Business processing method and device, communication equipment and readable storage medium

Abstract

The invention discloses a business processing method and device, communication equipment and a readable storage medium, and belongs to the technical field of artificial intelligence. The service processing method in the embodiment of the invention comprises the following steps: a first network function constructs a first network twin instance corresponding to a physical network; and performing model training in the first network twinning instance to obtain a first model for generating a service orchestration strategy. Therefore, the model for generating the service arrangement strategy can be effectively obtained.

Description

Technical Field

[0001] This application belongs to the technical field of artificial intelligence, and particularly relates to a service processing method, apparatus, communication device, and readable storage medium. Background Art

[0002] Considering that the computing power of mobile terminals is limited, but they have the characteristics of low latency and mobility requirements for computing, the concept of a mobile computing power network has been proposed in related technologies, that is, the mobile network will no longer only provide connection services, but will provide connection + computing services that integrate network and computing power. In order to determine a reasonable task scheduling and resource allocation strategy, an artificially trained artificial intelligence (AI) model is usually used to obtain the service scheduling strategy. In this case, how to effectively obtain a model for generating the service scheduling strategy is an urgent problem to be solved currently. Summary of the Invention

[0003] The purpose of the embodiments of this application is to provide a service processing method, apparatus, communication device, and readable storage medium to solve the problem of how to effectively obtain a model for generating the service scheduling strategy.

[0004] To solve the above technical problems, this application is implemented as follows:

[0005] In a first aspect, a service processing method is provided, which is applied to a first network function and includes:

[0006] The first network function constructs a first network twin instance corresponding to the physical network;

[0007] The first network function performs model training in the first network twin instance to obtain a first model for generating the service scheduling strategy.

[0008] In a second aspect, a service processing method is provided, which is applied to a third network function and includes:

[0009] The third network function receives a service request, and the service request includes service requirement information;

[0010] The third network function obtains relevant information of a second model for executing tasks according to the service requirement information;

[0011] The third network function processes the service requirement information and the relevant information of the second model by using the first model to obtain a second service scheduling strategy; wherein, the first model is trained in a network twin instance corresponding to the physical network and is used to generate the service scheduling strategy.

[0012] In a third aspect, a service scheduling apparatus is provided, which is applied to the first network function and includes:

[0013] A first construction module for constructing a first network twin instance corresponding to a physical network;

[0014] A training module for performing model training in the first network twin instance to obtain a first model for generating a service orchestration policy.

[0015] In a fourth aspect, a service orchestration apparatus is provided, which is applied to a third network function and includes:

[0016] A second receiving module for receiving a service request, where the service request includes service requirement information;

[0017] A second obtaining module for obtaining relevant information of a second model for performing a task according to the service requirement information;

[0018] A processing module for processing the service requirement information and the relevant information of the second model by using the first model to obtain a second service orchestration policy; the first model is trained in a network twin instance corresponding to a physical network and is used for generating a service orchestration policy.

[0019] In a fifth aspect, a communication device is provided, including a processor, a memory, and a program or instruction stored on the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the method described in the first aspect or the steps of the method described in the second aspect are implemented.

[0020] In a sixth aspect, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect or the steps of the method described in the second aspect are implemented.

[0021] Through the solution of the embodiments of the present application, model training can be performed in a network twin instance corresponding to a physical network, so as to obtain a model for generating a service orchestration policy without affecting the model performance and avoid the influence of physical network state changes on the model training process, and effectively obtain a model for generating a service orchestration policy. Description of the Drawings

[0022] Figure 1 is a schematic architecture diagram of a service orchestration system provided by an embodiment of the present application;

[0023] Figure 2 is a flowchart of a service processing method provided by an embodiment of the present application;

[0024] Figure 3 is a flowchart of another service processing method provided by an embodiment of the present application;

[0025] Figure 4 is the flowchart of the AI model training process in Embodiment 1 of the present application;

[0026] Figure 5 is the flowchart of the access and execution process of the AI service request in Embodiment 2 of the present application;

[0027] Figure 6 is the flowchart of the evaluation and verification process of the AI model in Embodiment 3 of the present application;

[0028] Figure 7 is the structural schematic diagram of a service orchestration device provided in an embodiment of the present application;

[0029] Figure 8 is the structural schematic diagram of another service orchestration device provided in an embodiment of the present application;

[0030] Figure 9 is the structural schematic diagram of a communication device provided in an embodiment of the present application. Detailed implementation manners

[0031] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, rather than all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0032] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such used data may be interchanged under appropriate circumstances so that the embodiments of the present application can be implemented in an order different from those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same category, and the number of objects is not limited. For example, the first object may be one or multiple. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / " generally represents an "or" relationship between the associated objects before and after.

[0033] To facilitate the understanding of the embodiments of the present application, the following content is first described.

[0034] Such as Figure 1As shown in the figure, an embodiment of the present application provides a schematic architecture diagram of a service orchestration system, which at least includes a digital twin, a model storage function, an intelligent service orchestration function, a network data platform function, a network access management unit, a network control unit, etc. Among them, the network data platform function is used to obtain the computing and networking resource information on the data plane, including but not limited to the statistical computing resource situation, the load situation of computing nodes, and the network situation, etc.; the intelligent service orchestration function is used for receiving computing tasks, comprehensively analyzing computing tasks and computing and networking information, formulating service orchestration strategies, and interacting with the digital twin, etc.; the model storage function is used to store trained service models (such as intelligent service models); the network control unit is used to control computing nodes to execute computing tasks according to the received service orchestration strategy, etc. In this way, a digital twin can be constructed to interact with the core network, and a bridge can be built between the underlying network and the core network by using digital twin technology to have real-time change perception characteristics.

[0035] Optionally, Figure 1 The digital twin shown in the figure at least includes a data acquisition module, a network twin instance orchestration module, a network twin instance, a model training module, an evaluation module, etc. Among them, the data acquisition module is used to collect network data to generate a network twin instance (or called a digital twin instance), the network twin instance orchestration module is used to orchestrate the processes and resources for generating a network digital mirror, the network twin instance is a twin object entity corresponding to the physical network orchestrated based on service requirements, and the model training module and the evaluation module are modules for performing typical service control on the network twin instance, and are respectively used for training a model (such as an AI model) for generating a service orchestration strategy and strategy evaluation. The digital twin can be used to generate a network twin instance that is a real-time mirror of the physical network, and perform iterative training and continuous optimization on the model for generating a service orchestration strategy in the network twin instance. At the same time, it performs strategy evaluation on the service orchestration strategy (or simply called the orchestration strategy) generated by the model; the corresponding process can include: First, the AI model for generating a service orchestration strategy performs iterative optimization in the digital twin until a global optimal strategy is obtained and the AI model training is completed; Then, the trained AI model is called by the intelligent service orchestration function to generate a service orchestration strategy online. At the same time, as the network state and service traffic change, the digital twin continuously updates the network twin instance, evaluates the service orchestration strategy generated by the AI model, further continuously optimizes the AI model, and regularly sends the new AI model to the model storage function for the intelligent service orchestration function to call.

[0036] Optionally, in the embodiments of the present application, the AI algorithm can be used for generating the orchestration strategy of the intelligent service orchestration function, where the orchestration strategy includes a task offloading strategy and a model deployment strategy. The task offloading strategy may include the matching relationship between computing tasks and computing nodes, the priority of the task queue, whether to enable lightweight models such as pruning and early exit (such as when the computing accuracy is not high or the available resources are limited, lightweight models such as pruning and early exit can be enabled), etc. The model deployment strategy may include the spatial deployment strategy and the temporal deployment strategy of the model (such as an AI model) for executing computing tasks. The spatial deployment strategy determines the spatial deployment of the corresponding model in the network, so that the model is distributed on computing nodes at an appropriate distance from the corresponding service request node, thereby reducing the overall inference latency between the service and the computing nodes and reducing the overall energy consumption of the system while ensuring the service quality requirements; the temporal deployment strategy can deploy the models with more service requests in the current time period to fast memories such as node memory / video memory and keep them in the "running state"; and / or, deploy the models with fewer service requests in the current time period to long-term memories such as hard disks and enter the "storage state"; and / or, delete the models with no service requirements for a long time from the memory of the node, mark them as the "removed state", and redeploy other models. Preferably, the above AI algorithm can be an AI algorithm based on reinforcement learning.

[0037] As Figure 1 shown, the interfaces between network functions and digital twins and between network functions may include:

[0038] (1) Interface a between the model storage function and the digital twin, which is used for the model training module in the digital twin to send the trained AI model for generating the orchestration strategy to the model storage function;

[0039] (2) Interface b between the intelligent service orchestration function and the digital twin, which is used for the intelligent service orchestration function to request the digital twin to verify the generated orchestration strategy. After the orchestration strategy is verified in the network twin instance, the evaluation module evaluates the evaluation results such as the overall latency and resource utilization rate of the current orchestration strategy, and then judges whether the accuracy of the AI model exceeds the due date and needs to start model optimization;

[0040] (3) Interface c between the network data platform function and the digital twin, which is used for the network data platform function to send various network information to the digital twin to build a mirror instance of the physical network;

[0041] (4) Interface between the intelligent service orchestration function and the model storage function, which is used for the model storage function to transfer the AI model parameter group to the intelligent service orchestration function.

[0042] The following will, in conjunction with the accompanying drawings, elaborate in detail on the service processing method, apparatus, communication device, and readable storage medium provided by the embodiments of the present application through specific embodiments and their application scenarios.

[0043] Please refer to Figure 2 , Figure 2 which is a flowchart of a service processing method provided by an embodiment of the present application. This method is applied to a first network function, such as the digital twin shown above Figure 1 or other network functions that can achieve similar functions. As Figure 2 shown, the method includes the following steps:

[0044] Step 21: The first network function constructs a first network twin instance corresponding to the physical network;

[0045] Step 22: The first network function performs model training in the first network twin instance to obtain a first model for generating service orchestration policies.

[0046] In the embodiments of the present application, the first model can be an AI model.

[0047] Optionally, when executing Step 21, a first network twin instance corresponding to or mirrored from the physical network can be constructed according to the relevant information of the physical network. This first network twin instance is a network twin instance for model training.

[0048] Optionally, the service orchestration policy may include but is not limited to at least one of the following:

[0049] Task offloading policy, which may include, for example, the matching relationship between computing tasks and computing nodes, task queue priorities, whether to enable lightweight models such as pruning and early termination (such as when the computing accuracy is not high or the available resources are limited, lightweight models such as pruning and early termination can be enabled), etc.;

[0050] Spatial deployment policy for the second model used to execute tasks; based on this spatial deployment policy, the spatial deployment of the second model in the network can be determined, so that the second model is distributed on computing nodes at an appropriate distance from the corresponding service request nodes, thereby reducing the overall inference latency between the service and the computing nodes and reducing the overall energy consumption of the system while ensuring the service quality requirements;

[0051] Temporal deployment policy for the second model used to execute tasks; based on this temporal deployment policy, the second model with more service requests in the current time period can be deployed to fast memories such as node memory / video memory and maintained in a "running state"; and / or, the second model with fewer service requests in the current time period can be deployed to long-term memories such as hard disks and enter a "stored state"; and / or, the second model with no service requirements for a long time can be deleted from the memory of the node.

[0052] It should be noted that the second model can be an AI model. The functions of the second model are different from those of the above-mentioned first model. For example, the second model is used to execute computing tasks, and the first model is used to generate business orchestration strategies. With the above-mentioned spatial deployment strategy and temporal deployment strategy, models can be dynamically deployed according to the relevant information of actual computing nodes, thereby reducing business latency and improving user experience.

[0053] Through the solution of the embodiments of the present application, model training can be carried out in the network twin instance corresponding to the physical network. Thus, without affecting the model performance, a model for generating business orchestration strategies can be obtained, and the influence of physical network state changes on the model training process can be avoided, effectively obtaining a model for generating business orchestration strategies.

[0054] Optionally, the business processing method in the embodiments of the present application may further include:

[0055] The first network function generates a first business orchestration strategy using the first model in the first network twin instance and executes the first business orchestration strategy, that is, executes the orchestration strategy in the mirror network;

[0056] The first network function evaluates the execution result of the first business orchestration strategy to obtain a first evaluation result; for example, the execution result may include computing latency, accuracy, etc.;

[0057] When the first evaluation result does not meet the first requirement, the first network function iteratively optimizes the first model, that is, determines that the first model cannot meet the corresponding business requirements and starts the training and tuning of the model; or, when the first evaluation result meets the first requirement, the first network function determines that the training of the first model is completed; the first requirement can be set based on actual needs, such as including that the execution result (such as computing latency, computing accuracy, etc.) exceeds a set threshold.

[0058] In this way, through the evaluation process, a model for generating business orchestration strategies that meets the requirements can be trained.

[0059] Optionally, the business processing method in the embodiments of the present application may further include:

[0060] The first network function sends the first model to the second network function for storage. The second network function is, for example, the model storage function shown above Figure 1 or other network functions that can achieve similar functions. In this way, by storing the first model in the second network function (such as the model storage function), it is convenient for intelligent business orchestration functions, etc. to call the first model to generate business orchestration strategies.

[0061] Optionally, the service processing method in the embodiments of the present application may further include:

[0062] The first network function receives a second service orchestration policy sent by the third network function; the second service orchestration policy is generated by using a first model for services in the physical network (i.e., actual services), and the third network function is, for example, the intelligent service orchestration function shown above Figure 1 or other network functions that can implement similar functions;

[0063] The first network function executes the second service orchestration policy and evaluates the execution result of the second service orchestration policy to obtain a second evaluation result; for example, the execution result may include computing delay, etc.;

[0064] When the second evaluation result does not meet the second requirement, the first network function iteratively optimizes the first model, that is, determines that the first model cannot meet the corresponding service requirements, starts the training and tuning of the model, and then sends the tuned model to the model storage function for storage; or, when the second evaluation result meets the second requirement, the first network function determines that the first model can meet the service requirements; the second requirement can be set based on actual requirements, for example, including that the execution result (such as computing delay, computing accuracy, etc.) exceeds a set threshold.

[0065] In this way, through the interaction between the first network function and the third network function, the first model used to generate the service orchestration policy can be continuously trained and optimized, thereby improving the real-time optimality and accuracy of the service orchestration policy, and ensuring the performance of the service orchestration policy.

[0066] Optionally, before executing the second service orchestration policy, the service processing method in the embodiments of the present application may further include:

[0067] The first network function obtains real-time network data from the fourth network function; the fourth network function is, for example, the network data platform function shown above Figure 1 or other network functions that can implement similar functions

[0068] The first network function constructs a second network twin instance corresponding to the physical network according to the real-time network data. This second network twin instance is a network twin instance for executing the policy.

[0069] The above-mentioned execution of the second service orchestration policy includes: the first network function executes the second service orchestration policy in the second network twin instance.

[0070] In this way, it is possible to mirror the network twin instance corresponding to the current physical network and execute the orchestration policy generated for the services in the current physical network, thereby ensuring the correctness of policy deployment and improving the evaluation reliability.

[0071] Please refer to Figure 3 , Figure 3 which is a flowchart of a service processing method provided by an embodiment of the present application. This method is applied to a third network function, and the third network function is, for example, the intelligent service orchestration function shown above Figure 1 or other network functions that can implement similar functions. As Figure 3 shown, the method includes the following steps:

[0072] Step 31: The third network function receives a service request, and the service request includes service requirement information; for example, a terminal can initiate a service request to the third network function.

[0073] Step 32: The third network function obtains relevant information of the second model for performing tasks according to the service requirement information.

[0074] Step 33: The third network function processes the service requirement information and the relevant information of the second model by using the first model to obtain a second service orchestration policy; the first model is trained in the network twin instance corresponding to the physical network and is used to generate service orchestration policies.

[0075] In an embodiment of the present application, the second model may be an AI model. The first model may be an AI model. The functions of the second model and the first model are different. For example, the second model is used to perform computing tasks, and the first model is used to generate service orchestration policies.

[0076] Optionally, the service requirement information may include, but is not limited to, at least one of the following: service identification ID, service type, maximum value of the end-to-end inference latency allowed by the service, minimum value of the network bandwidth required by the service, degree of quantization of service data (such as data compression degree), expected accuracy of the inference result, deployment method of service computing (such as centralized or distributed), etc.

[0077] Optionally, the relevant information of the second model is used to describe the model performance and may include, but is not limited to, at least one of the following: network architecture name, network structure parameters, relationship between model layers, parameters of each layer of the pre-trained model, training hyperparameters, model accuracy, service configuration description file, etc.

[0078] Optionally, the second service orchestration policy may include, but is not limited to, at least one of the following:

[0079] Task offloading strategies, which may include, for example, the matching relationship between computing tasks and computing nodes, the priority of task queues, whether to enable lightweight models such as pruning and early termination (e.g., when the computing accuracy is not high or the available resources are limited, lightweight models such as pruning and early termination can be enabled), etc.;

[0080] The spatial deployment strategy of the second model for executing tasks; based on this spatial deployment strategy, the spatial deployment of the second model in the network can be determined, so that the second model is distributed on computing nodes at an appropriate distance from the corresponding service request nodes, thereby reducing the overall inference latency between the service and the computing nodes and reducing the overall energy consumption of the system while ensuring the service quality requirements;

[0081] The temporal deployment strategy of the second model for executing tasks; based on this temporal deployment strategy, the second model with more service requests in the current time period can be deployed to fast memories such as node memory / video memory and maintained in the "running state"; and / or, the second model with fewer service requests in the current time period can be deployed to long-term memories such as hard disks and enter the "storage state"; and / or, the second model with no service requirements for a long time can be deleted from the memory of the node.

[0082] Through the solution of the embodiments of the present application, a service orchestration strategy can be obtained by using the model trained in the network twin instance corresponding to the physical network, thereby avoiding the influence of physical network state changes on the model training process and effectively obtaining a model for generating the service orchestration strategy.

[0083] Optionally, before the above-mentioned use of the first model to process the service demand information and the relevant information of the second model, the service processing method in the embodiments of the present application may further include:

[0084] The third network function obtains the first model from the second network function. The second network function is, for example, the above-mentioned Figure 1 model storage function shown or other network functions that can implement similar functions, and can receive the first model from the first network function (such as a digital twin) for storage, so as to facilitate the intelligent service orchestration function and other functions to call the first model to generate a service orchestration strategy.

[0085] Optionally, after the above-mentioned obtaining of the second service orchestration strategy, the service processing method in the embodiments of the present application may further include:

[0086] The third network function sends the second service orchestration strategy to the first network function; wherein, the second service orchestration strategy is used to evaluate whether the first model can meet the service requirements in the network twin instance constructed according to real-time network data. The first network function is, for example, the above-mentioned Figure 1The digital twin or other network functions capable of achieving similar functions as shown. In this way, the first model used to generate the service orchestration policy can be continuously trained and optimized, thereby improving the real-time optimality and accuracy of the service orchestration policy and ensuring the performance of the service orchestration policy.

[0087] Optionally, after obtaining the second service orchestration policy, the service processing method in the embodiments of the present application may further include:

[0088] The third network function sends the second service orchestration policy to the network control unit, and the second service orchestration policy is used for the network control unit to control the computing node to execute tasks; after the execution is completed, the calculation result can be fed back to the service request node.

[0089] The present application will be described below in conjunction with specific embodiments.

[0090] Embodiment 1

[0091] In this Embodiment 1, the training process of the AI model for generating the service orchestration policy is mainly described, as Figure 4 shown, including at least the following steps:

[0092] Step 1: The intelligent service orchestration function sends a request to the digital twin through interface b (as Figure 1 shown) to request an AI model (or simply referred to as a policy generation model) for generating the service orchestration policy.

[0093] Step 2: The network twin instance orchestration module of the digital twin obtains network historical data through the data acquisition module according to the request for generating the orchestration policy AI model, such as including historical computing network resource information, historical computing task information, etc., and constructs a network twin instance according to the obtained network historical data.

[0094] Step 3: The model training module of the digital twin trains an initial AI model (such as a reinforcement learning model) in the network twin instance, and issues a training control instruction for the policy generation model, performs orchestration policy generation and policy distribution in the network twin instance, and the evaluation module evaluates the effect of policy execution; when the evaluation result meets the set requirements, the training ends, and when the evaluation result does not meet the set requirements, continue the training iteration until the generated service orchestration policy meets the set requirements, and the model iterative training is completed.

[0095] Step 4: After the model training is completed, the digital twin sends the trained AI model for generating the orchestration policy to the model storage function through interface a (as Figure 1 shown) for storage; during subsequent service execution, the intelligent service orchestration function obtains the AI model for generating the orchestration policy from the model storage function.

[0096] Example 2

[0097] In this Example 2, the access and execution processes of AI service requests are mainly described. As Figure 5 shown, it includes at least the following steps:

[0098] Step 1: The end user registers;

[0099] Step 2: The end user initiates a service request, sends the AI service request to the network access management unit through Non-Access Stratum (NAS) signaling, and the network access management unit forwards this service request to the intelligent service orchestration function. Among them, the service request carries a parameter group for describing the AI service requirements, and the parameter group may include an AI service ID, an AI service type, a maximum value of the end-to-end inference delay allowed by the AI service, a minimum value of the network bandwidth required by the AI service, the degree of quantization of service data (data compression degree), the expected accuracy of the AI inference result, the deployment method of AI calculation (such as centralized or distributed), etc.;

[0100] Step 3: The intelligent service orchestration function obtains computing network resource information through the network data platform, such as including but not limited to the statistical computing resource situation, the load situation of computing nodes, and the network situation, etc.;

[0101] Step 4: The intelligent service orchestration function requests AI model information for task execution from the model storage function based on the AI service requirements;

[0102] Step 5: The model storage function returns a corresponding model parameter group to the intelligent service orchestration function. The specific parameters may include the name of the neural network architecture in the model, network structure parameters, the relationship between model layers, the parameters of each layer of the pre-trained model, training hyperparameters, model accuracy, service configuration description files, etc.;

[0103] Step 6: The intelligent service orchestration function calls an AI model for generating an orchestration strategy. Its input parameters include AI service requirement parameters, AI model information for task execution, and computing network resource information. After being processed by the AI model for generating an orchestration strategy, a service orchestration strategy is generated, and the strategy content may include a model deployment strategy and a task offloading strategy;

[0104] Step 7: The intelligent service orchestration function sends the generated service orchestration strategy to the network control unit, and the network control unit controls the corresponding computing nodes to complete the calculation execution, and feeds back the calculation result to the end user after the execution is completed;

[0105] Step 8: After the business execution is completed, the computing node can adjust the model status according to its model time deployment policy; for example, when the access popularity is lower than the threshold, the model status changes from the "running state" to the "stored state" until the "removed state".

[0106] Embodiment 3

[0107] In this Embodiment 3, the evaluation and verification process of the AI model for generating the business orchestration policy is mainly described, as Figure 6 shown, including at least the following steps:

[0108] Step 1: Based on the constructed network twin instance, the digital twin obtains real-time network status information from the network data platform through interface c (as Figure 1 shown), and the network twin instance orchestration module controls the generation of the network twin instance;

[0109] Step 2: The intelligent business orchestration function uploads the business orchestration policy generated based on the actual business to the digital twin through interface b (as Figure 1 shown), then the business orchestration policy is executed in the network twin instance, and the evaluation module evaluates the policy operation result;

[0110] Step 3: When the evaluation result of the evaluation module does not meet the set requirements, it means that the corresponding AI model cannot meet the current business needs and needs to be optimized; while when the evaluation result meets the set requirements, it means that the current policy generation AI model can continue to be used.

[0111] Step 4: When the evaluation result does not meet the set requirements, the model training module of the digital twin starts the training and optimization of the model, sends the current AI model to the network twin instance for iterative training, and the training and evaluation processes are the same as the initial training process;

[0112] Step 5: After the model training is completed, the digital twin sends the trained and updated AI model for generating the orchestration policy to the model storage function through interface a (as Figure 1 shown); during subsequent business execution, the intelligent business orchestration function obtains the new AI model for generating the orchestration policy from the model storage function.

[0113] It should be noted that for the business processing method provided in the embodiments of the present application, the execution subject can be a business orchestration device, or a control module in the business orchestration device for executing the business processing method. In the embodiments of the present application, the business orchestration device executing the business processing method is taken as an example to illustrate the business orchestration device provided in the embodiments of the present application.

[0114] Please refer to Figure 7 , Figure 7It is a schematic structural diagram of a service orchestration device provided by an embodiment of the present application. This device is applied to a first network function, such as the Figure 1 digital twin shown above or other network functions that can achieve similar functions. As Figure 7 shown, the service orchestration device 70 includes:

[0115] A first construction module 71, configured to construct a first network twin instance corresponding to a physical network;

[0116] A training module 72, configured to perform model training in the first network twin instance to obtain a first model for generating a service orchestration policy.

[0117] Optionally, the service orchestration device 70 further includes:

[0118] A first execution module, configured to generate a first service orchestration policy using the first model in the first network twin instance, and execute the first service orchestration policy; evaluate the execution result of the first service orchestration policy to obtain a first evaluation result; and, when the first evaluation result does not meet the first requirement, perform iterative optimization on the first model; or, when the first evaluation result meets the first requirement, determine that the training of the first model is completed.

[0119] Optionally, the service orchestration device 70 further includes:

[0120] A first sending module, configured to send the first model to a second network function for storage.

[0121] Optionally, the service orchestration device 70 further includes:

[0122] A first receiving module, configured to receive a second service orchestration policy sent by a third network function; the second service orchestration policy is generated using the first model for services in the physical network;

[0123] A second execution module, configured to execute the second service orchestration policy, and evaluate the execution result of the second service orchestration policy to obtain a second evaluation result; when the second evaluation result does not meet the second requirement, perform iterative optimization on the first model; or, when the second evaluation result meets the second requirement, determine that the first model can meet the service requirements.

[0124] Optionally, the service orchestration device 70 further includes:

[0125] A first acquisition module, configured to obtain real-time network data from a fourth network function before executing the second service orchestration policy;

[0126] A second construction module, configured to construct a second network twin instance corresponding to a physical network according to the real-time network data;

[0127] The second execution module is specifically configured to: execute the second service orchestration policy in the second network twin instance.

[0128] Optionally, the second service orchestration policy includes at least one of the following:

[0129] A task offloading policy;

[0130] A spatial deployment policy for a second model used to execute tasks;

[0131] A temporal deployment policy for a second model used to execute tasks.

[0132] The service orchestration device 70 according to the embodiments of the present application can implement each process of the method embodiment shown above, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here. Figure 2 Please refer to

[0133] Please refer to Figure 8 , Figure 8 FIG. is a schematic structural diagram of a service orchestration device provided by an embodiment of the present application. This device is applied to a third network function, and this third network function is, for example, the intelligent service orchestration function shown above Figure 1 or other network functions that can implement similar functions. As Figure 8 shown, the service orchestration device 80 includes:

[0134] A second receiving module 81, configured to receive a service request, where the service request includes service requirement information;

[0135] A second obtaining module 82, configured to obtain relevant information of a second model used to execute tasks according to the service requirement information;

[0136] A processing module 83, configured to process the service requirement information and the relevant information of the second model by using a first model to obtain a second service orchestration policy; the first model is trained in a network twin instance corresponding to a physical network and is used to generate a service orchestration policy.

[0137] Optionally, the second obtaining module 82 is further configured to: obtain the first model from a second network function before processing the service requirement information and the relevant information of the second model by using the first model.

[0138] Optionally, the service orchestration device 80 further includes:

[0139] A second sending module, configured to send the second service orchestration policy to a first network function; the second service orchestration policy is used to evaluate whether the first model can meet service requirements in a network twin instance constructed based on real-time network data.

[0140] Optionally, the second service orchestration policy includes at least one of the following:

[0141] A task offloading policy;

[0142] A spatial deployment policy for a second model used to execute tasks;

[0143] A temporal deployment policy for a second model used to execute tasks.

[0144] Optionally, the service orchestration device 80 further includes:

[0145] A third sending module, configured to send the second service orchestration policy to a network control unit, and the second service orchestration policy is used for the network control unit to control a computing node to execute a task.

[0146] Optionally, the service requirement information includes at least one of the following:

[0147] Service identifier, service type, maximum value of the end-to-end inference latency allowed by the service, minimum value of the network bandwidth required by the service, degree of quantization of service data, desired precision of the inference result, deployment method of service computing.

[0148] Optionally, the related information of the second model includes at least one of the following:

[0149] Network architecture name, network structure parameters, relationship between model layers, parameters of each layer of the pre-trained model, training hyperparameters, model accuracy, service configuration description file.

[0150] The service orchestration device 80 in the embodiments of the present application can implement each process of the method embodiment shown above, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here. Figure 3 To avoid repetition, it will not be elaborated here.

[0151] Optionally, as Figure 9 shown, the embodiments of the present application further provide a communication device 90, including a processor 91, a memory 92, a program or instruction stored on the memory 92 and executable on the processor 91. When the program or instruction is executed by the processor 91, each process of the above-mentioned service processing method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be elaborated here.

[0152] The embodiments of the present application also provide a readable storage medium, on which a program or instructions are stored. When the program or instructions are executed by a processor, each process of the above-mentioned business processing method embodiments can be implemented and the same technical effects can be achieved. To avoid repetition, details are not described here again.

[0153] Computer-readable media include both permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0154] It should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including that element.

[0155] The serial numbers of the embodiments of the present application above are only for description and do not represent the superiority or inferiority of the embodiments.

[0156] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions for causing a service classification device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present application.

[0157] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A service processing method, characterized in that, Including: The first network function constructs a first network twin instance corresponding to the physical network; The first network function performs model training in the first network twin instance to obtain a first model for generating service orchestration policies.

2. The method according to claim 1, wherein The method further includes: The first network function generates a first service orchestration policy using the first model in the first network twin instance and executes the first service orchestration policy; The first network function evaluates the execution result of the first service orchestration policy to obtain a first evaluation result; When the first evaluation result does not meet the first requirement, the first network function iteratively optimizes the first model; or when the first evaluation result meets the first requirement, the first network function determines that the training of the first model is completed.

3. The method according to claim 1, wherein The method further includes: The first network function distributes the first model to the second network function for storage.

4. The method according to any one of claims 1 to 3, characterized in that The method further includes: The first network function receives a second service orchestration policy sent by the third network function; wherein, the second service orchestration policy is generated using the first model for services in the physical network; The first network function executes the second service orchestration policy and evaluates the execution result of the second service orchestration policy to obtain a second evaluation result; When the second evaluation result does not meet the second requirement, the first network function iteratively optimizes the first model; or when the second evaluation result meets the second requirement, the first network function determines that the first model can meet the service requirements.

5. The method according to claim 4, characterized in that, Before executing the second service orchestration policy, the method further includes: The first network function obtains real-time network data from the fourth network function; The first network function constructs a second network twin instance corresponding to the physical network based on the real-time network data; Wherein, executing the second service orchestration policy includes: The first network function executes the second service orchestration policy in the second network twin instance.

6. The method according to claim 4, characterized in that, The second service orchestration policy includes at least one of the following: Task offloading policy; Spatial deployment policy of the second model for executing tasks; Temporal deployment policy of the second model for executing tasks.

7. A service processing method, characterized in that, Including: The third network function receives a service request, and the service request includes service requirement information; The third network function obtains relevant information of the second model for executing tasks according to the service requirement information; The third network function processes the service requirement information and the relevant information of the second model using the first model to obtain a second service orchestration policy; wherein, the first model is trained in a network twin instance corresponding to the physical network and is used to generate service orchestration policies.

8. The method according to claim 7, wherein Before processing the service requirement information and the relevant information of the second model using the first model, the method further includes: The third network function obtains the first model from the second network function.

9. The method according to claim 7, characterized in that After obtaining the second service orchestration policy, the method further includes: The third network function sends the second service orchestration policy to the first network function; wherein, the second service orchestration policy is used to evaluate whether the first model can meet the service requirements in a network twin instance constructed based on real-time network data.

10. The method according to any one of claims 7 to 9, characterized in that The second service orchestration policy includes at least one of the following: Task offloading policy; Spatial deployment policy of the second model for executing tasks; Temporal deployment policy of the second model for executing tasks.

11. The method according to claim 7, wherein After obtaining the second service orchestration policy, the method further includes: The third network function sends the second service orchestration policy to a network control unit, wherein the second service orchestration policy is used for the network control unit to control a computing node to execute a task.

12. The method according to claim 7, wherein The service requirement information includes at least one of the following: Service identifier, service type, maximum value of the end-to-end inference latency allowed by the service, minimum value of the network bandwidth required by the service, quantization degree of service data, desired accuracy of inference results, deployment method of service computing.

13. The method according to claim 7, characterized in that, The related information of the second model includes at least one of the following: Network architecture name, network structure parameters, inter-layer connection relationship of the model, parameters of each layer of the pre-trained model, training hyperparameters, model accuracy, service configuration description file.

14. A service orchestration device, characterized in that, Includes: A first construction module for constructing a first network twin instance corresponding to a physical network; A training module for performing model training in the first network twin instance to obtain a first model for generating a service orchestration policy.

15. A service orchestration device, characterized in that, Includes: A second receiving module for receiving a service request, where the service request includes service requirement information; A second obtaining module for obtaining related information of a second model for executing a task according to the service requirement information; A processing module for processing the service requirement information and the related information of the second model by using the first model to obtain a second service orchestration policy; The first model is trained in a network twin instance corresponding to a physical network and is used to generate a service orchestration policy.

16. A communication device, characterized in that, Includes a processor, a memory, and a program or instruction stored on the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the service processing method according to any one of claims 1 to 6 are implemented, or the steps of the service processing method according to any one of claims 7 to 13 are implemented.

17. A readable storage medium, characterized in that, A program or instruction is stored on the readable storage medium. When the program or instruction is executed by a processor, the steps of the service processing method according to any one of claims 1 to 6 are implemented, or the steps of the service processing method according to any one of claims 7 to 13 are implemented.

Images