Information processing method and device, storage medium and computer program product

Abstract

The embodiment of the invention discloses an information processing method and device and a storage medium, and the method comprises the steps: enabling an information processing device to transmit a to-be-used AI model to a network management node under the condition of receiving the to-be-used AI model, so as to carry out the conflict verification of the to-be-used AI model through a global digital twin module in the network management node; the network management node comprises a plurality of groups of AI models in a plurality of network nodes; the plurality of network nodes are in one-to-one correspondence with the plurality of groups of AI models; under the condition that the network management node determines that the AI model to be used conflicts with a part of AI models in the multiple groups of AI models, a new AI model transmitted by the network management node is received, and the new AI model is an AI model obtained after the network management node carries out combination training on the AI model to be used and the part of AI models successfully; the partial AI models are AI models of which the model performance is no longer within a performance threshold range in the multiple groups of AI models in the merging training; and replacing the AI model to be used with the new AI model.

Description

Technical Field

[0001] The present application relates to the field of network technology, and in particular to an information processing method and device, a storage medium, and a computer program product. Background Art

[0002] There are multiple base stations in a wireless network, and AI models between different base stations may conflict. Technical means are needed to resolve existing conflicts.

[0003] In the related technology, a virtual scene is created based on the scene configuration information. In the created virtual scene, multiple virtual agents collaborate to perform the target task; conflict information of the target conflict type corresponding to each virtual agent in the process of collaboratively performing the target task is obtained; based on the conflict information, a task collaboration strategy is generated, and the task collaboration strategy is used to control the physical agents corresponding to the multiple virtual agents to collaborate to perform the target task in the real scene, thereby reducing the AI ​​strategy issuance rate in the network node. Summary of the Invention

[0004] To solve the above technical problems, the embodiments of the present application hope to provide an information processing method and device, a storage medium, and a computer program product that can increase the delivery rate of AI strategies in network nodes.

[0005] The technical solution of this application is achieved as follows:

[0006] The present invention provides an information processing method, which includes:

[0007] Upon receiving the AI ​​model to be used, the AI ​​model to be used is sent to a network management node, so as to perform conflict verification on the AI ​​model to be used using a global digital twin module in the network management node; the network management node includes multiple groups of AI models in multiple network nodes; the multiple network nodes correspond one-to-one to the multiple groups of AI models;

[0008] When the network management node determines that the AI ​​model to be used conflicts with some of the AI ​​models in the multiple groups of AI models, a new AI model transmitted by the network management node is received, where the new AI model is an AI model obtained after the network management node successfully performs merged training on the AI ​​model to be used and the partial AI models; the partial AI model is an AI model in the multiple groups of AI models whose model performance is no longer within a performance threshold range during the merged training;

[0009] The AI ​​model to be used is replaced by the new AI model.

[0010] An embodiment of the present application provides an information processing device, the device comprising:

[0011] A sending unit is configured to, upon receiving the AI ​​model to be used, send the AI ​​model to be used to a network management node, so as to perform conflict verification on the AI ​​model to be used using a global digital twin module in the network management node; the network management node includes multiple groups of AI models in multiple network nodes; and the multiple network nodes correspond one-to-one to the multiple groups of AI models;

[0012] a receiving unit, configured to receive, when the network management node determines that the AI ​​model to be used conflicts with some of the AI ​​models in the multiple groups of AI models, a new AI model transmitted by the network management node, where the new AI model is an AI model obtained after the network management node successfully performs merged training on the AI ​​model to be used and the partial AI models; the partial AI model is an AI model in the multiple groups of AI models whose model performance is no longer within a performance threshold range during merged training;

[0013] A replacement unit is configured to replace the AI ​​model to be used with the new AI model.

[0014] An embodiment of the present application provides an information processing device, the device comprising:

[0015] A memory, a processor and a communication bus, wherein the memory communicates with the processor via the communication bus, and the memory stores an information processing program executable by the processor. When the information processing program is executed, the information processing method described above is executed by the processor.

[0016] An embodiment of the present application provides a storage medium having a computer program stored thereon, which is applied to an information processing device, and is characterized in that when the computer program is executed by a processor, the above-mentioned information processing method is implemented.

[0017] An embodiment of the present application further provides a computer program product, including a computer program, which can be executed by a processor in an information processing device to complete the steps of the aforementioned information processing method.

[0018] An embodiment of the present application provides an information processing method and device, a storage medium, and a computer program product. The information processing method includes: when an AI model to be used is received, sending the AI ​​model to be used to a network management node, so as to use the global digital twin module in the network management node to perform conflict verification on the AI ​​model to be used; the network management node includes multiple groups of AI models in multiple network nodes; the multiple network nodes correspond one-to-one to the multiple groups of AI models; when the network management node determines that there is a conflict between the AI ​​model to be used and some AI models in the multiple groups of AI models, receiving a new AI model transmitted by the network management node, the new AI model is an AI model obtained after the network management node successfully merges and trains the AI ​​model to be used and the partial AI model; the partial AI model is an AI model whose model performance in the multiple groups of AI models is no longer within the performance threshold range during the merged training; and replacing the AI ​​model to be used with the new AI model. The above-mentioned method is adopted to implement the solution. When the information processing device receives the AI ​​model to be used, it sends the AI ​​model to be used to the network management node, and uses the global digital twin module in the network management node to perform conflict verification on the AI ​​model to be used; when the network management node determines that the AI ​​model to be used conflicts with some AI models in multiple groups of AI models, the network management node is used to merge the AI ​​model to be used and some AI models to obtain a new AI model for training, and then use the new AI model to replace the AI ​​model to be used for application, so that the new AI model no longer conflicts with any model in the multiple groups of AI models, and the new AI model or any model in the multiple groups of AI models can be directly used to determine the AI ​​strategy and directly send it to the physical environment for operation, thereby improving the delivery rate of AI strategies in the network node. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flow chart of an information processing method provided in an embodiment of the present application;

[0020] Figure 2 A schematic diagram of an exemplary intelligent network architecture based on digital twins provided in an embodiment of the present application;

[0021] Figure 3 A schematic diagram of the state of an AI model in an exemplary conflict subset provided in an embodiment of the present application;

[0022] Figure 4 This is a schematic diagram of an exemplary action of an AI model in a conflict subset provided in an embodiment of the present application;

[0023] Figure 5 A schematic diagram of a target of an AI model in an exemplary conflict subset provided in an embodiment of the present application;

[0024] Figure 6An exemplary information processing method provided in this application embodiment Figure 1 ;

[0025] Figure 7 An exemplary information processing method provided in this application embodiment Figure 2 ;

[0026] Figure 8 A schematic diagram of an exemplary implementation of an AI conflict mitigation method between multiple network nodes in an ORAN according to an embodiment of the present application;

[0027] Figure 9 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application Figure 1 ;

[0028] Figure 10 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application Figure 2 . DETAILED DESCRIPTION

[0029] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

[0030] The present invention provides an information processing method, which is applied to an information processing device. Figure 1 A flow chart of an information processing method provided in an embodiment of the present application is as follows: Figure 1 As shown, the information processing method may include:

[0031] S101. When the AI ​​model to be used is received, the AI ​​model to be used is sent to the network management node, so as to utilize the global digital twin module in the network management node to perform conflict verification on the AI ​​model to be used.

[0032] An information processing method provided in an embodiment of the present application is applicable to scenarios where conflicts exist in AI models in multiple base stations.

[0033] In the embodiments of the present application, the information processing device may be implemented in various forms. For example, the information processing device described in the present application may be a network node, and specific network nodes include entities such as base stations and core network elements.

[0034] In the embodiments of this application, Figure 2As shown: Each of the multiple network nodes includes a control module, a local digital twin module and a decision execution module. Among them, the control module includes at least a data acquisition function (data acquisition) and a local AI decision-making function. The data acquisition function is used to realize the collection of data required for the local virtual environment in the digital twin module (local digital twin module) (the control module is a controller, connected to the actual network device, that is, connected to the network device in the physical environment, used to collect the input data of the AI ​​model from the actual network device, and send the input data to the local digital twin module), such as wireless channel conditions, network traffic size, etc., and input it into the local virtual environment of the local digital twin module. The local AI decision-making function is to receive the decision output by the AI ​​model in the local virtual environment of the local digital twin module, and send it to the decision execution module for execution.

[0035] It should be noted that if Figure 2 As shown, the local digital twin module includes at least an AI model management function and a local virtual environment. The AI ​​model management function (such as the AI ​​model management function in network node 1) is used to implement the maintenance and query of all AI models stored in the network node (such as network node 1) and the conflict subset information corresponding to the model. Based on the sub-state information of other network nodes received from the Xn interface (the communication interface between network nodes) and the state information sent by the local virtual environment, the AI ​​model is used to output a decision. After the decision is verified in the local virtual environment, it is sent to the local AI decision-making function through the T2 interface. After receiving the data collected by the data acquisition function of the control module, the local virtual environment realizes real-time digital twins of the control module, decision execution module and physical environment of the network node. The operation of data acquisition and virtual environment is continuous and real-time, and has the ability to predict the network environment and verify decisions.

[0036] It should be noted that the decision execution module is the executor of the AI ​​model (that is, it executes the AI ​​decision issued by the local AI decision issuance function). The decision execution module can be deployed in the core network, base station, etc., such as the router's forwarding module, CU / DU, etc., to execute decisions such as antenna adjustment and bandwidth allocation.

[0037] In the embodiment of the present application, the network management node includes a data acquisition function and a global digital twin module. The network management node can be deployed on entities such as network management equipment, edge cloud or base station in the network. Figure 2As shown: The global digital twin module includes at least AI model management functions and a global virtual environment. The AI ​​model management function implements the maintenance and query of AI models and conflict subset information corresponding to conflict subsets in the entire network, interacts with the global virtual environment to retrain the AI ​​model, and after retraining, uses the retrained model to output decisions. After the decision is verified in the global virtual environment, it is sent to the network node through the AI ​​interface together with the retrained AI model. The global virtual environment only maintains the digital twins of each network node in conflict and has the ability to predict the network environment and verify decisions. The data of the global virtual environment is not collected in real time. Instead, it receives data reported by the local digital twins of the network nodes when the AI ​​model needs to be retrained. After retraining is completed, it will no longer run.

[0038] In an embodiment of the present application, the AI ​​model to be used can be an AI model configured by the user into the network node, or it can be an AI model transmitted to the network node by other devices, or it can be an AI model obtained by the network node through other means. The specific way in which the network node obtains the AI ​​model to be used can be determined based on actual conditions, and the embodiment of the present application does not limit this.

[0039] In the embodiment of the present application, the network management node is a node that manages network nodes. The network management node includes multiple groups of AI models in multiple network nodes, wherein the multiple network nodes correspond to the multiple groups of AI models one-to-one, that is, one network node corresponds to one group of AI models.

[0040] It should be noted that the information processing device is any one of a plurality of network nodes.

[0041] In the embodiment of the present application, the AI ​​model to be used can be a deep reinforcement learning model, a self-supervised learning model, an image language model, etc., or other models. The specific model can be determined according to actual conditions, and the embodiment of the present application does not limit this.

[0042] In an embodiment of the present application, multiple groups of AI models may include deep reinforcement learning models, self-supervised learning models, image language models, etc.

[0043] In the embodiment of the present application, a communication connection is established between the network management node and each of the multiple network nodes to implement information interaction with each of the multiple network nodes.

[0044] In an embodiment of the present application, the information processing device sends the AI ​​model to be used to the network management node through a communication connection with the network management node.

[0045] In an embodiment of the present application, when the network management node receives the AI ​​model to be used transmitted by the information processing device, the network management node deploys the AI ​​model to be used to the global virtual environment of the global digital twin module, and starts running the AI ​​model to be used and the AI ​​models originally existing in the global virtual environment (i.e., multiple groups of AI models). The global virtual environment collects the operating data of these AI models in the operating environment and analyzes the performance results. When the performance results identify that the model performance of some AI models in the multiple groups of AI models is no longer within the performance threshold range, the network management node uses the partial AI models and the AI ​​model to be used to form a conflict subset, merges and trains all AI models in this conflict subset (i.e., partial AI models and AI models to be used) to obtain a new AI model, and then the network management node transmits the new AI model to the information processing device and the target network node to which the partial AI model belongs.

[0046] S102: When the network management node determines that the AI ​​model to be used conflicts with some AI models in the multiple groups of AI models, receive a new AI model transmitted by the network management node.

[0047] In an embodiment of the present application, after the information processing device sends the AI ​​model to be used to the network management node, if the network management node determines that the AI ​​model to be used conflicts with some AI models in multiple groups of AI models, it receives the new AI model transmitted by the network management node.

[0048] It should be noted that the new AI model is the AI ​​model obtained after the network management node successfully merges and trains the used AI model and the partial AI model; the partial AI model is the AI ​​model in which the model performance of multiple groups of AI models is no longer within the performance threshold range during the merged training.

[0049] It should also be noted that the performance threshold range can be a parameter configured in the network management node, a parameter transmitted to the network management node by other devices, or a parameter obtained by the network management node through other means. The specific way in which the network management node obtains the performance threshold range can be determined based on actual conditions, and the embodiments of the present application do not limit this.

[0050] In an embodiment of the present application, in the event of conflicts between AI models of different network nodes in the network (i.e., the AI ​​model to be used and some AI models in multiple groups of AI models), the conflicting AI models are divided into the same conflict subset (i.e., some AI models and the AI ​​model to be used are divided into the same conflict subset) and merged for training to alleviate conflicts between the AI ​​models. The conflict subset number is unique across the entire network.

[0051] It should be noted that merged training is to merge the states, actions (strategies), and goals of different AI models into one neural network for training to generate one AI model. Therefore, after merged training, the AI ​​models corresponding to the same conflict subset number in different network nodes are the same. For example, Figure 3 As shown: The AI ​​models in the conflict subset M belong to network node x, network node y and network node z respectively, then the corresponding sub-state of the AI ​​model is obtained in network node x, the corresponding sub-state of the AI ​​model is obtained in network node y, and the corresponding sub-state of the AI ​​model is obtained in network node z, thereby obtaining the state of the AI ​​model in the conflict subset M (it should be noted that the state of the neural network includes the states of all the original models). Figure 4 As shown: Get the corresponding AI model sub-action in network point x, get the corresponding AI model sub-action in network node y, get the corresponding AI model sub-action in network node z, and thus get the action of the AI ​​model in the conflict subset M (it should be noted that the action of the neural network includes the actions of all the original models). Figure 5 As shown: obtain the corresponding sub-goal of the AI ​​model in network point x, obtain the corresponding sub-goal of the AI ​​model in network node y, and obtain the corresponding sub-goal of the AI ​​model in network node z, so as to obtain the goal of the AI ​​model in the conflict subset M (it should be noted that the goal of this neural network includes the goals of all original models).

[0052] For example, the state of the AI ​​model can be the traffic information of each user obtained by the network node, the action (strategy) of the AI ​​model can be bandwidth allocation based on the collected traffic information of each user, and the goal of the AI ​​model can be the throughput value after bandwidth allocation.

[0053] S103: Replace the AI ​​model to be used with the new AI model.

[0054] In an embodiment of the present application, after the information processing device receives the new AI model transmitted by the network management node, it uses the new AI model to replace the AI ​​model to be used.

[0055] In an embodiment of the present application, after the network management node transmits the new AI model to the information processing device and the target network node to which the partial AI model belongs, the information processing device uses the new AI model to replace the AI ​​model to be used in the information processing device, and the target network node uses the new AI model to replace part of the AI ​​model in the target network node.

[0056] For example, before starting an unused AI model (AI model to be used), the network node N (information processing device) needs to verify in the global digital twin of the network management node whether there is any conflict with the started AI model. Figure 6 As shown: Network node n decides to start AI model n (the AI ​​model to be used) by monitoring network data or through external triggering. Network node N first sends AI model n to the network management node. Upon receiving AI model n from network node N, the network management node deploys AI model n to the global virtual environment and begins running it. The global virtual environment collects the operating data of the AI ​​model in the operating environment and analyzes the performance results. The network management node divides AI models whose performance no longer falls within the performance threshold (such as model 1 of network node 1) and AI model n into the same conflict subset, performs merge training, and outputs the AI ​​model q (new AI model) obtained after the merge training. AI model q is then sent to network node 1 and network node N, replacing the original AI model 1 and AI model n. Ultimately, the conflict subset contains (network node 1 and network node N), and AI model q is AI model q.

[0057] In an embodiment of the present application, after replacing the AI ​​model to be used with a new AI model, the information processing device, when collecting multiple sets of input data corresponding to multiple AI models in the virtual environment from the physical environment, determines, based on the operation results of the multiple AI models on the multiple sets of input data, a first AI model whose model performance is no longer within a performance threshold range, and a first moment when the model performance is no longer within the performance threshold range; determines a first conflict subset to which the first AI model belongs, and other network nodes to which other AI models in the first conflict subset belong; sends a decision update request to the other network nodes; obtains, from the other network nodes, other substates of the other AI models when running at the first moment based on the decision update request; determines a first hybrid strategy using the first AI model and based on the first substate and other substates of the first AI model when running at the first moment; and, upon determining, according to the first hybrid strategy, that the first model performance corresponding to the first AI model is within the performance threshold range and receiving prompt information indicating that the performance of other models determined according to other strategies of other AI models are within the performance threshold range, sends a first sub-strategy to the physical environment, so that the first sub-strategy is executed in the physical environment when the physical environment runs at the first moment.

[0058] It should be noted that the multiple AI models include new AI models.

[0059] It should be noted that the other AI models are AI models in the first conflict subset except the first AI model.

[0060] It should be noted that the first sub-strategy is the strategy corresponding to the first AI model in the first hybrid strategy.

[0061] In an embodiment of the present application, after a new AI model is used to replace the model to be used in a physical environment, when it is necessary to perform conflict verification on multiple AI models in the information processing device, the information processing device collects multiple sets of input data corresponding to the multiple AI models from the physical environment, inputs the multiple sets of input data into the local virtual environment, and uses the operation results of the multiple AI models on the multiple sets of input data in the local virtual environment to perform conflict verification on the multiple AI models.

[0062] In an embodiment of the present application, multiple AI models correspond one-to-one to multiple groups of input data, that is, one AI model corresponds to one group of input data.

[0063] In an embodiment of the present application, the number of first AI models can be one, or the number of first AI models can be multiple. The specific number of first AI models can be determined based on actual conditions, and the embodiment of the present application does not limit this.

[0064] In an embodiment of the present application, the number of other AI models can be one, or the number of other AI models can be multiple. The specific number of other AI models can be determined based on actual conditions, and the embodiment of the present application does not limit this.

[0065] In an embodiment of the present application, the information processing device uses the first AI model to determine the first hybrid strategy based on the first sub-state and other sub-states when the first AI model runs to the first moment, including: the information processing device inputs the first sub-state and other sub-states into the first AI model, and uses the first AI model to output the first hybrid strategy.

[0066] In an embodiment of the present application, after using the first AI model to determine the first hybrid strategy based on the first sub-state and other sub-states, the first sub-strategy corresponding to the first AI model is searched in the first hybrid strategy, and the information processing device runs the first sub-strategy to obtain a first running result, that is, the first model performance is obtained.

[0067] It should be noted that each of the other network nodes will obtain the sub-states of the AI ​​models sent by the network nodes other than its own network node in the first conflict subset, and input the obtained sub-states and the sub-states of the network model in its own network node into the other AI models, thereby obtaining other strategies of the other AI models.

[0068] Exemplarily, if the number of other AI models is one, the number of other network nodes is also one. When the information processing device obtains other sub-states of other AI models when running to the first moment from other network nodes according to the decision update request, it will also send the first sub-state of the first AI model when running to the first moment to other network nodes, so that other network nodes can use other AI models to determine the first hybrid strategy based on the first sub-state and other sub-states; determine other strategies of other AI models from the first hybrid strategy, and then other network nodes run the other strategies to obtain the operation results of other strategies, that is, the performance of other models.

[0069] Exemplarily, if the number of other AI models is two, the number of other network nodes may also be two. When the information processing device obtains the other sub-states (the first other sub-state and the second other sub-state) of the other AI models when running to the first moment from the two other network nodes according to the decision update request, it will also send the first sub-state of the first AI model when running to the first moment to the two other network nodes, so that the two other network nodes can use the two other AI models respectively to determine the first mixed strategy based on the first sub-state, the first other sub-state and the second other sub-state; the two other network nodes respectively determine the other strategies of the first other AI model and the other strategies of the first other AI model from the first mixed strategy, and the two other network nodes then respectively run the first other strategy and the second other strategy to obtain the operation results of the two other strategies, that is, the performance of the two other models.

[0070] In the embodiment of the present application, the physical environment is an environment opposite to the virtual environment. The virtual environment may be a simulation environment, and the physical environment may be an environment of actual application.

[0071] In the embodiment of the present application, a decision execution module is provided in the information processing device, and the decision execution module can be used to execute the first sub-strategy in a physical environment.

[0072] It can be understood that, based on the running results of multiple AI models on multiple sets of input data, the first AI model whose model performance is no longer within the performance threshold range and the first moment when the model performance is no longer within the performance threshold range are determined, and based on the decision update request, other sub-states of other AI models when running at the first moment are obtained from other network nodes. The first AI model is used, and based on the first sub-state and other sub-states when the first AI model runs at the first moment, a first hybrid strategy is determined to resolve the conflict between the AI ​​model and other AI models. The first AI model or any one of the other AI models can be directly used to determine the AI ​​strategy and directly send it to the physical environment for operation, thereby improving the delivery rate of AI strategies in network nodes.

[0073] In an embodiment of the present application, the information processing device determines the process of the first AI model whose model performance is no longer within the performance threshold range based on the running results of multiple AI models on multiple sets of input data, including: determining the model performance of multiple AI models based on the running results of each AI model in the multiple AI models on each set of input data; when the first performance among the multiple model performances is no longer within the performance threshold range, determining the first AI model corresponding to the first performance.

[0074] It should be noted that each set of input data is a set of input data corresponding to each AI model in multiple sets of input data.

[0075] In an embodiment of the present application, multiple sets of input data are input into multiple AI models respectively to obtain multiple model performances.

[0076] In the embodiment of the present application, the first performance is the performance of some models among the multiple models.

[0077] It should be noted that the number of first performances can be multiple, and the specific number can be determined according to actual conditions. The embodiments of the present application do not limit this.

[0078] In an embodiment of the present application, the first performance is the performance of the first AI model.

[0079] It can be understood that by setting up a digital twin module in the information processing device and using the digital twin module to verify multiple AI models based on multiple sets of input data, conflict mitigation can be achieved in the digital network. Various conflicts, including implicit conflicts, can be intuitively discovered, providing a reliable basis for conflict mitigation and reducing the risk of trial and error.

[0080] In an embodiment of the present application, the process of the information processing device determining the first conflict subset to which the first AI model belongs, and the other network nodes to which other AI models in the first conflict subset belong, includes: determining the first conflict subset to which the first AI model belongs; obtaining a conflict data set; and searching for other network nodes to which other AI models in the first conflict subset belong in the conflict data set.

[0081] It should be noted that the conflict data set records multiple conflict subsets, the AI ​​models that constitute the multiple conflict subsets, and the network nodes corresponding to the AI ​​models.

[0082] In the embodiment of the present application, the first conflict subset is a conflict subset including the first AI model.

[0083] It should be noted that the first conflict subset is established when a certain AI model in the first conflict subset is started because the AI ​​model conflicts with other existing AI models.

[0084] In an embodiment of the present application, a method for determining the first conflict subset to which the first AI model belongs may be to search for a conflict subset including the first AI model in multiple conflict subsets, thereby obtaining the first conflict subset.

[0085] In the embodiment of the present application, a network node may store multiple AI models, which can be divided into two categories: one type of AI model does not conflict with the AI ​​models of other network nodes and therefore does not belong to any conflict subset; the other type of AI model conflicts with the AI ​​models of other network nodes and belongs to a certain conflict subset. The AI ​​model and conflict subset information storage of a network node (i.e., conflict dataset) is shown in Table 1:

[0086] Table 1

[0087]

[0088] It should be noted that, with respect to Table 1, AI model 1 does not conflict with the AI ​​models of other network nodes, and no conflict subset is established based on the AI ​​model. That is, there is no conflict subset number that includes AI model 1, nor is there any information about the network nodes included in the conflict subset. The conflict subset to which AI model 2 belongs is conflict subset 1, and the network nodes included in conflict subset 1 are network nodes 1, 2, 4, and 3. The conflict subset to which AI model 3 belongs is conflict subset 4, and the network nodes included in conflict subset 4 are network nodes 1, 2, 4, and 5.

[0089] It can be understood that by dividing the conflict subsets, setting the conflict data set, and searching for other network nodes to which other AI models corresponding to the first conflict subset belong based on the conflict data set, the decision generation time can be reduced.

[0090] In an embodiment of the present application, the information processing device sends the first sub-strategy to the physical environment when it determines according to the first hybrid strategy that the first model performance corresponding to the first AI model is within the performance threshold range, and receives prompt information that the other model performance determined according to other strategies of other AI models is within the performance threshold range. The process includes: searching for the first sub-strategy corresponding to the first AI model from the first hybrid strategy; running the first sub-strategy to obtain a first running result; and sending the first sub-strategy to the physical environment when the first model performance determined according to the first running result is within the performance threshold range, and receiving prompt information that other performance corresponding to other running results obtained by other network nodes running other strategies are also within the performance threshold range.

[0091] In an embodiment of the present application, an AI decision-issuing function is provided in the information processing device, which can use the AI ​​decision-issuing function to issue the first sub-strategy to the decision execution module in a physical environment, and use the decision execution module to execute the first sub-strategy to obtain a first operating result.

[0092] It can be understood that by dividing the conflict subsets, a distributed global state information acquisition and decision output method is proposed to reduce the decision generation time.

[0093] In an embodiment of the present application, the information processing device uses the first AI model and the first sub-state and other sub-states when the first AI model runs to the first moment to determine the first hybrid strategy. When the performance of the first model is not within the performance threshold range, or when prompt information is received that other performance is not within the performance threshold range, the information processing device sends a retraining request to the network management node; uses the global digital twin module in the network management node to retrain the first AI model according to the retraining request to obtain a second AI model whose performance is within the performance threshold range, and a second hybrid strategy obtained according to the second AI model; receives the second AI model and the second hybrid strategy; searches for the second sub-strategy corresponding to the first sub-state from the second hybrid strategy; runs the second sub-strategy to obtain a second running result; when the performance of the second AI model determined according to the second running result is within the performance threshold range, and prompt information is received that the performance of other second AI models corresponding to the second other running results determined by other network nodes according to the second AI model are within the performance threshold range, when the physical environment runs to the first moment, the second sub-strategy is run using the second AI model in the physical environment.

[0094] In an embodiment of the present application, the retraining method is to retrain the first AI model with new data while the state, action (strategy), and goal of the first AI model remain unchanged.

[0095] In an embodiment of the present application, the network node x (information processing device) collects data from the physical environment and inputs it into the virtual environment. When the virtual environment runs to time T, it finds that the optimization target performance (such as rx1) is no longer within the threshold range (performance threshold range). Determine the type of AI model m (first AI model) corresponding to the optimization target. In the case where the type of AI model m is a second-class AI model (that is, it conflicts with the AI ​​models of other network nodes and belongs to a certain conflict subset), determine the conflict subset to which the AI ​​model m belongs, thereby implementing the conflict mitigation process for the AI ​​model m. For example, as Figure 7 As shown:

[0096] Step 1: If the AI ​​model m corresponding to the optimization objective belongs to the conflict subset M (determine the first conflict subset to which the first AI model belongs), network node x queries the conflict dataset and finds that it belongs to the same conflict subset M as network node y and network node z (in the conflict dataset, find other network nodes to which other AI models in the first conflict subset belong);

[0097] Step 2: Network node x sends a decision update request to network nodes y and z (sending a decision update request to other network nodes). The decision update request includes at least its own network node ID x, the conflict subset number M, and the timestamp T (first moment). The timestamp T is used to ensure time synchronization between different network nodes.

[0098] Step 3: After network nodes y and z receive the decision update request from network node x, they query the corresponding sub-state according to the conflict subset and collect the sub-state information of the virtual environment in the digital twin module running to time T. (other sub-states);

[0099] Step 4: Network nodes y and z feed back the sub-state information to all other network nodes in the conflict subset M. Taking network node y as an example, it sends sub-state information to network nodes x and z, including at least its own network node ID y, conflict subset number M, timestamp T and sub-state information.

[0100] Step 5: Network nodes x, y, and z obtain all states (the first substate and other substates) required by model m in the conflict subset M, input the model m, and output an overall action containing all node sub-actions (using the first AI model and the first substate and other substates when the first AI model runs to the first moment to determine the first hybrid strategy);

[0101] Step 6: Network nodes x, y, and z run their respective sub-actions at time T in the virtual environment to determine whether the performance meets the constraints.

[0102] Step 7: Network nodes x, y, and z send performance information to each other. For example, network node x sends performance information to y and z, which includes at least its own network node ID x, conflict subset number M, timestamp T, and whether any performance does not meet the threshold N / Y (binary value).

[0103] Step 8A: If, after executing the sub-action in step 6, the performance of network nodes x, y, and z all meet the threshold constraints (when the first model performance determined according to the first operation result is within the performance threshold range, and prompt information is received that other performances corresponding to other operation results obtained by other network nodes running other strategies are also within the performance threshold range), then when the physical environment runs to time T, the digital twin module sends the sub-action through the control module to the decision execution module for execution;

[0104] Step 8B: If, after executing the sub-action in step 6, at least one of the performances of network nodes x, y, and z does not meet the threshold constraint (when the performance of the first model is not within the performance threshold range, or when receiving prompt information that other performances are not within the performance threshold range), then network nodes x, y, and z send a retraining request to the global digital twin module (send a retraining request to the network management node), including at least the conflict subset number M and the digital twin data associated with the conflict subset;

[0105] Step 9: The global digital twin retrains the AI ​​model based on the digital twin data uploaded by network nodes x, y, and z to obtain the AI ​​model m' (the second AI model), inputs the sub-state of each network node at time T, and outputs the sub-action and performance (the second AI model and the second hybrid strategy);

[0106] Step 10A: If each performance satisfies the threshold constraint (if the performance of the second AI model determined based on the second operation result is within the performance threshold range, and prompt information is received indicating that the performance of other second AI models corresponding to second other operation results determined by other network nodes based on the second AI model are within the performance threshold range), the network management node sends a decision update reception response message to network nodes x, y, and z, including at least the conflicting subset label M, model m', and all sub-actions. Network nodes x, y, and z then execute step 8.A.

[0107] Step 10B: If at least one performance item does not meet the threshold constraint (if the second AI model performance determined based on the second operating result is not within the performance threshold range, or prompt information is received that the other second AI model performance corresponding to the second other operating result determined by another network node based on the second AI model is not within the performance threshold range), a decision update rejection response message is sent, including at least the conflicting subset number M, the timestamp T, and the rejection response indicator.

[0108] It can be understood that this application proposes a conflict mitigation architecture that combines centralized and distributed methods, where distributed local digital twins run in real time and centralized global digital twins are triggered when needed, saving data transmission of digital twins in the network and maintenance costs of large-scale digital twins.

[0109] In an embodiment of the present application, the information processing device determines the first AI model whose model performance is no longer within the performance threshold range, and after the first moment when the model performance is no longer within the performance threshold range, in the absence of a conflicting subset including the first AI model, the first sub-state is input into the first AI model to obtain a third sub-strategy; runs the third sub-strategy to obtain a third running result; and when the first model performance of the first AI model determined according to the third running result is within the performance threshold range, the third sub-strategy is issued to the physical environment, so that when the physical environment runs to the first moment, the third sub-strategy is executed in the physical environment.

[0110] In an embodiment of the present application, the information processing device, in a physical environment, uses the AI ​​decision-making distribution function to distribute the third sub-strategy to the decision execution module at a first moment, and uses the decision execution module to execute the third sub-strategy.

[0111] It can be understood that by setting up a digital twin module in the information processing device, using the digital twin module to perform conflict detection on the first AI model, and realizing conflict mitigation in the digital network, various conflicts including implicit conflicts can be intuitively discovered, providing a reliable basis for conflict mitigation and reducing the risk of trial and error.

[0112] In an embodiment of the present application, after the information processing device executes the third sub-strategy and obtains a third operation result, if the first model performance of the first AI model determined according to the third operation result is not within the performance threshold range, the first AI model is retrained to obtain a third AI model;

[0113] When the model performance corresponding to the third AI model is within the performance threshold range, the first sub-state is input into the third AI model to obtain a fourth sub-strategy; the fourth sub-strategy is executed to obtain a fourth operating result; when the model performance of the third AI model determined according to the fourth operating result is within the performance threshold, the fourth sub-strategy is issued to the physical environment, so that when the physical environment runs to the first moment, the fourth sub-strategy is executed in the physical environment.

[0114] In an embodiment of the present application, the information processing device obtains new training data, and retrains the first AI model using the new training data in a virtual environment, thereby obtaining a third AI model.

[0115] In this embodiment of the present application, the network node x (information processing device) collects data from the physical environment and inputs it into the virtual environment. When the virtual environment runs to time T and finds that the optimization target performance (such as rx1) is no longer within the threshold range (performance threshold range), it determines the type of AI model m (first AI model) corresponding to the optimization target. As shown in steps af:

[0116] In step a, if the AI ​​model m corresponding to the optimization target does not belong to any conflicting subset (there is no conflicting subset including the first AI model);

[0117] Step b: Network node x inputs the state to model m at time T, obtains the action, and executes the action (inputs the first sub-state into the first AI model to obtain the third sub-strategy);

[0118] Step c: If, after executing the sub-action in step b, the performance of network node x meets the threshold constraint (when the first model performance of the first AI model determined according to the third operation result is within the performance threshold range), then when the physical environment runs to time T, the digital twin module sends the sub-action to the decision executor for execution (sending the third sub-strategy to the physical environment so that when the physical environment runs to the first time, the third sub-strategy is executed in the physical environment);

[0119] In step d, if, after executing the sub-action in step b, at least one of the performance constraints of network node x does not meet the threshold (if the first model performance of the first AI model determined according to the third operation result is not within the performance threshold range), then network node x will retrain AI model m in the virtual environment to obtain model m' (retrain the first AI model to obtain the third AI model);

[0120] In step e, if each performance level satisfies the threshold constraint (if the model performance corresponding to the third AI model is within the performance threshold range), network node x executes step b.

[0121] In step f, if at least one performance does not meet the threshold constraint (when the model performance corresponding to the third AI model is not within the performance threshold range), no new action will be issued to the decision execution part.

[0122] It can be understood that by setting a digital twin module in the information processing device, when the first model performance of the first AI model determined according to the third operation result is not within the performance threshold range, the digital twin module is used to retrain the first AI model, so as to solve the conflict problem between the first AI model and other AI models. The first AI model can be directly used to determine the AI ​​strategy and directly sent to the physical environment for operation, thereby improving the delivery rate of AI strategies in network nodes.

[0123] For example, the AI ​​conflict mitigation method among multiple nodes in the network is implemented in ORAN as follows: Figure 8As shown: The AI ​​model of the present invention is ORAN's xAPP. Conflict mitigation between ORAN's xAPPs is accomplished by Conflict Mitigation, but the current ORAN standard documents do not have a specific implementation method, nor do they provide conflict mitigation between Near-RT RICs. In the ORAN architecture, new digital twins DT (Global DT and Local DT) are added. The network management node corresponds to the SMO, and the control module of the network node corresponds to the Near-RT RIC. A new interface T2 is added between DT and the SMO and Near-RT RIC. New AI and Xn port messages are added.

[0124] It can be understood that when the information processing device receives the AI ​​model to be used, it sends the AI ​​model to be used to the network management node, and uses the global digital twin module in the network management node to perform conflict verification on the AI ​​model to be used; when the network management node determines that the AI ​​model to be used conflicts with some AI models in multiple groups of AI models, the network management node is used to merge the AI ​​model to be used and some AI models to obtain a new AI model for training, and then use the new AI model to replace the AI ​​model to be used for application, so that the new AI model no longer conflicts with any model in the multiple groups of AI models, and the new AI model or any model in the multiple groups of AI models can be directly used to determine the AI ​​strategy and directly send it to the physical environment for operation, thereby improving the delivery rate of AI strategies in the network node.

[0125] Based on the same inventive concept as the above-mentioned information processing method, an embodiment of the present application provides an information processing device 1 corresponding to an information processing method; Figure 9 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application Figure 1 , the information processing device 1 may include:

[0126] The sending unit 11 is configured to, upon receiving the AI ​​model to be used, send the AI ​​model to be used to a network management node, so as to perform conflict verification on the AI ​​model to be used using the global digital twin module in the network management node; the network management node includes multiple groups of AI models in multiple network nodes; and the multiple network nodes correspond one-to-one to the multiple groups of AI models;

[0127] A receiving unit 12 is configured to receive, when the network management node determines that the AI ​​model to be used conflicts with some of the AI ​​models in the multiple groups of AI models, a new AI model transmitted by the network management node, where the new AI model is an AI model obtained after the network management node successfully performs merged training on the AI ​​model to be used and the partial AI models; the partial AI model is an AI model in the multiple groups of AI models whose model performance is no longer within a performance threshold range during merged training;

[0128] The replacement unit 13 is configured to replace the AI ​​model to be used with the new AI model.

[0129] In some embodiments of the present application, the apparatus further includes a determining unit;

[0130] The determining unit is configured to, when collecting multiple sets of input data corresponding to multiple AI models in a virtual environment from a physical environment, determine, based on the results of running the multiple AI models on the multiple sets of input data, a first AI model whose model performance is no longer within the performance threshold range, and a first moment at which the model performance is no longer within the performance threshold range; the multiple AI models include the new AI model; determine a first conflict subset to which the first AI model belongs, and other network nodes to which other AI models in the first conflict subset belong; the other AI models are AI models other than the first AI model in the first conflict subset; and determine a first hybrid strategy using the first AI model and based on a first substate and the other substates when the first AI model runs to the first moment;

[0131] The sending unit 11 is used to send a decision update request to the other network node; to obtain the other sub-states of the other AI model when it runs to the first moment from the other network node according to the decision update request; when it is determined according to the first hybrid strategy that the first model performance corresponding to the first AI model is within the performance threshold range, and prompt information is received that the other model performance determined according to other strategies of the other AI model is within the performance threshold range, the first sub-strategy is sent to the physical environment, so that when the physical environment runs to the first moment, the first sub-strategy is executed in the physical environment; the first sub-strategy is the strategy corresponding to the first AI model in the first hybrid strategy.

[0132] In some embodiments of the present application, the determination unit is used to determine the model performance of multiple AI models based on the operating results of each AI model in the multiple AI models on each set of input data; each set of input data is a set of input data corresponding to each AI model in the multiple sets of input data; when the first performance among the multiple model performances is no longer within the performance threshold range, determine the first AI model corresponding to the first performance.

[0133] In some embodiments of the present application, the apparatus further includes an acquiring unit and a searching unit;

[0134] The determining unit is configured to determine a first conflict subset to which the first AI model belongs;

[0135] The acquisition unit is configured to acquire a conflict data set, wherein the conflict data set records a plurality of conflict subsets, the AI ​​models constituting the plurality of conflict subsets, and the network nodes corresponding to the AI ​​models;

[0136] The search unit is configured to search the conflict data set for other network nodes to which other AI models in the first conflict subset belong.

[0137] In some embodiments of the present application, the apparatus further comprises an operating unit;

[0138] The searching unit is configured to search the first hybrid strategy for a first sub-strategy corresponding to the first AI model;

[0139] The running unit is configured to run the first sub-strategy to obtain a first running result;

[0140] The sending unit 11 is used to send the first sub-strategy to the physical environment when the first model performance determined according to the first operating result is within the performance threshold range and the prompt information received indicates that other performance corresponding to other operating results obtained by other network nodes running other strategies are also within the performance threshold range.

[0141] In some embodiments of the present application, the sending unit 11 is configured to send a retraining request to the network management node when the performance of the first model is not within the performance threshold range or when receiving prompt information that the other performance is not within the performance threshold range; so as to use the global digital twin module in the network management node to retrain the first AI model according to the retraining request to obtain a second AI model whose performance is within the performance threshold range, and a second hybrid strategy obtained based on the second AI model;

[0142] The receiving unit 12 is configured to receive the second AI model and the second hybrid strategy;

[0143] The searching unit is configured to search the second hybrid strategy for a second sub-strategy corresponding to the first sub-state;

[0144] The operating unit is configured to operate the second sub-strategy to obtain a second operating result; and when the physical environment operates to the first moment, when the performance of the second AI model determined according to the second operating result is within the performance threshold range and prompt information is received indicating that the performance of other second AI models corresponding to the second other operating result determined by the other network node according to the second AI model is within the performance threshold range, the second sub-strategy is operated using the second AI model in the physical environment.

[0145] In some embodiments of the present application, the apparatus further comprises an input unit;

[0146] the input unit is configured to input the first sub-state into the first AI model to obtain a third sub-strategy when there is no conflicting subset including the first AI model;

[0147] The running unit is configured to run the third sub-strategy to obtain a third running result;

[0148] The sending unit 11 is used to send the third sub-strategy to the physical environment when the first model performance of the first AI model determined according to the third operation result is within the performance threshold range, so as to execute the third sub-strategy in the physical environment when the physical environment runs to the first moment.

[0149] In some embodiments of the present application, the apparatus further comprises a training unit;

[0150] the training unit being configured to retrain the first AI model to obtain a third AI model when the first model performance of the first AI model determined according to the third operating result is not within the performance threshold range;

[0151] the input unit being configured to input the first sub-state into the third AI model to obtain a fourth sub-strategy when the model performance corresponding to the third AI model is within the performance threshold range;

[0152] The running unit is configured to execute the fourth sub-strategy to obtain a fourth running result;

[0153] The sending unit 11 is used to send the fourth sub-strategy to the physical environment when the model performance of the third AI model determined according to the fourth operation result is within the performance threshold, so as to execute the fourth sub-strategy in the physical environment when the physical environment runs to the first moment.

[0154] It should be noted that, in actual applications, the above-mentioned sending unit 11, receiving unit 12 and replacement unit 13 can be implemented by a processor 14 on the information processing device 1, specifically a CPU (Central Processing Unit), MPU (Microprocessor Unit), DSP (Digital Signal Processing) or Field Programmable Gate Array (FPGA); the above-mentioned data storage can be implemented by the memory 15 on the information processing device 1.

[0155] The present application also provides an information processing device 1, such as Figure 10 As shown, the information processing device 1 includes: a processor 14, a memory 15 and a communication bus 16. The memory 15 communicates with the processor 14 through the communication bus 16. The memory 15 stores a program executable by the processor 14. When the program is executed, the information processing method described above is executed by the processor 14.

[0156] In practical applications, the memory 15 may be a volatile memory, such as a random-access memory (RAM); or a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 14.

[0157] An embodiment of the present application provides a computer-readable storage medium having a computer program thereon, which implements the information processing method described above when the program is executed by the processor 14 .

[0158] The embodiment of the present application further provides a computer program product, including a computer program, which can be executed by the processor 14 in the information processing device to complete the steps of the aforementioned information processing method.

[0159] It can be understood that when the information processing device receives the AI ​​model to be used, it sends the AI ​​model to be used to the network management node, and uses the global digital twin module in the network management node to perform conflict verification on the AI ​​model to be used; when the network management node determines that the AI ​​model to be used conflicts with some AI models in multiple groups of AI models, the network management node is used to merge the AI ​​model to be used and some AI models to obtain a new AI model for training, and then use the new AI model to replace the AI ​​model to be used for application, so that the new AI model no longer conflicts with any model in the multiple groups of AI models, and the new AI model or any model in the multiple groups of AI models can be directly used to determine the AI ​​strategy and directly send it to the physical environment for operation, thereby improving the delivery rate of AI strategies in the network node.

[0160] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0161] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0162] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0164] The above description is merely a preferred embodiment of the present application and is not intended to limit the scope of protection of the present application.

Claims

1. An information processing method, characterized in that: The method comprises: Upon receiving the AI ​​model to be used, the AI ​​model to be used is sent to a network management node, so as to perform conflict verification on the AI ​​model to be used using a global digital twin module in the network management node; the network management node includes multiple groups of AI models in multiple network nodes; the multiple network nodes correspond one-to-one to the multiple groups of AI models; When the network management node determines that the AI ​​model to be used conflicts with some of the AI ​​models in the multiple groups of AI models, a new AI model transmitted by the network management node is received, where the new AI model is an AI model obtained after the network management node successfully performs merged training on the AI ​​model to be used and the partial AI models; the partial AI model is an AI model in the multiple groups of AI models whose model performance is no longer within a performance threshold range during the merged training; The AI ​​model to be used is replaced by the new AI model.

2. The method according to claim 1, characterized in that After replacing the AI ​​model to be used with the new AI model, the method further includes: In a case where multiple sets of input data corresponding to multiple AI models in a virtual environment are collected from a physical environment, determining, based on results of running the multiple AI models on the multiple sets of input data, a first AI model whose model performance is no longer within the performance threshold range, and a first moment when the model performance is no longer within the performance threshold range; the multiple AI models include the new AI model; Determine a first conflict subset to which the first AI model belongs, and other network nodes to which other AI models in the first conflict subset belong; the other AI models are AI models in the first conflict subset other than the first AI model; Sending a decision update request to the other network node; obtaining, from the other network node according to the decision update request, other sub-states of the other AI model when running to the first moment; Determining a first hybrid strategy using the first AI model and based on the first substate and the other substates when the first AI model runs to the first moment; When it is determined according to the first hybrid strategy that the first model performance corresponding to the first AI model is within the performance threshold range, and prompt information is received that the other model performances determined according to other strategies of the other AI models are within the performance threshold range, the first sub-strategy is issued to the physical environment, so that when the physical environment runs to the first moment, the first sub-strategy is executed in the physical environment; the first sub-strategy is the strategy corresponding to the first AI model in the first hybrid strategy.

3. The method according to claim 2, characterized in that The determining, based on the results of running the multiple AI models on the multiple sets of input data, the first AI model whose model performance is no longer within the performance threshold range includes: determining model performance of the multiple AI models based on an operation result of each AI model in the multiple AI models on each set of input data; wherein each set of input data is a set of input data corresponding to each AI model in the multiple sets of input data; When a first performance among the multiple model performances is no longer within the performance threshold range, determine a first AI model corresponding to the first performance.

4. The method according to claim 2, characterized in that The determining the first conflict subset to which the first AI model belongs, and other network nodes to which other AI models in the first conflict subset belong, includes: Determining a first conflict subset to which the first AI model belongs; Obtaining a conflict data set; the conflict data set records a plurality of conflict subsets, AI models constituting the plurality of conflict subsets, and network nodes corresponding to the AI ​​models; In the conflict data set, other network nodes to which other AI models in the first conflict subset belong are searched.

5. The method according to claim 2, characterized in that The sending of the first sub-strategy to the physical environment, when it is determined according to the first hybrid strategy that the first model performance corresponding to the first AI model is within the performance threshold range, and prompt information is received indicating that the performance of other models determined according to other strategies of the other AI models is within the performance threshold range, includes: Searching, from the first hybrid strategy, a first sub-strategy corresponding to the first AI model; Executing the first sub-strategy to obtain a first execution result; When the first model performance determined according to the first operating result is within the performance threshold range, and prompt information is received that other performances corresponding to other operating results obtained by other network nodes running other strategies are also within the performance threshold range, the first sub-strategy is sent to the physical environment.

6. The method according to claim 2, characterized in that After determining the first hybrid strategy using the first AI model and according to the first substate and the other substates when the first AI model runs to the first moment, the method further includes: When the performance of the first model is not within the performance threshold range, or when prompt information is received that the other performance is not within the performance threshold range, a retraining request is sent to the network management node; the first AI model is retrained according to the retraining request using the global digital twin module in the network management node to obtain a second AI model whose performance is within the performance threshold range, and a second hybrid strategy obtained based on the second AI model; receiving the second AI model and the second hybrid strategy; Searching for a second sub-strategy corresponding to the first sub-state from the second hybrid strategy; Running the second sub-strategy to obtain a second running result; When the performance of the second AI model determined according to the second operating result is within the performance threshold range, and prompt information is received that the performance of other second AI models corresponding to the second other operating result determined by the other network node according to the second AI model is within the performance threshold range, when the physical environment runs to the first moment, the second sub-strategy is run using the second AI model in the physical environment.

7. The method according to claim 2, characterized in that After determining the first AI model whose model performance is no longer within the performance threshold range, and the first moment when the model performance is no longer within the performance threshold range, the method further includes: In the absence of a conflicting subset including the first AI model, inputting the first sub-state into the first AI model to obtain a third sub-strategy; Running the third sub-strategy to obtain a third running result; When the first model performance of the first AI model determined according to the third operating result is within the performance threshold range, the third sub-strategy is issued to the physical environment, so that when the physical environment runs to the first moment, the third sub-strategy is executed in the physical environment.

8. The method according to claim 7, characterized in that After executing the third sub-strategy and obtaining a third execution result, the method further includes: If the first model performance of the first AI model determined according to the third running result is not within the performance threshold range, retrain the first AI model to obtain a third AI model; When the model performance corresponding to the third AI model is within the performance threshold range, inputting the first sub-state into the third AI model to obtain a fourth sub-strategy; Executing the fourth sub-strategy to obtain a fourth operation result; When the model performance of the third AI model determined according to the fourth operating result is within the performance threshold, the fourth sub-strategy is issued to the physical environment, so that when the physical environment runs to the first moment, the fourth sub-strategy is executed in the physical environment.

9. An information processing device, characterized in that The device comprises: A sending unit is configured to, upon receiving the AI ​​model to be used, send the AI ​​model to be used to a network management node, so as to perform conflict verification on the AI ​​model to be used using a global digital twin module in the network management node; the network management node includes multiple groups of AI models in multiple network nodes; and the multiple network nodes correspond one-to-one to the multiple groups of AI models; a receiving unit, configured to receive, when the network management node determines that the AI ​​model to be used conflicts with some of the AI ​​models in the multiple groups of AI models, a new AI model transmitted by the network management node, where the new AI model is an AI model obtained after the network management node successfully performs merged training on the AI ​​model to be used and the partial AI models; the partial AI model is an AI model in the multiple groups of AI models whose model performance is no longer within a performance threshold range during merged training; A replacement unit is configured to replace the AI ​​model to be used with the new AI model.

10. An information processing device, characterized in that: The device comprises: A memory, a processor, and a communication bus, wherein the memory communicates with the processor via the communication bus, the memory stores an information processing program executable by the processor, and when the information processing program is executed, the method according to any one of claims 1 to 8 is performed by the processor.

11. A storage medium having a computer program stored thereon, applied to an information processing device, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

12. A computer program product comprising a computer program, characterized in that The computer program implements the method according to any one of claims 1 to 8 when executed by a processor.

Images