Big data-based magmatic rock spatial distribution prediction method and system

Abstract

The invention discloses a magmatic rock spatial distribution prediction method and system based on big data, and the method comprises the steps: obtaining the state data and historical data of a predicted magmatic rock region, constructing a shape impact factor according to the analysis result of a sample of magmatic rock in the predicted magmatic rock region, carrying out the inversion analysis of geophysical data, and obtaining the spatial distribution of the magmatic rock in the predicted magmatic rock region. Matching the sample analysis result with the target state data to obtain a first prediction parameter, inputting the first prediction parameter and a magmatic rock historical database matching result into a magmatic rock spatial distribution prediction model to obtain a magmatic rock prediction parameter, and performing geophysical simulation according to the magmatic rock prediction parameter, and optimizing the magmatic rock spatial distribution prediction model according to a geophysical simulation result, and determining a magmatic rock spatial distribution prediction result. The method not only can improve the accuracy of magmatic rock spatial distribution prediction, but also has good interpretability, and can be directly applied to a magmatic rock spatial distribution prediction system based on big data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data prediction, and in particular to a magma rock spatial distribution prediction method and system based on big data. BACKGROUND

[0002] With the continuous progress of geological exploration technology, the requirement for the prediction accuracy of magma rock spatial distribution is higher and higher. As an important part of the crust, the spatial distribution of magma rock is influenced by many factors such as formation cause, year, surrounding geology and geophysical environment, and the corresponding prediction of the spatial distribution is relatively complex. Therefore, efficient and accurate prediction results of magma rock spatial distribution are of great significance to geological research and mineral resource exploration.

[0003] Traditional magma rock prediction methods mostly rely on the experience of geologists and field exploration data, which is not only time-consuming and laborious, but also has limited prediction accuracy. In addition, it involves multidimensional data such as geophysical characteristics, remote sensing information, geological structure and magma composition, which has a large amount of data and a complex processing flow. In recent years, the development of big data technology has provided new possibilities for magma rock prediction. By quickly and efficiently integrating and analyzing massive geophysical data, remote sensing data and geological sample analysis data, the spatial distribution of magma rock can be predicted. In addition, by integrating multi-source information and using geophysical simulation software to build a three-dimensional material structure model of magma rock, the geophysical environment around the magma rock can be simulated and visualized. Although big data technology provides a new perspective for magma rock research, there are still many challenges in how to effectively integrate and process massive geophysical, remote sensing and geological sample data, how to build an accurate magma rock material structure model, and how to improve the accuracy and generalization ability of the prediction model. Therefore, it is necessary to design an accurate, efficient and fast magma rock spatial distribution prediction method and system based on big data to overcome the shortcomings of existing spatial distribution prediction methods and systems, achieve efficient and high-precision spatial distribution prediction, and provide strong technical support for geological research and mineral resource exploration. SUMMARY

[0004] The purpose of the present application is to provide a magma rock spatial distribution prediction method and system based on big data.

[0005] To achieve the above purpose, the present application is implemented according to the following technical solutions:

[0006] The present application comprises the following steps:

[0007] Obtain state data and historical data of the predicted magma rock area, and preprocess the state data and the historical data; the state data includes geophysical data and remote sensing data;

[0008] Collecting a magmatic rock sample in a region where the magmatic rock is predicted, performing component analysis and structural analysis on the sample, and constructing a shape influence factor according to the analysis results;

[0009] Performing inversion analysis on the geophysical data to obtain first state data, matching the component analysis results with the first state data to obtain first state parameters, matching the structural analysis results with the remote sensing data to obtain second state parameters, and combining the first state parameters and the second state parameters to obtain first prediction parameters of the magmatic rock;

[0010] Establishing a magmatic rock history database according to the historical data, matching the first prediction parameters of the magmatic rock with the magmatic rock history database, and performing data fusion on the matching results to obtain prediction parameters of the magmatic rock;

[0011] Performing geophysical simulation according to the prediction parameters of the magmatic rock, and determining a spatial distribution prediction result of the magmatic rock according to the geophysical simulation result.

[0012] Further, the method for constructing a shape influence factor according to the analysis results comprises:

[0013] Collecting a magmatic rock sample and a surrounding rock layer sample in a region where the magmatic rock is predicted, performing component analysis on the magmatic rock sample to obtain component analysis results, performing structural analysis on the magmatic rock sample to obtain structural analysis results, and measuring the hardness and density of the surrounding rock layer sample;

[0014] Inputting the analysis results of the magmatic rock sample and the hardness and density of the surrounding rock layer sample into a shape influence function to obtain a shape influence factor, and the expression is:

[0015]

[0016] wherein Q F is the shape influence factor of the magmatic rock, w1, w2, and w3 are weight coefficients of the shape influence factor, f l is the hardness of the magmatic rock, f i is the hardness of the surrounding rock layer, is the average hardness of the surrounding rock layer, p l is the density of the magmatic rock, p i is the density of the surrounding rock layer, is the average density of the surrounding rock layer, n is the number of surrounding rock layers, Age is the formation year of the magmatic rock, C l is the category of the magmatic rock, C i is the category of the surrounding rock layer, S l is the structural feature of the magmatic rock.

[0017] Further, the method for performing inversion analysis on the geophysical data to obtain first state data comprises:

[0018] The geophysical data is input into the inversion model to obtain simulated values of the physical property parameters of the magma rock, and a single-objective function of the physical property parameters is constructed;

[0019] The expression of the single-objective function of the physical property parameters is:

[0020]

[0021] wherein Aim ρ1 is the first objective function of the density, Aim ρ2 is the second objective function of the density, n1 is the gravity sampling quantity, ρ i is the measured density data, ρ' i is the simulated density data, max is the maximum value, min is the minimum value, Aim v1 is the first objective function of the seismic wave velocity, Aim v2 is the second objective function of the seismic wave velocity, n2 is the seismic wave sampling quantity, vi is the measured seismic wave velocity data, ρ i is the simulated seismic wave velocity data, Aim B1 is the first objective function of the magnetic force, n3 is the magnetic force sampling data quantity, B i is the measured density data, B' i is the simulated density data;

[0022] The judgment matrix is constructed according to the single-objective function to perform consistency inspection, the evaluation value is calculated according to the judgment vector to obtain a transition objective function, and the total objective function is determined according to the transition objective function, and the expression is:

[0023]

[0024] wherein Aim final is the total objective function, n4 is the single-objective function quantity, w i is the transition objective function weight, Aim i is the transition objective function, and the transition objective function corresponds to the single-objective function one by one;

[0025] The inversion gradient is calculated according to the partial derivative of the single-objective function, the inversion model is optimized according to the total objective function and the inversion gradient, and the iteration is continuously performed until the total objective function is minimized to stop the iteration to obtain the optimized inversion model;

[0026] The physical property parameters of the magma rock are obtained by inputting the to-be-predicted geophysical data into the optimized inversion model, and the first state data of multiple groups of magma rocks are predicted according to the physical property parameters of the magma rock.

[0027] Further, the method for obtaining the first prediction parameter of the magma rock comprises:

[0028] The first state parameter is obtained by directly matching the component analysis result with the first state data according to the density data; and the rule of the direct matching is that when the density difference value is less than a set threshold value, it is determined as a group of matching data.

[0029] The principal component analysis is performed on the remote sensing data to obtain main remote sensing characteristic data, a plurality of groups of overhead morphological and distribution range prediction results are obtained according to the main remote sensing characteristic data, the correlation degrees of the main remote sensing characteristic data and the prediction results are calculated, and a data group is selected as the second state data according to the correlation degrees;

[0030] The ant colony search algorithm is adopted to match the analysis result and the second state data to obtain a shortest path determination data matching result, and the specific steps are as follows:

[0031] The original population is divided into a common search group and an exploration search group according to a ratio of 8:2, an initialization path from i point to j point is generated, and the common search group is searched and the pheromone is updated first;

[0032] After the pheromone of the common search group and the exploration search group is updated, the nodes with high pheromone of the common search group are punished, and the expression is as follows:

[0033]

[0034] Wherein is the pheromone of the common search group after punishment, is the pheromone of the common search group before punishment, is the pheromone of the exploration search group, t is the iteration number, when the current path of the common search group is less than 3 times the current shortest path, q=0 is set to update the path pheromone, and otherwise q=1 is set, and the path pheromone is not updated;

[0035] The probability of algorithm mutation is adjusted, and the probability expression of state transition is as follows:

[0036]

[0037] Wherein P ij is the transition probability, τ ij is the pheromone from i point to j point, η ij is the heuristic value from i point to j point, α is the pheromone weight, β is the heuristic value weight, a k is the search candidate point, is the punishment factor, δ3 is the bias term index, p is a random number, p0 is the mutation probability, and T is the maximum iteration number;

[0038] The path is searched, the pheromone and the shortest path are updated, and the iteration is completed until the algorithm converges, and the shortest path is output, and the data matching result is determined according to the shortest path, and the second state parameter is composed of the matching result;

[0039] The first prediction parameter of the magmatic rock is obtained by combining the first state parameter and the second state parameter.

[0040] Further, the method for matching the first prediction parameter of the magmatic rock and the historical database of the magmatic rock comprises:

[0041] Classifying and principal component analyzing the historical data to obtain main magma rock parameters and main surrounding geological parameters, and constructing a magma rock historical database;

[0042] Calculating the comprehensive similarity of the corresponding feature vectors of the aerial view shape and surrounding geological data in the first predicted parameters of the magma rock and the corresponding feature vectors of the main surrounding geological parameters to obtain first matching data, calculating the comprehensive similarity of the corresponding feature vectors of the remaining data in the first predicted parameters of the magma rock and the corresponding feature vectors of the main magma rock parameters to obtain second matching data, and combining the first matching data and the second matching data to obtain a matching data group.

[0043] Further, the matching result is data fused to obtain a method for predicting parameters of a magma rock, comprising:

[0044] Dividing the matching data group into a training set and a test set;

[0045] Constructing a magma rock spatial distribution prediction model, including an input layer, a fusion layer and an output layer, and setting an encoder and a decoder;

[0046] The input layer standardizes and processes two groups of matching data and outputs them to the encoder, the encoder sets a self-attention layer and a multi-head attention layer, uses the self-attention layer to process the data dependence within each database, and uses the multi-head self-attention layer to capture semantic information to enhance the expression of the model; the self-attention layer uses an efficient additive attention mechanism, and the specific steps are as follows:

[0047] The attention weight of the data is calculated, and the expression is:

[0048]

[0049] where w i is the attention weight of the input vector F i in , s(·) is a score function, l is the total number of input vectors, and q is a target vector;

[0050] A consistent global query vector is created, and an efficient additive attention mechanism is determined according to the query vector, and the expression is:

[0051]

[0052] where is the input vector processed by the efficient additive attention mechanism, Q is the normalized query matrix, L(·) is a linear projection function, K is the normalized key matrix, n is the sequence length, d is the embedding vector dimension, and Q i is the query vector corresponding to the input vector x i ;

[0053] The training set is processed by the encoder and then fused in the fusion layer using a multi-scale cross-axis attention mechanism. The multi-scale cross-axis attention mechanism includes multi-scale convolution and axis attention mechanism, and the steps are represented as:

[0054] The data is encoded and summed in three parallel one-dimensional convolutions in different directions, and the expression is:

[0055]

[0056] where F ix is a multi-scale convolution along the horizontal direction, F iy is a multi-scale convolution along the vertical direction, is a one-dimensional convolution layer with three different kernel sizes, i∈{1, 2, 3} is the index of the convolution layer, x, y are the directions of the spatial dimension, Conv 1×1 is a 1×1 convolution operation, is the normalized encoder input vector;

[0057] In the horizontal direction, F ix is used to generate the value vector V ix and the key vector K ix , F iy is used to generate the query vector Q iy , in the vertical direction, F iy is used to generate the value vector V iy and the key vector K iy , F ix is used to generate the query vector Q ix , and the attention mechanism is used for the three generated vectors, and the expression is:

[0058] F E1 = MHCA x (F ix , F ix , F iy )

[0059] F E2 = MHCA y (F ix , F iy , F iy )

[0060] where F E1 is the horizontal output feature processed by the multi-head cross attention mechanism, F E2 is the vertical output feature processed by the multi-head cross attention mechanism, MHCA x is a multi-head cross attention mechanism along the horizontal direction, and MHCA y is a multi-head cross attention mechanism along the vertical direction;

[0061] The multi-head cross attention mechanism processing result is convoluted, the vector output by the encoder is interacted with the fusion layer output vector in the decoder to obtain fusion data, and the fusion data is converted into an output result through an output layer, and the expression is:

[0062] F i out =Conv 1×1 (F E1 )+Conv 1×1 (F E2 )+F i in

[0063] Where F i out is the output data feature vector, Conv 1×1 (F E1 ) is the convolution result of F E1 , and Conv 1×1 (F E2 ) is the convolution result of F E2 .

[0064] The Huber Loss loss function is used to evaluate the difference between the predicted value and the true value, the Adam optimizer is used to update the model parameters, and the test set is used to evaluate the magma rock spatial distribution prediction model.

[0065] The target state data to be predicted is input into the magma rock spatial distribution prediction model to obtain magma rock prediction parameters.

[0066] Further, the method for determining the magma rock spatial distribution prediction result according to the geophysical simulation result comprises:

[0067] In the geophysical simulation, a geophysical model of the target area is constructed according to the magma rock prediction parameters, gravity simulation values and magnetic field simulation values of the target area are obtained, and the difference between the geophysical data and the simulation values is calculated;

[0068] When the difference is less than a set threshold, the magma rock prediction parameters are output as the magma rock spatial distribution prediction result;

[0069] When the difference is greater than the set threshold, the magma rock spatial distribution prediction model is optimized according to the difference between the geophysical data and the simulation values, the target state data to be predicted is input into the optimized magma rock spatial distribution prediction model again to obtain magma rock prediction parameters, and the geophysical simulation and the threshold determination step are repeated until the difference is less than the set threshold, and the optimization is stopped to output the magma rock spatial distribution prediction result;

[0070] The method for optimizing the hyperparameters of the magma rock spatial distribution prediction model according to the difference between the geophysical data and the simulation values comprises:

[0071] Initialize the parameter matrix of the magma rock spatial distribution prediction model, and record the corresponding parameter position, use chaotic mapping to initialize the population, the expression is:

[0072]

[0073] Wherein is the initial position of the ith individual, Z i is the sequence after chaotic mapping, x max is the upper limit of the search space, x min is the lower limit of the search space, and alpha is the chaotic parameter.

[0074] Calculate the fitness of the population, and record the best fitness value, randomly select three population individuals for mutation, crossover and selection operation, the expression is:

[0075] V i = X r1 + mu (X r2 -X r3 )

[0076]

[0077] Wherein V i is the new individual generated by mutation, X is the parent population individual, r1, r2, r3 are different random integers in [1, Q], Q is the population size, mu is the mutation factor, U ij is the new individual generated by crossover, P cr is the crossover probability, j rand is a random integer in [1, gamma], is the new population individual generated by greedy selection, t is the iteration number, f(·) is the objective function, X i is the original individual.

[0078] Update the nonlinear convergence factor, the expression is:

[0079] Theta = 1-ln[exp{sin(t / T) / sin2-1}] 1 / 2

[0080] Wherein theta is the nonlinear convergence factor, and T is the maximum iteration factor.

[0081] Update the position coefficient A, the position coefficient weight C, the random number l and the individual probability p, update the individual position according to the position coefficient A, calculate the individual fitness according to the objective function and record the best fitness, iterate until the maximum iteration number is reached, stop iteration, and output the global optimal solution and the fitness value.

[0082] Further, the method for obtaining the measured length comprises:

[0083] According to the distance measuring scale, the laser image length of the scanning group at different spatial positions is adjusted to obtain a comprehensive measurement length, and a comprehensive measurement set is formed by the transmission influence factor, the distance measuring scale and the first edge coordinate, the comprehensive measurement set is divided into a training set and a test set;

[0084] A high-precision length measurement model is constructed, and the high-precision length measurement model comprises an input layer, a base model layer, a strategy layer and an output layer;

[0085] The base model layer is composed of two multilayer perception machine models and two gradient boosting machine models in parallel;

[0086] In the multilayer perception machine model: two hidden layers are designed, the ReLU function is used as the activation function to accelerate the training and reduce the gradient disappearance, the Cross-Entropy Loss loss function is selected to solve the deviation of the cross-entropy loss function from the evaluation index and the overconfidence, and the Adam optimization model weight parameter is selected;

[0087] In the gradient boosting machine model: the mathematical model is constructed by iteration, the log loss function is used to measure the difference between the predicted probability distribution and the true distribution, and the learning rate is adjusted to control the optimization process;

[0088] The strategy layer uses a stacking method to input the base model prediction structure into a new model for prediction to obtain a prediction result, and the test set is used to evaluate the high-precision length measurement model;

[0089] The machine working data and state data to be measured are input into the high-precision length measurement model to obtain the measurement length.

[0090] In a second aspect, a magma rock spatial distribution prediction system based on big data includes:

[0091] A data acquisition module is configured to obtain state data and historical data of a predicted magma rock region, and to preprocess the state data and the historical data;

[0092] A data processing module is configured to construct a shape influence factor based on an analysis result, to obtain first state data by inverting analysis of the geophysical data, to obtain first state parameters, first state parameters, magma rock first prediction parameters and a magma rock historical database, and to match the magma rock first prediction parameters and the magma rock historical database;

[0093] A fusion model module is configured to construct a magma rock spatial distribution prediction model based on the matched data set, and to obtain magma rock prediction parameters by fusing the matched data set through the magma rock spatial distribution prediction model;

[0094] The simulation module is configured to perform geophysical simulation according to the magma rock prediction parameters, determine a magma rock spatial distribution prediction result according to a result of the geophysical simulation, and perform visual display of the magma rock spatial distribution prediction result to a user.

[0095] The optimization module is configured to optimize a magma rock spatial distribution prediction model according to a difference between the geophysical data and the result of the geophysical simulation.

[0096] The intelligent supervision module is configured to store, view and manage the target state data, the historical data and the magma rock spatial distribution prediction result.

[0097] The present application has the following beneficial effects:

[0098] Compared with the prior art, the present application has the following technical effects:

[0099] The present application can improve the accuracy of magma rock spatial distribution prediction, thereby improving the efficiency and precision of magma rock spatial distribution prediction based on big data, and can greatly save resources and improve the efficiency of spatial distribution prediction, can realize the prediction of the spatial distribution of magma rock in the magma rock region, and can quickly and accurately predict the spatial distribution of magma rock, thereby providing strong technical support for mining and geological exploration, and has important significance for magma rock spatial distribution prediction based on big data, and can adapt to different magma rock spatial distribution prediction systems based on big data and different magma rock spatial distribution prediction needs of different users, and has certain universality. BRIEF DESCRIPTION OF DRAWINGS

[0100] Figure 1 The present application is a method for predicting the spatial distribution of magma rock based on big data. DETAILED DESCRIPTION

[0101] The present application will be further described below through specific embodiments, and the illustrative embodiments of the present application and the description are used to explain the present application, but are not limited to the present application.

[0102] The present application is a method for predicting the spatial distribution of magma rock based on big data.

[0103] As shown in Figure 1 The present application is a method for predicting the spatial distribution of magma rock based on big data.

[0104] Obtaining state data and historical data of a predicted magmatic rock region, preprocessing the state data and the historical data; the state data includes geophysical data and remote sensing data;

[0105] Collecting magmatic rock samples of the predicted magmatic rock region, performing component analysis and structural analysis on the samples, and constructing a shape influence factor according to the analysis results;

[0106] Performing inversion analysis on the geophysical data to obtain first state data, matching the component analysis results with the first state data to obtain first state parameters, matching the structural analysis results with the remote sensing data to obtain second state parameters, and combining the first state parameters and the second state parameters to obtain magmatic rock first prediction parameters;

[0107] Establishing a magmatic rock historical database according to the historical data, matching the magmatic rock first prediction parameters with the magmatic rock historical database, and performing data fusion on the matching results to obtain magmatic rock prediction parameters;

[0108] Performing geophysical simulation according to the magmatic rock prediction parameters, and determining a magmatic rock spatial distribution prediction result according to a geophysical simulation result.

[0109] In this embodiment, the method for constructing a shape influence factor according to the analysis results includes:

[0110] Collecting magmatic rock samples and surrounding rock layer samples of the predicted magmatic rock region, performing component analysis on the magmatic rock samples to obtain component analysis results, performing structural analysis on the magmatic rock samples to obtain structural analysis results, and measuring the hardness and density of the surrounding rock layer samples; the component analysis results include formation year, mineral component, density, hardness, and type; the structural analysis results include magmatic rock joint, fracture, and fault conditions;

[0111] Inputting the analysis results of the magmatic rock samples and the hardness and density of the surrounding rock layer samples into a shape influence function to obtain a shape influence factor, and the expression is:

[0112]

[0113] wherein Q F is a magmatic rock shape influence factor, w1, w2, and w3 are weight coefficients of the shape influence factor, f l is magmatic rock hardness, f i is surrounding rock layer hardness, is an average value of the surrounding rock layer hardness, p l is magmatic rock density, p i is surrounding rock layer density, is an average value of the surrounding rock layer density, and n is the number of surrounding rock layers, Age is the formation year of the magmatic rock, and C lC is a category of magmatic rock i S is a category of surrounding rock l is a structural feature of the magmatic rock

[0114] In the actual evaluation, the magmatic rock sampling analysis sample analysis data (taking 3 groups of sample data mean) of a certain place are analyzed: the hardness of the magmatic rock is 75, the density is 2.68 g / cm 3 , the age is 1200 million years, the intrusive rock is granite, and the structure is complex; the hardness of the surrounding rock 1 is 50, the density is 2.60 g / cm 3 ; the hardness of the surrounding rock 2 is 65, the density is 2.75 g / cm 3 ; the hardness of the surrounding rock 3 is 55, the density is 2.65 g / cm 3 ; the hardness of the surrounding rock 4 is 60, the density is 2.7 g / cm 3 ; the analysis results of the magmatic rock sample, the hardness and the density of the surrounding rock sample are input into the shape influence function to obtain a shape influence factor Q F = 1.1961.

[0115] In this embodiment, the method for obtaining the first state data by inverting the geophysical data comprises the following steps:

[0116] The geophysical data is input into an inversion model to obtain a simulation value of a physical property parameter of the magmatic rock, and a single objective function of the physical property parameter is constructed; the physical property parameter comprises density, magnetic force and seismic wave velocity;

[0117] The expression of the single objective function of the physical property parameter is as follows:

[0118]

[0119] Wherein Aim ρ1 is the first objective function of the density, Aim ρ2 is the second objective function of the density, n1 is the gravity sampling quantity, ρ i is the measured data of the density, ρ' i is the simulation data of the density, max is the maximum value, min is the minimum value, Aim v1 is the first objective function of the seismic wave velocity, Aim v2 is the second objective function of the seismic wave velocity, n2 is the seismic wave sampling quantity, v i is the measured data of the seismic wave velocity, ρ' i is the simulation data of the seismic wave velocity, Aim B1 is the first objective function of the magnetic force, n3 is the magnetic force sampling data quantity, B i is the measured data of the density, B' i is the simulation data of the density;

[0120] The judgment matrix is constructed according to the single objective function to perform consistency test, the evaluation value is calculated according to the judgment vector to obtain a transition objective function, and the total objective function is determined according to the transition objective function, and the expression is as follows:

[0121]

[0122] wherein Aim final is the total objective function, n4 is the number of single objective functions, w i is the transition objective function weight, Aim i is the transition objective function, the transition objective function corresponds to the single objective function one by one;

[0123] The inversion gradient is calculated according to the partial derivative of the single objective function, the inversion model is optimized according to the total objective function and the inversion gradient, and the iteration is continuously iterated until the total objective function is minimized to stop iteration to obtain the optimized inversion model;

[0124] The predicted geophysical data are input into the optimized inversion model to obtain the physical parameters of the magma rock, and the first state data of the magma rock are predicted according to the physical parameters of the magma rock; the first state data includes the density, burial depth, thickness and shape of the magma rock;

[0125] In actual evaluation, a set of geophysical data of the predicted magma rock area is obtained: gravity field 9.813 m / s 2 , background magnetic field 50000 nT, seismic wave velocity 5000 m / s; the optimized inversion model is used to invert the data to obtain the physical parameters of the magma rock: magma rock density 2.65 g / cm 3 , magnetic parameter 0.5 A / m, magma rock body seismic wave velocity 6000 m / s; the first state data is predicted by combining the inversion results of multiple sets of geophysical data at different positions: magma rock density 2.70±0.1 g / cm 3 , top burial depth-200±20 m, thickness 100±5 m, long strip shape left thick right thin.

[0126] In this embodiment, the method for obtaining the first predicted parameters of the magma rock comprises:

[0127] The first state parameters are obtained by directly matching the composition analysis results with the first state data according to the density data; the rule of direct matching is that when the density difference is less than a set threshold, it is determined as a set of matching data; the first state parameters are the mineral composition, density, hardness, burial depth, thickness and shape of the magma rock;

[0128] The main remote sensing feature data are obtained by principal component analysis of the remote sensing data, the first state data are obtained by principal component analysis of the remote sensing data, the correlation degree of the main remote sensing feature data and the prediction results is calculated, and the data set is selected as the second state data according to the correlation degree; the main remote sensing feature data include exposed magma rock coordinates, shape characteristics, texture characteristics and surrounding geology;

[0129] The ant colony search algorithm is used to match the analysis result and the second state data, and the shortest path determination data matching result is obtained, and the specific steps are as follows:

[0130] The original population is divided into a common search group and an exploration search group according to 8:2, the initial path from i point to j point is generated, the common search group is searched and the pheromone is updated, and the expression is:

[0131]

[0132] Among them The pheromone of the common search group, t is the iteration number, h p The pheromone evaporation rate of the common search group, The common search group pheromone increment, δ1 is the bias item index, Q p The common search group pheromone hardness, L p The path length of the common search group, R p The common search group path;

[0133] Then the exploration search group is searched and the common search group pheromone is updated, and the expression is:

[0134]

[0135] Among them The pheromone of the exploration search group, t is the iteration number, h k The pheromone evaporation rate of the exploration search group, The exploration search group pheromone increment, δ2 is the bias item index, Q k The pheromone hardness of the exploration search group, λ is the hardness coefficient, L k The path length of the exploration search group, R k The exploration search group path;

[0136] After the pheromone of the common search group and the exploration search group is updated, the nodes with high common search group pheromone are punished, and the expression is:

[0137]

[0138] Among them The common search group pheromone after punishment, The common search group pheromone before punishment, The pheromone of the exploration search group, t is the iteration number, the current path of the common search group is less than 3 times the current shortest path, q=0 is set, the path pheromone is updated, and on the contrary, q=1 is set, and the path pheromone is not updated;

[0139] The probability of algorithm mutation is adjusted, and the probability expression of state transition is:

[0140]

[0141] wherein P ij is the transition probability, τ ij is the pheromone from i to j, η ij is the heuristic value from i to j, α is the pheromone weight, β is the heuristic weight, a k is the search candidate point, is the penalty factor, δ3 is the bias term index, p is the random number, p0 is the mutation probability, T is the maximum iteration number;

[0142] The path is constantly searched, the pheromone and the shortest path are updated, and the iteration is completed until the algorithm converges, the shortest path is output, the data matching result is determined according to the shortest path, and the second state parameter is composed of the matching result; the second state parameter includes magma rock joints, faults, top view shape, distribution range and surrounding geology;

[0143] The first predicted parameter of the magma rock is obtained by combining the first state parameter and the second state parameter, and the thickness, shape and distribution range of the magma rock in the first predicted parameter are corrected according to the shape influencing factor;

[0144] In actual evaluation, the first state parameter is obtained according to the direct matching of the density data and the component analysis result and the first state data: the magma rock of the intrusive granite structure, the main component is quartz and feldspar, the density is 2.69 g / cm 3 , the hardness is 75, the buried depth is 3000±500 m, the thickness is 500±50 m, and the long strip shape is left thick and right thin;

[0145] After principal component analysis and correlation calculation are performed on the remote sensing characteristic data of a predicted magma rock area, a group of data with the highest correlation is determined as the second state data: the surface top view shape is a long strip mountain, the distribution range is predicted to be expanded outward by 2 kilometers, there is no exposed magma rock coordinate, the shape feature is steep slope and flat top, the texture feature is not obvious, the north is gray / red meat Paleozoic rock mass, the south is mainly mixed metamorphic rock, the east is adjacent to a large fault, and the west is flat sedimentary rock. The second state data and the tectonic analysis result (the horizontal joint surface is rough, the crack width is large, the vertical joint is smooth, the cooling crack and the tectonic fissure are obvious, and there is an upward moving fault) form the second state parameter.

[0146] In this embodiment, the method for matching the first predicted parameter of the magma rock and the magma rock historical database comprises:

[0147] The historical data is classified, a historical data similarity matrix is constructed, K-Mean clustering and hierarchical clustering combined two-step clustering are performed based on the similarity matrix to obtain magma rock parameters and surrounding geological parameters, principal component analysis is performed on the classification results to obtain main magma rock parameters and main surrounding geological parameters, and a magma rock historical database is constructed;

[0148] The comprehensive similarity of the corresponding feature vectors of the top view shape and surrounding geological data in the first predicted parameters of the magma rock and the corresponding feature vectors of the main surrounding geological parameters is calculated to obtain first matching data, the comprehensive similarity of the corresponding feature vectors of the remaining data in the first predicted parameters of the magma rock and the corresponding feature vectors of the main magma rock parameters is calculated to obtain second matching data, and the first matching data and the second matching data are combined to obtain a matching data group;

[0149] In actual evaluation, the first predicted parameters of the predicted magma rock region are matched to obtain the corresponding matching data in the magma rock historical database by using the above method: the structure of the intrusive granite is 2.72 g / cm3, the density of the magma rock is 2.72 g / cm3, the top buried depth is -218 m, the thickness is 92 m, the long strip shape is right thick and left thin, the distribution range is about 2.5 kilometers outwardly expanding, the surface topography is wide mountain, the mountain is high in the middle and steep on both sides, the horizontal joint is rough, the crack is obvious, the vertical joint is smooth, the cooling crack is obvious, and there is a fault moving upward. 3

[0150] The second matching data is the local surrounding geological parameters: there are multiple faults and fold structures around the magma rock, rock crushing and displacement phenomena can be seen in the fault zone, the upper rock layer 1 of the magma rock is a 100±20m thick sandstone layer, the upper rock layer 2 of the magma rock is a 120±20m thick shale layer, the lower rock layer 1 of the magma rock is a 200±40m thick gneiss layer, and the lower rock layer 2 of the magma rock is a 100±20m thick schist layer.

[0151] In this embodiment, the matching results are fused to obtain the method of predicting the parameters of the magma rock, which comprises:

[0152] The matching data group is divided into a training set and a test set;

[0153] A magma rock spatial distribution prediction model is constructed, which comprises an input layer, a fusion layer and an output layer, and an encoder and a decoder are set;

[0154] The two groups of matching data are standardized and output to the encoder by the input layer, the encoder sets a self-attention layer and a multi-head attention layer, the self-attention layer is used to process the data dependence within each database, and the multi-head self-attention layer is used to capture semantic information to enhance the expression of the model; the self-attention layer adopts an efficient additive attention mechanism, and the specific steps are as follows:

[0155] The attention weight of the data is calculated, and the expression is: ​

[0156]

[0157] where w i is the attention weight of input vector F i in , s(·) is the score function, l is the total number of input vectors, and q is the target vector;

[0158] A consistent global query vector is created, and an efficient additive attention mechanism is determined according to the query vector, and the expression is:

[0159]

[0160] where is the input vector processed by the efficient additive attention mechanism, Q is the normalized query matrix, L(·) is the linear projection function, K is the normalized key matrix, n is the sequence length, d is the embedding vector dimension, and Q i is the corresponding query vector of input vector x i ;

[0161] After the training set is processed by the encoder, a multi-scale cross-axis attention mechanism is used in the fusion layer for data fusion; the multi-scale cross-axis attention mechanism includes multi-scale convolution and axis attention mechanism, and the steps are represented as:

[0162] Data is encoded and summed in different directions in three parallel one-dimensional convolutions, and the expression is:

[0163]

[0164] where F ix is the multi-scale convolution along the horizontal direction, F iy is the multi-scale convolution along the vertical direction, is a one-dimensional convolution layer with three different kernel sizes, i∈{1,2,3} is the index of the convolution layer, x and y are the directions of the spatial dimension, and Conv 1×1 is a 1×1 convolution operation, is the normalized encoder input vector;

[0165] In the horizontal direction, F ix generates value vector V ix and key vector K ix , F iy generates query vector Q iy , in the vertical direction, F iy generates value vector V iy and key vector K iy , and F ix generates query vector Q ixThe attention mechanism is used for the three generated vectors, and the expression is:

[0166] F E1 = MHCAx(F ix , F ix , F iy )

[0167] F E2 = MHCA y (F ix , F iy , F iy )

[0168] Where F E1 is the horizontal output feature after multi-head cross attention mechanism processing, F E2 is the vertical output feature after multi-head cross attention mechanism processing, MHCA x is the multi-head cross attention mechanism along the horizontal direction, and MHCA y is the multi-head cross attention mechanism along the vertical direction.

[0169] The multi-head cross attention mechanism processing result is convolved, the vector output by the encoder is interacted with the fusion layer output vector in the decoder to obtain fusion data, and the output result is converted through the output layer, and the expression is:

[0170] F i out = Conv 1×1 (F E1 ) + Conv 1×1 (F E2 ) + F i in

[0171] Where F i out is the data feature vector of the output, Conv 1×1 (F E1 ) is the convolution result of F E1 , and Conv 1×1 (F E2 ) is the convolution result of F E2 .

[0172] The Huber Loss loss function is used to evaluate the difference between the predicted value and the true value, the Adam optimizer is used to update the model parameters, and the test set is used to evaluate the magma rock spatial distribution prediction model.

[0173] The target state data to be predicted is input into the magma rock spatial distribution prediction model to obtain the magma rock prediction parameters.

[0174] In the actual evaluation, the matching data is input into the magma rock spatial distribution prediction model for data fusion to obtain magma rock prediction parameters: intrusive granite structure, magma rock density 2.69 g / cm 3 , hardness 75, top buried depth-230 m, thickness 90 m, long strip right thick left thin, distribution range about 2.3 kilometers outward expansion, surface topography is wide mountain, mountain middle high and both sides steep, magma rock has upward moving fault, there are many faults and fold structures around the magma rock, rock layer 1 above the magma rock is sandstone layer (100±10 m thick, density 2.60 g / cm 3 ), rock layer 2 above the magma rock is shale layer (120±10 m thick, density 2.75 g / cm 3 ), rock layer 1 below the magma rock is gneiss layer (200±10 m thick, density 2.65 g / cm 3 ), and rock layer 2 below the magma rock is schist layer (100±10 m thick, density 2.7 g / cm 3 ).

[0175] In the embodiment, the method for determining the magma rock spatial distribution prediction result according to the geophysical simulation result comprises the following steps:

[0176] In the geophysical simulation, a geophysical model of the target area is constructed according to the magma rock prediction parameters, the gravity simulation value and the magnetic field simulation value of the target area are obtained, and the difference between the geophysical data and the simulation value is calculated;

[0177] When the difference is less than the set threshold value, the magma rock prediction parameters are output as the magma rock spatial distribution prediction result;

[0178] When the difference is greater than the set threshold value, the magma rock spatial distribution prediction model is optimized according to the difference between the geophysical data and the simulation value, the target state data to be predicted is input into the optimized magma rock spatial distribution prediction model again to obtain the magma rock prediction parameters, the geophysical simulation and the threshold value determination step are repeated, and the optimization is stopped until the difference is less than the set threshold value, and the magma rock spatial distribution prediction result is output;

[0179] The method for optimizing the hyperparameters of the magma rock spatial distribution prediction model according to the difference between the geophysical data and the simulation value comprises the following steps:

[0180] The parameter matrix of the magma rock spatial distribution prediction model is initialized, and the corresponding parameter positions are recorded, the chaos mapping is used to initialize the population, and the expression is as follows:

[0181]

[0182] Wherein, x is the initial position of the ith individual, Z i is the sequence after the chaos mapping, x max is the upper bound of the search space, and xmin where α is a chaotic parameter;

[0183] The fitness of the population is calculated, and the best fitness value is recorded. Three population individuals are randomly selected for mutation, crossover and selection operations, and the expression is:

[0184] V i = X r1 + μ(X r2 -X r3 )

[0185]

[0186] where V i is a new individual generated by mutation, X is the parent population individual, r1, r2, r3 are different random integers in [1, Q], Q is the population size, μ is the mutation factor, U ij is a new individual generated by crossover, P cr is the crossover probability, j rand is a random integer in [1, γ], is a new population individual generated by greedy selection, t is the iteration number, f(·) is the objective function, X i is the original individual;

[0187] The nonlinear convergence factor is updated, and the expression is:

[0188] θ = 1 - ln[exp{sin(t / T) / sin2-1}] 1 / 2

[0189] where θ is the nonlinear convergence factor, and T is the maximum iteration factor;

[0190] The position coefficient A, the position coefficient weight c, the random number l and the individual probability p are updated. The individual position is updated according to the position coefficient A. The individual fitness is calculated according to the objective function, and the best fitness is recorded. The iteration is stopped until the maximum iteration number is reached. The global optimal solution and the fitness value are output;

[0191] The position updating rule is:

[0192] When |A|≥1, the individual position updating expression is:

[0193] X(t+1) = X rand (t)-A|2r2·X * (t)-X(t)|

[0194] A = (2r1-1)(2-2t / T)

[0195] wherein X(t+1) is the updated individual position in the search stage, t is the current iteration number, T is the maximum iteration number, X rand is a randomly selected individual position, r1, r2 are random numbers in [0, 1], X * (t) is the current best individual position, X(t) is the current position;

[0196] When |A|<1 and p<0.5, the individual position update expression is:

[0197]

[0198] wherein X(t+1) is the updated individual position in the search stage, Levy(y) is a random step length, used to expand the local search space, y is the problem dimension, β is a constant, σ μ , σ v is a standard deviation,

[0199] When |A|<1 and p≥0.5, the individual position update expression is:

[0200]

[0201] wherein X(t+1) is the updated individual position using the spiral update mechanism, b is a constant;

[0202] In the actual evaluation, according to the magmatic rock prediction parameters, a magmatic rock geophysical model is constructed to obtain the gravity simulation value 9.805 m / s 2 , the background magnetic field simulation value 48850 nT, and the seismic wave velocity 4969 m / s of the target area, and the corresponding threshold values are 0.005 m / s 2 , 1000 nT, and 100 m / s. At this time, the difference between two of the geophysical data and the simulation value is greater than the preset threshold value, and the magmatic rock spatial distribution prediction model needs to be optimized;

[0203] According to the difference between the geophysical data and the simulation value, the parameters of the magmatic rock spatial distribution prediction model are optimized, and according to the magmatic rock prediction parameters output by the optimized magmatic rock spatial distribution prediction model, Magma rock geophysical modeling obtained the gravity simulation value 9.815 m / s 2 , the background magnetic field simulation value 50500 nT, and the seismic wave velocity 5013 m / s of the target area, and at this time, the difference between three of the geophysical data and the simulation value is less than the preset threshold value, and the magmatic rock prediction parameters are output as the magmatic rock spatial distribution prediction result (intrusive granite structure, magmatic rock density 2.71 g / cm 3The top buried depth is 235 m±10 m, the thickness is 95±10 m, the long strip shape is right thick and left thin, the distribution range is about 2.3 kilometers outwardly expanded, and the magma rock has the fault condition of upward movement.

[0204] In a second aspect, a magma rock spatial distribution prediction system based on big data comprises:

[0205] A data acquisition module is configured to acquire state data and historical data of a predicted magma rock region, and to pre-process the state data and the historical data.

[0206] A data processing module is configured to construct a shape influence factor according to an analysis result, to perform inversion analysis on the geophysical data to obtain first state data, to obtain first state parameters, first state parameters, magma rock first prediction parameters, and a magma rock historical database, and to match the magma rock first prediction parameters and the magma rock historical database.

[0207] A fusion model module is configured to construct a magma rock spatial distribution prediction model according to the matched data set, to fuse the matched data set through the magma rock spatial distribution prediction model to obtain magma rock prediction parameters.

[0208] A simulation module is configured to perform geophysical simulation according to the magma rock prediction parameters, to determine a magma rock spatial distribution prediction result according to a geophysical simulation result, and to visually display the magma rock spatial distribution prediction result to a user.

[0209] An optimization module is configured to optimize the magma rock spatial distribution prediction model according to a difference between the geophysical data and the geophysical simulation result.

[0210] An intelligent supervision module is configured to store, view, and manage the target state data, the historical data, and the magma rock spatial distribution prediction result.

[0211] The above description is only a preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting the spatial distribution of igneous rocks based on big data, characterized in that: The following steps are involved: S1. Acquire state data and historical data of the predicted igneous rock area, and preprocess the state data and historical data; the state data includes geophysical data and remote sensing data; S2. Collect igneous rock samples from the predicted igneous rock area, perform composition analysis and structural analysis on the samples, and construct shape influencing factors based on the analysis results; S3. Performing inversion analysis on the geophysical data to obtain first state data, matching a component analysis result with the first state data to obtain a first state parameter, matching a structural analysis result with the remote sensing data to obtain a second state parameter, and combining the first state parameter and the second state parameter to obtain a first prediction parameter for igneous rock; S4. Establishing a magmatic rock historical database based on the historical data, matching the first prediction parameter of the magmatic rock with the magmatic rock historical database, and performing data fusion on the matching results to obtain the magmatic rock prediction parameter; S5. Performing geophysical simulation based on the igneous rock prediction parameters, and determining the igneous rock spatial distribution prediction result based on the geophysical simulation result.

2. The method for predicting the spatial distribution of igneous rocks based on big data according to claim 1, characterized in that: The method for constructing a shape influence factor according to the analysis result includes: Collecting magmatic rock samples from the predicted magmatic rock area and samples of rock formations surrounding the magmatic rock, performing composition analysis on the magmatic rock samples to obtain composition analysis results, performing structural analysis on the magmatic rock samples to obtain structural analysis results, and measuring the hardness and density of rock formation samples surrounding the magmatic rock; The analysis results of the igneous rock sample and the hardness and density of the surrounding rock samples are input into the shape influence function to obtain the shape influence factor, which is expressed as: where Q F is the igneous rock shape influencing factor, w1, w2, w3 are the shape influencing factor weight coefficients, f l is the hardness of igneous rock, f i is the hardness of the surrounding rock formation, is the mean hardness of the surrounding rock formation, ρ l is the density of igneous rock, ρ i is the density of the surrounding rock layer, is the mean density of the surrounding rock layers, n is the number of surrounding rock layers, Age is the year of formation of the igneous rock, C l For igneous rock type, C i is the type of surrounding rock formations, S l It is a characteristic of igneous rock structure.

3. The method for predicting the spatial distribution of igneous rocks based on big data according to claim 1, characterized in that: The method of performing inversion analysis on the geophysical data to obtain first state data includes: Input geophysical data into the inversion model to obtain simulated values ​​of magmatic rock physical parameters and construct a single objective function of physical parameters; The single objective function expression of physical property parameters is: Where Aim ρ1 is the first objective function of density, Aim ρ2 is the second objective function of density, n1 is the number of gravity sampling, ρ i is the measured density data, ρ' i is the density simulation data, max is the maximum value, min is the minimum value, Aim v1 is the first objective function of seismic wave velocity, Aim v2 is the second objective function of seismic wave velocity, n2 is the number of seismic wave samples, v i is the measured data of seismic wave velocity, ρ' i is the seismic wave velocity simulation data, Aim B1 is the first magnetic objective function, n3 is the number of magnetic sampling data, B i is the measured density data, B' i is the density simulation data; According to the single objective function, a judgment matrix is ​​constructed to perform consistency test. The transition objective function is obtained by calculating the evaluation value according to the judgment vector. The total objective function is determined according to the transition objective function. The expression is: Among them, Aim final is the total objective function, n4 is the number of single objective functions, w i is the weight of the transition objective function, Aim i is the transition objective function, and the transition objective function corresponds one-to-one with the single objective function; The inversion gradient is calculated based on the partial derivative of the single objective function, and the inversion model is optimized based on the total objective function and the inversion gradient. The iteration is continued until the total objective function is minimized, and then the optimized inversion model is obtained. The geophysical data to be predicted are input into the optimized inversion model to obtain the physical property parameters of the magmatic rock, and multiple groups of first-state data of the magmatic rock are predicted based on the physical property parameters of the magmatic rock.

4. The method for predicting the spatial distribution of igneous rocks based on big data according to claim 1, characterized in that: The method for obtaining the first prediction parameter of the igneous rock includes: Directly matching the component analysis result with the first state data according to the density data to obtain the first state parameter; the rule of the direct matching is that a group of matching data is determined when the density difference is less than a set threshold; Performing principal component analysis on remote sensing data to obtain main remote sensing feature data, obtaining multiple sets of bird's-eye view morphology and distribution range prediction results based on the main remote sensing feature data, calculating the correlation between the main remote sensing feature data and the prediction results, and selecting data groups as second state data according to the correlation; The ant colony search algorithm is used to match the construction analysis results with the second state data to obtain the shortest path determination data matching result. The specific steps are as follows: The original population is divided into a normal searching group and a pioneering searching group at a ratio of 8:

2. An initialization path from point i to point j is generated. The normal searching group is used to search and update pheromones first. After completing the pheromone update of the normal search group and the pioneering search group, the node with higher pheromone in the normal search group is penalized. The expression is: in For the ordinary search group pheromone after punishment, For the ordinary search group pheromone before punishment, To develop the pheromone of the search group, t is the number of iterations. When the current path of the ordinary search group is less than 3 times the current shortest path, set q = 0 to update the path pheromone. Otherwise, set q = 1 and do not update the path pheromone. Adjust the probability of algorithm mutation, and the probability expression of state transition is: Among them, P ij is the transition probability, τ ij is the pheromone from point i to point j, η ij is the heuristic value from point i to point j, α is the pheromone weight, β is the heuristic value weight, a k To search for candidate points, is the penalty factor, δ3 is the bias term index, p is a random number, p0 is the mutation probability, and T is the maximum number of iterations; Continuously search for paths, update pheromones and shortest paths, and iterate until the algorithm converges and completes the iteration. Output the shortest path, determine the data matching result based on the shortest path, and the matching result constitutes the second state parameter; The first state parameter and the second state parameter are combined to obtain a first prediction parameter of the igneous rock.

5. The method for predicting the spatial distribution of igneous rocks based on big data according to claim 1, characterized in that: The method for matching the first prediction parameter of the igneous rock with the igneous rock history database includes: Classify and perform principal component analysis on historical data to obtain the main magmatic rock parameters and surrounding geological parameters, thus forming a magmatic rock history database; The first matching data is obtained by calculating the comprehensive similarity between the feature vectors corresponding to the overhead morphology and surrounding geological data in the first prediction parameter of the igneous rock and the feature vectors corresponding to the main surrounding geological parameters. The second matching data is obtained by calculating the comprehensive similarity between the feature vectors corresponding to the remaining data in the first prediction parameter of the igneous rock and the feature vectors corresponding to the main igneous rock parameters. The first matching data and the second matching data are combined to obtain a matching data group.

6. The method for predicting the spatial distribution of igneous rocks based on big data according to claim 1, characterized in that: The method for performing data fusion on the matching results to obtain igneous rock prediction parameters includes: Divide the matching data set into training and testing sets; Construct a magmatic rock spatial distribution prediction model, including input layer, fusion layer and output layer, and set encoder and decoder; The input layer normalizes the two sets of matching data and outputs them to the encoder. The encoder sets up a self-attention layer and a multi-head attention layer. The self-attention layer is used to process the data dependencies within each database, and the multi-head self-attention layer is used to capture semantic information to enhance the expression of the model. The self-attention layer adopts an efficient additive attention mechanism. The specific steps are as follows: Calculate the attention weight of the data, the expression is: where w i is the input vector F i in The attention weight, s(·) is the score function, l is the total number of input vectors, and q is the target vector; Create a consistent global query vector and determine the efficient additive attention mechanism based on the query vector. The expression is: in is the input vector processed by the efficient additive attention mechanism, Q is the normalized query matrix, L(·) is the linear projection function, K is the normalized key matrix, n is the sequence length, d is the embedding vector dimension, Q i is the input vector x i The corresponding query vector; After the training set is processed by the encoder, a multi-scale cross-axial attention mechanism is used in the fusion layer to perform data fusion. The multi-scale cross-axial attention mechanism includes multi-scale convolution and axial attention mechanism. The steps are as follows: The data is encoded and summed in different directions in three parallel one-dimensional convolutions. The expression is: Among them F ix is the multi-scale convolution along the horizontal direction, F iy is the multi-scale convolution along the vertical direction, are three one-dimensional convolutional layers with different convolution kernel sizes, i∈{1,2,3} is the index of the convolutional layer, x and y are the directions of the spatial dimension, Conv 1×1 is a 1×1 convolution operation, is the normalized encoder input vector; Horizontally, use F ix Generate value vector V ix and key vector K ix , use F iy Generate query vector Q iy , vertically, use F iy Generate value vector V iy and key vector K iy , use F ix Generate query vector Q ix , using the attention mechanism for the three generated vectors, the expression is: F E1 =MHCA x (F ix ,F ix ,F iy ) F E2 =MHCA y (F ix ,F iy ,F iy ) Among them F E1 is the horizontal output feature after processing by the multi-head cross attention mechanism, F E2 The vertical output features after multi-head cross attention mechanism processing, MHCA x MHCA is a multi-head cross attention mechanism along the horizontal direction. y It is a multi-head cross attention mechanism in the vertical direction; The results of the multi-head cross attention mechanism are convolved, and the vector output by the encoder is interacted with the output vector of the fusion layer in the decoder to obtain fused data, which is converted into the output result through the output layer. The expression is: F i out =Conv 1×1 (F E1 )+Conv 1×1 (F E2 )+F i in Among them F i out is the output data feature vector, Conv 1×1 (F E1 ) is F E1 The convolution result, Conv 1×1 (F E2 ) is F E2 The convolution result of The Huber Loss function was used to evaluate the difference between the predicted value and the true value, the Adam optimizer was used to update the model parameters, and the test set was used to evaluate the igneous rock spatial distribution prediction model; The target state data to be predicted is input into the igneous rock spatial distribution prediction model to obtain the igneous rock prediction parameters.

7. The method for predicting the spatial distribution of igneous rocks based on big data according to claim 1, characterized in that: The method for determining the prediction result of the spatial distribution of igneous rocks based on the geophysical simulation result includes: In geophysical simulation, a geophysical model of the target area is constructed based on the predicted parameters of igneous rocks, the gravity simulation values ​​and magnetic field simulation values ​​of the target area are obtained, and the difference between the geophysical data and the simulation values ​​is calculated; When the difference is less than the set threshold, the magmatic rock prediction parameters are output as the magmatic rock spatial distribution prediction results; When the difference is greater than the set threshold, the magmatic rock spatial distribution prediction model is optimized according to the difference between the geophysical data and the simulation value, the target state data to be predicted is input into the optimized magmatic rock spatial distribution prediction model again to obtain the magmatic rock prediction parameters, and the geophysical simulation and threshold determination steps are repeated until the difference is less than the set threshold, and the optimization is stopped and the magmatic rock spatial distribution prediction result is output; The method for optimizing the hyperparameters of the magmatic rock spatial distribution prediction model based on the difference between geophysical data and simulation values ​​is as follows: Initialize the parameter matrix of the magmatic rock spatial distribution prediction model, record the corresponding parameter positions, and use chaotic mapping to initialize the population. The expression is: in is the initial position of the i-th individual, Z i is the sequence after chaotic mapping, x max is the upper bound of the search space, x min is the lower bound of the search space, α is the chaos parameter; Calculate the fitness of the population and record the best fitness value. Randomly select three individuals from the population for mutation, crossover, and selection operations. The expression is: V i =X r1 +μ(X r2 -X r3 ) Where V i is the new individual generated by mutation, X is the individual of the parent population, r1, r2, r3 are different random integers in [1, Q], Q is the population size, μ is the mutation factor, U ij is the new individual generated by crossover, P cr is the crossover probability, j rand is a random integer in [1,γ], is the new population individual generated by greedy selection, t is the number of iterations, f(·) is the objective function, X i For the original individual; Update the nonlinear convergence factor, the expression is: θ=1-ln[exp{sin(t / T) / sin2-1}] 1 / 2 Where θ is the nonlinear convergence factor and T is the maximum iteration factor; Update the position coefficient A, position coefficient weight C, random number l and individual probability p, update the individual position according to the position coefficient A, calculate the individual fitness according to the objective function and record the best fitness, iterate until the maximum number of iterations is reached and stop the iteration, output the global optimal solution and fitness value.

8. A big data-based igneous rock spatial distribution prediction system for executing the method according to any one of claims 1 to 7, characterized in that: include: Data acquisition module: used to obtain status data and historical data of the predicted igneous rock area, and pre-process the status data and historical data; Data processing module: used for constructing shape influencing factors according to the analysis results, for performing inversion analysis on the geophysical data to obtain first state data, for obtaining first state parameters, first state parameters, first predicted parameters of igneous rocks and a igneous rock history database, and for matching the first predicted parameters of igneous rocks with the igneous rock history database; Fusion model module: used for constructing a magmatic rock spatial distribution prediction model based on the matching data set, and obtaining magmatic rock prediction parameters by fusing the matching data set with the magmatic rock spatial distribution prediction model; Simulation module: used to perform geophysical simulation based on the igneous rock prediction parameters, determine the igneous rock spatial distribution prediction results based on the geophysical simulation results, and visualize the igneous rock spatial distribution prediction results to users; Optimization module: used to optimize the magmatic rock spatial distribution prediction model based on the difference between geophysical data and geophysical simulation results; Intelligent supervision module: used to store, view and manage the target status data, the historical data and the igneous rock spatial distribution prediction results.

Images