Enterprise digital operation management method and system

Abstract

This invention relates to the field of enterprise digital management technology, and particularly to an enterprise digital operation management method and system. The method includes: acquiring business activity data, extracting the interaction strength between business activities, and clustering multiple business activities into multiple intelligent agents; extracting technical metadata from the target system's data dictionary and interface documents, constructing a metadata knowledge graph using the technical metadata, and providing data interfaces for the intelligent agents using the data knowledge graph; based on the data interfaces, each intelligent agent uses a probability prediction model to generate a probability distribution of future external variables, and makes individual decisions based on the probability distribution; determining whether there is shared resource coupling in the individual decisions of two intelligent agents, and if so, introducing a collaborative mechanism to correct their independent decisions through shadow prices so that the individual decisions reach global optimality.

Description

Technical Field

[0001] This invention relates to the field of enterprise digital management technology, and in particular to an enterprise digital operation management method and system. Background Technology

[0002] As digital transformation deepens, enterprises face numerous challenges in digital operations management. Specifically: First, inconsistent data standards across business systems create severe data silos, hindering the effective utilization of global data. Second, when multiple business units share limited resources, the lack of efficient collaboration mechanisms leads to departmental competition resulting in local optima rather than global optima.

[0003] Therefore, this invention proposes a digital operation management method and system for enterprises to systematically solve the above problems. Summary of the Invention

[0004] The present invention aims to solve the technical problems mentioned in the background section above.

[0005] The technical solution proposed in this invention is: a digital operation management method for enterprises, the method comprising: acquiring business activity data, extracting the interaction strength between business activities, and clustering multiple business activities into multiple intelligent agents;

[0006] Technical metadata is extracted from the data dictionary and interface documents of the target system, a metadata knowledge graph is constructed using the technical metadata, and a data knowledge graph is used to provide data interfaces for intelligent agents.

[0007] Based on the data interface, each agent uses a probability prediction model to generate a probability distribution of future external variables and makes individual decisions based on the probability distribution.

[0008] Determine whether there is shared resource coupling in the individual decisions of two agents. If so, introduce a cooperative mechanism to correct their independent decisions through shadow prices so that the individual decisions reach the global optimum.

[0009] Preferably, the target system includes an enterprise resource planning system, a manufacturing execution system, and a customer relationship management system; the technical metadata includes data sources, tables, fields, and application interface APIs.

[0010] Preferably, the construction of a metadata knowledge graph using technical metadata includes:

[0011] An initial graph is constructed using technical metadata. The initial graph includes table nodes, field nodes, and API nodes. The inclusion relationship between tables and fields is used as the inclusion relationship edge between table nodes and field nodes.

[0012] For any two field nodes in different tables and ;

[0013] Calculation obtained and similarity ;

[0014] in, For field name similarity, For field type similarity, For field context similarity, For the similarity of field value distribution, , , , These are the similarity weight coefficients;

[0015] if Greater than the preset threshold Then judge and Similar, and will and Connect via equivalent edges;

[0016] Clustering is performed on the initial graph using equivalent edges to group field nodes that describe the same business into a group, forming field clusters. A business label is added to each field cluster, and the field clusters with business labels are used as business entity nodes.

[0017] The database data definition language (DDL) of the target system is used to parse the reference relationships of each table node and construct the reference relationship edges;

[0018] A metadata knowledge graph is constructed using table nodes, field nodes, equivalence edges, business entity nodes, and reference relationship edges.

[0019] Preferably, the provision of a data interface for the intelligent agent using a data knowledge graph includes:

[0020] When a knowledge graph receives a semantic query request from an intelligent agent, it locates the corresponding business entity node.

[0021] Starting from the business entity node, the corresponding API is called, and the equivalent edges are traversed to find all corresponding field nodes. Then, a standardized list of data access handles is returned. The list of data access handles includes the field ID and the data source, i.e., the target system to which the field ID belongs.

[0022] Preferably, the step of acquiring business activity data, extracting the interaction strength between business activities, and clustering multiple business activities into multiple intelligent agents includes:

[0023] Obtain access to all business activities of the enterprise;

[0024] Identify key business activities from business activity data based on the enterprise's value stream;

[0025] Calculate the interaction weight of any two key business activities ;in For key business activities and key business activities Interaction frequency between them For data coupling degree, To coordinate the tightness, To change the synchronization rate; , , , This refers to the business weighting coefficient.

[0026] A directed graph of value stream is constructed using key business activities as value nodes, the dependencies between key business activities as directed edges, and the interaction weights as the edge weights of the directed edges.

[0027] Calculating global modularity using interaction weights Intelligent cohesion Coupling between agents ;

[0028] In intelligent cohesion The degree of aggregation is greater than the threshold and the coupling degree between agents is greater than the threshold. Under the constraint of less than the coupling degree threshold, with the goal of maximizing the global modularity, multiple key business activities are obtained and assigned to the same intelligent agent.

[0029] Preferred options also include:

[0030] Maintain a time decay coefficient for the directed graph of value stream ;

[0031] The interaction weights are updated online using a time decay coefficient to obtain the updated interaction weights. ;

[0032] in, , , , In the sliding window respectively Internal business activities and business activities Number of interactions, average data packet size per interaction, average response time, and modification lag metrics; when key business activities and key business activities When both are modified, the modification lag indicator is set to 1; otherwise, it is set to 0.

[0033] use Update intelligent cohesion Coupling between agents If the coupling between the updated agents is... If it increases, then business activities will be affected. and business activities The communication protocol is changed from asynchronous to synchronous, or the communication priority and bandwidth are increased;

[0034] If the updated smart cohesion If the value decreases and falls below the aggregation degree threshold, then the updated value will be... Replace with the average of historical interaction weights and submit a hot migration request. Manual determination will then be made regarding whether to migrate critical business activities. and key business activities Migrate out of the current agent.

[0035] Preferably, based on the data interface, each agent uses a probabilistic prediction model to generate a probability distribution of future external variables, and makes individual decisions based on the probability distribution, including:

[0036] In each decision cycle, the agent uses a data interface to obtain multi-source state data, that is, it obtains the data corresponding to the field ID from the data source through a list of data access handles. The multi-source state data includes internal variables and external variables.

[0037] The internal variables include current inventory level and current price;

[0038] The external variables include the demand quantity, demand price, start time, and end time of the current key business activity;

[0039] Multiple state features are extracted from multi-source state data within a decision-making cycle, and a state feature vector is generated by encoding them through a Transformer encoder. The state features include maximum inventory, minimum price, maximum price, average price, average inventory, maximum demand, earliest start time of the current key business, and average duration of the business.

[0040] Input the state feature vector into a pre-trained probabilistic prediction model to output the future. The probability distribution of the external variables after the step; the probability prediction model is a deep neural network; the probability distribution of the external variables includes the predicted values ​​of the external variables and their confidence levels;

[0041] Decisions are made and corresponding actions are taken based on the predicted values ​​and confidence levels of external variables.

[0042] The actions include: adjusting inventory demand, adjusting prices, adjusting business start time, and adjusting business end time.

[0043] Preferably, the step of determining whether there is shared resource coupling in the individual decisions of the two agents, and if so, introducing a cooperative mechanism to correct their independent decisions through shadow prices so that the individual decisions reach the global optimum, includes:

[0044] When two agents share global resources in their decisions, a global negotiation is initiated, including:

[0045] Set up a lightweight digital coordination node;

[0046] At the start of negotiations, the digital coordinating node initializes the shadow price. And broadcast;

[0047] Let each intelligent agent In the Round reception ;

[0048] Construct the total cost optimization function ; Total cost;

[0049] The optimal solution is obtained by solving the total cost optimization function using a quadratic programming algorithm. ;

[0050] in, For intelligent agents Internal costs, This is the global resource limit. This is the penalty coefficient;

[0051] For intelligent agents In the The optimal action corresponding to the round of decision-making. Resource usage fees paid at shadow prices Penalty for overall supply and demand imbalance;

[0052] intelligent agent Demand Return to the coordinating node;

[0053] Coordination nodes summarize total requirements Calculate the original residuals And update the shadow price as follows: ; The number of agents;

[0054] When the original residual is less than the preset tolerance, negotiated convergence occurs, including:

[0055] Each agent uses the final shadow price and the optimal action to revise its independent decision and execute the corresponding optimal action to achieve global cost optimization.

[0056] A digital operation management system for enterprises, the system being used to execute the aforementioned digital operation management method for enterprises.

[0057] A computer-readable storage medium storing a computer program that is executed by a processor to implement the aforementioned enterprise digital operation management method.

[0058] The beneficial effects of this invention are:

[0059] 1. This invention avoids data silos by constructing a metadata knowledge graph. Key business activities are identified based on the enterprise's end-to-end value stream. A weighted directed interaction graph is constructed using four dimensions: interaction frequency, data coupling, coordination tightness, and change synchronization rate. Activities are clustered to maximize global modularity, resulting in multiple autonomous agents. This method ensures high cohesion among business activities within agents and low coupling between agents, guaranteeing the semantic integrity of each agent while minimizing communication and dependency overhead during collaboration.

[0060] 2. In this invention, when multiple agents' decisions involve shared resources and coupling, a collaborative mechanism is introduced. Shadow prices are used to correct each agent's independent decisions, ensuring that each decision achieves global optimality. While protecting the data sovereignty and autonomous decision-making rights of each agent, negotiation replaces traditional centralized scheduling, significantly improving the system's scalability and robustness. Attached Figure Description

[0061] Figure 1 This is a flowchart of a digital operation management method for enterprises according to the present invention. Detailed Implementation

[0062] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0063] It is understood that the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0064] refer to Figure 1 The technical solution provided by this invention is: a method for enterprise digital operation management, the method comprising:

[0065] Step 1: Acquire business activity data, extract the interaction strength between business activities, and cluster multiple business activities into multiple intelligent agents. This includes the following steps:

[0066] Step 1.1: Obtain all business activities of the enterprise. For example, order receipt, inventory inquiry, outbound shipment, cash receipt, invoicing, etc.

[0067] Identify key business activities from business activity data based on the enterprise's value stream (such as order receipt and cash inflow);

[0068] Calculate key business activities and key business activities Interaction weights ;in For key business activities and key business activities Interaction frequency between them For data coupling degree, To coordinate the tightness, This refers to the change synchronization rate (the probability that a change in one party historically causes a change in the other party). , , , This refers to the business weighting coefficient.

[0069] A directed graph of the value stream is constructed, with key business activities as value nodes, the dependencies between these activities defined by key business activities as directed edges, and the interaction weights as the edge weights. ;

[0070] Calculating global modularity using interaction weights Intelligent cohesion Coupling between agents The calculation formula is: ;in, It is the sum of the interaction weights within a directed graph of the same agent. This is the sum of the weights of key business activities within a directed graph of the same agent and key business interactions inside and outside the directed graph (regardless of the interaction direction). The closer this value is to 1, the higher the cohesion.

[0071] ;in To and A directed graph of value flows between different intelligent agents; It is a positive number, used to avoid the denominator being zero.

[0072] ;in, For key business activities Total output weights For key business activities The total input weights. Weights and sum , To adjust the coefficient, when key business activities and key business activities The value is 1 when the target belongs to the same intelligent agent, and 0 otherwise.

[0073] In intelligent cohesion The degree of aggregation is greater than the aggregation threshold (e.g., 0.7) and the coupling degree between agents is greater than the aggregation threshold. Under constraints less than a coupling threshold (e.g., 0.1) (high cohesion, low coupling), with the goal of maximizing global modularity, multiple key business activities are obtained and assigned to the same intelligent agent. That is, key business activities are... They are assigned to the same intelligent agent.

[0074] Step 2: Extract technical metadata from the target system's data dictionary and interface documents, construct a metadata knowledge graph using the technical metadata, and use the data knowledge graph to provide data interfaces for the intelligent agent. This specifically includes the following steps:

[0075] Step 2.1: Scan the data dictionary and interface documentation of the Enterprise Resource Planning System, Manufacturing Execution System, and Customer Relationship Management System to extract metadata such as data sources, tables, fields, and application interface APIs;

[0076] An initial graph is constructed using technical metadata. The initial graph includes table nodes, field nodes, and API nodes. The inclusion relationship between tables and fields is used as the inclusion relationship edge between table nodes and field nodes.

[0077] For any two field nodes in different tables and By using the pre-trained language model BERT, and Convert field names to semantic vectors .

[0078] Calculation obtained and similarity ;

[0079] if Greater than the preset threshold Then judge and Similar, and will and Connected by equivalent edges. In this embodiment... .

[0080] in, For field name similarity, ,For example The field name is cust_id. If the field name is customer_identifier, then these two field names are highly similar. , , , These are the similarity weight coefficients.

[0081] This represents field type similarity. If two fields have the same data type, such as both being integers or strings, then the field similarity is 1. Field similarity can be used to quickly filter out fields with clearly mismatched data types.

[0082] This is the field context similarity, which is the sum of the name similarities of the neighboring fields in the tables containing the two fields. For example, if two tables both contain fields such as "Customer ID", "Address", and "Phone", then the field context similarity of the "Customer ID" field is the sum of the cosine similarity of the "Address" field and the cosine similarity of the "Phone" field.

[0083] The similarity is denoted by the distribution of field values, and the Jaccard coefficient is the distribution of the values ​​of two fields. If the names of two fields are different, but the Jaccard coefficient is high, that is, the values ​​of the two fields have a high overlap rate, the two fields are also considered similar.

[0084] Step 2.2: Cluster the initial graph using equivalent edges to group field nodes that describe the same business into a group, forming field clusters. Add a business label to each field cluster and use the field clusters with business labels as business entity nodes.

[0085] Step 2.3: Parse the reference relationships of each table node from the target system's database DDL (Data Definition Language) and construct reference relationship edges. Use table nodes, field nodes, equivalence edges, business entity nodes, and reference relationship edges to construct a metadata knowledge graph.

[0086] This includes using data knowledge graphs to provide data interfaces for intelligent agents, including:

[0087] When a knowledge graph receives a semantic query request from an intelligent agent, it locates the corresponding business entity node.

[0088] Starting from the business entity node, the corresponding API is called, and the equivalent edges are traversed to find all corresponding field nodes. Then, a standardized list of data access handles is returned. The list of data access handles includes the field ID and the data source, i.e., the target system to which the field ID belongs.

[0089] By constructing a data knowledge graph, intelligent question-and-answer-style data discovery can be achieved. For example, when a query is received for "total sales of the sales business," the system can retrieve all fields associated with the "sales" business entity, find the corresponding fields, and obtain the relevant sales data.

[0090] Automating impact analysis. For example, when a field in a table needs to be changed, the metadata knowledge graph is traversed to find all similar fields and corresponding business entities to check whether a change has occurred, avoiding manual investigation.

[0091] This completely eliminates the problem of data silos, providing agents with a shared data space. Subsequent predictions by agents are all based on this data space.

[0092] In some preferred embodiments, a time decay coefficient is maintained for the directed graph of the value stream. ;

[0093] The interaction weights are updated online using a time decay coefficient to obtain the updated interaction weights. .

[0094] in, , , , In the sliding window respectively Internal business activities and business activities Number of interactions, average data packet size per interaction, average response time, and modification lag metrics; when key business activities and key business activities When both are modified, the modification lag indicator is set to 1; otherwise, it is set to 0.

[0095] use Update intelligent cohesion Coupling between agents If the coupling between the updated agents is... If it increases, then business activities will be affected. and business activities The communication protocol is changed from asynchronous to synchronous, or the communication priority and bandwidth are increased.

[0096] If the updated smart cohesion If the value decreases and falls below the aggregation degree threshold, then the updated value will be... Replace with the average of historical interaction weights and submit a hot migration request. Manual determination will then be made regarding whether to migrate critical business activities. and key business activities The current agent is migrated out. In this embodiment, the agents include a sales forecasting agent, an inventory management agent, a procurement and replenishment agent, and a scheduling agent. For example, in the critical business chain mentioned above—order receipt - inventory query - outbound - cash receipt - invoicing—the agents involved include the sales forecasting agent and the inventory management agent. The critical business activities such as order receipt and cash receipt will be assigned to the sales forecasting agent. Inventory query and outbound will be assigned to the inventory management agent.

[0097] Step 3: Based on the data interface, each agent uses a probabilistic prediction model to generate a probability distribution of future external variables, and makes individual decisions based on the probability distribution. This specifically includes the following steps:

[0098] In each decision cycle, the agent uses a data interface to obtain multi-source state data, that is, it obtains the data corresponding to the field ID from the data source through a list of data access handles. The multi-source state data includes internal variables and external variables.

[0099] The internal variables include current inventory level and current price;

[0100] The external variables include the inventory demand for the current key business activity, the product sales price, the start time of the current key business activity, and the end time of the current key business activity.

[0101] Multiple state features are extracted from multi-source state data within a decision-making cycle, and a state feature vector is generated by encoding them using a Transformer encoder. The state features include maximum inventory, minimum price, maximum price, average price, average inventory, maximum demand, earliest start time of the current key business, and average duration of the business.

[0102] Input the state feature vector into a pre-trained probabilistic prediction model to output the future. The probability distribution of the external variables after the step; the probability prediction model is a deep neural network; the probability distribution of the external variables includes the predicted values ​​of the external variables and their confidence levels;

[0103] Decisions are made and corresponding actions are taken based on the predicted values ​​and confidence levels of external variables. If the confidence level of the predicted value is greater than a preset confidence threshold, such as 0.8, then the appropriate action is taken based on the corresponding predicted value.

[0104] The actions include: adjusting inventory demand, adjusting prices, adjusting business start time, and adjusting business end time.

[0105] For example, if a sales forecasting agent predicts future inventory demand of 800 square meters with a confidence level greater than 0.8, the forecast is considered valid. If the current maximum inventory level in the warehouse is 700 square meters, an inventory adjustment action needs to be performed, i.e., increasing the inventory level, such as allocating spare warehouse space.

[0106] Step 4: Determine if there is resource coupling in the individual decisions of the two agents. If so, introduce a cooperative mechanism to correct their independent decisions through shadow prices, so that each individual decision reaches the global optimum. This includes the following steps:

[0107] When two agents share global resources in their decision-making, global negotiation is initiated. In this embodiment, the shared global resources include total inventory area and total budget.

[0108] The specific steps are as follows:

[0109] Set up a lightweight digital coordination node;

[0110] At the start of negotiations, the digital coordinating node initializes the shadow price. And broadcast;

[0111] Let each intelligent agent In the Round reception ;

[0112] Construct the total cost optimization function ; Total cost;

[0113] The optimal solution is obtained by solving the total cost optimization function using a quadratic programming algorithm. ;

[0114] in, For intelligent agents Internal costs, This is the global resource limit. This is the penalty coefficient;

[0115] For intelligent agents In the The optimal action corresponding to the round of decision-making. Resource usage fees paid at shadow prices Penalty for global supply and demand imbalance;

[0116] intelligent agent Demand Return to the coordinating node.

[0117] Coordination nodes summarize total requirements Calculate the original residuals And update the shadow price as follows: ; The number of agents.

[0118] When the original residual is less than the preset tolerance, negotiated convergence occurs, including:

[0119] Each agent uses the final shadow price and the optimal action to revise its independent decision and execute the corresponding optimal action to achieve global cost optimization.

[0120] For example, consider two agents: an inventory management agent and a scheduling agent. The inventory management agent predicts future inventory of 600 square meters, while the scheduling agent predicts future production will require 500 square meters, but the total capacity is 1000 square meters. These conditions cannot be met simultaneously.

[0121] Therefore, global negotiation needs to be initiated, specifically as follows:

[0122] Introduce a digital coordinator as the digital coordination node. Set the initial construction price. (RMB / square meter·cycle), and broadcast. This indicates that storage space is currently free. A penalty coefficient is also set. .

[0123] Inventory Management Intelligent Agent Local optimization: Constructing an augmented local cost function:

[0124] ; First item The first item is internal cost (holding cost + stockout loss); the second item, price cost, is 0; the third item, deviation penalty, is... In other words, the less you occupy, the greater the penalty. When the price is zero and the deviation penalty encourages over-occupancy, the inventory management agent... Still reporting ideal demand .

[0125] Generate scheduling agent Local optimization: Similarly, the scheduling agent also reports the ideal requirements. .

[0126] Coordinator Summary: Total Requirements Original residuals (Excess demand). Update shadow prices: .

[0127] This can be interpreted as you two need an additional 100 square meters, and the current warehousing unit price has increased to 200. (New price broadcast.)

[0128] Round 2 Negotiation: Coordinator Broadcast: Price And the other party's previous requirements (inform the inventory management AI). Generate scheduling agent 500 was requested; the scheduling agent was then informed. Inventory Management Intelligent Agent I asked for 600.

[0129] Inventory Management Intelligent Agent Local optimizations:

[0130] Now, occupying an extra square meter not only incurs a storage fee of 200 yuan, but also... Deviations are penalized (if) (Then the penalty increases). Under the combined effect of internal costs and these two additional costs, the inventory management agent... We must exercise greater restraint. (Assuming an inventory management intelligent agent...) Reported after re-optimization .

[0131] Generate scheduling agent Local optimizations:

[0132] Deviation penalty items like The penalty will then increase. Generate a scheduling agent. Reported after re-optimization .

[0133] Coordinator Summary: Total Requirements Original residuals (Insufficient demand, idle resources). Update shadow prices: .

[0134] Prices cannot be negative (negative prices encourage waste), and in actual engineering projects, a lower limit for prices is set at 0. .

[0135] This can be interpreted as you want less this time, there's idle space in the warehouse, and the price has dropped back to 0. The new price is being broadcast.

[0136] Round 3 Negotiation: Coordinator Broadcast: Price And the other party's needs.

[0137] Inventory Management Intelligent Agent Local optimizations: ;

[0138] The price is zero, but the deviation penalty is in effect. .like Penalties increase if the score exceeds 580. Inventory management intelligent agent. Reported after taking into account internal costs .

[0139] Generate scheduling agent Local optimizations: Similarly, in deviation punishment Report .

[0140] Coordinator Summary: Total Requirements Original residuals Supply and demand are basically balanced, and the residual is within the tolerance range (e.g., ±10).

[0141] The dual residual is also close to zero. Negotiation converges. The optimal solutions at this point are 500 and 490, and the inventory management agent... The optimal action to execute is to adjust the inventory demand to 500 square meters. A scheduling agent is then generated. The optimal action is to adjust inventory demand to 490 square meters.

[0142] During the collaboration process, neither agent was "commanded" to reduce their resource usage significantly. They simply perceived a shared price signal (and a deviation penalty), internalized the cost of using the shared resource in their respective local cost functions, and thus spontaneously adjusted their total demand to within capacity.

[0143] Therefore, through decentralized negotiation driven by price signals and independent decision-making adjustments by each agent, global optimum is ultimately achieved. This is precisely the core value of the multi-agent collaborative mechanism.

[0144] The present invention also provides an enterprise digital operation management system, the system being used to execute the aforementioned enterprise digital operation management method.

[0145] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the aforementioned enterprise digital operation management method.

[0146] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0148] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the principles described, the implementation of the present invention may have any changes or modifications.

Claims

1. A method for enterprise digital operation management, characterized in that, The method includes: Acquire business activity data, extract the interaction strength between business activities, and cluster multiple business activities into multiple intelligent agents; Extract technical metadata from the data dictionary and interface documents of the target system, construct a metadata knowledge graph using the technical metadata, and provide data interfaces for intelligent agents using the data knowledge graph; Based on the data interface, each agent uses a probability prediction model to generate a probability distribution of future external variables and makes individual decisions based on the probability distribution. Determine whether there is shared resource coupling in the individual decisions of two agents. If so, introduce a cooperative mechanism to correct their independent decisions through shadow prices so that the individual decisions reach the global optimum.

2. The enterprise digital operation management method according to claim 1, characterized in that, The target systems include Enterprise Resource Planning (ERP) systems, Manufacturing Execution System (MES) systems, and Customer Relationship Management (CRM) systems; the technical metadata includes data sources, tables, fields, and Application Programming Interfaces (APIs).

3. The enterprise digital operation management method according to claim 2, characterized in that, The construction of a metadata knowledge graph using technical metadata includes: An initial graph is constructed using technical metadata. The initial graph includes table nodes, field nodes, and API nodes. The inclusion relationship between tables and fields is used as the inclusion relationship edge between table nodes and field nodes. For any two field nodes in different tables and ; Calculation obtained and similarity in, For field name similarity, For field type similarity, For field context similarity, For the similarity of field value distribution, , , , These are the similarity weight coefficients; if Greater than the preset threshold Then judge and Similar, and and Connect via equivalent edges; Clustering is performed on the initial graph using equivalent edges to group field nodes that describe the same business into a group, forming field clusters. A business label is added to each field cluster, and the field clusters with business labels are used as business entity nodes. The database data definition language (DDL) of the target system is used to parse the reference relationships of each table node and construct the reference relationship edges; A metadata knowledge graph is constructed using table nodes, field nodes, equivalence edges, business entity nodes, and reference relationship edges.

4. The enterprise digital operation management method according to claim 3, characterized in that, The provision of data interfaces for intelligent agents using data knowledge graphs includes: When a knowledge graph receives a semantic query request from an intelligent agent, it locates the corresponding business entity node. Starting from the business entity node, the corresponding API is called, and the equivalent edges are traversed to find all corresponding field nodes. Then, a standardized list of data access handles is returned. The list of data access handles includes the field ID and the data source, i.e., the target system to which the field ID belongs.

5. The enterprise digital operation management method according to claim 4, characterized in that, The steps of acquiring business activity data, extracting the interaction strength between business activities, and clustering multiple business activities into multiple intelligent agents include: Obtain access to all business activities of the enterprise; Identify key business activities from business activity data based on the enterprise's value stream; Calculate the interaction weight of any two key business activities ;in For key business activities and key business activities Interaction frequency between them For data coupling degree, To coordinate the tightness, To change the synchronization rate; , , , This refers to the business weighting coefficient. A directed graph of value stream is constructed using key business activities as value nodes, the dependencies between key business activities as directed edges, and the interaction weights as the edge weights of the directed edges. Calculating global modularity using interaction weights Intelligent cohesion Coupling between agents ; In intelligent cohesion The degree of aggregation is greater than the threshold and the coupling degree between agents is greater than the threshold. Under the constraint of less than the coupling degree threshold, with the goal of maximizing the global modularity, multiple key business activities are obtained and assigned to the same intelligent agent.

6. The enterprise digital operation management method according to claim 5, characterized in that, Also includes: Maintain a time decay coefficient for the directed graph of value stream ; The interaction weights are updated online using a time decay coefficient to obtain the updated interaction weights. in, , , , In the sliding window Internal business activities and business activities Number of interactions, average data packet size per interaction, average response time, and modification lag metrics; when key business activities and key business activities When both are modified, the modification lag indicator is set to 1; otherwise, it is set to 0. use Update intelligent cohesion Coupling between agents If the coupling between the updated agents is... If it increases, then business activities will be affected. and business activities The communication protocol is changed from asynchronous to synchronous, or the communication priority and bandwidth are increased; If the updated smart cohesion If the value decreases and falls below the aggregation degree threshold, then the updated value will be... Replace with the average of historical interaction weights and submit a hot migration request. Manual determination will then be made regarding whether to migrate critical business activities. and key business activities Migrate out of the current agent.

7. The enterprise digital operation management method according to claim 6, characterized in that, Based on the data interface, each agent uses a probabilistic prediction model to generate a probability distribution of future external variables, and makes individual decisions based on the probability distribution, including: In each decision cycle, the agent uses a data interface to obtain multi-source state data, that is, it obtains the data corresponding to the field ID from the data source through a list of data access handles. The multi-source state data includes internal variables and external variables. The internal variables include current inventory level and current price; The external variables include the demand quantity, demand price, start time, and end time of the current key business activity; Multiple state features are extracted from multi-source state data within a decision-making cycle, and a state feature vector is generated by encoding them through a Transformer encoder. The state features include maximum inventory, minimum price, maximum price, average price, average inventory, maximum demand, earliest start time of the current key business, and average duration of the business. Input the state feature vector into a pre-trained probabilistic prediction model to output the future. The probability distribution of the external variables after the step; the probability prediction model is a deep neural network; the probability distribution of the external variables includes the predicted values ​​of the external variables and their confidence levels; Decisions are made and corresponding actions are taken based on the predicted values ​​and confidence levels of external variables. The actions include: adjusting inventory demand, adjusting prices, adjusting business start time, and adjusting business end time.

8. The enterprise digital operation management method according to claim 7, characterized in that, The process of determining whether the individual decisions of the two agents are coupled due to shared resources, and if so, introducing a collaborative mechanism to correct their independent decisions through shadow prices, so that the individual decisions reach the global optimum, includes: When two agents share global resources in their decisions, a global negotiation is initiated, including: Set up a lightweight digital coordination node; At the start of negotiations, the digital coordinating node initializes the shadow price. And broadcast; Let each intelligent agent In the Round reception ; Construct the total cost optimization function Total cost; The optimal solution is obtained by solving the total cost optimization function using a quadratic programming algorithm. in, For intelligent agents Internal costs, This is the global resource limit. This is the penalty coefficient; For intelligent agents In the The optimal action corresponding to the round of decision-making. Resource usage fees paid at shadow prices Penalty for global supply and demand imbalance; For the first Resource requirements for round-robin decision-making; intelligent agent Demand Return to the coordinating node; Coordination nodes summarize total requirements Calculate the original residuals And update the shadow price as follows: ; The number of agents; When the original residual is less than the preset tolerance, negotiated convergence occurs, including: Each agent uses the final shadow price and the optimal action to revise its independent decision and execute the corresponding optimal action to achieve global cost optimization.

9. A digital operation management system for enterprises, characterized in that, The system is used to execute an enterprise digital operation management method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement an enterprise digital operation management method according to any one of claims 1-8.

Images