Virtual drilling data processing method and device

Abstract

The embodiment of the application provides a kind of virtual drilling data processing method and device, method includes: by obtaining the range of target area and historical exploration data, the target area is discretized into regular grid model and the historical exploration data is associated with the regular grid model, by geostatistics method, the geological attribute information of each grid in regular grid model is spatially interpolated to obtain initial belief graph, according to initial belief graph, Markov decision process model is constructed, based on initial belief graph and Markov decision process model, virtual drilling is carried out by Monte Carlo tree search algorithm, according to the virtual observation result obtained by virtual drilling, initial belief graph is updated and immediate reward is determined, according to the expected cumulative reward of immediate reward and updated belief graph, according to the one or more candidate drill holes of the highest expected cumulative reward, determine drilling scheme, the application can improve the efficiency and accuracy of drilling.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a virtual drilling data processing method and apparatus. Background Technology

[0002] In the field of mineral resource exploration, traditional methods often rely on the experience of geologists and limited prior information, especially in scenarios involving concealed ore bodies in overburdened areas and deep prospecting (depths exceeding 100 meters). These methods face multiple challenges, including insufficient information, high costs, and low efficiency. Specific pain points include:

[0003] Lack of prior information: In covered areas, ore bodies are often covered by thick layers of sediment or vegetation, lacking direct outcrop information, which leads to unclear exploration targets.

[0004] High costs of deep mineral exploration: The cost of deep drilling increases sharply with increasing depth, and the risk of drilling failure also increases.

[0005] Green exploration needs: With increasing environmental awareness, reducing ineffective drilling and achieving less drilling and more precise drilling have become important directions for green exploration.

[0006] Low efficiency in target area reduction: From general survey to detailed survey, target area reduction often relies on a large amount of borehole data, resulting in high time and economic costs.

[0007] Traditional exploration methods mainly rely on geological mapping, geophysical exploration, and limited borehole verification. While these methods can guide mineral exploration to some extent, the information provided by geological mapping and geophysical exploration is often indirect when dealing with complex geological conditions and deep exploration. It is difficult to directly reflect the spatial distribution and grade variation of ore bodies. Furthermore, traditional methods often require a large number of borehole verifications, resulting in long exploration cycles and high costs.

[0008] Therefore, there is an urgent need for a virtual drilling data processing method that transforms resource exploration problems into a virtual data processing process, which can improve the efficiency and accuracy of real drilling. Summary of the Invention

[0009] To address the problems in the prior art, this application provides a virtual drilling data processing method and apparatus, which can improve the efficiency and accuracy of drilling.

[0010] To solve at least one of the above problems, this application provides the following technical solution:

[0011] Firstly, this application provides a virtual drilling data processing method, including:

[0012] Historical exploration data of the target area is obtained, the target area is discretized into a regular grid model, and the historical exploration data is associated with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resource. The borehole data includes borehole location and borehole results.

[0013] Geostatistical methods are used to spatially interpolate the geological attribute information of each grid in the regular grid model based on the historical exploration data to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0014] Using the initial belief graph as the state space, a corresponding Markov decision process model is determined. The initial belief graph and the Markov decision process model are input into a Monte Carlo tree search algorithm to perform virtual drilling at each candidate borehole location, determining the corresponding virtual observation results. Based on the virtual observation results, the initial belief graph is updated using a Bayesian algorithm to determine the corresponding updated belief graph. Based on the updated belief graph and the Markov decision process model, the expected cumulative reward of the corresponding candidate borehole is determined, and one or more candidate boreholes with the highest expected cumulative reward are output. Based on the output candidate boreholes, the corresponding drilling layout scheme is determined.

[0015] Further, the step of using the initial belief graph as a state space to determine the corresponding Markov decision process model includes:

[0016] Construct the state space of the Markov decision process model, wherein the state space is defined as the initial belief graph of the regular grid model, and each state of the state space corresponds to the probability distribution of the existence of the target resource in the grid of the initial belief graph;

[0017] Construct the action space of the Markov decision process model, wherein the action space is defined as a preset set of candidate borehole locations;

[0018] Construct the observation space for the Markov decision process model, whereby the observation space is defined as a set of binary observation results, including mineralization and non-mineralization;

[0019] A state transition sub-model of a Markov decision process model is constructed. The state transition probability of the state transition sub-model is set as an identity matrix to represent that the geological state itself remains unchanged under drilling operations.

[0020] An observation sub-model is constructed for a Markov decision process model, wherein the probability observed by the observation sub-model is defined as the prior probability or attribute strength of the current belief graph.

[0021] Construct a reward function for a Markov decision process model, wherein the reward function is used to quantify virtual observation results;

[0022] Construct a discount factor for a Markov decision process model, wherein the discount factor is a non-zero number.

[0023] Further, before inputting the initial belief map and the Markov decision process model into the Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location and determine the corresponding virtual observation results, the following steps are included:

[0024] The root node of the preset search tree is initialized based on the belief graph, and the corresponding Monte Carlo tree search algorithm is determined. The Monte Carlo tree search algorithm uses UCT scoring and constructs development items and exploration items, which are used to balance exploration and development.

[0025] Each node in the Monte Carlo tree search algorithm is used to count the total number of times the node is visited and the cumulative reward obtained from starting from that node.

[0026] Furthermore, the step of inputting the initial belief map and the Markov decision process model into a Monte Carlo tree search algorithm to perform virtual drilling at each candidate borehole location and determine the corresponding virtual observation results includes:

[0027] The initial belief map and the Markov decision process model are input into the Monte Carlo tree search algorithm. The Monte Carlo tree search algorithm selects a candidate borehole location from the action space of the Markov decision process model as the current simulation action, and performs simulation sampling based on the initial belief map to determine the geological realization of the resource distribution in the corresponding target area.

[0028] In the geological realization of the resource distribution in the target area, forward modeling is performed on the borehole location corresponding to the current simulation action to determine the theoretical observation value of the borehole location;

[0029] The theoretical observation values ​​are superimposed with a preset observation noise model to determine the corresponding virtual observation results. The observation noise model is used to simulate the measurement uncertainty in the actual drilling process.

[0030] Further, the step of updating the initial belief map using a Bayesian algorithm based on the virtual observation results to determine the corresponding updated belief map includes:

[0031] The initial belief map is used as a prior probability distribution;

[0032] The virtual observation results are quantified into virtual radiation intensity. Based on the virtual radiation intensity, the corresponding observation likelihood function is determined. The virtual radiation intensity is calculated using a physical-statistical coupling model based on the grade, mineralization thickness, and burial depth in the virtual observation results.

[0033] The prior probability distribution is multiplied by the observation likelihood function according to Bayes' theorem and then normalized to determine the corresponding updated belief map, which represents the posterior probability distribution.

[0034] Further, determining the expected cumulative reward of the corresponding candidate borehole based on the updated belief graph and the Markov decision process model includes:

[0035] The updated belief graph is used as the new state input of the Markov decision process model. Virtual drilling and belief updates are performed recursively, and the subsequent rewards are discounted according to the discount factor to determine the corresponding subsequent rewards.

[0036] The immediate reward of virtual drilling is added to the subsequent reward to determine the expected cumulative reward of the corresponding candidate borehole under the current simulation path. The immediate reward of virtual drilling is obtained from the reward function based on the virtual observation results. The reward function quantifies the virtual observation results into a virtual radiation intensity value that is proportional to the target resource intensity.

[0037] Furthermore, the step of spatially interpolating the geological attribute information of each grid in the regular grid model based on the historical exploration data using geostatistical methods to determine the corresponding initial belief map includes:

[0038] Based on the historical exploration data, the geological attribute dependence of each known grid in the regular grid model is calculated using geostatistical methods to determine the corresponding dependence value. The geostatistical methods include the variation function method and the covariance function method.

[0039] The geological attribute information of unknown grids in the regular grid model is weighted and calculated based on the dependency relationship value to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0040] Secondly, this application provides a virtual drilling data processing device, comprising:

[0041] The target area grid determination module is used to acquire historical exploration data of the target area, discretize the target area into a regular grid model, and associate the historical exploration data with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resources. The borehole data includes borehole location and borehole results.

[0042] The initial belief map determination module is used to perform spatial interpolation of the geological attribute information of each grid in the regular grid model based on the historical exploration data using geostatistical methods to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0043] The virtual drilling module is used to take the initial belief map as the state space, determine the corresponding Markov decision process model, input the initial belief map and the Markov decision process model into a Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location, determine the corresponding virtual observation results, update the initial belief map using a Bayesian algorithm based on the virtual observation results, determine the corresponding updated belief map, determine the expected cumulative reward of the corresponding candidate borehole based on the updated belief map and the Markov decision process model, output one or more candidate boreholes with the highest expected cumulative reward, and determine the corresponding drilling layout scheme based on the output candidate boreholes.

[0044] 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 virtual drilling data processing method.

[0045] 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 virtual drilling data processing method described above.

[0046] Fifthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the virtual drilling data processing method described above.

[0047] As can be seen from the above technical solution, this application provides a virtual drilling data processing method and apparatus. By acquiring the range and historical exploration data of the target area, the target area is discretized into a regular grid model, and the historical exploration data is associated with the regular grid model. The geological attribute information of each grid in the regular grid model is spatially interpolated using geostatistical methods to obtain an initial belief map. A Markov decision process model is constructed based on the initial belief map. Based on the initial belief map and the Markov decision process model, virtual drilling is performed using a Monte Carlo tree search algorithm. The initial belief map is updated based on the virtual observation results obtained from the virtual drilling, and the immediate reward is determined. The expected cumulative reward is calculated based on the immediate reward and the updated belief map. The drilling scheme is determined based on one or more candidate boreholes with the highest expected cumulative reward, thereby improving the efficiency and accuracy of drilling. Attached Figure Description

[0048] 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.

[0049] Figure 1 This is one of the flowcharts illustrating the virtual drilling data processing method in the embodiments of this application;

[0050] Figure 2 This is a structural diagram of the virtual drilling data processing device in the embodiments of this application;

[0051] Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0052] Figure label:

[0053] 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

[0054] 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.

[0055] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0056] Traditional mineral resource exploration relies heavily on geological mapping, geophysical exploration, and limited borehole verification, resulting in long exploration cycles and high costs. This application provides a virtual drilling data processing method and apparatus. By acquiring the scope and historical exploration data of the target area, the target area is discretized into a regular grid model, and the historical exploration data is associated with the regular grid model. Geostatistical methods are used to spatially interpolate the geological attribute information of each grid in the regular grid model to obtain an initial belief map. A Markov decision process model is constructed based on the initial belief map. Based on the initial belief map and the Markov decision process model, virtual drilling is performed using a Monte Carlo tree search algorithm. The initial belief map is updated based on the virtual observation results obtained from the virtual drilling, and an immediate reward is determined. The expected cumulative reward is calculated based on the immediate reward and the updated belief map. A drilling plan is determined based on one or more candidate boreholes with the highest expected cumulative reward, thereby improving the efficiency and accuracy of drilling.

[0057] The overall approach of this application is as follows:

[0058] First, spatial modeling of mineralization indicators is performed using the Kriging method and Gaussian process regression to construct a "mineralization probability map," also known as an initial belief map. In other words, a guiding map with initial mineralization probabilities is first built in the target area. This map is constructed based on historical drilling data, including the locations of historical drilling points, drilling results, and surrounding geological properties.

[0059] Next, the resource exploration problem is transformed into a partially observable Markov decision process, and a Markov decision process model is constructed. Based on the initial belief map and the Markov decision process model, a Monte Carlo tree search algorithm is used to perform virtual drilling on this initial belief map, obtaining the virtual drilling results. The reward of the Markov decision process model that can be obtained from the virtual drilling results is output, and the initial belief map is adaptively updated using Bayesian algorithm based on the virtual drilling results to update the global mineralization probability, resulting in an updated belief map. Throughout the virtual drilling, the Monte Carlo tree search algorithm uses the UCT scoring method for simulated exploration.

[0060] Finally, by continuously iterating and updating the belief graph, the cumulative reward of the Markov decision process model is output. The virtual drilling point with the highest cumulative reward value is then output as the recommended drilling point. By drilling at this recommended point, drilling efficiency can be maximized. In this process, the reward function of the Markov decision process model is adaptively adjusted to the virtual drilling radiation intensity, allowing the reward value to be directly determined from the virtual radiation intensity value obtained during virtual drilling.

[0061] To improve the efficiency and accuracy of drilling, this application provides an embodiment of a virtual drilling data processing method, see [link to embodiment]. Figure 1 The virtual drilling data processing method specifically includes the following:

[0062] Step S101: Obtain historical exploration data of the target area, discretize the target area into a regular grid model, and associate the historical exploration data with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resource. The borehole data includes borehole location and borehole results.

[0063] Optionally, in this embodiment, the task of this step is to convert complex, continuous, and irregular geological exploration information in the real world into a structured data format.

[0064] Optionally, in this embodiment, all available historical exploration data are systematically extracted from past exploration reports, geological maps, borehole databases, and geophysical and geochemical measurement results of the target area.

[0065] Specifically, borehole data records the spatial coordinates (such as latitude, longitude, and elevation) of each drilled borehole and its drilling results, i.e. whether the target resource was found at that location (e.g., mineralization or no mineralization). In digital processing, this result is quantified into a binary variable, such as "mineralization found" being marked as 1 and "no mineralization found" being marked as 0.

[0066] Specifically, the geological attribute information of the target resource includes various indirect or direct evidence related to its mineralization potential. This can be grade data measured in boreholes (e.g., gold content in g / t), anomaly intensity from geophysical methods (e.g., magnetic and electrical methods), elemental enrichment concentrations from geochemical methods, linear tectonic or alteration zone information interpreted by remote sensing, and so on. These attributes collectively constitute the basis for inferring the resource potential of unknown areas. In this embodiment, the geological attribute information is merely an enumeration and not an exhaustive list; adding other types of attribute information related to mineralization potential will not affect the implementation of this embodiment.

[0067] In this embodiment, the integrity of the information is ensured by using direct drilling results and indirect geological attribute information. This not only allows us to learn where mineralization has occurred, but also to understand the geological patterns that suggest mineralization may occur here.

[0068] Optionally, in this embodiment, a regular spatial grid is defined for the entire target area based on the exploration accuracy requirements and the balance of computing resources. Preferably, a two-dimensional rectangular grid (or a cubic grid in three-dimensional applications) is used, and each grid unit (or pixel) has a uniform size (e.g., 25 meters × 25 meters).

[0069] Optionally, in this embodiment, each known data point (such as borehole location or geochemical sampling point) is assigned to a specific grid cell based on its geographical coordinates. A grid cell may contain multiple data points or none.

[0070] For each grid cell, assign a corresponding attribute value based on the known data falling into it.

[0071] For meshes with drilled holes, the drill result (1 or 0) is directly used as the most authoritative "label" for that mesh.

[0072] For grids with other geological attribute data, their measured values ​​(such as grade value, radiation intensity value) are recorded.

[0073] For a grid with no known data, its state is marked as "unknown" (often represented by NaN or a specific identifier in the program).

[0074] Understandably, the grid space constructed in this step precisely defines the knowledge status of the current exploration stage, intuitively showing the distribution of known mineral deposits, anomaly zones, and large blank areas, providing decision-makers with a clear spatial understanding.

[0075] Step S102: Using geostatistical methods, spatial interpolation is performed on the geological attribute information of each grid in the regular grid model based on the historical exploration data to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0076] Optionally, in this embodiment, the purpose of this step is to transform the discrete known point space constructed in step S101 into a continuous global probability (or attribute strength) distribution map, i.e., the initial belief map. Essentially, it is a process combining spatial interpolation and statistical inference.

[0077] Optionally, in this embodiment, geostatistical methods are used to calculate the dependence of geological attributes. The fundamental assumption is that geographically close points have similar geological attributes (such as grade and mineralization), and this similarity weakens as the distance between points increases.

[0078] First, iterate through all pairwise combinations of known grid points (i.e., points with historical borehole data). For each pair of points, calculate the spatial distance between them, and simultaneously calculate the square of the difference in their geological attribute values ​​(e.g., 1 for mineralization, 0 for no mineralization; directly using grade values, etc.). This process ultimately yields a dataset containing multiple sets of "distances h" and corresponding "variance values ​​γ(h)," i.e., a scatter plot of geological attributes.

[0079] Next, a continuous mathematical function is used to quantitatively describe the relationship between "distance" and "attribute dependency". Preferably, the mathematical function can be a function of variation or a function of covariance.

[0080] By fitting the calculated geological attribute scatter plot to the mathematical function model, we can determine the hyperparameters of the mathematical function model that are most suitable for the current target area. Thus, we obtain a mathematical function model that can quantify the geological attribute dependency between any two points. For any given distance h, we can calculate its covariance C(h) or variation γ(h) through this function model.

[0081] Once a quantitative model describing spatial dependencies is obtained, the attribute values ​​of all unknown grid cells can be predicted, and interpolation can be performed on the unknown grid cells. Specifically, this involves precise weighted calculations based on spatial dependencies.

[0082] Specifically, for each unknown grid point in the regular grid model, the system searches for all known grid points within a set search radius (usually related to the correlation length), centered on that unknown point. The goal is to assign appropriate weights to the attribute values ​​of a set of known points so that the estimated value of the unknown point calculated using these weights and known values ​​is optimal (unbiased and with minimal variance). The weight assignment is solved using the dependency values ​​(variance function / covariance function) determined in the above steps through the Kriging equations (a mathematical system).

[0083] Specifically, the core of the Kriging equations is:

[0084] The closer a known point is to an unknown point, the greater its weight should be.

[0085] If the known points are close to each other (i.e., their information is redundant), their respective weights will be reduced accordingly to avoid duplicate calculations.

[0086] The sum of all weights is 1 to ensure the unbiasedness of the estimate;

[0087] For each unknown point, after solving for the optimal weights of all known points around it, the estimated value of the unknown point is calculated by weighted summation. This estimated value is the belief value of the unknown point grid.

[0088] For binary problems, the belief value is the prior probability of the existence of the target resource, ranging from [0, 1].

[0089] For continuous numerical values, the belief value represents the strength of the attribute.

[0090] Once all unknown grid points have been calculated, a continuous initial belief map covering the entire target area is generated. This map visually illustrates which areas are most likely to be rich in the target resources based on existing data.

[0091] Step S103: Using the initial belief map as the state space, determine the corresponding Markov decision process model. Input the initial belief map and the Markov decision process model into the Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location, determine the corresponding virtual observation results, update the initial belief map using a Bayesian algorithm based on the virtual observation results, determine the corresponding updated belief map, determine the expected cumulative reward of the corresponding candidate borehole based on the updated belief map and the Markov decision process model, output one or more candidate boreholes with the highest expected cumulative reward, and determine the corresponding drilling layout scheme based on the output candidate boreholes.

[0092] Optionally, in this embodiment, based on the initial belief graph described above, this step first constructs a decision engine. Specifically, the initial belief graph (belief(x,y)) is transformed into a decision optimization problem, and a formal mathematical model framework is established.

[0093] Specifically, we construct a Markov decision process model.

[0094] The state space is the initial belief graph. It refers to our current cognitive state regarding the probability of the existence of resources at all grid points, i.e., the entire belief graph itself. The probability value of each pixel (grid) in the belief graph collectively constitutes a high-dimensional state vector. This cognitive state continuously evolves as drilling progresses.

[0095] The action space is the set of all candidate borehole locations. Each action represents an instruction to drill at a specific coordinate (i, j) on a regular grid.

[0096] Observations are the results obtained after drilling operations are performed. In virtual drilling, they are simplified to binary results, such as "mineralization encountered" (represented by 1) or "mineralization not encountered" (represented by 0), or they can be continuous values ​​based on geological attributes (such as grade).

[0097] We designed the reward function as a numerical reward based on virtual observation results to quantify the immediate benefits obtained from drilling.

[0098] In other words, the following effects can be achieved with the design of the reward function:

[0099] Discovery Reward: A positive reward (e.g., +10) is given when drilling at a location and observing "mineralization". This value can be calibrated based on the estimated grade of the resource, the metal price, and the recovery rate.

[0100] Drilling cost: Each drilling operation, regardless of the outcome, generates a negative reward (i.e., cost, such as -1) to reflect the financial, time, and manpower costs of drilling.

[0101] This function forces the algorithm not only to find ore bodies but also to consider the cost of finding them. It opposes blind drilling and encourages obtaining the greatest geological discoveries and economic returns with the fewest drill holes and the lowest cost, thereby realizing the green exploration concept of "less drilling and more precise drilling".

[0102] State transition and observation model: In mineral exploration, it is usually assumed that the underground geological body is static and unchanging during the exploration period. Therefore, the state transition model describes the evolution of our cognitive state, rather than the changes in the geological body itself. The observation model is directly provided by the belief map: when drilling at a certain location (i, j), the observed probability of "mineralization" is the probability value belief(i, j) of the belief map at that location.

[0103] By constructing the POMDP model described above, we have precisely transformed a complex geological exploration problem into a sequential decision-making problem that can be processed and optimized by a computer. We have clarified the algorithm's input (current knowledge), possible actions (where to drill), possible outcomes (what to see), and ultimate goal (maximizing benefits), providing clear rules and objectives for the subsequent Monte Carlo tree search algorithm.

[0104] Optionally, in this embodiment, in order to pursue the optimal solution of the POMDP model, we use Monte Carlo Tree Search (MCTS) for virtual drilling simulation.

[0105] MCTS assesses the long-term value of each candidate borehole by simulating a large number of possible future scenarios.

[0106] Specifically, the selection steps for Monte Carlo Tree Search (MCTS).

[0107] Starting from the root node of the MCTS tree (the current belief state), the algorithm recursively selects the optimal child node and searches downwards. The selection criterion adopts the Upper Confidence Bound (UCT) formula, striking a balance between "utilization" and "exploration," where "utilization" is the "development" term in the formula, representing the development and utilization of the mineral resource.

[0108]

[0109] Where n is the node currently being evaluated (i.e., a candidate borehole location that we are considering).

[0110] n p It is the parent node of node n;

[0111] Q(n) is the total cumulative reward, which is the sum of the total rewards obtained from all subsequent simulations starting from this borehole position (node ​​n) (e.g., the sum of the rewards from all virtual drilling paths).

[0112] N(n) visit count, the total number of times this borehole location (node ​​n) is visited during the MCTS simulation;

[0113] N(n p Parent node visit count: The total number of times the parent node (i.e., the previous drilling decision) was visited;

[0114] c is the exploration constant, used to balance "development" and "exploration";

[0115] Development item: Prefer to select nodes with high average rewards in historical simulations (i.e., holes with good virtual drilling results in the past).

[0116] Exploration: Provide a "reward score" for nodes with relatively few visits to encourage the algorithm to try holes with unknown potential but possibly better.

[0117] Understandably, the UCT strategy ensures that the search process does not get stuck in local optima too early, and can systematically explore a wide decision space while focusing on decision paths that have already shown high returns.

[0118] Specifically, the extended steps of Monte Carlo Tree Search (MCTS).

[0119] When a node that has not yet been fully explored (i.e., a candidate borehole) is found, the algorithm creates one or more new child nodes for that node. These child nodes represent new belief states corresponding to different observations (such as "mineralized" or "mineralized") that may occur after the drilling action.

[0120] Specifically, the simulation steps of Monte Carlo Tree Search (MCTS).

[0121] Starting from the newly expanded node, the algorithm performs a "rapid virtual drilling" until the preset depth or the end of the simulation. During this process:

[0122] Based on the current belief map, a possible underground reality is randomly sampled.

[0123] In this sampled "virtual world," subsequent drilling actions are performed according to a certain strategy (such as random selection). For each virtual borehole, a virtual observation result is generated based on the "real" situation of that location in the virtual world (mineralized or unmineralized) and combined with a simple noise model. The total reward obtained for the entire simulated path is calculated cumulatively.

[0124] Specifically, the backhaul steps of Monte Carlo Tree Search (MCTS).

[0125] The cumulative reward obtained in the simulation steps is propagated back along the previously selected path to update the statistics of all nodes on the path from the leaf node to the root node (mainly the total reward and the number of visits).

[0126] After repeating the above four steps multiple times, the algorithm has fully evaluated all possible actions starting from the current state. Ultimately, the recommended drilling scheme selects the action (hole position) corresponding to the child node with the most visits under the root node. This is because the number of visits reflects the algorithm's overall trust in the value of that action after weighing exploration and utilization.

[0127] Optionally, in this embodiment, in each simulation of MCTS, after obtaining a virtual observation result, the algorithm uses a Bayesian algorithm to update the current belief state, thereby entering the next decision node. The principle of Bayesian update is: the "new belief" is proportional to the "observation likelihood" multiplied by the "old belief (prior)".

[0128] Specifically: if the virtual drilling result at location (i, j) is "mineralized", the probability of that location and its surrounding area is increased based on a spatial correlation model (such as Kriging variance); conversely, it is decreased. This updated belief graph, as the state of the child nodes, is used for subsequent simulation decisions.

[0129] Repeat the MCTS loop (select, expand, simulate, backpropagate) thousands or even millions of times until the result converges. Finally, the algorithm analyzes all direct child nodes under the root node of the decision tree (i.e., the first drilling action of all candidates). The node with the most visits is selected as the optimal action. A high number of visits indicates that the action has been explored extensively and deeply, and its average performance (expected cumulative reward) is trusted by the algorithm.

[0130] The final output of one or more (e.g., Top-K) candidate borehole locations represents the drilling layout with the highest expected economic value, based on extensive virtual validation under the current cognitive state. This layout serves as the starting point for a dynamic decision-making sequence. After actually drilling the first recommended hole, the obtained real observation data is used to update the initial belief map. The entire process can then be repeated to provide recommendations for the next borehole, forming a closed-loop intelligent exploration system of "drilling-updating-optimization".

[0131] This example demonstrates how this embodiment uses historical exploration data to describe the spatial correlation of mineralization indicators using geostatistics, constructs an initial belief map, updates the belief map with global mineralization probability based on Bayesian methods, and finally uses MCTS to simulate the expected net returns of different borehole sequences to select the optimal borehole scheme.

[0132] As described above, the virtual drilling data processing method provided in this application can obtain the range and historical exploration data of the target area, discretize the target area into a regular grid model, associate the historical exploration data with the regular grid model, spatially interpolate the geological attribute information of each grid in the regular grid model using geostatistical methods to obtain an initial belief map, construct a Markov decision process model based on the initial belief map, perform virtual drilling using the Monte Carlo tree search algorithm based on the initial belief map and the Markov decision process model, update the initial belief map and determine the immediate reward based on the virtual observation results obtained from the virtual drilling, calculate the expected cumulative reward based on the immediate reward and the updated belief map, and determine the drilling scheme based on one or more candidate boreholes with the highest expected cumulative reward, thereby improving the efficiency and accuracy of drilling.

[0133] In one embodiment of the virtual drilling data processing method of this application, it may further include the following:

[0134] Step S201: Construct the state space of the Markov decision process model. The state space is defined as the initial belief graph of the regular grid model, wherein each state of the state space corresponds to the probability distribution of the existence of the target resource in the grid of the initial belief graph.

[0135] Step S202: Construct the action space of the Markov decision process model, wherein the action space is defined as a preset set of candidate borehole locations;

[0136] Step S203: Construct the observation space of the Markov decision process model, wherein the observation space is defined as a set of binary observation results, including mineralization and non-mineralization;

[0137] Step S204: Construct a state transition sub-model of the Markov decision process model. The state transition probability of the state transition sub-model is set as an identity matrix to represent that the geological state itself remains unchanged under drilling operations.

[0138] Step S205: Construct the observation sub-model of the Markov decision process model, wherein the probability observed by the observation sub-model is defined as the prior probability or attribute strength of the current belief graph;

[0139] Step S206: Construct the reward function for the Markov decision process model, wherein the reward function is used to quantify the virtual observation results;

[0140] Step S207: Construct a discount factor for the Markov decision process model, wherein the discount factor is a non-zero number.

[0141] Optionally, in this embodiment, this step constructs a decision engine based on the initial belief map described above. Specifically, the initial belief map (belief(x,y)) is transformed into a decision optimization problem. A mathematical model framework is established for the exploration decision problem, closely integrating geological uncertainties with benefit objectives.

[0142] Specifically, POMDP seven-tuple modeling (Markov decision process model).

[0143] formula: .

[0144] Specifically, construct the state space S.

[0145] The state space represents the actual subsurface conditions, i.e., whether each grid cell contains economically exploitable ore bodies. In real exploration, this is a partially observable state, and we cannot directly know the whole picture. Therefore, in the POMDP model, "state" is defined as the "belief state" of the current understanding of the subsurface. This belief state is the initial belief map itself.

[0146] The initial belief map is a two-dimensional probability matrix that we have constructed, corresponding one-to-one with the regular grid. The value of each cell (between 0 and 1) represents the confidence that the target resource exists at that location. The entire state space is the set of all possible belief maps.

[0147] Specifically, construct action space A.

[0148] The action space is defined as a predefined set of candidate borehole locations, where each action represents drilling at a specific location within a regular grid model. In implementation, candidate locations are selected from all undrilled grids based on geological constraints, engineering feasibility, and exploration objectives, forming a finite set of actions. This ensures the algorithm can search and optimize within a clearly defined decision-making scope. A well-constructed action space is fundamental to subsequent strategy optimization.

[0149] Specifically, construct the observation space O.

[0150] The observation space defines the possible outcomes after performing an action (drilling a hole). Preferably, to simplify the model and highlight the main issues, we define observation as a binary set of observation results, i.e., obtaining either "mineralization encountered" or "mineralization not encountered" after each drilling operation. This definition originates from the basic conclusions of borehole core identification or rapid on-site analysis in actual exploration. Although the simplification of the observation space results in the loss of some information details, it ensures the stable operation of the algorithm under limited data conditions, making it particularly suitable for scenarios with high information uncertainty in the early exploration stages.

[0151] Specifically, construct the state transition sub-model T.

[0152] The state transition sub-model is used to describe the evolution of the geological state after the execution of actions. In this method, the model is set as an identity matrix, meaning that the properties of the geological body itself (such as ore body morphology and grade distribution) are assumed to remain unchanged during a single drilling operation. This assumption is based on the static characteristics of the geological body over a short timescale, which aligns with the physical context of actual exploration. From an implementation perspective, the identity matrix setting greatly simplifies the calculation of state transitions, avoids the uncertainties brought about by complex geological dynamic modeling, and allows the algorithm to focus on belief updates and strategy optimization, thereby improving the overall feasibility and efficiency of the solution.

[0153] Specifically, we construct the observation sub-model Z.

[0154] The observation sub-model defines the probability of obtaining a specific observation result after performing a drilling operation under a given geological condition. In its implementation, this probability is directly derived from the prior probability or attribute strength value of the corresponding grid in the current belief map. For example, the probability of encountering minerals at a certain location is equal to the mineralization probability of that location in the belief map. This model combines geological spatial cognition (represented in the form of a belief map) with the randomness of the observation process, establishing a probabilistic correlation between decision-making and observation. Its role is to provide a likelihood calculation basis for Bayesian updates, ensuring that the cognition of subsurface conditions can be reasonably adjusted based on the observation results after each virtual drilling operation.

[0155] Specifically, construct the reward function R.

[0156] The reward function quantifies the immediate benefits gained from each drilling operation and is a key mechanism guiding the algorithm to learn and optimize its strategy. Preferably, in this scheme, the reward function is designed as a numerical reward based on virtual observation results. For example, a positive reward is given if minerals are found (reflecting the value of resource discovery), and a negative reward is given if no minerals are found (reflecting drilling costs). Optionally, the reward function can also consider other factors such as borehole depth and location priority. By reasonably setting the reward value, the exploration objective can be transformed into a mathematical indicator that the algorithm can optimize. The design of the reward function directly affects the strategy's tendency.

[0157] Specifically, a discount factor γ is constructed.

[0158] A discount factor is used to adjust the weight of future rewards in current decisions. This is achieved by exponentially decaying the reward at future time steps when calculating cumulative rewards. For example, a discount factor of 0.9 means that the reward for the next step is discounted to 90% of the current reward, the reward for the step after that is discounted to 81%, and so on. This mechanism guides the algorithm to prioritize near-term gains, encouraging the achievement of goals with fewer drilling runs and shorter exploration cycles. It avoids excessively pursuing potentially high long-term returns while ignoring current costs and risks, thus reflecting the principle of "earning money sooner is better than earning it later."

[0159] Through step S207, this embodiment obtains a Markov decision process model with a complete structure and clear definition, transforming the highly uncertain sequential decision problem of geological exploration into a mathematical framework that can be efficiently solved by algorithms such as Monte Carlo tree search.

[0160] In one embodiment of the virtual drilling data processing method of this application, it may further include the following:

[0161] Step S301: Initialize the root node of the preset search tree according to the belief graph, determine the corresponding Monte Carlo tree search algorithm, the Monte Carlo tree search algorithm adopts UCT scoring and constructs development items and exploration items, the development items and exploration items are used to balance exploration and development;

[0162] Step S302: Each node of the Monte Carlo tree search algorithm is used to count the total number of times the node is visited and the cumulative reward obtained from the node.

[0163] Optionally, in this embodiment, this step is the construction step of the Monte Carlo tree search algorithm.

[0164] First, the search tree is initialized and its selection strategy is established. At the start of the algorithm, the "belief graph," representing the probability of the global resource distribution, is encapsulated as the root node of the search tree. This root node represents the starting point of the decision-making process, i.e., the initial state before any virtual drilling has been performed. Simultaneously, UCT (Upper Confidence Interval) is established as the standard scoring mechanism for tree node selection. The UCT score is composed of the sum of "development items" and "exploration items":

[0165]

[0166] The development item is calculated as the average reward a node receives (total cumulative reward divided by the number of visits). It tends to select nodes with high historical returns, guiding the algorithm to "cultivate" known high-potential areas, i.e., to maximize returns using existing knowledge. The exploration item, on the other hand, is proportional to the square root of the total number of visits to the parent node and inversely proportional to the number of visits to the node itself. Its purpose is to give extra "points" to nodes with relatively few visits, encouraging the algorithm to explore areas that have not yet been fully evaluated, thus avoiding getting trapped in local optima. By adjusting the weights of these two items with a constant coefficient, the UCT scoring mechanism cleverly balances the core contradiction between "developing" known high-value areas and "exploring" unknown potential areas.

[0167] Second, define the key statistical information that each node in the search tree needs to maintain and update. During the MCTS simulation, the algorithm creates and traverses a large number of nodes, each corresponding to a specific decision state (e.g., the belief state after drilling at a certain location and obtaining a specific result). To support the calculation of the UCT score and provide a basis for subsequent decisions, each node needs to dynamically record two types of core data: the first is the "total number of visits," which is a counter that records how many times the node has been selected and visited during the simulation. The second is the "cumulative reward," which is an accumulator that records the sum of all immediate rewards obtained from that node in all subsequent simulation paths. Each time a simulation is completed, the reward value obtained is backtracked along the access path, updating these two statistics for each node on the path.

[0168] Through step S302, this embodiment successfully constructed the Monte Carlo tree search algorithm, laying the foundation for subsequent application of Monte Carlo tree search for efficient virtual drilling.

[0169] In one embodiment of the virtual drilling data processing method of this application, it may further include the following:

[0170] Step S401: Input the initial belief map and the Markov decision process model into the Monte Carlo tree search algorithm. The Monte Carlo tree search algorithm selects a candidate borehole location from the action space of the Markov decision process model as the current simulation action, and performs simulation sampling according to the initial belief map to determine the geological realization of the resource distribution in the corresponding target area.

[0171] Step S402: In the geological realization of the resource distribution in the target area, perform forward modeling calculation on the borehole location corresponding to the current simulation action to determine the theoretical observation value of the borehole location;

[0172] Step S403: Superimpose the theoretical observation values ​​with the preset observation noise model to determine the corresponding virtual observation results. The observation noise model is used to simulate the measurement uncertainty in the actual drilling process.

[0173] Optionally, in this embodiment, this step is an application step of Monte Carlo tree search.

[0174] First, the MCTS algorithm selects a location from the Markov decision process model's action space—that is, from all preset candidate borehole locations—as the "simulated action" to be evaluated. This selection can be based on the tree search strategy from the previous step, or it can be randomly selected from a new simulation path.

[0175] Secondly, the simulation sampling based on the initial belief map involves treating the belief map as a probability distribution and randomly selecting a possible subsurface resource distribution scenario from this distribution, called a geological realization. By generating a large number of different geological realizations, the algorithm can explore all possible subsurface conditions that conform to current knowledge, thereby avoiding relying on a single model for decision-making. The geological realization provides a computational environment for subsequent virtual drilling.

[0176] Next, in the generated geological implementation, the algorithm will perform selected simulation actions to conduct virtual drilling.

[0177] Forward modeling of virtual drilling is a process of calculating observation results based on a known model (in this case, geological realization) and known physical laws. That is, along the set borehole trajectory, the geological attribute information of each grid passed through is queried and integrated, and finally one or more theoretical observation values ​​are output.

[0178] Finally, in actual exploration, any measurement inevitably contains errors and uncertainties. These uncertainties may stem from errors in laboratory analysis, sample depletion or dilution caused by core drilling, the subjectivity of geological logging, and the microscopic heterogeneity of the subsurface geological body itself. The theoretical observations obtained from the above steps are ideal values; to more realistically simulate reality, perturbations must be introduced.

[0179] In this embodiment, an observation noise model is introduced. The noise model is a mathematical probability distribution used to quantitatively describe the statistical characteristics of the observation error. The most common model is a Gaussian (normal) distribution with a mean of zero and a variance of σ², indicating that the error fluctuates around the true value, and the range of fluctuation is determined by σ². Specifically, the theoretical observation value is superimposed with a noise value randomly selected from the noise model, and the result of the superposition is the final virtual observation result.

[0180] Through step S403, this embodiment successfully completed a large number of virtual drilling activities based on the initial belief map and Markov decision process model using the Monte Carlo algorithm.

[0181] In one embodiment of the virtual drilling data processing method of this application, it may further include the following:

[0182] Step S501: Use the initial belief map as a prior probability distribution;

[0183] Step S502: Quantify the virtual observation results into virtual radiation intensity, and determine the corresponding observation likelihood function based on the virtual radiation intensity, wherein the virtual radiation intensity is calculated based on the grade, mineralization thickness and burial depth in the virtual observation results through a physical-statistical coupling model;

[0184] Step S503: Multiply the prior probability distribution with the observation likelihood function according to Bayes' theorem and normalize the result to determine the corresponding updated belief map, which represents the posterior probability distribution.

[0185] Optionally, in this embodiment, this step is based on a Bayesian reasoning-based belief graph dynamic update mechanism, which uses the observational evidence obtained after the virtual drilling in step S403 above to continuously correct the understanding of the spatial distribution of underground resources.

[0186] Specifically, the initial belief map is first defined as the prior probability distribution under the current state. This belief map is a two-dimensional probability field covering the entire regular grid, where the value of each grid cell represents the probability estimate of the existence of a target resource (such as an ore body) at that location based on all historical data. This prior distribution encapsulates the algorithm's comprehensive understanding and uncertainty of the geological conditions of the exploration area up to the current moment, providing a theoretical basis for incorporating new evidence.

[0187] Specifically, the "observation results" of virtual drilling are transformed into a mathematical form that can be used for Bayesian updates.

[0188] Original Bayesian belief update formula:

[0189]

[0190] Where b(s) is the old belief, which is also the prior belief, representing the probability of finding minerals at each grid point s before drilling;

[0191] T is the state transition matrix, representing the probability that the world scene will transition from s to s' after action a is performed;

[0192] Z is the observation likelihood, which represents the probability that if the actual underground state is s', we will drill at location a and observe o (e.g., mineralization = 1).

[0193] b'(s') is the new belief, which is also the posterior belief. It represents the probability distribution of the subsurface state s' after new observation data o is obtained.

[0194] We simplified the original Bayesian update formula, and the belief update is simplified to:

[0195] High virtual strength → High probability of finding minerals

[0196]

[0197] By introducing virtual drilling and virtual radiation intensity, we have greatly simplified the update process. The formula is no longer as theoretical and complex as the original formula, but has become more intuitive.

[0198] Where b(s) is the old belief, which is exactly the same as in the original formula, and is the mineralization probability diagram before the update;

[0199] I virt It is the radiation intensity observed virtually. In the simulation phase of MCTS, we do not randomly guess "mineralization or non-mineralization", but calculate the radiation intensity value through a physical model. This will be a continuous value with richer information.

[0200] It is a simplified version of the observation likelihood, which is a probability density function of a normal distribution, where, This is the radiation intensity that we believe should be observed in state s. It is the uncertainty of observation;

[0201] Proportional to means that "the result calculated on the right side, after normalization, is the same as b'(s) on the left side";

[0202] The new belief, which is exactly the same as the original formula, is an updated mineralization probability diagram.

[0203] The formula for measuring virtual strength is as follows:

[0204]

[0205] I: Virtual drill radiation intensity (any unit, linearly comparable to the "radiation flux value" in the report)

[0206] C: Predicted grade (g / t)

[0207] T: Predicted mineralization thickness (m)

[0208] Z: Predicted burial depth (m)

[0209] α: Calibration coefficient (calculated by back-calculating from existing boreholes to minimize the RMS error between I and the measured flux value)

[0210] First, the multidimensional virtual observation results (such as grade, mineralization thickness, burial depth, etc.) are aggregated into a single indicator with clear physical meaning—"virtual radiation intensity"—through a "physical-statistical coupling model".

[0211] Subsequently, based on this quantified virtual radiation intensity value, an "observation likelihood function" is constructed. This function defines the probability of observing the current virtual radiation intensity value under the assumption that resources exist (or do not exist) at a certain grid point. Its role is to establish a probabilistic relationship between geological conditions (whether the grid contains minerals) and observational data (radiation intensity), serving as a bridge connecting new evidence with prior knowledge.

[0212] Specifically, multiplying the prior probability distribution with the observation likelihood function means that the prior probability is enhanced in grid regions where the virtual observation result is "more likely" based on the likelihood function; conversely, the prior probability is weakened in regions where the observation result is "less likely".

[0213] Since the result of direct multiplication is usually no longer a standard probability distribution (the sum of probabilities is not 1), it is necessary to perform "normalization" to restore the sum of probabilities across the entire domain to 1.

[0214] The normalized result is the "updated belief map," which is also a posterior probability distribution. It is an updated global resource probability model that incorporates the latest virtual drilling evidence, providing the latest and less uncertain state basis for the Monte Carlo tree search algorithm to make the next round of decision-making and path optimization.

[0215] Through step S503, this embodiment successfully performs dynamic updates to the belief graph based on Bayesian inference.

[0216] In one embodiment of the virtual drilling data processing method of this application, it may further include the following:

[0217] Step S601: Use the updated belief map as the new state input of the Markov decision process model, recursively perform virtual drilling and belief updates, and discount the subsequent rewards according to the discount factor to determine the corresponding subsequent rewards;

[0218] Step S602: Add the immediate reward of virtual drilling to the subsequent reward to determine the expected cumulative reward of the corresponding candidate borehole under the current simulation path. The immediate reward of virtual drilling is obtained from the reward function based on the virtual observation results. The reward function quantifies the virtual observation results into a virtual radiation intensity value that is proportional to the target resource intensity.

[0219] Optionally, in this embodiment, this step describes how, during a Monte Carlo Tree Search (MCTS) simulation or rollout phase, the long-term expected cumulative reward of a candidate borehole action is estimated, starting from the current state (i.e., the updated belief graph). The core of this process lies in evaluating the potential long-term value of the current decision, rather than just the immediate one-step benefit, by simulating future exploration sequences.

[0220] Specifically, the belief map, updated by Bayesian analysis and representing the latest geological knowledge, is used again as the current state input for the Partially Observable Markov Decision Process (POMDP). The algorithm does not stop there, but based on this new belief state, it continues to select subsequent virtual drilling actions (e.g., using a tree strategy or a random strategy), performs virtual drilling again, obtains new virtual observation results, and updates the belief map accordingly.

[0221] Each recursive step receives an immediate reward from the reward function based on its virtual observations. This recursive process continues until the simulation terminates (e.g., reaching a preset simulation depth, exhausting the budget, or exploring the boundary). Finally, all immediate rewards generated by this series of future actions are discounted using a preset discount factor and summed to calculate the "subsequent reward." The discount factor (γ, typically less than 1) ensures that the algorithm favors near-term gains, adhering to the economic principle of "earning money sooner is better," while also guaranteeing that the sum of rewards over an infinite time series converges.

[0222] Specifically, after simulating future paths and calculating subsequent rewards, the value is summarized and backtracked. The immediate reward directly generated by this virtual drilling (i.e., the action currently being evaluated) is added to the sum of all calculated future rewards to obtain the expected cumulative reward for this candidate borehole under the current specific simulated path.

[0223] In the reward function of the Markov decision process model, we rewrite the reward function into a form that can be directly measured by virtual radiation intensity. The reward function is:

[0224]

[0225] β: Economic conversion factor

[0226] Cdrill: Drilling cost (RMB / hole)

[0227] The mathematical expression for the virtual drill radiation intensity I(a) is as follows:

[0228] Physical assumptions

[0229] Radiation intensity ∝ Grade

[0230] Radiation intensity ∝ mineralization thickness

[0231] Radiation intensity ∝ 1 / (burial depth + d0) (d0 is a small constant to prevent division by zero)

[0232]

[0233] I: Virtual drill radiation intensity (any unit, linearly comparable to the "radiation flux value" in the report)

[0234] C: Predicted grade (g / t)

[0235] T: Predicted mineralization thickness (m)

[0236] Z: Predicted burial depth (m)

[0237] α: Calibration coefficient (calculated by back-calculating from existing boreholes to minimize the RMS error between I and the measured flux value)

[0238] In terms of the resulting effects, the reward function quantifies the virtual observation results (such as mineral grade and thickness) into a virtual radiation intensity value that is proportional to the target resource intensity. This makes the immediate reward no longer a simple binary "mineral found / mineral not found" return, but a continuous and rich value signal that can more accurately reflect the economic potential of borehole discoveries.

[0239] Through step S602, this embodiment successfully enables MCTS to accurately identify the optimal borehole locations that can bring direct high returns and provide high information value for subsequent exploration, ultimately achieving the intelligent drilling target of maximizing global benefits.

[0240] In one embodiment of the virtual drilling data processing method of this application, it may further include the following:

[0241] Step S701: Using geostatistical methods, the geological attribute dependency of each known grid in the regular grid model is calculated based on the historical exploration data to determine the corresponding dependency value. The geostatistical methods include the variation function method and the covariance function method.

[0242] Step S702: Perform weighted calculations on the geological attribute information of the unknown grids in the regular grid model based on the dependency relationship value to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0243] Optionally, this embodiment describes the initial belief map construction process.

[0244] First, the geological attribute dependency calculation iterates through all the point pairs formed by these known points. For each pair, the spatial distance and semivariance are calculated. All point pairs are grouped according to their distance h (i.e., placed into different "distance tolerance" intervals), and then the average of all semivariances within each distance group is calculated. Finally, a series of data points consisting of "distance h" and the corresponding "mean semivariance γ(h)" is output. This "distance-semivariance" scatter plot is the geological attribute scatter plot, revealing the original statistical law of geological attribute differences varying with distance within the current region.

[0245] Next, the scatter plot of geological attributes is fitted to a preset function, which can be either a function of variation or a function of covariance. The function of variation measures "difference"; a larger value indicates a greater expected difference between two points, meaning they are less similar. The function of covariance measures "co-variance"; a larger value indicates that two points are more likely to be simultaneously above or below the average, meaning they are more similar.

[0246] It is understandable that the variation function and the covariance function are mathematically equivalent. The variation function is used to provide the optimal interpolation result for the subsequent process, while the covariance function model is used to provide the posterior probability distribution for the subsequent process.

[0247] Once the fitting is complete, the function model is officially determined. For any two grid points in the regular grid model, regardless of whether they are known or not, a definite dependency value can be calculated using the function model.

[0248] Finally, for each unknown grid point in the regular grid model, the system searches for all known grid points within a set search radius (usually related to the correlation length l), centered on that unknown point. The goal is to assign appropriate weights to the attribute values ​​of a set of known points so that the estimates of the unknown points calculated using these weights and known values ​​are optimal (unbiased and with minimal variance). The weight assignment is solved using the dependency values ​​(variance function / covariance function) determined in the above steps through the Kriging equations (a mathematical system).

[0249] Specifically, the core of the Kriging equations is:

[0250] The closer a known point is to an unknown point, the greater its weight should be.

[0251] If the known points are close to each other (i.e., their information is redundant), their respective weights will be reduced accordingly to avoid duplicate calculations.

[0252] The sum of all weights is 1 to ensure the unbiasedness of the estimate;

[0253] For each unknown point, after solving for the optimal weights of all known points around it, the estimated value of the point is calculated by weighted summation. This estimated value is the belief value of the grid for the unknown point.

[0254] For binary problems, the belief value is the prior probability of the existence of the target resource, ranging from [0, 1].

[0255] For continuous numerical values, the belief value represents the strength of the attribute.

[0256] Once all unknown grid points have been calculated, a continuous initial belief map covering the entire target area is generated. This map visually illustrates which areas are most likely to be rich in the target resources based on existing data.

[0257] Through step S702, this embodiment successfully constructed an initial belief map, laying a solid data foundation for subsequent virtual drilling.

[0258] To improve the efficiency and accuracy of drilling, this application provides an embodiment of a virtual drilling data processing apparatus for implementing all or part of the virtual drilling data processing method, see [link to embodiment]. Figure 2 The virtual drilling data processing device specifically includes the following components:

[0259] The target area grid determination module 10 is used to acquire historical exploration data of the target area, discretize the target area into a regular grid model, and associate the historical exploration data with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resource. The borehole data includes borehole location and borehole results.

[0260] The initial belief map determination module 20 is used to determine the corresponding initial belief map by spatially interpolating the geological attribute information of each grid in the regular grid model based on the historical exploration data using geostatistical methods. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0261] The virtual drilling module 30 is used to take the initial belief map as the state space, determine the corresponding Markov decision process model, input the initial belief map and the Markov decision process model into a Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location, determine the corresponding virtual observation results, update the initial belief map according to the virtual observation results using a Bayesian algorithm, determine the corresponding updated belief map, determine the expected cumulative reward of the corresponding candidate borehole according to the updated belief map and the Markov decision process model, output one or more candidate boreholes with the highest expected cumulative reward, and determine the corresponding drilling layout scheme according to the output candidate boreholes.

[0262] As described above, the virtual drilling data processing device provided in this application embodiment can acquire the range and historical exploration data of the target area, discretize the target area into a regular grid model, associate the historical exploration data with the regular grid model, spatially interpolate the geological attribute information of each grid in the regular grid model using geostatistical methods to obtain an initial belief map, construct a Markov decision process model based on the initial belief map, perform virtual drilling using the Monte Carlo tree search algorithm based on the initial belief map and the Markov decision process model, update the initial belief map and determine the immediate reward based on the virtual observation results obtained from the virtual drilling, calculate the expected cumulative reward based on the immediate reward and the updated belief map, and determine the drilling scheme based on one or more candidate boreholes with the highest expected cumulative reward, thereby improving the efficiency and accuracy of drilling.

[0263] To further illustrate this solution, this application also provides a specific application example of using the aforementioned virtual drilling data processing device to implement the virtual drilling data processing method, which specifically includes the following:

[0264] The core physical principles of this scheme include:

[0265] 1.1 Spatial correlation modeling: Geostatistics (such as Kriging and Gaussian processes) are used to describe the spatial correlation of mineralization indicators and construct a "belief map".

[0266] 1.2 Bayesian Update: For each hole drilled, the global mineralization probability is updated using the observation results (mineralization found / mineralization not found) through the Bayesian formula.

[0267] 1.3 Maximizing Information Value: MCTS Simulation of Expected Net Gains (Discovery Rewards) for Different Drilling Sequences (Drilling cost), select the optimal location for the next hole.

[0268] The nuclear-related mathematical formulas used in this scheme include:

[0269] 1. POMDP 7-tuple

[0270] S: State space (whether each grid contains minerals)

[0271] A: Motion space (candidate borehole location)

[0272] O: Observation space (drilling results)

[0273] T: State transition (only measurement state is updated, geological state remains unchanged)

[0274] Z: Observation probability (the probability of finding the mineral is given by the belief map)

[0275] R:

[0276] γ: Discount factor (0.9)

[0277] 2. Belief Renewal (Bayes)

[0278]

[0279] Formula: b′(s′) ∝ Observational Likelihood Z(o∣s′,a) Predicted Prior ∑s T(s′∣s,a) b(s)

[0280]

[0281] 3. UCT rating (MCTS selection)

[0282]

[0283]

[0284] 4. Spatial covariance (Kriging / GPR kernel function)

[0285] ; Relevance length (Gaussian kernel)

[0286] Numerator: 2 (empirical multiple, adjustable) Denominator: 1 (implicit) Unit: meter (m)

[0287]

[0288]

[0289] This solution can be applied in the following situations:

[0290]

[0291] The implementation steps and code examples for this solution are as follows:

[0292]

[0293] The concept of virtual drilling and its application in this scheme:

[0294] Virtual drilling refers to "drilling" a hole in a computer without actually drilling, using numerical simulation and rapid inversion to immediately obtain possible observations (lithology, mineralization, elemental content, etc.). These virtual observations are then fed into the POMDP belief update module, instantly refreshing the global mineralization probability map and updating the initial belief map, thereby assessing "what the expected net benefit would be if we actually drilled here."

[0295] Essentially, it equals differentiable / fast forward modeling + Bayesian inversion + observation noise model, returning results in milliseconds. This eliminates the need for simple random sampling in the "rollout" phase of MCTS. Instead, it uses "virtual real diamond" feedback based on physical-statistical coupling, significantly improving policy accuracy.

[0296] The underlying physical framework of virtual drilling is as follows:

[0297]

[0298] Virtual drilling in the closed-loop integration of POMDP-MCTS includes:

[0299] Tree node expansion → Candidate action a (hole position + hole depth)

[0300] Virtual Drill Rollout

[0301] Sample a geological realization si from the current belief b(s).

[0302] In this implementation, action a is executed, which calls the fast forward model ovirt.

[0303] Update belief using ovirt → new belief b′

[0304] Immediately calculate and accumulate the single-step reward r.

[0305] Backpropagation → Send back the accumulated rewards and update the UCT statistics.

[0306] Repeat this process hundreds of times, then select the node with the "most virtual visits and the highest average Q" as the next real borehole.

[0307] The steps to implement virtual drilling:

[0308]

[0309] The three technical forms of virtual drilling include:

[0310] 0D Quick Version - Only returns "Ore Found / Not Found" + Radiation Intensity (This version is the one with the uploaded MATLAB script).

[0311] 1D Logging Version – Returns grade-depth curves along the borehole track for updating the 3D block model.

[0312] 2D / 3D Differentiable Version – Uses differentiable earth modeling or deep learning proxy models, supports gradient backpropagation, and can directly perform end-to-end optimization.

[0313] Formula for virtual drill radiation intensity:

[0314] Physical assumptions

[0315] Radiation intensity ∝ Grade

[0316] Radiation intensity ∝ mineralization thickness

[0317] Radiation intensity ∝ 1 / (burial depth + d0) (d0 is a small constant to prevent division by zero)

[0318] Mathematical expression

[0319]

[0320] I: Virtual drill radiation intensity (any unit, linearly comparable to the "radiation flux value" in the report)

[0321] C: Predicted grade (g / t)

[0322] T: Predicted mineralization thickness (m)

[0323] Z: Predicted burial depth (m)

[0324] α: Calibration coefficient (calculated by back-calculating from existing boreholes to minimize the RMS error between I and the measured flux value)

[0325] Global Fast Calculation Process

[0326] Prior model

[0327] Grade surface C(x,y): Interpolation of the grade of observed boreholes using indicator kriging or GPR.

[0328] Thickness surface T(x,y): Abnormal vertical extension = Length of abnormal depth interval × Probability weight

[0329] Burial depth Z(x,y): The depth of the abnormal top plate is taken (the report has given the range of 5–145 m).

[0330] Grid-by-grid computation

[0331] The grid step size is 25 m (consistent with the minimum spacing of actual boreholes).

[0332] Substituting each (xi, yj) into the above equation yields the two-dimensional matrix Iij.

[0333] Cross-validation with satellite radiative flux

[0334] Linear regression: Ivirt = k Isat + b

[0335] If R² ≥ 0.8, accept the formula; otherwise, rewind and recalibrate α or adjust the prior.

[0336] For the adaptive application of virtual radiation intensity in this scheme:

[0337] 1. We rewrite the POMDP reward function into a form that can be directly measured by virtual radiation intensity.

[0338]

[0339] β: Economic conversion factor

[0340] Cdrill: Drilling cost (RMB / hole)

[0341] When MCTS nodes are expanded

[0342] Instead of randomly rolling out "mineral found / not found", it directly reads I(a).

[0343] 2. We simplify our belief to: High virtual strength → High probability of finding minerals.

[0344]

[0345] This eliminates the need for each "sampling-forward modeling" step, enabling a single preprocessing step followed by second-level optimization across the entire domain.

[0346] Finally, the "Ultimate Drilling Ranking" is output in one go.

[0347] Algorithm objective:

[0348]

[0349] step

[0350] Global computation I(x,y)

[0351] Sort by I in descending order, and select the top 3 × N candidate holes.

[0352] Use MCTS (or simply a greedy algorithm with distance constraints) to further refine the selection of N elements, ensuring the space does not cluster.

[0353] Output the final Top-N hole positions, expected profit, and cumulative metal quantity.

[0354] Let's take one example to illustrate:

[0355] (Already used α=1, β=0.3 million yuan·t) -1 Cdrill = 50,000 yuan, grid 25 m, distance ≥ 100 m (cluster removal)

[0356]

[0357] From a hardware perspective, in order to improve the efficiency and accuracy of drilling, this application provides an embodiment of an electronic device for implementing all or part of the virtual drilling data processing method, wherein the electronic device specifically includes the following:

[0358] The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the virtual drilling data processing method 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 virtual drilling data processing method in the present embodiment, and the contents of the embodiments of the virtual drilling data processing method are incorporated herein, and repeated details will not be described again.

[0359] 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.

[0360] In practical applications, some parts of the virtual drilling data processing method can be executed on the electronic device side as described above, or all operations can be completed on 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 on the client device, the client device may further include a processor.

[0361] 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.

[0362] 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.

[0363] In one embodiment, the virtual drilling data processing method function can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0364] Step S101: Obtain historical exploration data of the target area, discretize the target area into a regular grid model, and associate the historical exploration data with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resource. The borehole data includes borehole location and borehole results.

[0365] Step S102: Using geostatistical methods, spatial interpolation is performed on the geological attribute information of each grid in the regular grid model based on the historical exploration data to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0366] Step S103: Using the initial belief map as the state space, determine the corresponding Markov decision process model. Input the initial belief map and the Markov decision process model into the Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location, determine the corresponding virtual observation results, update the initial belief map using a Bayesian algorithm based on the virtual observation results, determine the corresponding updated belief map, determine the expected cumulative reward of the corresponding candidate borehole based on the updated belief map and the Markov decision process model, output one or more candidate boreholes with the highest expected cumulative reward, and determine the corresponding drilling layout scheme based on the output candidate boreholes.

[0367] As described above, the electronic device provided in this application obtains the range and historical exploration data of the target area, discretizes the target area into a regular grid model, and associates the historical exploration data with the regular grid model. It then uses geostatistical methods to spatially interpolate the geological attribute information of each grid in the regular grid model to obtain an initial belief map. Based on the initial belief map, a Markov decision process model is constructed. Based on the initial belief map and the Markov decision process model, virtual drilling is performed using a Monte Carlo tree search algorithm. The initial belief map is updated based on the virtual observation results obtained from the virtual drilling, and an immediate reward is determined. The expected cumulative reward is calculated based on the immediate reward and the updated belief map. Finally, a drilling plan is determined based on one or more candidate boreholes with the highest expected cumulative reward. This improves the efficiency and accuracy of drilling.

[0368] In another embodiment, the virtual drilling data processing method can be configured separately from the central processing unit 9100. For example, the virtual drilling data processing method can be configured as a chip connected to the central processing unit 9100, and the virtual drilling data processing method function can be implemented through the control of the central processing unit.

[0369] 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 technologies.

[0370] 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.

[0371] 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.

[0372] 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.

[0373] 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.

[0374] 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.).

[0375] 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.

[0376] 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.

[0377] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the virtual drilling data processing method with a server or client as the execution subject in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the virtual drilling data processing method with a server or client as the execution subject in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0378] Step S101: Obtain historical exploration data of the target area, discretize the target area into a regular grid model, and associate the historical exploration data with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resource. The borehole data includes borehole location and borehole results.

[0379] Step S102: Using geostatistical methods, spatial interpolation is performed on the geological attribute information of each grid in the regular grid model based on the historical exploration data to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0380] Step S103: Using the initial belief map as the state space, determine the corresponding Markov decision process model. Input the initial belief map and the Markov decision process model into the Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location, determine the corresponding virtual observation results, update the initial belief map using a Bayesian algorithm based on the virtual observation results, determine the corresponding updated belief map, determine the expected cumulative reward of the corresponding candidate borehole based on the updated belief map and the Markov decision process model, output one or more candidate boreholes with the highest expected cumulative reward, and determine the corresponding drilling layout scheme based on the output candidate boreholes.

[0381] As described above, the computer-readable storage medium provided in this application embodiment obtains the range and historical exploration data of the target area, discretizes the target area into a regular grid model, associates the historical exploration data with the regular grid model, spatially interpolates the geological attribute information of each grid in the regular grid model using geostatistical methods to obtain an initial belief map, constructs a Markov decision process model based on the initial belief map, performs virtual drilling using a Monte Carlo tree search algorithm based on the initial belief map and the Markov decision process model, updates the initial belief map and determines the immediate reward based on the virtual observation results obtained from the virtual drilling, calculates the expected cumulative reward based on the immediate reward and the updated belief map, and determines the drilling layout scheme based on one or more candidate boreholes with the highest expected cumulative reward, thereby improving the efficiency and accuracy of drilling layout.

[0382] Embodiments of this application also provide a computer program product capable of implementing all steps of the virtual drilling data processing method in the above embodiments, where the execution subject is a server or a client. When executed by a processor, this computer program / instruction implements the steps of the virtual drilling data processing method. For example, the computer program / instruction implements the following steps:

[0383] Step S101: Obtain historical exploration data of the target area, discretize the target area into a regular grid model, and associate the historical exploration data with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resource. The borehole data includes borehole location and borehole results.

[0384] Step S102: Using geostatistical methods, spatial interpolation is performed on the geological attribute information of each grid in the regular grid model based on the historical exploration data to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

[0385] Step S103: Using the initial belief map as the state space, determine the corresponding Markov decision process model. Input the initial belief map and the Markov decision process model into the Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location, determine the corresponding virtual observation results, update the initial belief map using a Bayesian algorithm based on the virtual observation results, determine the corresponding updated belief map, determine the expected cumulative reward of the corresponding candidate borehole based on the updated belief map and the Markov decision process model, output one or more candidate boreholes with the highest expected cumulative reward, and determine the corresponding drilling layout scheme based on the output candidate boreholes.

[0386] As described above, the computer program product provided in this application obtains the scope and historical exploration data of the target area, discretizes the target area into a regular grid model, and associates the historical exploration data with the regular grid model. It then uses geostatistical methods to spatially interpolate the geological attribute information of each grid in the regular grid model to obtain an initial belief map. Based on the initial belief map, it constructs a Markov decision process model. Based on the initial belief map and the Markov decision process model, it performs virtual drilling using a Monte Carlo tree search algorithm. The initial belief map is updated based on the virtual observation results obtained from the virtual drilling, and an immediate reward is determined. The expected cumulative reward is calculated based on the immediate reward and the updated belief map. Finally, a drilling plan is determined based on one or more candidate boreholes with the highest expected cumulative reward. This improves the efficiency and accuracy of drilling.

[0387] 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.

[0388] 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.

[0389] 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.

[0390] 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.

[0391] 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. A virtual drilling data processing method, characterized in that, The method includes: Historical exploration data of the target area is obtained, the target area is discretized into a regular grid model, and the historical exploration data is associated with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resource. The borehole data includes borehole location and borehole results. Geostatistical methods are used to spatially interpolate the geological attribute information of each grid in the regular grid model based on the historical exploration data to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid. Using the initial belief graph as the state space, a corresponding Markov decision process model is determined. The initial belief graph and the Markov decision process model are input into a Monte Carlo tree search algorithm to perform virtual drilling at each candidate borehole location, determining the corresponding virtual observation results. Based on the virtual observation results, the initial belief graph is updated using a Bayesian algorithm to determine the corresponding updated belief graph. Based on the updated belief graph and the Markov decision process model, the expected cumulative reward of the corresponding candidate borehole is determined, and one or more candidate boreholes with the highest expected cumulative reward are output. Based on the output candidate boreholes, the corresponding drilling layout scheme is determined.

2. The virtual drilling data processing method according to claim 1, characterized in that, The step of using the initial belief graph as the state space to determine the corresponding Markov decision process model includes: Construct the state space of the Markov decision process model, wherein the state space is defined as the initial belief graph of the regular grid model, and each state of the state space corresponds to the probability distribution of the existence of the target resource in the grid of the initial belief graph; Construct the action space of the Markov decision process model, wherein the action space is defined as a preset set of candidate borehole locations; Construct the observation space for the Markov decision process model, whereby the observation space is defined as a set of binary observation results, including mineralization and non-mineralization; A state transition sub-model of a Markov decision process model is constructed. The state transition probability of the state transition sub-model is set as an identity matrix to represent that the geological state itself remains unchanged under drilling operations. An observation sub-model is constructed for a Markov decision process model, wherein the probability observed by the observation sub-model is defined as the prior probability or attribute strength of the current belief graph. Construct a reward function for a Markov decision process model, wherein the reward function is used to quantify virtual observation results; Construct a discount factor for a Markov decision process model, wherein the discount factor is a non-zero number.

3. The virtual drilling data processing method according to claim 1, characterized in that, Before inputting the initial belief map and the Markov decision process model into the Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location and determine the corresponding virtual observation results, the process includes: The root node of the preset search tree is initialized based on the belief graph, and the corresponding Monte Carlo tree search algorithm is determined. The Monte Carlo tree search algorithm uses UCT scoring and constructs development items and exploration items, which are used to balance exploration and development. Each node in the Monte Carlo tree search algorithm is used to count the total number of times the node is visited and the cumulative reward obtained from starting from that node.

4. The virtual drilling data processing method according to claim 1, characterized in that, The step of inputting the initial belief map and the Markov decision process model into a Monte Carlo tree search algorithm to perform virtual drilling at each candidate borehole location and determine the corresponding virtual observation results includes: The initial belief map and the Markov decision process model are input into the Monte Carlo tree search algorithm. The Monte Carlo tree search algorithm selects a candidate borehole location from the action space of the Markov decision process model as the current simulation action, and performs simulation sampling based on the initial belief map to determine the geological realization of the resource distribution in the corresponding target area. In the geological realization of the resource distribution in the target area, forward modeling is performed on the borehole location corresponding to the current simulation action to determine the theoretical observation value of the borehole location; The theoretical observation values ​​are superimposed with a preset observation noise model to determine the corresponding virtual observation results. The observation noise model is used to simulate the measurement uncertainty in the actual drilling process.

5. The virtual drilling data processing method according to claim 4, characterized in that, The step of updating the initial belief map using a Bayesian algorithm based on the virtual observation results, and determining the corresponding updated belief map, includes: The initial belief map is used as a prior probability distribution; The virtual observation results are quantified into virtual radiation intensity. Based on the virtual radiation intensity, the corresponding observation likelihood function is determined. The virtual radiation intensity is calculated using a physical-statistical coupling model based on the grade, mineralization thickness, and burial depth in the virtual observation results. The prior probability distribution is multiplied by the observation likelihood function according to Bayes' theorem and then normalized to determine the corresponding updated belief map, which represents the posterior probability distribution.

6. The virtual drilling data processing method according to claim 2, characterized in that, The step of determining the expected cumulative reward of the corresponding candidate borehole based on the updated belief graph and the Markov decision process model includes: The updated belief graph is used as the new state input of the Markov decision process model. Virtual drilling and belief updates are performed recursively, and the subsequent rewards are discounted according to the discount factor to determine the corresponding subsequent rewards. The immediate reward of virtual drilling is added to the subsequent reward to determine the expected cumulative reward of the corresponding candidate borehole under the current simulation path. The immediate reward of virtual drilling is obtained from the reward function based on the virtual observation results. The reward function quantifies the virtual observation results into a virtual radiation intensity value that is proportional to the target resource intensity.

7. The virtual drilling data processing method according to claim 1, characterized in that, The step of determining the corresponding initial belief map by spatially interpolating the geological attribute information of each grid in the regular grid model based on the historical exploration data using geostatistical methods includes: Based on the historical exploration data, the geological attribute dependence of each known grid in the regular grid model is calculated using geostatistical methods to determine the corresponding dependence value. The geostatistical methods include the variation function method and the covariance function method. The geological attribute information of unknown grids in the regular grid model is weighted and calculated based on the dependency relationship value to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid.

8. A virtual drilling data processing device, characterized in that, The device includes: The target area grid determination module is used to acquire historical exploration data of the target area, discretize the target area into a regular grid model, and associate the historical exploration data with the regular grid model. The historical exploration data includes borehole data and geological attribute information of the target resources. The borehole data includes borehole location and borehole results. The initial belief map determination module is used to perform spatial interpolation of the geological attribute information of each grid in the regular grid model based on the historical exploration data using geostatistical methods to determine the corresponding initial belief map. The initial belief map represents the prior probability or attribute strength of the existence of target resources in each grid. The virtual drilling module is used to take the initial belief map as the state space, determine the corresponding Markov decision process model, input the initial belief map and the Markov decision process model into a Monte Carlo tree search algorithm to perform virtual drilling on each candidate borehole location, determine the corresponding virtual observation results, update the initial belief map using a Bayesian algorithm based on the virtual observation results, determine the corresponding updated belief map, determine the expected cumulative reward of the corresponding candidate borehole based on the updated belief map and the Markov decision process model, output one or more candidate boreholes with the highest expected cumulative reward, and determine the corresponding drilling layout scheme based on the output candidate boreholes.

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 virtual drilling data processing method according to 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 virtual drilling data processing method according to any one of claims 1 to 7.

Images