Conflict mitigation method of artificial intelligence module, network device and readable storage medium

Abstract

The invention provides a conflict relieving method of an artificial intelligence module, network equipment and a readable storage medium, and relates to the technical field of communication. The method executed by a first network device comprises the following steps: when the first network device needs to deploy a first artificial intelligence (AI) decision module, obtaining performance change data corresponding to all AI decision modules from a second network device; determining first information according to the performance change data; the first information is used for indicating whether the second network equipment trains the target AI decision module or not; the target AI decision-making modules are part of AI decision-making modules in all AI decision-making modules; sending a training request to a second network device; the training request at least carries a first number of the first network device, a second number of the target AI decision module and the first information. According to the scheme, the frequency and time of joint training can be reduced, and energy consumption and deployment time are reduced.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a conflict mitigation method for an artificial intelligence module, a network device, and a readable storage medium. Background Art

[0002] The selection or configuration of the goals of artificial intelligence (AI) algorithms may lead to conflicting actions, and conflict mitigation is necessary. There are many ways to conduct joint training for conflict mitigation: one is to combine multiple agents for training, that is, combine the input states and output actions, and use a neural network for training. When there are many agents, the state and action scales are too large, which will increase the training time and reduce the effect; the other is multi-agent distributed training, which requires a lot of interaction between agents, and also requires a lot of training time and overhead. Summary of the invention

[0003] The purpose of this application is to provide a conflict mitigation method, a network device and a readable storage medium for an artificial intelligence module to solve the number and scope of joint training or distributed training in related technologies to reduce energy consumption and deployment time.

[0004] To achieve the above-mentioned purpose, an embodiment of the present application provides a conflict mitigation method for an artificial intelligence module, which is executed by a first network device and includes:

[0005] When the first network device needs to deploy the first artificial intelligence AI decision module, the performance change data corresponding to all AI decision modules are obtained from the second network device;

[0006] Determine first information according to the performance change data; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules;

[0007] Send a training request to the second network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module and the first information.

[0008] Optionally, determining the first information according to the performance change data includes:

[0009] Determining, according to the performance change data, whether each performance parameter corresponding to the performance change data is less than a corresponding first preset range;

[0010] If each of the performance parameters is greater than or equal to the corresponding first preset range, it is determined that the first information carries a first parameter, and a first AI decision module is deployed; the first parameter is used to instruct the second network device not to perform training;

[0011] If some of the performance parameters are less than the corresponding first preset range, it is determined that the first information carries a second parameter; the second parameter is used to instruct the second network device to perform hierarchical training on the target AI decision module; the hierarchical training is to divide all AI decision modules into sets, and jointly train the target AI decision module sets after the divided sets.

[0012] Optionally, the above method further includes:

[0013] When each performance parameter in the performance change data is less than the corresponding first preset range, it is determined that the first information carries a third parameter; the third parameter is used to indicate that attribute information corresponding to the first AI decision module is deleted in the first network device and / or the second network device.

[0014] Optionally, when the first information carries the second parameter, the method further includes:

[0015] Receiving performance result data sent by the second network device after performing hierarchical training on the target AI decision module;

[0016] Determine a decision message of the first network device according to the performance result data; the decision message carries the first number, the second number, and second information; the second information is used to indicate whether the first AI decision module is deployed in the first network device;

[0017] The decision message is sent to the second network device.

[0018] Optionally, after sending the decision message to the second network device, the method further includes:

[0019] Receiving target parameter data corresponding to all AI decision modules after hierarchical training of the target AI decision module sent by the second network device;

[0020] According to all the target parameter data, modify all AI decision modules of the first network device and deploy the first AI decision module;

[0021] The first AI decision module is sent to the decision executor.

[0022] Optionally, when the first network device needs to deploy the first artificial intelligence AI decision module, obtaining performance change data corresponding to all AI decision modules from the second network device includes:

[0023] Sending a verification message for verifying the first AI decision module to the second network device, wherein the verification message carries at least the first number and a number corresponding to the first AI decision module;

[0024] Receive a verification response message sent by the second network device, wherein the verification response message carries performance change data corresponding to all the AI ​​decision modules.

[0025] To achieve the above-mentioned purpose, an embodiment of the present application further provides a conflict mitigation method for an artificial intelligence module, which is executed by a second network device and includes:

[0026] Sending performance change data of all artificial intelligence AI decision modules to the first network device;

[0027] receiving a training request sent by the first network device; the training request carries at least a first number of the first network device, a second number of a target AI decision module, and first information; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules;

[0028] According to the training request, an operation corresponding to the first information is executed.

[0029] Optionally, the above method further includes:

[0030] receiving a decision message sent by the first network device; the decision message carries the first number, the second number, and second information; the second information is used to indicate whether a first AI decision module is deployed in the first network device;

[0031] When the second information carries a fourth parameter, deleting the attribute information corresponding to the first AI decision module in the operating environment of the second network device; the fourth parameter is used to indicate that the first AI decision module is not deployed in the first network device;

[0032] When the second information carries a fifth parameter, the second network device performs hierarchical training on the target AI decision module; the fifth parameter is used to indicate that the first AI decision module is deployed in the first network device; the hierarchical training is to divide all AI decision modules into sets, and jointly train the target AI decision module sets after the divided sets;

[0033] The target parameter data corresponding to all AI decision modules after the hierarchical training are sent to the first network device.

[0034] Optionally, the above method further includes:

[0035] Determine attribute information of the first AI decision module; the attribute information includes: priority and performance threshold constraint parameters of the first AI decision module;

[0036] According to the priorities of the multiple AI decision modules, the multiple AI decision modules are divided into a first decision module set and a second decision module set; the first decision module set is used to represent all decision modules with a priority not higher than that of the first AI decision module; the second decision module set is used to represent all decision modules with a priority higher than that of the first AI decision module;

[0037] According to the performance threshold constraint parameters corresponding to each AI decision module in the second decision module set, jointly train the first decision module set, and determine the target parameters for the performance change of the first decision module set during the training process;

[0038] When it is determined that the target parameter meets the preset threshold range, the hierarchical training of the target AI decision module is completed.

[0039] To achieve the above object, an embodiment of the present application further provides a network device, wherein the network device is a first network device, including:

[0040] A first acquisition module is used to acquire performance change data corresponding to all AI decision modules from a second network device when the first artificial intelligence AI decision module needs to be deployed on the first network device;

[0041] A first determination module is used to determine first information according to the performance change data; the first information is used to indicate whether the second network device trains a target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules;

[0042] The first sending module is used to send a training request to the second network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module and the first information.

[0043] To achieve the above object, an embodiment of the present application further provides a network device, wherein the network device is a second network device, including:

[0044] The second sending module is used to send the performance change data of all artificial intelligence AI decision modules to the first network device;

[0045] A first receiving module is used to receive a training request sent by the first network device; the training request carries at least a first number of the first network device, a second number of a target AI decision module, and first information; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules;

[0046] A first processing module is used to execute the operation corresponding to the first information according to the training request. To achieve the above purpose, an embodiment of the present application also provides a network device, including: a transceiver, a processor, a memory, and a program or instruction stored in the memory and executable on the processor; when the processor executes the program or instruction, the conflict mitigation method of the artificial intelligence module as described in any one of the above items is implemented.

[0047] To achieve the above objectives, an embodiment of the present application further provides a readable storage medium having a program or instruction stored thereon, which, when executed by a processor, implements the steps in the conflict mitigation method of the artificial intelligence module as described in any of the above items.

[0048] The beneficial effects of the above technical solution of the present application are as follows:

[0049] The solution of the present application is that when the first network device needs to deploy the first artificial intelligence AI decision module, the performance change data corresponding to all AI decision modules is obtained from the second network device. Through the performance change data, it can be determined whether there is a network conflict between the first AI decision module and multiple AI decision modules in the first network device; according to the performance change data, the first information is determined, and the first information is used to indicate whether the second network device trains the target AI decision module, and the target AI decision module is a part of all AI decision modules; a training request is sent to the second network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module, and the first information. In the above scheme, the first information is determined by the performance change data, and when the deployment requirements are not met, the training can be instructed by the first information to resolve the conflict between the AI ​​decision modules; it can also reduce the number and time required for joint training, reduce energy consumption and deployment time. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of the structure of the conflict mitigation architecture provided in an embodiment of the present application;

[0051] Figure 2 A schematic flow chart of a conflict mitigation method for an artificial intelligence module of a first network device according to an embodiment of the present application;

[0052] Figure 3A schematic flow chart of a conflict mitigation method for an artificial intelligence module of a second network device according to an embodiment of the present application;

[0053] Figure 4 The AI ​​decision-making conflict mitigation process based on hierarchical training provided in the embodiment of the present application;

[0054] Figure 5 A schematic diagram of the structure of a first network device provided in an embodiment of the present application;

[0055] Figure 6 A schematic diagram of the structure of a second network device provided in an embodiment of the present application;

[0056] Figure 7 A schematic diagram of the structure of a network device according to another embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to make the technical problems, technical solutions and advantages to be solved by the present application clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0058] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present application. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0059] In the various embodiments of the present application, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0060] It should be known that an entity may have multiple AI algorithms (AI decision modules) to optimize different objectives. For example, in network intelligence, an AI algorithm is designed to solve each network optimization problem. AI algorithms usually change one or more parameters to optimize specific metrics. For example, in a network, the control target of a wireless resource management use case can be a cell, user equipment (UE), or bearer, etc. Wireless resource management control content may include access control, bearer control, switching control, quality of service (QoS) control, resource allocation, etc. Because the selection or configuration of AI algorithm objectives may lead to conflicting actions, conflict mitigation is necessary. Conflict mitigation is the resolution of conflicting interactions between different AI algorithms. Network conflicts can be divided into three categories: direct conflicts, indirect conflicts, and implicit conflicts.

[0061] (1) Direct conflict: Conflict mitigation allows for direct observation of conflicts. For example, two or more AI algorithms request different settings for the same configuration of one or more parameters of a control target. (2) Indirect conflict: Conflicts cannot be observed directly, but it is possible to observe that there are certain dependencies between the parameters and resources of the AI ​​use case targets. (3) Implicit conflict: Conflicts cannot be observed directly, even if the dependencies between AI use cases are not obvious.

[0062] In order to solve the network conflicts that may exist when there are multiple AI algorithms (AI decision modules) optimizing different goals, refer to Figure 1 As shown, an embodiment of the present application provides a conflict mitigation architecture, including a management / control body 1, a digital twin body 2 and a decision execution body 3.

[0063] Among them, the management / control body 1 is a logical body that manages or controls the decision execution body 3, and can be deployed on entities such as network management equipment, edge cloud, core network element or base station in the network. The management / control body 1 is composed of an AI decision module conflict mitigation decision unit 101, an AI decision module subscription unit 102, and an AI decision module repository 103. The AI ​​decision module is a module that outputs intelligent decisions based on the received data, contains an AI model and related business logic functions, and can be regarded as an application (APP). The AI ​​decision modules are independent and may cause conflicts when deployed at the same time. The AI ​​decision module conflict mitigation decision unit 101 interacts with the digital twin 2 and outputs conflict mitigation decisions based on the information fed back by the digital twin 2. The AI ​​decision module subscription unit 102 interacts with the decision execution body 3 to establish a relationship between the AI ​​decision module and the decision execution body 3, so that the AI ​​decision module can control the decision execution body 3. The AI ​​decision module repository 103 stores multiple AI decision modules, including attribute information of the AI ​​decision module and data of the AI ​​decision module.

[0064] The digital twin 2 is a logical body of a digital decision-making execution body and a network environment, which is used to pre-verify the AI ​​decision module before actual deployment, and can be deployed on entities such as network management equipment, core network, edge cloud or base station in the network. The digital twin 2 is composed of an AI decision module management unit 201, an AI decision module data analysis unit 202 and an AI decision module operating environment 203. The AI ​​decision module management unit 201 has the ability to create, delete, and decide on joint training of AI decision modules, and stores the attribute information of the AI ​​decision modules in the digital twin 2. The AI ​​decision module data analysis unit 202 collects the operating data of each AI decision module in the AI ​​decision module operating environment 203 and analyzes the performance results. The AI ​​decision module operating environment 203 is where the management / control body 1 decides to start / start the AI ​​decision module to run / train in the twin environment.

[0065] The decision executor 3 is an executor for executing the AI ​​decision module, which can be deployed in the core network, base stations, etc., such as the forwarding module of the router, CU (centralized units) / DU (distributed units), etc., to execute decisions such as antenna adjustment and bandwidth allocation.

[0066] It should be noted that the above-mentioned conflict mitigation architecture can be applied to the following conflict mitigation method of the artificial intelligence module, for example, the first network device can be Figure 1 The management / control body 1 in the second network device can be Figure 1 Digital Twins in 2.

[0067] Reference Figure 2 As shown, an embodiment of the present application provides a conflict mitigation method for an artificial intelligence module, which is executed by a first network device and includes:

[0068] Step 21, when the first network device needs to deploy the first artificial intelligence AI decision module, obtain the performance change data corresponding to all AI decision modules from the second network device.

[0069] It should be noted that the first AI decision module is an AI decision module that needs to be deployed in the first network device, and can also be understood as an AI decision module to be deployed in the first network device. The first AI decision module is preferably an AI decision module.

[0070] Correspondingly, after the relevant parameters of the first AI decision module are sent to the second network device, the second network device needs to verify the first AI decision module in a verification environment, that is, the second network device runs all AI decision modules in the operating environment, and compares the parameter data of each AI decision module after running with the relevant parameters of the first AI decision module to determine the performance change data corresponding to all AI decision modules.

[0071] Here, the performance change data is the actual result generated after running in the operating environment of the second network device, and the result data after comparing with the relevant parameters of the first AI decision module. For example, if the actual result generated after running is A, and the relevant parameter of the first AI decision module is B, then the performance change data after comparison is the difference between (AB). For another example, if B is 90% and A is 85%, then the performance change data is -5%. Of course, the -5% here is the performance change parameter of one AI decision module among all AI decision modules.

[0072] In an embodiment of the present application, it can be determined based on the acquired performance change data whether deploying the first AI decision module will cause a conflict with all AI decision modules in the first network device, and data support is provided for subsequently determining whether all AI decision modules in the first network device need to be trained.

[0073] Step 22: determine first information based on the performance change data; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules.

[0074] In an embodiment of the present application, the performance change data can be used to determine whether the performance change data meets the deployment conditions, that is, whether the current performance change data will have deployment conflicts with multiple AI decision modules in the first network device. If there is no conflict, it can be determined that the first information carries a parameter, indicating that the first AI decision module can be directly deployed. In the case of a conflict, it can be determined that the first information carries another parameter, indicating that the target AI decision module needs to be trained. The above-mentioned first number and second number are both unique numbers.

[0075] Optionally, the first number is the number of the first network device, such as can be represented by M / C_ID. The second number is the number of the target AI decision module that needs to be trained, which can carry the unique identifier of each AI decision module that needs to be trained, and the unique identifier can be represented by AI_ID. For example, the number AI_ID can be identified by numbers (0, 1, 2, 3...).

[0076] Step 23, sending a training request to the second network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module and the first information.

[0077] The present application can send a training request to the second network device, and the first information in the training request is used to indicate whether the second network device should perform training, which can reduce the number and scope of models that need to be jointly trained to make AI decisions, and reduce energy consumption and deployment time.

[0078] It should be noted that the second network device is a network device with a digital twin function.

[0079] Optionally, the above step 21 includes:

[0080] Sending a verification message for verifying the first AI decision module to the second network device, wherein the verification message carries at least the first number and a number corresponding to the first AI decision module;

[0081] Receive a verification response message sent by the second network device, wherein the verification response message carries performance change data corresponding to all the AI ​​decision modules.

[0082] In the embodiment of the present application, the first network device starts the AI ​​decision module by monitoring network data or external triggering, and sends a verification message for verifying the first AI decision module to the second network device. The verification message can be represented by M / C-AI-TestRequest, and the data code of the first AI decision module is sent to the second network device. The M / C-AI-TestRequest includes the first number M / C_ID of the first network device, the attribute information AI_APP-Attribute of the first AI decision module, and the number corresponding to the first AI decision module.

[0083] Correspondingly, the second network device runs and collects the operating data of all AI decision modules in the operating environment, compares each operating data with the parameters corresponding to the first AI decision module, determines the performance result of each AI decision module, sets the performance result field of each AI decision module after verification, and determines the performance result field of all AI decision modules as the performance change data corresponding to each AI decision module.

[0084] Here, a verification response message sent by the second network device is received. The response message can be represented by M / C-AI-TestResponce. The verification response message carries the performance change data, the first number and the second number of all AI decision modules.

[0085] It should be noted that the performance change data AI_threshold_Flag indicates the performance change of all AI decision modules, that is, whether the corresponding function of each AI decision module drops below the preset threshold, whether the corresponding amplitude value of each AI decision module exceeds the preset amplitude range, etc. Each performance change data includes at least one of the following: the number of the AI ​​decision module whose performance drops below the preset threshold, the percentage of performance drop, etc.

[0086] Optionally, the verification message sent by the first network device to the second network device also carries attribute information of the first AI decision module; the attribute information includes: priority and performance threshold constraint parameters of the first AI decision module.

[0087] Here, the priority of the first AI decision module can be represented by the priority AI_priority, which represents the importance of the decision output by the first AI decision module, which can be comprehensively judged by factors such as duration, the number of related decision execution bodies, and the importance of the business, and can be obtained by weighting the above numerical values. The AI_priority can be sequentially identified by numbers or the like, and can be sequentially identified from large to small, from small to large, or sequentially identified according to a certain preset rule.

[0088] The performance threshold constraint parameter can be represented by AI_threshold, which is a preset value of the execution result, representing the acceptable minimum value of the performance (i.e., optimization target) result of the model output decision under different environments, and can be considered as a constraint condition. The performance threshold constraint parameter can be represented in a variety of ways. If the dimension of the optimization target of the first AI decision module includes multiple aspects, such as the optimization target is throughput and latency, the preset value of the performance threshold constraint parameter can be a vector, each optimization target of the vector has a preset value; or a number, which represents the weighted value of all optimization targets.

[0089] Combination Figure 1 As shown, all AI decision modules of the first network device store attribute information and data of each AI decision module in the AI ​​decision module repository 13 of the management / control body 1, which can be shown in the following Table 1. The AI ​​decision module repository 13 will update relevant data when a new AI decision module arrives or an existing AI decision module changes.

[0090] Table 1: AI decision module repository of management / control body A

[0091] AI_ID AI_priority AI_threshold AI_location 1 2 [2，6] C:\Users\cmcc\AImodel / 1 2 1 5 C:\Users\cmcc\AImodel / 2

[0092] Among them, AI_location represents the storage location of each AI decision module code or data.

[0093] Optionally, the above step 22 includes:

[0094] Determining, according to the performance change data, whether each performance parameter corresponding to the performance change data is less than a corresponding first preset range;

[0095] If each of the performance parameters is greater than or equal to the corresponding first preset range, it is determined that the first information carries a first parameter, and a first AI decision module is deployed; the first parameter is used to instruct the second network device not to perform training;

[0096] If some of the performance parameters are less than the corresponding first preset range, it is determined that the first information carries a second parameter; the second parameter is used to instruct the second network device to perform hierarchical training on the target AI decision module; the hierarchical training is used to instruct all AI decision modules to be divided into sets, and the target AI decision module sets after the divided sets are jointly trained.

[0097] In the present application, the first preset range can indicate whether each performance parameter meets the threshold range for deploying the first AI decision-making module. If the performance change data includes three performance parameters, the first preset range corresponding to each performance parameter can be analyzed and compared respectively to determine the analysis result of each performance parameter, and multiple analysis results can be determined as performance change data.

[0098] In one implementation, if each performance parameter is greater than or equal to the corresponding first preset range, it means that each performance parameter meets the preset requirements, that is, the first network device meets the conditions for deploying the first AI decision-making module, then it is determined that the first information carries the first parameter, and the first information can be represented by a field, such as by the AI_Train field. The AI_Train field is set with a first parameter False. The first parameter False indicates that the conflict between the first AI decision module and other AI decision modules is small, and no training is required. The first network device chooses to deploy the first AI decision module.

[0099] In another implementation, if some performance parameters, but not all performance parameters, are less than the corresponding first preset range, it indicates that a network conflict may occur, that is, the performance change data shows that the performance drops to the threshold, indicating that the first AI decision module has a large conflict with other AI decision modules, and the target AI decision module needs to be graded and trained. The graded training is used to indicate that all AI decision modules are divided into sets, and the target AI decision module set after the divided sets is jointly trained. Here, it is determined that the first information carries a second parameter, that is, the AI_Train field is set to True, and the second parameter is used to instruct the second network device to perform graded training on the target AI decision module.

[0100] Optionally, the above method further includes:

[0101] When each performance parameter in the performance change data is less than the corresponding first preset range, it is determined that the first information carries a third parameter; the third parameter is used to indicate that attribute information corresponding to the first AI decision module is deleted in the first network device and / or the second network device.

[0102] In the present application, the first preset range can indicate whether each performance parameter meets the threshold range for deploying the first AI decision module, and can be analyzed and compared with the first preset range corresponding to each performance parameter, and the analysis result of each performance parameter can be determined, and the multiple analysis results can be determined as performance change data. When each performance parameter in the performance change data is less than the corresponding first preset range, it means that each performance parameter in the performance change data does not meet the preset requirements, and it is determined that the first information carries a third parameter, and the third parameter is set to False, indicating that the target AI decision module does not need to be graded and trained, and the first AI decision module is not deployed. The attribute information of the first AI decision module in the first network device can be deleted, and the attribute information of the first AI decision module in the second network device can be deleted at the same time, and step 21 is re-executed to design the attribute information of the new AI decision module again.

[0103] Optionally, when the first information carries the second parameter, the method further includes:

[0104] Receiving performance result data sent by the second network device after performing hierarchical training on the target AI decision module;

[0105] Determine a decision message of the first network device according to the performance result data; the decision message carries the first number, the second number, and second information; the second information is used to indicate whether the first AI decision module is deployed in the first network device;

[0106] The decision message is sent to the second network device.

[0107] In an embodiment of the present application, when the second parameter is carried in the first information, it indicates that the second network device is required to perform hierarchical training on the target AI decision module, and the target AI decision module is a part of all AI decision modules. The first network device receives the performance result data sent by the second network device after performing hierarchical training on the target AI decision module. The performance result data may be a training success result or a training failure result. According to the training success result or the training failure result carried in the performance result data, the decision message of the first network device is determined. Of course, the second information in the decision message corresponds to an indication of deploying the first AI decision module or not deploying the first AI decision module.

[0108] Optionally, after sending the decision message to the second network device, the method further includes:

[0109] Receiving target parameter data corresponding to all AI decision modules after hierarchical training of the target AI decision module sent by the second network device;

[0110] According to all the target parameter data, modify all AI decision modules of the first network device and deploy the first AI decision module;

[0111] The first AI decision module is sent to the decision executor.

[0112] It should be noted that if the second information is represented by the field Decision, after sending the field Decision to the second network device, if the field Decision carries a False parameter, the second network device deletes the AI ​​decision module from the operating environment of the second network device, and will not receive the target parameter data corresponding to all AI decision modules sent by the second network device.

[0113] In the embodiment of the present application, if the field Decision carries the True parameter, the AI ​​decision module after hierarchical training is retained, and the first network device can receive the target parameter data corresponding to all AI decision modules after hierarchical training, and modify the parameters corresponding to all AI decision modules of the first network device through all the target parameter data. After the modification is completed, the first AI decision module can be deployed on the first network device. After the deployment of the first AI decision module is completed, the first network device subscribes to the decision executor, obtains the control and management rights of the decision executor, and sends the first AI decision module to the decision executor, which can ensure that there is no network conflict when the decision executor performs the corresponding action.

[0114] The method of the present application can reduce the number and scope of AI decision-making modules that require joint training, reduce energy consumption and deployment time.

[0115] Reference Figure 3 As shown, the embodiment of the present application also provides a conflict mitigation method for an artificial intelligence module, which is executed by a second network device and includes:

[0116] Step 31: Send the performance change data of all artificial intelligence AI decision modules to the first network device.

[0117] In the embodiment of the present application, the first AI decision module is an AI decision module that needs to be deployed in the first network device. After receiving the relevant parameters of the first AI decision module, the second network device needs to verify the first AI decision module in a verification environment, that is, the second network device runs all AI decision modules in the running environment, and compares the parameter data after each AI decision module is run with the relevant parameters of the first AI decision module to determine the performance change data corresponding to all AI decision modules.

[0118] For example, if the actual result generated after running is A, and the relevant parameter of the first AI decision module is B, then the performance change data after comparison is the difference between (AB). For another example, if B is 90% and A is 85%, then the performance change data is -5%. Of course, the -5% here is the performance change parameter of one AI decision module among all AI decision modules.

[0119] Of course, when there are difference results corresponding to multiple parameters, they can be weighted and determined as a numerical representation according to the preset weights set accordingly. In this application, the performance change data of all AI decision modules are sent to the first network device, which is also used to determine whether the target AI decision module of the first network device needs subsequent hierarchical training.

[0120] Step 32, receiving a training request sent by the first network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module and the first information; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules.

[0121] Step 33: Execute the operation corresponding to the first information according to the training request.

[0122] In an embodiment of the present application, by receiving a training request, the training request carries a first number and a second number, and the first number and the second number are both unique numbers. Optionally, the first number is the number of the first network device, such as can be represented by M / C_ID. The second number is the number of the target AI decision module that needs to be trained, and can carry a unique identifier of each AI decision module that needs to be trained. The unique identifier can be represented by AI_ID, such as the number AI_ID can be identified by a number or the like (0,1,2,3...). According to the parameters carried by the first information, the present application can instruct the second network device to execute the information of deleting the first AI decision module, or to perform hierarchical training on the target AI decision module. The present application only performs hierarchical training when hierarchical training is required. When no training is required, the first AI decision module is directly deployed through the first network device, thereby reducing energy consumption and deployment time.

[0123] Optionally, the above step 31 includes:

[0124] Receiving a verification message sent by a first network device for verifying an artificial intelligence (AI) decision module; the verification message carries at least the first number and a number corresponding to the first AI decision module;

[0125] A verification response message is sent to the first network device; the verification response message carries the performance change data corresponding to all the AI ​​decision modules.

[0126] In an embodiment of the present application, a verification message sent by the first network is received, and the verification message can be represented by M / C-AI-TestRequest, and the data code of the AI ​​decision module sent by the first network device is received. The M / C-AI-TestRequest includes the first number M / C_ID of the first network device, the attribute information AI_APP-Attribute of the first AI decision module carried, and the number corresponding to the first AI decision module. The second network device runs all AI decision modules in the operating environment, collects the operating data of all AI decision modules during the operation process, and compares each operating data with the parameters corresponding to the first AI decision module, determines the performance result of each AI decision module, sets the performance result field of each AI decision module after verification, and determines the performance result field of all AI decision modules as the performance change data corresponding to each AI decision module.

[0127] The second network device sends a verification response message to the first network device, and the verification response message can be represented by M / C-AI-TestResponce, and the M / C-AI-TestResponce needs to include the first number M / C_ID, the number corresponding to the first AI decision module, and the performance change data of all AI decision modules. Here, the performance change data can be represented by AI_threshold_Flag.

[0128] It should be noted that the performance change data AI_threshold_Flag indicates the performance change of all AI decision modules, that is, whether the corresponding function of each AI decision module drops below the preset threshold, whether the corresponding amplitude value of each AI decision module exceeds the preset amplitude range, etc. Each performance change data includes at least one of the following: the number of the AI ​​decision module whose performance drops below the preset threshold, the percentage of performance drop, etc.

[0129] Optionally, the above step 33 includes:

[0130] The first information carries a first parameter, which is not processed by the second network device; the first parameter is used to instruct the second network device not to perform training;

[0131] The first information carries a second parameter, and the second network device performs hierarchical training on the target AI decision module; the second parameter is used to instruct the second network device to perform hierarchical training on the target AI decision module; the hierarchical training is to divide all AI decision modules into sets, and jointly train the target AI decision module sets after the divided sets.

[0132] In one implementation of the present application, the first information can be represented by a field, such as by an AI_Train field. The first information carries a first parameter, that is, the AI_Train field is set with a first parameter False. The first parameter False indicates that the conflict between deploying the first AI decision module and other AI decision modules is small, and no training is required. The first network device can deploy an AI decision module that meets the first parameter, so the first parameter can be used to instruct the second network device not to perform training.

[0133] In another implementation of the present application, the first information carries a second parameter, and a network conflict may occur. That is, the performance change data shows that the performance has dropped to a threshold, indicating that the first AI decision-making module has a large conflict with other AI decision-making modules and requires hierarchical training. That is, the AI_Train field is set with the second parameter True, and the second network device performs hierarchical training on the target AI decision module.

[0134] Optionally, the above method further includes:

[0135] The first information carries a third parameter, and the attribute information of the first AI decision module is deleted in the operating environment of the second network device; the third parameter is used to indicate that the attribute information corresponding to the first AI decision module is deleted in the second network device.

[0136] In the implementation method of the present application, the first information carries a third parameter, and the third parameter is set to False, indicating that there is no need to perform hierarchical training on the target AI decision module, and the first network device chooses not to deploy the first AI decision module. The attribute information of the first AI decision module in the second network device can be deleted, and the first network device can re-issue the attribute information of the new AI decision module.

[0137] In the embodiment of the present application, it is necessary to verify in the second network device whether there is a conflict with the started AI decision module. After the second network device receives the M / C-AI-Request message from the first network device to apply for the deployment of the first AI decision module, all AI decision modules in the operating environment are collected, and the operating data of all AI decision modules during the operation process are collected, and each operating data is compared with the parameters corresponding to the first AI decision module, and the performance change values ​​of each AI decision module are calculated and fed back to the first network device. If the conflict between the first AI decision module and other AI decision modules is large, and the first network device still decides to deploy the first AI decision module, it is necessary to start hierarchical training to alleviate the conflict.

[0138] Optionally, the above method further includes:

[0139] Determine attribute information of the first AI decision module; the attribute information includes: priority and performance threshold constraint parameters of the first AI decision module;

[0140] According to the priorities of the multiple AI decision modules, the multiple AI decision modules are divided into a first decision module set and a second decision module set; the first decision module set is used to represent all decision modules with a priority not higher than that of the first AI decision module; the second decision module set is used to represent all decision modules with a priority higher than that of the first AI decision module;

[0141] According to the performance threshold constraint parameters corresponding to each AI decision module in the second decision module set, jointly train the first decision module set, and determine the target parameters for the performance change of the first decision module set during the training process;

[0142] When it is determined that the target parameter meets the preset threshold range, the hierarchical training of the target AI decision module is completed.

[0143] It should be noted that the priority of the first AI decision module can be represented by the priority AI_priority, which represents the importance of the decision output by the AI ​​decision module, which can be comprehensively judged by factors such as duration, the number of related decision execution bodies, and the importance of the business, and can be obtained by weighting the above numerical values. The AI_priority can be sequentially identified by numbers or the like, and can be sequentially identified from large to small, from small to large, or sequentially identified according to a certain preset rule. The performance threshold constraint parameter can be represented by AI_threshold, which is a preset value of the execution result, representing the acceptable minimum value of the performance (i.e., optimization target) result of the model output decision under different environments, and can be considered as a constraint condition. The performance threshold constraint parameter can be represented by multiple ways. If the dimension of the optimization target of the first AI decision module contains multiple aspects, such as the optimization target is throughput and latency, the preset value of the performance threshold constraint parameter can be a vector, each optimization target of the vector has a preset value; or a number, which represents the weighted value of all optimization targets.

[0144] Here, the attribute information of all AI decision modules sent by the first network device is determined and can be stored in the second network device, and its storage form can be as shown in Table 2 below.

[0145] Table 2:

[0146] AI_ID M / C_ID AI_priority AI_threshold AI_location A-1 A 2 [2，6] C:\Users\cmcc\AImodel / A1 A-2 A 1 5 C:\Users\cmcc\AImodel / A2

[0147] In the present application, when the second network device performs hierarchical training, all AI decision modules in the above Table 2 can be divided into a first decision module set N and a second decision module set M according to the priority of the first AI decision module. The second decision module set M is all decision modules with a higher priority than the first AI decision module, and the first decision module set N is all decision modules with a priority not higher than the first AI decision module. Taking the performance threshold corresponding to each AI decision module in the second decision module set M as a constraint, the AI ​​decision modules in the first decision module set N are jointly trained to complete the hierarchical training of the target AI module.

[0148] For example, the multiple AI decision modules are 10 AI decision modules, and their priorities are divided into multiple priorities, such as being divided into ten priorities, each AI decision module has a different priority; or being divided into four priorities, among which some AI decision modules have the same priority, etc. Here, an example is taken with three priorities, such as being divided into a, b, and c, and their priorities decrease in sequence. The priority of the first AI decision module is n. If the priorities of a and b are higher than n, and the priority of c is not higher than n, then the AI ​​decision modules corresponding to the priorities of a and b are divided into the second decision module set M, and the remaining AI decision modules are divided into the first decision module set N; if there are only three AI decision modules in the first decision module set N, then these three AI decision modules need to be jointly trained, and there is no need to train the second decision module set M. In this way, all AI decision modules are divided into sets, and the target AI decision module set after the divided set (such as the first decision module set N whose priority is not higher than the first AI decision module) is jointly trained, that is, the hierarchical training method, which can solve the problem of conflicts in AI decision modules.

[0149] It should be noted that the AI ​​decision modules in set N are trained in stages. When the actions output during the training process cause the performance of the AI ​​decision modules in set M to drop to a threshold, a larger penalty factor is fed back to ensure the performance of the AI ​​decision modules in set M. The penalty factor here can be understood as a correction parameter used to ensure the performance of the AI ​​decision modules in set M.

[0150] It should also be noted that it is assumed that the state dimension of each AI decision-making module is S and the action dimension is F. If the hierarchical training method is not adopted, the state dimension of the neural network after the merger is (|M|+|N|)*S, and the action dimension is (|M|+|N|)*F. In this application, if the hierarchical training method is adopted, the state dimension is |N|*S, and the action dimension is |N|*F. It can be seen that the complexity of the neural network can be significantly reduced, and the training time and overhead can be reduced.

[0151] Optionally, the first information carries a second parameter, and the second network device performs hierarchical training on the target AI decision module, including:

[0152] When the first information carries the second parameter, the second network device starts hierarchical training;

[0153] The result after the hierarchical training is set as performance result data, and the performance result data is sent to the first network device.

[0154] In an embodiment of the present application, the first information is represented by the AI_Train field, and the first information carries a second parameter, that is, the second parameter True is set in the AI_Train field. If the AI_Train field is set to True, the second network device starts hierarchical training; the second network device sets the training result as the post-training performance result data and sends it to the first network device.

[0155] Correspondingly, the first network device decides whether to deploy the first AI decision module based on the result, and sets a decision message MC-AI-Decision, which must include the first number M / C_ID of the first network device, the second number AI_ID of the target AI decision module, and the field Decision of whether to deploy, and sends it to the second network device.

[0156] Optionally, the above method further includes:

[0157] receiving a decision message sent by the first network device; the decision message carries the first number, the second number, and second information; the second information is used to indicate whether a first AI decision module is deployed in the first network device;

[0158] When the second information carries a fourth parameter, deleting the attribute information corresponding to the first AI decision module in the operating environment of the second network device; the fourth parameter is used to indicate that the first AI decision module is not deployed in the first network device;

[0159] When the second information carries a fifth parameter, the second network device performs hierarchical training on the target AI decision module; the fifth parameter is used to indicate that the first AI decision module is deployed in the first network device;

[0160] The target parameter data corresponding to all AI decision modules after hierarchical training are sent to the first network device.

[0161] In one implementation of the present application, a decision message sent by a first network device is received, such as represented by an MC-AI-Decision message; the MC-AI-Decision message carries a first number M / C_ID, a second number AI_ID and second information; the second information can be represented by Decision, and the Decision is used to indicate whether the first AI decision module is deployed in the first network device; the second information Decision carries a fourth parameter, and the fourth parameter is used to indicate that the first AI decision module is not deployed in the first network device, and to delete the attribute information of the first AI decision module in the second network device, that is, when the fourth parameter is represented as False, the attribute information of the first AI decision module is deleted in the operating environment of the second network device.

[0162] In another implementation of the present application, if the second information Decision carries a fifth parameter and the fifth parameter is expressed as True, the second network device retains the first AI decision module, and the second network device performs hierarchical training on the target AI decision module, and sends the parameter data corresponding to all modified AI decision module codes corresponding to all AI decision modules after the hierarchical training to the first network device, so that the first network device can modify the parameters and then deploy the first AI decision module, thereby avoiding network conflicts when the first network device is deployed.

[0163] The present application also provides a process for mitigating conflicts between AI decisions based on hierarchical training, referring to Figures 1 to 4 As shown, the first network device is Figure 1 The management / control body 1 in the figure and the second network device are digital twins 2. The interaction process between the three is as follows:

[0164] Step 1: The management / control body decides to start the AI ​​decision module by monitoring network data or external triggers.

[0165] Step 2: The AI ​​decision module conflict mitigation decision function of the management / control body sends a verification request M / C-AI-TestRequest for the AI ​​decision module to the AI ​​decision module management function of the digital twin, and sends the data code of the AI ​​decision module from the AI ​​decision module repository to the AI ​​decision module management function.

[0166] Step 3: The AI ​​decision module management function of the digital twin deploys the AI ​​decision module to the AI ​​decision module operating environment and starts running.

[0167] Step 4: The AI ​​decision module data analysis function of the digital twin collects the operating data of each AI decision module in the AI ​​decision module operating environment, analyzes the performance results, and sets the verified performance result data.

[0168] Step 5: The digital twin sends a verification response M / C-AI-TestResponce to the conflict mitigation decision function of the AI ​​decision module of the management / control body, which must include the management / control body number M / C_ID, the AI ​​decision module number AI_ID and the verified performance result data AI_threshold_Flag.

[0169] The performance result data after verification AI_threshold_Flag indicates the performance change of all AI decision modules in the running environment after the new deployment, that is, whether it has dropped below the threshold and the extent of the drop. This field contains at least one of the following: the number of the AI ​​decision module whose performance has dropped below the threshold, the percentage of performance drop, etc.

[0170] In step 6, the conflict mitigation decision function of the AI ​​decision module of the management / control body decides whether joint training is required, and sends a MC-AI-TrainRequest message to the AI ​​decision module management function of the digital twin, which must include the management / control body number M / C_ID, the AI ​​decision module number AI_ID and the training field AI_Train. The AI_Train field indicates whether the management / control body requires the digital twin to jointly train the AI ​​decision module.

[0171] The basis for judging whether to conduct joint training can be as follows:

[0172] 1) If the AI_threshold_Flag field received by the management / control body shows that the performance has not dropped to the threshold, it means that the AI ​​decision module has little conflict with other AI decision modules and does not need training, and the AI_Train field is set to False;

[0173] 2) If the AI_threshold_Flag field received by the management / control body shows that the performance has dropped to the threshold, it means that the AI ​​decision module has a large conflict with other AI decision modules. You can choose not to deploy (delete) the module and set the AI_Train field to False;

[0174] 3) If the AI_threshold_Flag field received by the management / control body shows that the performance has dropped to the threshold, it means that the AI ​​decision module has a large conflict with other AI decision modules and requires joint training. The AI_Train field is set to True;

[0175] In a specific embodiment, if the AI_Train field is set to False, the management / control body deletes / starts the AI ​​decision module; if it is started (the AI ​​decision module has little conflict with other AI decision modules), the AI ​​decision module subscription function of the management / control body sends a subscription request to the decision executor to obtain management / control rights over the decision executor, so that the decision output by the AI ​​decision module can be sent to the decision executor.

[0176] In another specific embodiment, if the AI_Train field is set to True, the digital twin starts hierarchical training; the digital twin sets the training result as the post-training performance result data and sends it to the management / control body. The management / control body decides whether to deploy the AI ​​decision module based on the result, and sets the MC-AI-Decision message, which must include the management / control body's number M / C_ID, the AI ​​decision module's number AI_ID, and the field Decision of whether to deploy, and sends it to the digital twin.

[0177] If Decision is False, the digital twin deletes the AI ​​decision module from the AI ​​decision module operating environment.

[0178] If Decision is True, the digital twin retains the AI ​​decision module and sends all modified AI decision module codes to the management / control body; the management / control body deploys the modified AI decision module. The AI ​​decision module subscription function of the management / control body sends a subscription request to the decision executor to obtain management / control rights over the decision executor, so that the decisions output by the AI ​​decision module can be sent to the decision executor.

[0179] To summarize, the solution of the present application utilizes digital twins to achieve conflict mitigation in digital networks. It can intuitively discover various conflicts, including implicit conflicts, provide a reliable basis for conflict mitigation, and reduce the risk of trial and error. It also sets priorities and thresholds for AI decisions to minimize the number and scope of models that need to be jointly trained to make AI decisions, thereby reducing energy consumption and deployment time.

[0180] Reference Figure 5 As shown, the embodiment of the present application further provides a network device, wherein the network device is a first network device, including:

[0181] The first acquisition module 51 is used to acquire the performance change data corresponding to all AI decision modules from the second network device when the first artificial intelligence AI decision module needs to be deployed on the first network device;

[0182] A first determination module 52 is used to determine first information according to the performance change data; the first information is used to indicate whether the second network device trains a target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules;

[0183] The first sending module 53 is used to send a training request to the second network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module and the first information.

[0184] Optionally, the first determining module 52 includes:

[0185] A first determining unit, configured to determine, based on the performance change data, whether each performance parameter corresponding to the performance change data is less than a corresponding first preset range;

[0186] a second determination unit, configured to determine that the first information carries a first parameter and deploy a first AI decision module if each of the performance parameters is greater than or equal to a corresponding first preset range; the first parameter is used to instruct the second network device not to perform training;

[0187] The third determination unit is used to determine that the first information carries a second parameter if some of the performance parameters are less than the corresponding first preset range; the second parameter is used to instruct the second network device to perform hierarchical training on the target AI decision module; the hierarchical training is to divide all AI decision modules into sets, and jointly train the target AI decision module sets after the divided sets.

[0188] Optionally, the above network device further includes:

[0189] The second determination module is used to determine that the first information carries a third parameter when each performance parameter in the performance change data is less than the corresponding first preset range; the third parameter is used to indicate that the attribute information corresponding to the first AI decision module is deleted in the first network device and / or the second network device.

[0190] Optionally, the above network device further includes:

[0191] A third receiving module is used to receive performance result data sent by the second network device after performing hierarchical training on the target AI decision module;

[0192] A third determination module, configured to determine a decision message of the first network device according to the performance result data; the decision message carries the first number, the second number, and second information; the second information is used to indicate whether the first AI decision module is deployed in the first network device;

[0193] The third sending module is used to send the decision message to the second network device.

[0194] Optionally, the above network device further includes:

[0195] A fourth receiving module, configured to receive target parameter data corresponding to all AI decision modules after hierarchical training of the target AI decision module, sent by the second network device;

[0196] A second processing module, configured to modify all AI decision modules of the first network device according to all the target parameter data, and then deploy the first AI decision module;

[0197] The third processing module is used to send the first AI decision module to the decision executor.

[0198] Optionally, the first acquisition module 51 includes:

[0199] A first sending unit, configured to send a verification message for verifying the first AI decision module to the second network device, wherein the verification message carries at least the first number and a number corresponding to the first AI decision module;

[0200] The first receiving unit is used to receive a verification response message sent by the second network device, wherein the verification response message carries performance change data corresponding to all the AI ​​decision modules.

[0201] Among them, the implementation embodiments of the conflict mitigation method of the above-mentioned artificial intelligence module are all applicable to the embodiments of the network device and can achieve the same technical effect.

[0202] Reference Figure 6 As shown, the embodiment of the present application further provides a network device, which is a second network device, including:

[0203] The second sending module 61 is used to send the performance change data of all artificial intelligence AI decision modules to the first network device;

[0204] The first receiving module 62 is used to receive a training request sent by the first network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module and the first information; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules;

[0205] The first processing module 63 is used to execute the operation corresponding to the first information according to the training request.

[0206] Optionally, the above network device further includes:

[0207] a fifth receiving module, configured to receive a decision message sent by the first network device; the decision message carries the first number, the second number, and second information; the second information is used to indicate whether a first AI decision module is deployed in the first network device;

[0208] a fourth processing module, configured to delete the attribute information corresponding to the first AI decision module in the operating environment of the second network device when the second information carries a fourth parameter; the fourth parameter is used to indicate that the first AI decision module is not deployed in the first network device;

[0209] a fifth processing module, configured to, when the second information carries a fifth parameter, cause the second network device to perform hierarchical training on the target AI decision module; the fifth parameter is used to indicate that the first AI decision module is deployed in the first network device; the hierarchical training is to divide all AI decision modules into sets, and jointly train the target AI decision module sets after the divided sets;

[0210] The sixth processing module is used to send the target parameter data corresponding to all AI decision modules after hierarchical training to the first network device.

[0211] Optionally, the above network device further includes:

[0212] A fourth determination module, used to determine attribute information of the first AI decision module; the attribute information includes: a priority and a performance threshold constraint parameter of the first AI decision module;

[0213] a seventh processing module, configured to divide the plurality of AI decision modules into a first decision module set and a second decision module set according to the priorities of the plurality of AI decision modules; the first decision module set is configured to represent all decision modules with a priority not higher than that of the first AI decision module; and the second decision module set is configured to represent all decision modules with a priority higher than that of the first AI decision module;

[0214] an eighth processing module, configured to perform joint training on the first decision module set according to the performance threshold constraint parameters corresponding to each AI decision module in the second decision module set, and determine a target parameter for a performance change of the first decision module set during the training process;

[0215] The ninth processing module is used to complete the hierarchical training of the target AI decision module when determining that the target parameter meets the preset threshold range.

[0216] Among them, the implementation embodiments of the conflict mitigation method of the above-mentioned artificial intelligence module are all applicable to the embodiments of the network device and can achieve the same technical effect.

[0217] A network device according to another embodiment of the present application, such as Figure 7 As shown, it includes a transceiver 710, a processor 700, a memory 720, and a program or instruction stored in the memory 720 and executable on the processor 700; when the processor 700 executes the program or instruction, the conflict mitigation method applied to the artificial intelligence module is implemented.

[0218] The transceiver 710 is used to receive and send data under the control of the processor 700.

[0219] Among them, Figure 7 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 700 and memory represented by memory 720. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The transceiver 710 may be a plurality of components, i.e., including a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium. The processor 700 is responsible for managing the bus architecture and general processing, and the memory 720 may store data used by the processor 700 when performing operations.

[0220] A readable storage medium in an embodiment of the present application stores a program or instruction thereon. When the program or instruction is executed by a processor, the steps in the conflict mitigation method of the artificial intelligence module as described above are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0221] The processor is the processor in the conflict mitigation method of the artificial intelligence module described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0222] The above exemplary embodiments are described with reference to the accompanying drawings, and many different forms and embodiments are feasible without departing from the spirit and teachings of the present application. Therefore, the present application should not be constructed as a limitation of the exemplary embodiments proposed herein. More specifically, these exemplary embodiments are provided so that the present application will be perfect and complete, and the scope of the present application will be conveyed to those who are familiar with the technology. In these figures, the component sizes and relative sizes may be exaggerated for clarity. The terms used here are only based on the purpose of describing specific exemplary embodiments and are not intended to be limiting. As used herein, unless the text clearly indicates otherwise, the singular forms "one", "an" and "the" are intended to include these multiple forms. It will be further understood that the terms "comprising" and / or "including" when used in this specification indicate the presence of the features, integers, steps, operations, components and / or components, but do not exclude the presence or increase of one or more other features, integers, steps, operations, components, components and / or their groups. Unless otherwise indicated, when stated, a range of values ​​includes the upper and lower limits of that range and any subranges therebetween.

[0223] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A conflict mitigation method for an artificial intelligence module, characterized in that: The method is executed by a first network device and includes: When the first network device needs to deploy the first artificial intelligence AI decision module, the performance change data corresponding to all AI decision modules are obtained from the second network device; Determine first information according to the performance change data; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules; Send a training request to the second network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module and the first information.

2. The method according to claim 1, characterized in that Determining first information according to the performance change data includes: Determining, according to the performance change data, whether each performance parameter corresponding to the performance change data is less than a corresponding first preset range; If each of the performance parameters is greater than or equal to the corresponding first preset range, it is determined that the first information carries a first parameter, and a first AI decision module is deployed; the first parameter is used to instruct the second network device not to perform training; If some of the performance parameters are less than the corresponding first preset range, it is determined that the first information carries a second parameter; the second parameter is used to instruct the second network device to perform hierarchical training on the target AI decision module; the hierarchical training is to divide all AI decision modules into sets, and jointly train the target AI decision module sets after the divided sets.

3. The method according to claim 1, characterized in that The method further comprises: When each performance parameter in the performance change data is less than the corresponding first preset range, it is determined that the first information carries a third parameter; the third parameter is used to indicate that the attribute information corresponding to the first AI decision module is deleted in the first network device and / or the second network device.

4. The method according to claim 2, characterized in that: When the first information carries the second parameter, the method further includes: Receiving performance result data sent by the second network device after performing hierarchical training on the target AI decision module; Determine a decision message of the first network device according to the performance result data; the decision message carries the first number, the second number, and second information; the second information is used to indicate whether the first AI decision module is deployed in the first network device; The decision message is sent to the second network device.

5. The method according to claim 4, characterized in that After sending the decision message to the second network device, the method further includes: Receiving target parameter data corresponding to all AI decision modules after hierarchical training of the target AI decision module sent by the second network device; According to all the target parameter data, modify all AI decision modules of the first network device and deploy the first AI decision module; The first AI decision module is sent to the decision executor.

6. The method according to claim 1, characterized in that When the first network device needs to deploy the first artificial intelligence AI decision module, the performance change data corresponding to all AI decision modules are obtained from the second network device, including: Sending a verification message for verifying the first AI decision module to the second network device, wherein the verification message carries at least the first number and a number corresponding to the first AI decision module; Receive a verification response message sent by the second network device, wherein the verification response message carries performance change data corresponding to all the AI ​​decision modules.

7. A conflict mitigation method for an artificial intelligence module, characterized in that: The method is executed by the second network device and includes: Sending performance change data of all artificial intelligence AI decision modules to the first network device; receiving a training request sent by the first network device; the training request carries at least a first number of the first network device, a second number of a target AI decision module, and first information; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules; According to the training request, an operation corresponding to the first information is executed.

8. The method according to claim 7, characterized in that The method further comprises: receiving a decision message sent by the first network device; the decision message carries the first number, the second number, and second information; the second information is used to indicate whether a first AI decision module is deployed in the first network device; When the second information carries a fourth parameter, deleting the attribute information corresponding to the first AI decision module in the operating environment of the second network device; the fourth parameter is used to indicate that the first AI decision module is not deployed in the first network device; When the second information carries a fifth parameter, the second network device performs hierarchical training on the target AI decision module; the fifth parameter is used to indicate that the first AI decision module is deployed in the first network device; the hierarchical training is to divide all AI decision modules into sets, and jointly train the target AI decision module sets after the divided sets; The target parameter data corresponding to all AI decision modules after the hierarchical training are sent to the first network device.

9. The method according to claim 7, characterized in that: The method further comprises: Determine attribute information of the first AI decision module; the attribute information includes: priority and performance threshold constraint parameters of the first AI decision module; According to the priorities of the multiple AI decision modules, the multiple AI decision modules are divided into a first decision module set and a second decision module set; the first decision module set is used to represent all decision modules with a priority not higher than that of the first AI decision module; the second decision module set is used to represent all decision modules with a priority higher than that of the first AI decision module; According to the performance threshold constraint parameters corresponding to each AI decision module in the second decision module set, jointly train the first decision module set, and determine the target parameters for the performance change of the first decision module set during the training process; When it is determined that the target parameter meets the preset threshold range, the hierarchical training of the target AI decision module is completed.

10. A network device, the network device being a first network device, characterized in that: include: A first acquisition module is used to acquire performance change data corresponding to all AI decision modules from a second network device when the first artificial intelligence AI decision module needs to be deployed on the first network device; A first determination module is used to determine first information according to the performance change data; the first information is used to indicate whether the second network device trains a target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules; The first sending module is used to send a training request to the second network device; the training request carries at least the first number of the first network device, the second number of the target AI decision module and the first information.

11. A network device, the network device being a second network device, characterized in that: include: The second sending module is used to send the performance change data of all artificial intelligence AI decision modules to the first network device; A first receiving module is used to receive a training request sent by the first network device; the training request carries at least a first number of the first network device, a second number of a target AI decision module, and first information; the first information is used to indicate whether the second network device trains the target AI decision module; the target AI decision module is a part of the AI ​​decision modules among all the AI ​​decision modules; The first processing module is used to execute an operation corresponding to the first information according to the training request.

12. A network device comprising: A transceiver, a processor, a memory, and a program or instruction stored in the memory and executable on the processor; wherein the processor implements the conflict mitigation method for an artificial intelligence module as described in any one of claims 1 to 6 when executing the program or instruction, or implements the conflict mitigation method for an artificial intelligence module as described in any one of claims 7 to 9 when executing the program or instruction.

13. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by the processor, the conflict mitigation method for the artificial intelligence module as described in any one of claims 1 to 6 is implemented, or when the program or instruction is executed, the conflict mitigation method for the artificial intelligence module as described in any one of claims 7 to 9 is implemented.

Images