Non-metallic ore grade block segment adaptive correction method and system driven by while-drilling feedback

Abstract

The application discloses a non-metallic ore grade block segment adaptive correction method and system driven by while-drilling feedback, and relates to the field of mine digital modeling.The method comprises the following steps: obtaining a historical three-dimensional block segment grade model and while-drilling multi-source sensing data collected in real time during drilling; determining a target block segment sequence to be crossed based on a drilling space trajectory; and obtaining a while-drilling grade estimation profile through a response inversion model; then, block segment observation grades are obtained by aggregation according to the spatial attribution relationship between depth positions and block segments; further, spatial anisotropic variance constraints are formed by combining spatial proximity relationships, lithology heterogeneity degrees and prior uncertainty parameters, adaptive update weights of each target block segment are determined, and differences between block segment observation grades and prior grade means are assimilated to obtain posterior grade parameters, and then a corrected block segment grade model of a current mining period is generated. Thus, while-drilling dynamic correction of the non-metallic ore grade model can be realized, and the block segment grade estimation precision and the model update reliability are improved.

Description

Technical Field

[0001] This application relates to the field of digital modeling technology for mines, and in particular to a method and system for adaptive correction of non-metallic mineral grade blocks driven by drilling feedback. Background Technology

[0002] With the development of intelligent mining and digital twin technology, 3D geological modeling has become a crucial foundation for mine resource evaluation, mining engineering design, and production organization and management. In the digital modeling and resource estimation of non-metallic mines, it is typically necessary to construct 3D block grade models based on preliminary exploration borehole, core sampling data, and geological interpretation results. Different spatial blocks are then assigned corresponding grade attributes using spatial estimation methods such as kriging interpolation and inverse distance weighting. These block models provide fundamental data support for orebody delineation, reserve estimation, mining planning, and ore quality control.

[0003] However, traditional block grade models primarily rely on static data generated during the exploration phase. Influenced by the number of exploration boreholes, borehole spacing, sampling resolution, and uncertainties in geological interpretation, initial models often fail to fully reflect the true grade variations within a local space. This is especially true for non-metallic deposits with well-developed bedding, rapidly changing lithology, and strong heterogeneity in grade distribution; static models are prone to estimation biases in local enrichment zones, depletion zones, and near orebody boundaries. Even when the model is corrected during the production phase using data from blast hole analysis, supplementary boreholes, or local sampling, such corrections typically rely on offline detection, manual judgment, and periodic updates, making it difficult to adapt promptly to the constantly changing geological conditions during mining operations.

[0004] In recent years, technologies such as online sensing, measurement while drilling, geophysical exploration, and data assimilation have been gradually introduced into dynamic mine modeling scenarios to improve the consistency between resource models and field conditions. These technologies can enrich field data sources to some extent and shorten the time interval between data acquisition and model updates. However, in practical applications, mine field data is complex and subject to numerous interference factors. Furthermore, the model update process is still limited by data quality, update efficiency, and geological adaptability. Consequently, the updated grade model still struggles to reliably meet the refined application requirements for resource evaluation, production scheduling, and ore quality control in dynamic mining scenarios. Summary of the Invention

[0005] This application provides a method, system, storage medium, computer program product, and electronic device for adaptive correction of non-metallic mineral grade blocks driven by drilling feedback, which at least solves the problem that static grade models in the prior art are difficult to adapt to the dynamic changes of local geological conditions during the mining process in a timely, accurate, and geologically reasonable manner.

[0006] In a first aspect, embodiments of this application provide a non-metallic mineral grade block adaptive correction method driven by drilling feedback. The method includes: acquiring a historical three-dimensional block grade model of a target non-metallic mineral area and real-time drilling multi-source sensor data collected during drilling operations; the historical three-dimensional block grade model includes multiple spatial blocks and the spatial range, lithological properties, and prior grade parameters corresponding to each spatial block, the prior grade parameters including a prior grade mean and a prior uncertainty parameter used to characterize the uncertainty of the prior grade mean; the drilling multi-source sensor data includes the spatial trajectory of the current borehole and multi-dimensional physical response signals collected along the drilling depth; determining the target block sequence traversed by the current borehole based on the intersection relationship between the spatial trajectory and the spatial range of each spatial block, and inputting the multi-dimensional physical response signals into a preset response inversion model to obtain a drilling grade estimation profile distributed along the current borehole depth; and determining the target block sequence according to the depth position in the drilling grade estimation profile. The spatial attribution relationship between columns is used to aggregate the drilling grade estimates falling into the same target block to obtain the block observation grade of each target block. Based on the spatial range, the spatial proximity relationship between adjacent spatial blocks is determined, and based on the lithological properties, the degree of lithological heterogeneity between adjacent spatial blocks is determined. Spatial heteroscedasticity constraints are formed based on the spatial proximity relationship, the degree of lithological heterogeneity, and the prior uncertainty parameter. Adaptive update weights for each target block are determined based on these spatial heteroscedasticity constraints. The spatial heteroscedasticity constraints are used to adjust the intensity of grade assimilation between adjacent spatial blocks. The adaptive update weights are used to assimilate the observed difference between the block observation grade and the corresponding prior grade mean to obtain the posterior grade parameters of each target block. The historical three-dimensional block grade model is updated based on the posterior grade parameters to generate a corrected block grade model for the current mining cycle. The posterior grade parameters include the posterior grade mean and the corresponding posterior uncertainty parameter.

[0007] Secondly, embodiments of this application provide a non-metallic mineral grade block adaptive correction system driven by drilling feedback. The system includes: a model data acquisition unit, used to acquire a historical three-dimensional block grade model of the target non-metallic mineral area and real-time drilling multi-source sensor data collected during drilling operations; the historical three-dimensional block grade model includes multiple spatial blocks and the spatial range, lithological properties, and prior grade parameters corresponding to each spatial block, the prior grade parameters including a prior grade mean and a prior uncertainty parameter used to characterize the uncertainty of the prior grade mean, the drilling multi-source sensor data including the spatial trajectory of the current borehole and multi-dimensional physical response signals collected along the drilling depth; a drilling inversion matching unit, used to determine the target block sequence traversed by the current borehole according to the intersection relationship between the spatial trajectory and the spatial range of each spatial block, and input the multi-dimensional physical response signals into a preset response inversion model to obtain a drilling grade estimation profile distributed along the current borehole depth; and a block observation aggregation unit, used to aggregate the block observations according to the depth positions in the drilling grade estimation profile and the... The spatial attribution relationship between target block sequences is used to aggregate the drilling grade estimates falling into the same target block to obtain the block observation grade of each target block. A heteroscedasticity weighting unit is used to determine the spatial proximity relationship between adjacent spatial blocks based on the spatial range, and to determine the degree of lithological heterogeneity between adjacent spatial blocks based on the lithological properties. Spatial heteroscedasticity constraints are formed based on the spatial proximity relationship, the degree of lithological heterogeneity, and the prior uncertainty parameter, and the weighting of each target block is determined according to the spatial heteroscedasticity constraints. The adaptive update weights are used to adjust the intensity of grade assimilation between adjacent spatial blocks. The model assimilation correction unit is used to assimilate the observation difference between the observed grade of the block and the corresponding prior grade mean using the adaptive update weights to obtain the posterior grade parameters of each target block, and to update the historical three-dimensional block grade model based on the posterior grade parameters to generate the corrected block grade model for the current mining cycle. The posterior grade parameters include the posterior grade mean and the corresponding posterior uncertainty parameters.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the drilling feedback-driven adaptive correction method for non-metallic mineral grade blocks according to any embodiment of the present application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the non-metallic mineral grade block adaptive correction method driven by drilling feedback according to any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the non-metallic mineral grade block adaptive correction method driven by drilling feedback according to any embodiment of this application.

[0011] The non-metallic mineral grade block adaptive correction method and system driven by drilling feedback provided in this application can achieve at least the following technical effects: By establishing a correspondence between the current borehole spatial trajectory and spatial segments in the historical 3D segment grade model, and converting the depth-oriented grade information obtained from multi-source sensing data during drilling into segment-level observed grades, the continuously acquired field responses during drilling can be mapped to specific segment units. Thus, the drilling observation data can form grade observations that match the spatial scale of the segment model, providing feedback with clear spatial attribution for subsequent model corrections. Since this process focuses on the segments actually traversed by the borehole, it improves the accuracy of the correspondence between field sensing data and segment grade attributes, allowing model corrections to be more focused on the local areas affected by the current drilling exposure.

[0012] Furthermore, spatial heteroscedasticity constraints are formed based on spatial proximity, lithological heterogeneity, and prior uncertainty parameters, and adaptive update weights are determined accordingly. This allows the grade correction process for blocks to be simultaneously regulated by spatial continuity, geological differences, and prior confidence. For blocks that are spatially close and have similar lithology, the grade assimilation effect can be moderately transmitted; for areas with significant lithological differences or weak geological continuity, unreasonable cross-boundary diffusion can be suppressed. Meanwhile, blocks with higher prior uncertainty can more fully absorb drilling observation information, while blocks with higher prior confidence remain relatively stable. Thus, the model update can achieve differentiated adjustments based on the geological conditions of the block itself and its surrounding area, improving the geological adaptability and stability of the corrected grade distribution.

[0013] Overall, using feedback while drilling as the trigger and spatial heteroscedasticity constraints as the correction and control mechanism, real-time field observations, historical block models, lithological differences, and uncertainty expressions are unified into the adaptive update process of block grade. This allows for constrained dynamic correction of local grade deviations within the current mining area while maintaining the overall continuity of the historical model. This results in a corrected block grade model that better reflects the geological conditions of the current mining cycle, providing a more reliable model foundation for dynamic evaluation of non-metallic mineral resources, mining organization, and ore quality control. Attached Figure Description

[0014] 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.

[0015] Figure 1 A flowchart is shown as an example of a nonmetallic mineral grade block adaptive correction method driven by drilling feedback according to an embodiment of this application; Figure 2 The flowchart illustrates an example of the operation of the method according to an embodiment of this application, which outputs the grade prediction values ​​of each discrete depth node through a preset response inversion model. Figure 3 The flowchart illustrates an example of determining the adaptive update weights of each target block based on spatial heteroscedasticity constraints in a method according to an embodiment of this application. Figure 4 This paper presents an example of an operation flowchart for target borehole position feedback scheduling based on an information state diagram in a method according to an embodiment of the present application. Figure 5 A schematic diagram illustrating the system operation mechanism of an example of a non-metallic mineral grade block adaptive correction method driven by drilling feedback according to an embodiment of this application is shown. Figure 6 This illustration shows a comparative experimental simulation effect of an example of the three-dimensional block evolution of grade at different renewal stages when the method of the present application is applied to a non-metallic mining area; Figure 7 A simulation diagram illustrating the overall performance comparison of different methods and an example of the convergence process during drilling updates is shown. Figure 8 A comparative experimental simulation diagram shows an example of grade-tonnage evaluation of non-metallic mineral deposits using different methods based on different block models; Figure 9 A structural block diagram of an example of a nonmetallic mineral grade block adaptive correction system driven by drilling feedback according to an embodiment of this application is shown. Detailed Implementation

[0016] 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.

[0017] It should be noted that, in terms of the specific implementation of spatial grade estimation, current technologies, when constructing or revising block grade models, typically still rely on the test results of discrete sampling points as the primary data foundation, and combine spatial estimation methods such as IDW (Inverse Distance Weighting) or Kriging to infer block attributes. Among these, the IDW method is relatively simple to calculate, but the estimation results are easily affected by the spatial distribution of sampling points, the neighborhood range, and local anomalous samples; the Kriging method can incorporate spatial variation characteristics for estimation, but it is highly dependent on the construction of the variogram, the number of samples, and assumptions about spatial structure. In scenarios where non-metallic deposits exhibit significant stratigraphic fluctuations, interlayer development, local enrichment, and lithological abrupt changes, the above methods often struggle to balance overall continuity with the accuracy of local boundary representation.

[0018] During the production phase, some mines will revise existing grade models by combining blast hole analysis, local sampling, or supplementary borehole data. This method can supplement the deficiencies of previous exploration data to a certain extent, but it usually still requires processes such as sample testing, data processing, identification of discrepancies, and local re-estimation. For areas with relatively gentle grade changes, complex transition relationships between ore bodies and surrounding rocks, or uneven distribution of sampling points, the scope and intensity of correction are often difficult to determine stably, resulting in the model update process still being biased towards periodicity, localization, and empirical methods.

[0019] Regarding the dynamic updating of mine resource models, some studies have proposed ideas such as Real-Time Mining (RTM), attempting to utilize online sensors, production process data, and data assimilation methods to enable resource models to be gradually corrected as the mining process progresses. This research has driven the development of mine modeling from static estimation to dynamic updating; however, its application in engineering is still limited by the quality of field data, sensor stability, and model adaptability. Especially in non-metallic deposits, the layered structure of the ore body, the alternation of hard and soft layers, local heterogeneity, and factors such as dust, moisture, and vibration in the stope can all affect the correspondence between field observation data and grade attributes.

[0020] In terms of on-site detection and space exploration, technologies such as HSI (Hyperspectral Imaging), LIBS (Laser-Induced Breakdown Spectroscopy), 3D laser scanning, magnetic detection, surface wave detection, and electromagnetic detection have been used for mineral identification, elemental analysis, identification of concealed geological bodies, and reconstruction of goaf morphology. These technologies can provide supplementary data sources for mine modeling, but in practical applications, they are easily affected by factors such as sample surface condition, dust and water bodies, obstruction conditions, equipment calibration, construction and deployment costs, and the ambiguity of inversion interpretation. Overall, these technologies are more often used for rapid sample analysis, auxiliary geological interpretation, or safety monitoring, and are not yet capable of independently undertaking the task of continuously updating grade attributes during production.

[0021] From the perspective of the construction needs of smart mines and digital twin systems, a closer data linkage is needed between orebody models, mining conditions, equipment operating parameters, and production decisions. While current technologies have established a certain foundation in data acquisition, orebody modeling, on-site detection, and local model updates, issues such as scale inconsistencies, discontinuous spatial correspondences, insufficient update timeliness, and the inability of update results to effectively serve subsequent operational decisions still exist between different data sources and block models. Therefore, the dynamic application of non-metallic mine grade models still has room for further improvement.

[0022] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0023] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0024] Figure 1 A flowchart illustrating an example of a nonmetallic mineral grade block adaptive correction method driven by drilling feedback according to an embodiment of this application is shown.

[0025] Regarding the execution entity of the method in this application embodiment, it can be a mine digital twin platform, a mining production control platform, an edge computing device, a server, or a data processing system composed of the above devices. It is applicable to mining operations in non-metallic mineral areas such as phosphate, kaolin, limestone, and potash, and is particularly suitable for mining areas where the ore body is layered, has many interlayers, irregular boundaries between mine waste and other materials, or significant local grade variations. By converting real-time multi-source sensor data collected during drilling into observational information that can be used to correct the block model, and combining this information with the spatial range, lithological properties, and prior uncertainty in the historical three-dimensional block grade model, the grade parameters of the target block are adaptively updated, enabling the grade block model to reflect newly revealed ore body information within the current mining cycle.

[0026] like Figure 1 As shown, in step S110, the historical three-dimensional block grade model of the target non-metallic mining area and the real-time multi-source sensor data collected during drilling operations are obtained.

[0027] Here, the historical three-dimensional block grade model includes multiple spatial blocks and the corresponding spatial range, lithological properties and prior grade parameters of each spatial block. The prior grade parameters include the prior grade mean and the prior uncertainty parameter used to characterize the uncertainty of the prior grade mean. The drilling multi-source sensor data includes the spatial trajectory of the current borehole and the multi-dimensional physical response signal collected along the drilling depth.

[0028] In some implementations, the historical three-dimensional block grade model can be a three-dimensional resource model established based on previous geological exploration data, borehole core sampling data, geological mapping results, laboratory test results, and production exposure data. For example, in open-pit phosphate or kaolin mining operations, during the early exploration phase, core samples can typically only be obtained at certain borehole intervals, and the grade of ore bodies at different spatial locations can be estimated based on the test results. Thus, the resulting historical three-dimensional block grade model can reflect the overall resource distribution trend of the mining area, but significant uncertainties may exist at local waste boundaries, interlayer locations, thin ore bodies, or local enrichment areas. Therefore, in this embodiment, the historical three-dimensional block grade model serves as a priori model for drilling-while-drilling corrections during the current mining cycle.

[0029] The historical 3D block grade model includes multiple spatial blocks and their corresponding spatial extent, lithological properties, and prior grade parameters. The spatial extent defines the geometric boundary or coverage area of ​​the spatial block within the target non-metallic mining area's 3D coordinate system. This can be represented by the block's center location, block size, top and bottom surface extent, or 3D grid index. Lithological properties characterize the rock type, ore layer category, stratigraphic characteristics, interlayer properties, or mineral composition characteristics corresponding to the spatial block. Prior grade parameters represent the grade estimation state of the spatial block before the introduction of drilling feedback data from the current drilling operation. These parameters include the prior grade mean and a prior uncertainty parameter characterizing the uncertainty of the prior grade mean. For areas with dense historical sampling and continuous lithology, the prior uncertainty parameter can be relatively small; for areas with sparse historical sampling, complex ore body boundaries, or significant lithological variations, the prior uncertainty parameter can be relatively large.

[0030] Multi-source sensing data while drilling can be data acquired in real-time or near real-time during the current drilling operation. This multi-source sensing data includes the spatial trajectory of the current borehole and multi-dimensional physical response signals acquired along the drilling depth. Specifically, the spatial trajectory of the current borehole is used to characterize the borehole's advancement path in three-dimensional space and can be determined by drilling rig positioning information, inclination measurement information, drilling depth information, or trajectory measurement information. Furthermore, the multi-dimensional physical response signals are used to characterize the physical response changes of the current borehole location and its adjacent rock and ore media during the drilling process. These signals can reflect changes in ore composition, lithology, structural differences, or grade-related response characteristics. This allows historical static models and real-time sensing data from the current drilling process to be acquired and organized within the same data processing framework, thereby improving the completeness and timeliness of the input data.

[0031] In step S120, based on the intersection relationship between the spatial trajectory and the spatial range of each spatial block, the sequence of target blocks traversed by the current borehole is determined, and the multidimensional physical response signal is input into the preset response inversion model to obtain the drilling grade estimation profile distributed along the current borehole depth.

[0032] In some implementations, the spatial trajectory of the current borehole can be mapped to a unified three-dimensional coordinate system containing the historical three-dimensional block quality model, and it can be determined whether the spatial trajectory intersects with the spatial range of each spatial block. For example, each spatial block can be considered as a volume element with a defined three-dimensional boundary, and the spatial trajectory of the current borehole can be considered as a spatial curve extending with the drilling depth. When the spatial curve enters, passes through, or traverses the spatial range of a certain spatial block, that spatial block can be identified as the target block traversed by the current borehole. Furthermore, the traversed target blocks can be sorted according to the drilling direction, drilling depth sequence, or the order of passage on the spatial trajectory to obtain a target block sequence.

[0033] For non-metallic mining areas, ore bodies may be distributed in inclined layers, and borehole trajectories may be deviated due to drilling posture, bench location, or construction conditions. If drilling data is assigned to blocks solely based on vertical depth or planar projection, observation information may be incorrectly allocated to adjacent layers, interlayers, or waste rock blocks. This embodiment determines the target block sequence by identifying the intersection relationship between the spatial trajectory and the spatial block range. This allows the spatial location revealed by the current borehole to correspond to specific block units in the historical 3D block grade model, thereby improving the accuracy of spatial matching.

[0034] Simultaneously, multidimensional physical response signals collected along the drilling depth can be input into a preset response inversion model to obtain a grade estimation profile distributed along the current borehole depth. The preset response inversion model is used to characterize the mapping relationship between multidimensional physical response signals and non-metallic mineral grades. Since the physical response signals collected during drilling typically reflect the indirect response characteristics of the ore-rock medium rather than direct grade values, it is necessary to convert this type of physical response information into estimation results that can be used for grade characterization through the response inversion model. This allows the spatial exposure location of the current borehole and the grade estimation results along the borehole to be determined simultaneously, thus forming a grade profile expression corresponding to the borehole trajectory.

[0035] In step S130, according to the spatial attribution relationship between each depth position in the drilling grade estimation profile and the target block sequence, the drilling grade estimation values ​​falling into the same target block are aggregated to obtain the block observation grade of each target block.

[0036] In real-world drilling scenarios, the grade estimation profile while drilling (WWD) is typically a continuous or near-continuous profile data distribution along the borehole depth, and its sampling scale usually differs from the block scale of the 3D block model. For example, the WWD grade estimation profile can generate multiple grade estimates at small depth intervals, while each spatial block in the historical 3D block grade model typically corresponds to a 3D volume unit. Directly using the grade estimate at a single depth location to represent the entire spatial block is susceptible to local noise, short-segment anomalies, or sensor fluctuations; simply ignoring fine-grained variations within the profile reduces the usability of the WWD feedback data.

[0037] Therefore, based on the three-dimensional spatial location corresponding to each depth position in the drilling grade estimation profile, it can be determined which target block in the target block sequence each depth position belongs to. For multiple drilling grade estimates belonging to the same target block, aggregation processing can be performed to obtain the block observation grade corresponding to that target block. This aggregation processing can adopt statistical methods that can characterize the overall observation state of the block, or it can be comprehensively determined by combining depth coverage, observation stability, or effective observation ratio. Through this processing, the profile-level grade estimation results distributed along the borehole are converted into block-level observations that match the three-dimensional block model, thereby solving the problem of spatial scale inconsistency between the drilling profile data and the three-dimensional block model, and improving the representativeness of the block observation grades. Thus, by aggregating multiple drilling grade estimates within the same target block, the impact of single-point fluctuations on the block observation results can be reduced, making the obtained block observation grades more suitable as the observation basis for block-level model correction.

[0038] In step S140, the spatial proximity relationship between adjacent spatial blocks is determined based on the spatial range, and the degree of lithological heterogeneity between adjacent spatial blocks is determined based on lithological properties. Spatial heteroscedasticity constraints are formed based on spatial proximity relationship, degree of lithological heterogeneity and prior uncertainty parameters, and adaptive update weights of each target block are determined based on spatial heteroscedasticity constraints. The spatial heteroscedasticity constraints are used to adjust the intensity of grade assimilation influence between adjacent spatial blocks.

[0039] In some implementations, the adjacency, distance, or spatial connectivity relationships between blocks can be determined based on their spatial extent, thereby obtaining the spatial proximity relationships between adjacent blocks. This reflects the degree of proximity and potential spatial correlation between different blocks in the three-dimensional mining area. Simultaneously, based on the lithological properties of each block, differences in lithological type, ore layer properties, stratigraphic position, or mineral composition between adjacent blocks can be compared, thereby determining the degree of lithological heterogeneity between adjacent blocks. This reflects whether there are significant geological differences, changes in mine-waste boundaries, or abrupt changes in stratigraphic positions between adjacent blocks.

[0040] In non-metallic mineral areas, close spatial proximity does not necessarily imply continuous grade variation. For example, in layered phosphate deposits, adjacent blocks may be located in ore layers and interbedded rock layers, respectively; in kaolin deposits, adjacent blocks may exhibit significant grade differences due to variations in alteration, impurity content, or mineral composition. If update weights are determined solely based on spatial distance, drilling observation information from one block can unreasonably influence adjacent blocks with significant lithological differences. Therefore, this embodiment uses spatial proximity, lithological heterogeneity, and prior uncertainty parameters together to form spatial heteroscedasticity constraints. This allows blocks that are spatially close, have small lithological differences, and high prior uncertainty to receive a stronger assimilation effect, while blocks that are spatially distant, have large lithological differences, or weak geological continuity are subject to a weaker assimilation effect.

[0041] An adaptive update weight, determined based on spatial heteroscedasticity constraints, is used to control the degree of correction of the observed grade of a block to the corresponding prior grade parameters. This weight is not a fixed empirical coefficient, but rather dynamically determined based on the spatial relationships, lithological differences, and prior uncertainties of the spatial blocks within the current mining area model. This allows the assimilation weight to adapt to the heterogeneous distribution characteristics of non-metallic ore bodies, reduces the risk of excessive smoothing across lithological boundaries, and improves the responsiveness of high-uncertainty blocks to current drilling observation information.

[0042] In step S150, the observation difference between the observed grade of the block and the corresponding prior grade mean is assimilated using adaptive update weights to obtain the posterior grade parameters of each target block. Based on the posterior grade parameters, the historical three-dimensional block grade model is updated to generate the corrected block grade model for the current mining cycle. The posterior grade parameters include the posterior grade mean and the corresponding posterior uncertainty parameters.

[0043] Specifically, the observational difference between the observed grade of a block and its corresponding prior mean grade can be determined first. This difference characterizes the deviation between the block grade status revealed by the current drilling feedback data and the prior estimates in the historical 3D block grade model. Subsequently, an adaptive update weight is used to assimilate this observational difference, ensuring that the updated block grade parameters reflect both the new information provided by current drilling observations and the prior geological knowledge contained in the historical 3D block grade model. Therefore, the system does not simply replace the prior mean grade with the observed block grade, but rather determines the correction magnitude based on the adaptive update weight under spatial heteroscedasticity constraints.

[0044] In some implementations, the posterior grade parameters obtained after assimilation and update may include the posterior grade mean and the corresponding posterior uncertainty parameter. The posterior grade mean is used to characterize the grade estimation result of the target block after incorporating the current drilling observation data, and the posterior uncertainty parameter is used to characterize the degree of uncertainty of the grade estimation result after assimilation and update. For example, when the drilling observation information has good consistency with the prior model and the corresponding area has strong lithological continuity, the posterior grade mean can be stably corrected near the prior grade mean, and the posterior uncertainty parameter can be reduced accordingly; when the drilling observation information differs greatly from the prior model or the corresponding area has a high degree of lithological heterogeneity, the posterior uncertainty parameter can retain the corresponding uncertainty expression to reduce the risk of overcorrection caused by a single observation.

[0045] Furthermore, the posterior grade parameters of each target block can be written into the corresponding spatial block location in the historical 3D block grade model to generate a corrected block grade model for the current mining cycle. This model retains the spatial block structure of the historical 3D block grade model while incorporating local observation information provided by multi-source sensor data during the current drilling operation. This allows the grade parameters of the target blocks to be dynamically corrected while maintaining the stability of the prior model, thereby improving the accuracy and timeliness of block grade expression within the current mining cycle.

[0046] In some examples of embodiments of this application, the multidimensional physical response signals acquired along the drilling depth include gamma-ray signals while drilling, acoustic logging signals, and cuttings laser spectral signals.

[0047] In practice, the grade variation of ore bodies in non-metallic mining areas is typically influenced by factors such as lithology, mineral composition, interlayer distribution, pore structure, and local weathering or alteration. A single type of physical response signal during drilling is often insufficient to comprehensively characterize changes in ore and rock properties. Therefore, complementary multidimensional physical response signals can be collected during drilling operations. Gamma-ray signals during drilling can reflect differences in the natural radioactivity response of formations, thus aiding in the identification of variations in different lithological layers or impurity content; sonic logging signals can reflect the compactness, fracture development, and mechanical structure differences of the ore and rock medium; and laser spectral signals from drilling cuttings can reflect the elemental composition or spectral intensity distribution in the drilling cuttings. By using data from these different response dimensions as input for grade inversion, the ability to characterize the complex lithology and grade variations in non-metallic mining areas can be improved.

[0048] Regarding the implementation details of obtaining the drilling grade estimation profile through inversion, firstly, real-time drilling dynamics parameters of the current borehole during drilling operations are obtained. These real-time drilling dynamics parameters include at least the instantaneous drilling speed and the drill bit attitude angle. The instantaneous drilling speed is used to characterize the actual advance state of the drill bit per unit time, and the drill bit attitude angle is used to characterize the degree of deviation of the drill bit's advance direction from the preset reference direction. During actual drilling, the drill bit advance speed may fluctuate with changes in rock hardness, fracture development, and drilling conditions, and the drill bit attitude may also deviate due to formation anisotropy, drill string condition, or construction conditions. Therefore, incorporating real-time drilling dynamics parameters into the processing can make the subsequent depth registration results more consistent with the actual spatial advance state of the borehole.

[0049] Then, the multidimensional physical response signal is time-synchronized, and based on real-time drilling dynamics parameters, the synchronized multidimensional physical response signal undergoes time-depth mapping processing to transform the time-series response signal into a unified spatial depth domain composed of multiple discrete depth nodes. Specifically, the time-depth mapping processing determines the equivalent spatial depth corresponding to each acquisition moment based on the cumulative advance of the drilling speed over time and the projection effect of the drill bit attitude inclination angle on the advance direction. The equivalent spatial depth is used to characterize the projected depth of the current borehole along a preset reference direction.

[0050] For example, the time of data acquisition after synchronization Corresponding equivalent space depth Satisfy the following formula: Equation (1) In the formula, For a moment Instantaneous drilling speed, For a moment The drill bit's tilt angle.

[0051] The aforementioned time-depth mapping is used to convert sensor response signals indexed by acquisition time into depth domain data indexed by spatial depth. Since the drill bit's advance speed may vary across different acquisition periods, if the physical response signals are arranged solely according to the acquisition time interval, the actual spatial distances between adjacent sampling points may not be consistent. Furthermore, when the drill bit's attitude has an inclination angle, the actual advance distance of the drill bit also differs from the projected depth along a preset reference direction. Therefore, by accumulating the instantaneous drilling speed and drill bit attitude inclination angle, the equivalent spatial depth corresponding to each acquisition moment can be determined more reasonably, enabling drilling-while-drilling physical response signals from different sources to be mapped into a unified spatial depth domain.

[0052] In engineering implementation, time-depth mapping can be calculated based on discrete sampling sequences. Assume the time sampling interval of the data acquisition system is... , No. Discrete sampling time Corresponding equivalent space depth It can be determined by the following discrete accumulation method: Equation (2) In the formula, For discrete-time sampling point index, For the first The instantaneous drilling speed at each sampling moment, For the first The drill bit attitude inclination angle at each sampling moment.

[0053] Subsequently, to address the depth resolution differences caused by the different detection principles of multi-source sensors, spatial resampling, window aggregation, and scale normalization were performed on each physical response signal mapped to a unified spatial depth domain, constrained by a preset benchmark sampling interval and the effective detection window corresponding to each type of physical response signal. This combined the signals to form multi-source feature vectors corresponding to each discrete depth node. The multi-source feature vectors include at least the normalized count rate extracted from the gamma-ray signal during drilling, the amplitude attenuation gradient extracted from the sonic logging signal, and the spectral element intensity distribution extracted from the cuttings laser spectral signal.

[0054] In practice, different types of sensors may have different installation positions, sampling frequencies, detection ranges, and response delays on the drilling tool. For example, some sensors respond to the mineral and rock media within a certain volume around the borehole, while others mainly reflect the characteristics of samples or cuttings at more local locations. If different physical response signals are directly spliced ​​together according to their original timestamps, the actual spatial locations corresponding to different modal data in the same feature vector may be inconsistent. Therefore, discrete depth nodes can be divided first using a preset benchmark sampling interval, and then the response data within the corresponding depth range can be windowed and aggregated according to the effective detection windows of various physical response signals. Furthermore, features with different dimensions and numerical ranges can be scaled and normalized. Through this process, multi-source feature vectors that are spatially aligned and comparable in numerical scale can be formed.

[0055] Then, the multi-source feature vectors corresponding to each discrete depth node are input into the preset response inversion model according to the spatial sequence to output the grade prediction value of each discrete depth node. The grade estimation profile while drilling is formed by each discrete depth node and the corresponding grade prediction value.

[0056] After completing depth domain registration and multi-source feature vector construction, each discrete depth node has a corresponding comprehensive physical response expression. The system can input the multi-source feature vectors corresponding to each discrete depth node into a preset response inversion model according to the spatial order of the borehole depth direction. Based on the physical response information contained in the multi-source feature vectors, the preset response inversion model predicts the non-metallic mineral grade of the corresponding discrete depth node and outputs the corresponding grade prediction value. By combining each discrete depth node and its grade prediction value according to the borehole depth direction, a drilling grade estimation profile distributed along the current borehole depth can be obtained.

[0057] Through the embodiments of this application, the multi-source physical response signals acquired in the time domain are converted into multi-source feature expressions in a unified spatial depth domain, and further form grade estimation results corresponding to the borehole depth position. This can reduce the impact of drilling speed fluctuations, drill bit attitude deviations and differences in depth resolution of different sensors on the grade estimation profile, so that the generated grade estimation profile while drilling has good consistency in spatial position and physical response meaning.

[0058] In some examples of embodiments of this application, the multidimensional physical response signals acquired along the drilling depth include gamma-ray signals while drilling, acoustic logging signals, and cuttings laser spectral signals.

[0059] It should be noted that different types of physical response signals while drilling differ in their physical meaning, data scale, noise characteristics, and sensitivity to ore and rock properties. For example, gamma-ray signals while drilling can reflect differences in the natural radioactivity response of the formation, sonic logging signals can reflect the compactness, fracture development, and acoustic propagation characteristics of the ore and rock medium, and cuttings laser spectral signals can reflect the elemental composition or spectral intensity distribution in the drilled cuttings. If physical response signals from different sources are simply spliced ​​together and input into the same feature processing structure, it may be difficult to distinguish the information contributions between different modes, thus affecting the stability of the grade prediction results. Therefore, in this embodiment, different types of physical response signals can be divided into different physical sensing modes, and corresponding feature extraction channels can be configured for different physical sensing modes.

[0060] In some embodiments, the preset response inversion model includes multiple feature extraction channels, an attention allocation unit, and a grade output unit. Here, each feature extraction channel corresponds to a physical sensing mode, and the multiple physical sensing modes include the gamma response mode corresponding to the drilling gamma-ray signal, the acoustic response mode corresponding to the acoustic logging signal, and the spectral response mode corresponding to the cuttings laser spectrum signal. For different physical sensing modes, corresponding feature extraction structures can be set according to their data characteristics. For example, for signals with continuously varying characteristics along the depth direction, a network structure suitable for sequence feature extraction can be used; for signals with spectral line distribution characteristics, a network structure suitable for high-dimensional feature mapping can be used. In this way, various physical response signals can obtain feature expressions adapted to their physical properties before entering the fusion processing.

[0061] Figure 2 The diagram illustrates an example of an operation flowchart in which the grade prediction values ​​of each discrete depth node are output through a preset response inversion model according to an embodiment of the present application.

[0062] In step S210, for the multi-source feature vector corresponding to any discrete depth node, modal sub-features corresponding to each physical sensing modality are determined, and the multi-source feature vector is input into the attention allocation unit to determine the attention weights corresponding to each physical sensing modality. Each attention weight is used to characterize the degree of contribution of the corresponding physical sensing modality to the grade prediction result.

[0063] In actual drilling, the effectiveness of different physical sensing modes may vary at different geological locations. For example, when the borehole passes through strata with well-developed fractures or high degrees of fracturing, acoustic response signals may be affected by attenuation or scattering; when rock cuttings samples are affected by contamination or transport lag, the local stability of spectral response signals may decrease; while other physical response signals may still maintain good usability. To enable the model to adjust the contribution of different modes according to the comprehensive physical response state at the current discrete depth node, this embodiment can process multi-source feature vectors through an attention allocation unit and assign corresponding attention weights to each physical sensing mode.

[0064] In some implementations, the attention allocation unit may employ a learnable mapping structure with normalized output. For example, the first... At the discrete depth node, the first The attention weights corresponding to each physical perception modality can satisfy: Equation (3) In the formula, The first input attention allocation unit Multi-source feature vectors corresponding to discrete depth nodes and The attention allocation unit and the first Learnable parameters corresponding to each physical sensing mode This represents the vector transpose operation. To iterate through the summation index of each physical sensing mode, and They are respectively with the first Learnable parameters corresponding to each physical sensing mode It is an exponential function. This represents the number of physical sensing modes.

[0065] In step S220, each modal sub-feature is input into its corresponding feature extraction channel for nonlinear feature extraction. The modal features output from each feature extraction channel are then weighted and fused based on their respective attention weights. The grade prediction value for any discrete depth node is then output by the grade output unit. The grade prediction value is obtained by fusing the nonlinear feature results of each physical perception modality according to the normalized attention contribution relationship.

[0066] For example, the grade output unit outputs the first grade. The predicted grade value corresponding to a discrete depth node can be expressed by the following formula: Equation (4) In the formula, For the first Among the discrete depth nodes, the one with the first Modal sub-features corresponding to each physical sensing mode. For the first Grade prediction value corresponding to each discrete depth node. For the first The nonlinear feature extraction function corresponding to each feature extraction channel.

[0067] In equation (4) above, the nonlinear feature extraction function Used to the first Each physical perception modality corresponds to a modal sub-feature mapping that can participate in the grade output. Attention weights. This is used to adjust the contribution of the modal response result to the grade prediction value of the current discrete depth node. When a physical sensing modality has high stability at the current depth position or is strongly correlated with grade changes, the attention weight corresponding to that modality can be relatively large; when a physical sensing modality is affected by noise, attenuation, or local anomalies, the attention weight corresponding to that modality can be relatively small. Through this weighted fusion method, the grade output unit can integrate the response information of multiple physical sensing modalities to obtain the grade prediction value of the corresponding discrete depth node.

[0068] Here, the preset response inversion model is a model pre-trained based on a loss function with geological constraints. This loss function includes grade prediction error constraints, depth adjacent node continuity constraints, and geological response consistency constraints. Among them, the grade prediction error constraints are used to constrain the deviation between the grade prediction value and the true grade label, the depth adjacent node continuity constraints are used to adjust the smoothness of the grade prediction results according to the lithological continuity between adjacent discrete depth nodes, and the geological response consistency constraints are used to constrain the inversion results corresponding to the model parameters to be consistent with the preset lithology-grade response relationship.

[0069] In some implementations, the loss function can be expressed as follows: Equation (5) In the formula, This represents the total loss function value during model training. This represents the number of discrete depth nodes in a single training sequence. For the first The true grade label corresponding to each discrete depth node. For the first Grade prediction value corresponding to each discrete depth node. and For preset weighting coefficients, For characterizing the first The discrete depth node and the first The continuity constraint coefficient for the degree of lithological continuity between discrete depth nodes. To preset the model parameters of the response inversion model, This is a geological constraint term used to ensure that the inversion results corresponding to the model parameters are consistent with the preset lithology and grade response relationship.

[0070] In the loss function as shown in equation (5) above, the first term is used to constrain the error between the grade prediction value output by the model and the true grade label, enabling the model to learn the correspondence between multi-source physical response signals and grades. The second term is used to constrain the variation of grade prediction results between adjacent discrete depth nodes, and its constraint strength is determined by the continuity constraint coefficient. Adjustment. For adjacent depth nodes with high lithological continuity, A relatively large value can be used to maintain good continuity between adjacent grade prediction results; for adjacent depth nodes with significant lithological differences, A relatively small value can be chosen so that the model can retain possible grade abrupt changes. The third term is used to introduce geological constraints related to the relationship between lithology and grade response, so that the model training results are not entirely dependent on the sample fitting error, but are constrained by the preset geological response relationship.

[0071] In some implementations, the continuity constraint coefficient The continuity constraint coefficient can be determined based on the differences in lithological characteristics between adjacent discrete depth nodes. For example, the continuity constraint coefficient can satisfy: Equation (6) In the formula, For the first The discrete depth node and the first Lithological characteristic differences between discrete depth nodes This is a sensitivity adjustment parameter. When the lithological characteristics of adjacent discrete depth nodes are relatively different, the continuity constraint coefficient is relatively large, and the continuity constraint of adjacent depth nodes has a strong smoothing effect on the grade prediction results; when the lithological characteristics of adjacent discrete depth nodes are relatively different, the continuity constraint coefficient is relatively small, and the suppression effect of the continuity constraint of adjacent depth nodes on grade abrupt changes is correspondingly weakened.

[0072] Through the embodiments of this application, the preset response inversion model can adaptively allocate contributions among different physical sensing modes, and simultaneously consider grade prediction error, depth direction continuity and geological response consistency during the training process. This can reduce the impact of abnormal fluctuations of a single physical sensing mode on grade prediction results, and make the grade estimation profile while drilling have good smoothness in lithological continuous areas, while retaining the corresponding grade change characteristics in lithological change areas.

[0073] In some examples of embodiments of this application, spatial heteroscedasticity constraints are quantitatively characterized by spatial heteroscedasticity kernel matrix, the prior grade mean values ​​corresponding to each target block constitute the prior mean vector, and the block observation grade of each target block constitutes the drilling observation vector.

[0074] In some implementations, the prior grade mean values ​​of each target block can be arranged according to the spatial order, drilling sequence, or preset numbering order of the target blocks in the target block sequence to form a prior grade mean vector. Simultaneously, the observed grades of each target block are arranged accordingly to form a drilling observation vector. This vectorized representation allows the prior grade states and drilling observation results of multiple target blocks to be unified in the same data structure, facilitating subsequent spatial constraint calculations and assimilation updates.

[0075] In the 3D block modeling scenario of non-metallic mining areas, different target blocks are spatially related, but may also exhibit different uncertainty characteristics due to lithological differences, stratigraphic variations, interlayer distribution, or changes in mine waste boundaries. Therefore, this embodiment uses a spatial heteroscedasticity kernel matrix to quantitatively express the spatial correlation and geological heterogeneity effects between target blocks. This spatial heteroscedasticity kernel matrix is ​​not determined solely by geometric distance, but further incorporates the geological anisotropy and lithological heterogeneity between target blocks, enabling it to reflect the characteristics of spatial correlation in non-metallic ore bodies that vary with direction, distance, and lithological differences.

[0076] Figure 3 The diagram illustrates an example of an operation flowchart for determining the adaptive update weights of each target block based on spatial heteroscedasticity constraints in a method according to an embodiment of this application.

[0077] like Figure 3 As shown, in step S310, based on the spatial range of each target block, the spatial center coordinates corresponding to each target block are determined. Combining the spatial proximity relationship and according to the geological anisotropic characteristics of the target non-metallic mining area, the spatial center coordinate difference between any two target blocks is anisotropically scaled to obtain the anisotropic equivalent distance.

[0078] In some implementations, the spatial center coordinates of the target block can be determined first based on its spatial extent. For spatial blocks with regular volumetric structures, the center coordinates can be determined based on the three-dimensional boundary range of the block; for irregular spatial blocks, the spatial center coordinates can also be determined based on their spatial geometric center, volume center, or a preset representative point. Subsequently, for any two target blocks, the difference between their spatial center coordinates can be calculated and scaled in conjunction with the geological anisotropy characteristics of the target non-metallic mining area.

[0079] In layered or vein-like nonmetallic ore bodies, the continuity of the ore body along the strike, dip, and thickness directions may vary. For example, grade changes along the extension direction of the ore layer may be relatively slow, while grade changes may be more pronounced when crossing the thickness direction of the ore layer or the direction of interlayers. If only ordinary Euclidean distance is used to measure the spatial correlation between target blocks, this directional difference may not be reflected. Therefore, a geometric anisotropy matrix can be introduced to perform directional rotation and scale adjustment on the spatial center coordinate difference between target blocks. For example, two target blocks... and Anisotropic equivalent distance between It can satisfy: Equation (7) In the formula, and The target block segments are respectively and target block The three-dimensional spatial center coordinate vector, An anisotropic scaling matrix is ​​determined based on the geological anisotropic characteristics of the target non-metallic mining area. This matrix can be determined based on the main extension direction of the ore body, the dip angle of the strata, the direction of spatial variation, or other a priori geological information, so that the spatial distance measurement is more in line with the directional continuity characteristics of the non-metallic ore body.

[0080] In step S320, based on the lithological properties of each target block and the differences in lithological properties between adjacent spatial blocks, the lithological heterogeneity index corresponding to each target block is determined; wherein, the lithological heterogeneity index is used to quantify the degree of lithological difference between the corresponding target block and the adjacent spatial blocks.

[0081] Specifically, lithological attributes can include information such as rock type, ore layer category, interlayer attributes, stratigraphic number, mineral composition type, or host rock category. The system can determine the lithological heterogeneity index corresponding to a target block based on the differences in lithological attributes between the target block and its neighboring blocks. For example, when the target block is surrounded mainly by similar lithologies or continuous ore layers, its lithological heterogeneity index can be set to a lower value; when the target block is located at the boundary between ore and host rock, near interlayers, in abrupt stratigraphic changes, or in a mixed lithological region, its lithological heterogeneity index can be set to a higher value.

[0082] The lithological heterogeneity index is used to convert geological differences into numerical parameters that can be used in calculations. This allows the subsequent spatial heteroscedasticity kernel matrix to reflect not only the spatial distance relationships between target blocks but also the complexity of the local geological environment in which the target blocks are located. Therefore, it helps to retain the corresponding uncertainty expression in areas with significant lithological variations, avoiding the simplistic assumption that adjacent blocks have the same spatial stability.

[0083] In step S330, for any two target blocks in the target block set composed of various target blocks, corresponding spatial heteroscedasticity kernel matrix elements are constructed. The spatial heteroscedasticity kernel matrix elements decrease as the anisotropic equivalent distance between the two corresponding target blocks increases, and increase as the lithological heterogeneity index of the two corresponding target blocks increases, so that the influence of spatial distance and lithological difference jointly participate in the grade assimilation constraint.

[0084] For example, for any two target blocks in the set of target blocks composed of each of the aforementioned target blocks... and Construct the corresponding spatial heteroscedasticity kernel matrix elements : Equation (8) In the formula, For the target block segment With the target block segment The anisotropic equivalent distance between them Distance parameters are spatially correlated. and The target block segments are respectively and the target block segment The corresponding lithological heterogeneity index, after being scaled by grade standard deviation, and satisfying... and .

[0085] In equation (8) above, the exponential term characterizes the tendency of spatial correlation to decrease with increasing anisotropic equivalent distance. When two target blocks are close in spatial distance after geological anisotropy scaling, their spatial heteroscedasticity kernel matrix elements can be relatively large; when the anisotropic equivalent distance between them is large, the kernel matrix elements can be relatively small. The lithological heterogeneity index is used to adjust the influence of local geological differences on the expression of spatial heteroscedasticity. When the target block is located in an area with significant lithological changes, its corresponding lithological heterogeneity index can make the spatial heteroscedasticity kernel matrix elements reflect stronger uncertainty fluctuations. Through this construction method, the spatial heteroscedasticity kernel matrix can simultaneously express the effects of spatial distance attenuation and lithological differences.

[0086] In step S340, based on the spatial heteroscedasticity kernel matrix, the prior covariance matrix composed of the prior uncertainty parameters of each target block, the observation error covariance matrix composed of the preset observation error covariance, and the observation operator matrix, the target assimilation weight matrix, which serves as the adaptive update weight, is solved. The observation operator matrix is ​​used to map the grade parameters of the target block to the observation space where the drilling observation vector resides, and the target assimilation weight matrix is ​​used to adjust the correction strength of the drilling observation vector to the prior mean vector under spatial heteroscedasticity constraints and observation error constraints.

[0087] For example, the target assimilation weight matrix adaptively updates the weights. The solution can be obtained using the following formula: Equation (9) In the formula, To map the grade parameters of the target block to the observation operator matrix in the observation space where the drilling observation vector is located, To observe the transpose of the operator matrix, Let be the prior covariance matrix composed of the prior uncertainty parameters of each target block. It is a spatial heteroscedasticity kernel matrix composed of the elements of the spatial heteroscedasticity kernel matrix. It is the observation error covariance matrix composed of the preset observation error covariance.

[0088] In the above equation (9), It can be used as an equivalent prior covariance expression to simultaneously reflect the existing prior uncertainty of the target block and the geological heterogeneity introduced by the spatial heteroscedasticity constraint, observation operator matrix. This describes the correspondence between the grade status of a target block and the observations while drilling (WWP) vectors. When WWP observations only correspond to a portion of the target block, or when there is an aggregation, mapping, or scaling relationship between the observed values ​​and the block status, it can be represented by the observation operator matrix. Observation error covariance matrix. It is used to characterize the uncertainties introduced in drilling observation data by sensor errors, response inversion errors, or scale transformation errors.

[0089] Target assimilation weight matrix The solution process is used to weigh prior uncertainty, spatial heteroscedasticity constraints, and observation errors. Generally, when the observation error is small and the equivalent prior uncertainty is large, the correction effect of the observation-while-drilling vector on the grade state can be relatively enhanced; when the observation error is large or the prior state is relatively stable, the correction effect can be relatively weakened. Therefore, the target assimilation weight matrix can adjust the assimilation update intensity according to the prior state, geological heterogeneity, and observation reliability of the current target block.

[0090] In step S350, the observation residuals between the drilling observation vector and the prior mean vector mapped by the observation operator matrix are assimilated and fused using the target assimilation weight matrix to obtain the posterior mean vector and the posterior covariance matrix. The posterior mean vector is obtained by adaptively correcting the prior mean vector along the direction indicated by the observation residuals, and the posterior covariance matrix is ​​updated under the joint constraints of the spatial heteroscedasticity kernel matrix and the target assimilation weight matrix.

[0091] For example, using the target assimilation weight matrix For drilling observation vectors With the observed operator matrix Mapped prior mean vector The observation residuals between the two observations are assimilated and fused to calculate the posterior mean vector. and posterior covariance matrix : Equation (10) Equation (11) In the formula, It is the identity matrix. The drilling observation vector is composed of the block observation grades of each target block. It is the prior mean vector composed of the prior grade mean values ​​of each target block.

[0092] In the above posterior mean update formula, The target assimilation weight matrix represents the difference between the observation vector and the prior mean vector in the observation space. This is used to adjust the magnitude of the correction to the prior mean vector based on this difference. Thus, the posterior mean vector can absorb drilling observation information while retaining prior model information. The posterior covariance matrix is ​​used to represent the uncertainty state after assimilation and update. Its calculation process ensures that the updated uncertainty is simultaneously affected by the equivalent prior covariance and the target assimilation weight matrix, thereby unifying the grade mean update and uncertainty update into the same assimilation calculation process.

[0093] In step S360, the elements corresponding to the target block in the posterior mean vector are extracted as the posterior grade mean of each target block, and the posterior uncertainty parameters of each target block are determined based on the uncertainty components of the target block in the posterior covariance matrix.

[0094] In some implementations, each element in the posterior mean vector can correspond to the posterior grade mean of a target block. The diagonal component or marginal uncertainty component in the posterior covariance matrix corresponding to that target block can be used to determine the posterior uncertainty parameter of that target block. The system can write the corresponding results in the posterior mean vector and posterior covariance matrix back to the corresponding target block attribute fields in the 3D block grade model according to the block number, spatial order, or borehole crossing order in the target block set. Through this processing, the conversion from matrix calculation results to block model attribute results can be completed, enabling each target block to obtain updated posterior grade mean and posterior uncertainty parameters.

[0095] Through the embodiments of this application, the assimilation and updating process of the target block can simultaneously consider spatial anisotropy, lithological heterogeneity, prior uncertainty and observation error, so that the grade correction result can be adjusted according to the geological correlation and observation reliability between the target blocks, and the updated posterior grade parameters have clear numerical sources in terms of both grade mean and uncertainty.

[0096] In some examples of embodiments of this application, after obtaining the posterior grade parameters of each target block, it is also possible to determine the blocks to be diffused whose spatial range does not intersect with the spatial trajectory of the current borehole from the historical three-dimensional block grade model; for any block to be diffused, determine the updated set of target blocks located within the preset diffusion distance threshold range of the block to be diffused.

[0097] It should be noted that the current borehole typically only traverses a portion of the target non-metallic mineral area. For the target block directly traversed by the current borehole, posterior grade parameters can be obtained based on drilling observation data. However, for spatial blocks whose spatial extent does not intersect with the current borehole's trajectory, although they are not directly exposed by the current borehole, they may have some correlation with the updated target blocks in terms of spatial location, stratigraphic extension direction, or lithological properties. Therefore, these spatial blocks not directly traversed by the current borehole can be identified as blocks to be diffused. Within a preset diffusion distance threshold around the blocks to be diffused, target blocks that have undergone assimilation and updating are selected to obtain a set of updated target blocks.

[0098] Specifically, the preset diffusion distance threshold can be determined based on the spatial correlation scale of the target non-metallic mining area, the continuity of the ore body, the block size, or historical modeling experience. For example, for non-metallic ore bodies with relatively continuous layered distribution, the preset diffusion distance threshold can be appropriately increased along the direction of ore layer extension; for areas with many interlayers or significant changes in the boundary of mining waste, the preset diffusion distance threshold can be appropriately decreased. Through the above processing, the spatial range of the assimilation results during drilling can be limited to propagate to surrounding blocks, avoiding the unconstrained extension of local observation updates to areas with weak spatial correlation.

[0099] Then, for any updated target block in the updated target block set, the changes in mean grade and uncertainty before and after assimilation are calculated. The change in mean grade characterizes the grade correction magnitude of the updated target block, and the change in uncertainty characterizes the change in the reliability of the grade estimate of the updated target block.

[0100] Specifically, the change in mean grade can be determined by the difference between the updated posterior mean grade and the prior mean grade of the target block, and the change in uncertainty can be determined by the signed difference between the updated posterior uncertainty parameter and the prior uncertainty parameter of the target block. Here, the change is used instead of the posterior mean grade directly for diffusion to preserve the regional geological trends already present in the historical 3D block grade model. This ensures that the block to be diffused only absorbs the incremental correction information brought about by the current borehole assimilation process, preventing the original prior grade background of the block from being directly replaced by the absolute grade values ​​of neighboring blocks.

[0101] In some cases, the posterior uncertainty parameter of the updated target block is less than its prior uncertainty parameter, and the corresponding change in uncertainty can be negative, representing the reduction in uncertainty after assimilation. In other cases, if the current observation differs significantly from the prior or the observation reliability is low, the change in uncertainty can be retained as a non-negative value. Thus, by expressing the change in uncertainty through signed differences, the direction of the reliability change of the block to be diffused can be kept consistent with the assimilation result of the updated target block.

[0102] Subsequently, based on the posterior grade mean of the updated target block and its adjacent spatial blocks, the local grade spatial gradient is determined. Based on the geological anisotropy characteristics of the target non-metallic mining area, the spatial center coordinate difference between the updated target block and the block to be diffused is anisotropically scaled to obtain the anisotropic equivalent distance between the updated target block and the block to be diffused.

[0103] In some implementations, the local grade spatial gradient can be used to characterize the severity of grade changes around the updated target block. For example, the magnitude of local grade changes can be estimated along one or more spatial directions based on the difference between the posterior mean grade of the updated target block and its neighboring blocks. A small local grade spatial gradient indicates that the grade change in the area is relatively gradual; a large local grade spatial gradient indicates that the area may be located at a mine-waste boundary, interlayer change zone, stratigraphic abrupt change zone, or local anomaly zone. Therefore, the local grade spatial gradient can serve as an important basis for determining whether the assimilation increment is suitable for outward propagation.

[0104] Furthermore, the spatial continuity of non-metallic ore bodies is often directional. For example, the grade correlation along the strike or dip direction of a layered ore body may be stronger, while the grade variation along the cross-layer direction may be more pronounced. Therefore, based on the geological anisotropy characteristics of the target non-metallic mining area, the spatial center coordinate difference between the updated target block and the block to be diffused can be scaled to obtain the anisotropic equivalent distance, which better reflects the effective correlation between target blocks in geological space compared to ordinary geometric distance.

[0105] Next, based on the anisotropic equivalent distance and the local grade spatial gradient, the adaptive diffusion influence coefficient between the updated target block and the block to be diffused is determined. The adaptive diffusion influence coefficient decreases with increasing anisotropic equivalent distance and with increasing local grade spatial gradient. The local grade spatial gradient is used to characterize the degree of local abrupt changes in geological grade, in order to suppress excessive smoothing effects across geological abrupt boundary boundaries during spatial diffusion.

[0106] For example, the adaptive diffusion effect coefficient The following requirements must be met: Equation (12) In the formula, For the updated target segment With the block to be diffused The anisotropic equivalent distance between them The diffusion attenuation reference length and , The gradient characterization coefficients are set and , For the updated target segment The corresponding local grade spatial gradient.

[0107] In the above (12), This term characterizes the tendency of diffusion effects to decrease with increasing anisotropic equivalent distance. When the anisotropic equivalent distance between the updated target block and the block to be diffused is small, this term takes a relatively large value; when the anisotropic equivalent distance between them is large, this term takes a relatively small value. In the denominator... This setting is used to suppress diffusion effects based on local grade spatial gradients. When the local grade spatial gradient is large, the adaptive diffusion effect coefficient is reduced to decrease the diffusion intensity near areas of abrupt grade changes or lithological boundaries. This setting allows updated information to have a strong propagation effect in areas with gentle grade changes, while being more constrained in areas with drastic grade changes.

[0108] Subsequently, based on the adaptive diffusion influence coefficient, the changes in the mean grade and uncertainty of the updated target block before and after assimilation are propagated to the block to be diffused, so as to perform diffusion correction on the prior mean grade and prior uncertainty parameters of the block to be diffused. Specifically, the diffusion correction makes the updated target block, which is closer and has a more gradual grade change, have a stronger influence on the block to be diffused, while making the updated target block, which is farther away or has a more obvious grade change, have a weaker influence on the block to be diffused.

[0109] For example, the block to be diffused The updated grade mean and updated uncertainty parameter after diffusion correction can be expressed by the following formula: Equation (13) Equation (14) In the formula, and These are the segments to be diffused. The prior grade mean and prior uncertainty parameters before the update. and These are the segments to be diffused. Updated grade mean and updated uncertainty parameters after diffusion correction. For the updated target segment The change in mean grade before and after assimilation For the updated target segment The change in uncertainty before and after assimilation For the segment to be diffused The set of updated target blocks within the preset diffusion distance threshold range. This indicates traversing the updated target block set. The summation operation is performed on all updated target blocks in the dataset.

[0110] In the above expression, the updated grade mean of the block to be diffused is determined jointly by its prior grade mean and the change in grade mean of the updated target blocks in the neighborhood. The changes in each grade mean are adjusted using an adaptive diffusion influence coefficient. Therefore, updated target blocks that are closer in distance and have relatively gentle local grade changes contribute more to the correction of the grade mean of the block to be diffused. For the propagation of uncertainty parameters, the squared term of the adaptive diffusion influence coefficient is used to adjust the change in uncertainty. Since the diffusion influence coefficient is usually in the non-negative range, its squared term ensures that the propagation intensity of the uncertainty change is not higher than the propagation intensity of the grade mean change, thereby reducing the risk of excessively large uncertainty corrections in areas not directly traversed by the current borehole.

[0111] Then, the updated average grade and updated uncertainty parameters of the block to be diffused after diffusion correction are written into the grade model of the corrected block in the current mining cycle.

[0112] In some implementations, the updated mean grade and update uncertainty parameters can be written into the corresponding block attribute fields according to the block number, spatial index, or coordinate position of the block to be diffused in the historical 3D block grade model. After writing, the corrected block grade model for the current mining cycle includes not only the assimilation update results of the target block directly traversed by the current borehole, but also the diffusion correction results of the block to be diffused based on the adjacent updated target blocks. This allows the model to present a more continuous block grade change within the current mining cycle, while retaining corresponding boundary constraints in areas of abrupt local grade changes.

[0113] Through the embodiments of this application, the changes in the average grade and the changes in uncertainty generated by the current borehole assimilation can be propagated to the block to be diffused according to the diffusion influence coefficient controlled by the anisotropic equivalent distance and the local grade spatial gradient. This can enhance the propagation of updated information in areas where the grade changes are gentle and the spatial correlation is strong, and weaken the propagation of updated information in areas where the grade changes are drastic or the spatial correlation is weak, so that the update results of the block to be diffused are more consistent with the spatial continuity and local mutation characteristics of the target non-metallic mining area.

[0114] In some examples of embodiments of this application, after obtaining the posterior grade parameters of each target block, an equivalent prior covariance matrix can be constructed based on the prior covariance matrix and the spatial heteroscedasticity kernel matrix. The equivalent prior covariance matrix is ​​used to simultaneously characterize the prior uncertainty of the target block and the spatial constraint uncertainty introduced by geological heterogeneity.

[0115] For example, the equivalent prior covariance matrix It can be expressed by the following formula: Equation (15) Here, the prior covariance matrix The spatial heteroscedasticity kernel matrix is ​​used to represent the uncertainty in grade estimation of the target block before the introduction of current drilling observation data. This is used to represent spatial constraint uncertainties introduced by spatial proximity, geological anisotropy, and lithological heterogeneity. By combining these two factors to form an equivalent prior covariance matrix, the prior benchmark for information gain evaluation can simultaneously include the original model uncertainty and the influence of spatial heteroscedasticity constraints, thus providing a consistent comparison benchmark for uncertainty evaluation before and after assimilation.

[0116] Then, for any target block segment, extract the equivalent prior edge covariance submatrix corresponding to the target block segment from the equivalent prior covariance matrix, and extract the posterior edge covariance submatrix corresponding to the target block segment from the posterior covariance matrix.

[0117] Here, the equivalent prior covariance matrix and the posterior covariance matrix can simultaneously contain the joint uncertainty relationships between multiple target blocks. To evaluate the uncertainty change of a single target block after assimilating the current drilling observation data, the equivalent prior marginal covariance submatrix corresponding to the target block can be extracted from the equivalent prior covariance matrix, and the corresponding posterior marginal covariance submatrix can be extracted from the posterior covariance matrix, based on the index position of the target block in the target block set. For the case where the target block is characterized by only a single grade parameter, the marginal covariance submatrix can be represented as the corresponding variance term; for the case where multiple attribute parameters are simultaneously represented, the marginal covariance submatrix can be represented as the attribute covariance submatrix of the corresponding target block.

[0118] Subsequently, the equivalent prior marginal covariance submatrix and the posterior marginal covariance submatrix are positive definite, and based on the Gaussian information entropy relationship, the grade estimation uncertainty of the target block segment is expressed as the covariance uncertainty domain determined by the corresponding marginal covariance submatrix.

[0119] In some implementations, the extracted marginal covariance submatrix may be non-strictly positive definite, close to singular, or lack numerical stability due to sampling errors, matrix inversion errors, or floating-point operation errors during numerical computation. To ensure the stability of subsequent determinant and logarithmic operations, the equivalent prior marginal covariance submatrix and posterior marginal covariance submatrix can be positive definite. For example, a lower limit threshold can be set for excessively small or non-positive eigenvalues ​​through eigenvalue decomposition, or a preset regularization term can be added to the main diagonal of the submatrix. The equivalent prior marginal covariance submatrix and posterior marginal covariance submatrix after positive definite processing are still denoted as […]. and .

[0120] Under Gaussian uncertainty expression, the marginal covariance submatrix can be used to describe the range of uncertainty in the local state space of the target block grade estimation result. The covariance uncertainty domain can be understood as the range of local uncertainty distribution determined by the marginal covariance submatrix, and its magnitude is related to the determinant of the marginal covariance submatrix. Therefore, by comparing the changes in the uncertainty domain corresponding to the equivalent prior marginal covariance submatrix and the posterior marginal covariance submatrix, the extent to which the current drilling observation data reduces the uncertainty of the target block grade estimation can be evaluated.

[0121] Subsequently, using the reduction in the covariance uncertainty domain by the current borehole-introduced observation vector as a metric, the local information gain index corresponding to the target block is calculated. The local information gain index increases with the degree of reduction between the equivalent prior covariance uncertainty domain and the posterior covariance uncertainty domain, and is used to quantitatively characterize the reduction in grade estimation uncertainty of the target block after assimilating the observation vector.

[0122] For example, the reduction in the covariance uncertainty domain by the observation-while-drilling vector introduced by the current borehole is used as a metric, and the target block is calculated using a logarithmic determinant. Corresponding local information gain index : Equation (16) In the formula, This represents the operator for solving the determinant of a matrix. To correspond to the target block segment The equivalent prior marginal covariance submatrix, To correspond to the target block segment The posterior marginal covariance submatrix This is a local information gain index determined based on the ratio of covariance determinants. The local information gain index... Used for quantitative characterization of the target block segment The reduction in grade estimation uncertainty after assimilating the drilling observation vector; In the above equation (16), Used to characterize target blocks The local covariance uncertainty domain measure is used before assimilating the current drilling observation vector. Used to characterize target blocks The local information gain index is measured by the uncertainty domain of the local covariance after assimilating the current drilling observation vector. The difference between the two represents the degree to which the current drilling observation vector reduces the uncertainty of the target block grade estimation. When the uncertainty domain corresponding to the posterior marginal covariance submatrix is ​​significantly reduced compared to the equivalent prior marginal covariance submatrix, the local information gain index increases accordingly; when the change in uncertainty before and after assimilation is small, the local information gain index decreases accordingly.

[0123] Then, the local information gain index is associated with the corresponding target block segment for storage.

[0124] Specifically, the local information gain index corresponding to the target block can be used as the attribute information of the target block, associated with the spatial index, block number, or three-dimensional coordinate position of the target block, and stored in the three-dimensional block grade model or mine geological spatial database. Through this storage method, the contribution of current drilling observation data to the degree of uncertainty reduction of different target blocks can be recorded at the spatial block level, so that each target block not only has a posterior grade mean and posterior uncertainty parameters, but also corresponding information gain evaluation results.

[0125] Through the embodiments of this application, the information gain evaluation of the target block can be calculated based on the change in the covariance uncertainty domain before and after assimilation, so that the local information gain index has a clear source of covariance and information content meaning, and can reflect the degree of reduction of the uncertainty of the target block grade estimation by the current drilling observation data.

[0126] Figure 4 A flowchart illustrating an example of target borehole position feedback scheduling for the next drilling cycle based on an information state diagram in a method according to an embodiment of this application is shown.

[0127] like Figure 4 As shown, in step S410, after associating and storing the local information gain index with the corresponding target block, an information state diagram is constructed based on the local information gain index corresponding to each target block and the current uncertainty parameters of each spatial block in the modified block quality model. Here, the information state diagram includes graph nodes corresponding to each spatial block, and the node attributes of the graph nodes include at least the observed information gain and residual uncertainty of the corresponding spatial block.

[0128] Specifically, each spatial block in the modified block grade model can be mapped to a graph node in the information state diagram, and the edges between the graph nodes can be determined based on the proximity, connectivity, or geological extension relationships between the spatial blocks. For target blocks that have been updated through current borehole assimilation, their local information gain index can be written into the corresponding graph node as the observed information gain; for each spatial block in the modified block grade model, its current uncertainty parameter can be written into the corresponding graph node as the residual uncertainty. Thus, the information state diagram can express the information acquisition status and uncertainty distribution status of different spatial blocks in a graph data structure.

[0129] By constructing an information state diagram, acquired information and information still needing to be explored can be uniformly expressed at the spatial block level. The observed information gain characterizes the contribution of current or historical drilling observations to the reduction of uncertainty in the corresponding spatial block, while the remaining uncertainty characterizes the degree of uncertainty in grade estimation that still exists in the corresponding spatial block. Thus, the uncertainty information in the modified block grade model is transformed into a data structure that facilitates spatial retrieval and candidate location evaluation, enabling the information state of different spatial blocks to be uniformly read and compared.

[0130] In step S420, candidate borehole locations are selected, and the set of covered spatial segments is determined according to the corresponding predicted drilling trajectory. The residual uncertainty of each covered spatial segment is extracted from the information state diagram. Combined with the preset observation error covariance and the predicted observation relationship, the predicted posterior uncertainty parameter of each covered spatial segment at the candidate borehole location is estimated.

[0131] Specifically, candidate borehole locations are selected from a preset set of candidate borehole locations. Based on the candidate borehole locations and the corresponding predicted drilling trajectories, a set of spatial segments covered by the predicted drilling trajectories is determined. For any spatial segment in the set of spatial segments, the residual uncertainty of the corresponding graph node is extracted from the information state diagram as the current uncertainty parameter of the spatial segment. Based on the current uncertainty parameter, the preset observation error covariance, and the prediction observation relationship corresponding to the predicted drilling trajectory, the predicted posterior uncertainty parameter of the spatial segment at the candidate borehole locations is estimated.

[0132] In some implementations, the preset set of candidate borehole locations can be determined based on the current mining face, the area accessible to the drilling rig, the preset borehole spacing, the operational safety boundary, or the mining area production plan. For any candidate borehole location, a corresponding predicted drilling trajectory can be generated based on the borehole azimuth, dip angle, depth, or construction parameters corresponding to that candidate borehole location, and the set of spatial segments that the predicted drilling trajectory will traverse or cover can be determined. For each spatial segment in the set of spatial segments, the residual uncertainty of the corresponding graph node can be extracted from the information state diagram as the basis for evaluating the current uncertainty of the potential exploration benefits of that candidate borehole location.

[0133] In implementations employing linear assimilation or local linearization assimilation, the predicted posterior uncertainty parameter can be estimated based on the current uncertainty parameter, the preset observation error covariance, and the predicted observation relationship corresponding to the predicted drilling trajectory, without needing to obtain the actual drilling observation values ​​beforehand. The predicted observation relationship characterizes the observation correspondence between the predicted drilling trajectory and the spatial block state when drilling is carried out at the candidate borehole location; the preset observation error covariance characterizes the measurement errors, inversion errors, or spatial scale conversion errors that may exist during the expected drilling observation process. Thus, before actual drilling, the uncertainty changes that different candidate borehole locations may bring can be estimated, making the evaluation of candidate locations have a calculable basis.

[0134] In step S430, the expected information gain estimation model is invoked, and the expected local information gain corresponding to the spatial block is calculated based on the degree of uncertainty reduction between the current uncertainty parameter and the predicted posterior uncertainty parameter of the spatial block.

[0135] In some implementations, the expected information gain estimation model may employ a logarithmic determinant difference metric consistent with local information gain indices. For example, candidate borehole locations... Corresponding to spatial block segment Expected local information gain It can be calculated using the following formula: Equation (17) In the formula, For spatial blocks The covariance matrix corresponding to the current uncertainty parameter. For spatial blocks At candidate borehole locations The covariance matrix corresponding to the predicted posterior uncertainty parameter.

[0136] In the above formula (17), Used to characterize spatial blocks Covariance uncertainty domain measure before candidate borehole location evaluation Used to characterize the candidate borehole location The covariance uncertainty domain measure for the space block is expected to be obtained after observation. Therefore, the difference between the two is used to represent the relationship between the candidate borehole location and the space block. The degree of reduction in potential uncertainty allows the potential information gain of candidate borehole locations for different spatial blocks to be measured using a unified information content index.

[0137] In step S440, a comprehensive utility objective function for evaluating the detection benefits of candidate borehole locations is constructed based on the expected local information gain corresponding to each candidate borehole location.

[0138] Here, the overall utility objective function increases with the increase of the weighted expected local information gain of the spatial blocks covered by the predicted drilling trajectory, and decreases with the increase of the transfer cost of the on-site drilling rig moving from its current location to the candidate borehole location. The weighted expected local information gain is jointly determined by the resource priority weight of the spatial block and the corresponding expected local information gain.

[0139] For example, the comprehensive utility objective function It can be expressed by the following formula: Equation (18) In the formula, Candidate borehole locations The corresponding set of spatial blocks covered by the predicted drilling trajectory. For spatial blocks The corresponding resource priority weight coefficient, This is a preset coefficient representing the cost of drilling rig transfer. This indicates the current location of the drilling rig on site. From the current location to the candidate borehole location The actual transfer path length or transfer cost.

[0140] In equation (18) above, the summation term represents the weighted expected information gain corresponding to the set of spatial blocks covered by the predicted drilling trajectory. Resource priority weight coefficient The expected information gain of different spatial blocks can be weighted differently in the comprehensive utility evaluation based on their resource value, mining priority, production impact, or mineral waste boundary sensitivity. Deductions characterize the engineering costs required for the drilling rig to move from its current location to a candidate borehole location. These costs can be determined by the actual path length, transfer time, equipment energy consumption, or operational organization costs. Thus, the information gain and equipment transfer costs of candidate borehole locations are incorporated into the same evaluation function, ensuring that the evaluation results simultaneously reflect both geological information value and on-site implementation costs.

[0141] In step S450, among the candidate borehole locations that satisfy the preset equipment accessibility constraints, safe operation constraints, and drilling boundary constraints, the comprehensive utility objective function is optimized and solved, and the candidate borehole location with the highest comprehensive utility is determined as the target borehole location for the next drilling cycle.

[0142] Specifically, equipment accessibility constraints can be used to limit the spatial range that the drilling rig can reach or operate in; safe operation constraints can be used to limit the safe distance between the borehole location and slopes, roadways, goafs, equipment passages, or hazardous areas; and drilling boundary constraints can be used to limit the borehole location, drilling direction, and drilling depth from exceeding the operating boundary of the target non-metallic mining area or the preset mining range. Among the candidate borehole locations that satisfy the above constraints, enumeration search, heuristic optimization, swarm intelligence optimization, gradient optimization, or other constraint optimization methods can be used to optimize and solve the comprehensive utility objective function.

[0143] By using the above-mentioned constraint optimization, the target borehole location with higher comprehensive utility can be selected from multiple candidate borehole locations. This ensures that the determination of the target borehole location is simultaneously influenced by geological information benefits, resource priorities, equipment relocation costs, and on-site operational constraints, thereby improving the consistency between the borehole layout results and actual mining operation conditions.

[0144] In step S460, a feedback scheduling instruction is generated based on the target borehole location and sent to the on-site drilling rig or drilling rig scheduling system so that the on-site drilling rig can perform new multi-source sensor data acquisition while drilling for the target borehole location in the next drilling cycle and trigger the next round of adaptive update of the correction block grade model.

[0145] Here, the feedback scheduling instructions can include the three-dimensional coordinates of the target borehole location, borehole azimuth, borehole inclination angle, target borehole depth, work sequence, or equipment scheduling information. For drilling rig systems with automatic scheduling capabilities, the feedback scheduling instructions can be converted into equipment control parameters; for work scenarios requiring manual confirmation, the feedback scheduling instructions can also be generated as borehole work orders, scheduling tasks, or construction suggestions. After receiving the feedback scheduling instructions, the on-site drilling rig or drilling rig scheduling system can arrange the drilling operations for the next drilling cycle based on the target borehole location.

[0146] Through the embodiments of this application, the candidate location optimization results are converted into field-executable drilling operation information, enabling the target borehole location to be used for production organization and drilling data acquisition in the next drilling cycle. By converting the calculation results of the information state diagram, expected information gain, and comprehensive utility objective function into feedback scheduling instructions, the borehole layout process can be kept consistent with the current information state of the modified block grade model.

[0147] Figure 5 A schematic diagram illustrating the system operation mechanism of an example of a non-metallic mineral grade block adaptive correction method driven by drilling feedback according to an embodiment of this application is shown.

[0148] like Figure 5As shown, the system uses real-time multi-source signals and historical 3D block grade models as dual inputs. The real-time multi-source signals can include drilling physical response characteristics such as gamma ray, acoustic waves, and laser spectroscopy, and are input as multi-source signals to the depth attention response inversion module for multi-source feature extraction and initial grade estimation, yielding dynamic prediction results during drilling. The historical 3D block grade model provides prior parameters such as the mean, covariance, or uncertainty of the grade, and these prior parameters are input to the heteroscedasticity adaptive assimilation module. The heteroscedasticity adaptive assimilation module, based on the dynamic prediction results during drilling and prior parameters, combined with the spatial geological differences of the target non-metallic mining area, determines the assimilation update weights and outputs the updated error and covariance results. Subsequently, the adaptive neighborhood diffusion module, based on the updated error and covariance results, performs constrained propagation of the correction amount along the orebody distribution direction or spatial correlation direction to generate a corrected block grade model. Here, the corrected block grade model includes the updated grade mean and uncertainty parameters. Furthermore, the heteroscedasticity adaptive assimilation module and the adaptive neighborhood diffusion module can also output an information gain field. The information gain feedback scheduling module calculates the uncertainty reduction based on the information gain field and determines the target borehole location, then generates dynamic borehole scheduling instructions and sends them to the on-site drilling rig or drilling rig scheduling system to drive the next cycle of multi-source signal acquisition while drilling. Thus, the system can form a closed-loop update process between data acquisition while drilling, grade model correction, and borehole feedback scheduling.

[0149] To verify the effectiveness and engineering applicability of the proposed drilling feedback-driven adaptive correction method for non-metallic mineral grade blocks, this experiment constructed a three-dimensional geostatistical simulation model of a heterogeneous non-metallic deposit. Considering the real-world characteristics of non-metallic deposits such as phosphate rock and kaolinite, including bedding development, grade spatial anisotropy, and local abrupt changes, a high-resolution three-dimensional synthetic block model at a realistic scale was generated using Sequential Gaussian Simulation (SGS). The model orebody has a strike length of 1000 m, a dip width of 800 m, and a vertical depth of 200 m. It is discretized into 320,000 spatial blocks with a grid resolution of 10 m × 10 m × 5 m. The actual grade field follows a log-normal distribution. An anisotropic variability function with a strike range of 150 m, a dip range of 100 m, and a vertical range of 15 m was introduced. Furthermore, blocky high-grade anomaly zones were superimposed at specific strata to realistically simulate complex local mineralization enrichment scenarios.

[0150] Regarding the experimental setup and baseline comparison, the initial prior model was set as a static model generated using ordinary kriging interpolation based on a 100 m × 100 m sparse exploration borehole grid, which retained a high initial uncertainty. In simulating the actual acquisition-while-drilling (AWD) and dynamic assimilation process, a multi-source physical response vector containing gamma, acoustic transit time, and dimensionality-reduced spectral features was extracted every 1 m of drilling, and Gaussian white noise with a signal-to-noise ratio of 20 dB was superimposed to reproduce measurement interference from the engineering site. To highlight the performance advantages of the proposed method, the experiment selected the inverse distance weighting (IDW) static update method and the integrated Kalman filter (Standard EnKF) as comparative baselines to verify the assimilation correction performance of the proposed method under spatial heteroscedasticity constraints.

[0151] To comprehensively evaluate the model correction effect from three dimensions—numerical accuracy, spatial morphology, and engineering economics—this experiment established three core evaluation indicators. First, the relative root mean square error (rRMSE) is used to measure the overall deviation between the estimated grade field and the actual grade field, effectively eliminating evaluation bias caused by the absolute magnitude of grade values. Second, the grade-tonnage curve consistency (GT Concordance) index is introduced to assess the goodness of fit between the updated block model's predicted ore quantity and average grade at different boundary grades and the actual situation, forming the core foundation of mine economic evaluation. Finally, the local information gain (IG) index is used to quantitatively evaluate the reduction in covariance uncertainty of the target block after each borehole update. To maintain consistency with the theoretical framework above, this index uses the difference in logarithmic determinant calculated from the equivalent prior marginal covariance submatrix before and after assimilation with the posterior marginal covariance submatrix, i.e., satisfying… This allows for a direct verification of the actual information contribution of drilling feedback observations to eliminating local geological blind spots.

[0152] Figure 6 This diagram illustrates a comparative simulation of the evolution of three-dimensional blocks of grade at different update stages when the method of this application is applied to a non-metallic mining area. The diagram displays three grade block models within the same three-dimensional spatial reference frame, where the spatial reference directions include depth, strike, and dip. Different grade ranges are represented by varying color depths and transparency. Low-grade areas are displayed in light colors or with high transparency, medium-grade areas in semi-transparent light blue, and high-grade anomaly areas in dark voxels, facilitating observation of the spatial distribution of high-grade anomalies within the three-dimensional block models.

[0153] like Figure 6 Part (a) of the paper illustrates a prior model constructed based on historical sparse exploration data. In this model, high-grade anomaly areas are distributed as multiple relatively dispersed blocks or clusters, with weak continuity between local high-grade areas and diffuse boundaries in some medium- and high-grade areas. This indicates that when relying solely on historical sparse exploration data for spatial estimation, the model is susceptible to insufficient sampling density and spatial smoothing, resulting in unclear representations of the extension direction and boundary morphology of high-grade anomalies.

[0154] like Figure 6 Part (b) shows the drilling update model updated by the drilling feedback adaptive correction method according to embodiments of this application. Figure 6 Compared to part (a) of the previous model, the high-grade anomaly region in the drilling update model gradually converges from a relatively scattered blocky distribution into a banded or vein-like structure extending along a specific spatial direction, while the medium-grade region also forms a more continuous transition zone around the high-grade anomaly region. This result indicates that by introducing multi-source sensing data during drilling and combining it with spatial heteroscedasticity constraints for assimilation correction, the model can adjust for local dispersion errors in the prior model, making the spatial distribution of high-grade anomalies more concentrated and continuous.

[0155] Figure 6 Section (c) shows the reference real model used as the simulation verification benchmark. In this reference real model, the high-grade anomaly region appears as a vein-like structure extending continuously along the spatial direction, while the medium-grade region is distributed around the high-grade anomaly region, forming a transition envelope. Figure 6 Part (b) of the text is related to Figure 6 A comparison of part (c) shows that the drilling update model is closer to the reference real model than the prior model in terms of the extension direction, continuity, and distribution of the peripheral medium-grade transition zone of the high-grade anomaly.

[0156] Depend on Figure 6 A comparison of parts (a), (b), and (c) in the document reveals that the method in this embodiment can adaptively correct the historical three-dimensional block grade model using drilling feedback data, causing the originally diffuse prior grade distribution to converge towards a spatial morphology that better conforms to the reference real model. This comparison demonstrates that the method in this embodiment helps reduce local smoothing errors caused by sparse historical exploration data and improves the ability of the three-dimensional block grade model in non-metallic mineral areas to represent the spatial continuity and boundary morphology of high-grade anomalies.

[0157] Figure 7 The diagram shows a comparison of the overall performance of different methods and a simulation result of an example of the convergence process during drilling updates.

[0158] like Figure 7 As shown in the main view, the horizontal axis represents the single-step assimilation delay, and the vertical axis represents the relative root mean square error (rRMSE). The figure shows the experimental results distribution for the Inverse Distance Weighted Interpolation (IDW), the Integrated Kalman Filter (EnKF), and the ADGC (Adaptive Drilling-driven Grade Correction) method of this application. Specifically, the data points corresponding to the IDW method are mainly distributed in the low-latency but relatively high-error region, the data points corresponding to the EnKF method are mainly distributed in the medium-latency and medium-error region, and the data points corresponding to the ADGC method of this application are mainly distributed in the low-error region, forming a corresponding Pareto front. These results indicate that the method of this application, with a certain increase in single-step assimilation delay, can achieve a relatively low rRMSE, thus achieving a better overall performance between computational overhead and model error.

[0159] Furthermore, Figure 7 The inset subplot in the upper right corner illustrates the trends in the overall rRMSE and information gain of the model as the number of boreholes increases. The line graph represents the overall relative root mean square error (rRMSE), and the bar graph represents the information gain during the corresponding borehole update process. The inset subplot shows that in the early stages with fewer boreholes, the information gain is relatively high, and the model's rRMSE decreases significantly. As the number of boreholes continues to increase, the information gain gradually decreases, and the rate of decrease in the model's rRMSE also gradually diminishes, showing an overall convergence trend.

[0160] Depend on Figure 7 It can be seen that the method of this application can reduce the relative error of the three-dimensional block grade model through drilling feedback updates, and the trend of its information gain change can reflect the marginal improvement of the model uncertainty by the new borehole observations. This comparative result shows that the method of this application has good coordination between grade model error control, computational overhead, and borehole information benefit evaluation, and can provide experimental support for borehole position feedback scheduling in the next drilling cycle.

[0161] Figure 8This diagram illustrates a comparative simulation of the grade-tonnage evaluation of non-metallic mineral deposits using different methods based on different block models. The grade-tonnage comparison curves are used to compare the tonnage estimation results and average grade estimation results of different block models under different boundary grade conditions. The tonnage curve corresponds to the left axis and is represented by a solid line, while the average grade curve corresponds to the right axis and is represented by a dashed or dotted line to distinguish it from the left-axis tonnage curve. In the figure, the horizontal axis represents the boundary grade, the left vertical axis represents the ore tonnage, and the right vertical axis represents the average grade. The tonnage curve characterizes the ore tonnage variation that meets the grade requirements under the corresponding boundary grade conditions, while the average grade curve characterizes the average grade variation corresponding to the ore block that meets the boundary grade conditions. The figure shows the evaluation results corresponding to the reference real model, the prior model, and the ADGC model updated by the method of this application embodiment.

[0162] like Figure 8 As shown in the tonnage curves, the ore tonnage corresponding to each model decreases with increasing boundary grade. The tonnage curves corresponding to the prior model are generally higher than those of the reference model, indicating that prior models relying solely on sparse historical exploration data may overestimate the amount of ore meeting the boundary grade conditions. In contrast, the tonnage curve corresponding to the ADGC model is closer to the reference model, indicating that the block grade model corrected by drilling feedback data can more reasonably reflect changes in ore tonnage under different boundary grade conditions.

[0163] As can be seen from the average grade curves, the average grade corresponding to each model shows an upward trend as the boundary grade increases. The average grade curves corresponding to the prior models are generally lower than those of the reference model, indicating that the prior models may underestimate the average grade of the screened ore due to spatial smoothing or insufficient representation of local high-grade anomalies. The average grade curve corresponding to the ADGC model is closer to that of the reference model, indicating that the method in this application can improve the representation effect of the prior models on the spatial distribution of high-grade blocks.

[0164] Depend on Figure 8 It can be seen that the updated ADGC model of the method in this embodiment is closer to the reference real model in both ore tonnage curve and average grade curve than the prior model. This result shows that the method of this embodiment can reduce the grade-tonnage evaluation bias caused by historical sparse data modeling, and helps to improve the data support capability of non-metallic mineral block grade models in dynamic reserve assessment, boundary grade analysis and mining plan formulation.

[0165] In summary, addressing the problem that traditional non-metallic mineral grade block models rely on historical static exploration data and struggle to reflect newly revealed information during mining, this application proposes a drilling feedback-driven adaptive correction method for non-metallic mineral grade blocks. This method incorporates real-time drilling multi-source sensor data acquired during drilling operations into the block grade model update process. Specifically, a pre-defined response inversion model converts multi-dimensional physical response signals such as gamma, acoustic, and laser spectra into drilling grade estimation results distributed along the borehole depth, forming the observed block grade after spatial attribution matching. Furthermore, by combining prior grade parameters, spatial proximity relationships, lithological heterogeneity, and prior uncertainty parameters from the historical three-dimensional block grade model, spatial heteroscedasticity constraints are constructed, and adaptive update weights are determined, thereby achieving assimilation and correction between drilling observation information and the historical block model.

[0166] Building upon this foundation, this application can further propagate the updated target block's correction information to neighboring, unexposed blocks via a neighborhood diffusion mechanism, and use information gain metrics to schedule subsequent borehole locations, enabling the grade block model to be continuously updated throughout the mining process. Simulation results show that, compared to prior models relying solely on historical sparse exploration data, the method presented in this application can reduce grade estimation errors, improve the spatial morphological representation of high-grade anomaly zones, and enhance the consistency between grade-tonnage evaluation results and the reference model. Therefore, this application can provide more reliable data support for dynamic grade modeling, mining plan adjustment, mine waste identification, and digital mine management in non-metallic mines such as potash, phosphate, and limestone.

[0167] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0168] Figure 9 A structural block diagram of an example of a nonmetallic mineral grade block adaptive correction system driven by drilling feedback according to an embodiment of this application is shown.

[0169] like Figure 9 As shown, the non-metallic mineral grade block adaptive correction system 900 driven by drilling feedback includes a model data acquisition unit 910, a drilling inversion matching unit 920, a block observation aggregation unit 930, a heteroscedasticity weight determination unit 940, and a model assimilation correction unit 950.

[0170] The model data acquisition unit 910 is used to acquire the historical three-dimensional block grade model of the target non-metallic mining area and the real-time multi-source sensing data collected during drilling operations. The historical three-dimensional block grade model includes multiple spatial blocks and the spatial range, lithological properties and prior grade parameters corresponding to each spatial block. The prior grade parameters include the prior grade mean and the prior uncertainty parameter used to characterize the uncertainty of the prior grade mean. The multi-source sensing data collected during drilling includes the spatial trajectory of the current borehole and the multi-dimensional physical response signal collected along the drilling depth.

[0171] The drilling inversion matching unit 920 is used to determine the sequence of target blocks traversed by the current borehole based on the intersection relationship between the spatial trajectory and the spatial range of each spatial block, and input the multidimensional physical response signal into a preset response inversion model to obtain a drilling grade estimation profile distributed along the current borehole depth.

[0172] The block observation aggregation unit 930 is used to aggregate the drilling grade estimates falling into the same target block according to the spatial attribution relationship between each depth position in the drilling grade estimation profile and the target block sequence, so as to obtain the block observation grade of each target block.

[0173] The heteroscedasticity weight determination unit 940 is used to determine the spatial proximity relationship between adjacent spatial blocks based on the spatial range, and to determine the degree of lithological heterogeneity between adjacent spatial blocks based on the lithological properties. It forms spatial heteroscedasticity constraints based on the spatial proximity relationship, the degree of lithological heterogeneity, and the prior uncertainty parameter, and determines the adaptive update weight of each target block based on the spatial heteroscedasticity constraints. The spatial heteroscedasticity constraints are used to adjust the intensity of grade assimilation influence between adjacent spatial blocks.

[0174] The model assimilation correction unit 950 is used to assimilate the observation difference between the observed grade of the block and the corresponding prior grade mean using the adaptive update weights to obtain the posterior grade parameters of each target block, and to update the historical three-dimensional block grade model based on the posterior grade parameters to generate the corrected block grade model for the current mining cycle; the posterior grade parameters include the posterior grade mean and the corresponding posterior uncertainty parameters.

[0175] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the steps of any of the above-described drilling feedback-driven adaptive correction methods for non-metallic mineral grade blocks.

[0176] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described drilling feedback-driven adaptive correction methods for non-metallic mineral grade blocks.

[0177] In some embodiments, this application also provides an electronic device comprising: at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of a non-metallic mineral grade block adaptive correction method driven by drilling feedback.

[0178] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0179] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0180] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0181] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0182] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A non-metallic mineral grade block adaptive correction method driven by drilling feedback, characterized in that, The method includes: The historical three-dimensional block grade model of the target non-metallic mining area and real-time drilling multi-source sensor data are acquired during drilling operations. The historical three-dimensional block grade model includes multiple spatial blocks and the spatial range, lithological properties and prior grade parameters corresponding to each spatial block. The prior grade parameters include the prior grade mean and the prior uncertainty parameter used to characterize the uncertainty of the prior grade mean. The drilling multi-source sensor data includes the spatial trajectory of the current borehole and multi-dimensional physical response signals acquired along the drilling depth. Based on the intersection relationship between the spatial trajectory and the spatial range of each spatial block, the sequence of target blocks traversed by the current borehole is determined, and the multidimensional physical response signal is input into a preset response inversion model to obtain a drilling grade estimation profile distributed along the current borehole depth. Based on the spatial attribution relationship between each depth position in the drilling grade estimation profile and the target block sequence, the drilling grade estimation values ​​falling into the same target block are aggregated to obtain the block observation grade of each target block; Based on the spatial range, the spatial proximity relationship between adjacent spatial blocks is determined, and based on the lithological properties, the degree of lithological heterogeneity between adjacent spatial blocks is determined. Spatial heteroscedasticity constraints are formed according to the spatial proximity relationship, the degree of lithological heterogeneity, and the prior uncertainty parameter. The adaptive update weight of each target block is determined according to the spatial heteroscedasticity constraints. The spatial heteroscedasticity constraints are used to adjust the intensity of grade assimilation influence between adjacent spatial blocks. The observed difference between the observed grade of the block and the corresponding prior grade mean is assimilated using the adaptive update weights to obtain the posterior grade parameters of each target block. The historical three-dimensional block grade model is then updated based on the posterior grade parameters to generate a corrected block grade model for the current mining cycle. The posterior grade parameters include the posterior grade mean and the corresponding posterior uncertainty parameters.

2. The method according to claim 1, characterized in that, The multidimensional physical response signals acquired along the drilling depth include gamma-ray signals while drilling, acoustic logging signals, and cuttings laser spectral signals. The step of inputting the multidimensional physical response signal into a preset response inversion model to obtain a drilling grade estimation profile along the current borehole depth includes: The real-time drilling dynamics parameters of the current borehole during the drilling operation are obtained, and the real-time drilling dynamics parameters include at least the instantaneous drilling speed and the drill bit attitude inclination angle; The multidimensional physical response signal is time-synchronized, and based on the real-time drilling dynamics parameters, the synchronized multidimensional physical response signal is subjected to time-depth mapping processing to transform the response signal under the time series into a unified spatial depth domain composed of multiple discrete depth nodes; wherein, the time-depth mapping processing determines the equivalent spatial depth corresponding to each acquisition moment based on the cumulative advance of the drilling speed in the time dimension and the projection effect of the drill bit attitude inclination angle on the advance direction, and the equivalent spatial depth is used to characterize the projection depth of the current borehole along a preset reference direction; To address the depth resolution differences caused by the different detection principles of multi-source sensors, and constrained by a preset benchmark sampling interval and the effective detection window corresponding to various physical response signals, spatial resampling, window aggregation, and scale normalization are performed on each physical response signal mapped to the unified spatial depth domain to form a multi-source feature vector corresponding to each discrete depth node. The multi-source feature vector includes at least the normalized count rate extracted from the drilling gamma-ray signal, the amplitude attenuation gradient extracted from the sonic logging signal, and the spectral element intensity distribution extracted from the cuttings laser spectral signal. The multi-source feature vectors corresponding to each discrete depth node are input into the preset response inversion model in a spatial sequence to output the grade prediction value of each discrete depth node. The grade estimation profile while drilling is formed by each discrete depth node and the corresponding grade prediction value.

3. The method according to claim 2, characterized in that, The preset response inversion model includes multiple feature extraction channels, an attention allocation unit, and a grade output unit; wherein, each feature extraction channel corresponds to a physical sensing mode, and the multiple physical sensing modes include a gamma response mode corresponding to the drilling gamma ray signal, an acoustic response mode corresponding to the acoustic logging signal, and a spectral response mode corresponding to the cuttings laser spectrum signal; The grade prediction values ​​for each discrete depth node are output through the preset response inversion model, including: For any discrete depth node, a multi-source feature vector is used to determine the modal sub-features corresponding to each of the physical perception modes. The multi-source feature vectors are then input into an attention allocation unit to determine the attention weights corresponding to each of the physical perception modes. Each attention weight is used to characterize the degree of contribution of the corresponding physical perception mode to the grade prediction result. Each modal sub-feature is input into its corresponding feature extraction channel for nonlinear feature extraction. The modal features output by each feature extraction channel are then weighted and fused based on the attention weights of each modality. The grade prediction value corresponding to any discrete depth node is then output by the grade output unit. The grade prediction value is obtained by fusing the nonlinear feature results of each physical perception modality according to the normalized attention contribution relationship. The preset response inversion model is a model pre-trained based on a loss function with geological constraints. The loss function includes grade prediction error constraints, depth adjacent node continuity constraints, and geological response consistency constraints. The grade prediction error constraints are used to constrain the deviation between the predicted grade value and the true grade label. The depth adjacent node continuity constraints are used to adjust the smoothness of the grade prediction results according to the lithological continuity between adjacent discrete depth nodes. The geological response consistency constraints are used to constrain the inversion results corresponding to the model parameters to be consistent with the preset lithology-grade response relationship.

4. The method according to claim 1, characterized in that, The spatial heteroscedasticity constraint is quantitatively characterized by the spatial heteroscedasticity kernel matrix. The a priori grade mean values ​​corresponding to each target block together constitute the a priori mean vector, and the block observation grade of each target block together constitutes the drilling observation vector. The adaptive update weights for each target block segment, determined based on the spatial heteroscedasticity constraint, include: Based on the spatial range of each target block, the spatial center coordinates corresponding to each target block are determined. Combining the spatial proximity relationship and according to the geological anisotropic characteristics of the target non-metallic mining area, the spatial center coordinate difference between any two target blocks is anisotropically scaled to obtain the anisotropic equivalent distance. Based on the lithological properties of each target block and the differences in lithological properties between adjacent spatial blocks, a lithological heterogeneity index is determined for each target block; wherein, the lithological heterogeneity index is used to quantify the degree of lithological difference between the corresponding target block and its neighboring spatial blocks. For any two target blocks in the target block set composed of the aforementioned target blocks, a corresponding spatial heteroscedasticity kernel matrix element is constructed; wherein, the spatial heteroscedasticity kernel matrix element decreases as the anisotropic equivalent distance between the two corresponding target blocks increases, and increases as the lithological heterogeneity index of the two corresponding target blocks increases, so that the influence of spatial distance and the influence of lithological difference jointly participate in the grade assimilation constraint; Based on the spatial heteroscedasticity kernel matrix, the prior covariance matrix composed of the prior uncertainty parameters of each target block, the observation error covariance matrix composed of the preset observation error covariance, and the observation operator matrix, the target assimilation weight matrix, which serves as the adaptive update weight, is solved. The observation operator matrix is ​​used to map the grade parameters of the target block to the observation space where the drilling observation vector resides, and the target assimilation weight matrix is ​​used to adjust the correction strength of the drilling observation vector to the prior mean vector under spatial heteroscedasticity constraints and observation error constraints. Using the target assimilation weight matrix, the observation residuals between the drilling observation vector and the prior mean vector mapped by the observation operator matrix are assimilated and fused to obtain the posterior mean vector and the posterior covariance matrix; wherein, the posterior mean vector is obtained by adaptively correcting the prior mean vector along the direction indicated by the observation residuals, and the posterior covariance matrix is ​​updated by updating the prior covariance matrix under the joint constraints of the spatial heteroscedasticity kernel matrix and the target assimilation weight matrix; The elements corresponding to the target block segments in the posterior mean vector are extracted as the posterior grade mean of each target block segment, and the posterior uncertainty parameters of each target block segment are determined based on the uncertainty components of the target block segments in the posterior covariance matrix.

5. The method according to claim 4, characterized in that, After obtaining the posterior grade parameters of each target block, the method further includes: From the historical three-dimensional segment quality model, identify segments whose spatial range does not intersect with the spatial trajectory of the current borehole; for any segment to be diffused, determine the updated set of target segments within the preset diffusion distance threshold range of the segment to be diffused; For any updated target block in the set of updated target blocks, calculate the change in mean grade and the change in uncertainty before and after assimilation; wherein, the change in mean grade is used to characterize the grade correction magnitude of the updated target block, and the change in uncertainty is used to characterize the change in the reliability of the grade estimation of the updated target block. Based on the posterior grade mean of the updated target block and its adjacent spatial blocks, the local grade spatial gradient is determined. Based on the geological anisotropic characteristics of the target non-metallic mining area, the spatial center coordinate difference between the updated target block and the block to be diffused is anisotropically scaled to obtain the anisotropic equivalent distance between the updated target block and the block to be diffused. Based on the anisotropic equivalent distance and the local grade spatial gradient, an adaptive diffusion influence coefficient is determined between the updated target block and the block to be diffused; wherein, the adaptive diffusion influence coefficient decreases as the anisotropic equivalent distance increases and decreases as the local grade spatial gradient increases, and the local grade spatial gradient is used to characterize the degree of local abrupt change in geological grade, so as to suppress the excessive smoothing effect across the boundary of geological abrupt change during spatial diffusion; Based on the adaptive diffusion influence coefficient, the changes in the mean grade and uncertainty of the updated target block before and after assimilation are propagated to the block to be diffused, so as to perform diffusion correction on the prior mean grade and prior uncertainty parameters of the block to be diffused; wherein, the diffusion correction makes the updated target block that is closer and has a more gradual grade change have a stronger influence on the block to be diffused, and makes the updated target block that is farther away or has a more obvious grade change have a weaker influence on the block to be diffused. The updated mean grade and updated uncertainty parameters of the block to be diffused after diffusion correction are written into the grade model of the corrected block in the current mining cycle.

6. The method according to claim 4, characterized in that, After obtaining the posterior grade parameters of each target block, the method further includes: An equivalent prior covariance matrix is ​​constructed based on the prior covariance matrix and the spatial heteroscedasticity kernel matrix; wherein, the equivalent prior covariance matrix is ​​used to simultaneously characterize the prior uncertainty of the target block and the spatial constraint uncertainty introduced by geological heterogeneity; For any target block segment, extract the equivalent prior edge covariance submatrix corresponding to the target block segment from the equivalent prior covariance matrix, and extract the posterior edge covariance submatrix corresponding to the target block segment from the posterior covariance matrix; The equivalent prior marginal covariance submatrix and the posterior marginal covariance submatrix are positive definite, and based on the Gaussian information entropy relationship, the grade estimation uncertainty of the target block is expressed as the covariance uncertainty domain determined by the corresponding marginal covariance submatrix. Using the reduction in the covariance uncertainty domain by the current borehole-introduced observation-while-drilling vector as a metric, the local information gain index corresponding to the target block is calculated; wherein, the local information gain index increases with the degree of reduction between the equivalent prior covariance uncertainty domain and the posterior covariance uncertainty domain, and is used to quantitatively characterize the reduction in the grade estimation uncertainty of the target block after assimilating the observation-while-drilling vector; The local information gain index is associated with and stored with the corresponding target block segment.

7. The method according to claim 6, characterized in that, After associating and storing the local information gain index with the corresponding target block segment, the method further includes: Based on the local information gain index corresponding to each target block and the current uncertainty parameter of each spatial block in the modified block quality model, an information state diagram is constructed; the information state diagram includes graph nodes corresponding to each spatial block, and the node attributes of the graph nodes include at least the observed information gain and residual uncertainty of the corresponding spatial block; Candidate borehole locations are selected from a preset set of candidate borehole locations. Based on the candidate borehole locations and the corresponding predicted drilling trajectories, a set of spatial segments covered by the predicted drilling trajectories is determined. For any spatial segment in the set of spatial segments, the residual uncertainty of the corresponding graph node is extracted from the information state diagram as the current uncertainty parameter of the spatial segment. Based on the current uncertainty parameter, the preset observation error covariance, and the prediction observation relationship corresponding to the predicted drilling trajectory, the predicted posterior uncertainty parameter of the spatial segment at the candidate borehole location is estimated. The expected information gain estimation model is invoked, and the expected local information gain of the candidate borehole location corresponding to the spatial block is calculated based on the degree of uncertainty reduction between the current uncertainty parameter and the predicted posterior uncertainty parameter of the spatial block. Based on the expected local information gain corresponding to each of the candidate borehole locations, a comprehensive utility objective function is constructed to evaluate the detection benefits of the candidate borehole locations; wherein, the comprehensive utility objective function increases with the increase of the weighted expected local information gain of the spatial block covered by the predicted drilling trajectory, and decreases with the increase of the transfer cost of the on-site drilling rig moving from its current location to the candidate borehole location; the weighted expected local information gain is jointly determined by the resource priority weight of the spatial block and the corresponding expected local information gain; Among the candidate borehole locations that satisfy the preset equipment accessibility constraints, safe operation constraints, and drilling boundary constraints, the comprehensive utility objective function is optimized and solved, and the candidate borehole location with the highest comprehensive utility is determined as the target borehole location for the next drilling cycle. A feedback scheduling instruction is generated based on the target borehole location, and the feedback scheduling instruction is sent to the on-site drilling rig or drilling rig scheduling system so that the on-site drilling rig performs new multi-source sensor data acquisition while drilling for the target borehole location in the next drilling cycle, and triggers the next round of adaptive update of the correction block grade model.

8. A non-metallic mineral grade block adaptive correction system driven by drilling feedback, characterized in that, The system includes: The model data acquisition unit is used to acquire the historical three-dimensional block grade model of the target non-metallic mining area and the real-time multi-source sensing data collected during drilling operations. The historical three-dimensional block grade model includes multiple spatial blocks and the spatial range, lithological properties and prior grade parameters corresponding to each spatial block. The prior grade parameters include the prior grade mean and the prior uncertainty parameter used to characterize the uncertainty of the prior grade mean. The multi-source sensing data collected during drilling includes the spatial trajectory of the current borehole and the multi-dimensional physical response signal collected along the drilling depth. The drilling inversion matching unit is used to determine the sequence of target blocks traversed by the current borehole based on the intersection relationship between the spatial trajectory and the spatial range of each spatial block, and input the multidimensional physical response signal into a preset response inversion model to obtain a drilling grade estimation profile distributed along the current borehole depth. The block observation aggregation unit is used to aggregate the drilling grade estimates falling into the same target block according to the spatial attribution relationship between each depth position in the drilling grade estimation profile and the target block sequence, so as to obtain the block observation grade of each target block. The heteroscedasticity weight determination unit is used to determine the spatial proximity relationship between adjacent spatial blocks based on the spatial range, and to determine the degree of lithological heterogeneity between adjacent spatial blocks based on the lithological properties. It then forms spatial heteroscedasticity constraints based on the spatial proximity relationship, the degree of lithological heterogeneity, and the prior uncertainty parameter, and determines the adaptive update weight of each target block according to the spatial heteroscedasticity constraints. The spatial heteroscedasticity constraints are used to adjust the intensity of grade assimilation effects between adjacent spatial blocks. The model assimilation and correction unit is used to assimilate the observation difference between the observed grade of the block and the corresponding prior grade mean using the adaptive update weights to obtain the posterior grade parameters of each target block, and to update the historical three-dimensional block grade model based on the posterior grade parameters to generate the corrected block grade model for the current mining cycle; the posterior grade parameters include the posterior grade mean and the corresponding posterior uncertainty parameters.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor, characterized in that the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method as described in any one of claims 1-7.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-7.

Images