Intelligent prospecting method and device based on data control theory

Abstract

This application provides an intelligent mineral exploration method and apparatus based on data control theory. The method includes: collecting multimodal raw mineral exploration data, constructing a variable relationship graph with each exploration variable as a node and the statistical relationship between variables as an edge, and using dissipative structure theory as guidance to measure the centrality of the variable relationship graph through a graph neural network to obtain a subset of key variables in the raw mineral exploration data; based on the Haken enslavement principle, performing feature dimensionality reduction on the subset of key variables to construct a feature parameter space, and predicting the feature states in the feature parameter space through ordinary differential equations. After the state is verified by a particle filtering algorithm, the corresponding feature parameters are output; based on the cusp catastrophe theory, the feature parameters are input as control parameters into a preset physical information neural network, and the state is judged through a state discriminant to determine the mineral exploration decision signal. This application can improve the accuracy and efficiency of mineralization decision-making.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to an intelligent mineral exploration method and apparatus based on data control theory. Background Technology

[0002] Against the backdrop of continuously growing global demand for mineral resources and increasing difficulty in deep mineral exploration, traditional mineral exploration methods face fundamental challenges. Traditional mineral exploration techniques primarily rely on linear statistical models and qualitative empirical judgments, and their core limitations are reflected in the following aspects:

[0003] Information attenuation problem: Deep mineralization systems are highly complex, and traditional geophysical and geochemical exploration methods are prone to information loss during data transmission and analysis, resulting in the obscuring of key mineralization signals.

[0004] The dilemma of multiple interpretations: The heterogeneous nature of geological data from multiple sources means that a single indicator can be interpreted in multiple ways. For example, the same geophysical anomaly may correspond to multiple mineralization types, which significantly reduces the reliability of predictions.

[0005] Limitations of static understanding: It simplifies the mineralization process into a linearly decomposable system, ignoring its dynamic evolution characteristics. Actual mineralization involves the nonlinear coupling of multiple factors such as hydrothermal migration, tectonic activity, and material precipitation, making it difficult for traditional methods to capture the critical state of the system.

[0006] These limitations significantly reduce the prediction success rate of traditional mineral exploration in deep "black box" systems, making it unable to effectively guide the precise location of "deep, blind, and hidden" ore bodies. Therefore, there is an urgent need for an intelligent mineral exploration method based on data control theory that can improve the accuracy and efficiency of mineralization decisions. Summary of the Invention

[0007] To address the problems in the existing technology, this application provides an intelligent mineral exploration method and apparatus based on data control theory, which can improve the accuracy and efficiency of mineralization decision-making.

[0008] To solve at least one of the above problems, this application provides the following technical solution:

[0009] Firstly, this application provides an intelligent mineral exploration method based on data control theory, comprising:

[0010] Multimodal raw data from mineral exploration is collected and standardized. A variable relationship graph is constructed using each standardized exploration variable as a node and the statistical correlation between exploration variables as an edge. The variable relationship graph is input into a preset graph neural network. Through node embedding learning and centrality measurement, the corresponding key variable subset is determined. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification.

[0011] The subset of key variables is input into a pre-trained variational autoencoder for feature dimensionality reduction to determine the corresponding low-dimensional feature parameter space. The low-dimensional feature parameter space is then used as the initial state to input a preset neural ordinary differential equation for evolution law prediction to determine the corresponding feature parameter state. Finally, the posterior distribution of the feature parameter state is dynamically updated using a particle filter algorithm combined with new observation data to determine the corresponding feature parameters.

[0012] A physical information neural network is constructed based on the cusp catastrophe theory. The feature parameters are input into the physical information neural network as control parameters. A discriminant is calculated based on the preset control parameters to determine the state of the control parameters. The system state is divided according to the value of the discriminant after the state determination, and the corresponding mineral exploration decision signal is determined based on the system state.

[0013] Furthermore, the process of collecting multimodal raw data from mineral exploration and standardizing the raw data includes:

[0014] Collect geophysical exploration data, including at least one of gravity, magnetic, electromagnetic or seismic data, and record its corresponding spatial coordinates, acquisition time and exploration depth attributes;

[0015] Collect geochemical analysis data, including at least elemental content data in rocks, soils, or stream sediments, and record the sampling location, sampling stratigraphic level, and analysis time attribute;

[0016] Collect geological exploration data, including at least the spatial distribution information of stratigraphy, tectonic lines, alteration zones or mineral outcrops, and record their observation points, geological age and descriptive accuracy attributes;

[0017] The above data is aligned using a unified geographic coordinate system and time reference. The aligned data is then standardized in terms of dimensions to determine the corresponding exploration variables. The data alignment includes spatial interpolation and time synchronization. The dimension standardization includes one of Z-score standardization, maximum and minimum value normalization, or mean and variance normalization.

[0018] Further, the step of inputting the variable relationship graph into a preset graph neural network, and determining the corresponding key variable subset through node embedding learning and centrality measurement, includes:

[0019] The variable relationship graph is input into a preset graph neural network, which is a graph attention network or a graph convolutional network.

[0020] The graph neural network learns the low-dimensional embedding representation of each exploration variable node in the variable relationship graph through a multi-layer message passing mechanism and an attention weighting mechanism, and determines the centrality index of the node based on the embedding representation. The centrality index includes at least one of eigenvector centrality, betweenness centrality, and proximity centrality.

[0021] Further, the step of using the low-dimensional feature parameter space as the initial state to input a preset neural ordinary differential equation for evolution law prediction, and determining the corresponding feature parameter state, includes:

[0022] The low-dimensional feature parameter space is used as the initial state input to a preset neural network model. The network model simulates the dynamic evolution of the feature parameters in the continuous time domain through a parameterized function, wherein the mathematical form of the parameterized function is defined by an ordinary differential equation.

[0023] The ordinary differential equation is solved by numerical integration to determine the trajectory of the state change of the characteristic parameter at its future spatial location.

[0024] Furthermore, the step of dynamically updating the posterior distribution of the feature parameter state using a particle filtering algorithm in conjunction with new observation data to determine the corresponding feature parameters includes:

[0025] The characteristic parameter state change trajectory predicted by the ordinary differential equation is used as the state transition model to determine the corresponding set of state particles;

[0026] The newly acquired multi-source exploration observation data is used as input to the measurement model to calculate the likelihood probability of each particle in the state particle set with the current observation data.

[0027] The state particle set is resampled according to the likelihood probability, the weight distribution of the particles is updated, and the posterior probability distribution of the corresponding feature parameter at the current time is determined.

[0028] Based on the updated posterior probability distribution, the statistical estimate of the corresponding feature parameter is determined.

[0029] Furthermore, the step of constructing a physical information neural network based on cusp catastrophe theory, and inputting the feature parameters as control parameters into the physical information neural network, includes:

[0030] A neural network is constructed using feature parameters as network input and state prediction as network output. The cusp mutation equilibrium equation is embedded into the loss function of the neural network to determine the corresponding physical information neural network.

[0031] The feature parameters are converted into control parameter pairs through a learnable mapping layer and input into the physical information neural network, wherein the control parameter pairs include a splitting factor and a regularization factor.

[0032] Further, the step of calculating a discriminant based on preset control parameters to determine the state of the control parameters, and classifying the system state based on the value of the discriminant after state determination, includes:

[0033] The state of the control parameters is determined by calculating a discriminant based on the preset control parameters.

[0034] If the discriminant value after state judgment is greater than zero, the system is determined to be in a monostable low-ore potential zone.

[0035] If the discriminant value after state judgment is equal to zero, then the system is determined to be in the critical state abrupt boundary region.

[0036] If the discriminant value after state judgment is less than zero, the system is determined to be in a bistable rich ore potential zone.

[0037] Secondly, this application provides an intelligent mineral exploration device based on data control theory, comprising:

[0038] A key subset determination module based on dissipative structure theory is used to collect multimodal raw data in mineral exploration and standardize the raw data. Using each exploration variable after standardization as a node and the statistical correlation between exploration variables as edges, a variable relationship graph is constructed. The variable relationship graph is input into a preset graph neural network, and the corresponding key variable subset is determined through node embedding learning and centrality measurement. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification.

[0039] The feature parameter determination module based on the Haken enslavement principle is used to input the subset of key variables into a pre-trained variational autoencoder for feature dimensionality reduction, determine the corresponding low-dimensional feature parameter space, input the low-dimensional feature parameter space as the initial state into a preset neural ordinary differential equation for evolution law prediction, determine the corresponding feature parameter state, and dynamically update the posterior distribution of the feature parameter state through a particle filtering algorithm combined with new observation data to determine the corresponding feature parameters.

[0040] The mining area discrimination module based on cusp catastrophe theory is used to construct a physical information neural network based on cusp catastrophe theory, input the feature parameters as control parameters into the physical information neural network, calculate the discriminant expression based on the preset control parameters to judge the state of the control parameters, classify the system state based on the value of the discriminant expression after the state judgment, and determine the corresponding mineral exploration decision signal based on the system state.

[0041] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the intelligent mineral exploration method based on data control theory.

[0042] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the intelligent mineral exploration method based on data control theory.

[0043] Fifthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the intelligent mineral exploration method based on data control theory.

[0044] As can be seen from the above technical solution, this application provides an intelligent mineral exploration method and device based on data control theory. By collecting multimodal raw mineral exploration data, a variable relationship graph is constructed using each exploration variable as a node and the statistical relationship between variables as edges. Guided by dissipative structure theory, a graph neural network is used to measure the centrality of the variable relationship graph, obtaining a subset of key variables in the raw mineral exploration data. Based on the Haken enslavement principle, the key variable subset is subjected to feature dimensionality reduction to construct a feature parameter space. The feature states in the feature parameter space are predicted using ordinary differential equations. These states are verified by a particle filtering algorithm, and the corresponding feature parameters are output. Guided by cusp catastrophe theory, the feature parameters are input as control parameters into a preset physical information neural network. A state discrimination equation is used to determine the state and identify the mineral exploration decision signal, thereby improving the accuracy and efficiency of mineralization decisions. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is one of the flowcharts illustrating the intelligent mineral exploration method based on data control theory in the embodiments of this application;

[0047] Figure 2 This is a structural diagram of the intelligent mineral exploration device based on data control theory in the embodiments of this application;

[0048] Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0049] Figure label:

[0050] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0053] Considering the problems of information attenuation, multiple solutions, and limitations of static cognition in existing mineral exploration, this application provides an intelligent mineral exploration method and device based on data control theory. By collecting multimodal raw mineral exploration data, a variable relationship graph is constructed using each exploration variable as a node and the statistical relationship between variables as edges. Guided by dissipative structure theory, a graph neural network is used to measure the centrality of the variable relationship graph, obtaining a subset of key variables from the raw mineral exploration data. Based on the Haken enslavement principle, the key variable subset is subjected to feature dimensionality reduction to construct a feature parameter space. The feature states in the feature parameter space are predicted using ordinary differential equations, and these states are verified by a particle filter algorithm to output the corresponding feature parameters. Guided by cusp catastrophe theory, the feature parameters are input as control parameters into a preset physical information neural network. A state discriminant is used to determine the state and identify the mineral exploration decision signal, thereby improving the accuracy and efficiency of mineralization decisions.

[0054] To improve the accuracy and efficiency of mineralization decision-making, this application provides an embodiment of an intelligent mineral exploration method based on data control theory, see [link to relevant documentation]. Figure 1 The intelligent mineral exploration method based on data control theory specifically includes the following:

[0055] Step S101: Collect multimodal raw data in mineral exploration and standardize the raw data. Using each exploration variable after standardization as a node and the statistical correlation between exploration variables as an edge, construct a variable relationship graph. Input the variable relationship graph into a preset graph neural network. Through node embedding learning and centrality measurement, determine the corresponding key variable subset. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification.

[0056] Optionally, in this embodiment, this step is the key variable screening stage, which is guided by dissipative structure theory.

[0057] Specifically, dissipative structure theory involves selecting variables and defining the system.

[0058] The core problem of dissipative structure theory is how to identify the truly critical few from thousands of exploration variables. The underlying logic is that mineralization is the process by which an open system forms a "dissipative structure" (ore body) far from equilibrium. Variables are divided into "fast variables" (which change rapidly and have no decisive effect on the macroscopic order) and "slow variables" (which change slowly and dominate the direction of system evolution).

[0059] The mathematical embodiment of dissipative structure theory, taking the Brusselator model as an example:

[0060]

[0061]

[0062] in:

[0063] X and Y: are the state variables of the system (concentration of intermediate reactants).

[0064] A and B: These are the system's control parameters (initial concentrations of reactants, which are usually considered constants).

[0065] t: is time

[0066] : These represent the rates of change of variables X and Y over time, respectively.

[0067] This system of equations describes the condition over a specific parameter range (e.g., B>1+A). 2 Within this range, the system can move far from equilibrium, through the nonlinear term X. 2 The interaction of Y spontaneously generates stable spatiotemporal ordered structures (such as chemical oscillations or Turing patterns), which is a typical mathematical manifestation of dissipative structures.

[0068] This autocatalytic reaction model demonstrates how a system can generate an ordered structure through nonlinear interactions 2X + Y -> 3X. This suggests that we should focus on identifying variables involved in nonlinear interactions that can drive system instability and rebuild a new order.

[0069] Guided by the aforementioned dissipative structure theory, this embodiment integrates graph neural networks to screen key slow variables.

[0070] Specifically, multimodal raw data is collected from mineral exploration activities. Originating from various exploration methods such as geophysical exploration, geochemical analysis, and geological surveys, this data possesses different physical meanings, dimensions, spatiotemporal resolutions, and forms of expression, constituting typical heterogeneous data. To eliminate dimensional differences and scale effects between different data sources, the raw data needs to be standardized. Methods such as normalization, standardization, or deviation standardization are typically used to transform all variables into a uniform numerical range. This ensures the comparability of variables in subsequent modeling and avoids model bias towards variables with larger numerical values ​​due to differences in magnitude.

[0071] After standardization, each exploration variable is treated as a node in the graph structure. Based on this, the correlation between any two variables is calculated using statistical methods and represented as edges in the graph. The weights of the edges are quantified using statistical measures such as mutual information, nonlinear correlation coefficients, or causal strength. This helps to reveal the complex and nonlinear interaction networks within metallogenic systems.

[0072] Next, the constructed variable relationship graph is input into a pre-defined graph neural network. During this process, node embedding learning maps each exploration variable to a low-dimensional dense vector. This vector not only preserves the characteristics of the variable itself but also encodes its relationships with other variables. Based on this, nodes with important topological positions in the graph are identified using centrality measures (such as eigenvector centrality, betweenness centrality, and proximity centrality). Eigenvector centrality is particularly important, as it reflects the recursive relationship between the importance of a node and its connected nodes, effectively identifying variables that are connected to multiple important variables and occupy core positions in the network.

[0073] Through node embedding learning and centrality analysis using graph neural networks, a subset of key variables was ultimately identified. This subset of key variables consists of "slow variables" with high centrality indices that, in a physical sense, correspond to the dominant forces in the evolution of the mineralization system, such as tectonic activity intensity, fluid migration rate, and element enrichment efficiency. These variables occupy pivotal positions in the network and can significantly influence the state of other variables and the overall dynamic behavior of the system. Therefore, they are considered key control factors driving the system from disorder to order, ultimately forming ore bodies.

[0074] Step S102: Input the subset of key variables into a pre-trained variational autoencoder for feature dimensionality reduction to determine the corresponding low-dimensional feature parameter space. Use the low-dimensional feature parameter space as the initial state to input a preset neural ordinary differential equation for evolution law prediction to determine the corresponding feature parameter state. Then, use a particle filter algorithm to dynamically update the posterior distribution of the feature parameter state in combination with new observation data to determine the corresponding feature parameters.

[0075] Optionally, in this embodiment, this step is the feature parameter extraction stage, which is guided by the Haken enslavement principle.

[0076] Specifically, the core issue of Haken's enslavement principle is how key slow variables interact and which dominates. The underlying logic is that as a system approaches a critical point, a very small number of characteristic parameters emerge from among numerous slow variables. These characteristic parameters are generated by the cooperation of subsystems and, in turn, govern the behavior of all other subsystems (the enslavement principle). For example, "tectonic-fluid coupling strength" and "intrinsic element concentration efficiency" may become the dominant characteristic parameters of the entire mineralization system. This step condenses dozens of key variables into 3-7 characteristic parameters with clear geological significance.

[0077] The mathematical core of Haken's principle of enslavement, the linearized evolution equation of the system near the critical point, is:

[0078]

[0079] It is the state variable q i The rate of change of time t (derivative);

[0080] It is the feature value (or linear growth rate / decrease rate) of the i-th mode;

[0081] It is the i-th state variable (or modal amplitude);

[0082] : Represents the nonlinear term acting on the i-th variable, which is a function of all state variables q = (q1,q2,…,qn);

[0083]

[0084] Based on the eigenvalue λ i :

[0085] Fast variable (u): Re(λ) u If the value is less than or equal to 0, rapid decay occurs.

[0086] Slow variable / feature parameter (v): Re(λ) v )≈0, dominating long-term behavior;

[0087] Applying the adiabatic approximation du / dt ≈ 0, we obtain the slave equation:

[0088] u(t) = f(v(t))

[0089] This indicates that the fast variable is instantaneously dominated by the slow variable, and the system dynamics are simplified to the evolution equation of the characteristic parameters.

[0090] Guided by the aforementioned Haken enslavement principle, this embodiment integrates synergetics to extract feature parameters based on this theory.

[0091] Specifically, the selected subset of key variables is first input into a pre-trained variational autoencoder. The encoder part of the variational autoencoder maps high-dimensional, potentially complexly correlated key variables to a low-dimensional, continuous feature parameter space. Each point in this space (i.e., a low-dimensional vector) corresponds to a compressed representation of the original high-dimensional data, capturing the essential characteristics driving the macroscopically ordered state of the mineralization system. In this way, dozens or even hundreds of geological, geophysical, and geochemical variables are condensed into a few (usually 3-7) feature parameters with clear physical meaning, realizing the transformation of data from "numerous and complex" to "few and precise," laying a foundation of dimensionality-reduced, information-dense data for subsequent analysis.

[0092] Specifically, after obtaining the low-dimensional feature parameter space, it is used as the initial state of the system and input into a pre-defined neural ordinary differential equation network. The neural ordinary differential equation network is a model that combines a neural network with an ordinary differential equation solver. It does not directly learn discrete input-output mappings, but rather learns the continuous-time derivatives of the hidden states, i.e., the "dynamic laws" of the system.

[0093] Specifically, the network parameterizes a function f_φ(z) to describe the evolution rate of the feature parameter z over time t:

[0094] dz / dt = f_φ(z)

[0095] By numerically integrating this differential equation, the model can predict the state of characteristic parameters at any future time, starting from any initial state. This is essentially a mathematical model of the evolution trajectory of the ore-forming system, making dynamic deduction of the system state possible and breaking through the limitations of traditional static statistical analysis models.

[0096] The introduction of constant differential equations allows models to learn and simulate the dynamic behavior of mineralization systems in a continuous-time manner. This allows geologists not only to understand "what the system is in now" but also to predict "how it will evolve," providing a powerful quantitative tool for understanding the temporal dimension of mineralization processes and predicting their future development.

[0097] The first two steps construct a system evolution model learned from historical data. However, during actual exploration, new observational data (such as new drilling results and geophysical data) will continuously be obtained. To integrate this new information into the system's understanding in real time, this step employs a particle filtering algorithm (based on Bayesian filtering techniques from sequential Monte Carlo simulations), and the technical process is as follows:

[0098] First, the next-time characteristic parameter state predicted by the constant differential equation is taken as the "prior probability distribution". Then, a particle filtering algorithm generates a large number of random samples (called "particles") to approximate this prior distribution. Each particle represents a possible state hypothesis of the system. Next, the algorithm compares the newly acquired actual observation data with the predicted observation value of each particle and calculates a weight value. The higher the weight, the better the fit between the particle and the actual observation data. Finally, the particle set is resampled according to these weights. High-weight particles are retained and replicated, while low-weight particles are eliminated. The resampled particle set forms the updated "posterior probability distribution", which integrates model prediction (prior) and new evidence (observation) and is the optimal probabilistic estimate of the current state of the system. Through this process, the system cognition achieves dynamic calibration and continuous updating. The model can "digest" new data, correct the predicted trajectory, and enable the overall mineral exploration system to have a "closed-loop feedback" capability of continuous learning and self-optimization during the exploration process.

[0099] Step S103: Construct a physical information neural network based on the cusp catastrophe theory, input the feature parameters as control parameters into the physical information neural network, calculate the discriminant according to the preset control parameters to judge the state of the control parameters, divide the system state according to the value of the discriminant after the state judgment, and determine the corresponding mineral exploration decision signal according to the system state.

[0100] Optionally, in this embodiment, this step is the mineralization state judgment stage, which is guided by the cusp catastrophe theory.

[0101] Specifically, the core issue of cusp catastrophe theory is how to determine whether a system has "large mineral deposits" or "small mineral deposits" based on characteristic parameters. The underlying logic is that changes in characteristic parameters drive discontinuous jumps in the system's state. The cusp catastrophe model perfectly simulates the qualitative change process from "no mineral deposits to large mineral deposits." Geologically, the "split factor" and "regular factor" correspond to two characteristic parameters, and their combination determines whether the system is in a low-mineral state, a high-mineral state, or a critical state.

[0102] The mathematical framework of cusp catastrophe theory states that the system state is described by a potential function:

[0103]

[0104] in:

[0105] V(x): is the potential function of the system.

[0106] x: is the state variable of the system (e.g., mineralization intensity).

[0107] a: is the splitting factor control parameter (usually corresponding to order parameter 1, such as construction control).

[0108] b: is the regularity factor control parameter (usually corresponding to order parameter 2, such as material supply).

[0109] This potential function is the core model in catastrophe theory for studying bimodal, jump, and hysteresis phenomena. The equilibrium state of the system is determined by the equation dV / dx=0, that is:

[0110] The system equilibrium state satisfies dV / dx = 0:

[0111] x³ + ax + b = 0

[0112] The discriminant Δ = 4a³ + 27b² determines the system behavior: Δ = 4a³ + 27b²

[0113] Δ>0: Monostable (no mineral deposits)

[0114] Δ<0: Bistable (cusp region), state x may undergo abrupt bifurcation (the larger the negative value - large mine).

[0115] Δ = 0 is the boundary of the state transition. (Small mine - No mine)

[0116] Based on the cusp catastrophe theory, this stage establishes a dynamic-based discriminant model to quantitatively assess the potential and criticality of the target area to reach a large mineral deposit.

[0117] Specifically, the system constructs a physical information neural network based on a mathematical model of cusp catastrophe theory.

[0118] The network's input layer is explicitly set to two nodes, corresponding to two control parameters in the cusp catastrophe theory model: the "split factor a" and the "regularity factor b". These two parameters are derived from the feature parameter space obtained by dimensionality reduction in the previous stage through a defined mathematical mapping relationship. For example, the mapping relationship can be a simple linear transformation or a more complex nonlinear fitting, with the aim of transforming the abstract feature parameters z1, z2, ..., etc., into a and b with explicit physical meaning in catastrophe theory.

[0119] Among them, a is related to the degree of “split” or “asymmetry” of the system (such as the difference in structural stress or the heterogeneity of fluid channels), while b is related to the strength of the system’s “regularity” or “driving force” (such as the mass supply flux or energy input rate).

[0120] The output of the network's output layer is the system's "state variable x," which in this embodiment can be a comprehensive representation of "mineralization intensity," "orderliness of the ore-forming system," or "resource enrichment." The weights and bias parameters of the neural network are randomly initialized, but the key to the entire model lies in the special constraints of its training process.

[0121] The unique feature of the physical information neural network constructed in this step lies in the design of its loss function. The loss function consists of two parts, forming a balance between "data fitting" and "observation of physical laws".

[0122] The first part is the data fitting term. If there are known exploration areas (training samples) with "mineralization states" x_obs estimated or observed through other means (e.g., calibration values ​​based on existing borehole grades), the state x_pred predicted by the network needs to be as close as possible to these observations. This part is measured using mean squared error.

[0123] The second part is the physical constraint term. According to cusp catastrophe theory, the equilibrium state of the system (i.e., the condition for the occurrence of a stable mineralized state) is determined by the derivative of the potential function being zero. Embedding the cusp catastrophe equilibrium equation into the loss function is a typical example of constructing a loss function in a Physics-Informed Neural Network (PINN). Its standard mathematical form is as follows:

[0124]

[0125] Or, more concisely:

[0126]

[0127] Through this loss function design, the network training process is not merely about learning the complex mapping relationship from input (a, b) to output (x), but rather being guided to discover solutions that simultaneously satisfy the observed data and the fundamental physical laws of catastrophe theory. This ensures that even when the network makes predictions in sparse, unknown regions (which is the norm in deep mineral exploration), its inference results remain within a physically reasonable range, greatly enhancing the model's extrapolation ability and the physical interpretability of the results.

[0128] Within a pre-trained physical information neural network framework, for any exploration target area to be evaluated, we map its feature parameters to the corresponding (a, b) input network to obtain the predicted equilibrium state x. However, a single value of x is insufficient to fully determine the system's potential behavior.

[0129] At this point, the discriminant Δ from cusp catastrophe theory is introduced as the core criterion:

[0130] Δ = 4a³ + 27b²

[0131] The value of the discriminant Δ directly determines the topological shape of the system's potential energy surface, and thus the number of possible stable states of the system.

[0132] Based on the value of the discriminant Δ, the system clearly divides the exploration target area into three states:

[0133] When Δ>0: the system potential function has only one minimum point, corresponding to a monostable state. In the context of mineral exploration, this can be interpreted as the system being in an "inert" state, difficult to undergo qualitative change. Regardless of the initial perturbation, the system will eventually stabilize in a "poor ore" or "ore-free" state with low mineralization. This area is marked as a low-potential zone.

[0134] When Δ < 0: The system potential function has two minimum points (bistable) and one maximum point, and the system is in a mutation-sensitive region. These two steady states correspond to the "poor ore state" and the "rich ore state," respectively. The current state of the system depends on its historical path (hysteresis effect). Being in the target region of this area means that the combination of its control parameters (a, b) has met the necessary conditions for a qualitative change, and the system has the inherent potential to transition to the rich ore state. This region is identified as a high-concern area. If the predicted x value is located in the potential well corresponding to the rich ore state, it is a high-confidence rich ore target area; if the x value is located in the potential well of the poor ore state but close to the bifurcation boundary, it is a critical transition precursor region.

[0135] When Δ ≈ 0: The system is exactly on the bifurcation set, which is the critical boundary between monostable and bistable states. Systems in this state are extremely sensitive; even small parameter perturbations can cause a sudden jump in state. During exploration, this marks a critical zone, a key location where verification engineering needs to be deployed to "trigger" or confirm the system's state.

[0136] Based on the above state classification, the process ultimately generates a structured mineral exploration decision signal. This signal not only includes qualitative category labels (such as "low-potential area," "high-potential target area," and "critical area of ​​interest"), but also a series of quantitative indicators to support refined decision-making:

[0137] Potential level: The value is assigned based on the magnitude of the negative value of Δ and the location of the potential well where the predicted x value is located.

[0138] Criticality index: Calculate the geometric or probabilistic distance D from the (a, b) coordinates of the target area to the boundary of the theoretical bifurcation set. The smaller the D value, the closer the system is to the mutation threshold, and the more sensitive the possibility of a mineral breakthrough is to exploration activities.

[0139] Uncertainty measurement: By combining the posterior distribution information passed through processes such as particle filtering, confidence intervals for state judgment and potential assessment can be given.

[0140] This example demonstrates how this embodiment integrates dissipative structure theory, synergetics, and catastrophe theory to construct an intelligent mineral exploration model based on cybernetics and systems science, which is used for data-controlled intelligent mineral exploration decision-making, thereby improving the accuracy and efficiency of mineralization decisions.

[0141] As described above, the intelligent mineral exploration method based on data control theory provided in this application can collect multimodal raw mineral exploration data, construct a variable relationship graph with each exploration variable as a node and the statistical relationship between variables as an edge, and, guided by dissipative structure theory, use a graph neural network to measure the centrality of the variable relationship graph to obtain a subset of key variables in the raw mineral exploration data. Based on the Haken enslavement principle, the key variable subset is used to perform feature dimensionality reduction to construct a feature parameter space, and the feature states in the feature parameter space are predicted by ordinary differential equations. After the state is verified by a particle filtering algorithm, the corresponding feature parameters are output. Guided by cusp catastrophe theory, the feature parameters are used as control parameters and input into a preset physical information neural network. The state is judged by a state discriminant to determine the mineral exploration decision signal, thereby improving the accuracy and efficiency of mineralization decision-making.

[0142] In one embodiment of the intelligent mineral exploration method based on data control theory in this application, it may further include the following:

[0143] Step S201: Collect geophysical exploration data, including at least one of gravity, magnetic, electromagnetic or seismic data, and record its corresponding spatial coordinates, acquisition time and exploration depth attributes;

[0144] Step S202: Collect geochemical analysis data, including at least elemental content data in rocks, soils or stream sediments, and record the sampling location, sampling stratigraphic level and analysis time attribute;

[0145] Step S203: Collect geological exploration data, including at least the spatial distribution information of stratigraphy, tectonic lines, alteration zones or mineral outcrops, and record the observation points, geological age and descriptive accuracy attributes.

[0146] Step S204: Align the above data with a unified geographic coordinate system and time reference, and standardize the aligned data in terms of dimensions to determine the corresponding exploration variables. The data alignment includes spatial interpolation and time synchronization, and the dimension standardization includes one of Z-score standardization, maximum and minimum value normalization, or mean and variance normalization.

[0147] Optionally, in this embodiment, steps S201, S202, and S203 systematically collect raw data from the three major exploration fields of geophysics, geochemistry, and geological survey, respectively. Geophysical exploration data covers subsurface physical field information obtained by gravity, magnetic, electromagnetic, or seismic methods, and records key attributes such as spatial location, collection time, and exploration depth to characterize the spatial variation characteristics of the physical field within different depth ranges. Geochemical analysis data mainly obtains information on the content of various elements in rocks, soils, or stream sediments, while recording the location, stratigraphy, and analysis time of sampling points to reflect the composition and elemental distribution patterns of materials from the surface to shallow layers. Geological exploration data includes the spatial distribution characteristics of geological entities such as stratigraphic lithology, tectonic lines, alteration zones, or mineralized outcrops, and is accompanied by attributes such as observation point location, geological age, and descriptive precision to provide direct evidence of geological structure and mineralization background.

[0148] The aforementioned data differ significantly in source, scale, precision, and representation. Step S204 aims to systematically integrate and standardize them. First, all data are aligned using a unified geographic coordinate system and time benchmark. Spatial alignment requires methods such as Kriging interpolation and inverse distance weighting to interpolate data of different resolutions to a unified grid. Temporal alignment requires synchronization or time window division based on the collection time series to ensure data comparability in both spatiotemporal dimensions. Subsequently, methods such as Z-score standardization, maximum-minimum normalization, or mean-variance normalization are used to eliminate dimensional differences between variables and transform all data to the same numerical range.

[0149] Through step S204, this embodiment establishes a unified framework for multimodal data in terms of time, space and dimensions, providing a consistent data foundation for subsequent variable correlation analysis.

[0150] In one embodiment of the intelligent mineral exploration method based on data control theory in this application, it may further include the following:

[0151] Step S301: Input the variable relationship graph into a preset graph neural network, wherein the graph neural network is a graph attention network or a graph convolutional network;

[0152] Step S302: The graph neural network learns the low-dimensional embedding representation of each exploration variable node in the variable relationship graph through a multi-layer message passing mechanism and an attention weighting mechanism, and determines the centrality index of the node based on the embedding representation. The centrality index includes at least one of eigenvector centrality, betweenness centrality, and proximity centrality.

[0153] Optionally, in this embodiment, this step implements node representation learning and centrality analysis through network architecture.

[0154] Specifically, graph neural networks are selected from graph attention networks or graph convolutional networks. Graph convolutional networks aggregate neighborhood information of each node through convolution operations defined on the graph, progressively learning the local and global features of the nodes. Graph attention networks, on the other hand, introduce an attention mechanism on top of graph convolution, enabling the network to dynamically assign different aggregation weights to different neighbors based on their importance. Both structures can effectively handle non-Euclidean data such as exploration variable relationship graphs, capturing complex dependencies between variables.

[0155] Specifically, in each layer, each node updates its feature representation by aggregating the features of its neighbors (possibly with attention weighting). After multiple layers of stacking, the low-dimensional embedding representation finally learned by each node not only contains its own attribute information but also encodes the structural information of its multi-hop neighbors and even the entire graph. This representation can effectively characterize the topological role of a node in the graph and its association patterns with other variables. Based on the learned node embeddings, various centrality indices of the nodes can be further calculated. Eigenvector centrality reflects the degree of connectivity between a node and other important nodes in the network and is a key indicator for identifying network "hub" variables; betweenness centrality measures the ability of a node to act as a "bridge" to control the flow of information or materials; proximity centrality represents the reciprocal of the average distance from a node to all other nodes in the network, reflecting the propagation efficiency of its influence. By analyzing these centrality indices, the global status and structural influence of each exploration variable in the metallogenic system network can be systematically evaluated.

[0156] Through step S302, this embodiment goes beyond simple statistical correlation, capturing complex, high-order nonlinear dependencies between variables using a deep learning model. This allows for a more reliable identification of the key slow variables that truly dominate the mineralization dynamic system. This provides a high-quality subset of input variables, filtered through deep feature learning and structural importance, for subsequent systems science-based modeling and decision-making.

[0157] In one embodiment of the intelligent mineral exploration method based on data control theory in this application, it may further include the following:

[0158] Step S401: Input the low-dimensional feature parameter space as the initial state into the preset neural network model. The network model simulates the dynamic evolution of the feature parameters in the continuous time domain through parameterization functions, wherein the mathematical form of the parameterization functions is defined by ordinary differential equations.

[0159] Step S402: Solve the ordinary differential equation using numerical integration to determine the trajectory of the state change of the characteristic parameter at its future spatial location.

[0160] Optionally, in this embodiment, the feature parameter space refers to the condensation of key slow variables into feature parameters with clear geological significance.

[0161] Specifically, key slow variables can be:

[0162] Geophysical data (sensing physical fields): magnetic susceptibility intensity, gravity Bouguer gravity anomaly, seismic wave velocity, etc.

[0163] Geochemical data (sensing material fields): rock / soil elemental content, isotope ratios, fluid inclusions, etc.;

[0164] Geological and remote sensing data (sensing structure and lithology): tectonics, lithology, remote sensing, etc.

[0165] Characteristic parameters are extracted from the aforementioned key slow variables and are core parameters that govern the macroscopic ordered state and evolutionary direction of the entire mineralization system. For example:

[0166] The tectonic-fluid coupling intensity (a) integrates the coupling efficiency of tectonic driving forces (fracture opening, connectivity, and activity phases) and fluid energy (fluid pressure, temperature, and velocity). It answers the core question of whether fluid can be effectively pumped to favorable locations and remain there. It "enslaves" several key variables such as fracture density, resistivity anomalies (reflecting fluid dynamics), homogenization temperature, and wave velocity anomalies. A high value of this parameter indicates that an efficient mineralization "transport-trap" system is in operation.

[0167] The super-aggregation efficiency of ore-forming elements (b) describes the net efficiency of gold extraction, transport, and precipitation from massive amounts of host rock. It considers not only the gold content but also the strength and scale of the physicochemical gradients leading to gold precipitation (such as sudden temperature drops, phase separation, and water-rock reactions). It influences the spatial zoning of elements like Au, As, and Sb, pyrite content, silicification intensity, and δ³⁺. 4 S-value, etc. A high value indicates that the system is in a highly efficient "chemical precipitation factory" state.

[0168] These extracted feature parameters are used as the initial state of the system and input into a pre-defined neural network model. A parameterized function is used to simulate the dynamic evolution of the feature parameters in the continuous time domain.

[0169] Specifically, dynamic learning utilizes Neural ODE to learn the evolution laws of feature parameters:

[0170] dz / dt = f_φ(z)

[0171] After establishing the parameterized ordinary differential equation, the equation is solved using numerical integration. Numerical integration starts from the current initial state (state z0 at known time t0) and proceeds along the time axis with discrete but sufficiently fine steps to calculate the system state at various future time points. Commonly used methods include the Runge-Kutta method. Through this numerical integration, the trajectory of characteristic parameters' state changes over future time series can be determined. Importantly, this "future" is not limited to the time dimension. In mineral exploration applications, since exploration data is often associated with spatial location (e.g., different borehole locations, different depths), by parameterizing spatial coordinates or exploration sequences as time-like variables, this model can also predict the state changes of characteristic parameters in unexplored spatial locations, enabling continuous extrapolation of the system behavior in unknown areas.

[0172] Through step S402, the parameterized function successfully learned in this embodiment essentially reveals the intrinsic force field or rules driving the evolution of the system, and can predict the state of the feature parameters at unsampled spatial locations.

[0173] In one embodiment of the intelligent mineral exploration method based on data control theory in this application, it may further include the following:

[0174] Step S501: Use the characteristic parameter state change trajectory predicted by the ordinary differential equation as the state transition model to determine the corresponding set of state particles;

[0175] Step S502: Use the newly acquired multi-source exploration observation data as input to the measurement model, and calculate the likelihood probability of each particle in the state particle set with the current observation data;

[0176] Step S503: Resample the set of state particles according to the likelihood probability, update the weight distribution of the particles, and determine the posterior probability distribution of the corresponding feature parameters at the current time.

[0177] Step S504: Based on the updated posterior probability distribution, determine the statistical estimate of the corresponding feature parameter.

[0178] Optionally, this step uses the state change trajectory of the characteristic parameters predicted by the constant differential equation as the state transition model of the system. Specifically, based on the prior probability distribution of the current state predicted by this model, a large number (usually thousands to tens of thousands) of random samples are generated using the Monte Carlo sampling method. Each sample is called a "particle". These particles together constitute the state particle set, where each particle carries a specific set of characteristic parameter values, representing a hypothetical state that the system may be in. The overall distribution of the particle set reflects the uncertainty of the system state predicted based on historical data and the dynamic model.

[0179] Once new multi-source exploration observation data (such as geophysical anomalies, borehole core analysis results, etc.) is acquired, it is input into the measurement model. This model defines the theoretical relationship between the system state (i.e., the characteristic parameter values ​​of particles) and the actual observation values. For each particle in the particle set, the algorithm calculates its corresponding theoretical observation value and compares this theoretical value with the actual observation data. It then calculates the degree of matching between the particle and the current actual observation data, i.e., the likelihood probability, using a preset likelihood function (usually based on the measurement error distribution). The higher this probability value, the better the state assumption represented by the particle matches the actual observation results.

[0180] Based on the calculated likelihood probability of each particle (as weights), the original particle set is resampled. The core of the resampling process is random sampling with replacement according to the weight ratios: particles with higher weights have a greater chance of being sampled multiple times, while particles with lower weights may be eliminated. After resampling, the distribution of the newly generated particle set in the state space changes—particle density increases in regions that better match the observation data, and particle density decreases in regions that deviate from the observation data. This updated particle set and its distribution represent the posterior probability distribution of the system's feature parameters at the current time step, integrating model predictions (prior) and new observational evidence.

[0181] Finally, based on the updated posterior probability distribution particle set, the final estimates of the feature parameters are extracted through statistical computation. The most common method is to calculate the mean of the particle set as the optimal estimate of the state, while simultaneously calculating the variance or confidence interval to quantify the uncertainty of the estimate. This output not only provides the most probable value of the system's current state, but more importantly, it provides a quantitative measure of the reliability of this estimate.

[0182] Through step S504, this embodiment successfully constructed a complete Bayesian inference closed loop through the particle filtering process, enabling the system to integrate intermittently acquired new observational evidence into the continuously evolving dynamic model in real time and quantitatively, and continuously correct the state perception of the deep mineralization system.

[0183] In one embodiment of the intelligent mineral exploration method based on data control theory in this application, it may further include the following:

[0184] Step S601: Construct a neural network using feature parameters as network input and state prediction as network output, and embed the cusp mutation equilibrium equation into the loss function of the neural network to determine the corresponding physical information neural network;

[0185] Step S602: The feature parameters are converted into control parameter pairs through a learnable mapping layer and input into the physical information neural network, wherein the control parameter pairs include a splitting factor and a regularization factor.

[0186] Optionally, in this embodiment, this step is the construction phase of the Physical Information Neural Network (PINN).

[0187] Specifically, the basic architecture of the neural network is defined first. The "feature parameters" that describe the essence of the mineralization system are used as inputs, and the output of the network is set as the "state prediction" of the system, that is, a scalar or vector that comprehensively represents the mineralization intensity or enrichment degree of the target area.

[0188] The key difference lies in the fact that, in this embodiment, the network is not trained through conventional supervised learning, but rather using a "physical constraint learning" paradigm. Specifically, the core equilibrium equation of cusp catastrophe theory (x³ + ax + b = 0) is directly incorporated as a strong constraint into the neural network's loss function. This means that during training, the network must not only minimize the error between the predicted state and (if) finite observation states (the data fitting term), but also simultaneously minimize the degree to which its predictions violate the aforementioned physical equation (the physical constraint term). In this way, the network is forced to learn input-output mappings that both conform to sparse observation data and strictly adhere to the stability laws described by catastrophe theory. This ensures that even when predicting in unknown regions with extremely limited exploration data, the model's output remains within a physically reasonable solution space, thereby greatly improving the model's extrapolation reliability and the physical interpretability of the prediction results, overcoming the fundamental deficiency of pure black-box models in predicting low reliability in areas without data.

[0189] Next, the refinement and necessary additions to S601 resolved the interface problem between the characteristic parameters and the control parameters of the catastrophe theory.

[0190] A separate "learnable mapping layer" is added before the neural network. This layer (which can be one or more fully connected layers) receives the original feature parameter vector as input and outputs a pair of specific values, namely "split factor a" and "regularization factor b". The weight parameters of the mapping layer are optimized and learned together with other parts during the overall training of the model.

[0191] The geological significance lies in acknowledging the potential for complex, nonlinear transformations between characteristic parameters abstracted from actual observational data and control parameters in theoretical models. By learning this mapping through a data-driven approach, the model can adaptively find the (a, b) parameter combination that best explains the current observational data while conforming to the catastrophe theory framework.

[0192] Through step S602, this embodiment successfully completed state prediction and physical equation constraint calculation.

[0193] In one embodiment of the intelligent mineral exploration method based on data control theory in this application, it may further include the following:

[0194] Step S701: Calculate the discriminant based on the preset control parameters and determine the state of the control parameters;

[0195] Step S702: If the discriminant value after state judgment is greater than zero, then the system is determined to be in a monostable low-ore potential zone;

[0196] Step S703: If the discriminant value after state judgment is equal to zero, then the system is determined to be in the critical state abrupt boundary region;

[0197] Step S704: If the discriminant value after state judgment is less than zero, the system is determined to be in a bistable rich ore potential zone.

[0198] Optionally, in this embodiment, the discriminant Δ from cusp catastrophe theory is introduced as the core criterion:

[0199] Δ = 4a³ + 27b²

[0200] The value of the discriminant Δ directly determines the topological shape of the system's potential energy surface, and thus the number of possible stable states of the system.

[0201] Based on the value of the discriminant Δ, the system clearly divides the exploration target area into three states:

[0202] When Δ>0: the system potential function has only one minimum point, corresponding to a monostable state. In the context of mineral exploration, this can be interpreted as the system being in an "inert" state, difficult to undergo qualitative change. Regardless of the initial perturbation, the system will eventually stabilize in a "poor ore" or "ore-free" state with low mineralization. This area is marked as a low-potential zone.

[0203] When Δ < 0: The system potential function has two minimum points (bistable) and one maximum point, and the system is in a mutation-sensitive region. These two steady states correspond to the "poor ore state" and the "rich ore state," respectively. The current state of the system depends on its historical path (hysteresis effect). Being in the target region of this area means that the combination of its control parameters (a, b) has met the necessary conditions for a qualitative change, and the system has the inherent potential to transition to the rich ore state. This region is identified as a high-concern area. If the predicted x value is located in the potential well corresponding to the rich ore state, it is a high-confidence rich ore target area; if the x value is located in the potential well of the poor ore state but close to the bifurcation boundary, it is a critical transition precursor region.

[0204] When Δ ≈ 0: The system is exactly on the bifurcation set, which is the critical boundary between monostable and bistable states. Systems in this state are extremely sensitive; even small parameter perturbations can cause a sudden jump in state. During exploration, this marks a critical zone, a key location where verification engineering needs to be deployed to "trigger" or confirm the system's state.

[0205] Based on the above state division, the process ultimately generates a structured mineral exploration decision signal.

[0206] Through step S704, this embodiment successfully determines the state of the mining area based on the state judgment discriminant, which is used to generate accurate mineralization decisions.

[0207] To improve the accuracy and efficiency of mineralization decision-making, this application provides an embodiment of an intelligent mineral exploration device based on data control theory for implementing all or part of the aforementioned intelligent mineral exploration method based on data control theory. See [link to embodiment]. Figure 2 The intelligent mineral exploration device based on data control theory specifically includes the following components:

[0208] The key subset determination module 10 based on dissipative structure theory is used to collect multimodal raw data in mineral exploration and standardize the raw data. The standardized exploration variables are used as nodes and the statistical correlation between exploration variables are used as edges to construct a variable relationship graph. The variable relationship graph is input into a preset graph neural network. Through node embedding learning and centrality measurement, the corresponding key variable subset is determined. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification.

[0209] The feature parameter determination module 20 based on the Haken enslavement principle is used to input the subset of key variables into a pre-trained variational autoencoder for feature dimensionality reduction, determine the corresponding low-dimensional feature parameter space, input the low-dimensional feature parameter space as the initial state into a preset neural ordinary differential equation for evolution law prediction, determine the corresponding feature parameter state, and dynamically update the posterior distribution of the feature parameter state through a particle filtering algorithm combined with new observation data to determine the corresponding feature parameters.

[0210] The mining area discrimination module 30 based on cusp catastrophe theory is used to construct a physical information neural network based on cusp catastrophe theory, input the feature parameters as control parameters into the physical information neural network, calculate the discriminant formula according to the preset control parameters to judge the state of the control parameters, divide the system state according to the value of the discriminant formula after the state judgment, and determine the corresponding mineral exploration decision signal according to the system state.

[0211] As described above, the intelligent mineral exploration device based on data control theory provided in this application can collect multimodal raw mineral exploration data, construct a variable relationship graph with each exploration variable as a node and the statistical relationship between variables as an edge, and, guided by dissipative structure theory, perform a centrality measurement on the variable relationship graph through a graph neural network to obtain a subset of key variables in the raw mineral exploration data. Based on the Haken enslavement principle, the key variable subset is used to perform feature dimensionality reduction to construct a feature parameter space, and the feature states in the feature parameter space are predicted through ordinary differential equations. After the state is verified by a particle filtering algorithm, the corresponding feature parameters are output. Guided by cusp catastrophe theory, the feature parameters are input as control parameters into a preset physical information neural network, and the state is judged through a state discriminant to determine the mineral exploration decision signal, thereby improving the accuracy and efficiency of mineralization decision-making.

[0212] From a hardware perspective, in order to improve the accuracy and efficiency of mineralization decision-making, this application provides an embodiment of an electronic device for implementing all or part of the intelligent mineral exploration method based on data control theory. The electronic device specifically includes the following components:

[0213] The system comprises a processor, memory, a communication interface, and a bus; wherein the processor, memory, and communication interface communicate with each other via the bus; the communication interface is used to realize information transmission between the intelligent mineral exploration method based on data control theory and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the intelligent mineral exploration method based on data control theory in the present embodiment, and the contents of the embodiments of the intelligent mineral exploration method based on data control theory are incorporated herein, and repeated details will not be described again.

[0214] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0215] In practical applications, parts of the intelligent mineral exploration method based on data control theory can be executed on the electronic device side as described above, or all operations can be completed in the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0216] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0217] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0218] In one embodiment, the intelligent mineral exploration method based on data control theory can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0219] Step S101: Collect multimodal raw data in mineral exploration and standardize the raw data. Using each exploration variable after standardization as a node and the statistical correlation between exploration variables as an edge, construct a variable relationship graph. Input the variable relationship graph into a preset graph neural network. Through node embedding learning and centrality measurement, determine the corresponding key variable subset. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification.

[0220] Step S102: Input the subset of key variables into a pre-trained variational autoencoder for feature dimensionality reduction to determine the corresponding low-dimensional feature parameter space. Use the low-dimensional feature parameter space as the initial state to input a preset neural ordinary differential equation for evolution law prediction to determine the corresponding feature parameter state. Then, use a particle filter algorithm to dynamically update the posterior distribution of the feature parameter state in combination with new observation data to determine the corresponding feature parameters.

[0221] Step S103: Construct a physical information neural network based on the cusp catastrophe theory, input the feature parameters as control parameters into the physical information neural network, calculate the discriminant according to the preset control parameters to judge the state of the control parameters, divide the system state according to the value of the discriminant after the state judgment, and determine the corresponding mineral exploration decision signal according to the system state.

[0222] As described above, the electronic device provided in this application collects multimodal raw mineral exploration data, constructs a variable relationship graph with each exploration variable as a node and the statistical relationship between variables as an edge, and uses dissipative structure theory as guidance to measure the centrality of the variable relationship graph through a graph neural network to obtain a subset of key variables in the raw mineral exploration data; based on the Haken enslavement principle, the key variable subset is used to perform feature dimensionality reduction to construct a feature parameter space, and the feature states in the feature parameter space are predicted through ordinary differential equations. After the state is verified by a particle filtering algorithm, the corresponding feature parameters are output; based on the cusp catastrophe theory, the feature parameters are used as control parameters to input into a preset physical information neural network, and the state is judged through a state discriminant to determine the mineral exploration decision signal, thereby improving the accuracy and efficiency of mineralization decision.

[0223] In another implementation, the intelligent mineral exploration method based on data control theory can be configured separately from the central processing unit 9100. For example, the intelligent mineral exploration method based on data control theory can be configured as a chip connected to the central processing unit 9100, and the functions of the intelligent mineral exploration method based on data control theory can be realized through the control of the central processing unit.

[0224] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technology.

[0225] like Figure 3 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0226] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0227] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0228] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0229] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device for communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0230] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0231] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Furthermore, the audio processor 9130 is also coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored sound via the speaker 9131.

[0232] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the intelligent mineral exploration method based on data control theory, where the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the intelligent mineral exploration method based on data control theory, where the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0233] Step S101: Collect multimodal raw data in mineral exploration and standardize the raw data. Using each exploration variable after standardization as a node and the statistical correlation between exploration variables as an edge, construct a variable relationship graph. Input the variable relationship graph into a preset graph neural network. Through node embedding learning and centrality measurement, determine the corresponding key variable subset. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification.

[0234] Step S102: Input the subset of key variables into a pre-trained variational autoencoder for feature dimensionality reduction to determine the corresponding low-dimensional feature parameter space. Use the low-dimensional feature parameter space as the initial state to input a preset neural ordinary differential equation for evolution law prediction to determine the corresponding feature parameter state. Then, use a particle filter algorithm to dynamically update the posterior distribution of the feature parameter state in combination with new observation data to determine the corresponding feature parameters.

[0235] Step S103: Construct a physical information neural network based on the cusp catastrophe theory, input the feature parameters as control parameters into the physical information neural network, calculate the discriminant according to the preset control parameters to judge the state of the control parameters, divide the system state according to the value of the discriminant after the state judgment, and determine the corresponding mineral exploration decision signal according to the system state.

[0236] As described above, the computer-readable storage medium provided in this application collects multimodal raw mineral exploration data, constructs a variable relationship graph with each exploration variable as a node and the statistical relationship between variables as an edge, and uses dissipative structure theory as guidance to measure the centrality of the variable relationship graph through a graph neural network to obtain a subset of key variables in the raw mineral exploration data; based on the Haken enslavement principle, the subset of key variables is dimensionality reduced to construct a feature parameter space, and the feature states in the feature parameter space are predicted by ordinary differential equations. After the state is verified by a particle filter algorithm, the corresponding feature parameters are output; based on the cusp catastrophe theory, the feature parameters are input as control parameters into a preset physical information neural network, and the state is judged by a state discriminant to determine the mineral exploration decision signal, thereby improving the accuracy and efficiency of mineralization decision-making.

[0237] Embodiments of this application also provide a computer program product capable of implementing all steps of the intelligent mineral exploration method based on data control theory, where the execution subject is a server or client, as described in the above embodiments. When executed by a processor, this computer program / instruction implements the steps of the intelligent mineral exploration method based on data control theory. For example, the computer program / instruction implements the following steps:

[0238] Step S101: Collect multimodal raw data in mineral exploration and standardize the raw data. Using each exploration variable after standardization as a node and the statistical correlation between exploration variables as an edge, construct a variable relationship graph. Input the variable relationship graph into a preset graph neural network. Through node embedding learning and centrality measurement, determine the corresponding key variable subset. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification.

[0239] Step S102: Input the subset of key variables into a pre-trained variational autoencoder for feature dimensionality reduction to determine the corresponding low-dimensional feature parameter space. Use the low-dimensional feature parameter space as the initial state to input a preset neural ordinary differential equation for evolution law prediction to determine the corresponding feature parameter state. Then, use a particle filter algorithm to dynamically update the posterior distribution of the feature parameter state in combination with new observation data to determine the corresponding feature parameters.

[0240] Step S103: Construct a physical information neural network based on the cusp catastrophe theory, input the feature parameters as control parameters into the physical information neural network, calculate the discriminant according to the preset control parameters to judge the state of the control parameters, divide the system state according to the value of the discriminant after the state judgment, and determine the corresponding mineral exploration decision signal according to the system state.

[0241] As described above, the computer program product provided in this application collects multimodal raw mineral exploration data, constructs a variable relationship graph with each exploration variable as a node and the statistical relationship between variables as an edge, and uses dissipative structure theory as guidance to measure the centrality of the variable relationship graph through a graph neural network to obtain a subset of key variables in the raw mineral exploration data; based on the Haken enslavement principle, the subset of key variables is dimensionality reduced to construct a feature parameter space, and the feature states in the feature parameter space are predicted through ordinary differential equations. After the state is verified by a particle filter algorithm, the corresponding feature parameters are output; based on the cusp catastrophe theory, the feature parameters are input as control parameters into a preset physical information neural network, and the state is judged through a state discriminant to determine the mineral exploration decision signal, thereby improving the accuracy and efficiency of mineralization decision.

[0242] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0243] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0244] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0245] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0246] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. An intelligent mineral exploration method based on data control theory, characterized in that, The method includes: Multimodal raw data from mineral exploration is collected and standardized. A variable relationship graph is constructed using each standardized exploration variable as a node and the statistical correlation between exploration variables as an edge. The variable relationship graph is input into a preset graph neural network. Through node embedding learning and centrality measurement, the corresponding key variable subset is determined. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification. The subset of key variables is input into a pre-trained variational autoencoder for feature dimensionality reduction to determine the corresponding low-dimensional feature parameter space. The low-dimensional feature parameter space is then used as the initial state to input a preset neural ordinary differential equation for evolution law prediction to determine the corresponding feature parameter state. Finally, the posterior distribution of the feature parameter state is dynamically updated using a particle filter algorithm combined with new observation data to determine the corresponding feature parameters. A physical information neural network is constructed based on the cusp catastrophe theory. The feature parameters are input into the physical information neural network as control parameters. A discriminant is calculated based on the preset control parameters to determine the state of the control parameters. The system state is divided according to the value of the discriminant after the state determination, and the corresponding mineral exploration decision signal is determined based on the system state.

2. The intelligent mineral exploration method based on data control theory according to claim 1, characterized in that, The process of collecting multimodal raw data from mineral exploration and standardizing the raw data includes: Collect geophysical exploration data, including at least one of gravity, magnetic, electromagnetic or seismic data, and record its corresponding spatial coordinates, acquisition time and exploration depth attributes; Collect geochemical analysis data, including at least elemental content data in rocks, soils, or stream sediments, and record the sampling location, sampling stratigraphic level, and analysis time attribute; Collect geological exploration data, including at least the spatial distribution information of stratigraphy, tectonic lines, alteration zones or mineral outcrops, and record their observation points, geological age and descriptive accuracy attributes; The above data is aligned using a unified geographic coordinate system and time reference. The aligned data is then standardized in terms of dimensions to determine the corresponding exploration variables. The data alignment includes spatial interpolation and time synchronization. The dimension standardization includes one of Z-score standardization, maximum and minimum value normalization, or mean and variance normalization.

3. The intelligent mineral exploration method based on data control theory according to claim 1, characterized in that, The step of inputting the variable relationship graph into a preset graph neural network, and determining the corresponding key variable subset through node embedding learning and centrality measurement, includes: The variable relationship graph is input into a preset graph neural network, which is a graph attention network or a graph convolutional network. The graph neural network learns the low-dimensional embedding representation of each exploration variable node in the variable relationship graph through a multi-layer message passing mechanism and an attention weighting mechanism, and determines the centrality index of the node based on the embedding representation. The centrality index includes at least one of eigenvector centrality, betweenness centrality, and proximity centrality.

4. The intelligent mineral exploration method based on data control theory according to claim 1, characterized in that, The step of using the low-dimensional feature parameter space as the initial state to input a preset neural ordinary differential equation for evolution law prediction and determining the corresponding feature parameter state includes: The low-dimensional feature parameter space is used as the initial state input to a preset neural network model. The network model simulates the dynamic evolution of the feature parameters in the continuous time domain through a parameterized function, wherein the mathematical form of the parameterized function is defined by an ordinary differential equation. The ordinary differential equation is solved by numerical integration to determine the trajectory of the state change of the characteristic parameter at its future spatial location.

5. The intelligent mineral exploration method based on data control theory according to claim 4, characterized in that, The step of dynamically updating the posterior distribution of the feature parameter states using a particle filtering algorithm, combined with new observation data, to determine the corresponding feature parameters includes: The characteristic parameter state change trajectory predicted by the ordinary differential equation is used as the state transition model to determine the corresponding set of state particles; The newly acquired multi-source exploration observation data is used as input to the measurement model to calculate the likelihood probability of each particle in the state particle set with the current observation data. The state particle set is resampled according to the likelihood probability, the weight distribution of the particles is updated, and the posterior probability distribution of the corresponding feature parameter at the current time is determined. Based on the updated posterior probability distribution, the statistical estimate of the corresponding feature parameter is determined.

6. The intelligent mineral exploration method based on data control theory according to claim 1, characterized in that, The step of constructing a physical information neural network based on cusp catastrophe theory, and inputting the feature parameters as control parameters into the physical information neural network, includes: A neural network is constructed using feature parameters as network input and state prediction as network output. The cusp mutation equilibrium equation is embedded into the loss function of the neural network to determine the corresponding physical information neural network. The feature parameters are converted into control parameter pairs through a learnable mapping layer and input into the physical information neural network, wherein the control parameter pairs include a splitting factor and a regularization factor.

7. The intelligent mineral exploration method based on data control theory according to claim 1, characterized in that, The step of calculating a discriminant based on preset control parameters to determine the state of the control parameters, and classifying the system state based on the value of the discriminant after state determination, includes: The state of the control parameters is determined by calculating a discriminant based on the preset control parameters. If the discriminant value after state judgment is greater than zero, the system is determined to be in a monostable low-ore potential zone. If the discriminant value after state judgment is equal to zero, then the system is determined to be in the critical state abrupt boundary region. If the discriminant value after state judgment is less than zero, the system is determined to be in a bistable rich ore potential zone.

8. An intelligent mineral exploration device based on data control theory, characterized in that, The device includes: A key subset determination module based on dissipative structure theory is used to collect multimodal raw data in mineral exploration and standardize the raw data. Using each exploration variable after standardization as a node and the statistical correlation between exploration variables as edges, a variable relationship graph is constructed. The variable relationship graph is input into a preset graph neural network, and the corresponding key variable subset is determined through node embedding learning and centrality measurement. The weight of the edge is determined by mutual information, nonlinear correlation coefficient or causal relationship strength quantification. The feature parameter determination module based on the Haken enslavement principle is used to input the subset of key variables into a pre-trained variational autoencoder for feature dimensionality reduction, determine the corresponding low-dimensional feature parameter space, input the low-dimensional feature parameter space as the initial state into a preset neural ordinary differential equation for evolution law prediction, determine the corresponding feature parameter state, and dynamically update the posterior distribution of the feature parameter state through a particle filtering algorithm combined with new observation data to determine the corresponding feature parameters. The mining area discrimination module based on cusp catastrophe theory is used to construct a physical information neural network based on cusp catastrophe theory, input the feature parameters as control parameters into the physical information neural network, calculate the discriminant expression based on the preset control parameters to judge the state of the control parameters, classify the system state based on the value of the discriminant expression after the state judgment, and determine the corresponding mineral exploration decision signal based on the system state.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the intelligent mineral exploration method based on data control theory as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the intelligent mineral exploration method based on data control theory as described in any one of claims 1 to 7.

Images