Supply chain data quality automatic verification and closed-loop treatment method and system

Abstract

The invention discloses a supply chain data quality automatic verification and closed-loop management method and system, and relates to the technical field of supply chain digitization, and the method comprises the following steps: inputting an obtained supply chain authority source list into a preset verification rule set, and extracting a problem field set according to a rule verification result; injecting the problem field set into a DCAN responsibility mapping mechanism, outputting a capability matrix through a pre-constructed multi-modal responsibility agent, and calculating a responsibility proportion weight of a responsibility subject based on the capability matrix; inputting the responsibility proportion weight into a responsibility allocation decision model, and outputting the allocation responsibility of the corresponding responsibility subject; distributing a problem rectification task to a corresponding responsibility subject according to the distribution responsibility; and if the rectification failure is detected, extracting a rectification failure field set, transmitting the rectification failure field set to the secondary responsibility subject according to a preset responsibility transmission path, and executing circulation until the problem is solved. The method is used for solving the problems that data responsibility affiliation is not clear, tracking is lacked after rectification tasks are distributed, and the rectification process is broken.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supply chain digitization, and more particularly, to a supply chain data quality automatic checking and closed-loop management method and system. BACKGROUND

[0002] With the deepening of enterprise digital transformation, supply chain management systems are increasingly complex, covering multiple business systems. Modern supply chain management relies on the collaborative work of multiple heterogeneous information systems, generating massive structured and semi-structured data. These data run through procurement, manufacturing, warehousing, transportation, and after-sales, and are the basis for supporting inventory management, order fulfillment, cost control, and risk monitoring. To ensure the correctness of business decisions and automated processes, the integrity, accuracy, consistency, and timeliness of supply chain data quality attributes must be checked and managed. Current supply chain data quality management methods include: improving data quality through conversion, mapping, standardization, etc. at the data access stage; recording metadata, blood relationship, and responsible subjects to provide viewing and auditing functions; using historical distribution, time series models, or supervised / unsupervised algorithms to detect abnormal records or events; manually analyzing anomalies, locating responsibilities, and promoting repair, often relying on cross-department coordination.

[0003] Currently, the management of supply chain data quality has the following problems: mostly relying on manual sampling or static checking methods, data in various business systems is scattered and lacks unified standards and authoritative data sources, and data responsibility is unclear, lacking self-adaptation and intelligent capabilities, resulting in low supply chain data quality, restricting the upgrading of supply chain digitization and the improvement of operational efficiency.

[0004] To solve the above problems, the present application provides a solution. SUMMARY

[0005] To overcome the above-mentioned defects of the prior art, embodiments of the present application provide a supply chain data quality automatic checking and closed-loop management method and system, which builds an authoritative source list, uses a DCAN responsibility mapping mechanism and a responsibility allocation decision model, and combines a responsibility transmission path to solve the problems of unclear data responsibility in various business systems, lack of tracking after rectification task assignment, and broken rectification process.

[0006] To achieve the above-mentioned purposes, the present application provides the following technical solutions: The application discloses a supply chain data quality automatic checking and closed-loop management method, which comprises the following steps: inputting an obtained supply chain authority source list into a preset checking rule set, extracting a problem field set according to a rule checking result; injecting the problem field set into a DCAN responsibility mapping mechanism, outputting an ability matrix through a pre-constructed multi-modal responsibility intelligent agent, and calculating a responsibility proportion weight of a responsibility subject based on the ability matrix; inputting the responsibility proportion weight into a responsibility allocation decision model to output an allocated responsibility of the corresponding responsibility subject; according to the allocated responsibility, assigning a problem rectification task to the corresponding responsibility subject; if it is detected that the rectification fails, extracting a rectification failure field set, and conducting the rectification failure field set to a secondary responsibility subject according to a preset responsibility conduction path, and performing a cycle until the problem is solved.

[0007] In a preferred embodiment, the step of inputting the obtained supply chain authority source list into the preset checking rule set and extracting the problem field set according to the rule checking result specifically comprises the following steps: fusing obtained multi-source supply chain system data, calculating a field similarity of the fused system data, generating a standard dictionary based on a similarity clustering result; calculating a data source authority score according to the standard dictionary, and screening to obtain an authority source list; inputting the authority source list into the preset rule checking set for checking to generate a data quality score; and when the data quality score is lower than a preset authority threshold, positioning a problem field to construct a problem field set.

[0008] In a preferred embodiment, the step of injecting the problem field set into the DCAN responsibility mapping mechanism, outputting the ability matrix through the pre-constructed multi-modal responsibility intelligent agent, and calculating the responsibility proportion weight of the responsibility subject based on the ability matrix specifically comprises the following steps: inputting the problem field set into the pre-constructed multi-modal responsibility intelligent agent; outputting an ability matrix of the responsibility subject based on the multi-modal responsibility intelligent agent, wherein the ability matrix comprises processing ability parameters of each subject on the field type; calculating a subject efficiency value of the responsibility subject in processing the problem field according to a preset subject efficiency formula and combining the parameters in the ability matrix, and calculating a marginal contribution value based on the efficiency value; calculating a Shapley value based on the marginal contribution value, and normalizing the Shapley value of each subject according to the field to obtain the responsibility proportion weight.

[0009] In a preferred embodiment, the construction of the multi-modal responsibility intelligent agent specifically comprises the following steps: performing feature extraction on cross-system metadata to obtain a field vector and a subject vector; constructing an initial ability matrix according to a matching degree of the field vector and the subject vector; and constructing an intelligent agent template according to the field vector and the ability matrix, combining an identity core, a reputation value and a historical task rectification qualification rate.

[0010] In a preferred embodiment, the responsibility proportion weight is input into a responsibility allocation decision model, and an allocated responsibility corresponding to a responsibility subject is output, specifically: the responsibility proportion weight, the subject technical capability strength, the penalty amount, and the repair workload are input into a responsibility allocation decision model, the subject technical capability strength is calculated according to a capability matrix; based on the subject technical capability strength and the responsibility proportion weight, a normalized allocation of the repair workload is calculated to obtain a repair responsibility; and based on the responsibility proportion weight, a weighted allocation of the penalty amount is performed to obtain a penalty responsibility.

[0011] In a preferred embodiment, if the rectification fails, a rectification failure field set is extracted, and the rectification failure field set is transmitted to a secondary responsibility subject according to a preset responsibility conduction path, and a loop is executed until the problem is solved, specifically: a rectification qualification rate is calculated according to a task assignment result and rectification feedback data; when the rectification qualification rate is lower than a preset threshold, a rectification failure field set is extracted from the rectification feedback data; a cross-domain responsibility conduction weight of the rectification failure field is calculated to obtain a responsibility conduction path; based on the responsibility conduction path, the rectification failure field is input into a DCAN responsibility mapping mechanism to allocate a new responsibility subject until the rectification failure field after re-rectification is rechecked by a verification rule set and a data quality score is not lower than an authoritative threshold.

[0012] In a preferred embodiment, the multi-modal responsibility intelligent agent further comprises an intelligent agent evolution mechanism, specifically: a responsibility subject reputation increment is calculated by a reputation increment calculation model based on a rectification qualification rate, and a reputation value is dynamically updated in combination with a decay index; a predicted loss function is calculated according to a rectification qualification rate and a predicted rectification qualification rate, and a capability matrix parameter is adjusted by a gradient descent method, the predicted rectification qualification rate is calculated from the reputation value; according to a constructed cross-domain capability migration formula, a capability matrix of a similar domain is migrated to a new domain.

[0013] A system of a supply chain data quality automatic verification and closed-loop governance method, comprising: a problem extraction module, configured to input an obtained supply chain authoritative source list into a preset verification rule set, and extract a problem field set according to a rule verification result; A responsibility mapping module, configured to inject the problem field set into a DCAN responsibility mapping mechanism, and output a capability matrix through a pre-constructed multi-modal responsibility intelligent agent, and calculate a responsibility proportion weight of a responsibility subject based on the capability matrix; A responsibility allocation module, configured to input the responsibility proportion weight into a responsibility allocation decision model, and output an allocated responsibility corresponding to the responsibility subject; A task assignment module, configured to assign a problem rectification task to a corresponding responsibility subject according to the allocated responsibility; A closed-loop management module is configured to extract a rectification failure field set if rectification failure is detected, and to conduct the rectification failure field set to a secondary responsibility subject according to a preset responsibility conduction path, and to perform a loop until the problem is solved.

[0014] The technical effects and advantages of the supply chain data quality automatic verification and closed-loop management method and system of the present application are as follows: The present application fuses multi-source data to construct reliable authoritative source data, uses a DCAN responsibility mapping mechanism and a responsibility allocation decision model to allocate data responsibility to relevant responsibility subjects, and combines a responsibility conduction path to ensure that, in the event of rectification failure, responsibility can be effectively transferred to the relevant field according to the conduction path, and responsibility allocation is re-performed, thereby avoiding blind spots in data governance. The present application effectively solves the problems of unclear data governance responsibility division and data quality problem rectification lag in the prior art, and can effectively improve the digitalization upgrade and operation efficiency of the supply chain. Through an automatic quality scoring and problem reporting mechanism, combined with dynamic reputation adjustment and intelligent task allocation, the efficiency and accuracy of data governance are greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A supply chain data quality automatic verification and closed-loop management method flowchart is provided for the embodiments of the present application.

[0016] Figure 2 A system structure diagram of a supply chain data quality automatic verification and closed-loop management method is provided for the embodiments of the present application.

[0017] Figure 3 A logic diagram of a supply chain data quality automatic verification and closed-loop management method is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0019] Embodiment 1, Figure 1 A supply chain data quality automatic verification and closed-loop management method of the present application is provided, which comprises the following steps: S1, input the obtained supply chain authoritative source list into a preset verification rule set, and extract a problem field set according to the rule verification result; S2, inject the problem field set into the DCAN responsibility mapping mechanism, output the capability matrix of the pre-constructed multi-modal responsibility agent, and calculate the responsibility proportion weight of the responsibility subject based on the capability matrix; S3, input the responsibility proportion weight into the responsibility allocation decision model, and output the allocated responsibility of the corresponding responsibility subject; S4, according to the allocated responsibility, assign the problem rectification task to the corresponding responsibility subject; S5, if the rectification fails, extract the rectification failure field set, and conduct the rectification failure field set to the secondary responsibility subject according to the preset responsibility conduction path, and execute the cycle until the problem is solved.

[0020] In this embodiment, by fusing multi-source data, reliable authoritative source data is constructed, and the data responsibility is allocated to the relevant responsibility subject by using the DCAN responsibility mapping mechanism and the responsibility allocation decision model, and combined with the responsibility conduction path, it is guaranteed that when the rectification fails, the responsibility can be effectively transferred to the related field according to the conduction path, and the responsibility is allocated again, avoiding the blind spot in data governance. The problems of unclear data governance responsibility division and data quality problem rectification lag in the prior art are effectively solved, which can effectively improve the digitalization upgrade and operation efficiency of the supply chain. Through the automatic quality scoring and problem reporting mechanism, combined with dynamic reputation adjustment and task intelligent allocation, the efficiency and accuracy of data governance are greatly improved.

[0021] S1, input the obtained supply chain authoritative source list into the preset verification rule set, and extract the problem field set according to the rule verification result.

[0022] In this embodiment, the obtained supply chain authoritative source list is input into the preset verification rule set, and the problem field set is extracted according to the rule verification result, which is specifically: Fuse the obtained multi-source supply chain system data, calculate the field similarity of the fused system data, and generate a standard dictionary based on the similarity clustering result; According to the standard dictionary, calculate the data source authority score, and filter to obtain the authoritative source list; Input the authoritative source list into the preset rule verification set for verification, and generate a data quality score. When the data quality score is lower than the preset authority threshold, locate the problem field to construct the problem field set.

[0023] In this embodiment, the multi-source supply chain system data is fused, the field similarity of the fused data is calculated, and the standard dictionary is generated by merging the high similarity fields, which is specifically: Connect multiple supply chain business systems through an API gateway, extract data from each system, fuse the field information from different data sources, and calculate the similarity of the fields in different data sources. When the similarity is greater than a preset similarity threshold When, the fields are clustered into unified standard fields, referred to as a standard dictionary .

[0024] The data fusion formula is specifically as follows:

[0025] In the formula, is the fused multi-source supply chain system data; is a field alignment operator; is a field set in the i-th data source; is the data content in the i-th data source. The field similarity formula is specifically as follows:

[0026]

[0027] In the formula, is the similarity of fields and ; , are the value ranges of fields and ; , are the data types of fields and , is the field type matching degree.

[0028] In this embodiment, the data source authority score is calculated according to the standard dictionary, and data sources exceeding a high-score data source threshold are screened to form an authority source list.

[0029] The data source authority score formula is specifically as follows:

[0030] In the formula, is the data source, is the data source authority score, is the data completeness score, is the field consistency score, is the data timeliness score, , , are weight coefficients, and .

[0031] When the data source authority score is , the data source is retained in the authority source list, wherein is a high-score data source threshold.

[0032] ​​In this embodiment, the authoritative source list is input into the preset rule verification set for verification to generate a data quality score, specifically: The rule verification formula is specifically as follows:

[0033] In the formula, is the authoritative source list data is the verification result of the rule .

[0034] The data quality score formula is specifically as follows:

[0035] In the formula, is the data quality score, is the total number of data entries in the authoritative source list, is the total number of rules in the preset rule verification set, is the verification result of the authoritative source list data.

[0036] In this embodiment, when the data quality score is lower than a preset authoritative threshold, a problem field set is located, specifically: The preset authoritative threshold is If , the system triggers an alarm to prompt attention and processing; The rule verification violation problem field is located to generate a problem field set.

[0037] It should be noted that the problem field refers to a field that is determined to have data abnormalities, errors or non-compliance with standard specifications.

[0038] S2, the problem field set is injected into the DCAN responsibility mapping mechanism, and the ability matrix is output by the pre-constructed multi-modal responsibility agent, and the responsibility proportion weight of the responsibility subject is calculated based on the ability matrix.

[0039] In this embodiment, the problem field set is injected into the DCAN responsibility mapping mechanism, and the ability matrix is output by the pre-constructed multi-modal responsibility agent, and the responsibility proportion weight of the responsibility subject is calculated based on the ability matrix, specifically: The problem field set is input into the pre-constructed multi-modal responsibility agent; The ability matrix of the responsibility subject is output based on the multi-modal responsibility agent, and the ability matrix includes the processing ability parameters of each subject to the field type; According to the preset subject efficiency formula, the subject efficiency value of the responsibility subject processing the problem field is calculated in combination with the parameters in the ability matrix, and the marginal contribution value is calculated based on the efficiency value; The Shapley value is calculated based on the marginal contribution value, and the Shapley value of each subject is normalized according to the field to obtain a responsibility proportion weight.

[0040] In the embodiment, the multi-modal responsibility intelligent agent is constructed, in particular: Feature extraction is performed on the cross-system metadata to obtain a field vector and a subject vector. An initial capability matrix is constructed according to the matching degree of the field vector and the subject vector. According to the field vector and the capability matrix, an intelligent agent template is constructed in combination with an identity core, a reputation value and a historical task rectification qualification rate.

[0041] In the embodiment, metadata in the HR system, the permission system and the operation log is extracted, redundant or low-correlation features are removed from the metadata, metadata features are extracted through the operation type in the log, the skill label in the HR and the role name in the permission system, the field vector and the subject vector are obtained, and the capability matrix is constructed according to the matching degree of the field vector and the subject vector, as follows: For each field , the number of occurrences of each label in the field is counted to obtain a field vector.

[0042] The field vector is specifically as follows: , wherein is the number of occurrences of each label in the field.

[0043] For each subject , the number of occurrences of the associated labels is counted to obtain a subject vector.

[0044] The subject vector is specifically as follows: , wherein is the number of occurrences of the labels associated with the subject.

[0045] It should be noted that the field vector reflects the importance of the operation in the field by counting the association number of the field and the operation label representing the field; The subject vector reflects the use frequency of the responsibility subject on the operation by counting the association number of the responsibility subject and the operation label to determine the association frequency of the subject and the operation.

[0046] The formula for constructing the capability matrix is specifically as follows: , wherein, is the capability matrix, is the matching degree of the field vector and the subject vector.

[0047] In the embodiment, the agent template is constructed by identity core, reputation value, field vector, capability matrix and historical task rectification qualified rate, wherein the identity core is obtained through the CA digital certificate authentication system.

[0048] The agent template is specifically as follows:

[0049] wherein, is an agent template, is an identity core, is a reputation value, is a field vector, is a capability matrix, is a historical task rectification qualified rate.

[0050] In the embodiment, the subject performance value of the responsibility subject in the problem field is calculated according to a preset subject performance formula combined with the parameters in the capability matrix, and the marginal contribution value is calculated based on the performance value; The Shapley value is calculated based on the marginal contribution value, and the Shapley value of each subject is normalized according to the field to obtain the responsibility proportion weight, which is specifically: The problem field set is input into the multi-modal responsibility agent, and the responsibility subject set is ; For each problem field , the task type associated with each problem field is identified, for each subject, the capability vector of the subject is extracted, and is the capability matrix; The subject performance value of the responsibility subject in the problem field is calculated according to a preset subject performance formula; The marginal contribution of the responsibility subject subset to the problem field is calculated; According to the marginal contribution value, the Shapley value is calculated for each responsibility subject and the problem field ; According to the Shapley value, the responsibility proportion weight is generated.

[0051] The subject performance formula is specifically as follows:

[0052] wherein, is the problem field of the responsibility subject subset The main performance value, A subset of the responsible entities, For the question field The importance weight of the business, As the responsible party, It is the L2 norm. This is the capability vector.

[0053] It should be noted that the subject effectiveness value refers to the overall performance of the responsible subject in handling the problem field.

[0054] The specific formula for marginal contribution is as follows:

[0055] In the formula, This is the marginal contribution value.

[0056] The specific formula for calculating the Shapley value is as follows:

[0057] In the formula, Shapley value, The total number of responsible entities, A subset of the responsible entities, To remove the responsible party The complete list of responsible parties.

[0058] The specific formula for calculating the responsibility ratio weighting is as follows: , in

[0059] In the formula, as the main body For the question field Responsibility proportion weighting Shapley value, This is the sum of the Shapley values ​​for all subjects.

[0060] S3 inputs the responsibility proportion weights into the responsibility allocation decision model and outputs the allocation responsibility of the corresponding responsible subject.

[0061] In this embodiment, the step of inputting the responsibility proportion weights into the responsibility allocation decision model and outputting the allocated responsibility of the corresponding responsible subject specifically involves: The responsibility allocation decision model is input with the responsibility ratio weight, the main technical capability strength, the penalty amount and the repair workload. The main technical capability strength is calculated based on the capability matrix. Based on the strength of the main technical capabilities and the weight of the responsibility ratio, the repair responsibility is obtained by calculating the normalized allocation of the repair workload; Based on the weighted proportion of responsibility, the penalty amount is weighted and allocated to determine the penalty responsibility.

[0062] It should be noted that the strength of the main technical capabilities is calculated based on the capability matrix, the actual repair workload, and the dynamic complexity factor; the penalty amount is obtained through the contract management system and the financial settlement system; and the repair workload is obtained through the data quality management system and the work order management system.

[0063] In this embodiment, a responsibility ratio weight matrix is ​​constructed based on the calculated responsibility ratio weights. Obtain the task repair workload vector Vector of penalty amount and the matrix of technical capabilities of responsible entities The responsibilities for restoration and punishment are quantified.

[0064] It should be noted that only responsible entities with the capability to perform repair tasks can allocate repair responsibilities. The capability to perform repair tasks is judged based on the strength of the responsible entity's technical ability to repair the problem field. When the strength of the technical ability is higher than the technical feasibility threshold, repair responsibility is allocated.

[0065] The formula for normalizing repair responsibility is as follows:

[0066] in,

[0067] In the formula, the constraints are satisfied. , To quantify the responsibility for repair, As a standard repair workload, As a weighted proportion of responsibility, As the responsible party Fix the problematic fields Technical capabilities As a technical feasibility threshold, Responsible Entity Fix the problematic fields The capability vector, This is a capability matrix. This is a technical strength adjustment factor.

[0068] The weighted formula for calculating penalties is as follows:

[0069] In the formula, To quantify punitive liability and meet constraints , The standard penalty amount, This refers to the weighting of responsibility proportions.

[0070] S4. Based on the assigned responsibilities, assign problem rectification tasks to the corresponding responsible entities.

[0071] In this embodiment, the step of assigning problem rectification tasks to the corresponding responsible parties based on the allocated responsibilities specifically includes: The intelligent dispatch formula generates subject-level task instructions, and the task instructions are used to dispatch problem rectification tasks to the responsible subjects. The intelligent distribution formula is as follows:

[0072] In the formula, As the main task, As the responsible party, This represents the number of fields in the question.

[0073] S5. If rectification failure is detected, the rectification failure field set is extracted and transmitted to the secondary responsible entity according to the preset responsibility transmission path. The loop continues until the problem is resolved.

[0074] In this embodiment, if rectification failure is detected, the rectification failure field set is extracted and transmitted to the secondary responsible entity according to the preset responsibility transmission path, and the process is repeated until the problem is resolved. Specifically: Calculate the rectification pass rate based on the task assignment results and rectification feedback data; When the rectification pass rate is lower than the preset threshold, extract the rectification failure field set from the rectification feedback data; Calculate the cross-domain responsibility transmission weight for the rectification failure field to obtain the responsibility transmission path; Based on the responsibility transmission path, the rectification failure field is re-entered into the DCAN responsibility mapping mechanism to assign a new responsible subject until the rectification failure field after re-rectification is reviewed by the verification rule set and the data quality score is not lower than the authoritative threshold.

[0075] It should be noted that the purpose of calculating the cross-domain responsibility transfer weight is to re-enter the rectification failure field into other domains similar to the original domain in the DCAN responsibility mapping mechanism to re-map responsibility when the problem rectification fails.

[0076] In this embodiment, the formula for the rectification pass rate is as follows:

[0077] In the formula, The current rectification compliance rate of the responsible party. To assign to the subject The problem field subset This is the result of field verification based on the set of validation rules.

[0078] In this embodiment, the preset rectification qualified rate threshold is 90%. When the rectification qualified rate is less than 90%, the rectification failure field set is extracted from the rectification feedback data. When the rectification qualified rate is less than 90%, the rectification failure field set is extracted from the rectification feedback data. And trigger the cross-domain responsibility conduction process, conduct responsibility for the rectification failure problem field, re-input the conducted rectification failure field into the DCAN responsibility mapping mechanism and responsibility allocation decision model, allocate a new responsibility subject, until the rectification data passes the review of the set of verification rules and the quality score is not less than the authority threshold.

[0079] The cross-domain responsibility conduction formula is as follows:

[0080]

[0081] In the formula, is the weight of the rectification failure field conducted to the domain , is the field original domain vector, is the domain vector of the target domain, is the candidate domain vector in iteration.

[0082] The conduction path formula is as follows:

[0083] In the formula, is the optimal conduction path of the rectification failure field , is the weight of the rectification failure field conducted to the domain .

[0084] In this embodiment, the multi-modal responsibility intelligent agent further includes an intelligent agent evolution mechanism, specifically: Based on the rectification qualified rate, the responsibility subject reputation increment is calculated through a reputation increment calculation model, and the reputation value is dynamically updated combined with the decay index; According to the rectification qualified rate and the predicted rectification qualified rate, a prediction loss function is calculated, and the ability matrix parameters are adjusted by gradient descent method, and the predicted rectification qualified rate is calculated from the reputation value; According to the constructed cross-domain ability migration formula, the ability matrix of the similar domain is migrated to the new domain added.

[0085] In this embodiment, the reputation increment calculation model is as follows:

[0086] In the formula, To increase reputation; For the number of fields in the question, The current rectification compliance rate of the responsible parties is the actual problem. The percentage of issues rectified by the responsible party in the previous instance; For capability marker matrix elements, As the responsible party Repair fields Technical capabilities This is the threshold for technical feasibility. As a weight of responsibility, For adaptive reputation learning rate and , , This represents the current credit score of the responsible entity.

[0087] Furthermore, the specific formula for calculating the reputation score update is as follows:

[0088] Attenuation coefficient dynamically adjusted:

[0089] In the formula, For the updated reputation value, , To increase reputation, The reputation value before the main update. The time decay coefficient, For time step.

[0090] In this embodiment, the specific formula for calculating the prediction loss function is as follows:

[0091] The logit function:

[0092] The formula for calculating the predicted rectification pass rate is:

[0093] In the formula, For the prediction loss function, For the number of fields in the question, The predicted rectification pass rate for this field. The actual rectification pass rate for this field. For fields The predicted rectification pass rate This represents the current credit score of the responsible entity.

[0094] Furthermore, the specific formula for the gradient of the computational capability matrix is ​​as follows:

[0095] in, , This is achieved through automatic differentiation.

[0096] In the formula, Let be the gradient matrix of the loss function. For the elements of the capability matrix, The coordinates of the matrix elements For partial differential operators, As the responsible party Repair fields Technical capabilities This is the technical feasibility threshold.

[0097] Furthermore, the formula for updating the capability matrix is ​​as follows:

[0098] In the formula, For the updated capability matrix, This is the current capability matrix. Let be the learning rate, and the constraint be... , From the perspective of capabilities, For the number of fields in the question, The gradient matrix of the loss function .

[0099] In this embodiment, the updated capability matrix Feedback is sent to the multimodal responsible agent in the responsibility mapping mechanism; the updated agent reputation value is then... The multimodal responsible agent is fed back to the responsibility mapping mechanism for prediction in the next cycle.

[0100] In this embodiment, the capability matrix of similar domains is migrated to the newly added domain according to the constructed cross-domain capability transfer formula.

[0101] The specific formula for cross-domain capability transfer is as follows:

[0102] The formula for the transfer factor is:

[0103] In the formula, To update the capability score of the target domain after migration, In the source domain capability matrix, the first... The ability in the first The capability score of each source domain For migration strength scalar, For the first The ability to the first an adaptable factor of the source domain, is a source domain is a similarity of the target domain , is a transfer factor, is a confidence of the first source domain, is a time difference of the last time when the source domain capability is updated to current, is a time decay coefficient, is a capability vector of the source domain , is a feature subspace of the target domain , is a linear projection operator for projecting a vector to a subspace , is a non-zero constant.

[0104] It should be noted that the cross-domain capability transfer formula is obtained by calculating the similarity of the source domain capability matrix and the target domain capability matrix to obtain the original capability score of the target domain. On the basis of the original capability score, the transfer increment of the source domain capability to the target domain capability is calculated by the similarity of the source domain capability matrix and the target domain capability matrix, the source domain confidence, the source domain adaptable factor, the transfer degree scalar and the transfer factor, and the transfer increment is added to the original capability score, thereby realizing effective transfer of cross-domain capability.

[0105] Embodiment 2, Figure 2 The system of the supply chain data quality automatic checking and closed-loop management method of the application comprises: A problem extraction module is configured to input the obtained supply chain authoritative source list into a preset checking rule set, and extract a problem field set according to a rule checking result. A responsibility mapping module is configured to inject the problem field set into a DCAN responsibility mapping mechanism, output an ability matrix through a pre-constructed multi-modal responsibility intelligent agent, and calculate a responsibility proportion weight of a responsibility subject based on the ability matrix. A responsibility allocation module is configured to input the responsibility proportion weight into a responsibility allocation decision model, and output an allocated responsibility of the corresponding responsibility subject. A task dispatching module is configured to dispatch a problem rectification task to the corresponding responsibility subject according to the allocated responsibility. A closed-loop management module is configured to extract a rectification failure field set if the rectification fails, and conduct the rectification failure field set to a secondary responsibility subject according to a preset responsibility conduction path, and execute a loop until the problem is solved.

[0106] The above formulas are all dimensionless values, and the formulas are obtained by collecting a large amount of data to simulate the latest real situation, and the preset parameters in the formulas are set by the person skilled in the art according to the actual situation.

[0107] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product.

[0108] Those of ordinary skill in the art can realize that the modules and algorithm steps of the examples described in connection with the embodiments disclosed herein can be realized by electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are performed by hardware or software depends on the specific application and design constraints of the technical solution. Those of ordinary skill in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0109] In addition, each functional module in each embodiment of the present application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0110] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0111] Finally, the above is merely preferred embodiments of the present application, and is not intended to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for automatic verification and closed-loop management of supply chain data quality, characterized in that, Includes the following steps: Input the obtained authoritative supply chain source list into the preset verification rule set, and extract the problem field set based on the rule verification results; The problem field set is injected into the DCAN responsibility mapping mechanism, and the capability matrix is ​​output through the pre-built multimodal responsibility agent. The responsibility proportion weight of the responsible subject is calculated based on the capability matrix. Input the responsibility proportion weights into the responsibility allocation decision model, and output the allocated responsibility of the corresponding responsible subject; Based on the assigned responsibilities, problem rectification tasks are assigned to the corresponding responsible entities; If rectification failure is detected, the rectification failure field set is extracted and passed to the secondary responsible entity according to the preset responsibility transmission path, and the cycle is executed until the problem is resolved.

2. The automatic verification and closed-loop management method for supply chain data quality according to claim 1, characterized in that, The process involves inputting the obtained authoritative supply chain source list into a preset verification rule set, and extracting a set of problematic fields based on the rule verification results. Specifically: The system integrates multi-source supply chain system data, calculates the field similarity of the integrated system data, and generates a standard dictionary based on the similarity clustering results. The authority score of the data source is calculated based on the standard dictionary, and a list of authoritative sources is obtained by filtering. The authoritative source list is input into the preset rule verification set for verification, and a data quality score is generated. When the data quality score is lower than the preset authoritative threshold, the problem fields are located and a problem field set is constructed.

3. The automatic verification and closed-loop management method for supply chain data quality according to claim 2, characterized in that, The process of injecting the problem field set into the DCAN responsibility mapping mechanism, outputting a capability matrix through a pre-constructed multimodal responsibility agent, and calculating the responsibility proportion weight of the responsible subject based on the capability matrix is ​​as follows: Input the problem field set into the pre-built multimodal responsible agent; The multimodal responsible intelligent agent outputs a capability matrix of the responsible subjects, which includes the processing capability parameters of each subject for field types; Based on the preset subject effectiveness formula, the subject effectiveness value of the responsible subject's problem-handling field is calculated by combining the parameters in the capability matrix, and the marginal contribution value is calculated based on the effectiveness value. The Shapley value is calculated based on the marginal contribution value, and the Shapley values ​​of each entity are normalized by field to obtain the responsibility proportion weight.

4. The automatic verification and closed-loop management method for supply chain data quality according to claim 3, characterized in that, The construction of the multimodal responsible intelligent agent is specifically as follows: Feature extraction is performed on cross-system metadata to obtain domain vectors and subject vectors; Construct an initial capability matrix based on the matching degree between the domain vector and the subject vector; Based on the domain vector and capability matrix, an intelligent agent template is constructed by combining the identity core, reputation value, and historical task rectification pass rate.

5. The automatic verification and closed-loop management method for supply chain data quality according to claim 4, characterized in that, The process of inputting responsibility proportion weights into the responsibility allocation decision model and outputting the allocated responsibility of the corresponding responsible entity is as follows: The responsibility allocation decision model is input with the responsibility ratio weight, the main technical capability strength, the penalty amount and the repair workload. The main technical capability strength is calculated based on the capability matrix. Based on the strength of the main technical capabilities and the weight of the responsibility ratio, the repair responsibility is obtained by calculating the normalized allocation of the repair workload; Based on the weighted proportion of responsibility, the penalty amount is weighted and allocated to determine the penalty responsibility.

6. The automatic verification and closed-loop management method for supply chain data quality according to claim 5, characterized in that, The process of assigning rectification tasks to the relevant responsible parties based on the allocation of responsibilities is as follows: The intelligent dispatch formula generates subject-level task instructions, and the task instructions are used to dispatch problem rectification tasks to the responsible subjects. The intelligent distribution formula is as follows: In the formula, As the main task, As the responsible party, This represents the number of fields in the question.

7. The automatic verification and closed-loop management method for supply chain data quality according to claim 6, characterized in that, If rectification failure is detected, the rectification failure field set is extracted and passed to the secondary responsible entity according to the preset responsibility transmission path. This process is repeated until the problem is resolved. Specifically: Calculate the rectification pass rate based on the task assignment results and rectification feedback data; When the rectification pass rate is lower than the preset threshold, extract the rectification failure field set from the rectification feedback data; Calculate the cross-domain responsibility transmission weight for the rectification failure field to obtain the responsibility transmission path; Based on the responsibility transmission path, the rectification failure field is re-entered into the DCAN responsibility mapping mechanism to assign a new responsible subject until the rectification failure field after re-rectification is reviewed by the verification rule set and the data quality score is not lower than the authoritative threshold.

8. The automatic verification and closed-loop management method for supply chain data quality according to claim 7, characterized in that, The multimodal responsible intelligent agent also includes an agent evolution mechanism, specifically: Based on the rectification pass rate, the credit increment of the responsible entity is calculated through the credit increment calculation model, and the credit value is dynamically updated in combination with the decay index. The prediction loss function is calculated based on the rectification pass rate and the predicted rectification pass rate. The capability matrix parameters are adjusted by the gradient descent method. The predicted rectification pass rate is calculated from the reputation value. Based on the constructed cross-domain capability transfer formula, the capability matrix of similar domains is transferred to the newly added domain.

9. A system using the automatic verification and closed-loop management method for supply chain data quality as described in any one of claims 1-8, comprising: The problem extraction module is used to input the obtained authoritative supply chain source list into a preset set of verification rules, and extract a set of problem fields based on the rule verification results; The responsibility mapping module is used to inject the problem field set into the DCAN responsibility mapping mechanism, output the capability matrix through the pre-built multimodal responsibility agent, and calculate the responsibility proportion weight of the responsible subject based on the capability matrix; The responsibility allocation module is used to input the responsibility proportion weights into the responsibility allocation decision model and output the allocated responsibilities of the corresponding responsible entities. The task assignment module is used to assign problem rectification tasks to the corresponding responsible parties based on the assigned responsibilities. The closed-loop governance module is used to extract the set of rectification failure fields if rectification failure is detected, and to pass the set of rectification failure fields to the secondary responsible entities according to the preset responsibility transmission path, and execute the loop until the problem is resolved.

Images