A knowledge graph construction method, device and medium for gold mine beneficiation

Abstract

The application discloses a kind of gold mine selection metallurgy-oriented knowledge graph construction method, equipment and medium, it is related to mining informatization technical field, including, based on the topological establishment of selection metallurgy process Directional connection is generated and point position and test mapping;Access selection metallurgy data, according to cumulative ore quality generation process batch and determine batch time window, form batch instance to standard variable aggregation;In batch, according to the intensity of variable change, split working condition event and generate event statistical vector;Determine circuit inlet and outlet stream, carry out event level metal amount closure and interval check, fusion obtains evidence intensity;With evidence intensity update the reliability of constraint interval, intervention action and cause-effect hypothesis, incremental writing and re-verification to affected subgraph.The application carries out event level metal amount closure and interval check by circuit inlet and outlet stream and fuses to generate evidence intensity, realizes with calculable evidence quantization constraint knowledge reliability, improves the reliability and error-correcting property of entry graph knowledge.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of informationization of mining industry, and particularly relates to a knowledge graph construction method for gold ore dressing and smelting, equipment and medium. BACKGROUND

[0002] In recent years, the digitization, networking and data elementization of the gold ore dressing and smelting process continue to advance, and the process control and monitoring means are generally deployed in the dressing plant, forming process data, test data and production record data covering ore supply, ore grinding and classification, leaching and adsorption, desorption and electric accumulation and tailing treatment and the like. At the same time, the knowledge organization technology for complex processes gradually evolves from 'document archiving and index reporting' to'semantic expression and associated retrieval', and the process topology is taken as a carrier to structurally describe the process, equipment, stream, index and process constraints, so that the cross-system data can be aggregated around the process object, providing support for operation analysis, abnormality tracing and process management, and the related method has been concerned and applied in the semantic application of mining informationization and process industry.

[0003] However, the existing related technology still has two key deficiencies: firstly, multi-source data is often loosely associated by time stamp or system field, lacks a unified semantic caliber and batch organization mechanism for the ore dressing and smelting process topology, and it is difficult to realize the unique alignment of test samples and document evidence within the batch time window, resulting in the batch instance being uncalculated, the evidence chain being incomplete, and further unable to support the fine segmentation of batch working conditions and event-level feature precipitation. Secondly, the existing knowledge extraction and storage rely on text rules or statistical confidence, lack of mechanism consistency constraints such as event-level metal quantity closure and interval verification of loop inlet and outlet streams, and it is difficult to form a computable evidence strength to quantify the knowledge credibility. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a knowledge graph construction method for gold ore dressing and smelting to solve the problems that the batch, sample and document evidence cannot be uniquely aligned within the same time window and how to realize evidence-driven credibility update under the constraint of process topology.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a knowledge graph construction method for gold ore dressing and smelting, which comprises: establishing a directed connection table according to the process topology of ore dressing and smelting, and generating a mapping table;

[0008] Accessing smelting data based on a mapping table, triggering generation of a process batch according to cumulative ore quality and determining a batch time window, performing statistics and cumulative aggregation on standard variables within the batch time window, forming a batch traceability identifier, and obtaining a batch instance set;

[0009] In the batch instance set, extracting a key standard variable sequence for a process node and calculating a change intensity, determining a candidate boundary time set according to the change intensity, generating and merging a global boundary within the batch, obtaining a strongly bound working condition event set, and generating a statistical vector for each working condition event;

[0010] Determine the inlet and outlet streams of the loop through the working condition event set, perform metal amount closure accounting and interval verification, calculate the closure residual and interval verification data through the statistical vector, and generate the evidence strength;

[0011] Update the credibility of the constraint interval, intervention action and causal hypothesis through the evidence strength, and write into the knowledge graph according to the forgetting rule, and reverify the affected local subgraph.

[0012] As a preferred scheme of the knowledge graph construction method for gold ore smelting of the present application, wherein: the directed connection table is established according to the smelting process topology, and the mapping table is generated, and the specific steps are as follows,

[0013] Read the point list, assay field list and process file, identify the material flow connection in the process file, define each connection as a stream, and record the starting point and ending point, and obtain the directed connection table;

[0014] Fix the object type set and the relationship type set and specify the direction and mandatory fields, retrieve the corresponding cumulative point for each stream, map to the standard variable symbol and write into the mapping table;

[0015] Map the situation of each stream requiring grade to the standard variable symbol, and map the control point belonging to the node of each device to the standard variable symbol set and write into the mapping table.

[0016] As a preferred scheme of the knowledge graph construction method for gold ore smelting of the present application, wherein: the smelting data is accessed based on the mapping table, the process batch is generated according to the cumulative ore quality, and the batch time window is determined, and the specific steps are as follows,

[0017] Integrate and accumulate the main ore quality flow, and when the cumulative value reaches the batch quality threshold, determine the batch end time as the reaching time, and obtain the batch time window;

[0018] For each batch time window, the process point value is converted into a standard variable sequence according to the mapping table, and a time average and a key stream cumulative amount are generated within the batch time window to obtain a batch feature record;

[0019] The test sample is uniquely aligned with the batch time window by adopting a gating rule of containing the time window and minimizing the distance from the midpoint of the time window, so that the target batch time window is obtained.

[0020] As a preferred scheme of the knowledge graph construction method for gold ore dressing and smelting, the batch traceability identifier is formed, and a batch instance set is obtained, and the specific steps are as follows,

[0021] If the target batch is uniquely determined, the sampling point name, detection item name, detection result and sample number of the sample are written into the sample set of the batch corresponding to the target batch time window;

[0022] The evidence positioning information is generated for each document evidence, and the evidence set is obtained by uniquely aligning the document occurrence time with the batch time window;

[0023] The batch traceability identifier is generated for each batch based on the batch feature record, the batch traceability identifier, the batch time window, the sample set and the evidence set are combined into the batch instance set.

[0024] As a preferred scheme of the knowledge graph construction method for gold ore dressing and smelting, the batch traceability identifier is formed, and a batch instance set is obtained, and the specific steps are as follows,

[0025] In the directed connection table, the process nodes in each batch in the batch instance set which need to generate working condition events are fixed to form a node list;

[0026] The key standard variable sequence is extracted through each process node in the node list and normalized, and the change intensity is calculated.

[0027] When the change intensity is continuously not less than the change threshold and the duration is not less than the sliding window length, the first time that meets the condition is recorded as the candidate boundary time of the node.

[0028] As a preferred scheme of the knowledge graph construction method for gold ore dressing and smelting, the batch traceability identifier is formed, and a batch instance set is obtained, and the specific steps are as follows,

[0029] The candidate boundary moments generated by all process nodes within the same batch are merged into a global boundary set, and time tolerance deduplication is performed. The set of working condition events is obtained by dividing the batch by the batch start time and the batch end time.

[0030] Add the batch start time and batch end time to the beginning and end of the deduplicated global boundary list, and use the time periods between adjacent boundaries as the time windows of the working condition events in sequence.

[0031] Calculate statistical vectors for the key standard variable sequences within each working condition event time window.

[0032] As a preferred embodiment of the knowledge graph construction method for gold ore beneficiation and metallurgy described in this invention, the steps of determining the loop inlet and outlet streams through a set of operating events, performing metal quantity closure calculation and interval verification, calculating the closure residual and interval verification data through statistical vectors, and generating evidence strength are as follows.

[0033] In the set of operating events, determine the set of loop inlet and outlet streams for each operating event, extract the cumulative quality according to the event time window, and generate a grade representative value for each stream;

[0034] The metal quantity is closed-loop accounting is performed by accumulating representative values ​​of mass and grade, and the closed-loop residual is calculated.

[0035] Perform interval verification on each event at the device node, calculate the interval violation degree based on the statistical vector, and summarize it into interval verification data;

[0036] The closure residuals and interval verification data are combined to form the strength of evidence.

[0037] As a preferred embodiment of the knowledge graph construction method for gold ore beneficiation and metallurgy described in this invention, the steps of updating the credibility of constraint intervals, intervention actions, and causal hypotheses based on evidence strength, writing them into the knowledge graph according to a fixed forgetting rule, and re-verifying the affected local subgraphs are as follows.

[0038] The constraint interval, intervention actions, and causal assumptions are used to generate a candidate knowledge set through rule extraction.

[0039] The evidence strength of candidate knowledge associations is fused into a single candidate knowledge evidence value using a strong evidence priority fusion method.

[0040] Maintain a credibility state value for each piece of knowledge content that has been written into the knowledge graph. Generate an updated credibility state value based on the forgetting rule using candidate knowledge evidence values ​​and write it into the knowledge graph. Re-verify the affected local subgraphs.

[0041] The forgetting rule means that the updated credibility state value is composed of a fixed proportion of the previous version credibility state value and a remaining proportion of the current round candidate knowledge evidence value.

[0042] In a second aspect, the present application provides a computer device comprising a memory and a processor, and the memory stores a computer program, wherein: the computer program is executed by the processor to implement any step of the knowledge graph construction method for gold ore dressing and smelting according to the first aspect of the present application.

[0043] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein: the computer program is executed by the processor to implement any step of the knowledge graph construction method for gold ore dressing and smelting according to the first aspect of the present application.

[0044] The present application has the beneficial effects that: by triggering the generation of batches according to the cumulative ore quality and forming batch traceability identifiers, the batch caliber is stable and unified, and the batch instance is traceable and recalculable; by generating global boundaries based on variable change intensity within the batch and cutting work condition events and generating event statistical vectors, automatic segmentation of work conditions within the batch and event-level feature precipitation are realized, supporting the refinement of knowledge into the graph from the batch level to the event level; by carrying out event-level metal quantity closure and interval verification according to the circuit inlet and outlet streams and fusing to generate evidence strength, the knowledge credibility is quantitatively constrained by calculable evidence, and the reliability and error-correcting property of the knowledge into the graph are improved. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0046] Figure 1 The flowchart of the knowledge graph construction method for gold ore dressing and smelting.

[0047] Figure 2 The flowchart for obtaining the batch instance set.

[0048] Figure 3 The flowchart for obtaining the work condition event set and generating the statistical vector.

[0049] Figure 4 The flowchart for knowledge graph incremental writing and re-verification rollback.

[0050] Figure 5 The comparative data graph of the change intensity multi-curve.

[0051] Figure 6A comparison data graph of the credibility state value multi-curve. DETAILED DESCRIPTION

[0052] In order to make the above objectives, features and advantages of the present application more apparent, specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0053] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details given herewith. In other instances, well-known methods have not been described in detail in order to avoid obscuring aspects of the present application.

[0054] Secondly, the "one embodiment" or "embodiment" referred to herein can include specific features, structures or characteristics in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments.

[0055] Reference Figures 1-6 For one embodiment of the present application, the embodiment provides a knowledge graph construction method for gold mine beneficiation and smelting, including the following steps:

[0056] S1, a directed connection table is established according to the beneficiation and smelting process topology, and a mapping table is generated.

[0057] Read the point list (including point name, engineering quantity unit, range and sampling period), the test field list (including sampling point, detection item, unit and output time field) and the process file (including the connection relationship of process and equipment), identify the process name and equipment position number in the process file, assign a unique identifier to each process or equipment, record the name, the process segment to which it belongs and the drawing positioning information; identify the material flow connection in the process file, define each connection as a stream, assign a unique identifier to each stream, and record the start and end points to obtain a directed connection table.

[0058] A set of fixed object types is included, including batches (for carrying material in the time window of aggregation information), working condition events (for carrying internal running segment information of the batch), processes, devices, streams, indicators (such as grade, recovery rate, particle size, etc.), constraints (for carrying closed accounting and interval checking results), and evidence (for carrying document positioning or data window index); a set of fixed relationship types is defined, and the direction and mandatory fields are specified. Specifically, the stream starting point relationship is stream→process or device; the stream end point relationship is stream→process or device; the working condition event associated batch relationship is working condition event→batch; the working condition event occurs relationship is working condition event→process or device; the constraint acts on relationship is constraint→working condition event; the evidence supports relationship is evidence→any object or relationship; the mandatory field list is fixed for batch, working condition event, constraint, and evidence (for example, start time, end time, source fingerprint, window index, and checking mark), and the semantic skeleton list is obtained.

[0059] For each point or test field, a unique standard variable symbol is assigned, and the standard unit, effective range upper and lower limit, sampling period or output period, and corresponding original field name are recorded.

[0060] For each stream, the corresponding cumulative point in the point list is retrieved, mapped to the standard variable symbol, and written to the mapping table; for each stream that needs grade, the corresponding sampling point and detection item in the test field list are retrieved, mapped to the standard variable symbol, and written to the mapping table; for each process or device node, the control points (such as pH, reagent flow, temperature) belonging to the node are mapped to a set of standard variable symbols and written to the mapping table; device aliases, reagent aliases, and process aliases in the procedure or record are unified to standard variable symbols through a preset synonym table, and the term→target identifier is written to the mapping table; perform uniqueness verification: for any original field name or term, if there are two or more target identifiers, it is determined as failed and the to-be-verified entry is output, and only after manual confirmation is it allowed to be written to the mapping table.

[0061] S2, based on the mapping table, access the smelting data, trigger the generation of the process batch according to the cumulative ore quality, determine the batch time window, and aggregate the standard variables in the batch time window to form the batch traceability identifier and obtain the batch instance set.

[0062] Integrate the main feed ore mass flow between the batch start time and the current time, and when the accumulated value reaches the preset batch mass threshold, determine the batch end time as the time when the accumulated value reaches the preset batch mass threshold, and obtain the batch time window; for each batch time window, convert the process point value into a standard variable sequence according to the mapping table, and generate statistics and accumulations within the batch time window, take the time average of the control variables, and take the accumulations of the flow variables and the time average; generate a batch feature record for each batch, including the batch start time, the batch end time, the main feed ore cumulative mass, the time average of the key variables, and the cumulative amount of the key flow.

[0063] It should be noted that the preset batch mass threshold is formed by taking the main feed ore belt scale data in the last 30 days, excluding downtime, failure and ore blending switching period, calculating the cumulative ore mass in the remaining data with a 1-hour window, taking the 80th percentile value of the sample set and rounding up as the preset batch mass threshold.

[0064] The gating rule of time window inclusion and minimum distance from the time window midpoint determines the unique attribution, which uniquely aligns the test sample with the batch time window, and the expression is:

[0065] ;

[0066] ;

[0067] Wherein, represents the target batch time window, represents the candidate batch time window, represents the sample sampling time, represents the batch midpoint time, represents the batch start time, represents the batch end time, represents the alignment time tolerance, represents the constraint condition.

[0068] It should be noted that, The absolute value of the difference between the sampling time of the same sampling point in the last 30 days and the corresponding test record storage time is taken as the sample set, and the 95th percentile value of the sample set is taken as the alignment time tolerance.

[0069] If there is no batch that meets the constraint condition, or there are two or more batches that meet the constraint condition at the same time and the distance is parallel minimum, the sample is written into the to-be-checked list and not written into any batch of valid sample set; if the target batch is uniquely determined, the sample point name, detection item name, detection result and sample number of the sample are written into the sample set of the batch corresponding to the target batch time window.

[0070] For each piece of document evidence (including procedure clauses, operation record entries, and maintenance record entries), evidence location information is generated, and unique alignment is performed according to the document occurrence time and batch time window. Specifically, if the document evidence has a clear occurrence time (e.g., operation record time, maintenance start time), the unique batch to which it belongs is determined according to the gating rules. If only the document generation time or release date exists, the document generation time or release date is used as the document time. The document name, version number, page number, paragraph number, original text fragment location information, standardized process name or equipment name, and standardized logistics channel name or indicator name are written into the evidence set.

[0071] For each batch, a batch traceability identifier is generated, which includes batch time window information, batch aggregate summary, uniquely aligned samples within the batch, and document location summary. Specifically, the batch time window information includes the batch start time, batch end time, and cumulative mass of the main feed ore; the batch aggregate summary is obtained by splicing batch feature records in a fixed order; and the document location summary is a summary string written in a fixed order from evidence fragments in the evidence set.

[0072] The batch traceability identifier, batch time window, sample set, and evidence set are combined into a batch instance set.

[0073] S3. In the batch instance set, extract the key standard variable sequence for the process node and calculate the change intensity. Determine the candidate boundary moment set based on the change intensity, generate and merge the global boundary within the batch, obtain the set of strongly bound working condition events of the batch, and generate a statistical vector for each working condition event.

[0074] In the directed connection table, read the correspondence between each process node and its adjacent logistics channel, fix the process nodes that need to generate working condition events in each batch in the batch instance set. Specifically, the process or equipment node that has at least one control measurement point data sequence and is located on the critical loop path within the batch time window is arranged in the order of process flow to form a node list.

[0075] For each process node in the node list, extract key standard variable sequences, including quality flow variables of the logistics channels associated with the node, reagent addition flow variables associated with the node, and key chemical environment variables of the node (acidity, alkalinity, temperature, and presence of dissolved oxygen), and normalize each variable sequence according to its effective range.

[0076] Within each process node, the intensity of change is calculated for the normalized sequence of key variables, expressed as follows:

[0077] ;

[0078] in, Indicates the first a flow node at time a change intensity, a number of key variables participating in the calculation of the flow node, a sliding integral window length, a time at the right end of the sliding window, a key variable number, an integral time variable, a normalized key variable value of the key variable.

[0079] When the change intensity is continuously not lower than the change threshold value and the duration is not less than the sliding window length, the first time that meets the condition is recorded as the candidate boundary time of the node, and all candidate boundary times are collected as the candidate boundary time set of the node.

[0080] It should be noted that the change threshold value is obtained by taking the stable running data of the last 30 days without "no ore matching switching, no maintenance downtime, no fault alarm of point position", calculating the change intensity, obtaining the change intensity time series of each node, collecting to form a change intensity sample set, and taking the 95th percentile value of the change intensity sample set as the change threshold value.

[0081] As Figure 5 illustrates the key process of generating global boundaries based on variable change intensity within a batch, splitting the working condition event, and generating event statistical vectors; the upper figure is a comparison of multivariate change intensity curves within the same batch time window; the horizontal dashed line is the change threshold value, the red dashed rectangle and the double-headed arrow indicate the selected local comparison interval, which is used to highlight the period when the working condition changes more concentrated; the lower figure is a local magnification result of the interval, the vertical dashed line and the arrow indicate the time when the difference between the two curves is the largest, and the change intensity value of the corresponding variable is also given at this time; by comparing the overview and local magnification, it can be seen that when the key variable change intensity presents obvious difference or synchronous transition near the threshold value, the global boundary set can be formed based on the candidate boundary time, and the batch time window can be divided into a working condition event set accordingly, so as to realize the automatic segmentation of working conditions within the batch.

[0082] The candidate boundary time generated by all process nodes in the same batch is combined into a global boundary set, and time tolerance is removed. The working condition event set is divided according to the batch start time and the batch end time. The batch start time is taken as the starting boundary of the global boundary list, and the batch end time is taken as the terminal boundary of the global boundary list. The batch start time and the batch end time are added to the head and tail of the global boundary list after deduplication, and the time period between adjacent boundaries is sequentially taken as the working condition event time window. The statistical vector of the key standard variable sequence in each working condition event time window is calculated. The statistical vector includes a time mean item and an accumulation item. The time mean item is the ratio of the sum of all sampling record values to the number of sampling records in the time period from the event start time to the event end time. The accumulation item is obtained by arranging the sampling record sequence in ascending order of sampling time from the event start time to the event end time. For each pair of adjacent sampling records in the sampling record sequence, the time difference between the two records is calculated, and the average of the variable values of the two records is calculated. The product of the time difference and the average is taken as the increment of the adjacent sampling interval. The increments of all adjacent sampling intervals in the event time window are summed to obtain the accumulation.

[0083] An evidence identifier is generated for each event, including batch traceability identifier, participating variable list and original sampling index range.

[0084] S4, determine the loop inlet and outlet streams through the working condition event set, perform metal amount closure calculation and interval verification, calculate the closure residual and interval verification data through the statistical vector, and generate the evidence strength.

[0085] In the working condition event set, the loop inlet and outlet stream set is determined for each working condition event. Specifically, the loop containing the process or device node is retrieved in the loop definition table. If only one loop is retrieved, the loop is taken as the accounting loop of the event. If multiple loops are retrieved, the loop with the maximum sum of in-degree and out-degree of the selected node in the directed connection table is selected as the accounting loop, and the unselected loops are written into the to-be-checked list. The inlet stream list and outlet stream list are read from the selected loop, and the mass flow point mapping and gold grade sampling point mapping corresponding to each stream are further read from the stream definition table. If a certain outlet stream lacks a gold grade sampling point mapping, the event is marked as grade missing and written into the to-be-checked list. The inlet stream set and outlet stream set corresponding to each event are obtained.

[0086] It should be noted that the loop definition table is automatically generated by retrieving closed paths from the directed connection list and determining the loop entry and exit logistics channels according to the rules. The rules for determining the loop entry and exit logistics channels are as follows: for each logistics channel in the directed connection list, check whether the start and end points belong to the loop node set respectively. If the end point of the logistics channel belongs to the loop node set and the start point does not belong to the loop node set, the logistics channel is determined to be the loop entry logistics channel; if the start point of the logistics channel belongs to the loop node set and the end point does not belong to the loop node set, the logistics channel is determined to be the loop exit logistics channel. The flow stream definition table is obtained by registering each logistics channel in the process topology as a flow stream record, and writing the name of the logistics channel, the starting process or equipment, and the ending process or equipment. Then, according to the mapping table, the names of the quality flow points, sampling points, and gold content test fields corresponding to the logistics channels are added to the same record, and finally, the flow stream definition table is obtained by summarizing them.

[0087] For each event, the cumulative mass of the inlet and outlet streams is extracted according to the event time window, and a grade representative value is generated for each stream. Specifically, for each stream, based on the mass flow rate point mapping, the flow rate sampling sequence within the event time window is read from the batch instance, and the product of the time difference between adjacent sampling times and the average of the two adjacent flow rates is used as the interval increment. Then, all interval increments are summed to obtain the cumulative mass within the event time window. For each stream, based on the gold grade sampling point mapping, the sample records of the sampling points are selected from the sample set of the batch, and the gold grade of the sample whose sampling time is closest to the midpoint of the event time window is taken as the grade representative value.

[0088] For each event that meets the sample availability condition, perform a metal quantity closure check and calculate the closure residual, expressed as:

[0089] ;

[0090] ;

[0091] in, This indicates the amount of metal in a given stream within the event time window. This indicates the cumulative mass of the stream within the event time window. This represents the grade value of the flowing stock within the event time window. Indicates closed residual. This represents the total amount of metal flowing into the inlet stream. This represents the total amount of exported metals. This represents the amount of metal quantity that can be accounted for during the event time window, whether it represents a loss or inventory change.

[0092] Interval check is performed on each event at the associated process or equipment node, the lower bound and upper bound of the allowed range of the node key control variable are read from the standard variable dictionary, the time mean value in the statistical vector is compared with the allowed range, the interval violation degree is calculated and the arithmetic mean of all interval violation degrees is taken as the interval check data, the expression is:

[0093] ;

[0094] wherein, represents the interval violation degree of the i-th control variable in the event time window, represents the lower bound of the control variable allowed range, represents the mean value of the control variable in the event time window, represents the upper bound of the control variable allowed range.

[0095] It should be noted that the standard variable dictionary is obtained by reading each instrument point and each test field one by one, first determining the physical meaning name (for example, mass flow, cumulative mass, pH, reagent addition flow, gold grade) and standard unit caliber (for example, tons per hour, tons, dimensionless, liters per hour, grams per ton), and then writing the original field name, standard name, standard unit, range upper and lower limit, sampling period or output period, data source identification information into the corresponding entry to obtain the standard variable dictionary.

[0096] The closed residual and the interval check data are fused into the evidence strength, the expression is:

[0097] ;

[0098] wherein, represents the interval check data, represents the evidence strength, represents the scale parameter of the closed residual, represents the scale parameter of the interval check data.

[0099] It should be noted that, is the median value of the closed residual sample set in the stable running event in the last 30 days, such as the 50th percentile value; is the median value of the interval violation degree sample set in the stable running event in the last 30 days, such as the 50th percentile value.

[0100] S5, the credibility of the constraint interval, the intervention action and the causal hypothesis is updated by the evidence strength, and the knowledge graph is written according to the forgetting rule, and the affected local subgraph is reverified.

[0101] ​Based on the document evidence, a rule extraction method is used to generate a candidate knowledge set, which includes constraint intervals, intervention actions and causal assumptions. The constraint interval is described as the allowable range of a control variable on a certain process or device. The intervention action is described as the trigger condition, execution action, object and expected impact. The causal assumption is described as the influence direction of the change of a certain factor on the change of a certain index under the applicable conditions. An action scope identifier is added to each candidate knowledge, which is composed of process or device name, logistics channel name or index name, and control variable name.

[0102] The rule extraction method is a deterministic text analysis based on fixed sentence patterns and keyword patterns. For each document evidence, the original text segment is extracted using page number and paragraph positioning, and sentence segmentation and word segmentation are performed on the original text segment. Synonym normalization rules are called to convert process names, device names, logistics channel names, index names and control variable names in the original text segment into standardized names.

[0103] For each candidate knowledge, a set of constraint entities of the same process or device, the same loop and time close to the occurrence time of the document evidence is retrieved in the knowledge graph according to the action scope identifier, and the corresponding evidence strength of the constraint entity is read. The multiple evidence strengths associated with the candidate knowledge are fused into a candidate knowledge evidence value using a strong evidence priority fusion form, which is expressed as:

[0104] ;

[0105] Among them, represents the evidence value of the th candidate knowledge, represents the th evidence strength associated with the th candidate knowledge, represents the number of evidence strengths associated with the th candidate knowledge, represents the candidate knowledge number, represents the evidence number.

[0106] It should be noted that the constraint entity is written into the knowledge graph as a complete record with the name of the loop, the inlet and outlet stream set, the event start and end time, the batch start and end time, the closed residual, the interval check data and the evidence strength. The record is the constraint entity.

[0107] A credibility state value is maintained for each piece of knowledge content written into the knowledge graph. For a new candidate knowledge extracted, firstly, a search comparison is performed in the knowledge graph according to "the same process or equipment, the same logistics channel or index, the same control variable, the same semantic content", if no existing knowledge is searched, the candidate knowledge is written as new knowledge, and the credibility state value is fixedly assigned as a preset initial value; if the existing knowledge is searched, the credibility state value of the last version of the existing knowledge is read, and the candidate knowledge evidence value of this round is read, and then an updated credibility state value is generated according to a fixed forgetting rule; the fixed forgetting rule is that the updated credibility state value is composed of "the last version credibility state value in a fixed proportion" and "the candidate knowledge evidence value of this round in the remaining proportion", and the fixed proportion is a forgetting factor configured in advance; after the update is completed, the updated credibility state value is written back to the new version record of the candidate knowledge.

[0108] It should be noted that the preset initial value and the forgetting factor are obtained by selecting knowledge entries that have been confirmed correct by human and have not been rolled back in the last 30 days as calibration samples, calculating the level of the credibility state value when it tends to be stable after continuous multiple batches of updates for each sample, and taking the median of the stable level as the preset initial value; the credibility state value change amplitude distribution of these calibration samples from the last batch to the next batch is calculated, and the fixed proportion that makes the median change amplitude of single update fall within a preset target interval (for example, 0.02 to 0.05) is taken as the forgetting factor.

[0109] As Figure 6 The effect of quantifying the credibility of constrained knowledge with computable evidence and improving the reliability and error-correcting ability of the knowledge in the graph is presented; the multiple curves in the graph respectively correspond to the update trajectories of the credibility state values of different candidate knowledge over time, and the credibility state values are iteratively updated according to the fixed forgetting rule, in which the forgetting factor controls the proportion of the contribution of the last version credibility and the contribution of the evidence of this round; the upper graph shows the overall evolution trend of the credibility state values of each candidate knowledge in the complete time range, and the red dashed rectangle and the double-headed arrow mark the selected local comparison interval to highlight the response differences of different candidate knowledge in a short period of time; the lower graph locally magnifies the interval, and the vertical dashed line and the arrow mark the moment when the difference between the two curves is the largest, thereby intuitively displaying the convergence speed and stability difference of different knowledge entries under the evidence driving; it can be seen from the overview and local magnification comparison that when the evidence strength continuously supports a candidate knowledge, its credibility state value rises and tends to be stable; when the evidence is insufficient or inconsistent with the mechanism, the credibility state value growth is limited or falls.

[0110] For each piece of knowledge content determined as a valid new version, revalidation is only performed in the affected range to control the amount of calculation, the affected range being centered on the process or equipment, the material flow channel or the index name involved in the knowledge content, and the process or equipment directly connected to the knowledge content is searched in the same loop and the corresponding inlet and outlet streams to form an affected object set; when revalidation is performed, events in which the process or equipment associated with the event belongs to the affected object set and the event loop is consistent with the loop where the new knowledge is located are retained as a working condition event set, the closed residual and interval check data of each working condition event in the working condition event set are read, and the event value of each working condition event is calculated, and the sum of the closed residual and interval check data of the working condition event is taken as the event value; then the arithmetic mean of the event values of all working condition events in the working condition event set is calculated to obtain the revalidation deviation of the new version of the knowledge content; if the revalidation deviation is greater than the preset deviation threshold, it is determined that the new version of the knowledge content causes the mechanism consistency to deteriorate, the new version of the knowledge content is set as invalid, and the previous version is restored as valid.

[0111] It should be noted that the preset deviation threshold is obtained by statistically collecting a revalidation deviation sample set corresponding to all normal version updates in the last 30 days, and taking the 95th percentile value of the revalidation deviation sample set as the preset deviation threshold.

[0112] The embodiment also provides a computer device suitable for the knowledge graph construction method for gold ore dressing and smelting, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the knowledge graph construction method for gold ore dressing and smelting proposed in the above embodiment.

[0113] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, touchpad or mouse, etc.

[0114] The embodiment also provides a storage medium on which a computer program is stored, the program being executed by a processor to implement the knowledge graph construction method for gold mine beneficiation and smelting as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0115] To sum up, the application realizes stable and unified batch caliber and traceable and recalculation batch instances by triggering batch generation and forming batch traceable identification according to cumulative ore quality; realizes automatic segmentation of working conditions and event-level feature precipitation in batches by generating global boundaries based on variable change intensity, cutting working condition events, and generating event statistical vectors; supports knowledge graphing from batch level to event level; realizes quantifiable evidence constraint on knowledge reliability by event-level metal quantity closure and interval verification based on loop inlet and outlet streams and fusion to generate evidence strength; and improves the reliability and error-correcting property of knowledge graphing.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the application rather than limit the application. Although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the application can be modified or replaced equivalently without departing from the spirit and scope of the application, and all should be covered in the scope of the claims of the application.

Claims

1. A method for constructing a knowledge graph for gold mine beneficiation, characterized in that: The application relates to a batch traceability method and system for a mineral processing system. According to the mineral processing flow topology, a directed connection table is established, and a mapping table is generated; Based on the mapping table, mineral processing data is accessed, a process batch is triggered to be generated according to accumulated ore quality, a batch time window is determined, standard variables are statistically accumulated and aggregated in the batch time window, a batch traceability identifier is formed, and a batch instance set is obtained; In the batch instance set, key standard variable sequences are extracted from process nodes, and change intensity is calculated, a candidate boundary time set is determined according to the change intensity, a global boundary in the batch is generated and merged, a working condition event set with strong binding is obtained, and a statistical vector is generated for each working condition event; Through the working condition event set, an inlet and an outlet flow of a loop are determined, metal amount closed accounting and interval checking are performed, closed residual and interval checking data are calculated through the statistical vector, and evidence strength is generated; Through the evidence strength, the credibility of constraint intervals, intervention actions and cause-effect assumptions is updated, and the knowledge graph is written in accordance with a forgetting rule, and a local subgraph affected is reverified.

2. The gold mine-oriented knowledge graph construction method of claim 1, wherein: The directed connection table is established according to the mineral processing flow topology, and the mapping table is generated, and the specific steps are as follows: Point lists, assay field lists and process files are read, material flow connections in the process files are identified, each connection is defined as a flow, and a starting point and an ending point are recorded to obtain a directed connection table; A fixed object type set and a relationship type set are determined, directions and compulsory fields are specified, corresponding cumulative point positions of each flow are searched, are mapped to standard variable symbols and are written into the mapping table; For each flow, a situation requiring grade is mapped to a standard variable symbol, and control point positions belonging to a node of each device are mapped to a standard variable symbol set and are written into the mapping table.

3. The method for constructing a knowledge graph for gold mine beneficiation according to claim 2, characterized in that: Based on the mapping table, mineral processing data is accessed, a process batch is triggered to be generated according to accumulated ore quality, a batch time window is determined, standard variables are statistically accumulated and aggregated in the batch time window, and the specific steps are as follows: The main ore quality flow is integrated and accumulated, when the accumulated value reaches a batch quality threshold, the reaching time is determined as the batch end time, and a batch time window is obtained; For each batch time window, process point values are converted into standard variable sequences according to the mapping table, time average and key flow cumulative amount in the batch time window are generated, and a batch feature record is obtained; An alignment rule of containing a time window and minimizing the distance of a time window midpoint is adopted, assay samples and batch time windows are uniquely aligned, and a target batch time window is obtained.

4. The gold mine-oriented knowledge graph construction method of claim 3, wherein: The batch traceability identifier is formed, and the batch instance set is obtained, and the specific steps are as follows: If the target batch is uniquely determined, the sampling point name, detection item name, detection result and sample number of the sample are written into the sample set of the batch corresponding to the target batch time window; Evidence positioning information is generated for each document evidence, and the document occurrence time and the batch time window are uniquely aligned to obtain an evidence set; According to the batch feature record, a batch traceability identifier is generated for each batch, the batch traceability identifier, the batch time window, the sample set and the evidence set are jointly combined into a batch instance set.

5. The gold mine-oriented knowledge graph construction method of claim 4, wherein: The method comprises the following steps of: In the batch instance set, key standard variable sequences are extracted for the process nodes and the change strength is calculated, and the candidate boundary time set is determined according to the change strength, and the specific steps are as follows: In the directed connection table, the process nodes in each batch in the batch instance set that need to generate the working condition event are fixed to form a node list; The key standard variable sequences are extracted through each process node in the node list and normalized processing is performed, and the change strength is calculated; 6. The gold mine-oriented knowledge graph construction method of claim 5, wherein: When the change strength is continuously not lower than the change threshold and the duration is not less than the sliding window length, the first time that meets the condition is recorded as the candidate boundary time of the node. The batch global boundary is generated and merged, the batch strongly bound working condition event set is obtained, and the statistical vector is generated for each working condition event, and the specific steps are as follows: The candidate boundary times generated by all the process nodes in the same batch are merged into a global boundary set, and time tolerance is removed, and the working condition event set is obtained by being cut according to the batch start time and the batch end time; The batch start time and the batch end time are added to the head and tail of the global boundary list after the removal, and the time period between adjacent boundaries is sequentially taken as the working condition event time window; 7. The gold mine-oriented knowledge graph construction method of claim 6, wherein: The statistical vector is calculated for the key standard variable sequence in each working condition event time window. The loop inlet and outlet streams are determined through the working condition event set, the metal amount closed accounting and interval verification are performed, the closed residual and interval verification data are calculated through the statistical vector, and the evidence strength is generated, and the specific steps are as follows: The loop inlet and outlet stream sets are determined for each working condition event in the working condition event set, the cumulative mass is extracted according to the event time window, and the grade representative value is generated for each stream; The metal amount closed accounting is performed through the cumulative mass and the grade representative value, and the closed residual is calculated; The interval verification is performed on each event on the device node, the interval violation degree is calculated based on the statistical vector and is summarized as the interval verification data; 8. The gold mine-oriented knowledge graph construction method of claim 7, wherein: The closed residual and the interval verification data are fused into the evidence strength. The credibility of the constraint interval, the intervention action and the causal hypothesis is updated through the evidence strength, and the knowledge graph is written according to the fixed forgetting rule, and the affected local subgraph is reverified, and the specific steps are as follows: The constraint interval, the intervention action and the causal hypothesis are generated into a candidate knowledge set by using the rule extraction method; The evidence strength of the candidate knowledge is fused into a candidate knowledge evidence value in a strong evidence first fusion form; The credibility state value is maintained for each piece of knowledge content written into the knowledge graph, the updated credibility state value is generated through the candidate knowledge evidence value according to the forgetting rule and is written into the knowledge graph, and the affected local subgraph is reverified; 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The forgetting rule means that the updated credibility state value is composed of a fixed proportion of the previous version of the credibility state value and a remaining proportion of the current candidate knowledge evidence value.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The processor executes the computer program to realize the steps of the knowledge graph construction method for gold ore dressing in claim 1-8. The computer program is executed by the processor to realize the steps of the knowledge graph construction method for gold ore dressing in claim 1-8.

Images