, that is, different mineralization condition parameters. Based on continuous cyclic iterative optimization calculations, the corresponding forward and inversion calculations of the next parameter combination value are performed until a model with the maximum prediction accuracy index value is obtained.

[0067] Further: Combined with the historical data of the ore deposit area, the mineralization background, exploration degree and exploration conditions of the ore deposit area are obtained, the predicted target areas are prioritized, the prospecting prospects and exploration risks are evaluated, and prospecting suggestions are put forward.

[0068] Embodiment 2:

[0069] Figure 2 A flow chart of the mineralization numerical calculation model in the embodiment of the present invention is shown, in which the geology of the ore deposit is explored by the exploration component, drilling operations are performed at corresponding positions in the mining area by the drilling rig, and the geology of the mining area at the drilling position is sampled and analyzed. Based on the differences in the density, magnetism, electricity, elasticity, radioactivity and other physical properties of various rocks and ores, different physical methods and geophysical instruments are used to detect the changes in geological physical properties at different positions and depths in the mining area.

[0070] A number of index parameters are extracted from the exploration data, and a mineralization numerical calculation model of the ore deposit is constructed based on the number of parameter indicators.

[0071] Specifically, the construction process of the mineralization numerical calculation model includes:

[0072] The method of extracting a plurality of index parameters from the exploration data and constructing a mineralization numerical calculation model of the ore deposit based on the plurality of index parameters includes:

[0073] S111: Extracting physical property parameters of the ore deposit from the exploration data.

[0074] By comprehensively analyzing the exploration data of the ore deposit area, such as the profile and drill holes of the ore deposit, a shallow geometric model of the ore deposit is constructed, that is, a geometric model of the geological characteristics of the surface of the ore deposit is constructed based on the exploration data of the ore deposit area.

[0075] Query the historical data of the mineral deposit area, measure the physical characteristics of the mineral deposit area such as gravity field, magnetic field, electric field, etc., and construct a deep geological structure geometric model of the mineral deposit area by combining the physical characteristics of gravity field, magnetic field, electric field, etc. in the historical data;

[0076] Based on the historical data of the ore deposit area, combined with the exploration data of the ore deposit area, a restoration model of the ore deposit area is constructed, combined with the shallow geometric model, the deep geological structure geometric model and the restoration model, combined with the initial geometric model of the ore deposit area.

[0077] Furthermore, by acquiring the physical characteristics of the geology of the mining area and constructing an initial geometric model of the mining area in combination with the historical data of the mining area, the structural characteristics of the mining area can be accurately reflected.

[0078] S112: Constructing a constitutive model of the ore deposit according to the exploration data.

[0079] The mineralization mechanism of the ore deposit area is summarized based on the exploration data, and the ore-controlling structure, magma, hydrothermal and other mineralization processes in the ore deposit area are analyzed through the exploration data. By querying historical data, based on the dynamic models of the three physical fields of force, heat and flow and their coupling methods, a force-heat-flow multi-field coupling numerical simulation constitutive model describing the ore deposit to be explored is established.

[0080] In this embodiment, the dynamic models of the three physical fields of force, heat and flow are coupled by multi-physics simulation software to obtain a force-heat-flow multi-field coupled numerical simulation constitutive model of the exploration deposit in the deposit area.

[0081] S113: setting conditional parameters of the mineralization numerical calculation model based on the exploration data;

[0082] According to the exploration data of the ore deposit area, the range of rock and ore physical properties and fluid parameters are obtained, and the mineralization pattern of the ore deposit area is analyzed according to the geological physical properties and fluid parameters in the ore deposit area. That is, according to the numerical changes of various physical parameters in the ore deposit area, the numerical change amplitude of each physical parameter is obtained and the amplitude change value of each physical parameter is converted, and the stress loading conditions and boundary constraints of the mineralization simulation are set in combination with the relevant research data of the mineralization background and pattern.

[0083] Furthermore, the physical parameters include: density, porosity, permeability, thermal conductivity, Poisson's ratio, fluid viscosity, fluid temperature, fluid pressure, etc. By analyzing the physical parameters of the geology of the ore deposit area, accurate prediction results of the geology of the ore deposit area can be obtained.

[0084] The metallogenic background and metallogenic model of the ore deposit area are analyzed by the physical parameters and the fluid parameters, and the stress loading conditions and boundary constraint conditions of the mineralization simulation are formulated, providing a basis for determining the model parameters, boundary conditions and initial conditions in the numerical simulation of the mineralization process.

[0085] S114: Constructing a numerical calculation model of the mineralization of the ore deposit by combining the physical property parameters, the constitutive model and the conditional parameters.

[0086] Based on the above-mentioned geometric model, as well as model parameters, boundary conditions and initial conditions, an initial numerical calculation model is established. Multi-physics field numerical simulation software is used to realize numerical simulation of the force-heat-fluid multi-field coupled mineralization dynamics process of the proposed exploration deposit, and an initial force-heat-fluid multi-field coupled numerical calculation model is obtained.

[0087] The feasibility and rationality of the force-heat-fluid multi-field coupling numerical calculation model are analyzed, that is, multi-case operation tests are performed based on multiple groups of test data, the debugging of the force-heat-fluid multi-field coupling numerical calculation model is completed, and the mineralization numerical calculation model is obtained.

[0088] Furthermore, the mineralization history data of the approximate ore deposit is extracted from the database, the geological physical parameters and fluid parameter data of the approximate ore deposit are extracted from the mineralization history data of the approximate ore deposit, and the geological physical parameters and fluid parameter data of the approximate ore deposit are input into the mineralization numerical calculation model of the ore deposit for computational verification. The computational result of the mineralization numerical calculation model is compared with the mineralization value of the approximate ore deposit. If the error value between the two is less than 0.3, the mineralization numerical calculation model meets the mineralization simulation calculation requirements of the ore deposit area, thereby improving the accuracy of the mineralization numerical calculation model.

[0089] After resampling the model parameters, the model parameters are forward modeled using the trained deep neural network proxy model to obtain forward modeling data.

[0090] The model parameters of the mineralization numerical calculation model are forward calculated through deep neural networks, and the forward calculation data are obtained by verifying the calculation accuracy of the mineralization condition parameters in the ore deposit area.

[0091] Specifically, Figure 3 A schematic diagram of the training process of the deep neural network model in an embodiment of the present invention is shown, wherein the calculation results of the mineralization condition parameters are output by the mineralization numerical calculation model, and the deep neural network model is trained according to the calculation results of the mineralization condition parameters, including:

[0092] S121: Simulation parameter sensitivity analysis.

[0093] The COMSOL Multiphysics finite element solver is used to define the sensitive functions of force, heat, flow, and chemical fields. The geometric model and physical field interface of the ore deposit area are configured in the COMSOL Multiphysics finite element solver. According to the nature and scale of the problem, the appropriate solver and iterator are selected, and the physical field parameters of the geometric model of the ore deposit area are set, such as boundary conditions, source terms, etc. The finite element simulation calculation is then performed by setting parameters such as solution time and step size.

[0094] Furthermore, COMSOL Multiphysics is the flagship product of COMSOL, a powerful multi-physics simulation software that is based on the finite element method and simulates real physical phenomena by solving partial differential equations (single field) or groups of partial differential equations (multiple fields).

[0095] S122: Obtain sensitive parameters.

[0096] According to the COMSOL Multiphysics finite element solver, the simulation parameters are subjected to sensitivity analysis, and sensitive parameters are extracted. By defining objective functions such as average stress and temperature extreme value, the model parameters, boundary conditions and initial conditions are subjected to sensitivity analysis, and the simulation parameters (i.e., parameters of mathematical-physical equations, such as elastic modulus E, density ρ, porosity ε, Poisson's ratio λ, thermal conductivity kp, constant pressure heat capacity Cp, etc.) with strong sensitivity to the simulation results are identified. The forward sensitivity calculation method is adopted, and the differential of the scalar objective function relative to a set of specified control variables is calculated according to the COMSOL Multiphysics finite element solver. The formula is as follows: wherein the objective function is Q(ζ), which is usually not an explicit expression of a single control variable ζ (parameter of mathematical-physical equation), but Q(ζ) is a function of the solution variable u, and u is an implicit expression of the control variable ζ (parameter of mathematical-physical equation), and the physical field partial differential equation is expressed as L(u(ζ),ζ)=0.

[0097]

[0098] Among them, ζ is the control variable, u is the solution variable, u(ζ) is the implicit expression of the control variable ζ, and Q(ζ) is the objective function; is the explicit partial derivative of Q with respect to ζ, d is the differential symbol, dζ is the differential of ζ, is the partial derivative of Q with respect to u, is the partial derivative of L with respect to ζ, is the partial derivative of u with respect to ζ.

[0099] S123: Resample sensitive parameters.

[0100] After inputting several physical property parameters of the ore deposit area into the COMSOL Multiphysics finite element solver for sensitivity analysis, several sensitive parameters are retained among the several physical property parameters, and a resampling operation is performed on the several sensitive parameters at equal intervals.

[0101] For example, numerical sampling is performed on each sensitive parameter at a preset time interval, and operation data samples of the sensitive parameters are obtained based on resampling.

[0102] S124: Inputting the resampled sensitive parameters into the mineralization numerical calculation model.

[0103] The mineralization numerical calculation model is used to calculate the resampled sensitive parameter data samples, and the corresponding calculation results are output in the mineralization numerical calculation model based on the sensitive parameters.

[0104] S125: Perform high-fidelity numerical simulation operations on sensitive simulation parameters and output simulation results.

[0105] MATLAB is used to write a script program to automatically resample the highly sensitive numerical simulation parameters, control COMSOL Multiphysics to solve, batch output stress, strain, fluid flux, temperature, ore-forming material concentration and other mineralization condition parameters (i.e. high-fidelity simulation data), and save the input parameters and output results of each time as a table file in CSV format.

[0106] S126: Generate a training set and a test set according to the simulation results, and train the deep neural network based on the training set and the test set.

[0107] The CSV data files of the above high-fidelity numerical simulation results are automatically read in batches, merged into a data set, and the data set is divided into a training set and a test set. The deep neural network is trained based on the training set, and the trained deep neural network model is tested and verified through the test set to ensure the operational accuracy of the deep neural network model.

[0108] Firstly, the numerical simulation input parameters, simulation unit spatial position coordinates (x, y), and time (t) in the automatically batch-read data set are used as feature variables. The simulation results are used as prediction variables, and the data set is divided into training set and test set according to a certain ratio. The deep neural network model is trained with the training set, and tested based on the test set. Deep neural network proxy models are constructed respectively for high-precision and fast solution of force field, thermal field, flow field, and chemical field in the multi-field coupled numerical model of ore-forming dynamics.

[0109] The simulation input parameters include parameters of mathematical-physical equations, such as elastic modulus E, density ρ, porosity ε, Poisson's ratio λ, thermal conductivity kp, constant pressure heat capacity Cp and other parameter data.

[0110] The simulation results include parameter data such as principal strain (Ep), principal stress (Sp), equivalent stress (Mises), shear stress (Evol) in the force field, temperature (T) in the thermal field, fluid velocity (U), flux (Q) in the flow field, etc.

[0111] Read the simulation condition parameters in CSV format in step S12, link the exploration results with the simulation condition parameters by spatial superposition, compile the simulation condition parameters and exploration data into a data set for the machine learning model, perform data fusion on the simulation condition parameters and exploration data based on the machine learning model, use the binary variable of "with or without ore" in the data set as the target variable, use the data marked with ore and the data marked without ore as the data for training machine learning,

[0112] Other data with unknown mineral properties are used as data to be predicted. The random forest algorithm is used to split the data for machine learning training into training sets and test sets, train the machine learning model and optimize the hyperparameters to obtain the trained machine learning model.

[0113] Through the trained machine learning model, the mineralization prediction variable information of the above-mentioned unknown mineral-bearing data points is integrated, the posterior probability of mineralization is calculated, and the preliminary prediction results are output.

[0114] According to the number of mineralization condition parameters recorded in the CSV data table, the mineralization condition parameters of the CSV data table are input into the machine learning model in sequence. The machine learning model uses the input mineralization condition parameters as variables to output the mineralization probability of the ore deposit area corresponding to the mineralization condition parameters, and counts the parameter operations of the machine learning model.

[0115] According to the parameter calculation counting results of the machine learning model, it is detected whether the prediction index quantity of the preliminary prediction result reaches the maximum value, that is, whether the machine learning model completes the mineralization prediction calculation of all mineralization condition parameters in the CSV data table.

[0116] If not, return to step S12 and construct a mineralization prediction model for the ore deposit area based on the next mineralization condition parameter data, so as to ensure that the machine learning model can complete the mineralization prediction of each mineralization condition parameter in the ore deposit area, thereby realizing the inversion calculation of the mineralization conditions of the ore deposit area.

[0117] Through particle swarm intelligent optimization iterative inversion calculation, the particle swarm is initialized by using discrete resampled simulation parameters, and the model construction and calculation of steps S12 to S14 are repeatedly executed, so that the various mineralization condition parameters of the ore deposit area can complete the prediction of the mineralization calculation of the ore deposit area.

[0118] Furthermore, by continuously performing iterative optimization calculations and performing forward and inverse calculations of the corresponding next parameter combination value until a model with a maximum prediction accuracy index value is obtained, the accuracy of the mineralization prediction of the machine learning model can be effectively improved, thereby generating accurate mineral exploration suggestions for the ore deposit area.

[0119] The embodiment of the present invention provides a method for predicting mineral resources based on a deep network agent, which accurately predicts the mineralization conditions in the mineral deposit area by forward modeling and inversion testing of the mineralization condition parameters in the mineral deposit area. The mineralization data in the mineral deposit area is calculated by forward modeling. The most likely mineralization conditions and mineralization conditions are inverted through the mineralization data, thereby achieving accurate prediction of the mineralization conditions in the mineral deposit area.

[0120] Embodiment three:

[0121] Figure 4 A schematic diagram of a mineral resource prediction device based on a deep network agent in an embodiment of the present invention is shown, and the mineral resource prediction device based on a deep network agent includes:

[0122] Exploration module 10: used to survey the mineral deposit through the exploration component, obtain the exploration data of the mineral deposit, and extract model parameters based on the exploration data.

[0123] A number of index parameters are extracted from the exploration data, and a mineralization numerical calculation model of the ore deposit is constructed based on the number of parameter indicators.

[0124] The geology of the ore deposit is explored through exploration components, drilling operations are carried out at the corresponding positions in the mining area through drilling rigs, and sampling and analysis are carried out on the geology of the mining area at the drilling positions. Based on the differences in the density, magnetism, electricity, elasticity, radioactivity and other physical properties of various rocks and ores, different physical methods and geophysical instruments are used to detect the changes in geological physical properties at different positions and depths in the mining area.

[0125] The physical property parameters of the ore deposit area are extracted based on the exploration data, and a constitutive model is generated based on the exploration data of the ore deposit area. The boundary conditions, initial conditions, etc. of the mineralization calculation are analyzed based on the exploration data of the ore deposit area, so as to generate a numerical calculation model of the mineralization of the ore deposit area based on the exploration data.

[0126] The calculation model building module 20 is used to extract a plurality of index parameters from the exploration data, and to build a mineralization numerical calculation model of the ore deposit based on the plurality of parameter indicators.

[0127] After resampling the model parameters, the model parameters are forward modeled using the trained deep neural network proxy model to obtain forward modeling data.

[0128] Under the constraints of the reasonable value range of model parameters, boundary conditions and initial conditions, the model parameters are resampled at high density and precision with equal intervals. By automatically reading the value of a certain combination of resampled parameters and combining it with a deep neural network proxy model, the gravity field, thermal field, flow field and chemical field of the ore deposit area are quickly solved to obtain the mineralization condition parameter values ​​corresponding to the corresponding input parameters, thus completing a forward calculation of the mineralization condition parameters.

[0129] Training module 30: used to output the calculation results of mineralization condition parameters through the mineralization numerical calculation model, and train the deep neural network model according to the calculation results of mineralization condition parameters.

[0130] The calculation results of each mineralization condition parameter are obtained through the mineralization numerical calculation model, and a data set is generated by combining the exploration data in the ore deposit area. The machine learning model is trained based on the data set, so as to optimize the accuracy of the machine learning model in mineralization prediction of the ore deposit area.

[0131] Other data with unknown mineral properties are used as data to be predicted. The random forest algorithm is used to split the data for machine learning training into training sets and test sets, train the machine learning model and optimize the hyperparameters to obtain the trained machine learning model.

[0132] Forward modeling module 40: After resampling the model parameters, forward modeling is performed on the model parameters through the trained deep neural network proxy model to obtain forward modeling data.

[0133] By constructing a model and training a machine learning model for each mineralization condition parameter in the ore deposit area, and by combining the calculation results of the mineralization numerical calculation model to inversely calculate different mineralization condition parameters, a mineralization prediction model with high prediction accuracy is obtained.

[0134] Machine learning module 50: generating a data set according to the forward modeling data combined with the exploration data, and training a machine learning model based on the data set to obtain a trained machine learning model;

[0135] Determination module 60: used to output a prediction result through the trained machine learning model, and detect whether the prediction index of the prediction result reaches a maximum value;

[0136] Output module 70: used to output prediction results.

[0137] Through particle swarm intelligent optimization iterative inversion calculation, the particle swarm is initialized using the discrete resampled simulation parameters, and the training and construction of the machine learning model are repeated based on different simulation parameters, that is, different mineralization condition parameters. Based on continuous cyclic iterative optimization calculations, the corresponding forward and inversion calculations of the next parameter combination value are performed until a model with the maximum prediction accuracy index value is obtained.

[0138] Further: Combined with the historical data of the ore deposit area, the mineralization background, exploration degree and exploration conditions of the ore deposit area are obtained, the predicted target areas are prioritized, the prospecting prospects and exploration risks are evaluated, and prospecting suggestions are put forward.

[0139] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0140] In addition, the above provides a detailed introduction to the embodiments of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for predicting mineral resources based on deep network agents, characterized in that: The prediction method comprises: S11: surveying the mineral deposit through the exploration component, acquiring exploration data of the mineral deposit, and extracting model parameters based on the exploration data; Extracting a number of index parameters from the exploration data, and constructing a mineralization numerical calculation model of the ore deposit based on the number of parameter indicators; S12: outputting calculation results of metallogenic condition parameters through the metallogenic numerical calculation model, and training the deep neural network model according to the calculation results of the metallogenic condition parameters; After resampling the model parameters, forward modeling is performed on the model parameters through the trained deep neural network proxy model to obtain forward modeling data; S13: generating a data set according to the forward modeling data combined with the exploration data, and training the machine learning model based on the data set to obtain a trained machine learning model; S14: Output the prediction result through the trained machine learning model, and detect whether the prediction index of the prediction result reaches the maximum value. If not, return to step S12; if so, proceed to step S15; S15: Output prediction results.

2. The method for predicting mineral resources based on deep network agent according to claim 1, characterized in that: The method of extracting a plurality of index parameters from the exploration data and constructing a mineralization numerical calculation model of the ore deposit based on the plurality of index parameters includes: Extracting physical property parameters of the ore deposit from the exploration data; constructing a constitutive model of the ore deposit based on the exploration data; Setting conditional parameters of the mineralization numerical calculation model based on the exploration data; A numerical calculation model of the mineralization of the ore deposit is constructed by combining the physical property parameters, the constitutive model and the conditional parameters.

3. The method for predicting mineral resources according to claim 2, characterized in that: The conditional parameters for setting the mineralization numerical calculation model based on exploration data include: According to the numerical changes of various physical parameters in the deposit area, by obtaining the numerical change amplitude of each physical parameter and converting the amplitude change value of each physical parameter, the stress loading conditions and boundary constraint conditions of the mineralization simulation are set in combination with the relevant research data of the mineralization background and model.

4. The method for predicting mineral resources based on deep network agent according to claim 1, characterized in that: Outputting the calculation results of the mineralization condition parameters through the mineralization numerical calculation model, and training the deep neural network model according to the calculation results of the mineralization condition parameters include: Simulation parameter sensitivity analysis to obtain sensitive simulation parameters; Resample sensitive simulation parameters and output simulation results based on high-fidelity numerical simulation; A deep neural network model is trained according to the simulation results.

5. The method for predicting mineral resources based on deep network agent according to claim 4, characterized in that: The simulation parameter sensitivity analysis, obtaining sensitive simulation parameters includes: The sensitivity analysis of simulation parameters is performed according to the COMSOL Multiphysics finite element solver to extract sensitive parameters.

6. The method for predicting mineral resources based on deep network agent according to claim 4, characterized in that: The sensitive simulation parameters are resampled, based on high-fidelity numerical simulation, and the output simulation results include: Retain several sensitive parameters among several physical property parameters, and resample several sensitive parameters; By operating the resampled sensitive parameter data samples on the mineralization numerical calculation model, the corresponding operation results are output in the mineralization numerical calculation model based on the sensitive parameters; Perform high-fidelity numerical simulation operations on sensitive simulation parameters and output simulation results.

7. The method for predicting mineral resources based on deep network agent according to claim 4, characterized in that: Outputting the calculation results through the mineralization numerical calculation model and training the deep neural network model based on the calculation results also include: A training set and a test set are generated according to the simulation results, and a deep neural network is trained based on the training set and the test set.

8. The method for predicting mineral resources based on deep network agent according to claim 1, characterized in that: The step of generating a data set based on the forward modeling data combined with the exploration data, and training the machine learning model based on the data set to obtain the trained machine learning model comprises: It is planned to use the random forest algorithm to split the data for training machine learning into training sets and test sets, train the machine learning model and optimize the hyperparameters to obtain the trained machine learning model.

9. The method for predicting mineral resources based on deep network agent according to claim 1, characterized in that: Outputting the prediction result through the trained machine learning model and detecting whether the prediction index amount of the prediction result reaches the maximum value includes: Through the trained machine learning model, the mineralization prediction variable information of unknown mineralization data points is integrated, the posterior probability of mineralization is calculated, and the preliminary prediction results are output; The counting results are calculated based on the parameters of the machine learning model, and it is detected whether the prediction index quantity of the preliminary prediction result reaches the maximum value.

10. A mineral resource prediction device based on deep network agent, characterized in that: The prediction device is used to perform the prediction method according to any one of claims 1 to 9; Exploration module: used to survey the mineral deposit through the exploration component, obtain the exploration data of the mineral deposit, and extract model parameters based on the exploration data; Computational model building module: used to extract several index parameters from the exploration data, and build a mineralization numerical calculation model of the ore deposit based on the several parameter indicators; Training module: used to output calculation results through the mineralization numerical calculation model and train the deep neural network model according to the calculation results; Forward calculation module: after resampling the model parameters, forward calculation is performed on the model parameters through the trained deep neural network proxy model to obtain forward modeling data; Machine learning module: generating a data set according to the forward modeling data combined with the exploration data, training a machine learning model based on the data set, and obtaining a trained machine learning model; Judgment module: used to output the prediction result through the trained machine learning model and detect whether the prediction index of the prediction result reaches the maximum value; Output module: used to output prediction results.

Images