A method and apparatus for modeling an entity network based on an agent system

Abstract

The application provides an entity network modeling method and device based on an agent system, including a first agent and a second agent in communication with each other, the first agent is used for constructing a data relationship graph of the entity network, and the second agent is used for generating a data model of the entity network according to the data relationship graph. In the application, the problems that a traditional data modeling method needs to master a large amount of business field and data statistical knowledge, has poor generalization ability, and is not timely, etc. are solved, the global data is automatically and timely preprocessed, relationship is recognized, and data modeling is performed through the agent and the prompt word, obvious advantages are obtained in operability, automation, construction cost, generalization, and real-time performance, and the bidirectional mapping, dynamic interaction, and real-time connection of the 6G physical network and the digital twin network can be effectively supported.

Description

TECHNICAL FIELD

[0001] One or more embodiments of the present specification relate to the technical field of network, and in particular, to a modeling method and device of an entity network based on an agent system. BACKGROUND

[0002] 6G Integrated Space-Air-Ground Networking for 6G is a seamless coverage communication network system that combines satellite communication, air unmanned aerial vehicle network and ground 5G / 6G infrastructure. In the face of such complex networking environment modeling, common data modeling techniques and methods, including statistical modeling, physical modeling, machine learning and data mining, have technical problems such as complex calculation process, high cost, the need for a large amount of training data, overly complex models and overfitting training data leading to poor generalization ability. At present, there is an urgent need for a data modeling method for 6G Integrated Space-Air-Ground Networking that has low operational complexity, high automation, low investment cost and high accuracy.

[0003] SUMMARY

[0004] The present application describes a modeling method and device of an entity network based on an agent system, which can solve the above technical problems.

[0005] According to a first aspect, a modeling method of an entity network based on an agent system is provided, applied to an agent system, the agent system comprising a first agent and a second agent in communication with each other, the first agent being configured to construct a data relationship graph of the entity network, and the second agent being configured to generate a data model of the entity network according to the data relationship graph, the method comprising:

[0006] The first agent perceives multi-modal data of the entity network, the multi-modal data comprising data information of an environment of the entity network;

[0007] The first agent establishes a data relationship graph of the entity network according to the multi-modal data, and sends the data relationship graph to the second agent, the data relationship graph being configured to represent the relationship type between network entity data and network operation data in the entity network;

[0008] The second agent generates a data model of the entity network according to the data relationship graph.

[0009] In some embodiments, the first agent establishes a data relationship graph of the entity network according to the multi-modal data, specifically comprising:

[0010] The first intelligent agent utilizes a large model to process the multi-modal data according to a preset data processing prompt word, to obtain full entities and network operation data of the entity network.

[0011] The large model is utilized to perform relationship identification on the full entities and network operation data according to a preset relationship identification prompt word.

[0012] According to the identification result, a data relationship graph of the entity network is generated, which is a triple graph taking network entity data and network operation data in the entity network as nodes and a relationship type between each network entity data and network operation data as edges.

[0013] In some more specific embodiments, the first intelligent agent utilizes a large model to process the multi-modal data according to a preset data processing prompt word, to obtain full entities and network operation data of the entity network, specifically including:

[0014] The first intelligent agent utilizes a large model to perform missing value judgment on the multi-modal data through a first preset data processing prompt word, and according to the judgment result, processes the multi-modal data through a second preset data processing prompt word, the first preset data processing prompt word is used to prompt a data value missing judgment standard, and the second preset data processing prompt word is used to prompt a processing mode for multi-modal data with missing data values;

[0015] The first intelligent agent utilizes a large model to perform duplicate item judgment on the multi-modal data through a third preset data processing prompt word, and according to the judgment result, processes the multi-modal data through a fourth preset data processing prompt word, the third preset prompt word is used to prompt a duplicate item judgment standard, and the fourth preset data processing prompt word is used to prompt a processing mode for multi-modal data with duplicate items;

[0016] The first intelligent agent utilizes a large model to perform abnormal value judgment on the multi-modal data through a fifth preset data processing prompt word, and according to the judgment result, processes the multi-modal data through a sixth preset data processing prompt word, the fifth preset prompt word is used to prompt an abnormal value judgment standard, and the sixth preset data processing prompt word is used to prompt a processing mode for multi-modal data with abnormal values.

[0017] In some more specific embodiments, the first intelligent agent utilizes a large model to perform relationship identification on the full entities and network operation data according to a preset relationship identification prompt word, specifically including:

[0018] The first intelligent agent uses a first preset relationship identification prompt word to determine the containment relationship of the full-amount entity and network running data, and according to the determination result, uses a second preset relationship identification prompt word to integrate the data with the containment relationship in the full-amount entity and network running data, the first preset relationship identification prompt word is used to prompt the containment relationship determination standard, and the second preset relationship identification prompt word is used to prompt the processing mode of the data with the containment relationship;

[0019] The first intelligent agent uses a third preset relationship identification prompt word to determine the association relationship of the full-amount entity and network running data, and according to the determination result, uses a fourth preset relationship identification prompt word to model the relationship of the data with the association relationship in the full-amount entity and network running data, the third preset relationship identification prompt word is used to prompt the association relationship determination standard, and the fourth preset relationship identification prompt word is used to prompt the relationship modeling mode of the data with the association relationship;

[0020] The first intelligent agent uses a fifth preset relationship identification prompt word to determine the causal relationship of the full-amount entity and network running data, and according to the determination result, uses a sixth preset relationship identification prompt word to jointly model the data with the causal relationship in the full-amount entity and network running data, the fifth preset relationship identification prompt word is used to prompt the causal relationship determination standard, and the sixth preset relationship identification prompt word is used to prompt the processing mode of the data with the causal relationship.

[0021] In some embodiments, the first intelligent agent perceives the multi-modal data of the entity network, specifically including:

[0022] The first intelligent agent perceives the network entity attribute data in the entity network environment by means of a multi-modal sensor and an external vector database, perceives geometric data and network running data by means of the multi-modal sensor, and perceives rules by means of the external vector database;

[0023] The network entity attribute data is used to represent the data information of the entity device in the entity network, the geometric data is used to represent the spatial position of the entity device in the entity network, the network running data is used to represent the alarm, performance index and log information in the entity network, and the rules are used to represent the management rules and service quality requirements of the entity network.

[0024] In some embodiments, the second intelligent agent generates a data model of the entity network according to the data relationship graph, specifically including:

[0025] The second agent generates a basic data model and a functional data model of the entity network according to the input data relationship graph, by calling a database query tool and a code generation tool in the tool, the basic data model is used to represent a basic data model of an entity device in the entity network, and the functional data model is used to represent a running logic of each entity device in the entity network.

[0026] In some embodiments, the agent system further comprises a third agent, which is used to monitor the first agent and the second agent.

[0027] According to a second aspect, a modeling device of an entity network based on an agent system is provided, which is applied to an agent system, the agent system comprises a first agent and a second agent which communicate with each other, the first agent is used to build a data relationship graph of an entity network, and the second agent is used to generate a data model of the entity network according to the data relationship graph, and the device comprises:

[0028] A first processing module is used for the first agent to perceive multi-modal data of an entity network, and the multi-modal data comprises data information of an entity network environment;

[0029] A second processing module is used for the first agent to build a data relationship graph of the entity network according to the multi-modal data, and send the data relationship graph to the second agent, and the data relationship graph is used to represent a relationship type between network entity data and network running data in the entity network;

[0030] A third processing module is used for the second agent to generate a data model of the entity network according to the data relationship graph.

[0031] According to a third aspect, an agent system is provided, which is used for modeling of an entity network, and the agent system comprises a first agent and a second agent which communicate with each other, the first agent is used to build a data relationship graph of an entity network, and the second agent is used to generate a data model of the entity network according to the data relationship graph;

[0032] The first agent is used to perceive multi-modal data of an entity network, and the multi-modal data comprises data information of an entity network environment;

[0033] The first agent is used to build a data relationship graph of the entity network according to the multi-modal data, and send the data relationship graph to the second agent, and the data relationship graph is used to represent a relationship type between network entity data and network running data in the entity network;

[0034] The second intelligent agent is configured to generate a data model of the entity network according to the data relationship graph.

[0035] In the system and method provided by the embodiments of the present specification, the problems of the conventional data modeling method, such as the need to master a large amount of business domain and data statistical knowledge, poor generalization ability, and low timeliness, are solved. The global data is automatically and real-timely preprocessed, the relationship is identified, and the data modeling is performed through the intelligent agent and the prompt word, which has obvious advantages in operability, automation, construction cost, generalization, and real-time performance, and can effectively support the bidirectional mapping, dynamic interaction, and real-time connection of the 6G physical network and the digital twin network. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0037] Figure 1 The system schematic diagram of the intelligent agent system for modeling of the entity network provided by the embodiments of the present specification is shown.

[0038] Figure 2 The flow schematic diagram of the modeling method of the entity network based on the intelligent agent system provided by the embodiments of the present specification is shown.

[0039] Figure 3 The module schematic diagram of the modeling device of the entity network based on the intelligent agent system provided by the embodiments of the present specification is shown.

[0040] Figure 4 The schematic diagram of the entity network based on the intelligent agent system provided by the embodiments of the present specification is shown. DETAILED DESCRIPTION

[0041] The schemes provided by the present specification will be described below with reference to the drawings.

[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be described below with reference to the drawings.

[0043] In the description of the embodiments of the present application, the words such as "exemplary", "for example", or "for instance" are used to represent an example, illustration or description. Any embodiment or design scheme described as "exemplary", "for example", or "for instance" in the embodiments of the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. In fact, the words such as "exemplary", "for example", or "for instance" are intended to present the relevant concept in a specific way.

[0044] In the description of the embodiments of the present application, the term "and / or" is merely an association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent three cases of A alone, B alone, and A and B together. In addition, unless otherwise specified, the term "multiple" means two or more.

[0045] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly indicating the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more features. The terms "include", "contain", "have" and their variants mean "include but are not limited to", unless otherwise specifically emphasized.

[0046] Currently, the modeling methods of entity networks include statistical modeling, physical modeling, machine learning modeling and data mining modeling.

[0047] Statistical modeling is a method of describing and predicting various phenomena in the network based on statistical data, constructing a model by analyzing a large amount of network data, and estimating model parameters and verifying hypotheses using statistical methods, but it requires professional personnel to model.

[0048] Physical modeling is based on physical principles and mathematical formulas to construct accurate data models in order to better understand and optimize network performance, but the process cost of physical modeling is high, and in some cases it may not meet the user's needs. In addition, some three-dimensional modeling software may have the disadvantages of complex user interface, software stability problems, etc., which may increase the learning cost and affect the work efficiency.

[0049] Machine learning modeling uses algorithms to learn patterns from a large amount of network historical data and establishes a data model accordingly. Network event prediction or decision-making is made using the data model. However, machine learning modeling requires training a large amount of data, and may have the problems of overfitting and being called "black box".

[0050] Data mining modeling extracts valuable information from a large amount of network historical data to obtain a data model to support decision-making, performance optimization, fault detection and other application scenarios. However, there are problems such as overfitting of training data leading to poor generalization ability, low algorithm efficiency caused by multiple sequential scanning and sorting of data sets during tree construction in the modeling process.

[0051] 6G Integrated Space-Air-Ground Networking (ISAGN) refers to the combination of satellite communication, aerial unmanned aerial vehicle network, and ground 5G / 6G infrastructure to form a seamless communication network system. This networking approach overcomes the limitations of single network coverage, transmission delay, and other issues, providing wider coverage, higher transmission rates, and better service quality. In the face of the complex environment of 6G ISAGN, a method of integrated data modeling with low operational complexity, high automation, low investment cost, and high accuracy is urgently needed.

[0052] The present application aims to improve user experience and 6G network data modeling automation by using an intelligent agent system to dynamically model 6G physical networks. The system mainly includes one control agent, i.e., the central control agent, and two business agents, i.e., the data relationship graph agent and the data modeling agent. The data relationship graph is a triple graph composed of data types as entities and data relationships as edges. The 6G data model includes a basic data model and a functional data model. The basic model is a model definition for 6G physical network entities, a digital twin model established from multiple dimensions, and is strongly related to the type of 6G network entities. The functional model is used to describe the internal operation logic of the 6G network twin, supports feedback and control of the twin network, and different dimensions of functions are created by matching and combining to create functional models for different twin application scenarios.

[0053] As Figure 1 As shown, the present application dynamically models 6G entity networks through an intelligent agent system, which includes one control agent, i.e., the central control agent, and two business agents, i.e., the data relationship graph agent and the data modeling agent.

[0054] The central control agent acts as a leader and monitors and coordinates the two agents below, does not undertake specific business functions, is responsible for coordinating the work of other agents, and ensures the efficient operation of the entire system. Possible tasks include but are not limited to resource scheduling, error detection and recovery, standardization of communication protocols, etc. In the 6G network environment, the central control agent may also need to handle multi-device collaboration, data flow management, etc.

[0055] The data relationship graph agent perceives multi-modal data of the 6G entity network environment with the help of multi-modal sensors and external vector databases, etc. The multi-modal data includes network entity and attribute data, geometric data, network operation data, and rules. Network entity and attribute data includes device information, connection status, location information, etc. Geometric data includes spatial layout, relative position between devices, etc. Network operation data includes alarms, performance indicators, and logs, etc. Rules include network management rules and quality of service requirements, etc.

[0056] The data relationship graph agent performs data preprocessing and relationship graph construction on the perceived multi-modal data. Data preprocessing mainly includes handling missing values, deleting duplicates, handling outliers, etc. After data preprocessing, the data relationship graph agent identifies relationships for the full entity and network operation data. Relationship types include inclusion relationship, association relationship and causal relationship. According to the cosine similarity of the full entity and network operation data, it is determined whether there is an inclusion relationship. The data with inclusion relationship is integrated to eliminate redundant information and provide consistent and accurate data model representation. According to whether the data of the full entity and network operation data have the same attributes, it is determined whether there is an association relationship, and the specific association relationship type is subdivided. For example, the network topology relationship composed of network devices, boards, ports and circuits. It is determined whether the data of the full entity and network operation data have a causal relationship, and the data with a causal relationship are jointly modeled, such as the causal relationship between network performance degradation and network alarms.

[0057] The data relationship graph agent generates a complete relationship graph according to the association relationship and causal relationship of the full entity and network operation data, and performs automatic pruning based on reinforcement learning with human feedback (RLHF) to form the final data relationship graph. The graph takes data types as nodes and relationships between data as edges to form a complex network graph. The data relationship graph is dynamically updated according to the data to reflect the latest data state and changes in the 6G entity network. The data relationship graph agent outputs the data relationship graph to the data modeling agent.

[0058] The data modeling agent automatically generates basic data models and functional data models based on the input data relationship graph by calling database queries and code generation tools in the tool. The basic data model includes entities in the 6G physical network, such as base stations and user equipment. These models digitally twin the network entities from multiple dimensions such as geographic location and performance indicators to better understand and predict the behavior of network device entities. The functional data model mainly includes the running logic of the network, such as traffic distribution and quality of service guarantee mechanisms. Functional models can be used to simulate different aspects of network operations and support dynamic feedback and control of the network. As the data relationship graph is updated, the data modeling agent continuously updates existing models and optimizes these models based on new insights.

[0059] The Data Modeling Agent and the Data Relationship Graph Agent play complementary roles in constructing the dynamic data model of the 6G network. They have a data sharing relationship, and the graph provided by the Data Relationship Graph Agent is one of the important foundations for the Data Modeling Agent to model. Through the graph, the Data Modeling Agent can obtain a comprehensive view of the data elements and their relationships, which is crucial for building accurate models. The two agents have a complementary relationship, with the Data Relationship Graph Agent focusing on discovering and representing relationships between data, while the Data Modeling Agent focuses on using these relationships to create and optimize models. The two agents complement each other and work together to build a complete 6G network data model. They have a feedback relationship, with the Data Relationship Graph Agent detecting new relationships or changes in existing relationships and promptly notifying the Data Modeling Agent to update the model. Conversely, the Data Modeling Agent may also adjust the data relationship graph based on the results of model predictions. The two agents have a joint decision-making relationship, and in some cases, they may need to work together to make decisions, such as when optimizing the network, the Data Modeling Agent may propose optimization solutions, and the Data Relationship Graph Agent is responsible for verifying whether these solutions take into account all relevant data relationships.

[0060] The Data Modeling Agent and the Data Relationship Graph Agent work closely together in constructing the dynamic data model of the 6G network, with the former using the data relationships provided by the latter to create more accurate models, and the latter relying on the former's understanding of data relationships to keep the graph up-to-date.

[0061] This system can be applied in various scenarios such as network fault diagnosis, performance optimization, security monitoring, etc. By using digital twin technology and data updates, the management and operation efficiency of the 6G network can be greatly improved.

[0062] Figure 2 A flowchart of a method for modeling an entity network based on an agent system is shown.

[0063] The agent system includes a first agent and a second agent that communicate with each other, the first agent is used to construct a data relationship graph of the entity network, and the second agent is used to generate a data model of the entity network based on the data relationship graph.

[0064] It also includes a third agent that monitors and coordinates the first and second agents, including but not limited to communication with external interfaces, resource scheduling, error detection and recovery, standardization of communication protocols, etc.

[0065] The method includes the following steps:

[0066] Step S110, the first intelligent agent perceives the multi-modal data of the entity network, and the multi-modal data includes data information of the entity network environment.

[0067] In some embodiments, specifically including:

[0068] The first intelligent agent perceives the network entity attribute data in the entity network environment by means of multi-modal sensors and an external vector database, perceives geometric data and network running data by means of the multi-modal sensors, and perceives rules by means of the external vector database.

[0069] The network entity attribute data is used to represent data information of the entity devices in the entity network, the geometric data is used to represent spatial positions of the entity devices in the entity network, the network running data is used to represent alarm, performance index and log information in the entity network, and the rules are used to represent management rules and quality of service requirements of the entity network.

[0070] For example, the network entity and attribute data includes device information, connection state, position information, etc. The geometric data includes spatial layout, relative position between devices, etc. The network running data includes alarm, performance index and log, etc. The rules include network management rules and quality of service requirements, etc.

[0071] The 6G network global data is acquired by the way that the intelligent agent perceives the environment, which greatly improves the completeness, accuracy and timeliness of data acquisition.

[0072] Step S120, the first intelligent agent establishes a data relationship graph of the entity network according to the multi-modal data, and sends the data relationship graph to the second intelligent agent, and the data relationship graph is used to represent the relationship type between the network entity data and the network running data in the entity network.

[0073] In some embodiments, specifically including:

[0074] Step S121, the first intelligent agent processes the multi-modal data according to a preset data processing prompt word by means of a large model, to obtain full-amount entities and network running data of the entity network.

[0075] Step S122, the full-amount entities and network running data are subjected to relationship identification by means of a large model according to a preset relationship identification prompt word.

[0076] Step S123, a data relationship graph of the entity network is generated according to the identification result, and the data relationship graph of the entity network is a triple graph composed of network entity data and network running data in the entity network as nodes and relationship types between the network entity data and the network running data as edges.

[0077] In some more specific embodiments, step S121 specifically includes:

[0078] The first agent utilizes a large model to judge missing values of the multi-modal data through a first preset data processing prompt word, and processes the multi-modal data through a second preset data processing prompt word according to a judgment result, the first preset data processing prompt word is used to prompt a data value missing judgment standard, and the second preset data processing prompt word is used to prompt a processing mode of the multi-modal data with missing data values.

[0079] The first agent utilizes a large model to judge repeated items of the multi-modal data through a third preset data processing prompt word, and processes the multi-modal data through a fourth preset data processing prompt word according to a judgment result, the third preset prompt word is used to prompt a repeated item judgment standard, and the fourth preset data processing prompt word is used to prompt a processing mode of the multi-modal data with repeated items.

[0080] The first agent utilizes a large model to judge abnormal values of the multi-modal data through a fifth preset data processing prompt word, and processes the multi-modal data through a sixth preset data processing prompt word according to a judgment result, the fifth preset prompt word is used to prompt an abnormal value judgment standard, and the sixth preset data processing prompt word is used to prompt a processing mode of the multi-modal data with abnormal values.

[0081] Specifically, data preprocessing mainly includes processing missing values, deleting repeated items, processing abnormal values, etc. The prompt word method is used, and the details are as follows:

[0082] The prompt words for processing missing values are as follows:

[0083] 1. How to judge data value missing? When the data value is NULL or empty, it means that there is data value missing.

[0084] 2. How to process missing values? The missing values of numerical type are processed by moving average, the missing values of text type are processed by mode, and the missing values of time type are processed by arithmetic difference.

[0085] The prompt words for processing repeated items are as follows:

[0086] 1. How to judge data repeated items? Two data records are exactly the same.

[0087] 2. How to process repeated items? Keep the first line or any line.

[0088] The prompt words for processing abnormal values are as follows:

[0089] 1. How to judge abnormal values? Invalid data items or outliers, invalid data items such as filling numbers or other texts in date items, outliers such as filling 2099-12-30 in date.

[0090] 2、How to deal with outliers? Numerical outliers use moving average, text outliers use mode, and time outliers use arithmetic difference.

[0091] In some more specific embodiments, step S122 specifically includes:

[0092] The first intelligent agent uses a first preset relationship identification prompt word to determine the containing relationship of the full entity and network operation data, and according to the determination result, a second preset relationship identification prompt word is used to integrate the data with a containing relationship in the full entity and network operation data. The first preset relationship identification prompt word is used to prompt the containing relationship determination standard, and the second preset relationship identification prompt word is used to prompt the processing method of the data with a containing relationship.

[0093] The first intelligent agent uses a third preset relationship identification prompt word to determine the association relationship of the full entity and network operation data, and according to the determination result, a fourth preset relationship identification prompt word is used to model the relationship of the data with an association relationship in the full entity and network operation data. The third preset relationship identification prompt word is used to prompt the association relationship determination standard, and the fourth preset relationship identification prompt word is used to prompt the way of modeling the relationship of the data with an association relationship.

[0094] The first intelligent agent uses a fifth preset relationship identification prompt word to determine the causal relationship of the full entity and network operation data, and according to the determination result, a sixth preset relationship identification prompt word is used to jointly model the data with a causal relationship in the full entity and network operation data. The fifth preset relationship identification prompt word is used to prompt the causal relationship determination standard, and the sixth preset relationship identification prompt word is used to prompt the processing method of the data with a causal relationship.

[0095] Specifically, the containing relationship is processed, and the prompt words are as follows:

[0096] 1. How to determine the data containing relationship? Determine the cosine similarity of two types of data. For data categories exceeding a certain cosine similarity threshold, determine the containing relationship.

[0097] 2. How to process containing relationship data? Integrate the containing relationship data, eliminate redundant information, and provide consistent and accurate data model representation.

[0098] The association relationship is processed, and the prompt words are as follows:

[0099] 1. How to determine the data association relationship? Determine whether two types of data have the same attributes. If so, determine the association relationship and subdivide the specific association relationship type.

[0100] 2. How to process data association relationship? Model the relationship of the data, such as the network topology relationship of network devices, boards, ports, and circuits.

[0101] The prompt words for processing the causal relationship are as follows:

[0102] 1. How to judge the causal relationship of data? Judge whether two types of data have a causal relationship, and if so, judge it as a correlation relationship and subdivide the specific correlation relationship type.

[0103] 2. How to process causal relationship data? Joint modeling of causal relationship data, causal relationship between network performance degradation and network alarms.

[0104] In some embodiments, step S123 specifically comprises:

[0105] The first intelligent agent generates a complete relationship graph based on the data correlation relationship, performs automatic pruning based on reinforcement learning RLHF of human feedback, forms a final data relationship graph, and dynamically updates the data according to the data.

[0106] Step S130, the second intelligent agent generates a data model of the entity network according to the data relationship graph.

[0107] In some embodiments, the implementation of this step can include:

[0108] The second intelligent agent generates a basic data model and a functional data model of the entity network by calling the database query tool and the code generation tool in the tool according to the input data relationship graph, the basic data model is used to represent the basic data model of the entity device in the entity network, and the functional data model is used to represent the running logic of each entity device in the entity network.

[0109] Based on the intelligent agent and the prompt word, the 6G network global data is preprocessed, data relationship is identified, and relationship graph is constructed and dynamically updated, which provides accurate reference basis for subsequent data modeling.

[0110] Corresponding to the above method provided by the application, Figure 3 A structure schematic diagram of an entity network modeling device based on an intelligent agent system is shown.

[0111] The device is applied to an intelligent agent system, and the intelligent agent system includes a first intelligent agent and a second intelligent agent in communication with each other, the first intelligent agent is used to construct a data relationship graph of an entity network, and the second intelligent agent is used to generate a data model of the entity network according to the data relationship graph.

[0112] The device comprises:

[0113] The first processing module is configured to enable the first intelligent agent to perceive multi-modal data of the entity network, and the multi-modal data includes data information of an entity network environment.

[0114] The first processing module is specifically configured to perceive, by the first intelligent agent, network entity attribute data in an entity network environment with the aid of a multi-modal sensor and an external vector database, perceive geometric data and network operation data through the multi-modal sensor, and perceive rules through the external vector database.

[0115] The network entity attribute data is used to represent data information of entity devices in the entity network, the geometric data is used to represent spatial positions of the entity devices in the entity network, the network operation data is used to represent alarms, performance indicators, and log information in the entity network, and the rules are used to represent management rules and quality of service requirements of the entity network.

[0116] For example, the network entity and attribute data includes device information, connection states, and position information. The geometric data includes spatial layouts and relative positions between devices. The network operation data includes alarms, performance indicators, and logs. The rules include network management rules and quality of service requirements.

[0117] In this way, the 6G network global data is acquired through the intelligent agent perceiving the environment, and the completeness, accuracy, and timeliness of data acquisition are greatly improved.

[0118] The second processing module is specifically configured to establish, by the first intelligent agent, a data relationship graph of the entity network according to the multi-modal data, and send the data relationship graph to the second intelligent agent. The data relationship graph is used to represent relationship types between network entity data and network operation data in the entity network.

[0119] In some embodiments, the second processing module specifically includes a first processing unit, a second processing unit, and a third processing unit.

[0120] The first processing unit is configured to process, by the first intelligent agent, the multi-modal data according to a preset data processing prompt word with the aid of a large model, to obtain full-amount entities and network operation data of the entity network.

[0121] The second processing unit is configured to perform relationship identification on the full-amount entities and network operation data according to a preset relationship identification prompt word with the aid of the large model.

[0122] The third processing unit is configured to generate a data relationship graph of the entity network according to the identification result. The data relationship graph of the entity network is a triple graph composed of network entity data and network operation data in the entity network as nodes and relationship types between the network entity data and the network operation data as edges.

[0123] In some more specific embodiments, the first processing unit, specifically for the first intelligent agent, utilizes the large model to perform missing value judgment on the multi-modal data through a first preset data processing prompt word, and according to the judgment result, processes the multi-modal data through a second preset data processing prompt word. The first preset data processing prompt word is used to prompt the data value missing judgment standard, and the second preset data processing prompt word is used to prompt the processing mode of the multi-modal data with data value missing.

[0124] The first intelligent agent utilizes the large model to perform duplicate item judgment on the multi-modal data through a third preset data processing prompt word, and according to the judgment result, processes the multi-modal data through a fourth preset data processing prompt word. The third preset prompt word is used to prompt the duplicate item judgment standard, and the fourth preset data processing prompt word is used to prompt the processing mode of the multi-modal data with duplicate items.

[0125] The first intelligent agent utilizes the large model to perform abnormal value judgment on the multi-modal data through a fifth preset data processing prompt word, and according to the judgment result, processes the multi-modal data through a sixth preset data processing prompt word. The fifth preset prompt word is used to prompt the abnormal value judgment standard, and the fourth preset data processing prompt word is used to prompt the processing mode of the multi-modal data with abnormal values.

[0126] In some more specific embodiments, the second processing unit, specifically for the first intelligent agent, utilizes a first preset relationship identification prompt word to perform inclusion relationship judgment on the full-amount entity and network running data, and according to the judgment result, integrates the data with inclusion relationship in the full-amount entity and network running data through a second preset relationship identification prompt word. The first preset relationship identification prompt word is used to prompt the inclusion relationship judgment standard, and the second preset relationship identification prompt word is used to prompt the processing mode of the data with inclusion relationship.

[0127] The first intelligent agent utilizes a third preset relationship identification prompt word to perform association relationship judgment on the full-amount entity and network running data, and according to the judgment result, performs relationship modeling on the data with association relationship in the full-amount entity and network running data through a fourth preset relationship identification prompt word. The third preset relationship identification prompt word is used to prompt the association relationship judgment standard, and the fourth preset relationship identification prompt word is used to prompt the relationship modeling mode of the data with association relationship.

[0128] The first intelligent agent utilizes a fifth preset relationship identification prompt word to perform causal relationship judgment on the full-amount entity and network running data, and according to the judgment result, performs joint modeling on the data with causal relationship in the full-amount entity and network running data through a sixth preset relationship identification prompt word. The fifth preset relationship identification prompt word is used to prompt the causal relationship judgment standard, and the sixth preset relationship identification prompt word is used to prompt the processing mode of the data with causal relationship.

[0129] The third processing unit is specifically configured to generate a complete relationship graph by the first agent according to the data correlation relationship, perform automatic pruning based on reinforcement learning of human feedback (RLHF), form a final data relationship graph, and dynamically update the data relationship graph according to the data.

[0130] The third processing module is configured to generate a data model of the entity network by the second agent according to the data relationship graph.

[0131] In some embodiments, the third processing module is specifically configured to generate a basic data model and a functional data model of the entity network by the second agent according to the input data relationship graph by calling a database query tool and a code generation tool in the tool, the basic data model is used to represent a basic data model of an entity device in the entity network, and the functional data model is used to represent a running logic of each entity device in the entity network.

[0132] Corresponding to the above method provided by the application, Figure 4 A structural schematic diagram of an agent system is shown.

[0133] For modeling of an entity network, the agent system includes a first agent and a second agent that communicate with each other, the first agent is configured to construct a data relationship graph of the entity network, and the second agent is configured to generate a data model of the entity network according to the data relationship graph.

[0134] The first agent is configured to perceive multi-modal data of the entity network, and the multi-modal data includes data information of an environment of the entity network.

[0135] The first agent is specifically configured to perceive network entity attribute data in the entity network environment by the first agent with the aid of multi-modal sensors and an external vector database, perceive geometric data and network running data by the multi-modal sensors, and perceive rules by the external vector database.

[0136] The network entity attribute data is used to represent data information of an entity device in the entity network, the geometric data is used to represent a spatial position of the entity device in the entity network, the network running data is used to represent traffic information in the entity network, and the rules are used to represent management rules and quality of service requirements of the entity network.

[0137] For example, the network entity and attribute data include device information, connection state, position information, etc. The geometric data includes spatial layout, relative position between devices, etc. The network running data includes alarm, performance index, and log, etc. The rules include network management rules and quality of service requirements, etc.

[0138] Through the above mode, the 6G network global data is acquired in a manner that the intelligent agent perceives the environment, and the integrity, accuracy and timeliness of data acquisition are greatly improved.

[0139] The first intelligent agent is configured to establish a data relationship graph of the entity network according to the multi-modal data, and send the data relationship graph to the second intelligent agent, where the data relationship graph is used to represent the relationship types between the network entity data and the network operation data in the entity network.

[0140] In some embodiments, the first intelligent agent is configured to utilize a large model to process the multi-modal data according to a preset data processing prompt word, to obtain full-amount entities and network operation data of the entity network.

[0141] The large model is utilized to perform relationship identification on the full-amount entities and network operation data according to a preset relationship identification prompt word.

[0142] According to the identification result, a data relationship graph of the entity network is generated, where the data relationship graph of the entity network is a triple graph composed of network entity data and network operation data in the entity network as nodes, and relationship types between the network entity data and the network operation data as edges.

[0143] In some more specific embodiments, the first intelligent agent is specifically configured to utilize the large model to perform missing value judgment on the multi-modal data through a first preset data processing prompt word, and according to the judgment result, process the multi-modal data through a second preset data processing prompt word, where the first preset data processing prompt word is used to prompt a missing value judgment standard, and the second preset data processing prompt word is used to prompt a processing manner for the multi-modal data with missing data values.

[0144] The first intelligent agent is specifically configured to utilize the large model to perform duplicate item judgment on the multi-modal data through a third preset data processing prompt word, and according to the judgment result, process the multi-modal data through a fourth preset data processing prompt word, where the third preset prompt word is used to prompt a duplicate item judgment standard, and the fourth preset data processing prompt word is used to prompt a processing manner for the multi-modal data with duplicate items.

[0145] The first intelligent agent is specifically configured to utilize the large model to perform abnormal value judgment on the multi-modal data through a fifth preset data processing prompt word, and according to the judgment result, process the multi-modal data through a sixth preset data processing prompt word, where the fifth preset prompt word is used to prompt an abnormal value judgment standard, and the sixth preset data processing prompt word is used to prompt a processing manner for the multi-modal data with abnormal values.

[0146] In some more specific embodiments, the first intelligent agent is specifically configured to identify the containment relationship between the full-quantity entity and the network operation data by using a first preset relationship identification prompt word, and according to the judgment result, the data having the containment relationship in the full-quantity entity and the network operation data is integrated by using a second preset relationship identification prompt word. The first preset relationship identification prompt word is used to prompt the containment relationship judgment standard, and the second preset relationship identification prompt word is used to prompt the processing mode of the data having the containment relationship.

[0147] The first intelligent agent is specifically configured to identify the association relationship between the full-quantity and network operation entity data by using a third preset relationship identification prompt word, and according to the judgment result, the data having the association relationship in the full-quantity entity and the network operation data is subjected to relationship modeling by using a fourth preset relationship identification prompt word. The third preset relationship identification prompt word is used to prompt the association relationship judgment standard, and the fourth preset relationship identification prompt word is used to prompt the relationship modeling mode of the data having the association relationship.

[0148] The first intelligent agent is specifically configured to identify the causal relationship between the full-quantity and network operation entity data by using a fifth preset relationship identification prompt word, and according to the judgment result, the data having the causal relationship in the full-quantity entity and the network operation data is subjected to joint modeling by using a sixth preset relationship identification prompt word. The fifth preset relationship identification prompt word is used to prompt the causal relationship judgment standard, and the sixth preset relationship identification prompt word is used to prompt the processing mode of the data having the causal relationship.

[0149] The first intelligent agent is specifically configured to generate a complete relationship graph according to the data association relationship, perform automatic pruning based on human feedback reinforcement learning (RLHF), form a final data relationship graph, and dynamically update the data according to the data.

[0150] The second intelligent agent is specifically configured to generate a data model of the entity network according to the data relationship graph.

[0151] In some embodiments, the second intelligent agent is specifically configured to generate a basic data model and a functional data model of the entity network by calling a database query tool and a code generation tool in the tool according to the input data relationship graph. The basic data model is used to represent the basic data model of the entity device in the entity network, and the functional data model is used to represent the operation logic of each entity device in the entity network.

[0152] According to another aspect, embodiments also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method described above. Figure 2 The method described above.

[0153] According to an embodiment of still another aspect, there is also provided a computing device comprising a memory having executable code stored therein and a processor that, when executing the executable code, implements the method described above in connection with Figure 2 the method described above.

[0154] Those skilled in the art should be aware that, in the above one or more examples, the functions described in the present application can be implemented in hardware, software, firmware or any combination thereof. When implemented in software, the functions can be stored in a computer readable medium or transmitted as one or more instructions or code on a computer readable medium.

[0155] The above detailed description has further described the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above detailed description is only a specific implementation of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made on the basis of the technical solutions of the present application shall be included in the protection scope of the present application.

Claims

1. A method for modeling an entity network based on an agent system, characterized in that, The method is applied to an agent system, the agent system includes a first agent and a second agent in communication with each other, the first agent is configured to construct a data relationship graph of an entity network, and the second agent is configured to generate a data model of the entity network according to the data relationship graph, and the method comprises the following steps: The first agent perceives multi-modal data of the entity network, and the multi-modal data comprises data information of an environment of the entity network; The first agent establishes a data relationship graph of the entity network according to the multi-modal data, and sends the data relationship graph to the second agent, wherein the data relationship graph is used to represent a relationship type between network entity data and network operation data in the entity network; The second agent generates a data model of the entity network according to the data relationship graph; The second agent generates a data model of the entity network according to the data relationship graph, specifically comprising: The second agent generates a basic data model and a functional data model of the entity network by calling a database query tool and a code generation tool in the tool according to the input data relationship graph, wherein the basic data model is used to represent a basic data model of an entity device in the entity network, and the functional data model is used to represent an operation logic of each entity device in the entity network.

2. The method of claim 1, wherein, The first agent establishes a data relationship graph of the entity network according to the multi-modal data, specifically comprising: The first agent uses a large model to process the multi-modal data according to a preset data processing prompt word, to obtain full-amount entities and network operation data of the entity network; The large model is used to perform relationship identification on the full-amount entities and network operation data according to a preset relationship identification prompt word; According to the identification result, a data relationship graph of the entity network is generated, wherein the data relationship graph of the entity network is a triple graph with network entity data and network operation data in the entity network as nodes and a relationship type between each network entity data and network operation data as edges.

3. The method of claim 2, wherein, The first agent uses a large model to process the multi-modal data according to a preset data processing prompt word, to obtain full-amount entities and network operation data of the entity network, specifically comprising: The first agent uses a large model to perform missing value judgment on the multi-modal data through a first preset data processing prompt word, and according to the judgment result, the multi-modal data is processed through a second preset data processing prompt word, wherein the first preset data processing prompt word is used to prompt a missing value judgment standard, and the second preset data processing prompt word is used to prompt a processing mode for the multi-modal data with missing data values; The first agent uses a large model to perform duplicate item judgment on the multi-modal data through a third preset data processing prompt word, and according to the judgment result, the multi-modal data is processed through a fourth preset data processing prompt word, wherein the third preset data processing prompt word is used to prompt a duplicate item judgment standard, and the fourth preset data processing prompt word is used to prompt a processing mode for the multi-modal data with duplicate items; The first intelligent agent utilizes a large model to perform outlier value judgment on the multi-modal data through a fifth preset data processing prompt word, and according to a judgment result, processes the multi-modal data through a sixth preset data processing prompt word. The fifth preset data processing prompt word is used to prompt an outlier value judgment standard, and the sixth preset data processing prompt word is used to prompt a processing mode of multi-modal data with an outlier value.

4. The method of claim 2, wherein, The first intelligent agent utilizes a large model to perform relationship identification on the full-quantity entity and network running data according to a preset relationship identification prompt word, specifically including: The first intelligent agent utilizes a first preset relationship identification prompt word to perform a containment relationship judgment on the full-quantity entity and network running data, and according to a judgment result, integrates data with a containment relationship in the full-quantity entity and network running data through a second preset relationship identification prompt word. The first preset relationship identification prompt word is used to prompt a containment relationship judgment standard, and the second preset relationship identification prompt word is used to prompt a processing mode of data with a containment relationship. The first intelligent agent utilizes a third preset relationship identification prompt word to perform an association relationship judgment on the full-quantity entity and network running data, and according to a judgment result, performs relationship modeling on data with an association relationship in the full-quantity entity and network running data through a fourth preset relationship identification prompt word. The third preset relationship identification prompt word is used to prompt an association relationship judgment standard, and the fourth preset relationship identification prompt word is used to prompt a relationship modeling mode of data with an association relationship. The first intelligent agent utilizes a fifth preset relationship identification prompt word to perform a cause-and-effect relationship judgment on the full-quantity entity and network running data, and according to a judgment result, performs joint modeling on data with a cause-and-effect relationship in the full-quantity entity and network running data through a sixth preset relationship identification prompt word. The fifth preset relationship identification prompt word is used to prompt a cause-and-effect relationship judgment standard, and the sixth preset relationship identification prompt word is used to prompt a processing mode of data with a cause-and-effect relationship.

5. The method of claim 1, wherein, The first intelligent agent perceives multi-modal data of an entity network, specifically including: The first intelligent agent perceives network entity attribute data in an entity network environment by means of a multi-modal sensor and an external vector database, perceives geometric data and network running data through the multi-modal sensor, and perceives rules through the external vector database; The network entity attribute data is used to represent data information of an entity device in an entity network, the geometric data is used to represent a spatial position of an entity device in an entity network, the network running data is used to represent alarm, performance index and log information in an entity network, and the rules are used to represent management rules and service quality requirements of an entity network.

6. The method of claim 1, wherein, The intelligent agent system further includes a third intelligent agent, which is used to monitor the first intelligent agent and the second intelligent agent.

7. A modeling device for entity networks based on intelligent agent systems, characterized in that, The application is applied to an agent system, the agent system comprises a first agent and a second agent which communicate with each other, the first agent is used for constructing a data relationship graph of an entity network, and the second agent is used for generating a data model of the entity network according to the data relationship graph, and the device comprises: A first processing module is used for the first agent to perceive multi-modal data of the entity network, and the multi-modal data comprises data information of an entity network environment; A second processing module is used for the first agent to establish a data relationship graph of the entity network according to the multi-modal data, and the data relationship graph is sent to the second agent, and the data relationship graph is used for representing a relationship type between network entity data and network operation data in the entity network; A third processing module is used for the second agent to generate a data model of the entity network according to the data relationship graph; The third processing module is used for the second agent to generate a data model of the entity network according to the data relationship graph, and specifically comprises: The second agent generates a basic data model and a functional data model of the entity network by calling a database query tool and a code generation tool in a tool according to the input data relationship graph, the basic data model is used for representing a basic data model of an entity device in the entity network, and the functional data model is used for representing operation logic of each entity device in the entity network.

8. An agent system for modeling an entity network, characterized in that: The agent system comprises a first agent and a second agent which communicate with each other, the first agent is used for constructing a data relationship graph of an entity network, and the second agent is used for generating a data model of the entity network according to the data relationship graph; The first agent is used for perceiving multi-modal data of the entity network, and the multi-modal data comprises data information of an entity network environment; The first agent is used for establishing a data relationship graph of the entity network according to the multi-modal data, and the data relationship graph is sent to the second agent, and the data relationship graph is used for representing a relationship type between network entity data and network operation data in the entity network; The second agent is used for generating a data model of the entity network according to the data relationship graph; The second agent is used for generating a data model of the entity network according to the data relationship graph, and specifically comprises: The second agent generates a basic data model and a functional data model of the entity network by calling a database query tool and a code generation tool in a tool according to the input data relationship graph, the basic data model is used for representing a basic data model of an entity device in the entity network, and the functional data model is used for representing operation logic of each entity device in the entity network.

9. The system of claim 8, wherein, The agent system further comprises a third agent, and the third agent is used for monitoring the first agent and the second agent.

Images