Efficient migration method of vertical large model for whole process of mineral processing

Abstract

This invention discloses an efficient migration method for a large-scale vertical model of the entire mineral processing process, relating to the field of artificial intelligence model migration technology. The method includes collecting raw ore property data, online instrument data, video image data, equipment status data, and laboratory sampling records from a gold mine's "lights-out" processing plant, and establishing a mineral flow identity chain according to the material transfer sequence. Based on the mineral flow identity chain, a segment inheritance migration sample is generated according to the process sequence, consisting of a current observation block, a previous inheritance block, a process action block, and a subsequent attribution block. Based on the segment inheritance migration sample, efficient segment inheritance migration is performed on the general basic large model according to the process sequence, and efficient migration of the current segment is completed based on the previous inheritance block, resulting in a large-scale vertical model of the entire gold mineral processing process. Late laboratory test results are backfilled along the mineral flow identity chain to the subsequent attribution block. The responsible segment is located based on the backfilled segment inheritance migration sample, and local re-migration is performed on the corresponding segment in the large-scale vertical model of the entire gold mineral processing process.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence model transfer technology, and in particular to an efficient transfer method for large vertical models of the entire mineral processing process. Background Technology

[0002] In the field of gold ore beneficiation and intelligent operation of beneficiation plants, existing technologies typically acquire production information by combining online instrument monitoring, industrial video monitoring, equipment status acquisition, laboratory sampling and analysis, and production scheduling records, focusing on processes such as crushing, grinding and classification, flotation, middlings regrinding, and thickening and dewatering. Based on this, they combine mechanistic analysis, empirical rules, statistical modeling, and general intelligent models to carry out condition identification, quality prediction, and operational support decision-making. With the advancement of the construction of "lights-out" beneficiation plants, the data acquisition capabilities, graphic and text fusion capabilities, and model application capabilities of the production site are continuously improving. The relevant methods can systematically reflect changes in process status, support the information management, intelligent analysis, and full-process operation optimization of the beneficiation process, and have good engineering application value in complex production environments.

[0003] Conventional methods in the entire gold ore beneficiation process still suffer from several drawbacks: multi-source records are often organized by location or time, the material transfer relationship across work sections is not adequately expressed, late test results are difficult to reliably correspond to previous working conditions, and general model migration is often processed uniformly as a whole sample, lacking work section inheritance and local update mechanisms, making it difficult to simultaneously take into account both the continuity of the entire process and migration efficiency. Summary of the Invention

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

[0005] Therefore, this invention provides an efficient migration method for large vertical models of the entire mineral processing process to solve the problems of insufficient ore flow correlation in the entire gold ore processing process, which makes it difficult to reliably trace back to previous working conditions for delayed test results, and the lack of segment inheritance and local update mechanisms in the migration of general models.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an efficient transfer method for a large-scale vertical model of the entire mineral processing process, comprising: Step S1: Collect raw ore property data, online instrument data, video image data, equipment status data, and laboratory sampling records from the gold mine's black-light processing plant, and establish an ore flow identity chain according to the material transfer sequence.

[0007] Step S2: Based on the mineral flow identity chain, generate a segment inheritance migration sample according to the process sequence, consisting of the current observation block, the previous segment inheritance block, the process action block, and the subsequent attribution block.

[0008] Step S3: Based on the segment inheritance migration sample, perform efficient segment inheritance migration on the general basic model according to the process sequence, and complete the efficient migration of the current segment according to the previous segment inheritance block to obtain the vertical class model of the entire gold mine beneficiation process.

[0009] Step S4: Backfill the late test results along the mineral flow identity chain to the subsequent ownership block, locate the responsible section based on the backfilled section inheritance migration sample, and perform local re-migration on the corresponding section in the large vertical model of the entire gold ore beneficiation process.

[0010] As a preferred embodiment of the efficient transfer method for the vertical category large model of the entire mineral processing process described in this invention, the specific steps for collecting raw ore property data, online instrument data, video image data, equipment status data, and laboratory sampling records from a gold mine's "dark lights" processing plant are as follows: At the raw ore feed location, crushing discharge location, grinding feed location, classification overflow location, roughing feed location, scavenging feed location, cleaning feed location, middlings regrinding reflux location, concentrate thickening feed location, and tailings discharge location, connect the raw ore property data acquisition interface, online instrument data acquisition interface, video image data acquisition interface, equipment status data acquisition interface, and laboratory sampling record interface to the same clock source, and scan the records at each acquisition location according to a unified time identifier and process sequence.

[0011] When the equipment status remains unchanged, the material flow direction does not change, and the feed rate or slurry flow rate remains continuous, adjacent records are merged to form a mineral flow segment. When the equipment stops, starts, pumps and valves are switched, loops are switched, flow is interrupted, feeding is interrupted, or the sampling location is changed, the current mineral flow segment is cut off and a new mineral flow segment is generated.

[0012] As a preferred embodiment of the efficient migration method for the entire mineral processing vertical category model described in this invention, the specific steps for establishing the mineral flow identity chain according to the material transfer sequence are as follows: For each downstream ore flow segment, upstream candidate ore flow segments that are temporally connected to the downstream ore flow segments are searched in the upstream collection location according to the material transfer sequence. The ore flow attribution degree is determined based on the actual transfer time, material quantity, solid concentration and stable operation history.

[0013] The upstream candidate mineral flow segment with the highest mineral flow attribution is selected as the main upstream source, and the upstream candidate mineral flow segments with non-zero mineral flow attribution are retained as auxiliary sources. The main upstream sources are sequentially connected to form a mineral flow identity chain according to the material transfer order. When the middlings are refluxed or the loop is switched, the main upstream source is used to maintain the continuity of the main chain of the mineral flow identity chain and the auxiliary sources are retained. A unique identity identifier is written for each mineral flow identity chain according to the formation order.

[0014] As a preferred embodiment of the efficient migration method for the entire mineral processing vertical model described in this invention, the step of generating a segment inheritance-based migration sample composed of the current segment observation block, the previous segment inheritance block, the process action block, and the subsequent effect attribution block according to the process sequence includes: The ore flow identity chain is read one by one according to the unique identifier, and divided into crushing section, grinding and classification section, flotation section and thickening and dewatering section according to the material transfer order. For the same ore flow identity chain, a section inheritance migration sample is generated for each section.

[0015] For each section of the inherited migration sample, a current section observation block, a previous section inheritance block, a process action block, and a subsequent attribution block are generated sequentially. The current section observation block contains the raw ore property data, online instrument data, video image data, and equipment status data records of the corresponding ore flow segment in the current section. The previous section inheritance block of the crushing section sample contains the entry condition record of the ore flow segment corresponding to the raw ore selection position in the same ore flow identity chain. The previous section inheritance block of the other section samples contains the entry condition record of the main upstream source ore flow segment corresponding to the previous section in the same ore flow identity chain. Records in the previous section inheritance block, except for image records and operation status records, are extracted as previous section inheritance value records in their original order. The previous section inheritance value records serve as the entry conditions for the current section. Image records and operation status records are retained in the previous section inheritance block. The process action block contains the records of actual changes within the current section. The subsequent attribution block contains the downstream laboratory sampling records on the same ore flow identity chain.

[0016] As a preferred embodiment of the efficient migration method for the entire mineral processing vertical model described in this invention, the step of generating a segment inheritance-based migration sample composed of the current segment observation block, the previous segment inheritance block, the process action block, and the subsequent effect attribution block according to the process sequence further includes: When multiple downstream testing and sampling records are connected on the same mineral flow identity chain, the path attribution value between the current section and the downstream testing and sampling record is determined according to the mineral flow attribution degree. The concentrate testing and sampling record with the largest path attribution value and the tailings testing and sampling record with the largest path attribution value are written into the subsequent attribution block respectively. The stage testing and sampling records are appended to the subsequent attribution block in the order of their appearance in the mineral flow identity chain. For affiliated sources, the connection relationship and path attribution value corresponding to the affiliated source are retained, but the test results corresponding to the affiliated source are not recorded as the real subsequent attribution value and written into the subsequent attribution block.

[0017] After all the observation blocks, previous inheritance blocks, process action blocks, and subsequent attribution blocks are generated, they are written into the same section inheritance migration sample in the order of unique identifier, current section name, observation block, previous inheritance block, process action block, and subsequent attribution block.

[0018] As a preferred embodiment of the efficient migration method for the entire mineral processing vertical class large model described in this invention, the specific steps of performing efficient migration of the general basic large model according to the process sequence are as follows: The Chinese instruction-following general basic model is selected as the general basic model. The original parameters of the general basic model are not updated. At the matrix parameters that represent the current input record, the corresponding section migration parameters for crushing, grinding and classification, flotation and thickening and dewatering are set respectively. The section migration parameters for each section include a front matrix and a back matrix. The product of the front matrix and the back matrix is ​​used as a low-rank incremental matrix with the same matrix shape as the corresponding original matrix parameters.

[0019] Each segment inheritance migration sample is constructed into a current segment migration input record and a current segment migration output record. The current segment migration input record is written in the order of unique identifier, current segment name, previous segment inheritance block, current segment observation block, and process action block. The current segment migration output record consists of the current segment entry condition value record and the subsequent effect assignment block. The current segment entry condition value record is generated according to the field names and writing order of the previous segment inheritance value record, and the previous segment inheritance value record in the previous segment inheritance block is used as the target record. The subsequent effect assignment block uses the actual subsequent effect assignment block in the current segment inheritance migration sample as the target record. The segment inheritance efficient migration is performed in the order of crushing segment, grinding and classification segment, flotation segment, and thickening and dewatering segment. Each time, only the segment migration parameters corresponding to the current segment are updated, and the current segment efficient migration is completed by using the previous segment inheritance block as the starting part of the current segment migration input record.

[0020] As a preferred embodiment of the efficient migration method for the entire mineral processing vertical category model described in this invention, the specific steps for completing the efficient migration of the current section based on the previous inherited block are as follows: During the migration process of each work segment, the differences between the consequential value records in the current work segment migration output records generated by the general basic model and the consequential value records of the same field in the actual consequential value block are compared simultaneously. The differences between the current work segment entry condition value records in the current work segment migration output records and the previous inheritance value records of the same field in the previous inheritance block are also compared. All work segment inheritance migration samples of the current work segment are repeatedly traversed, and the two types of differences obtained from two adjacent traversals are compared. When the difference obtained from two adjacent traversals is not less than the difference obtained from the previous traversal, it is determined that the corresponding difference will no longer decrease. When both types of differences no longer decrease, it is determined that the current work segment has completed efficient migration.

[0021] As a preferred embodiment of the efficient transfer method for the vertical category large model of the entire mineral processing process described in this invention, the specific steps for obtaining the vertical category large model of the entire gold ore processing process are as follows: After the current section completes the efficient migration, the migration parameters corresponding to the current section are not updated, and the same section inheritance efficient migration is immediately performed on the next section. If the current section has not yet completed the efficient migration, the migration of the next section will not be started in advance. For the crushing section samples, grinding and classifying section samples, flotation section samples, and thickening and dewatering section samples on the same ore flow identity chain, the migration order of crushing section, grinding and classifying section, flotation section, and thickening and dewatering section is always maintained. After all sections have been migrated, the general basic model and the section migration parameters corresponding to the crushing section, grinding and classifying section, flotation section, and thickening and dewatering section that have completed the section inheritance efficient migration are jointly solidified into a vertical large model of the entire gold ore beneficiation process.

[0022] As a preferred embodiment of the efficient migration method for the vertical category large model of the entire mineral processing process described in this invention, the specific steps of backfilling late test results to the subsequent attribution block along the mineral flow identity chain are as follows: Late test results are continuously read in the order of being written with the time stamp. For each late test result, the sampling location, sampling time and sample number corresponding to the current late test result are read. The mineral flow segment connected to the sampling location, sampling time and sample number is searched in the mineral flow identity chain according to the unique identity. The process is backtracked upstream segment by segment along the main chain already formed by the main upstream source. The inherited migration samples of the work segment hit during the backtracking process and their connection relationships are formed into candidate backfill paths. At the same time, the connection relationships of the auxiliary sources in the mineral flow identity chain are preserved.

[0023] When there is only one candidate backfill path, the current late test result is written into the subsequent ownership block of the inherited migration sample of the corresponding work section on the candidate backfill path in the order of sampling location, sampling time, sample number and test result. When there are multiple candidate backfill paths, the candidate backfill path with the largest path ownership value is selected as the backfill path, and the current late test result is written into the subsequent ownership block of the inherited migration sample of the corresponding work section on the backfill path as the real subsequent ownership value record. For the other candidate backfill paths that are not selected, only the connection relationship and path ownership value in the mineral flow identity chain are retained, and the current late test result is not written into the subsequent ownership block as the real subsequent ownership value record. If there is already a record with the same sampling location, the same sampling time and the same sample number in the subsequent ownership block, the current late test result replaces the original record.

[0024] As a preferred embodiment of the efficient migration method for the vertical category large model of the entire mineral processing process described in this invention, the specific steps of locating the responsible section based on the inherited migration sample of the backfilled section and performing local re-migration on the corresponding section in the vertical category large model of the entire gold ore processing process are as follows: For the inherited migration samples of the work section updated by the current late test result, the current work section migration input record and the current work section migration output record are reconstructed and input into the large vertical model of the entire gold ore beneficiation process. The differences between the aftereffect attribution value record in the current work section migration output record and the aftereffect attribution value record of the same field in the updated aftereffect attribution block are compared, as well as the differences between the current work section entry condition value record in the current work section migration output record and the previous section inheritance value record of the same field in the previous section inheritance block. For the crushing section, grinding and classification section, flotation section and thickening and dewatering section, the current work section responsibility value is calculated for each work section with an aftereffect attribution block updated by the current late test result. The work section with the largest current work section responsibility value is determined as the responsible work section.

[0025] Local re-migration is performed only on the migration parameters of the responsible section, and the inherited migration samples of the current responsible section and the inherited migration samples of the directly connected sections are used together as local re-migration samples. When the differences between the two types of results obtained in two adjacent traversals no longer continue to decrease, the local re-migration of the responsible section is considered to be completed, and the full-process vertical model of gold ore beneficiation after the local re-migration is used as the current full-process vertical model of gold ore beneficiation.

[0026] The beneficial effects of this invention are as follows: By collecting raw ore property data, online instrument data, video image data, equipment status data, and laboratory sampling records from a gold mine's "dark light" beneficiation plant, and establishing a mineral flow identity chain according to the material transfer sequence, the continuous expression of cross-section material transfer relationships and the stable correspondence between late test results and preceding conditions are achieved, improving the accuracy of data organization throughout the entire process. By backfilling late test results along the mineral flow identity chain to the subsequent attribution block, locating the responsible section based on the backfilled section inheritance migration sample, and performing local re-migration on the corresponding section in the large vertical model of the entire gold mine beneficiation process, targeted updates based on section inheritance relationships are achieved, balancing migration efficiency and model adaptability. Attached Figure Description

[0027] 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.

[0028] Figure 1 A flowchart for an efficient transfer method for a large-scale vertical model of the entire mineral processing process.

[0029] Figure 2 This is a schematic diagram illustrating the formation of the mineral flow identity chain.

[0030] Figure 3 This is a schematic diagram of the inheritance-based migration and efficient migration of work sections.

[0031] Figure 4 This is a schematic diagram illustrating the backfilling and local re-migration of late test results.

[0032] Figure 5 This is a curve showing the hit rate of backfilling late test results in the mineral flow identity chain.

[0033] Figure 6 The curve showing the convergence of the difference between the location of the responsible section and the local relocation. Detailed Implementation

[0034] 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.

[0035] 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.

[0036] 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.

[0037] Reference Figures 1-6 This is one embodiment of the present invention, which provides an efficient transfer method for a large vertical model of the entire mineral processing process, including the following steps: S1. Collect raw ore property data, online instrument data, video image data, equipment status data, and laboratory sampling records from the gold mine's dark processing plant, and establish an ore flow identity chain according to the material transfer sequence.

[0038] In the gold mine's "dark lights" beneficiation plant, data collection points are set up according to the material transfer sequence at the following locations: raw ore inlet, crushing discharge, grinding feed, classification overflow, roughing feed, scavenging feed, cleaning feed, middlings regrinding return, concentrate thickening feed, and tailings discharge. At each collection point, the raw ore property data acquisition interface, online instrument data acquisition interface, video image data acquisition interface, equipment status data acquisition interface, and laboratory sampling record interface are all connected to the same clock source. A unified time stamp is written to the output records of each acquisition interface. When different data corresponding to the same moment have acquisition delays, the delay records are corrected using the unified time reference corresponding to the written time stamp.

[0039] The raw ore property data includes the gold grade, associated sulfur content, particle size distribution, and grindability characterization of the raw ore. Online instrument data includes feed rate, slurry flow rate, slurry concentration, mill current, hydrocyclone feed pressure, classification overflow fineness, flotation pH, aeration rate, reagent addition, and thickened mud layer interface height. Video image data includes crushing discharge images, flotation froth images, and images of key conveying locations. Equipment status data includes belt operation status, mill operation status, pump and valve switching status, and flotation machine operation status. Laboratory sampling records include the sampling location, sampling time, and sample number. All collection records are marked with a sampling location identifier and a uniform time identifier, and are stored continuously in chronological order.

[0040] Starting with the feeding records generated at the raw ore inlet during continuous production, the records at each collection point are scanned in the same time frame according to the process sequence: raw ore inlet, crushing discharge, grinding feed, classifying overflow, roughing feed, scavenging feed, cleaning feed, middlings regrinding reflux, concentrate thickening feed, and tailings discharge. At the same collection point, two consecutive records directly connected after being sorted by a unified time identifier are considered adjacent records. When the equipment status and material flow remain unchanged between adjacent records... When the feed rate or slurry flow rate at the corresponding location remains continuous without switching, adjacent records are merged into a single ore flow segment. When shutdown, startup, pump / valve switching, loop switching, flow interruption, feed interruption, or sampling location change occurs, the current ore flow segment is truncated at the corresponding time, and a new ore flow segment is generated from the corresponding time. Each ore flow segment includes the segment start time, segment end time, corresponding sampling location, corresponding material quantity, slurry concentration, raw ore property data corresponding to the sampling location, online instrument data, video image data, and equipment status data records.

[0041] The difference between the start time of the downstream ore flow segment and the start time of the upstream candidate ore flow segment is taken as the actual transmission time. When the start time of the downstream ore flow segment is later than the start time of the upstream candidate ore flow segment, and the actual transmission time is between the minimum and maximum actual transmission time in the stable operation history of the current material transmission path, the upstream candidate ore flow segment and the downstream ore flow segment are considered to be sequentially connected. The feed rate is taken as the solid material mass flow rate, the slurry flow rate is taken as the slurry mass flow rate, and the slurry concentration is taken as the solid mass fraction. For solid material conveying locations, the accumulated solid material mass obtained from the start time to the end time of the segment is taken as the corresponding material quantity. For slurry conveying locations, the accumulated solid material mass obtained from the product of the slurry flow rate and the slurry concentration under the same unified time identifier from the start time to the end time of the segment is taken as the corresponding material quantity.

[0042] After the ore flow segment is generated, for each downstream ore flow segment at a downstream collection location, according to the material transfer sequence, upstream candidate ore flow segments that are temporally connected to the downstream ore flow segments are searched in the upstream collection locations, and the attribution value between the upstream candidate ore flow segments and the downstream ore flow segments is calculated. The expression is as follows: ; ; in, This represents the original assignment value from the upstream candidate ore flow segment to the downstream ore flow segment. This represents the normalized ore flow attribution degree. This indicates the actual transmission time from the upstream candidate ore flow segment to the downstream ore flow segment. This represents the median delivery time of the current material handling path in the history of stable operation. This represents the discrete scale of the current material transfer path's transfer time. This indicates the amount of material in the downstream ore flow segment. This indicates the average material transfer rate of the current material transfer path in the history of stable operation. This indicates the amount of material in the upstream candidate ore flow segment. This represents the discrete scale of the material quantity in the current material delivery path. Indicates the solids concentration of the downstream ore flow segment. This indicates the solids concentration of the upstream candidate ore flow segment. This represents the median concentration difference of the current material transport path within the historical stable operating records. The concentration difference is the difference between the solids concentration of the downstream ore flow segment in history and the solids concentration of the corresponding upstream ore flow segment in history. This represents the median absolute deviation of each historical concentration difference from the median concentration difference. This represents the set of all upstream candidate ore flow segments that are temporally connected to downstream ore flow segments. Indicates the upstream candidate ore flow segment identifier, Indicates the downstream ore flow segment identifier. This indicates the material transfer path identifier connecting upstream candidate ore flow segments and downstream ore flow segments. This represents the identifier of any upstream candidate mineral flow segment in the upstream candidate mineral flow segment set.

[0043] Stable operation history refers to historical production records where there are no downtimes, no loop switching, and continuous material flow along the same path.

[0044] The current material transfer path is paired with all historical upstream candidate ore flow segments and historical downstream ore flow segments in the stable operation history to form a historical pairing sample set. For each historical pairing sample, the actual transfer time, material quantity of the upstream candidate ore flow segment, material quantity of the downstream ore flow segment, solid concentration of the upstream candidate ore flow segment, and solid concentration of the downstream ore flow segment are read.

[0045] A transmission time sequence is constructed from all actual transmission times in the historical pairing sample set. The median transmission time of the transmission time sequence is obtained. Then, the absolute deviation of each historical actual transmission time from the median transmission time is calculated. The median of all absolute deviations is used as the discrete scale of the transmission time of the current material transmission path.

[0046] A historical correspondence is established between the material quantity of downstream ore flow segments and the material quantity of upstream candidate ore flow segments in the historical paired sample set. The average material transfer ratio of the current material transfer path is obtained based on all historical correspondences. For each historical paired sample, the absolute value of the difference between the material quantity of the downstream ore flow segment and the product of the average material transfer ratio and the material quantity of the upstream candidate ore flow segment is calculated. The median of all absolute values ​​is used as the discrete scale of the material quantity of the current material transfer path.

[0047] For each historical paired sample set, the solid concentration of the downstream ore flow segment is matched with the solid concentration of the upstream candidate ore flow segment. The concentration difference of each historical paired sample is calculated, the median of the concentration difference of all concentration differences is obtained, and then the absolute deviation of each concentration difference from the median of the concentration difference is calculated. The median of all absolute deviations is used as the concentration dispersion scale of the current material transfer path.

[0048] When the transmission duration discrete scale is zero, the minimum time interval between adjacent moments that can be distinguished by a unified time identifier is used as the transmission duration discrete scale. When the material quantity discrete scale is zero, the minimum effective resolution of the feed rate or the minimum distinguishable change of the slurry flow rate and slurry concentration used in the corresponding material quantity calculation is converted into the minimum solid material mass according to the duration of the ore flow segment and the corresponding material quantity calculation method, and the minimum solid material mass is used as the material quantity discrete scale. When the concentration discrete scale is zero, the minimum difference between adjacent output values ​​that can be distinguished by the slurry concentration acquisition interface is used as the concentration discrete scale.

[0049] For each downstream ore stream segment, the upstream candidate ore stream segment with the highest ore stream affiliation is selected as the main upstream source, and the remaining upstream ore stream segments with non-zero ore stream affiliation are retained as auxiliary sources. According to the material transfer sequence, the main upstream sources in the following locations are sequentially connected: raw ore feeding location, crushing discharge location, grinding feed location, classification overflow location, roughing feed location, scavenging feed location, cleaning feed location, middlings regrinding reflux location, concentrate thickening feed location, and tailings discharge location, forming an ore stream identity chain that runs through the entire beneficiation process. When middlings regrinding reflux or loop switching causes any ore stream segment to have both a main upstream source and an auxiliary source at the same time, only the main upstream source is used to maintain the continuity of the main chain of the ore stream identity chain, and the auxiliary sources are retained in the same ore stream identity chain.

[0050] Each mineral flow identity chain is written with a unique identifier in the order of its formation, and the unique identifier is simultaneously written into the corresponding mineral flow segment, associated raw ore property data, online instrument data, video image data, equipment status data, and laboratory sampling records.

[0051] S2. Based on the mineral flow identity chain, generate a segment inheritance migration sample according to the process sequence, consisting of the current observation block, the previous inheritance block, the process action block, and the subsequent attribution block.

[0052] The ore flow identity chain is read one by one according to the unique identifier, and divided into crushing section, grinding and classification section, flotation section and thickening and dewatering section according to the material transfer order. The crushing section corresponds to the raw ore feeding position and crushing discharge position. The grinding and classification section corresponds to the grinding feed position and classification overflow position. The flotation section corresponds to the roughing feed position, scavenging feed position, cleaning feed position and middlings regrinding return position. The thickening and dewatering section corresponds to the concentrate thickening feed position and tailings discharge position. For the same ore flow identity chain, a section inheritance migration sample is generated for each of the crushing section, grinding and classification section, flotation section and thickening and dewatering section. Each section inheritance migration sample uniquely corresponds to a section on the same ore flow identity chain.

[0053] For each work section, the inherited migration sample is sequentially generated into the current section observation block, the previous section inheritance block, the process action block, and the subsequent effect attribution block.

[0054] The observation blocks for the crushing section samples record the gold grade, associated sulfur content, particle size distribution, grindability characterization, feed rate, crushing discharge image, and belt operation status of the raw ore in the corresponding ore flow segments at the raw ore inlet and crushing discharge locations. The observation blocks for the grinding and classification section samples record the slurry flow rate, slurry concentration, mill current, hydrocyclone feed pressure, classification overflow fineness, and mill operation status of the ore flow segments at the grinding feed and classification overflow locations. The observation blocks for the flotation section samples record the flotation froth image, flotation pH, aeration rate, reagent dosage, pump and valve switching status, and flotation machine operation status of the ore flow segments at the rougher feed, scavenger feed, cleaner feed, and middlings regrinding reflux locations. The observation blocks for the thickening and dewatering section samples record the slurry concentration, thickened mud layer interface height, and pump and valve switching status of the ore flow segments at the concentrate thickening feed and tailings discharge locations.

[0055] The preceding segment of the crushing section sample is generated from the ore stream segment corresponding to the raw ore feed location in the same ore stream identity chain. The preceding segment of the grinding and classification section sample is generated from the upstream source ore stream segment corresponding to the crushing discharge location in the same ore stream identity chain. The preceding segment of the flotation section sample is generated from the upstream source ore stream segment corresponding to the classification overflow location in the same ore stream identity chain. The preceding segment of the thickening and dewatering section sample is generated from the ore stream segments corresponding to the concentrate feed location and the middlings regrinding reflux location, which continue along the upstream source to the tailings discharge location in the same ore stream identity chain. The preceding segment of the crushing section sample reads the input condition records according to the record order in the ore stream segment corresponding to the raw ore feed location. Except for the crushing section sample, the preceding segment of the other section samples is generated according to the current process. At the start of each segment, the records from the previous segment are read backwards, maintaining the original order of the records. Records in the inherited block from the previous segment, excluding image records and operational status records, are extracted in their original order as inherited numerical records. These inherited numerical records are used as the entry conditions for the current segment. Image records and operational status records are retained in the inherited block. For the crushing segment sample, the inherited block contains records of the gold grade, associated sulfur content, particle size distribution, and grindability of the raw ore. For the grinding and classification segment sample, the inherited block contains records of feed rate, crushing discharge images, and belt conveyor operation status. For the flotation segment sample, the inherited block contains records of slurry flow rate, slurry concentration, mill current, hydrocyclone feed pressure, and classification overflow fineness. For the thickening and dewatering segment sample, the inherited block contains flotation froth images, flotation pH, aeration rate, reagent dosage, and flotation machine operation status.

[0056] The process action block is extracted from the ore flow segment corresponding to the observation block of the current section, and only records that have actually changed within the current section are retained. In the unified time identifier sorting of the current section, the subsequent continuous record is compared with the previous continuous record item by item. When the feed rate, slurry flow rate, slurry concentration, mill current, hydrocyclone feed pressure, flotation pH, aeration rate, reagent addition, thickened mud layer interface height, belt running status, mill running status, pump valve switching status, or flotation machine running status changes relative to the previous continuous record, the time of change, the record before the change, and the record after the change are extracted and written into the process action block in chronological order. If there is no change between the continuous records of the current section, the record of the current section maintaining continuous production status is written into the process action block.

[0057] The subsequent attribution block is generated based on the downstream laboratory sampling records on the same mineral flow identity chain. Starting from the current section's position in the mineral flow identity chain, the laboratory sampling records corresponding to the concentrate thickening feed position and tailings discharge position are sequentially retrieved downstream along the main chain already formed by the upstream source. When there is no corresponding laboratory result on the same mineral flow identity chain, the already generated observation block, previous inherited block, and process action block are stored continuously according to the unique identity identifier, and the subsequent attribution block is generated after the corresponding laboratory result is written into the laboratory sampling record.

[0058] When multiple interconnected downstream laboratory sampling records exist on the same mineral flow identity chain, the path attribution value is calculated for the entire mineral flow identity chain path from the current section to the downstream laboratory sampling record based on the mineral flow attribution degree. The expression is as follows: ; in, This indicates the path assignment value between the current ore flow segment and the downstream laboratory sampling record. This represents the mineral flow identity chain path connecting the corresponding mineral flow segment in the current work section with the downstream laboratory sampling records. This indicates the connection relationship between two adjacent segments of mineral flow along the path. This indicates the identifier of the starting ore flow segment corresponding to the current work section. This indicates the identifier for downstream laboratory sampling records.

[0059] The sampling location, sampling time, sample number, and test results corresponding to the concentrate and tailings sampling records with the highest path attribution values ​​are written into the subsequent attribution block in sequence. If there are other stage sampling records on the same mineral flow identity chain that are located after the current section and before the concentrate thickening feed position and tailings discharge position, they are added to the subsequent attribution block in the order of their appearance in the mineral flow identity chain. For secondary sources, the path attribution value is calculated by using the connection relationship retained in the mineral flow identity chain, and the connection relationship and path attribution value corresponding to the secondary source are retained. The test results corresponding to the secondary source are not recorded as the actual subsequent attribution value and written into the subsequent attribution block.

[0060] After all the observation blocks, previous inheritance blocks, process action blocks, and subsequent attribution blocks are generated, they are written into the same segment inheritance migration sample in the order of unique identity identifier, current segment name, observation block, previous inheritance block, process action block, and subsequent attribution block. The records within the same segment inheritance migration sample maintain the chronological order under the same time identifier and do not change the chronological relationship of the already formed records. For the same ore flow identity chain, crushing segment samples, grinding and classification segment samples, flotation segment samples, and thickening and dewatering segment samples are generated in the process order. The crushing segment samples, grinding and classification segment samples, flotation segment samples, and thickening and dewatering segment samples together constitute the segment inheritance migration sample set corresponding to the ore flow identity chain.

[0061] S3. Based on the segment inheritance migration sample, perform efficient segment inheritance migration on the general basic model according to the process sequence, and complete the efficient migration of the current segment according to the previous segment inheritance block to obtain the vertical class model of the entire gold mine beneficiation process.

[0062] The crushing section sample, grinding and classification section sample, flotation section sample, and thickening and dewatering section sample corresponding to the ore flow identity chain are read one by one according to the unique identifier. The Chinese instruction following general basic model (e.g., Qwen2.5-VL-7B-Instruct) that can directly receive the image records in the observation block of this section and the inherited block of the previous section and the other records is selected as the general basic model. The general basic model that only receives the other records is not used because the crushing section sample and the flotation section sample contain crushing discharge image and flotation foam image, respectively. After the general basic model is determined, the original parameters of the general basic model are not updated.

[0063] The section migration parameters originate from the matrix parameter positions within the general basic large model that transform the representation of the current input record. For each layer of representation transformation matrices jointly represented by the received image record and other records in the general basic large model, section migration parameters are set for the crushing section, the grinding and classification section, the flotation section, and the thickening and dewatering section. The section migration parameters for each section are composed of a front matrix and a back matrix. The number of columns in the front matrix is ​​consistent with the number of rows in the back matrix (e.g., 16), and it serves as the rank value of the low-rank increment matrix. The number of rows in the front matrix is ​​consistent with the number of rows in the corresponding original matrix parameter, and the number of columns in the back matrix is ​​consistent with the number of columns in the corresponding original matrix parameter, so that the section migration parameters can be inserted into the corresponding matrix parameter positions and participate in the representation transformation of the current input record.

[0064] The matrix parameter position refers to the query projection matrix, key projection matrix, value projection matrix, and output projection matrix in each layer of self-attention after the image record and other records enter the same representation space, as well as the input projection matrix and output projection matrix in the feedforward network. The word embedding parameter, image encoding parameter, layer normalization parameter, and output header parameter do not have the segment migration parameter set.

[0065] The product of the first and second matrices is used as a low-rank increment matrix. The shape of the low-rank increment matrix is ​​consistent with the shape of the corresponding original matrix parameters. It is mapped to the corresponding original matrix parameters by adding them position by position, so that the matrix parameters after adding the section migration parameters are still in the original matrix parameter positions for calculation. The first matrix is ​​initialized as a zero matrix and the second matrix is ​​initialized as a small-amplitude random matrix. This ensures that the low-rank increment matrix does not change the original output of the general basic model before the efficient section inheritance migration begins. When the efficient migration of the current section is performed, only the first and second matrices corresponding to the current section are updated. The original parameters of the general basic model, the first and second matrices corresponding to the completed sections, and the first and second matrices corresponding to the unreached sections are all kept unchanged.

[0066] When performing efficient migration and reasoning for the current work section, the low-rank increment matrix corresponding to the completed work section and the low-rank increment matrix corresponding to the current work section are superimposed on the corresponding original matrix parameters according to the process order. The front and back matrices corresponding to the completed work section only participate in the forward calculation and are not updated, while the front and back matrices corresponding to the unreached work section do not participate in the forward calculation.

[0067] When constructing a general basic model for Chinese instruction following using a self-training method, general pre-training, Chinese instruction alignment training, preference and constraint alignment training, and scene adaptation training are performed. When a general basic model for Chinese instruction following that has already been trained is selected, the above training is not repeated, and the original parameters of the selected general basic model are kept unchanged.

[0068] The general pre-training data for the Chinese instruction-following general basic model includes general Chinese corpus and industrial scenario-related text and image data. The general Chinese corpus is selected from Chinese web pages, news reports, encyclopedia entries, standards and specifications, patent texts, technical paper abstracts, equipment manuals, process operation procedures, and safety management documents. The industrial scenario-related text and image data is selected from equipment interface screenshots, process flow diagrams, instrument trend charts, crushing and discharge images, flotation foam images, inspection photos, and corresponding Chinese explanatory texts. All training data undergoes deduplication, removal of garbled characters, removal of missing pages, removal of low-resolution images, removal of samples with unclear image-text correspondence, removal of samples with self-contradictory content, and removal of samples that induce unsafe outputs. Only samples with complete Chinese expression, clear image-text correspondence, consistent professional terminology, and continuous training capability are retained. The label organization format uniformly adopts instruction fields and input fields. The training samples consist of an output field, a modal field, and a constraint field. The instruction field is used to write Chinese instructions. The input field is used to write image records and other records. The output field is used to write the target output content. The modal field identifies the order of image records and other records. The constraint field is used to write security constraints and format constraints. For preference and constraint alignment training samples, in addition to the instruction, input, output, modal, and constraint fields, a preferred result field and a secondary result field are added. The general pre-training stage uses an autoregressive prediction loss function. The Chinese instruction alignment training stage uses a supervised fine-tuned cross-entropy loss function. The preference and constraint alignment training stage uses a pairwise preference loss function. The scene adaptation training stage uses a supervised fine-tuned cross-entropy loss function. The general pre-training stage uses the AdamW optimizer, setting the learning rate corresponding to the general Chinese corpus (e.g., ...). ), set the learning rate for relevant graphic and textual materials in the industrial scenario (e.g. During the Chinese instruction alignment training phase, the AdamW optimizer is used, and the learning rate is set (e.g., ...). The AdamW optimizer is used during the preference and constraint alignment training phase, and the learning rate is set (e.g., ...). During the scene adaptation training phase, the AdamW optimizer is used, and the learning rate is set (e.g., ...). Each training phase employs a learning rate adjustment method based on cosine decay after warm-up.

[0069] For each section of the inherited migration sample, a unified current section migration input record and a current section migration output record are constructed. The current section migration input record is written in the order of unique identifier, current section name, previous inherited block, current section observation block, and process action block. The current section migration output record consists of the current section entry condition value record and the subsequent attribution block. The current section entry condition value record is generated according to the field names and writing order of the previous inherited value record, and the previous inherited value record in the previous inherited block is used as the target record. The subsequent attribution block uses the actual subsequent attribution block in the current section's inherited migration sample as the target record. The records within the same section's inherited migration sample continue to maintain the established unified time identifier order, without disrupting the record order within the previous inherited block, current section observation block, and process action block. The previous inherited block is used as the starting part of the current section migration input record, and the current section observation block and process action block are sequentially connected.

[0070] When generating the current section entry condition numerical record, the current section migration input record is copied. The numerical content of the previous inherited numerical record in the copied current section migration input record is replaced with a unified missing identifier. The field names, writing order, and image records and running status records in the previous inherited block are retained. The replaced current section migration input record is used as the input for generating the current section entry condition numerical record, and the previous inherited numerical record in the previous inherited block is used as the target record. The unreplaced current section migration input record is used as the input for generating the subsequent belonging block, and the actual subsequent belonging block is used as the target record. The generated current section entry condition numerical record and subsequent belonging block together constitute the current section migration output record.

[0071] Only the segment inheritance migration samples that have been generated in the subsequent ownership block and contain test results are used for efficient segment inheritance migration. Segment inheritance migration samples that have not yet been generated in the subsequent ownership block or do not yet contain test results do not participate in the current efficient segment migration, and local re-migration is performed after the corresponding late test results are written.

[0072] For efficient migration of the current work section, the supervised fine-tuning cross-entropy loss is calculated separately for the generation process of the numerical record of the entry condition of the current work section and the generation process of the subsequent belonging block. The sum of the two supervised fine-tuning cross-entropy losses is used as the migration loss of the current work section. The AdamW optimizer is then used to update the front and back matrices corresponding to the current work section in reverse according to the migration loss of the current work section. The learning rate is set (e.g., The original parameters of the general basic model and the front and back matrices corresponding to non-current work sections are not updated.

[0073] After the current section migration input record and current section migration output record are formed, the section inheritance-based efficient migration is executed according to the process sequence. The sequence of the section inheritance-based efficient migration is: crushing section, grinding and classification section, flotation section, and thickening and dewatering section. For the crushing section, the current section migration input record and current section migration output record of all crushing section samples are read. The original parameters of the general basic model are not updated; only the section migration parameters corresponding to the crushing section are updated. This allows the general basic model to generate the corresponding current section entry condition numerical record and output record based on the previous section inheritance block, the current section observation block, and the process action block in the crushing section sample. After the crushing section is completed, the migration parameters of the crushing section are kept unchanged. The general basic model and the migration parameters of the crushing section that have completed efficient migration are used together as the starting model of the grinding and classification section. For the grinding and classification section, flotation section and thickening and dewatering section, the same operation is performed to perform efficient migration of the section inheritance. That is, the original parameters of the general basic model and the migration parameters of the completed section are kept unchanged. Only the migration parameters of the current section are updated. The previous inheritance block is used as the starting part of the current section migration input record to complete the efficient migration of the current section.

[0074] During the migration process of each work segment, the differences between the consequential value records in the current work segment migration output records generated by the general basic model and the consequential value records of the same field in the actual consequential value block are compared simultaneously. The differences between the current work segment entry condition value records in the current work segment migration output records and the previous inheritance value records of the same field in the previous inheritance block are also compared. All work segment inheritance migration samples of the current work segment are repeatedly traversed, and the two types of differences obtained from two adjacent traversals are compared. When the difference obtained from two adjacent traversals is not less than the difference obtained from the previous traversal, it is determined that the corresponding difference will no longer decrease. When both types of differences no longer decrease, it is determined that the current work segment has completed efficient migration.

[0075] The difference between the aftereffect attribution numerical records refers to the following: according to the field names and writing order of the test results in the aftereffect attribution block, each aftereffect attribution numerical record generated by the general basic model is matched one by one with the aftereffect attribution numerical records of the same field in the real aftereffect attribution block. For each field, the absolute value of the difference between the aftereffect attribution numerical records of the same field is calculated as the ratio of the larger value between the absolute value of the real aftereffect attribution numerical record and the minimum effective resolution of the corresponding test data. The arithmetic mean of the ratios obtained for all fields is then used as the aftereffect attribution difference value.

[0076] The difference between the current section entry condition numerical records refers to the following: according to the field names and writing order of the previous section inherited numerical records, each current section entry condition numerical record generated by the general basic model is matched one by one with the previous section inherited numerical records of the same field in the previous section inherited block. For each field, the absolute value of the difference between the previous section inherited numerical records of the same field is calculated as the ratio of the larger value between the absolute value of the previous section inherited numerical record and the minimum effective resolution of the corresponding collected data. The arithmetic mean of the ratios obtained for all fields is then used as the entry condition difference value.

[0077] Once the current segment is completed, the segment migration parameters corresponding to the current segment are not updated, and the same segment inheritance-style efficient migration is immediately performed on the next segment. If the current segment is not yet completed, the migration of the next segment is not started in advance. For crushing segment samples, grinding and classifying segment samples, flotation segment samples, and thickening and dewatering segment samples on the same ore flow identity chain, the migration order of crushing segment, grinding and classifying segment, flotation segment, and thickening and dewatering segment is always maintained. After all segments are migrated, the general basic model and the segment migration parameters corresponding to the crushing segment, grinding and classifying segment, flotation segment, and thickening and dewatering segment that have completed the segment inheritance-style efficient migration are jointly solidified into a vertical category model for the entire gold ore beneficiation process.

[0078] S4. Backfill the late test results along the mineral flow identity chain to the subsequent ownership block, locate the responsible section based on the backfilled section inheritance migration sample, and perform local re-migration on the corresponding section in the gold ore beneficiation vertical category model.

[0079] Late test results are continuously read in the order of being written with the time stamp. For each late test result, the sampling location, sampling time and sample number corresponding to the current late test result are read. The mineral flow segment connected to the sampling location, sampling time and sample number is searched in the mineral flow identity chain according to the unique identity. The process is backtracked upstream segment by segment along the main chain already formed by the main upstream source. The inherited migration samples of the work segment hit during the backtracking process and their connection relationships are formed into candidate backfill paths. At the same time, the connection relationships of the auxiliary sources in the mineral flow identity chain are preserved.

[0080] When there is only one candidate backfill path, the current late test result is written into the subsequent ownership block of the inherited migration sample of the corresponding work section on the candidate backfill path in the order of sampling location, sampling time, sample number and test result. When there are multiple candidate backfill paths, the candidate backfill path with the largest path ownership value is selected as the backfill path, and the current late test result is written into the subsequent ownership block of the inherited migration sample of the corresponding work section on the backfill path as the real subsequent ownership value record. For the other candidate backfill paths that are not selected, only the connection relationship and path ownership value in the mineral flow identity chain are retained, and the current late test result is not written into the subsequent ownership block as the real subsequent ownership value record. If there is already a record with the same sampling location, the same sampling time and the same sample number in the subsequent ownership block, the current late test result replaces the original record.

[0081] After late test results have been backfilled into the subsequent attribution block, all work segment inheritance migration samples updated by the current late test results are reconstructed into current work segment migration input records. Following the same numerical replacement method, the current work segment migration input records are copied, and the numerical content of the previous inherited numerical records in the copied current work segment migration input records is replaced with a unified missing identifier. The replaced current work segment migration input records are input into the gold mine beneficiation process vertical model to generate current work segment entry condition numerical records. The unreplaced current work segment migration input records are input into the gold mine beneficiation process vertical model to generate subsequent attribution blocks. The two generated results together constitute the current work segment migration output record. The differences between the subsequent attribution numerical records in the current work segment migration output record and the subsequent attribution numerical records of the same field in the updated subsequent attribution block are compared for each work segment, as well as the differences between the current work segment entry condition numerical records in the current work segment migration output record and the previous inherited numerical records of the same field in the previous inherited block. For work segments with subsequent attribution blocks updated by the current late test results, the current work segment responsibility value is calculated, expressed as: ; ; ; in, Indicates the first The subsequent attribution difference value of each current work section inherited migration sample. Indicates the first The difference in entry conditions for each current section of the inherited migration sample. This represents the generation of the large-scale vertical model for the entire gold ore beneficiation process. Each subsequent attribution value This indicates the subsequent attribution value of the same field in the updated subsequent attribution block. This represents the generation of the large-scale vertical model for the entire gold ore beneficiation process. The current entry condition value for the work section. This indicates the inherited value of the same field in the previous inheritance block. Indicates the first The minimum effective resolution of each test data field. Indicates the first The minimum effective resolution of each collected data field. Indicates the number of numerical fields to which the effect belongs. This indicates the number of numeric fields inherited from the previous page. This represents the set of inherited migration samples for the block to which the delayed test results belong after the current work section is updated. Indicates the number of samples in the set. This indicates the current responsibility value for the work section. Indicates the current work section identifier. This represents the sequence number of any inherited migration sample in the set.

[0082] The current section responsibility value is calculated only for sections whose inherited migration sample set corresponding to the delayed test results update block is not empty. When the inherited migration sample set corresponding to any section is empty, the current section responsibility value is not calculated, and the responsible section is determined from the sections whose inherited migration sample set is not empty.

[0083] The higher the current section responsibility value, the more likely the current section is to be the responsible section in the mineral flow identity chain corresponding to the current late test result, causing the inconsistency between the subsequent attribution value record in the current section migration output record and the subsequent attribution value record of the same field in the updated subsequent attribution block. For crushing section, grinding and classification section, flotation section and thickening and dewatering section on the same mineral flow identity chain, the section with the largest current section responsibility value is selected as the responsible section. When multiple sections have the same current section responsibility value, the section that is closer to the sampling position of the late test result in the mineral flow identity chain is selected as the responsible section.

[0084] After the responsible work section is determined, only the section migration parameters corresponding to the responsible work section in the large vertical model of the entire gold ore beneficiation process are locally re-migrated. During the local re-migrating, the responsible work section inheritance migration sample of the updated effect of the delayed test result is extracted, and at the same time, the section inheritance migration samples that are directly connected to the current responsible work section inheritance migration sample under the unified time identifier and belong to the same responsible work section and the same ore flow identity chain are extracted. The current responsible work section inheritance migration sample and the section inheritance migration samples directly connected to it are used together as the local re-migrating sample of the responsible work section. The original parameters of the large vertical model of the entire gold ore beneficiation process and the section migration parameters corresponding to non-responsible work sections are not updated. Only the front matrix and back matrix corresponding to the responsible work section are updated. The front inheritance block in the responsible work section inheritance migration sample is used as the starting part of the current work section migration input record. The observation block and process action block of this section are connected in sequence, and the local re-migrating is performed on the updated effect of the block.

[0085] The local re-migration of the responsible work section adopts the supervised fine-tuning cross-entropy loss function and AdamW optimizer. The current work section migration input record corresponding to the local re-migration sample of the responsible work section is used as input, and the current work section migration output record composed of the current work section entry condition numerical record and the updated aftereffect belonging block is used as target output. The front matrix and back matrix corresponding to the responsible work section are updated only according to the supervised fine-tuning cross-entropy loss function.

[0086] During the partial re-migration of the responsible section, the partial re-migration samples of the responsible section are repeatedly traversed. After each traversal, the differences between the consequential value records in the current section migration output records generated by the gold ore beneficiation vertical class model and the consequential value records of the same field in the updated consequential value block are compared, as well as the differences between the current section entry condition value records generated by the same value replacement method and the preceding inherited value records of the same field in the preceding inherited block. When the differences of the two types obtained in two adjacent traversals no longer continue to decrease, the partial re-migration of the responsible section is considered to be completed. After the partial re-migration of the responsible section is completed, the preceding and following matrices corresponding to the responsible section are kept unchanged, and the gold ore beneficiation vertical class model after the partial re-migration is completed is used as the current gold ore beneficiation vertical class model. If new late test results are written later, backfilling, responsible section positioning, and partial re-migration are continued along the ore flow identity chain.

[0087] like Figure 5As shown, the horizontal axis represents the delay time of late test results, and the vertical axis represents the hit rate of late test result backfilling. The curve of this invention represents the hit rate of writing late test results into the subsequent attribution block after forming mineral flow segments according to a unified time identifier, determining the main upstream source and auxiliary source according to the mineral flow attribution degree and forming a mineral flow identity chain, and then selecting the backfilling path according to the path attribution value. Compared with the curve of Scheme 1, which represents the hit rate when matching is performed only according to the proximity of sampling time, compared with the curve of Scheme 2, which represents the hit rate when backtracking only according to a fixed time window, and compared with the curve of Scheme 3, which represents the hit rate when merging only according to sampling position, from Figure 5 As can be seen from the graph, as the delay time of late test results increases, all curves generally decline, indicating that the increase in delay time increases the difficulty of corresponding to the preceding conditions. However, the curve of the present invention is always at the top and has the smallest downward slope, indicating that the ore flow identity chain can maintain the continuity of the main chain under the conditions of mid-ore regrinding and return, loop switching and multi-path coexistence, and improve the stable correspondence between late test results and preceding conditions through path attribution values. No local rebound peaks or abnormal sudden peaks appear in the figure, indicating that there is no obvious chain jitter in the backfilling process in different delay intervals. The continuous vertical spacing between the curves further indicates that the present invention has a stable advantage in improving the accuracy of data organization throughout the entire process.

[0088] like Figure 6 As shown, the horizontal axis represents the number of re-migration traversal rounds, and the vertical axis represents the normalized difference value. In this invention's scheme: the updated aftereffect attribution block difference curve represents the difference between the output result of the gold mine beneficiation vertical class model and the updated aftereffect attribution block after reconstructing the current section migration input record and current section migration output record of the section inheritance migration sample updated by the delayed test results. In this invention's scheme: the updated preceding inheritance block difference curve represents the difference between the current section entry condition value record in the current section migration output record formed after the model reads the preceding inheritance block and the preceding inheritance value record in the preceding inheritance block. Comparing the curve of Scheme 1, this shows the change in the direction difference of the aftereffect attribution block when re-migration is performed directly on all sections without locating the responsible section. Comparing the curve of Scheme 2, this shows the change in the direction difference of the aftereffect attribution block when re-migration is performed randomly on selected sections without considering the current section responsibility value. Figure 6As can be seen from the figure, the two difference curves in the scheme of the present invention decrease rapidly and stabilize earlier within a few re-migration traversal rounds. This indicates that after determining the responsible work section based on the backfilled work section inheritance migration sample, performing local re-migration only on the work section migration parameters corresponding to the responsible work section can simultaneously reduce the difference in the direction of the subsequent belonging block and the difference in the direction of the preceding inherited block, and reach the local re-migration completion state more quickly. No obvious rebound peak appears in the figure, indicating that there is no significant overshoot or unstable oscillation in the local re-migration process. However, the curves corresponding to the overall re-migration of the entire work section and the random work section re-migration decrease more slowly or have larger residual differences. This shows that when the present invention performs directional updates according to the work section inheritance relationship, it can maintain the adaptability of the model while taking into account the migration efficiency.

[0089] In summary, this invention achieves continuous expression of cross-section material transfer relationships and stable correspondence between delayed test results and preceding conditions by collecting raw ore property data, online instrument data, video image data, equipment status data, and laboratory sampling records from a gold mine's darkroom beneficiation plant, and establishing a mineral flow identity chain according to the material transfer sequence. This improves the accuracy of data organization throughout the entire process. By backfilling delayed test results along the mineral flow identity chain to the subsequent attribution block, locating the responsible section based on the inherited migration samples after backfilling, and performing local re-migration on the corresponding section in the large vertical model of the entire gold mine beneficiation process, this invention achieves targeted updates based on section inheritance relationships, balancing migration efficiency and model adaptability.

[0090] 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 highly efficient transfer method for a large-scale vertical model of the entire mineral processing process, characterized in that, include: Step S1: Collect raw ore property data, online instrument data, video image data, equipment status data and laboratory sampling records from the gold mine's black-light beneficiation plant, and establish an ore flow identity chain according to the material transfer sequence; Step S2: Based on the mineral flow identity chain, generate a segment inheritance migration sample according to the process sequence, consisting of the current observation block, the previous segment inheritance block, the process action block, and the subsequent attribution block. Step S3: Based on the segment inheritance migration sample, perform segment inheritance migration on the general basic model according to the process sequence. Keep the original parameters of the general basic model and the segment migration parameters corresponding to the completed segments unchanged. Only update the segment migration parameters corresponding to the current segment. Use the previous segment inheritance block as the starting part of the current segment migration input record to complete the current segment migration. Complete the current segment migration according to the previous segment inheritance block to obtain the vertical class model of the entire gold mine beneficiation process. Step S4: Backfill the late test results along the mineral flow identity chain to the subsequent ownership block, locate the responsible section based on the backfilled section inheritance migration sample, and perform local re-migration on the corresponding section in the gold mine beneficiation vertical category model. The specific steps for collecting raw ore property data, online instrument data, video image data, equipment status data, and laboratory sampling records from a gold mine's "dark light" beneficiation plant are as follows: At the raw ore inlet, crushing discharge, grinding feed, classification overflow, roughing feed, scavenging feed, cleaning feed, middlings regrinding reflux, concentrate thickening feed, and tailings discharge locations, connect the raw ore property data acquisition interface, online instrument data acquisition interface, video image data acquisition interface, equipment status data acquisition interface, and laboratory sampling record interface to the same clock source, and scan the records at each acquisition location according to a unified time identifier and process sequence; When the equipment status remains unchanged, the material flow direction does not change, and the feed rate or slurry flow rate remains continuous, adjacent records are merged to form a ore flow segment. When the equipment stops, starts, pumps and valves are switched, loops are switched, flow is interrupted, feed is interrupted, or the sampling location is changed, the current ore flow segment is cut off and a new ore flow segment is generated. The specific steps for establishing the mineral flow identity chain according to the material transfer order are as follows: for each downstream mineral flow segment, search for upstream candidate mineral flow segments that are temporally connected to the downstream mineral flow segment in the upstream collection location according to the material transfer order, and determine the mineral flow attribution degree based on the actual transfer time, material quantity, solid concentration and stable operation history. The upstream candidate mineral flow segment with the highest mineral flow attribution is selected as the main upstream source, and the upstream candidate mineral flow segments with non-zero mineral flow attribution are retained as auxiliary sources. The main upstream sources are sequentially connected to form a mineral flow identity chain according to the material transfer order. When the middlings are refluxed or the loop is switched, the main upstream source is used to maintain the continuity of the main chain of the mineral flow identity chain and the auxiliary sources are retained. A unique identity identifier is written for each mineral flow identity chain according to the formation order.

2. The efficient transfer method for a large-scale vertical model of the entire mineral processing process as described in claim 1, characterized in that, The process inheritance migration sample, generated according to the process sequence and consisting of the current observation block, the previous inherited block, the process action block, and the subsequent attribution block, includes: Read the ore flow identity chain one by one according to the unique identifier, and divide it into crushing section, grinding and classification section, flotation section and thickening and dewatering section according to the material transfer order. For the same ore flow identity chain, generate a section inheritance migration sample for each section. For each section of the inherited migration sample, a current section observation block, a previous section inheritance block, a process action block, and a subsequent attribution block are generated sequentially. The current section observation block contains the raw ore property data, online instrument data, video image data, and equipment status data records of the corresponding ore flow segment in the current section. The previous section inheritance block of the crushing section sample contains the entry condition record of the ore flow segment corresponding to the raw ore selection position in the same ore flow identity chain. The previous section inheritance block of the other section samples contains the entry condition record of the main upstream source ore flow segment corresponding to the previous section in the same ore flow identity chain. Records in the previous section inheritance block, except for image records and operation status records, are extracted as previous section inheritance value records in their original order. The previous section inheritance value records serve as the entry conditions for the current section. Image records and operation status records are retained in the previous section inheritance block. The process action block contains the records of actual changes within the current section. The subsequent attribution block contains the downstream laboratory sampling records on the same ore flow identity chain.

3. The efficient transfer method for a large-scale vertical model of the entire mineral processing process as described in claim 2, characterized in that, The process of generating a segment inheritance migration sample based on the process sequence, consisting of the current observation block, the previous segment inheritance block, the process action block, and the subsequent effect attribution block, further includes: When there are multiple interconnected downstream testing and sampling records on the same mineral flow identity chain, the path attribution value between the current section and the downstream testing and sampling record is determined according to the mineral flow attribution degree. The concentrate testing and sampling record with the largest path attribution value and the tailings testing and sampling record with the largest path attribution value are written into the subsequent attribution block respectively. The stage testing and sampling records are appended to the subsequent attribution block in the order of their appearance in the mineral flow identity chain. For affiliated sources, the connection relationship and path attribution value corresponding to the affiliated source are retained, and the testing results corresponding to the affiliated source are not recorded as the real subsequent attribution value and written into the subsequent attribution block. After all the observation blocks, previous inheritance blocks, process action blocks, and subsequent attribution blocks in this segment are generated, they are written into the same segment inheritance migration sample in the order of unique identifier, current segment name, observation block, previous inheritance block, process action block, and subsequent attribution block.

4. The efficient transfer method for a large-scale vertical model of the entire mineral processing process as described in claim 1 or 3, characterized in that, The specific steps for performing segment inheritance migration on the general basic large model according to the process sequence are as follows: The Chinese instruction following general basic model is selected as the general basic model. The original parameters of the general basic model are not updated. The corresponding section migration parameters for crushing, grinding and classification, flotation and thickening and dewatering are set at the matrix parameters that represent the current input record. The section migration parameters for each section include a front matrix and a back matrix. The product of the front matrix and the back matrix is ​​used as a low-rank incremental matrix with the same matrix shape as the corresponding original matrix parameter. Each segment inheritance migration sample is constructed into a current segment migration input record and a current segment migration output record. The current segment migration input record is written in the order of unique identifier, current segment name, previous segment inheritance block, current segment observation block, and process action block. The current segment migration output record consists of the current segment entry condition value record and the subsequent attribution block. The current segment entry condition value record is generated according to the field names and writing order of the previous segment inheritance value record, and the previous segment inheritance value record in the previous segment inheritance block is used as the target record. The subsequent attribution block uses the actual subsequent attribution block in the current segment inheritance migration sample as the target record. The segment inheritance migration is performed in the order of crushing segment, grinding and classification segment, flotation segment, and thickening and dewatering segment. Each time, only the segment migration parameters corresponding to the current segment are updated, and the current segment migration is completed by using the previous segment inheritance block as the starting part of the current segment migration input record.

5. The efficient transfer method for a large-scale vertical model of the entire mineral processing process as described in claim 4, characterized in that, The specific steps for migrating the current work section based on the previous inherited block are as follows: During the migration process of each work segment, the differences between the consequential value records in the migration output records of the current work segment generated by the general basic model and the consequential value records of the same field in the actual consequential value block are compared simultaneously. The differences between the current work segment entry condition value records in the migration output records of the current work segment and the previous inheritance value records of the same field in the previous inheritance block are also compared. All work segment inheritance migration samples of the current work segment are repeatedly traversed, and the two types of differences obtained from two adjacent traversals are compared. When the difference obtained from two adjacent traversals is not less than the difference obtained from the previous traversal, it is determined that the corresponding difference will no longer decrease. When both types of differences no longer decrease, it is determined that the current work segment has completed the migration.

6. The efficient transfer method for a large-scale vertical model of the entire mineral processing process as described in claim 5, characterized in that, The specific steps for obtaining the large-scale vertical model of the entire gold ore beneficiation process are as follows: After the current section is migrated, the migration parameters corresponding to the current section are not updated, and the same section inheritance migration is immediately performed on the next section. If the current section has not yet been migrated, the migration of the next section is not started in advance. For the crushing section samples, grinding and classifying section samples, flotation section samples, and thickening and dewatering section samples on the same ore flow identity chain, the migration order of crushing section, grinding and classifying section, flotation section, and thickening and dewatering section is always maintained. After all sections are migrated, the general basic model and the section migration parameters corresponding to the crushing section, grinding and classifying section, flotation section, and thickening and dewatering section that have completed the section inheritance migration are jointly solidified into a vertical large model of the entire gold ore beneficiation process.

7. The efficient transfer method for a large-scale vertical model of the entire mineral processing process as described in claim 1 or 6, characterized in that, The specific steps for backfilling late test results into the subsequent attribution block along the mineral flow identity chain are as follows: Late test results are continuously read in the order of being written with the time stamp. For each late test result, the sampling location, sampling time and sample number corresponding to the current late test result are read. The mineral flow segment connected to the sampling location, sampling time and sample number is searched in the mineral flow identity chain according to the unique identity. The process is traced upstream segment by segment along the main chain that has been formed by the main upstream source. The inherited migration samples of the work segment hit during the backtracking process and their connection relationship are formed into candidate backfill paths. At the same time, the connection relationship of the auxiliary source in the mineral flow identity chain is preserved. When there is only one candidate backfill path, the current late test result is written into the subsequent ownership block of the inherited migration sample of the corresponding work section on the candidate backfill path in the order of sampling location, sampling time, sample number and test result. When there are multiple candidate backfill paths, the candidate backfill path with the largest path ownership value is selected as the backfill path, and the current late test result is written into the subsequent ownership block of the inherited migration sample of the corresponding work section on the backfill path as the real subsequent ownership value record. For the other candidate backfill paths that are not selected, only the connection relationship and path ownership value in the mineral flow identity chain are retained, and the current late test result is not written into the subsequent ownership block as the real subsequent ownership value record. If there is already a record with the same sampling location, the same sampling time and the same sample number in the subsequent ownership block, the current late test result replaces the original record.

8. The efficient transfer method for a large-scale vertical model of the entire mineral processing process as described in claim 7, characterized in that, The steps for locating the responsible work section based on the inherited migration sample of the backfilled work section, and performing local re-migration on the corresponding work section in the large vertical model of the entire gold ore beneficiation process, are as follows: For the inherited migration samples of the work section updated by the current late test result, the current work section migration input record and the current work section migration output record are reconstructed and input into the large vertical model of the entire gold mine beneficiation process. The differences between the aftereffect attribution value record in the current work section migration output record and the aftereffect attribution value record of the same field in the updated aftereffect attribution block are compared, as well as the differences between the current work section entry condition value record in the current work section migration output record and the previous section inheritance value record in the previous section inheritance block. For the crushing section, grinding and classification section, flotation section and thickening and dewatering section, the current work section responsibility value is calculated for each work section with an aftereffect attribution block updated by the current late test result. The work section with the largest current work section responsibility value is determined as the responsible work section. Local re-migration is performed only on the migration parameters of the responsible section, and the inherited migration samples of the current responsible section and the inherited migration samples of the directly connected sections are used together as local re-migration samples. When the differences between the two types of results obtained in two adjacent traversals no longer continue to decrease, the local re-migration of the responsible section is considered to be completed, and the full-process vertical model of gold ore beneficiation after the local re-migration is used as the current full-process vertical model of gold ore beneficiation.

Images