Semiautogenous grinding process stockpile particle size distribution prediction method based on digital twinning

Abstract

The application discloses a kind of semi-autogenous grinding process ore stock heap particle size distribution prediction methods based on digital twinning, it is related to mineral processing technical field, including: in combination with feed opening opening, feed opening flow, conveying belt speed and material level field, each voxel block in the causal voxel particle size state field of ore stock heap is executed out of pile release, particle size state migration and discharge accessible relationship synchronous update, generate the candidate discharge voxel set corresponding to each feed opening;In combination with semi-autogenous grinding feed actual measurement particle size distribution, each feed opening propagation delay, each voxel block to the release time delay in the corresponding feed opening in the pile, candidate discharge voxel set, voxel block release amount threshold and surface layer stable direct observation exclusion condition, the inversion voxel set corresponding to each feed opening is screened, and only in inversion voxel set inside execution local reverse inversion, obtain the residual particle size state of each voxel block.The application improves semi-autogenous grinding feed particle size distribution prediction reliability.

Description

Technical Field

[0001] This invention relates to the field of mineral processing technology, and in particular to a method for predicting the particle size distribution of ore stockpiles in a semi-autogenous grinding process based on digital twins. Background Technology

[0002] In the field of mineral processing, semi-autogenous grinding (SAG) technology is widely used due to its advantages such as simple process and large processing capacity. As a key buffer link connecting crushing and grinding processes, the particle size distribution of the material inside the ore stockpile directly affects the feed properties of the SMG, thus profoundly impacting grinding efficiency, energy consumption, and final beneficiation indicators. Therefore, accurate prediction of the particle size distribution of the ore stockpile discharge (i.e., SMG feed) is a core prerequisite for achieving pre-grinding ore blending optimization and intelligent full-process control. Current conventional methods mainly rely on statistical models based on historical data, simplified mass balance calculations, or discrete element simulations under offline operating conditions. These methods aim to establish estimation or simulation models of the ore stockpile particle size distribution through analysis of information such as the properties of the ore fed into the stockpile and stockpile / reclaiming operation records.

[0003] In existing methods, the particle size results detected at the semi-autogenous grinding feed end usually lag behind the ore release process inside the ore heap. Predictions based solely on the current feed inlet flow rate or surface observation results are difficult to accurately characterize the time delay correlation between the voxel block and the feed inlet, and between the feed inlet and the detection position. Furthermore, there are incomplete observation characteristics in the internal region of the ore heap. Relying solely on positive state updates makes it difficult to trace back the measured particle size deviation of the semi-autogenous grinding feed to the specific region within the heap, affecting the continuous correction of the particle size status in subsequent cycles. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for predicting the particle size distribution of ore storage piles in a semi-autogenous grinding process based on digital twins, which solves the problems of insufficient correlation between feed particle size prediction time delay and the difficulty in accurately backtracking and correcting particle size deviation under the conditions of incomplete in-pile observation in the existing technology.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for predicting the particle size distribution of a semi-autogenous grinding (SAG) ore stockpile based on digital twins. The method includes: determining the amount of material written into the stockpile and the initial discharge reachability of each voxel based on the effective particle size state of the stockpile in the previous cycle, the discharge reachability of the stockpile in the previous cycle, the current material level field, the current coarse crushing feed mass, the current coarse crushing feed particle size distribution, and the current stockpile location; constructing a causal voxel particle size state field for the current cycle's stockpile; and, in conjunction with the feed inlet opening, feed inlet flow rate, conveyor belt speed, and material level field, performing stockpile release, particle size state migration, and synchronous update of discharge reachability for each voxel in the causal voxel particle size state field of the stockpile, generating candidate discharge bodies corresponding to each feed inlet. Voxel sets are used to filter the inversion voxel sets corresponding to each feed port, combining the measured particle size distribution of the semi-autogenous grinding feed, the propagation delay of each feed port, the release delay of each voxel block to the corresponding feed port, the candidate discharge voxel sets, the voxel block release threshold, and the direct observation exclusion conditions for surface stability. Local inverse inversion is performed only within the inversion voxel sets to obtain the residual particle size state of each voxel block. The residual particle size state is written back to the causal voxel particle size state field of the ore storage pile in the current period to form the effective particle size state of the current period. The effective particle size state of the current period and the discharge reachability relationship of the current period are used as the input of the causal voxel particle size state field of the ore storage pile in the next period, and the semi-autogenous grinding feed particle size distribution prediction results are output in a rolling manner.

[0007] As a preferred embodiment of the method for predicting the particle size distribution of a semi-autogenous grinding process ore storage pile based on digital twins as described in this invention, the construction of the causal voxel particle size state field of the ore storage pile in the current period includes: establishing a unified spatial coordinate system for the ore storage pile, and mapping the outer boundary of the ore storage pile, the current pile surface boundary, the coarse crushing feed boundary, the positions of each feed port, the positions of each transfer point, and the positions of the material level detection points to the same coordinate system; discretizing the effective volume of the ore storage pile to form several voxel blocks; recording the spatial boundary of each voxel block, the geometric center position of the voxel block, and the adjacency relationship with adjacent voxel blocks to form a voxel topology table of the ore storage pile; reconstructing the current pile surface of the ore storage pile according to the current period material level field, and judging the voxel blocks according to the positional relationship between the geometric center of the voxel block and the current pile surface and the outer boundary of the ore storage pile, distinguishing between currently effective voxel blocks and currently invalid voxel blocks.

[0008] As a preferred embodiment of the method for predicting the particle size distribution of a semi-autogenous grinding process ore stockpile based on digital twins as described in this invention, the determination of the amount of voxel blocks written into the stockpile includes: identifying the set of surface voxel blocks directly covered by the current coarse crushing batch in the current effective voxel blocks, centered on the current stockpile location; constructing spatial bearing weights for each surface voxel block based on the horizontal distance, vertical height difference, and local slope of the current stockpile surface from the geometric center of each surface voxel block to the current stockpile location, and normalizing the spatial bearing weights of all surface voxel blocks participating in the bearing, to obtain the proportion of voxel blocks written into the stockpile; calculating the particle size mass vector written to the corresponding voxel block in the current cycle based on the total mass of the current coarse crushing batch, the particle size mass fraction vector corresponding to the current coarse crushing particle size distribution, and the proportion of voxel blocks written into the stockpile, to obtain the amount of voxel blocks written into the stockpile.

[0009] As a preferred embodiment of the method for predicting the particle size distribution of ore storage piles based on digital twins in the semi-autogenous grinding process described in this invention, the initial discharge reachability relationship includes: constructing the geometric connectivity factor, historical release path factor, and operating condition constraint factor of the voxel block facing the feed port and performing weighted fusion, and normalizing the weighted fusion result of all feed ports to obtain the initial discharge reachability relationship of the voxel block facing each feed port.

[0010] As a preferred embodiment of the method for predicting the particle size distribution of ore stockpiles in a semi-autogenous grinding process based on digital twins as described in this invention, the step of combining the feed port opening, feed port flow rate, conveyor belt speed, and material level field includes: performing effectiveness processing on the flow rate of each feed port according to the conveyor belt speed of the current cycle; when the conveyor belt speed on the corresponding conveyor path of the feed port is not less than the effective operating speed threshold of the belt, the effective feed flow rate of the feed port is taken as the feed port flow rate of the current cycle; when the conveyor belt speed on the corresponding conveyor path of the feed port is less than the effective operating speed threshold of the belt, the effective feed flow rate of the feed port is taken as zero; and calculating the required mass of each feed port for the current cycle based on the effective feed flow rate of the feed port and the time interval of the current cycle.

[0011] As a preferred embodiment of the digital twin-based semi-autogenous grinding process ore stockpile particle size distribution prediction method described in this invention, the stockpile release includes: calculating and normalizing the release contribution weight of the voxel block to the feed port based on the discharge accessibility of the updated voxel block facing the feed port and the total ore mass within the updated voxel block, to obtain the release allocation ratio of the voxel block to the feed port; calculating the pre-release particle size mass vector of the voxel block facing the feed port based on the stockpile discharge demand mass and the release allocation ratio of the voxel block to the feed port; and applying a release upper limit constraint to the pre-release particle size mass vector of the voxel block to obtain the actual release particle size mass vector of the voxel block facing the feed port.

[0012] As a preferred embodiment of the method for predicting the particle size distribution of a semi-autogenous grinding process ore stockpile based on digital twins as described in this invention, the synchronous update of particle size state migration and discharge reachability includes: calling a particle size migration coefficient that matches the current operating conditions based on the current cycle feed port opening, feed port flow rate, conveyor belt speed, and material level field; updating the particle size state of the voxel block based on the remaining ore after stockpile release and the particle size migration coefficient; and performing a quality-weighted fusion of the discharge reachability of the source voxel block based on the remaining ore mass of the source voxel block migrating to the target voxel block to obtain the updated discharge reachability of the target voxel block.

[0013] As a preferred embodiment of the method for predicting the particle size distribution of ore stockpiles in a semi-autogenous grinding process based on digital twins as described in this invention, the generation of candidate discharge voxel sets corresponding to each feed port includes: a first condition for candidate discharge is that the total ore mass of the voxel block after the current cycle update is greater than zero; a second condition for candidate discharge is that the discharge accessibility of the voxel block facing the feed port is not less than a candidate accessibility threshold; and a third condition for candidate discharge is that the actual release mass of the voxel block facing the feed port in the current cycle is not less than a voxel block release amount threshold. Voxel blocks that simultaneously meet the first, second, and third conditions for candidate discharge are included in the candidate discharge voxel set of the corresponding feed port.

[0014] As a preferred embodiment of the method for predicting the particle size distribution of a semi-autogenous grinding (SAG) ore stockpile based on digital twins as described in this invention, the method of combining the measured particle size distribution of the SAG feed, the propagation delay of each feed port, the release delay of each voxel block to the corresponding feed port within the stockpile, the candidate discharge voxel set, the voxel block release threshold, and the direct observation exclusion condition for surface stability includes: reading the measured particle size distribution of the SAG feed obtained at the feed detection position in the current cycle, and converting the measured particle size distribution of the SAG feed into a measured particle size mass vector of the SAG feed; aligning the actual release particle size mass vector of each voxel block facing the feed port in the corresponding historical cycle to the current cycle according to the feed port propagation delay of each feed port, to obtain the predicted particle size mass vector of the SAG feed; using the difference between the measured particle size mass vector of the SAG feed and the predicted particle size mass vector of the SAG feed as the total error vector of the current cycle, and distributing the total error vector of the current cycle to each feed port to obtain the local error vector corresponding to each feed port.

[0015] As a preferred embodiment of the digital twin-based method for predicting the particle size distribution of ore stockpiles in a semi-autogenous grinding process as described in this invention, the method for obtaining the residual particle size state of each voxel block includes: determining the historical alignment time of the voxel block facing the feed port based on the feed port propagation delay from the feed port to the semi-autogenous grinding feed detection position and the in-stock release delay of the voxel block to the corresponding feed port; including voxel blocks that belong to the candidate discharge voxel set at the historical alignment time, whose actual release mass facing the corresponding feed port at the historical alignment time is not less than the voxel block release amount threshold, and which do not belong to the stable directly observed surface area, into the inversion voxel set of the corresponding feed port; determining the error attribution weight only within the inversion voxel set based on the discharge accessibility and actual release mass of the voxel block at the historical alignment time, and obtaining the residual particle size state of the voxel block based on the error attribution weight and the local error vector of the corresponding feed port.

[0016] The beneficial effects of this invention are as follows: By synchronously updating the particle size state migration and discharge reachability relationship, the particle size state of the voxel block is kept in line with the outlet of the feed port during the release process, which improves the accuracy of generating candidate discharge voxel sets; by screening the inversion voxel sets corresponding to each feed port and performing local inverse inversion only within the inversion voxel sets, the deviation corresponding to the measured particle size distribution of the semi-autogenous grinding feed can be traced back to the relevant voxel blocks along the propagation delay and the release delay in the pile, which improves the reliability of the particle size distribution prediction of the semi-autogenous grinding feed. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for predicting the particle size distribution of ore stockpiles in a semi-autogenous grinding process based on digital twins.

[0019] Figure 2 A flowchart for constructing the causal voxel-level state field of the current period's ore reservoir.

[0020] Figure 3 A flowchart for synchronously updating the discharge release, particle size state migration, and discharge reachability relationship.

[0021] Figure 4 The flowchart shows the process of selecting the voxel set for inversion, local inverse inversion, and writing back the residuals.

[0022] Figure 5 This is a convergence curve of multi-period prediction error.

[0023] Figure 6Heatmap for residual particle size distribution. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0027] Reference Figures 1-6 As one embodiment of the present invention, this embodiment provides a method for predicting the particle size distribution of ore reserves in a semi-autogenous grinding process based on digital twins, comprising the following steps: S1. Based on the effective particle size state of the ore storage pile in the previous cycle, the reachability relationship of the ore storage pile in the previous cycle, the current material level field, the current coarse crushing input quality, the current coarse crushing input particle size distribution, and the current material pile location, determine the input writing amount and initial output reachability relationship of each voxel block, and construct the causal voxel particle size state field of the ore storage pile in the current cycle.

[0028] Furthermore, using the geometric center of the bottom plate of the ore storage pile as the origin of the three-dimensional coordinate system, a unified spatial coordinate system for the ore storage pile is established. The outer boundary of the ore storage pile, the current pile surface boundary, the coarse crushing feed boundary, the positions of each feed port, the positions of each transfer point, and the positions of the material level detection points are uniformly mapped to the same coordinate system. The effective volume of the ore storage pile is discretized according to the length, width, and height directions to form several voxel blocks. For each voxel block, the spatial boundary, the position of the geometric center of the voxel block, and the adjacency relationship with adjacent voxel blocks are recorded to form a voxel topology table of the ore storage pile. After obtaining the spatial discretization results of the voxel blocks, the current periodic material level field is read, the current pile surface of the ore storage pile is reconstructed, and according to the reconstructed pile surface shape, if the geometric center of the voxel block is below the current pile surface and within the outer boundary of the ore storage pile, the voxel block is determined to be a valid voxel block; if the geometric center of the voxel block is above the current pile surface or outside the outer boundary of the ore storage pile, the voxel block is determined to be an invalid voxel block.

[0029] It should be noted that currently invalid voxel blocks are not included in the writing of coarse crushed ore into the pile and the assignment of the initial discharge reachability relationship in this cycle.

[0030] Furthermore, the total mass of the current coarse crushing batch, the particle size mass fraction vector of the current coarse crushing batch, and the current stockpile location are read, and the particle size class corresponding to the current coarse crushing batch is recorded sequentially as follows: When determining the spatial coverage of the current coarse crushing batch, the set of surface voxels directly covered by the current coarse crushing batch is identified in the current effective voxel blocks, with the current stockpile location as the center. Based on the horizontal distance, vertical height difference, and local slope of the current stockpile surface from the geometric center of each surface voxel block to the current stockpile location, a spatial bearing weight is constructed for each surface voxel block. The spatial bearing weights of all surface voxel blocks participating in the bearing are normalized to obtain the stockpile writing ratio of the voxel blocks.

[0031] in, Indicates the total number of particle size levels; for voxel blocks The corresponding spatial acceptance weight is denoted as , Indicates the sequence number is voxel blocks, It represents the current periodic moment measured at a unified time zero point.

[0032] It should be noted that the spatial acceptance weight is constructed as follows: the horizontal distance from the geometric center of the voxel block to the current stockpile location, the absolute value of the vertical height difference between the geometric center of the voxel block and the current stockpile location, and the local slope of the current stockpile surface at the location of the voxel block are obtained. The horizontal distance, the absolute value of the vertical height difference, and the local slope of the current stockpile surface are normalized according to the maximum value within the set of surface voxel blocks participating in the acceptance, to obtain the normalized horizontal distance, the normalized vertical height difference, and the normalized slope. Then, the spatial acceptance weight of the voxel block is constructed based on the normalized horizontal distance, the normalized vertical height difference, and the normalized slope. The voxel block with the smaller horizontal distance, the smaller absolute value of the vertical height difference, and the closer the local slope of the current stockpile surface is to the natural stockpile slope of the current coarse crushing batch is, the larger the spatial acceptance weight is obtained.

[0033] It should be noted that the current particle size mass fraction vector of the coarse crushed material fed into the pile is represented as: ; in, This represents the particle size mass fraction vector of the current coarse crushed particles fed into the pile. Indicates the particle size level as The mass percentage of the current coarse crushed batches fed into the stockpile. Indicates the particle size level as The mass percentage of the current coarse crushed batches fed into the stockpile. Indicates the particle size level as The mass percentage of the current coarse crushing batches fed into the stockpile.

[0034] It should be noted that, All components in the equation are not less than 0 and satisfy the following conditions: .

[0035] in, This represents a vector of all 1s with the same dimension as the granularity level.

[0036] It should be noted that the spatial acceptance weights of all participating surface voxel blocks are normalized and expressed as follows: ; in, Represented as voxel blocks The percentage of writes into the heap. q represents the total number of voxel blocks, and q represents the voxel block number. Voxel blocks The spatial carrying capacity of the current coarse crushing batch is given. Indicates the sequence number is voxel blocks.

[0037] It should be noted that for surface voxel blocks directly covered by the current coarse crushing batch, the corresponding spatial acceptance weight is calculated according to the spatial acceptance weight construction method; for voxel blocks that do not participate in acceptance, the corresponding spatial acceptance weight is set to 0.

[0038] Furthermore, the particle size mass vector written to the voxel block in the current cycle is calculated, and the particle size mass vector written to the voxel block in the current cycle is used as the amount of material written to the voxel block in the current cycle. After determining the amount of material written to each voxel block, an initial material discharge reachability relationship is established for each voxel block facing each feed port.

[0039] It should be noted that the granularity mass vector written to the voxel block in the current cycle is expressed as: ; in, This indicates that the current cycle is written to the voxel block. granularity mass vector, This indicates the total mass of the current batch of coarse crushed material entering the stockpile.

[0040] It should be noted that, This is used to calculate the mass distribution of the current coarse crushing batch at each particle size level. Then, based on the writing ratio of the voxel blocks, the current batch of ore is written into the corresponding voxel blocks according to the spatial connection relationship, thus obtaining the amount of voxel blocks written into the pile. By calculating the particle size mass vector written into the voxel blocks in the current cycle, the amount of writing obtained by different voxel blocks reflects not only the particle size composition of the current coarse crushing batch, but also the influence of the current stockpile location and the current pile surface geometry on the ore falling and diffusion.

[0041] The initial output reachability relationship is used to represent the probability that the ore written into the voxel block in the current cycle will be released through each feed port in the future, so that the voxel block state includes the particle size composition at the current position and the future output correlation of the ore at the current position.

[0042] It should be noted that, to establish the initial discharge reachability relationship for each voxel block towards each feed port, specifically, geometric connectivity factor, historical release path factor, and operating condition constraint factor are constructed.

[0043] The feed port number is recorded as follows: ; The serial number is The feed port is recorded as voxel blocks Facing the feed port The geometric connectivity factor is denoted as Used to characterize voxel blocks to the feed port Geometric reachability of the current ore reservoir voxel topology; voxel blocks Facing the feed port The historical release path factor is denoted as Used to characterize voxel blocks under historical operating conditions and offline mechanism template conditions. Internal ore through the feed port The long-term tendency to release; the feed port The operating condition constraint factor under the current operating conditions is denoted as: Used to characterize the current operating conditions at the feed inlet The degree of constraint on continuous output capacity.

[0044] It should be noted that the geometric connectivity factor is calculated as follows: starting from the voxel block and ending at the discharge area corresponding to the feed port, the shortest feasible path is searched along the voxel path that does not cross the current invalid voxel block and satisfies the flow constraint from high to low. The distances between the geometric centers of adjacent voxel blocks on the shortest feasible path are accumulated to obtain the path length from the voxel block to the feed port. The height decrease of the geometric centers of adjacent voxel blocks on the shortest feasible path is accumulated to obtain the cumulative height difference from the voxel block to the feed port. The path length and cumulative height difference are normalized according to the maximum path length and the maximum cumulative height difference among all currently valid voxel blocks corresponding to the same feed port. The normalized path length result is inverted and then weighted and summed with the normalized cumulative height difference result to obtain the geometric connectivity factor of the voxel block facing the feed port.

[0045] Among them, the shorter the path length and the greater the cumulative height difference of the voxel block, the larger the corresponding geometric connectivity factor; the geometric connectivity factor ranges from 0 to 1. The larger the geometric connectivity factor, the easier it is for the ore in the voxel block to be released through the feed port in space.

[0046] It should be noted that the historical release path factor is obtained by statistically analyzing the release paths of discrete units with the same or similar spatial positions as the voxel block under similar working conditions based on discrete element simulation results and historical measured data. This yields the frequency of ore being released through the feed port in the current spatial region, and the frequency of ore being released through the feed port in the current spatial region is normalized to the historical release path factor.

[0047] It should be noted that the setting method for the operating condition constraint factors is as follows: The current feed port opening, current feed port operating status, and current material level in the area covered by the feed port are read to construct the feed port activation coefficient and the material level support coefficient. The feed port activation coefficient for the current cycle is denoted as... When the feed port is open, take When the feed port is closed, take Record the actual material level in the area covered by the feed inlet as... The minimum release level is denoted as Record the reference material level as Set the operating condition constraint factor to the product of the feed port activation coefficient and the material level support coefficient.

[0048] It should be noted that when When the material level support coefficient is reached, it is expressed as: ; in, This indicates the material level support coefficient at the feed port in the current cycle.

[0049] It should be noted that the minimum release level is set according to the minimum material layer thickness required to form a continuous material flow in the area covered by the feed port. The minimum release level is set to 1.2 times the effective opening height of the corresponding feed port. The reference level is set according to the target material layer thickness under stable continuous discharge conditions. The reference level is set to 2.5 times the effective opening height of the corresponding feed port.

[0050] It should be noted that when At that time, the material level support coefficient is taken as 0.

[0051] Furthermore, after obtaining the geometric connectivity factor, historical release path factor, and operating condition constraint factor, the initial discharge reachability fusion algorithm is used to determine the voxel block's orientation towards the feed port. The initial output can reach a certain level.

[0052] It should be noted that the initial output reachability fusion algorithm is expressed as: ; ; in, Voxel blocks Facing the feed port The initial discharge can reach a certain degree. The non-negative weight coefficients of the geometric connectivity factor are represented. This represents the non-negative weighting coefficient of the historical release path factor. This represents the non-negative weighting coefficient of the operating condition constraint factor. Indicates the feed port number. Indicates the total number of feed ports. Voxel blocks Facing the feed port geometric connectivity factor, Indicates the sequence number is The feed port, Voxel blocks Facing the feed port Historical release path factors, Indicates the feed port Operating condition constraint factors under the current operating conditions.

[0053] It should be noted that , as well as The calibration was determined using historical production data and trial operation data. Specifically, historical periods including feed inlet opening, feed inlet flow rate, conveyor belt speed, material level, measured particle size distribution of semi-autogenous mill feed, and historical release path records were selected as calibration samples; while meeting the following conditions... and , , All are not less than Under the given conditions, several candidate weight combinations are generated according to the step size; the initial discharge reachability of the voxel block facing each feed port within the historical period is calculated using each candidate weight combination, and the prediction deviation is calculated based on the actual release path and the measured particle size distribution of the semi-autogenous grinding feed in the corresponding historical period; the candidate weight combination with the smallest prediction deviation is taken as the current ore storage pile operating condition category. , as well as The prediction deviation is evaluated by the difference between the predicted particle size mass vector of the semi-autogenous grinding feed and the measured particle size mass vector of the semi-autogenous grinding feed; when the difference in the internal spatial channels of the ore pile is greater, the candidate weight combination... The range of values ​​for this increases accordingly; when the historical release path has higher repetition, the candidate weight combination... The range of values ​​for this parameter increases accordingly; when the feed inlet starts and stops more frequently or the material level fluctuates more significantly, the candidate weight combination... The range of values ​​for increases accordingly.

[0054] It should be noted that, through the initial discharge accessibility fusion algorithm, the geometric connectivity of the voxel block facing the feed port, the historical release path tendency, and the current working condition allowance are weighted and fused, and then the weighted results of all feed ports are normalized to obtain the initial accessibility of the ore written into the voxel block in the current period that will be released through the feed port in the future; the larger the initial discharge accessibility of the voxel block facing the feed port, the greater the possibility that the ore currently written into the voxel block will be released through the current feed port in the future.

[0055] It should be noted that when the number of fusion molecules facing all feed ports of the voxel block is 0, the initial discharge accessibility of the voxel block is determined according to the principle of prioritizing the nearest feed port. Specifically, when there is an active feed port in the current cycle, the feed port with the shortest distance to the geometric center of the voxel block is selected as the selected feed port, and the initial discharge accessibility of the selected feed port is assigned a value of 1. The initial discharge accessibility of feed ports that are inconsistent with the selected feed port is assigned a value of 0. When there is no active feed port in the current cycle, the feed port with the shortest distance to the geometric center of the voxel block is selected as the selected feed port, and the initial discharge accessibility of the selected feed port is assigned a value of 1. The initial discharge accessibility of feed ports that are inconsistent with the selected feed port is assigned a value of 0. When the number of feed ports with the same distance to the geometric center of the voxel block is not less than 2 and all meet the selection conditions, the feed port with the highest sequence number is selected as the selected feed port.

[0056] Furthermore, after obtaining the infeed writing amount and initial discharge reachability of each voxel block in the current cycle, a causal voxel grain size state field of the ore storage heap for the current cycle is constructed. Specifically, the grain size state after writing in the current cycle and the discharge reachability of the current cycle are simultaneously registered in the voxel block state table. For voxel blocks participating in the writing in the current cycle, the grain size state of the voxel block in the current cycle is updated to the superposition result of the effective grain size state of the ore storage heap in the previous cycle and the infeed writing amount in the current cycle. Based on the total ore mass corresponding to the effective grain size state of the ore storage heap in the previous cycle and the total ore mass corresponding to the infeed writing amount in the current cycle, the discharge reachability of the ore storage heap in the previous cycle and the initial discharge reachability of the current cycle are adjusted. The reachability relationships are weighted and fused to obtain the current cycle's output reachability relationship. Specifically, when the total ore mass corresponding to the effective particle size state of the ore heap in the previous cycle is 0, the initial output reachability relationship of the current cycle is used as the current cycle's output reachability relationship. When the total ore mass corresponding to the current cycle's input write amount is 0, the output reachability relationship of the ore heap in the previous cycle is used as the current cycle's output reachability relationship. For voxel blocks that did not participate in the current cycle's write but are still located within the current effective heap, they inherit the effective particle size state of the ore heap in the previous cycle and the output reachability relationship of the ore heap in the previous cycle. For voxel blocks located outside the current effective heap, their corresponding particle size state and output reachability relationship are both set to zero.

[0057] It should be noted that each effective voxel block in the causal voxel particle size state field of the current cycle ore storage heap simultaneously possesses the particle size state information within the voxel block corresponding to the current cycle and the initial discharge reachability information of the voxel block facing each feed port in the current cycle.

[0058] S2. Combining the feed port opening, feed port flow rate, conveyor belt speed, and material level field, perform out-of-stock release, particle size state migration, and synchronous update of discharge reachability relationship for each voxel block in the causal voxel particle size state field of the ore storage pile, and generate a candidate discharge voxel set corresponding to each feed port.

[0059] Furthermore, the opening degree of each feed port, the flow rate of each feed port, the speed of the conveyor belt, and the material level field of the current cycle are read. The particle size state and discharge reachability of each voxel block are read from the causal voxel particle size state field of the current cycle. The flow rate of each feed port is processed according to the speed of the conveyor belt in the current cycle to obtain the effective feed flow rate of the feed port.

[0060] Among them, the precursor block will be updated. The granularity state in the current cycle is denoted as The precursor block will be updated. The reachability of material discharge to each feed port is denoted as: ;Will Corresponding feed port The output can reach the quantity denoted as Used to indicate the update of the precursor block Facing the feed port The output can reach a certain level.

[0061] It should be noted that the flow rate of each feed port is processed based on the current conveyor belt speed. Specifically, when the conveyor belt speed on the corresponding conveyor path of the feed port is not less than the effective belt running speed threshold, the effective feed flow rate of the feed port is taken as the feed flow rate of the current cycle; when the conveyor belt speed on the corresponding conveyor path of the feed port is less than the effective belt running speed threshold, the effective feed flow rate of the feed port is taken as 0.

[0062] It should be noted that the effective operating speed threshold of the belt is set according to the zero speed protection setting value of the corresponding conveyor belt and the minimum stable operating speed setting of the frequency converter. The value range is usually 3% to 10% of the rated belt speed of the corresponding conveyor belt.

[0063] It should be noted that the effective feed flow rate at the feed port is used to characterize the actual discharge demand that can enter the subsequent conveying chain within the current cycle; by introducing the conveyor belt speed to effectively process the feed port flow rate, it is prevented that the feed port flow rate will still be regarded as the effective discharge release demand when the conveyor belt stops or operates at a low speed.

[0064] Furthermore, based on the effective feed flow rate of each feed port, the required mass of each feed port for the current cycle is calculated.

[0065] It should be noted that the required mass of material leaving the pile at each feed port in the current cycle is expressed as: ; in, This indicates the required feed quality at the feed inlet during the current cycle. This indicates the effective feed flow rate at the feed port in the current cycle. This indicates the time interval of the current cycle.

[0066] Furthermore, based on the discharge availability relationship and the current cycle's discharge demand mass, the discharge release amount for the current cycle is allocated among each voxel block. For both voxel blocks and feed ports, the release contribution weight of the voxel blocks to the feed ports is calculated. The release contribution weight of each voxel block under the same feed port is normalized to obtain the release allocation ratio of the voxel blocks to the feed ports. Based on the discharge demand mass of the feed ports and the release allocation ratio of the voxel blocks, the pre-release particle size mass vector of the voxel blocks facing the feed ports is calculated.

[0067] It should be noted that the weighting of the voxel block's contribution to the release at the feed port is expressed as follows: ; in, Voxel blocks For the feed port Release contribution weight, Voxel blocks Total ore mass before update.

[0068] It should be noted that the release contribution weights of each bulk block under the same feed port are normalized and expressed as follows: ; in, Voxel blocks For the feed port The release distribution ratio, Indicates the voxel block number that participated in the normalization calculation. Voxel blocks For the feed port The release contribution weight.

[0069] It should be noted that when When the value is set to 0, it indicates that no voxel block contributes effectively to the release of the feed port in the current cycle. In this case, the release distribution ratio of the feed port to all voxel blocks is set to 0.

[0070] It should be noted that the pre-release particle size mass vector of the voxel block facing the feed port is expressed as: ; in, Voxel blocks Facing the feed port The pre-release particle size mass vector.

[0071] It should be noted that when At that time, Set it to the zero vector.

[0072] Furthermore, a release upper limit constraint is imposed on the pre-release particle size mass vector of the voxel block. Specifically, the release correction coefficient of the voxel block is calculated. Based on the release correction coefficient of the voxel block, the actual release particle size mass vector of the voxel block facing the feed port is calculated. The actual release particle size mass vectors of the voxel block facing all feed ports are summarized to obtain the voxel block's out-of-pile release particle size mass vector in the current cycle.

[0073] It should be noted that the upper limit constraint on the pre-release particle size mass vector of the voxel block is used to prevent the release amount of the same voxel block from exceeding the actual ore amount in the voxel block in the current cycle due to the simultaneous discharge demand from several feed ports from the associated area of ​​the same voxel block.

[0074] It should be noted that the release correction factor for the voxel block is expressed as: ; in, Voxel blocks The release correction factor.

[0075] It should be noted that when At that time, Set to 0.

[0076] It should be noted that the actual released particle size mass vector of the voxel block facing the feed port is expressed as: ; in, Voxel blocks The actual released particle size mass vector facing the feed port in the current cycle.

[0077] Furthermore, the actual release particle size mass vectors of the voxel block facing all feed ports are summarized to obtain the voxel block's discharge particle size mass vector for the current cycle: ; in, Voxel blocks The total granularity mass vector released in the current cycle.

[0078] It should be noted that by calculating the particle size mass vector of the voxel block released from the pile in the current cycle, the release of the voxel block is constrained by the current feed inlet flow rate, the operating status of the conveyor belt, the current ore mass of the voxel block, and the discharge reachability of the voxel block facing the feed inlet, so as to prevent the average deduction of the entire ore pile based solely on the feed inlet flow rate.

[0079] Furthermore, after calculating the heap release, particle size migration is performed on the remaining ore in the voxel block. Specifically, based on the current cycle feed port opening, feed port flow rate, conveyor belt speed, and material level field, a particle size migration coefficient matching the current working condition is called from a pre-established offline mechanism migration template library.

[0080] Among them, the source voxel block To target voxel block The particle size transfer coefficient is denoted as .

[0081] It should be noted that the particle size migration coefficient is determined jointly by discrete element simulation results, historical measured verification results, and current operating conditions for each source voxel. The corresponding migration coefficients of each target voxel block satisfy: .

[0082] It should be noted that the offline mechanism migration template library is obtained through discrete element simulation results and historical production verification data. Specifically, it is divided into several working condition types according to the combination of feed port opening degree, feed port flow rate, conveyor belt speed level, and material level zoning. For each working condition type, discrete element simulation consistent with the geometric boundary of the ore storage pile, feed port arrangement, ore angle of repose, ore particle size composition, and feed port discharge conditions is used to statistically analyze the proportion of remaining ore in the source voxel block that migrates to each target voxel block in one cycle, thus obtaining the initial particle size migration coefficient. The initial particle size migration coefficient is verified and corrected by the material level changes, feed port flow rate, conveyor belt speed, and semi-autogenous grinding feed particle size detection results recorded during historical production, forming an offline mechanism migration template library indexed by working condition type, source voxel block, and target voxel block. During online prediction, the current working condition type is determined based on the feed port opening degree, feed port flow rate, conveyor belt speed, and material level field of the current cycle, and the corresponding particle size migration coefficient is read from the offline mechanism migration template library.

[0083] Furthermore, the voxel block is updated based on the remaining ore and particle size migration coefficient after being released from the pile. The granularity state.

[0084] It should be noted that updating voxel blocks The granularity state is represented as: ; in, This indicates the voxel block updated in the current cycle. Particle size state, Indicates the original voxel block before the update. Particle size state, Represents source voxel block The granularity mass vector of the current cycle's heap release. Represents source voxel block To target voxel block The particle size transfer coefficient.

[0085] Furthermore, while the particle size state is shifting, the output accessibility relationship of each voxel block is updated synchronously. Specifically, the calculation of the output accessibility relationship from the source voxel block is performed. Migration to target voxel block The remaining ore mass; based on the remaining ore mass of different source voxels migrating to the target voxel block, the output accessibility relationship of the source voxels is weighted and fused to obtain the target voxel block. Updated output reachability relationship.

[0086] It should be noted that the calculation is performed from the source voxel block. Migration to target voxel block The remaining ore mass is expressed as: ; in, Indicates that the current cycle is generated by the source block. Migration to target voxel block The remaining ore quality.

[0087] It should be noted that the mass-weighted fusion of the output reachability relationship of the source block is expressed as follows: ; in, This indicates the voxel block updated in the current cycle. The output can reach a certain relationship. Indicates the original voxel block before the update. The output can reach the relationship.

[0088] It should be noted that when At that time, Set it to the zero vector.

[0089] Furthermore, based on the updated particle size distribution, the updated discharge reachability relationship, and the particle size mass vector of the current cycle's discharge release, voxel blocks that simultaneously meet the candidate discharge criteria are identified. The set of candidate discharge volumetric elements included in the feed port.

[0090] It should be noted that the candidate discharge determination conditions include the first candidate discharge condition, the second candidate discharge condition, and the third candidate discharge condition. When the first candidate discharge condition, the second candidate discharge condition, and the third candidate discharge condition are all met simultaneously, the candidate discharge determination conditions are determined to be met.

[0091] It should be noted that the first condition for candidate material output is specifically, voxel blocks. The total ore mass is greater than zero after the current cycle update.

[0092] It should be noted that the second condition for candidate output is specifically, voxel blocks. The reachability of the material discharged from the feed port is not less than the candidate reachability threshold.

[0093] It should be noted that the candidate reachability threshold is determined based on the opening of the feed port, the flow rate of the feed port, and the material level in the area covered by the feed port, and the value range is usually 0.05 to 0.3.

[0094] It should be noted that the third condition for candidate output is specifically, voxel blocks. The actual release mass facing the feed port in the current cycle is not less than the voxel block release threshold.

[0095] It should be noted that the voxel block release threshold is determined based on the minimum effective resolution of the semi-autogenous grinding feed particle size detection device and the minimum stable discharge rate of the feed port. The voxel block release threshold is usually set with the minimum effective resolution of the semi-autogenous grinding feed particle size detection device as the lower limit and 20% of the minimum stable discharge rate of the feed port within a sampling period as the upper limit.

[0096] Furthermore, the particle size status updated in the current cycle, the discharge reachability relationship updated in the current cycle, the particle size mass vector of the discharge in the current cycle, the actual discharge particle size mass vector of each feed port in the current cycle, and the candidate discharge voxel set corresponding to each feed port are stored in the voxel block status table of the current cycle.

[0097] S3. Combining the measured particle size distribution of the semi-autogenous grinding feed, the propagation delay of each feed port, the release delay of each voxel block to the corresponding feed port in the pile, the candidate discharge voxel set, the voxel block release threshold, and the direct observation exclusion conditions for surface stability, the inversion voxel set corresponding to each feed port is screened, and local inverse inversion is performed only within the inversion voxel set to obtain the residual particle size state of each voxel block.

[0098] Furthermore, the measured particle size distribution of the semi-autogenous grinding feed obtained at the semi-autogenous grinding feed detection position in the current cycle is read, and combined with the mass of the semi-autogenous grinding feed belt scale in the current cycle, the measured particle size distribution of the semi-autogenous grinding feed is converted into a measured particle size mass vector of the semi-autogenous grinding feed. The measured particle size mass vector of the semi-autogenous feed in the current cycle is denoted as: .

[0099] Furthermore, the propagation delay from each feed port to the semi-autogenous grinding feed detection position is read, and the feed port... The propagation time delay from the feed inlet to the semi-autogenous grinding feed detection position is denoted as .

[0100] in, According to the feed port The length of the conveying path between the semi-autogenous grinding feed detection position, the speed of the conveyor belt in the current cycle, and the dwell time at the transfer point are determined.

[0101] Furthermore, the candidate discharge voxel set corresponding to each feed port, the actual release particle size mass vector of each voxel block facing each feed port, and the effective feed flow rate of each feed port are read from the current cycle voxel block status table. Based on the feed port propagation delay of each feed port, the actual release particle size mass vector of each voxel block facing the feed port in the corresponding historical cycle is aligned to the current cycle. The alignment results of all feed ports are summarized to obtain the predicted particle size mass vector of the semi-autogenous grinding feed for the current cycle. This predicted particle size mass vector of the semi-autogenous grinding feed for the current cycle is denoted as... .

[0102] It should be noted that when the historical release cycle corresponding to the feed port propagation delay is inconsistent with the current cycle, the candidate discharge voxel set of the corresponding feed port and the actual release particle size mass vector of each voxel block facing the corresponding feed port are read from the voxel block status table corresponding to the historical release cycle, and the actual release particle size mass vector in the historical release cycle is aligned to the current cycle; when the historical release cycle corresponding to the feed port propagation delay is consistent with the current cycle, the candidate discharge voxel set of the corresponding feed port and the actual release particle size mass vector of each voxel block facing the corresponding feed port are read from the voxel block status table of the current cycle.

[0103] It should be noted that the predicted particle size mass vector of the semi-autogenous grinding feed is obtained by aligning and summarizing the particle size mass vectors released by each feed port in the corresponding historical period, and is used to compare with the measured particle size mass vector of the semi-autogenous grinding feed.

[0104] Furthermore, the difference between the measured particle size mass vector and the predicted particle size mass vector of the semi-autogenous grinding feed in the current cycle is used as the total error vector for the current cycle. Based on the effective feed flow rate ratio after alignment through the feed port propagation delay, the total error vector for the current cycle is distributed to each feed port, resulting in the local error vector corresponding to each feed port. The in-pile release delay of each voxel block to the corresponding feed port is determined. The total error vector of the current period is denoted as... ; feed port The corresponding local error vector is denoted as ; voxel blocks to the feed port The in-heap release delay is denoted as .

[0105] It should be noted that the in-stack release delay is obtained through an offline mechanism template lookup table. Specifically, the current cycle operating condition type is determined based on the feed port opening combination, feed port flow combination, material level zone, and stack point zone; then, voxel blocks are read from the offline mechanism template library. Under the current cycle operating condition, the material is fed into the feed port. The average release time is used as the in-pile release delay. .

[0106] Furthermore, based on the feed port propagation delay and the in-stack release delay, a dual-delay history alignment algorithm is used to reverse-engineer the voxel block. Historical alignment moments for the current cycle's semi-autogenous feeding error.

[0107] Among them, voxel blocks Facing the feed port Historical alignment moments are recorded as , It should be noted that the dual-delay history alignment algorithm is expressed as: ; in, Voxel blocks Facing the feed port This allows us to explain the historical alignment moments of the current cycle's semi-autogenous grinding feed error. Indicates according to periodic time intervals The calculation result is taken as the closest historical period moment.

[0108] It should be noted that the dual-delay history alignment algorithm is used to trace back the current cycle error at the semi-autogenous grinding feed detection position along the conveying propagation path and the in-pile release path to the historical cycle. Based on the fact that the error observed at the semi-autogenous grinding feed detection position in the current cycle is formed by the combined delay of the voxel block's in-pile release process and the conveying propagation process within the historical cycle, the feed port propagation delay and the in-pile release delay are subtracted from the current cycle time to obtain the voxel block... The corresponding historical alignment time.

[0109] Furthermore, based on the exclusion criteria for direct observation of surface stability, the direct observation stability markers of the voxel block were determined.

[0110] Among them, voxel blocks The direct observation stability marker for the current period is denoted as .

[0111] It should be noted that the direct observation stability indicator for determining voxel blocks is specifically that when the voxel block... Located within the visible shell of the current stack surface, continuously covered by the surface block size detection device or stack surface reconstruction results within a certain number of recent sampling periods (e.g., within the last 5 to 20 sampling periods), and voxel blocks within a certain number of recent sampling periods When the corresponding surface grain size fluctuation is not greater than the fluctuation threshold, take ; when the voxel block When located outside the visible shell of the current stack surface, take ; when the voxel block If it has not been continuously covered within a certain number of recent sampling periods, take ; when the voxel block When the surface granularity fluctuation exceeds the fluctuation threshold within a certain number of recent sampling periods, take... .

[0112] The reason for using the most recent 5 to 20 sampling periods is that when there are fewer than 5 sampling periods, the surface block size detection results are easily affected by single material drop impacts, changes in lighting, or local shading; when there are more than 20 sampling periods, the surface morphology of the ore pile may have changed significantly, and the representativeness of the early observation results for the current surface stability is reduced. Therefore, the most recent 5 to 20 sampling periods can achieve a balance between observation stability and pile shape timeliness.

[0113] It should be noted that the current visible shell layer of the stockpile surface is determined based on the current stockpile surface obtained from the current periodic material level field reconstruction, the effective observation range of the surface block size detection device, and the voxel block height. Specifically, a vertical projection is made onto the geometric center of the voxel block to obtain the height of the current stockpile surface at the vertical projection position. When the geometric center of the voxel block is below the current stockpile surface, and the vertical distance from the geometric center of the voxel block to the current stockpile surface is not greater than the thickness of the visible shell layer, the voxel block is determined to be located within the visible shell layer of the current stockpile surface. When the vertical distance from the geometric center of the voxel block to the current stockpile surface is greater than the thickness of the visible shell layer, or when the geometric center of the voxel block is above the current stockpile surface, the voxel block is determined to be located outside the visible shell layer of the current stockpile surface. The thickness of the visible shell layer is determined based on the height of the voxel block and the stable identification depth of the stockpile surface block size by the surface block size detection device, and is taken as one to three times the height of the voxel block.

[0114] It should be noted that the surface particle size fluctuation is the statistical value of the fluctuation of the surface particle size mass fraction of the same voxel block within the most recent sampling period; the fluctuation threshold is set according to the repeatability test results of the surface block size detection device and the statistical results of the natural fluctuation of the surface particle size of the ore pile, and the value range is usually 0.03 to 0.1.

[0115] It should be noted that, Voxel blocks It belongs to the surface region that can be directly observed stably; Voxel blocks It does not belong to the surface region that is stable and directly observable.

[0116] Furthermore, based on the voxel block release threshold and combined with historical alignment times, candidate discharge voxel sets, and directly observed stability indicators, the inverted voxel set corresponding to the feed port is selected.

[0117] It should be noted that the set of inverted voxels corresponding to the screening feed port is represented as: ; in, Indicates the feed port In the set of inverted voxels corresponding to the current period Indicates the feed port at the historical alignment point. The corresponding candidate output voxel set, Voxel blocks Facing the feed port at the historical alignment time The actual release particle size mass vector, Voxel blocks Facing the feed port at the historical alignment time The actual release mass, This indicates the threshold for voxel block release.

[0118] Furthermore, local inverse inversion is performed only within the inversion voxel set. Specifically, for voxel blocks within the inversion voxel set, the voxel blocks are calculated. For the feed port Error attribution weights.

[0119] It should be noted that the calculation voxel block For the feed port The error attribution weight is expressed as: ; in, Voxel blocks For the feed port Error attribution weights, Voxel blocks Facing the feed port at the historical alignment time The output can reach a certain degree. Voxel blocks Facing the feed port at the corresponding historical alignment time The output can reach a certain degree. Voxel blocks Facing the feed port at the corresponding historical alignment time The actual release particle size mass vector.

[0120] It should be noted that the error attribution weight is jointly determined by the discharge accessibility and actual release mass at the historical alignment time; the voxel block with a larger discharge accessibility and a larger actual release mass will have a greater impact on the feed port in the current cycle. Local errors have higher attribution weights; when the inverted voxel set is empty, error attribution weights are not calculated, and the feed port is... The local residual allocation for each voxel block is set to a zero vector.

[0121] Furthermore, based on voxel blocks For the feed port Error attribution weights and feed inlet The local error vector is used to calculate the feed port. voxel blocks The local residual allocation.

[0122] It should be noted that the calculation of the feed inlet... voxel blocks The local residual distribution is expressed as: ; in, Indicates the feed port in the current cycle. voxel blocks The local residual allocation.

[0123] It should be noted that when voxel blocks Not a feed port When retrieving the corresponding voxel set, Set it to the zero vector.

[0124] Furthermore, align each feed port with the voxel block. The local residual allocations are summarized to obtain the voxel blocks. The residual granularity state.

[0125] It should be noted that each feed port should be aligned with the voxel block. The local residual distributions are summarized and expressed as follows: ; in, Indicates the current periodic voxel block The residual granularity state.

[0126] It should be noted that the residual particle size state is used to characterize the voxel block particle size correction amount obtained by back-calculating the error between the measured particle size distribution of the semi-autogenous grinding feed and the predicted particle size distribution of the semi-autogenous grinding feed.

[0127] Furthermore, the inverted voxel set, the local error vector of the feed port, the local residual allocation, and the residual granularity state of each voxel block are stored in the voxel block state table of the current period.

[0128] S4. Write the residual particle size state back to the causal voxel particle size state field of the ore storage pile in the current cycle to form the effective particle size state of the current cycle. Use the effective particle size state of the current cycle and the discharge reachability relationship of the current cycle as the input of the causal voxel particle size state field of the ore storage pile in the next cycle, and output the prediction results of the semi-autogenous grinding feed particle size distribution in a rolling manner.

[0129] Furthermore, the voxel block particle size status, the voxel block output reachability relationship, and the voxel block residual particle size status are read from the voxel block status table of the current cycle; the residual particle size status is written back to the particle size status of the current cycle to form the effective particle size status of the current cycle.

[0130] Among them, the current periodic voxel block The effective granularity state is denoted as .

[0131] It should be noted that writing the residual granularity state back to the granularity state updated in the current period is represented as: ; in, Indicates the current periodic voxel block The effective granularity state, This indicates that non-negativity constraints are applied to each particle size level component in the particle size quality vector. It should be noted that, through When the write-back result corresponding to the current granularity level is less than 0, the quality corresponding to the current granularity level is set to 0; when the write-back result corresponding to the current granularity level is not less than 0, the write-back result corresponding to the current granularity level is retained.

[0132] Furthermore, based on the current effective particle size state, the current cycle's voxel block output reachability relationship is processed for state consistency. Specifically, when... At that time, retain the voxel block of the current period. The output can reach the relationship; when At that time, the current period voxel block The material output reachability relationship is set to zero vector.

[0133] It should be noted that the state consistency processing is used to prevent voxel blocks that no longer have effective ore quality from still retaining the discharge reachability relationship facing the feed port.

[0134] Furthermore, the current period's effective particle size state and the current period's output reachability relationship are simultaneously registered in the current period's voxel block state table, serving as input data for the next period's ore storage heap causal voxel particle size state field. Specifically, at the start of the next period's calculation, the current period's effective particle size state is used as the previous period's ore storage heap effective particle size state, and the current period's output reachability relationship is used as the previous period's ore storage heap output reachability relationship. After the residual particle size state is written back in each period, based on the current period's effective particle size state, the current period's output reachability relationship, and the next prediction... The feed port opening, feed port flow rate, conveyor belt speed, and material level field of the time window are updated synchronously according to the relationship between the discharge release and the discharge reachability to predict the discharge particle size mass of each feed port in the next prediction time window. Then, according to the feed port propagation delay from each feed port to the semi-autogenous mill feed detection position, the predicted release results of each feed port are aligned to the semi-autogenous mill feed detection position to obtain the predicted particle size mass vector of the semi-autogenous mill feed in the next prediction time window. The predicted particle size mass vector of the semi-autogenous mill feed in the next prediction time window is converted into the predicted particle size distribution of the semi-autogenous mill feed and output in a rolling manner.

[0135] It should be noted that the predicted particle size distribution of the semi-autogenous grinding feed includes the predicted mass corresponding to each particle size class and the predicted mass percentage corresponding to each particle size class. The predicted mass percentage corresponding to each particle size class is obtained by normalizing the predicted mass corresponding to each particle size class.

[0136] It should be noted that rolling output means that after each cycle completes the determination of the feed volume and initial discharge reachability, particle size state migration and discharge release and discharge reachability synchronous update, inversion voxel set screening and local inverse inversion, and residual particle size state write-back, the current cycle's effective particle size state and current cycle's discharge reachability continue to be used as the state input for the next cycle's calculation. In subsequent cycles, the feed volume writing, particle size state migration, discharge release, discharge reachability synchronous update, local inverse inversion, and residual particle size state write-back are repeatedly executed to continuously output the semi-autogenous grinding feed particle size distribution prediction results.

[0137] In this embodiment, in order to verify the continuous correction capability of the ore storage pile particle size state in multi-cycle prediction, four schemes were constructed: complete scheme, scheme with no in-pile release delay, scheme with full pile voxel residual allocation, and scheme without residual write-back. The prediction error of semi-autogenous grinding feed particle size distribution of each scheme in continuous cycles was recorded. Figure 5 This is the curve showing the change in prediction error over the entire cycle; from Figure 5 It can be seen that the prediction error corresponding to the complete scheme decreases faster as the cycle progresses and remains at a low level in the later stages of the cycle; the error reduction of the schemes for canceling the in-pile release delay and the schemes for all-pile voxel residual allocation is smaller, and the error of the scheme without residual write-back remains at a high level in the later stages. Figure 5 This indicates that after introducing the propagation delay of each feed port, the release delay within the pile, the inversion voxel set screening, and the write-back of the residual particle size state, the causal voxel particle size state field of the ore storage pile can continuously absorb the detection deviation of the semi-autogenous grinding feed end in the rolling calculation, making the prediction results of the semi-autogenous grinding feed particle size distribution in subsequent cycles more stable.

[0138] In this embodiment, to verify the positioning and distribution capability of the semi-autogenous grinding feed end detection deviation within the ore storage pile, the residual particle size distribution ratio of each voxel partition was recorded based on the complete scheme over a continuous period, and a heat map of the residual particle size distribution was displayed. Figure 6 In the diagram, the horizontal axis represents voxel partitions, the vertical axis represents periods, and the color intensity indicates the proportion of residuals allocated. The closer the color is to a high value region, the higher the proportion of residuals allocated to the corresponding voxel partition in the current period. Figure 6It is evident that the residual granularity state is not uniformly diffused across all voxel partitions, but is mainly concentrated in the relevant voxel partitions in the middle, forming a continuous high-value band as the cycle progresses; the maximum residual distribution ratio marked in the figure is also located in the corresponding concentrated area; the results indicate that after historical alignment through the propagation delay of each feed port and the release delay within the stack, the local error vector can be constrained within the range of voxel blocks that are more closely related to the actual release process, thereby making the distribution of residual granularity state more concentrated and directional.

[0139] In summary, this invention improves the accuracy of candidate discharge voxel set generation by: synchronously updating particle size state migration and discharge reachability relationships, ensuring that the particle size state of voxel blocks remains aligned with the feed port during the discharge release process; and by screening the inversion voxel sets corresponding to each feed port and performing local inverse inversion only within the inversion voxel sets, the deviation corresponding to the measured particle size distribution of semi-autogenous grinding feed can be traced back to the relevant voxel blocks along the propagation delay and the in-pile release delay, thus improving the reliability of semi-autogenous grinding feed particle size distribution prediction.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting particle size distribution in semi-autogenous grinding ore reserves based on digital twins, characterized in that, include: Based on the effective particle size state of the ore storage pile in the previous cycle, the reachability of the ore storage pile in the previous cycle, the current material level field, the current coarse crushing input quality, the current coarse crushing input particle size distribution, and the current material pile location, the input writing amount and initial output reachability of each voxel block are determined, and the causal voxel particle size state field of the ore storage pile in the current cycle is constructed. By combining the feed port opening, feed port flow rate, conveyor belt speed and material level field, the release from the stockpile, particle size state migration and discharge reachability relationship of each voxel block in the causal voxel particle size state field of the ore storage pile are synchronously updated to generate a candidate discharge voxel set corresponding to each feed port. Combining the measured particle size distribution of the semi-autogenous grinding feed, the propagation delay of each feed port, the release delay of each voxel block to the corresponding feed port in the pile, the candidate discharge voxel set, the voxel block release threshold, and the direct observation exclusion conditions for surface stability, the inversion voxel set corresponding to each feed port is screened, and local inverse inversion is performed only within the inversion voxel set to obtain the residual particle size state of each voxel block; The residual particle size state is written back to the causal voxel particle size state field of the ore storage pile in the current period to form the effective particle size state of the current period. The effective particle size state of the current period and the discharge reachability relationship of the current period are used as the input of the causal voxel particle size state field of the ore storage pile in the next period, and the prediction results of the semi-autogenous grinding feed particle size distribution are output in a rolling manner.

2. The method for predicting particle size distribution of semi-autogenous grinding process ore stockpiles based on digital twins as described in claim 1, characterized in that, The construction of the current periodic ore reservoir causal voxel-level state field includes: Establish a unified spatial coordinate system for the ore storage pile, and map the outer boundary of the ore storage pile, the current pile surface boundary, the coarse crushing pile boundary, the positions of each feed port, the positions of each transfer point, and the positions of the material level detection points to the same coordinate system. Discretize the effective volume of the ore storage pile to form several individual blocks. For each voxel block, record the voxel block spatial boundary, the voxel block geometric center position, and the adjacency relationship with adjacent voxel blocks to form a voxel topology table of the ore storage pile. The current pile surface of the ore storage heap is reconstructed based on the current periodic material level field. Based on the positional relationship between the geometric center of the voxel block and the current pile surface and the outer boundary of the ore storage heap, the voxel blocks are judged to distinguish between the current valid voxel blocks and the current invalid voxel blocks.

3. The method for predicting particle size distribution of semi-autogenous grinding process ore reserves based on digital twins as described in claim 2, characterized in that, The determination of the amount of data written into the heap for each voxel block includes: Centered on the current stockpile location, identify the set of surface voxel blocks directly covered by the current coarse crushing batch in the current effective voxel blocks; Based on the horizontal distance, vertical height difference and local slope of the current stacking point from the geometric center of each surface voxel block, a spatial bearing weight is constructed for each surface voxel block and the spatial bearing weight of all surface voxel blocks participating in the bearing is normalized to obtain the stacking writing ratio of the voxel block. Based on the total mass of the current coarse crushing batch, the particle size mass fraction vector corresponding to the current coarse crushing particle size distribution, and the write ratio of the voxel block, calculate the particle size mass vector written to the corresponding voxel block in the current cycle, and obtain the write amount of each voxel block.

4. The method for predicting particle size distribution of semi-autogenous grinding process ore stockpiles based on digital twins as described in claim 3, characterized in that, The initial output attainability relationship includes: The geometric connectivity factor, historical release path factor, and working condition constraint factor of the voxel block facing the feed port are constructed and weighted and fused. The weighted fusion result of all feed ports is normalized to obtain the initial discharge reachability relationship of the voxel block facing each feed port.

5. The method for predicting particle size distribution of semi-autogenous grinding process ore stockpiles based on digital twins as described in claim 1 or 4, characterized in that, The combination of feed port opening, feed port flow rate, conveyor belt speed, and material level field includes: The flow rate at each feed port is effectively processed based on the current conveyor belt speed. When the speed of the conveyor belt on the corresponding conveying path of the feed port is not less than the effective running speed threshold of the belt, the effective feed flow rate of the feed port is taken as the feed flow rate of the current cycle. When the speed of the conveyor belt on the corresponding conveying path of the feed port is less than the effective running speed threshold of the belt, the effective feed flow rate of the feed port is set to zero. Based on the effective feed flow rate of the feed inlet and the time interval of the current cycle, calculate the required mass of each feed inlet for the current cycle.

6. The method for predicting particle size distribution of semi-autogenous grinding process ore stockpiles based on digital twins as described in claim 1, characterized in that, The unloading process includes: Based on the discharge accessibility of the pre-updated voxel block facing the feed port and the total ore mass in the pre-updated voxel block, the release contribution weight of the voxel block to the feed port is calculated and normalized to obtain the release allocation ratio of the voxel block to the feed port. Based on the required mass of the feed outlet and the release distribution ratio of the voxel block to the feed outlet, calculate the pre-release particle size mass vector of the voxel block facing the feed outlet. By applying a release upper limit constraint to the pre-release particle size mass vector of the voxel block, the actual release particle size mass vector of the voxel block facing the feed port is obtained.

7. The method for predicting particle size distribution of semi-autogenous grinding process ore reserves based on digital twins as described in claim 6, characterized in that, The synchronous update of particle size state migration and discharge reachability includes: Based on the current cycle feed opening, feed flow rate, conveyor belt speed, and material level, call up the particle size migration coefficient that matches the current working conditions; Update the particle size state of the voxel block based on the remaining ore and particle size migration coefficient after it is released from the pile. Based on the remaining ore mass of the source voxel that migrates to the target voxel, the output accessibility relationship of the source voxel is mass-weighted and fused to obtain the updated output accessibility relationship of the target voxel.

8. The method for predicting particle size distribution of semi-autogenous grinding process ore reserves based on digital twins as described in claim 7, characterized in that, The generation of candidate discharge voxel sets corresponding to each feed port includes: The first condition for candidate discharge is that the total mass of ore after the current cycle update of the voxel block is greater than zero. The second condition for candidate discharge is that the discharge accessibility of the voxel block facing the feed port is not less than the candidate accessibility threshold. The third condition for candidate discharge is that the actual release mass of the voxel block facing the feed port in the current cycle is not less than the voxel block release amount threshold. Voxel blocks that simultaneously meet the first, second, and third conditions for candidate discharge will be included in the candidate discharge voxel set of the corresponding feed port.

9. The method for predicting particle size distribution of semi-autogenous grinding process ore reserves based on digital twins as described in claim 1, characterized in that, The combination of measured particle size distribution of semi-autogenous feed, propagation delay at each feed port, in-pile release delay of each voxel block to its corresponding feed port, candidate discharge voxel set, voxel block release threshold, and direct observation exclusion conditions for surface stability includes: Read the measured particle size distribution of the semi-autogenous grinding feed at the detection position in the current cycle, and convert the measured particle size distribution of the semi-autogenous grinding feed into a measured particle size mass vector of the semi-autogenous grinding feed. Based on the feed port propagation delay of each feed port, the actual release particle size mass vector of each voxel block facing the feed port in the corresponding historical period is aligned to the current period to obtain the predicted particle size mass vector of the semi-autogenous grinding feed. The difference between the measured particle size mass vector of the semi-autogenous grinding feed and the predicted particle size mass vector of the semi-autogenous grinding feed is used as the total error vector of the current cycle. The total error vector of the current cycle is then distributed to each feed port to obtain the local error vector corresponding to each feed port.

10. The method for predicting particle size distribution of semi-autogenous grinding process ore reserves based on digital twins as described in claim 9, characterized in that, The process of obtaining the residual granularity state of each voxel block includes: Based on the feed port propagation delay from the feed port to the semi-autogenous grinding feed detection position and the pile release delay from the voxel block to the corresponding feed port, the historical alignment time of the voxel block facing the feed port is determined. Voxels that belong to the candidate discharge voxel set at the historical alignment time, whose actual release mass towards the corresponding feed port at the historical alignment time is not less than the voxel block release amount threshold, and do not belong to the stable direct observation surface area, are included in the inversion voxel set of the corresponding feed port. Within the inverted voxel set only, the error attribution weight is determined based on the material accessibility and actual release quality of the voxel block at the historical alignment time. Based on the error attribution weight and the local error vector of the corresponding feed port, the residual particle size state of the voxel block is obtained.

Images