Prospecting method for deep copper-nickel sulfide deposit based on multi-source geochemical evidence

Abstract

The present application relates to the technical field of multi-source data processing, in particular to a deep copper-nickel sulfide deposit prospecting method based on multi-source geochemical evidence, comprising the following steps: in the present application, the distribution probability is extracted and the branch replacement continuous operation gradient is selected by substituting the cooling attenuation function into the simulated annealing algorithm, so as to force the multi-dimensional scalar space mapping to approach the global extreme threshold of physical and chemical evolution, the state scalar is deduced by substituting the state value into the Markov decision process and continuously executing the multi-order time step algebra mapping operation, so as to solve the problem that the static structured geological index array is difficult to handle the dynamic silica crystallization disturbance residual value drift problem, the transition distribution matrix is constructed by extracting the crustal contamination extreme value arrangement extreme value through the operation residual, the state is confirmed by mapping the grid operation drilling nickel isotope distribution quantity comparison threshold limit, and the scheme effect of quantifying the three-dimensional Euclidean solid boundary envelope range of the deep blind ore body occurrence space is achieved.

Description

Technical Field

[0001] This invention relates to the field of multi-source data processing technology, and in particular to a method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence. Background Technology

[0002] The field of multi-source data processing technology is a computational paradigm that executes multiple mathematical and logical operations, including heterogeneous matrix variable fusion operations, multivariate scalar nonlinear mapping, and spatial coordinate system targeted projection. The specific execution steps include calling the Pearson correlation coefficient to calculate the collinearity index of multiple independent source input scalars, constructing a Markov transition probability matrix to smooth the deviation of discrete spatial observation matrix, inputting the multivariate scalar set to the hidden layer nodes of a multilayer perceptron with ridge regression regularization penalty term, performing nonlinear transformation operations with logistic activation function, and outputting a quantized fusion result scheme with a preset root mean square error loss function valley value limit. The technology focuses on the hierarchical deep algebraic operations of arithmetic logic units of static structured geological index arrays.

[0003] The prospecting method for deep copper-nickel sulfide deposits based on multi-source geochemical evidence specifically employs a multi-variable evaluation and calculation scheme that couples rock major and trace isotope deviations, chalcophile element fractionation coefficients, and isotope mass ratios. The scheme aims to solve the problem of spatial three-dimensional Euclidean coordinate positioning of deep, concealed blind ore bodies. To achieve the spatial target area envelope positioning effect of the comprehensive mineralization potential index, the scheme extracts the chalcophile element concentration from the rock mass and inputs Rayleigh fractionation law differential equations to indicate the unknown degree of segregation in deep sulfide melts. It also inputs the comprehensive alteration degree of the whole rock and the cobalt-nickel element mass ratio scalars to calculate the approximate quantitative index using a similarity preference ranking mathematical model. Furthermore, it uses the analytic hierarchy process (AHP) to construct a multi-source evidence judgment matrix, employs a power iteration algorithm to solve for suitable eigenvalues ​​and associated eigenvectors, and weights and superimposes multiple independent geochemical scalars to output a quantitative prediction model indicating the occurrence state of deep blind ore bodies.

[0004] Existing technologies focus on deep algebraic operations at the arithmetic logic unit level of static structured geological index arrays. Numerical evaluation paradigms are limited to capturing the absolute linear mapping trajectories between multiple variables. Faced with the high-dimensional nonlinear isotopic mass ratio scalar fluctuation environment in deep space, they cannot quantify the gradient changes in complex curve partial derivative correlations. This leads to the omission of deep nonlinear isomorphic feature terms in the multidimensional spatial numerical projection matrix, causing anomalies in potential mineralization target areas to be smoothed out by background noise. Inputting a multivariate scalar set to the hidden nodes of a multilayer perceptron with ridge regression regularization penalty terms and relying on a preset root mean square error loss function to perform gradient descent optimization is prone to getting stuck in local valleys when dealing with non-convex optimization target spaces with multiple extrema, causing backpropagation weight parameter updates to stop. The lag and gradient vanishing phenomenon in the saturation region of the logistic activation function of the multilayer perceptron (MLP) cause the extreme values ​​of the output quantization fusion result to deviate significantly from the global optimal solution. This leads to the delineation of the Euclidean coordinate boundary in three-dimensional space deviating from the actual physical occurrence center point of the blind ore body. At the same time, the scheme of constructing a multi-source evidence judgment matrix using the analytic hierarchy process (AHP) and solving the eigenvalues ​​using the power iteration algorithm is constrained by the initial static matrix structure and lacks a time-step-level dynamic environmental evolution state inference mechanism. As a result, in the context of dealing with multiple dynamic melting extremum tensor operations of residuals caused by magmatic silicate crystallization disturbances, the output weighted superimposed scalar cannot dynamically adapt to the spatial discrete drift of multiple independent geochemical scalars, causing the numerical value of the three-dimensional envelope range of the target area of ​​the quantization prediction model to seriously exceed the limit and lose its engineering guidance significance. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence, comprising the following steps: Step 1: Based on the geochemical content of borehole elements and the dual diluent array, calculate the deviation of nickel isotopes in the borehole, calculate the partial derivative matrix of fractionation coefficient combination, calculate the Hessian condition number and compare the extreme value limits, aggregate the difference scalar arrangement spatial array, and obtain the fractionation residual array. Step 2: Based on the fractionation residual array, superimpose and calculate the silica crystallization perturbation residual, compare the residual terms to determine the deviation state, introduce the simulated annealing algorithm to extract the distribution probability, select the replacement branch to continuously calculate the gradient descent, aggregate the optimization sequence to arrange the tensor elements, and obtain the melt separation extreme value tensor. Step 3: Based on the aforementioned melting extremum tensor, calculate the overall alteration degree of the borehole rock and the mass ratio of cobalt and nickel elements, arrange scalars to construct a sulfide feature matrix and calculate the consistency ratio, compare the ratio terms with the out-of-bounds interruption beacon, extract the abnormal discrete scalars of the matrix, and locate the discrete conflict nodes. Step 4: Based on the discrete conflict nodes, extract the crustal contamination extreme values ​​by calculating the residuals, arrange the extreme values ​​to construct the transition distribution matrix, deduce the state scalar through the Markov decision process, map and replace to reconstruct the feature matrix, iteratively compare the threshold to extract the feature roots, and construct the potential weight matrix. Step 5: Based on the discrete conflict nodes and the potential weight matrix, map the grid to calculate the distribution of nickel isotopes in the borehole, compare the distribution scalar with the threshold boundary to confirm the status, remove the unmineralized conventional truncated three-dimensional spatial coordinates, arrange the connecting lines to construct a multi-dimensional three-dimensional bounding box, and establish the three-dimensional coordinates of the target area.

[0007] As a further embodiment of the present invention, the fractionation residual array includes isotope deviation, elemental partial derivative distribution matrix, and multidimensional three-dimensional structure array; the melt separation extreme value tensor includes silicate crystallization perturbation residual, exponential distribution probability, and grid structure feature array; the discrete conflict nodes include whole-rock comprehensive alteration degree, cobalt-nickel element mass ratio, and anomalous coordinate points; the potential weight matrix includes crustal contamination extreme values, multi-level transition state scalars, and weight coefficient arrays; and the target area three-dimensional coordinates include a set of cross-boundary spatial nodes, borehole element geochemical content and borehole nickel isotope Z-axis scalar, multidimensional three-dimensional border, and deep spatial envelope range.

[0008] As a further aspect of the present invention, the specific steps for obtaining the fractionation residual array are as follows: Based on the geochemical content of borehole elements and the dual diluent array, the continuous spectral bands were extracted by calibrating the peak overlap state, the discrete mass-charge ratio sequence of borehole nickel isotopes was analyzed and the deviation of borehole nickel isotopes was calculated, the row and column partial derivatives of the multivariate constant product arrangement matrix were calculated, and the element partial derivative distribution matrix was established. Based on the element partial derivative distribution matrix, the diagonal scalar of the matrix is ​​intercepted at the wandering diagonal position, the Hessian condition number is determined by comparing the numerical boundary, the deviation difference is calculated to update the grid translation coordinate point, the translation difference is combined to arrange a multi-dimensional three-dimensional structure array, and the fractionation residual array is obtained.

[0009] As a further aspect of the present invention, the specific steps for obtaining the melt-off extreme value tensor are as follows: Based on the fractionation residual array, the step size is allocated to call the geochemical content of borehole elements to derive the silica crystallization constant, the constant is superimposed to calculate the disturbance of the residual scalar and the scalar is compared to determine the boundary, the spatial position of the abnormal node is extracted, the position item is combined to arrange the deviation set, and the residual deviation state scalar is obtained. Based on the residual deviation state scalar, the simulated annealing algorithm is introduced to map and extract the exponential distribution probability. The probability terms are compared to verify the numerical boundary and the out-of-bounds parameters are truncated to trigger coordinate replacement. The partial derivative of the replacement terms is substituted to calculate the scalar descent. The scalars are accumulated and arranged into an evolution queue to establish a gradient descent optimization sequence. Based on the gradient descent optimization sequence, the spatial convergence term at the end of the sequence is extracted, the multidimensional optimization coordinate points are reconstructed, the coordinates are substituted to fill the missing nodes of the tensor, the combined filled nodes are reshaped to reshape the multi-dimensionality, and the grid structure feature array is aggregated to obtain the fused extreme value tensor.

[0010] As a further aspect of the present invention, the simulated annealing algorithm extracts the residual deviation state scalar and sets the initial system temperature parameters. It then substitutes the initial system temperature parameters into the cooling decay function to deduce the current stage annealing temperature scalar corresponding to the initial system temperature parameters. The residual deviation state scalar and the current stage annealing temperature scalar are used to perform a quotient operation to extract the negative exponential term parameter. The natural constant base is combined with the negative exponential term parameter to perform a nonlinear mapping algebra operation to extract the exponential distribution probability.

[0011] As a further aspect of the present invention, the specific steps for locating the discrete conflict nodes are as follows: Based on the aforementioned melting extremum tensor, the tensor data is segmented and the borehole element geochemical content is called to calculate the whole-rock comprehensive alteration scalar, the built-in cobalt-nickel element mass ratio sequence of borehole element geochemical content is analyzed, the combined parameters are arranged in a multi-dimensional row and column space structure, the determinant is solved to remove the orthogonal basis vector, and the sulfide feature matrix is ​​established. Based on the sulfide feature matrix, the matrix product term is calculated to separate the associated feature vectors, the feature parameter output consistency ratio is measured, the out-of-bounds interruption beacon comparison ratio term is called, the abnormal coordinate points where the value deviates from the center position are extracted, and the discrete conflict nodes are located.

[0012] As a further aspect of the present invention, the specific steps for constructing the potential weight matrix are as follows: Based on the discrete conflict nodes, the residual is calculated to obtain the node-related scalar deviation difference. The boundary parameters are extracted and the borehole nickel isotope deviation is called to derive the crustal contamination peak parameters. The peak parameters are aggregated and arranged into a multi-dimensional row and column grid. The internal parameters of the grid are combined to construct a probability space mapping array and obtain the transition distribution matrix. Based on the transition distribution matrix, the matrix element operation of the Markov decision process is used to deduce the multi-level transition state scalar, the state scalar is compared with the convergence boundary conditions, the non-boundary stationary parameters are selected and mapped to the original conflict coordinates, the values ​​corresponding to the abnormal coordinates are replaced and the multidimensional determinant is reorganized to establish the reconstructed feature matrix. Based on the reconstructed feature matrix, the deviation scalar is obtained by multiplying the elements within the matrix. The deviation scalar is continuously compared with the convergence threshold boundary. The eigenvalue parameters of the matrix corresponding to the numerical constraints are extracted. The distribution parameters of the eigenvalue arrangement weight coefficient matrix are extracted to obtain the potential weight matrix.

[0013] As a further aspect of the present invention, the Markov decision process extracts the transition distribution matrix, sets the matrix elements built into the transition distribution matrix as discrete state node sets, defines the transition decision action items corresponding to the discrete state node sets, extracts the probability mapping parameters associated with the transition decision action items by combining the transition distribution matrix, constructs a state value inference equation, substitutes the discrete state node sets, transition decision action items and probability mapping parameters into the state value inference equation and continuously performs multi-order time step algebraic mapping operations to obtain the state value parameters under each order of time step, combines the state value parameters, and calculates multi-level transition state scalars.

[0014] As a further aspect of the present invention, the replacement of abnormal coordinates and the corresponding numerical reconstruction of the multidimensional determinant involves first extracting the original abnormal coordinates associated with the multidimensional tensor index boundaries, calling the memory pointer direct addressing instruction to locate the underlying memory block storing the abnormal values, and then using a pointer overwrite mechanism to overwrite the filtered and non-boundary stable parameters into the memory block to perform numerical replacement. This process aggregates the replacement operation's multiple parameter scalars, performs tensor dimensionality upgrading and reshaping mapping based on a preset Euclidean orthogonal basis sequence, constructs a full-rank multidimensional square matrix structure, and calls the Gaussian elimination algorithm to perform elementary row transformation matrix algebra on the full-rank multidimensional square matrix structure, mapping to generate an upper triangular matrix array. The multiple element scalars on the main diagonal of the upper triangular matrix array are then extracted and continuously subjected to product algebra operations to output the reconstructed multidimensional determinant scalar.

[0015] As a further aspect of the present invention, the specific steps for establishing the three-dimensional coordinates of the target area are as follows: Based on the discrete conflict nodes and the potential weight matrix, the scalar of the borehole nickel isotope distribution is calculated by mapping the three-dimensional mesh of the borehole, the distribution parameters are obtained by multiplying the matrices, the distribution parameters are compared with the critical boundary to confirm the state, the spatial structure of the out-of-limit mesh node arrangement is extracted, and the out-of-limit spatial node set is established. Based on the set of out-of-bounds spatial nodes, the grid state is identified and the non-out-of-bounds conventional parameters are removed. The two-dimensional plane coordinates corresponding to the remaining parameters are extracted. The geochemical content values ​​of borehole elements and the nickel isotope content values ​​of borehole elements are used to arrange the Z-axis scalar, thereby reconstructing the multi-dimensional three-dimensional border. The border parameters are aggregated to define the deep space envelope range and establish the three-dimensional coordinates of the target area.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the distributed probability is extracted by substituting the cooling decay function into the simulated annealing algorithm and the gradient is continuously calculated by selecting the replacement branch. This overcomes the limitation that conventional non-convex optimization objectives are prone to getting stuck in local valley stationary points. The algorithm relies on the built-in exponential distribution probability mapping mechanism to cross the local valley boundary, so that the multidimensional scalar space mapping approaches the global extreme value threshold of physical and chemical evolution. In this invention, the state scalar is derived by substituting the Markov decision process into the state value derivation equation and continuously performing multi-order time step algebraic mapping operations. This gives the discrete conflict nodes a time step level dynamic evolution derivation attribute, thus solving the problem of static structured geological index arrays being unable to handle the numerical drift of dynamic silica crystallization disturbance residuals. In this invention, the extreme values ​​of crustal contamination are extracted by calculating residuals to construct a transfer distribution matrix. The distribution of nickel isotopes in boreholes is compared with threshold limits to confirm the state. Unmineralized conventional cutting is removed, and three-dimensional spatial coordinates are arranged and connected to construct a multi-dimensional three-dimensional border to establish the three-dimensional coordinates of the target area. This achieves the effect of a scheme for delineating the three-dimensional Euclidean three-dimensional boundary quantization envelope range of the deep blind ore body occurrence space. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the main steps of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Example 1 Please see Figure 1 This invention provides a technical solution: a method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence, comprising the following steps: Step 1: Based on the geochemical content of borehole elements and the dual diluent array, calculate the deviation of nickel isotopes in the borehole, calculate the partial derivative matrix of fractionation coefficient combination, calculate the Hessian condition number and compare the extreme value limits, aggregate the difference scalar arrangement spatial array, and obtain the fractionation residual array. Step 2: Based on the fractionation residual array, superimpose and calculate the silica crystallization perturbation residual, compare the residual terms to determine the deviation state, introduce the simulated annealing algorithm to extract the distribution probability, select the replacement branch to continuously calculate the gradient descent, aggregate the optimization sequence to arrange the tensor elements, and obtain the melt separation extreme value tensor. Step 3: Based on the melting extremum tensor, calculate the overall alteration degree of the borehole rock and the mass ratio of cobalt and nickel elements, arrange scalars to construct the sulfide feature matrix and calculate the consistency ratio, compare the ratio terms with the out-of-bounds interruption beacon, extract the abnormal discrete scalars of the matrix, and locate the discrete conflict nodes. Step 4: Based on discrete conflict nodes, extract crustal contamination extrema from the residuals, arrange the extrema to construct a transition distribution matrix, deduce the state scalar through a Markov decision process, map and replace to reconstruct the feature matrix, iteratively compare the threshold to extract the feature roots, and construct the potential weight matrix. Step 5: Based on discrete conflict nodes and potential weight matrix, map the grid to calculate the distribution of nickel isotopes in the borehole, compare the distribution scalar with the threshold boundary to confirm the status, remove the unmineralized conventional truncation to extract the three-dimensional spatial coordinates, arrange the connection lines to construct a multi-dimensional three-dimensional bounding box, and establish the three-dimensional coordinates of the target area.

[0020] The fractionation residual array includes isotope deviation, elemental partial derivative distribution matrix, and multidimensional three-dimensional structure array; the melting extremum tensor includes silicate crystallization perturbation residual, exponential distribution probability, and grid structure feature array; the discrete conflict nodes include whole-rock comprehensive alteration degree, cobalt-nickel element mass ratio, and anomalous coordinate points; the potential weight matrix includes crustal contamination extremum, multi-level transition state scalar, and weight coefficient array; and the target area three-dimensional coordinates include the boundary space node set, borehole element geochemical content and borehole nickel isotope Z-axis scalar, multidimensional three-dimensional border, and deep space envelope range.

[0021] The specific steps for obtaining the fractionation residual array are as follows: Based on the geochemical content of borehole elements and a dual-diluent array, a Gaussian mixture model decomposition algorithm was adopted. The initial number of Gaussian distribution components was set to 5, the upper limit for the expected value maximization iteration termination was set to 500, the tolerance parameter for the log-likelihood absolute difference was set to 0.001, and the covariance matrix type was configured as full covariance mode. The borehole element geochemical content was input into the corresponding mixed Gaussian distribution component, and the expected value calculation and maximization update steps were continuously executed to remove overlapping peaks. Continuous spectral bands were extracted, and then a quadratic derivative peak-finding algorithm was used based on the continuous spectral bands, with a smoothing window set. The width parameter is 15 data points, the local fitting polynomial order parameter is set to 3, the peak signal-to-noise ratio threshold parameter is set to 10 dB, the borehole nickel isotope discrete mass-to-charge ratio sequence is continuously smoothed and the coordinates of the zero cross points of the second derivative are obtained, and the corresponding coordinate isotope deviation is extracted. Then, based on the borehole nickel isotope deviation, the forward finite difference algorithm is used, the spatial difference step size constant is set to 0.05 micrometers, the Taylor expansion truncation error order is set to 1, and the first-order difference quotient algebra operation is performed in combination with the multivariate constant product term to arrange the row and column partial derivative terms of the matrix, and the element partial derivative distribution matrix is ​​established. Based on the element-partial derivative distribution matrix, a singular value decomposition algorithm is employed. The parameter for truncating singular values ​​is set to 10, and the randomization seed constant is set to 42. A bilateral orthogonal transformation operation is continuously performed on the element-partial derivative distribution matrix, traversing the diagonal position and truncating the matrix diagonal scalar. The division operation between the maximum and minimum absolute value terms is then called to process the matrix diagonal scalar and output the Hessian condition number. Based on the Hessian condition number, a Boolean logic decision operator is used to extract a preset numerical boundary. Specifically, the preset numerical boundary is calculated by continuously calculating the arithmetic mean of 100 sets of known background benchmark rock sample test parameters and superimposing three times the standard deviation to obtain a decision threshold constant of 150. The Hessian condition number and the decision threshold constant 150 are input into a comparison register, and an absolute value subtraction test is performed on the sign bit to determine if the Hessian condition number exceeds the limit. The determination of the excessive Hessian condition number is based on the fractionation residual array and the Hessian condition number anomaly identification map. The scalar value of the fractionation residual corresponding to depth layer R4 of grid C5 is extracted as 0.88, the scalar value of depth layer R4 of grid C6 is 0.91, and the scalar value of depth layer R5 of grid C6 is 0.96. The Hessian condition number associated with the multivariate spatial grid is determined to be greater than the judgment threshold constant 150. Then, based on the excessive Hessian condition number, an affine transformation algorithm is used to extract the absolute difference between the excessive Hessian condition number and the judgment threshold constant 150. The spatial translation vector scalar parameter is set, the rotation angle parameter is set to 0 degrees, and the spatial omnidirectional scaling ratio constant parameter is set to 1. The spatial translation vector scalar parameter is used to update the grid translation coordinate points. The combined translation difference is substituted into the homogeneous coordinate transformation matrix to arrange a multidimensional three-dimensional structure array and obtain the fractionation residual array.

[0022] The specific steps to obtain the melting extremum tensor are as follows: Based on the fractionation residual array, a fourth-order Runge-Kutta numerical integration algorithm is employed. The initial iteration step size is set to 0.1 meters, and the upper limit of the integration interval is set to 1 kilometer. Four consecutive weighted slope space infinitesimal algebraic superposition operations are performed on the elements within the fractionation residual array to allocate the step size. The silicate crystallization constant is derived by calling the geochemical content of borehole elements. Then, a scalar addition allocation operator is used to perform arithmetic addition of the silicate crystallization constant and the original residual term to obtain the perturbed residual scalar. Based on the perturbed residual scalar, a unit step comparison operator is used. Extract the preset boundary constant. Specifically, the preset boundary constant is generated by extracting the core data of fifty sets of historical exploration boreholes of the same type of mineral deposit and calculating the mean of extreme residuals to generate a fixed value of 0.85. The disturbance residual scalar and the preset boundary constant 0.85 are input into the register to perform absolute value subtraction to test the sign bit and determine the boundary. Based on the boundary determination instruction, the memory pointer offset retrieval instruction is called to extract the spatial position associated with the abnormal node. The combined position items are substituted into the dynamic array to open the instruction arrangement deviation set and obtain the residual deviation status scalar. Based on the residual deviation state scalar, a simulated annealing algorithm is employed. The initial system temperature is set to 1000 degrees Celsius, the cooling decay ratio is set to 0.95, the Markov chain iterative equilibrium step count is set to 200 steps, and the minimum termination temperature is set to 1 degree Celsius. An exponential decay operation is performed on the residual deviation state scalar in conjunction with the Boltzmann constant term to extract the exponential distribution probability. Based on the exponential distribution probability, a Mahalanobis distance algorithm is used to extract a preset numerical boundary threshold. Specifically, the preset numerical boundary threshold is generated and locked using a Mason rotation pseudo-random number generator, and a constant of 0.5 is used. The exponential distribution probability is compared with the constant 0.5 to verify the numerical boundary, and out-of-bounds parameters are truncated to trigger coordinate replacement. The operation, specifically the coordinate replacement operation, is based on the convergence curve of simulated annealing optimization. At the node where the number of iterations reaches 55 and 115, the value of the objective function is extracted and falls below the convergence tolerance reference line of 0.2. The coordinate random perturbation branch is forced to jump out of the local extreme stationary point. Then, based on the coordinate replacement result, a nonlinear conjugate gradient descent algorithm is adopted. The initial search step size parameter is set to 0.01. The Fletcher-Reeves conjugate update coefficient is set. The coordinates of the replacement item are input into the objective cost function and the first-order partial derivative algebra operation is continuously performed to extract the partial derivative descent scalar. The vector accumulation operator is used to merge the search directions of the predecessor nodes and accumulate the scalar to arrange the evolution queue, and establish the gradient descent optimization sequence. Based on the gradient descent optimization sequence, a floating-point difference comparison instruction is used, with a spatial convergence tolerance parameter set to 0.01%. The absolute distance difference between the last ten adjacent elements of the gradient descent optimization sequence is continuously calculated, and the terminal spatial convergence term is extracted. Based on the terminal spatial convergence term, an affine space transformation algorithm is used, with a translation offset parameter set to 0.5 meters for the three-dimensional vector and a three-dimensional rotation Euler angle parameter set to 0 degrees. The multi-dimensional optimization coordinate points are reconstructed and mapped. Then, based on the multi-dimensional optimization coordinate points, a K-dimensional tree spatial nearest neighbor index algorithm is used, with a leaf node capacity limit parameter set to thirty points and a spatial partitioning criterion set to the dimensional bisecting plane with the largest variance. The coordinates are searched and substituted to fill the missing nodes in the tensor. Subsequently, based on the filled nodes, a tensor Kronecker product reshaping algorithm is used to extract the third-order identity matrix as the tensor multiplication basis. The filled nodes and the third-order identity matrix are combined to perform inner product expansion to reshape the multi-order dimension. The aggregation splicing operator is called to arrange the tensor slice aggregation mesh structure feature array along the first dimension axis to obtain the fused extreme value tensor.

[0023] The specific steps for locating discrete conflict nodes are as follows: Based on the fusion extremum tensor, a tensor slice addressing instruction is used, with a step size parameter set to 2 and the initial slice dimension index constant set to 0. The fusion extremum tensor is continuously subjected to a leapfrog memory extraction and segmentation of tensor data along the third spatial dimension. The borehole elemental geochemical content is then used to calculate the whole-rock comprehensive alteration scalar. Based on the whole-rock comprehensive alteration scalar, a one-dimensional floating-point array parsing operator is used, with a byte alignment length parameter set to 8 bytes and a floating-point precision standard set to double precision 64 bits. The built-in cobalt-nickel elemental mass ratio sequence of the borehole elemental geochemical content is analyzed. Specifically, the cobalt-nickel elemental mass ratio sequence is constructed based on the multi-source geochemical evidence fusion heat map, extracting the normalized scalar values ​​of borehole number ZK05 at a depth of 700 meters (0.72), borehole number ZK06 at a depth of 760 meters (0.91), and borehole number ZK07 at a depth of 820 meters (0.87) to construct a numerical sequence. Subsequently, based on the cobalt-nickel elemental mass ratio sequence... The ratio sequence is spliced ​​using a spatial mapping operator. The row dimension aggregation constant parameter is set to 5, and the column dimension aggregation constant parameter is set to 5. The whole rock comprehensive alteration scalar and the cobalt-nickel element mass ratio sequence are filled into the preset memory space in column-major order and arranged into a multi-dimensional row and column space structure. Based on the multi-dimensional row and column space structure, the principal component threshold limit constant is set to 0.0001, and the maximum number of elimination iterations is set to 100. Elementary row transformation operations are continuously performed and transformed into an upper triangular matrix array to solve the determinant scalar. Based on the determinant scalar, the Gram-Schmidt orthogonalization algorithm is used to extract the preset inner product tolerance boundary. Specifically, the preset inner product tolerance boundary is set to a constant of 0.05 by reading the dot product mean of 10 sets of standard silicate melt sample vectors. The orthogonality between the dot products of the residual vectors is continuously calculated and tested with the preset inner product tolerance boundary of 0.05. Vector subtraction iteration is performed to remove non-independent projection terms and orthogonal basis vectors, and a sulfide feature matrix is ​​established. Based on the sulfide feature matrix, the Block matrix multiplication algorithm is adopted, with the block matrix order parameter set to 4 and the cache prefetch step size constant set to 16 bytes. Block-parallel addition-multiplication algebraic operations are performed on the sulfide feature matrix and its transpose array to calculate the matrix product. Based on the matrix product, the Haushold transform QR decomposition algorithm is adopted, with the iterative convergence precision constant set to 0.00001 and the maximum decomposition step size set to 200 times. Orthogonal similarity transformations are continuously performed to extract the main diagonal elements of the upper triangular matrix to separate the associated feature vectors. Based on the feature vectors, the weighted summation operator with the largest eigenvalue is used to extract the preset consistency random index parameter. Specifically, the preset consistency random index parameter is the test value of the standard order of a 5th-order fixed matrix input to the system. A fixed constant of 1.12 is generated. The difference between the largest eigenvalue and the matrix order is divided algebraically to calculate the characteristic parameter and output the consistency ratio value. Then, based on the consistency ratio value, a conditional branch comparison operator is used to call a preset out-of-bounds interrupt beacon. The preset out-of-bounds interrupt beacon is specifically injected into the system kernel hardware register with a fixed floating-point threshold constant of 0.1 to set a trigger signal. The consistency ratio value is compared with the fixed floating-point threshold constant of 0.1 using arithmetic comparison logic to determine the comparison ratio. If the result is greater than the fixed floating-point threshold constant of 0.1, a memory addressing interception instruction is called. The center position reference coordinate vector is set to the origin 0. The absolute distance in Euclidean space is calculated to extract the abnormal coordinate points where the intercepted value deviates from the center position and to locate discrete conflict nodes.

[0024] The specific steps for constructing the potential weight matrix are as follows: Based on discrete conflict nodes, the Euclidean distance residual calculation method is adopted. The background reference coordinate vector parameter is set to the origin 0, and the absolute value floating-point operation precision parameter is set to double precision 64-bit. The residuals of the square and square root algebraic mapping operations are continuously performed on multiple coordinate variables within the discrete conflict nodes to obtain the node-related scalar deviation difference. Based on the scalar deviation difference, an extreme value threshold filtering algorithm is used to extract preset boundary parameters. Specifically, the preset boundary parameters are obtained by continuously calculating the standard deviation of 30 sets of known crustal contaminated rock sample parameters input into the system to obtain a judgment cutoff threshold constant of 6.5. The scalar deviation difference is then compared with the judgment cutoff threshold. The constant 6.5 input comparison register performs high-bit truncation filtering to extract boundary parameters and calls the borehole nickel isotope deviation to derive the crustal contamination peak parameters. Based on the crustal contamination peak parameters, a multidimensional tensor splicing instruction is used, setting the row dimension span parameter to 10 and the column dimension step parameter to 10. The peak parameters are aggregated and arranged into a multidimensional row and column grid according to the main order traversal rule. Based on the multidimensional row and column grid, a normalization operator is used, setting the natural constant base extension precision parameter to 6 decimal places. Nonlinear exponential summation and quotient transformation are performed on the internal parameters of the grid to convert and combine the internal parameters of the grid to construct a probability space mapping array and obtain the transition distribution matrix. Based on the transition distribution matrix, the Bellman expectation iterative equation is adopted, the Markov time step parameter is set to 100 steps, and the state transition discount constant factor parameter is set to 0.9. The state space nodes of the built-in element attributes of the transition distribution matrix are defined. Combined with the state transition discount constant factor parameter of 0.9, multi-level transition state scalars are continuously operated on by performing multi-order time step algebraic mapping deduction matrix element operations. Specifically, the multi-level transition state scalars are operated based on the Markov decision process state transition matrix diagram, substituting background S0, weak anomaly S1, contamination peak S2, discrete conflict S3, and stable target region S4 multiple state node features. The self-transition distribution probability parameter 0.65 of the stable target region S4 is extracted. Combined with the state transition matrix operation, the stable convergent state scalar V100 corresponding to 100 time steps is obtained, specifically including 0.18, 0.36, and 0.63. The element values ​​are 0.79 and 0.91. Based on the multi-level transition state scalar, a floating-point absolute difference comparison operator is used to extract the preset convergence boundary conditions. Specifically, the preset convergence boundary conditions are to set the absolute difference limit constant of the two-order states to 0.000001 to generate a floating-point tolerance benchmark. The multi-level transition state scalar is continuously compared with the preset convergence boundary conditions to filter out non-boundary stationary parameters. Based on the stationary parameters, a memory pointer overwrite replacement instruction is used to extract the memory hexadecimal base address corresponding to the original conflict coordinates. The pointer offset parameter is called to overwrite the stationary parameter memory address data to the original conflict coordinates and replace the corresponding values ​​of the abnormal coordinates. Based on the numerical replacement results, an orthogonal basis reconstruction operator is used, the tensor dimension expansion cardinality is set to 3, the multi-dimensional determinant is reorganized, and a reconstructed feature matrix is ​​established. Based on the reconstructed feature matrix, the Hadamard matrix multiplication operator is used, and the number of parallel computing threads is set to 16. The reconstructed feature matrix and the positional scalar of the same-dimensional adjoint array are extracted and multiplied point-to-point to obtain the deviation scalar. Based on the deviation scalar, the Jacobi eigenvalue diagonalization algorithm is used, and the upper limit of the number of iterations for eliminating off-diagonal elements by planar rotation is set to 500. A preset convergence threshold boundary is extracted. Specifically, the preset convergence threshold boundary is determined by performing 100 iterations of blank comparison matrix evolution to obtain the absolute minimum non-zero constant 0.0001 as the calculation limit. The deviation scalar is continuously compared with the convergence threshold boundary, and elementary transformations are performed to eliminate variables and extract the corresponding matrix eigenvalue parameters that meet the numerical limits. Based on the matrix eigenvalue parameters, the diagonalization spatial topology arrangement instruction is used, and the non-main diagonal filling constant is set to 0. The eigenvalues ​​are extracted and filled to the coordinate positions of the main diagonal of the multi-dimensional square matrix to arrange the weight coefficient matrix output distribution parameters, thereby obtaining the potential weight matrix.

[0025] The specific steps for establishing the target area's three-dimensional coordinates are as follows: Based on discrete conflict nodes and a potential weight matrix, a three-dimensional Kriging space interpolation algorithm is employed. The variogram model is set to a spherical model, the nugget effect constant is set to 0.1, the sill variance is set to 1.5, the maximum search radius is set to 500 meters, the spatial anisotropy principal axis deflection angle is set to 45 degrees, and the mesh generation step size is set to 10 meters. Discrete conflict nodes are input into a Euclidean mesh generator to continuously perform spatial weight allocation calculations, mapping the spatial covariance matrix between discrete points and the points to be estimated, borehole three-dimensional mesh calculations, and scalar calculations of borehole nickel isotope distribution. Based on these scalars, a conjugate tensor dot product operator is used, with the tensor order constant set to 3 and the memory alignment bytes set to 6. The 4-byte algorithm performs element-wise multiplication and summation algebraic operations on the potential weight matrix and scalars to obtain distribution parameters. Based on the distribution parameters, a binary logic judgment instruction is used to extract the preset critical boundary. Specifically, the preset critical boundary is generated by continuously calculating the 95th percentile value of 50 sets of test data of ore-free surrounding rock samples and generating a fixed limit constant of 0.85. The distribution parameters and the fixed limit constant of 0.85 are input into the kernel comparison register to test the high and low level limits. The distribution parameters and the critical boundary are compared to confirm the status. The memory truncation addressing instruction is called to set the truncation offset constant to 4 bytes to cut off the memory block below the limit constant and extract the spatial structure of the out-of-limit grid node arrangement to establish the out-of-limit spatial node set. Based on the out-of-bounds spatial node set, a conditional mask filtering operator is used. The Boolean truth retention constant is set to 1, the Boolean false value removal constant is set to 0, and the concurrent thread group size parameter is set to 32. The out-of-bounds spatial node set and the mask matrix are bitwise ANDed to clear all-zero vector blocks, identify the grid state, and remove non-out-of-bounds regular parameters. A floating-point coordinate inverse solution instruction is called to set the coordinate system projection reference origin vector to 0. Hexadecimal memory data is parsed to extract the remaining parameters corresponding to the two-dimensional plane coordinates. The borehole element geochemical content values ​​and borehole nickel isotope content values ​​are used to arrange the Z-axis scalar, thereby reconstructing the multi-dimensional optimization 3D point position. Based on the multi-dimensional optimization 3D point position, a three-dimensional Delaunay triangulation algorithm is used, setting the minimum circumscribed sphere radius parameter to 10 meters, the boundary line segment length constraint constant to 50 meters, and the tetrahedral volume skewness tolerance constant to 0.2. The empty sphere method is continuously executed. The test excludes points containing the remaining points, reassembles the circumscribed sphere, generates a tetrahedral mesh, arranges the points, connects the topology, and reconstructs the multidimensional closed bounding box. Based on the multidimensional closed bounding box, a fast 3D convex hull algorithm is used, setting the coplanar tolerance parameter to 0.001, the upper limit parameter for extreme point search iteration to 1000 times, and the step size parameter for calculating the distance of the external point set to 5 meters. It continuously searches for the spatial vertex farthest from the initial simplex, performs boundary polygon expansion loops, eliminates internal redundant coplanar nodes, aggregates bounding box parameters, and delineates the deep spatial envelope range. Specifically, the deep spatial envelope range is delineated based on the potential weight matrix and the three-dimensional coordinate envelope map of the deep target area. The X-axis spatial coordinate boundary is extracted to define the 0 to 1000-meter range, the Y-axis spatial coordinate boundary is defined to define the 0 to 1000-meter range, and the depth Z-axis scalar is extracted to define the 0 to 1000-meter sinking range. The three-dimensional envelope closed interval is arranged, and the three-dimensional coordinates of the target area are established.

[0026] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence, characterized in that, Includes the following steps: Step 1: Based on the geochemical content of borehole elements and the dual diluent array, calculate the deviation of nickel isotopes in the borehole, calculate the partial derivative matrix of fractionation coefficient combination, calculate the Hessian condition number and compare the extreme value limits, aggregate the difference scalar arrangement spatial array, and obtain the fractionation residual array. Step 2: Based on the fractionation residual array, superimpose and calculate the silica crystallization perturbation residual, compare the residual terms to determine the deviation state, introduce the simulated annealing algorithm to extract the distribution probability, select the replacement branch to continuously calculate the gradient descent, aggregate the optimization sequence to arrange the tensor elements, and obtain the melt separation extreme value tensor. Step 3: Based on the aforementioned melting extremum tensor, calculate the overall alteration degree of the borehole rock and the mass ratio of cobalt and nickel elements, arrange scalars to construct a sulfide feature matrix and calculate the consistency ratio, compare the ratio terms with the out-of-bounds interruption beacon, extract the abnormal discrete scalars of the matrix, and locate the discrete conflict nodes. Step 4: Based on the discrete conflict nodes, extract the crustal contamination extreme values ​​by calculating the residuals, arrange the extreme values ​​to construct the transition distribution matrix, deduce the state scalar through the Markov decision process, map and replace to reconstruct the feature matrix, iteratively compare the threshold to extract the feature roots, and construct the potential weight matrix. Step 5: Based on the discrete conflict nodes and the potential weight matrix, map the grid to calculate the distribution of nickel isotopes in the borehole, compare the distribution scalar with the threshold boundary to confirm the status, extract the three-dimensional spatial coordinates, introduce the borehole elements and the Z-axis of the nickel isotope arrangement to construct a multi-dimensional three-dimensional bounding box, and establish the three-dimensional coordinates of the target area.

2. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 1, characterized in that, The fractionation residual array includes isotope deviation, elemental partial derivative distribution matrix, and multidimensional three-dimensional structure array. The melting separation extreme value tensor includes silicate crystallization perturbation residual, exponential distribution probability, and grid structure feature array. The discrete conflict nodes include whole-rock comprehensive alteration degree, cobalt-nickel element mass ratio, and anomalous coordinate points. The potential weight matrix includes crustal contamination extreme values, multi-level transition state scalars, and weight coefficient arrays. The target area three-dimensional coordinates include a set of cross-boundary spatial nodes, borehole element geochemical content and borehole nickel isotope Z-axis scalar, multidimensional three-dimensional border, and deep spatial envelope range.

3. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 1, characterized in that, The specific steps for obtaining the fractionation residual array are as follows: Based on the geochemical content of borehole elements and the dual diluent array, the continuous spectral bands were extracted by calibrating the peak overlap state, the discrete mass-charge ratio sequence of borehole nickel isotopes was analyzed and the deviation of borehole nickel isotopes was calculated, the row and column partial derivatives of the multivariate constant product arrangement matrix were calculated, and the element partial derivative distribution matrix was established. Based on the element partial derivative distribution matrix, the diagonal scalar of the matrix is ​​intercepted at the wandering diagonal position, the Hessian condition number is determined by comparing the numerical boundary, the deviation difference is calculated to update the grid translation coordinate point, the translation difference is combined to arrange a multi-dimensional three-dimensional structure array, and the fractionation residual array is obtained.

4. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 1, characterized in that, The specific steps to obtain the melting-separation extremum tensor are as follows: Based on the fractionation residual array, the step size is allocated to call the geochemical content of borehole elements to derive the silica crystallization constant, the constant is superimposed to calculate the disturbance of the residual scalar and the scalar is compared to determine the boundary, the spatial position of the abnormal node is extracted, the position item is combined to arrange the deviation set, and the residual deviation state scalar is obtained. Based on the residual deviation state scalar, the simulated annealing algorithm is introduced to map and extract the exponential distribution probability. The probability terms are compared to verify the numerical boundary and the out-of-bounds parameters are truncated to trigger coordinate replacement. The partial derivative of the replacement terms is substituted to calculate the scalar descent. The scalars are accumulated and arranged into an evolution queue to establish a gradient descent optimization sequence. Based on the gradient descent optimization sequence, the spatial convergence term at the end of the sequence is extracted, the multidimensional optimization coordinate points are reconstructed, the coordinates are substituted to fill the missing nodes of the tensor, the combined filled nodes are reshaped to reshape the multi-dimensionality, and the grid structure feature array is aggregated to obtain the fused extreme value tensor.

5. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 1, characterized in that, The simulated annealing algorithm extracts the residual deviation state scalar and sets the initial system temperature parameters. It then substitutes the initial system temperature parameters into the cooling decay function to deduce the current stage annealing temperature scalar corresponding to the initial system temperature parameters. The residual deviation state scalar and the current stage annealing temperature scalar are used to perform a quotient operation to extract the negative exponential term parameter. The natural constant base is combined with the negative exponential term parameter to perform a nonlinear mapping algebra operation to extract the exponential distribution probability.

6. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 1, characterized in that, The specific steps for locating the discrete conflict nodes are as follows: Based on the aforementioned melting extremum tensor, the tensor data is segmented and the borehole element geochemical content is called to calculate the whole-rock comprehensive alteration scalar, the built-in cobalt-nickel element mass ratio sequence of borehole element geochemical content is analyzed, the combined parameters are arranged in a multi-dimensional row and column space structure, the determinant is solved to remove the orthogonal basis vector, and the sulfide feature matrix is ​​established. Based on the sulfide feature matrix, the matrix product term is calculated to separate the associated feature vectors, the feature parameter output consistency ratio is measured, the out-of-bounds interruption beacon comparison ratio term is called, the abnormal coordinate points where the value deviates from the center position are extracted, and the discrete conflict nodes are located.

7. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 1, characterized in that, The specific steps for constructing the potential weight matrix are as follows: Based on the discrete conflict nodes, the residual is calculated to obtain the node-related scalar deviation difference. The boundary parameters are extracted and the borehole nickel isotope deviation is called to derive the crustal contamination peak parameters. The peak parameters are aggregated and arranged into a multi-dimensional row and column grid. The internal parameters of the grid are combined to construct a probability space mapping array and obtain the transition distribution matrix. Based on the transition distribution matrix, the matrix element operation of the Markov decision process is used to deduce the multi-level transition state scalar, the state scalar is compared with the convergence boundary conditions, the non-boundary stationary parameters are selected and mapped to the original conflict coordinates, the values ​​corresponding to the abnormal coordinates are replaced and the multidimensional determinant is reorganized to establish the reconstructed feature matrix. Based on the reconstructed feature matrix, the deviation scalar is obtained by multiplying the elements within the matrix. The deviation scalar is continuously compared with the convergence threshold boundary. The eigenvalue parameters of the matrix corresponding to the numerical constraints are extracted. The distribution parameters of the eigenvalue arrangement weight coefficient matrix are extracted to obtain the potential weight matrix.

8. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 1, characterized in that, The Markov decision process extracts the transition distribution matrix, sets the matrix elements built into the transition distribution matrix as discrete state node sets, defines the transition decision action items corresponding to the discrete state node sets, extracts the probability mapping parameters associated with the transition decision action items by combining the transition distribution matrix, constructs the state value inference equation, substitutes the discrete state node sets, transition decision action items and probability mapping parameters into the state value inference equation and continuously performs multi-order time step algebraic mapping operations to obtain the state value parameters under each order of time step, combines the state value parameters, and calculates multi-level transition state scalars.

9. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 7, characterized in that, The process of replacing the abnormal coordinates and reconstructing the multidimensional determinant involves first extracting the original abnormal coordinates and their associated multidimensional tensor index boundaries. Then, a direct addressing instruction using memory pointers is called to locate the underlying memory block storing the abnormal values. Selected, non-boundary stable parameters are overwritten into the memory block using a pointer overwrite mechanism to perform numerical replacement. This process aggregates the replacement operation's multiple parameter scalars. Based on a preset Euclidean orthogonal basis sequence, a tensor dimensionality-upgrading and reshaping mapping is performed to construct a full-rank multidimensional square matrix structure. Gaussian elimination is then used to perform elementary row operations and matrix algebra on the full-rank multidimensional square matrix structure, generating an upper triangular matrix array. The scalars of multiple elements on the main diagonal of the upper triangular matrix array are extracted and continuously subjected to product algebra operations to output the reconstructed multidimensional determinant scalar.

10. The method for prospecting deep copper-nickel sulfide deposits based on multi-source geochemical evidence according to claim 1, characterized in that, The specific steps for establishing the three-dimensional coordinates of the target area are as follows: Based on the discrete conflict nodes and the potential weight matrix, the scalar of the borehole nickel isotope distribution is calculated by mapping the three-dimensional mesh of the borehole, the distribution parameters are obtained by multiplying the matrices, the distribution parameters are compared with the critical boundary to confirm the state, the spatial structure of the out-of-limit mesh node arrangement is extracted, and the out-of-limit spatial node set is established. Based on the set of out-of-bounds spatial nodes, the grid state is identified and the non-out-of-bounds conventional parameters are removed. The two-dimensional plane coordinates corresponding to the remaining parameters are extracted. The geochemical content values ​​of borehole elements and the nickel isotope content values ​​of borehole elements are used to arrange the Z-axis scalar, thereby reconstructing the multi-dimensional three-dimensional border. The border parameters are aggregated to define the deep space envelope range and establish the three-dimensional coordinates of the target area.

Images