Communication network control method and device based on vertical federated learning, and storage medium
By using a method based on vertical federated learning in the communication network to segment and allocate communication tasks, the problem of low resource utilization efficiency of existing communication networks is solved, and efficient communication task execution and resource utilization are achieved.
Patent Information
- Application Number
- CN202510101720.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-22
- Publication Date
- 2025-05-06
AI Technical Summary
Existing communication networks have low resource utilization efficiency when performing communication tasks, making it difficult to cope with complex task requirements and dynamically changing network environments.
Using a communication network control method based on vertical federated learning, the communication task is segmented through the first network element, allocated to the corresponding second network element for processing, and vertical federated learning is performed to improve resource utilization efficiency.
Multiple second network elements are implemented to process communication tasks in parallel, improve the execution efficiency of communication tasks, maintain the confidentiality of the vertical federated learning system, and improve the efficient utilization of communication network resources.
Smart Images
Figure CN119945923A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of communication technology, and in particular to a communication network control method, a computer device and a storage medium based on vertical federated learning. Background Art
[0002] Communication networks usually need to perform multiple communication tasks. Current communication networks often provide fixed communication resources when performing communication tasks, which leads to low resource utilization efficiency of the communication network and low execution efficiency of communication tasks, making it difficult to cope with complex task requirements and dynamically changing network environments. Summary of the invention
[0003] In view of the technical problems faced by current communication networks such as low resource utilization efficiency and low execution efficiency of communication tasks, the purpose of the present invention is to provide a communication network control method, computer device and storage medium based on vertical federated learning.
[0004] On the one hand, an embodiment of the present invention includes a communication network control method based on vertical federated learning, and the communication network control method based on vertical federated learning includes:
[0005] The first network element acquires a communication task;
[0006] The first network element divides the communication task into multiple communication subtasks;
[0007] The first network element sends each of the communication subtasks to the corresponding second network element respectively;
[0008] Each of the second network elements respectively runs its own trainable model, and respectively performs and processes the communication subtask received by itself;
[0009] The first network element is used as a collaborator, and each of the second network elements is used as a participant, and vertical federated learning is performed according to each of the communication subtasks.
[0010] Furthermore, the communication network further includes a third network element and a fourth network element, and the first network element acquiring the communication task includes:
[0011] The third network element obtains the communication task from an external application;
[0012] The third network element queries the fourth network element for available second network elements according to the communication task;
[0013] When the available second network element is found, the third network element sends the communication task to the first network element.
[0014] Furthermore, the first network element divides the communication task to obtain a plurality of communication subtasks, including:
[0015] Obtaining data types corresponding to each of the trainable models;
[0016] The first network element divides the communication task according to each of the data types to obtain a plurality of the communication subtasks.
[0017] Further, the first network element divides the communication task according to each of the data types to obtain a plurality of the communication subtasks, including:
[0018] For any of the data types, determining a portion matching the data type from the communication tasks;
[0019] Slice out the determined part of the communication task to obtain a corresponding communication subtask;
[0020] The communication subtask is assigned to the second network element corresponding to the data type.
[0021] Further, the first network element divides the communication task according to each of the data types to obtain a plurality of the communication subtasks, further comprising:
[0022] Acquire the remaining part of the communication task; the remaining part is the part that does not match any of the data types;
[0023] obtaining the urgency of the remaining portion;
[0024] When the urgency of the remaining part is lower than a threshold, performing data obfuscation processing on the remaining part to obtain a corresponding communication subtask;
[0025] The communication subtask obtained by processing the remaining part is randomly assigned to one of the second network elements.
[0026] Further, the first network element divides the communication task according to each of the data types to obtain a plurality of the communication subtasks, further comprising:
[0027] When the urgency of the remaining part is higher than a threshold, the remaining part is allocated to the first network element for processing.
[0028] Further, allocating the communication subtask to the second network element corresponding to the data type includes:
[0029] For any of the communication subtasks, when the data type corresponding to the communication subtask corresponds to a second network element, assigning the communication subtask to the second network element;
[0030] When the data type corresponding to the communication subtask corresponds to a plurality of the second network elements, the priority of each of the second network elements is determined, and the communication subtask is allocated according to each priority.
[0031] Furthermore, the first network element is used as a collaborator, and each of the second network elements is used as a participant, and vertical federated learning is performed according to each of the communication subtasks, including:
[0032] The first network element distributes respective key information to each of the second network elements;
[0033] For any second network element, the second network element inputs the data from the communication subtask into the trainable model running by itself for processing, obtains intermediate information, and encrypts the intermediate information using the key information allocated to it, to obtain encrypted intermediate information;
[0034] The second network elements exchange the respective encrypted intermediate information;
[0035] For any second network element, the second network element uses all the obtained encrypted intermediate information to train the trainable model, obtains gradient information, and uploads the gradient information to the first network element;
[0036] The first network element decrypts the gradient information using the key information to obtain decrypted gradient information, and sends the decrypted gradient information to the second network element;
[0037] For any of the second network elements, the second network element uses the received decrypted gradient information to update parameters of the trainable model running on itself.
[0038] On the other hand, an embodiment of the present invention also includes a computer device, including a memory and a processor, the memory is used to store at least one program, and the processor is used to load at least one program to execute the communication network control method based on vertical federated learning in the embodiment.
[0039] On the other hand, an embodiment of the present invention also includes a computer-readable storage medium, which stores a program executable by a processor, and the program executable by the processor is used to execute the communication network control method based on vertical federated learning in the embodiment when executed by the processor.
[0040] The beneficial effects of the present invention are as follows: in the communication network control method based on vertical federated learning in the embodiment, the collaborators in the vertical federated learning system slice and divide the communication tasks, so that each second network element can obtain the corresponding communication subtask for processing respectively, and realize the use of multiple second network elements to process the communication tasks in parallel, thereby improving the execution efficiency of the communication tasks; moreover, the first network element divides the communication tasks according to the data characteristics of each participant in the vertical federated learning system, that is, each second network element, so that the communication subtask obtained by each second network element can match the position of the second network element in the vertical federated learning system, thereby maintaining the advantages of the vertical federated learning system such as strong confidentiality; by performing vertical federated learning in the communication network, the trainable data run by each second network element can be trained using data from the communication task, thereby improving the communication data processing performance of each second network element, realizing resource scheduling of each network element in the communication network based on vertical federated learning, and improving the efficient use of communication network resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 A schematic diagram of the steps of a communication network control method based on vertical federated learning in an embodiment;
[0042] Figure 2 It is a flowchart of a communication network control method based on vertical federated learning in an embodiment;
[0043] Figure 3 This is a schematic diagram of the principles of steps S20201-S20208 in the embodiment;
[0044] Figure 4 Schematic diagram of the principles of steps S501-S506 in the embodiment. DETAILED DESCRIPTION
[0045] Terminology explanation:
[0046] VFL: Vertical Federated Learning is a method of federated learning. It is suitable for the case where the user (source) overlap of data owned by multiple participants is high and the feature (type) overlap is low. Participants do not directly exchange data. The data of all participants is used to train the models owned by each participant. It can make full use of data from different aspects for training while protecting data security.
[0047] NWDAF: Network Data Analytics Function, a data-aware analysis network element, automatically senses and analyzes the network based on network data, and participates in the entire life cycle of network planning, construction, operation and maintenance, network optimization, and operation, making the network easier to maintain and control, improving the efficiency of network resource utilization, and enhancing user service experience;
[0048] AF: Application Function, which refers to various apps that can interact directly or indirectly with the 5G network. The so-called indirect interaction means that AF interacts with other functional network elements of 5G through NEF.
[0049] MANO: Management and Network Orchestration, management and network orchestration, plays a vital role in the network function virtualization (NFV) architecture. MANO is a layer composed of multiple functional entities that are responsible for managing and orchestrating cloud infrastructure, resources, and services.
[0050] This embodiment provides a communication network control method based on vertical federated learning. Figure 1 The communication network control method based on vertical federated learning includes at least one round of deployment process, and any round of deployment process includes the following steps:
[0051] The communication network control method based on vertical federated learning in this embodiment can be applied to a communication network. The communication network can be a core network, such as a 5G core network, which includes a first network element, multiple second network elements, a third network element, and a fourth network element. Among them, the first network element can be a VFL server, the second network element can be multiple NWDAFs such as NWDAF1, NWDAF2, NWDAF3, etc., the third network element can be an AF, and the fourth network element can be a MANO.
[0052] In this embodiment, each second network element (NWDAF) can be centrally set or distributedly set, and each second network element (NWDAF) runs a trainable model, for example, NWDAF1 runs trainable model 1, NWDAF2 runs trainable model 2, NWDAF2 runs trainable model 3, etc., so that image processing, speech processing, text map or automatic question and answer functions and applications are executed through the trainable model. The trainable model can specifically be an artificial intelligence model such as a neural network.
[0053] Reference Figure 1 , the communication network control method based on vertical federated learning includes the following steps:
[0054] S1. The first network element obtains the communication task;
[0055] S2. The first network element divides the communication task to obtain multiple communication subtasks;
[0056] S3. The first network element sends each communication subtask to the corresponding second network element;
[0057] S4. Each second network element runs its own trainable model and executes and processes the communication subtasks received by each network element;
[0058] S5. With the first network element as a collaborator and each second network element as a participant, vertical federated learning is performed according to each communication subtask.
[0059] In this embodiment, the process of steps S1-S5 is as follows Figure 2 As shown, it includes the following processes:
[0060] 1. External applications (such as smart city applications and traffic optimization systems) send task requests to AF (application functions), specifying the task type, required computing resources, data volume, time requirements, etc.
[0061] 2. After receiving the task request, AF analyzes the task type, computing resource requirements, data volume, time requirements, etc., and evaluates whether multiple NWDAFs are needed to process the task in parallel.
[0062] 3. AF requests IMANO (IPLOOK MANO management and orchestration function) to obtain the NWDAF information that has been instantiated in the current network, including the client's functions, processing capabilities, supported task types, resource status information, etc.
[0063] 4. IMANO returns basic information of NWDAF, including client ID, function, load status, etc.
[0064] 5. Based on the task requirements, NWDAF resource status information and AF, the complexity of the task is evaluated and the VFL server is requested to perform intelligent task segmentation.
[0065] 6. The VFL server intelligently divides tasks according to the complexity of the task, the client resource status and the system load, while ensuring data privacy. VFL uses local calculation results to update the global model without exchanging original data.
[0066] 7. Based on the algorithm, a list of tasks that need to be split is obtained. The VFL server provides an optimized task allocation list and returns it to the AF.
[0067] 8. Based on the task list provided by the VFL server, the AF concurrently initiates a task subscription (Nnwdaf_Analytics_Subscribe) request to the selected NWDAF to clarify the task type and related data that each client needs to perform.
[0068] 9. Each NWDAF confirms the subscription result and returns feedback to the AF.
[0069] 10. Each NWDAF starts executing according to the assigned task and feeds back the task progress, calculation status, and results to the AF in real time through Nnwdaf_Analytics_Notify.
[0070] 11. IMANO monitors the resource usage of NWDAF clients in real time and reports the data to AF so that AF can make dynamic adjustments during task execution.
[0071] 12. Based on the resource usage and real-time monitoring data fed back by NWDAF, AF dynamically adjusts the priority of tasks. For example, urgent tasks (such as fault detection) may be promoted to high-priority tasks to ensure that critical tasks are executed first.
[0072] 13. Multiple NWDAFs execute tasks in parallel and provide real-time feedback on task progress to AF based on their respective computing resources and task requirements.
[0073] 14. After all NWDAFs complete their tasks, AFs summarize the calculation results of each client and send them to the VFL server.
[0074] 15. The VFL server updates the global model or generates the final result based on the calculation results of each client. VFL aggregates the local models of each client but does not exchange the original data, thus ensuring data privacy.
[0075] 16. VFL returns the summarized global model or final results to AF.
[0076] 17.AF generates reports based on task execution status and returns them to external applications. It also optimizes task scheduling and resource allocation strategies based on task feedback and execution status.
[0077] 18.AF cancels the analysis tasks that are no longer needed through the Nnwdaf_Analytics_UnSubscribe interface and releases related resources.
[0078] 19. Each NWDAF returns the result of resource release.
[0079] 20. AF requests the VFL server to further optimize the task scheduling and resource allocation strategy.
[0080] 21. The VFL server optimizes the task allocation strategy based on historical data and real-time feedback, dynamically adjusts task scheduling and resource utilization, ensures load balancing and avoids overload, and finally returns the optimization results to AF.
[0081] Figure 2 In the process shown, the processing performed by each part is as follows:
[0082] 1. External Applications:
[0083] Send a task request to AF, specifying the task type, required computing resources, data volume, time requirements, etc.
[0084] 2.AF(Application Function):
[0085] After receiving the task request, it analyzes the task requirements and requests client information from the IMANO management and orchestration system.
[0086] Request the VFL server to perform task segmentation and load balancing.
[0087] Initiate a task subscription request to the NWDAF client.
[0088] Dynamically adjust task priorities and resource allocation.
[0089] 3.MANO (management and orchestration function):
[0090] Provides the AF with information about the NWDAF clients registered in the network (functions, loads, etc.).
[0091] Monitor the resource usage of each NWDAF client in real time and provide feedback to AF to help dynamically adjust tasks.
[0092] 4. VFL Server (Virtual Federated Learning):
[0093] Intelligently divide tasks and assign them to NWDAF clients based on task requirements and resource conditions.
[0094] Aggregate client computation results and update the global model, ensuring data privacy.
[0095] 5.NWDAF client (network data analysis function):
[0096] Execute tasks and calculate the results according to the assigned tasks and feed them back to AF.
[0097] Reference Figure 2, when executing step S1, processes 1-4 may be executed. Specifically, an external application (which may be run by a terminal such as a mobile phone used by a user, or by a device such as a server) may send a task request to the third network element (AF), so that the third network element (AF) establishes a communication task. A communication task may be a task executed in response to a task request, and its content may be to establish and maintain a session for an external application, transmit voice content, or transmit image content. The third network element (AF) queries the fourth network element (MANO) for an available second network element. If an available second network element is found, the third network element (AF) sends the communication task to the first network element (VFL server).
[0098] In this embodiment, each NWDAF, such as NWDAF1, NWDAF2, NWDAF3, ..., is an available second network element.
[0099] In step S2, the first network element (VFL server) divides the communication task to obtain multiple communication subtasks. For example, if the communication task requested by the external application is to transmit and process data of types such as text, voice, and image, then the first network element (VFL server) can divide the communication task into communication subtask 1 (transmission and processing of text type data), communication subtask 2 (transmission and processing of voice type data), and communication subtask 3 (transmission and processing of image type data).
[0100] Step S2 corresponds to Figure 2 Process 6 in .
[0101] In this embodiment, when executing step S2, that is, the first network element divides the communication task to obtain multiple communication subtasks, the following steps may be specifically performed:
[0102] S201. Obtain the data type corresponding to each trainable model;
[0103] S202. The first network element divides the communication task according to each data type to obtain multiple communication subtasks.
[0104] In this embodiment, the data types corresponding to the trainable models run by different second network elements (NWDAF) are not exactly the same. Specifically, the data type corresponding to the trainable model may refer to the data type that the trainable model can process, or the data type output by the trainable model, etc. Taking the data type that the trainable model can process as an example, the trainable model 1 run by the second network element NWDAF1 can process text type data, so the data type corresponding to the trainable model 1 is the text type; the trainable model 2 run by the second network element NWDAF2 can process voice type data, so the data type corresponding to the trainable model 2 is the voice type; the trainable model 3 run by the second network element NWDAF3 can process image type data, so the data type corresponding to the trainable model 3 is the image type...
[0105] In step S202, the first network element (VFL server) divides the communication task according to each data type to obtain multiple communication subtasks. For example, the first network element (VFL server) extracts the part of the communication task belonging to the "text type" to obtain communication subtask 1; extracts the part of the communication task belonging to the "voice type" to obtain communication subtask 2; extracts the part of the communication task belonging to the "image type" to obtain communication subtask 3... thereby obtaining multiple communication subtasks.
[0106] By executing steps S201-S202, the communication task can be divided into multiple communication subtasks, each of which has a corresponding data type and can be processed by a trainable model of the same data type. For example, when executing step S3, the first network element (VFL server) sends communication subtask 1 (of "text type") to the second network element NWDAF1 (the trainable model 1 run by NWDAF1 can process text type data), sends communication subtask 2 (of "voice type") to the second network element NWDAF2 (the trainable model 2 run by NWDAF2 can process voice type data), and sends communication subtask 3 (of "image type") to the second network element NWDAF3 (the trainable model 3 run by NWDAF3 can process image type data)... So when executing step S4, each second network element runs its own trainable model and executes and processes the communication subtasks received by each, for example, the trainable model 1 run by NWDAF1 processes communication subtask 1.
[0107] In this embodiment, when executing step S202, that is, the first network element divides the communication task according to each data type to obtain multiple communication subtasks, the following steps may be specifically performed:
[0108] S20201. For any data type, determine the portion that matches the data type from the communication task;
[0109] S20202. Divide the determined part of the communication task to obtain a corresponding communication subtask;
[0110] S20203. Assign the communication subtask to the corresponding second network element of the corresponding data type;
[0111] S20204. Get the remaining part of the communication task;
[0112] S20205. Get the urgency of the remaining part;
[0113] S20206. When the urgency of the remaining part is lower than the threshold, the remaining part is subjected to data obfuscation processing to obtain a corresponding communication subtask;
[0114] S20207. Randomly assign the remaining processed communication subtasks to a second network element;
[0115] S20208. When the urgency of the remaining part is higher than the threshold, the remaining part is allocated to the first network element for processing.
[0116] The principles of steps S20201-S20208 are as follows Figure 3 shown.
[0117] Reference Figure 3 When executing step S20201, a clustering algorithm can be used to divide the content in the communication task into parts such as text type, voice type or image type through clustering. There may be some content in the communication task that cannot be clustered and divided into any known type such as text type, voice type or image type. In this embodiment, such a part is referred to as the remaining part. Specifically, the remaining part may not belong to the type such as text type, voice type or image type, but belongs to other types other than these types (such as control instruction type, etc.), or the remaining part may be content containing multiple types such as text type, voice type or image type, but cannot be classified into any specific type, and cannot be further segmented due to the need to maintain integrity.
[0118] Reference Figure 3When executing steps S20202 and S20203, the part of the communication task that belongs to the text type is divided into communication subtask 1, and in this embodiment, only the second network element NWDAF1 among all the second network elements (NWDAF1, NWDAF2, NWDAF3...) runs a trainable model of text type, so communication subtask 1 can be assigned to the second network element NWDAF1; the part of the communication task that belongs to the voice type is divided into communication subtask 2, and in this embodiment, only the second network element NWDAF2 among all the second network elements (NWDAF1, NWDAF2, NWDAF3...) runs a trainable model of voice type, so communication subtask 2 can be assigned to the second network element NWDAF2.
[0119] Reference Figure 3 , when executing steps S20202 and S20203, the image-type part of the communication task is divided into the communication subtask 3, and in this embodiment, among all the second network elements (NWDAF1, NWDAF2, NWDAF3, ...), the trainable models run by multiple second network elements (NWDAF3 and NWDAF4) are all of the image type, that is, both the second network elements NWDAF3 and NWDAF4 can be used to execute and process the communication subtask 3. In this embodiment, the priorities of the two second network elements NWDAF3 and NWDAF4 can be determined according to factors such as the urgency of the communication subtask 3 and the processing capabilities (including idleness and performance) of the two second network elements NWDAF3 and NWDAF4, and the communication subtask 3 is preferentially allocated to the second network element with the highest priority. In the case where the second network element with the highest priority fails, the communication subtask 3 is allocated to the second network element with a lower priority.
[0120] Reference Figure 3 When executing steps S20204 and S20205, the urgency of the remaining part is obtained. Specifically, the urgency can be quantitatively expressed by values such as "urgent" and "normal". For example, when the remaining part is a fault detection subtask, the urgency of the remaining part is generally a value such as "urgent" indicating a higher urgency. When the remaining part is a fault information uploading subtask, the urgency of the remaining part is generally a value such as "normal" indicating a lower urgency.
[0121] If the urgency of the remaining part is low, then steps S20206-S20207 may be performed. Specifically, the first network element (VFL server) may perform data obfuscation processing on the remaining part. In this embodiment, the data obfuscation processing performed may be at least one of the following processing:
[0122] Randomization: making data unreadable by randomly changing some of its properties;
[0123] Fuzzification: blurring the specific content of data by introducing uncertainty or noise into the data;
[0124] Substitution: obfuscate data by replacing certain elements in the data;
[0125] Perturbation: Slightly modifying the data, such as adding noise or random changes, so that the data remains statistically the same.
[0126] Noise injection: By introducing random noise into the data, some features of the data become unclear;
[0127] Transformation: Change the representation of data through mathematical transformations. For example, linear or nonlinear transformations can be used to confuse data.
[0128] Aggregation: combining multiple data items into a broader data category to reduce the level of detail in the data;
[0129] Segmentation: Split sensitive information into multiple parts and store them in different locations. These fragmented data can only be merged and restored into complete information when needed.
[0130] After the data obfuscation process in step S20206, the remaining part becomes a communication subtask. When executing step S20207, the communication subtask obtained by the process in step S20206 is randomly assigned to a second network element. For example, a second network element can be randomly selected from NWDAF1, NWDAF2, NWDAF3, etc., and the communication subtask can be assigned to the second network element. Alternatively, priority calculation can be performed based on the processing capabilities of each second network element, and the communication subtask can be assigned to the corresponding second network element based on the priority.
[0131] If the urgency of the remaining part is low, step S20208 may be performed. Specifically, when the urgency of the remaining part is higher than a threshold, the remaining part is allocated to the first network element (VFL server) for execution processing.
[0132] In this embodiment, the principle of executing steps S20201-S20208 is that by executing steps S20201-S20203, the content with clear types in the communication task can be divided into corresponding communication sub-tasks. When the communication sub-tasks correspond to multiple second network elements, the communication sub-tasks are preferentially allocated according to the priority, so that the communication sub-tasks with higher urgency can be allocated for execution processing first, and the communication sub-tasks can be preferentially allocated to the second network element with stronger processing capabilities for execution processing, thereby improving the execution efficiency of the communication task; by executing steps S20204-S20208, for the remaining part with lower urgency, by performing data obfuscation processing on it, the characteristics of different types of data contained in the remaining part can be reduced, so that the remaining part can be processed and obtained. When a communication subtask is assigned to a second network element for execution, other types of data features obtained by the assigned second network element from the remaining part are reduced (for example, if the remaining part contains both text-type content and image-type content, and the second network element NWDAF1 is generally used to process text-type communication subtasks, if the remaining part is directly assigned to the second network element NWDAF1 as a communication subtask, the second network element NWDAF1 may obtain image-type data features, thereby causing data privacy leakage, which is not conducive to the implementation of the vertical federated learning system. However, data obfuscation processing can reduce the significance of image-type data features in the remaining part, which is conducive to reducing data privacy leakage and maintaining the confidentiality performance of the vertical federated learning system).
[0133] After executing steps S20201-S20208, execute steps S3-S4 (corresponding to Figure 2 For example, by executing steps S3-S4, the second network element NWDAF2 runs the trainable model 2 and executes the assigned voice type communication subtask 2.
[0134] After executing steps S3-S4, execute step S5. Step S5 corresponds to Figure 2 Process 15 in .
[0135] In this embodiment, when executing step S5, that is, taking the first network element as a collaborator and each second network element as a participant, and performing vertical federated learning according to each communication subtask, the following steps may be specifically performed:
[0136] S501. The first network element distributes respective key information to each second network element;
[0137] S502. For any second network element, the second network element inputs the data from the communication subtask into the trainable model running by itself for processing, obtains intermediate information, encrypts the intermediate information using the assigned key information, and obtains encrypted intermediate information;
[0138] S503. Each second network element exchanges its own encrypted intermediate information;
[0139] S504. For any second network element, the second network element uses all the encrypted intermediate information obtained to train the trainable model, obtains gradient information, and uploads the gradient information to the first network element;
[0140] S505. The first network element decrypts the gradient information using the key information, obtains the decrypted gradient information, and sends the decrypted gradient information to the second network element;
[0141] S506. For any second network element, the second network element uses the received decrypted gradient information to update parameters of the trainable model running on itself.
[0142] Steps S501-S506 are steps of performing a vertical federated learning process in a vertical federated learning system composed of a first network element (VFL server) and each second network element such as NWDAF1, NWDAF2, NWDAF3, etc. The principles of steps S501-S506 are as follows: Figure 4 shown.
[0143] Reference Figure 4 In step S501, the first network element (VFL server) can generate multiple public keys such as public key 1, public key 2, public key 3, etc. based on the public key encryption algorithm, and assign public key 1 as key information to the second network element NWDAF1, assign public key 2 as key information to the second network element NWDAF2, and assign public key 3 as key information to the second network element NWDAF3... The first network element (VFL server) retains the corresponding private key, so that each second network element can use the assigned public key to encrypt the data to be sent to the first network element (VFL server), and the first network element (VFL server) uses the private key for decryption.
[0144] In step S502, taking the second network element NWDAF1 as an example, NWDAF1 inputs the data from the communication subtask 1 into the trainable model 1 that runs itself for processing (for example, processing text maps, automatic question answering, etc.), obtains the intermediate information embedding1, and uses the assigned key information (public key 1) to encrypt the intermediate information to obtain the encrypted intermediate information (still recorded as embedding1). Similarly, the second network element NWDAF2 can obtain the encrypted intermediate information embedding2, and the second network element NWDAF3 can obtain the encrypted intermediate information embedding3...
[0145] Reference Figure 4 In step S503, the second network elements exchange their own encrypted intermediate information, so that each second network element obtains the encrypted intermediate information processed by other second network elements in addition to the encrypted intermediate information processed by itself. For example, the second network element NWDAF1 obtains the encrypted intermediate information embedding1 processed by itself and the encrypted intermediate information embedding2, embedding3, etc. processed by other second network elements.
[0146] In step S504, taking the second network element NWDAF1 as an example, NWDAF1 uses all the encrypted intermediate information such as embedding1, embedding2, embedding3, etc. to train the trainable model 1 it runs, obtains gradient information 1, and uploads gradient information 1 to the first network element (VFL server). Similarly, the second network element NWDAF2 can obtain gradient information 2, the second network element NWDAF3 can obtain gradient information 3, and the first network element (VFL server) obtains all gradient information (gradients) such as gradient information 1, gradient information 2, gradient information 3, etc.
[0147] In step S505, the first network element (VFL server) uses the key information (private key) to decrypt the gradient information uploaded by each second network element to obtain decrypted gradient information. Figure 4 , the first network element (VFL server) sends the decrypted gradient information to each second network element. Each second network element executes step S506 respectively, and uses the received decrypted gradient information to update the parameters of the trainable model running on itself.
[0148] In this embodiment, the vertical federated learning implemented by steps S501-S506 can call each second network element with the trusted first network element (VFL server) in the core network as a collaborator without directly exchanging original data (such as the assigned communication subtasks) between multiple second network elements such as NWDAF1, NWDAF2, NWDAF3..., and the second network elements exchange encrypted intermediate information, and the encrypted intermediate information can only be decrypted and known by the first network element (VFL server) and the second network element that generates it; the second network elements exchange decrypted gradient information to complete the training of the trainable model running on themselves, and the decrypted gradient information does not directly reflect the characteristics of the original data itself, so the privacy and security of the original data can be guaranteed.
[0149] In this embodiment, by executing steps S1-S5, resource scheduling of each network element in the communication network can be performed based on vertical federated learning. Specifically, the collaborators in the vertical federated learning system slice and divide the communication tasks, so that each second network element can obtain the corresponding communication sub-task for processing separately, thereby realizing the use of multiple second network elements to process the communication tasks in parallel, thereby improving the execution efficiency of the communication tasks; moreover, the first network element divides the communication tasks according to the data characteristics of each participant in the vertical federated learning system, that is, each second network element, so that the communication sub-task obtained by each second network element can match the position of the second network element in the vertical federated learning system, thereby maintaining the advantages of the vertical federated learning system such as strong confidentiality; by performing vertical federated learning in the communication network, the data from the communication tasks can be used to train the trainable data run by each second network element, thereby improving the communication data processing performance of each second network element.
[0150] In this embodiment, by executing steps S1-S5, the following three load balancing strategies of the communication network can be implemented:
[0151] 1.VFL server intelligent segmentation task
[0152] ●Input data:
[0153] Task requirements (from AF): including computational complexity, data volume, task priority, etc.
[0154] Resource information (from IMANO): including resource usage, processing capacity,
[0155] Task support type, etc.
[0156] Historical task execution data: such as task completion time, resource consumption, etc.
[0157] ●Processing logic:
[0158] Task breakdown:
[0159] ■Break down complex tasks into multiple subtasks.
[0160] ■The size of the subtask is determined by the computational complexity and the resource status of each NWDAF client.
[0161] Subtask priority assignment:
[0162] ■ Urgent tasks are assigned to the clients with the strongest computing power or the lightest load.
[0163] ■ Regular tasks are assigned to clients with moderate resource utilization.
[0164] Avoid overload:
[0165] ■Avoid assigning tasks to clients whose load is close to the upper limit through real-time monitoring.
[0166] Privacy protection:
[0167] ■Using federated learning technology, the original data is not exchanged, and only the local model update results are shared.
[0168] ■Form a global model or final output by aggregating subtask results.
[0169] Output data:
[0170] Optimized task allocation list: including the NWDAF client, task type, data volume, etc. of the task allocation.
[0171] 2.AF dynamic task scheduling and priority adjustment
[0172] ●Input data:
[0173] Task progress (from NWDAF client): including task completion status, calculation status, etc.
[0174] Resource status (from IMANO): including CPU, memory, bandwidth usage, etc.
[0175] ●Processing logic:
[0176] Real-time priority adjustment:
[0177] ■ Raise tasks to high priority based on their urgency (such as fault detection tasks).
[0178] ■Non-urgent tasks can be delayed to free up resources.
[0179] Resource reallocation:
[0180] ■When the resource consumption of a certain NWDAF client is too high, AF will transfer part of the tasks to other clients.
[0181] ■ Combined with IMANO resource data, ensure balanced resource usage for each client.
[0182] Dynamic task revocation and reallocation:
[0183] ■During the task execution process, low-priority tasks are dynamically revoked based on real-time monitoring conditions, and resources are allocated to high-priority tasks.
[0184] Output data:
[0185] Adjusted task assignment list.
[0186] Dynamically adjusted resource usage strategy.
[0187] 3. Joint Optimization of VFL and AF
[0188] ●Input data:
[0189] Historical execution data: task duration, resource utilization efficiency, etc.
[0190] Real-time feedback data: current task progress, client resource status, etc.
[0191] ●Processing logic:
[0192] Historical data analysis:
[0193] ■Analyze the efficiency of historical task allocation strategies and optimize future task scheduling.
[0194] Task prediction:
[0195] ■ Based on historical data and task types, predict the resource requirements of the task.
[0196] Global model update:
[0197] ■Update the global model using the feedback from the NWDAF client.
[0198] ■The updated model guides the next round of task allocation and load balancing.
[0199] Output data:
[0200] Optimized task scheduling strategy.
[0201] Dynamically adjusted load balancing model.
[0202] In this embodiment, by executing steps S1-S5, the communication network can implement the following mechanism:
[0203] 1. Dynamic selection mechanism for multiple NWDAF clients: Allows the VFL server to flexibly select multiple NWDAF clients, breaking the limitation of single client selection in traditional systems and improving task processing efficiency. Combined with existing load balancing and resource allocation algorithms, dynamic selection of multiple clients can be achieved by expanding the AF selection strategy.
[0204] 2. Intelligent task allocation and load balancing mechanism: The intelligent task allocation and load balancing mechanism is introduced to ensure the dynamic balance of tasks among multiple NWDAF clients, and maximize the efficiency of VFL task processing. The task allocation function can be integrated into AF using the existing load balancing algorithm and task scheduling system.
[0205] 3. Multi-client parallel task execution coordination mechanism: This mechanism monitors the task execution progress of multiple clients in real time, and for the first time realizes the dynamic adjustment of task allocation, ensuring the efficient collaboration and execution of parallel tasks. Combined with the existing distributed computing coordination technology, it can realize the parallel task management and coordination of multiple clients.
[0206] 4. Dynamic adjustment mechanism of task priority: It provides the dynamic adjustment function of task priority, ensuring that key tasks can be executed first in a multi-client environment, improving the flexibility and response speed of task allocation. The dynamic priority adjustment function of tasks can be realized by using existing priority management and task scheduling technologies.
[0207] 5. Real-time feedback and optimization mechanism of task completion status: This mechanism provides real-time feedback and task optimization functions, ensuring rapid response during task execution and improving the accuracy and overall efficiency of task allocation. By combining real-time monitoring and feedback mechanisms, continuous monitoring and optimization of task status can be achieved.
[0208] In this embodiment, by executing steps S1-S5, a VFL-based multi-task parallel processing and network slice resource optimization scheduling method is implemented, combined with an intelligent task allocation and scheduling mechanism, aiming to optimize task execution and resource usage. By real-time analysis of client resource information and system load, VFL can dynamically adjust task allocation to ensure load balancing and avoid system overload; an intelligent priority adjustment mechanism is constructed to ensure priority execution of key tasks, and real-time monitoring of task progress and resource utilization through a dynamic feedback mechanism to further improve system efficiency; a fully automatic task and resource management system is implemented, and through intelligent analysis and dynamic optimization, the system's intelligence level and resource utilization efficiency are improved to ensure efficient operation in a complex task environment.
[0209] A computer program for executing the communication network control method based on vertical federated learning in this embodiment can be written and written into a computer device or storage medium. When the computer program is read out and executed, the communication network control method based on vertical federated learning in this embodiment is executed, thereby achieving the same technical effect as the communication network control method based on vertical federated learning in the embodiment.
[0210] It should be noted that, unless otherwise specified, when a feature is referred to as being "fixed" or "connected" to another feature, it may be directly fixed or connected to the other feature, or it may be indirectly fixed or connected to the other feature. In addition, the descriptions of up, down, left, right, etc. used in the present disclosure are only relative to the relative positional relationship of the components of the present disclosure in the accompanying drawings. The singular forms of "a", "" and "the" used in the present disclosure are also intended to include the plural forms, unless the context clearly indicates other meanings. In addition, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as those generally understood by those skilled in the art. The terms used in the specification of this embodiment are only for describing specific embodiments and are not intended to limit the present invention. The term "and / or" used in this embodiment includes any combination of one or more related listed items.
[0211] It should be understood that, although the term first, second, third etc. may be adopted to describe various elements in the present disclosure, these elements should not be limited to these terms. These terms are only used to distinguish the same type of elements from each other. For example, without departing from the scope of the present disclosure, the first element may also be referred to as the second element, and similarly, the second element may also be referred to as the first element. The use of any and all examples or exemplary language ("for example", "such as" etc.) provided by the present embodiment is only intended to better illustrate embodiments of the present invention, and unless otherwise required, the scope of the present invention will not be limited.
[0212] It should be appreciated that embodiments of the present invention may be implemented or enforced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The method may be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and drawings described in the specific embodiments. Each program may be implemented in a high-level procedural or object-oriented programming language to communicate with a computer system. However, if desired, the program may be implemented in assembly or machine language. In any case, the language may be a compiled or interpreted language. In addition, the program may be run on a programmed dedicated integrated circuit for this purpose.
[0213] In addition, the operations of the process described in this embodiment may be performed in any suitable order, unless otherwise indicated in this embodiment or otherwise clearly contradicted by the context. The process described in this embodiment (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as a code (e.g., executable instructions, one or more computer programs, or one or more applications) executed on one or more processors in common, by hardware or a combination thereof. A computer program includes a plurality of instructions that may be executed by one or more processors.
[0214] Further, the method can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a minicomputer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, a RAM, a ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the process described herein. In addition, the machine-readable code, or part thereof, can be transmitted via a wired or wireless network. When such media includes instructions or programs that implement the above steps in conjunction with a microprocessor or other data processor, the invention of this embodiment includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques of the present invention, the present invention also includes the computer itself.
[0215] The computer program can be applied to input data to perform the functions of the present embodiment, thereby converting the input data to generate output data stored in a non-volatile memory. The output information can also be applied to one or more output devices such as a display. In a preferred embodiment of the present invention, the converted data represents a physical and tangible object, including a specific visual depiction of the physical and tangible object produced on the display.
[0216] The above are only preferred embodiments of the present invention. The present invention is not limited to the above embodiments. As long as the technical effects of the present invention are achieved by the same means, any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention. Within the scope of protection of the present invention, its technical solutions and / or implementation methods may have various modifications and changes.
Claims
1. A communication network control method based on vertical federated learning, applied to a communication network, characterized in that: The communication network includes a first network element and a plurality of second network elements, and the communication network control method based on vertical federated learning includes: The first network element acquires a communication task; The first network element divides the communication task into multiple communication subtasks; The first network element sends each of the communication subtasks to the corresponding second network element respectively; Each of the second network elements respectively runs its own trainable model, and respectively performs and processes the communication subtask received by itself; The first network element is used as a collaborator, and each of the second network elements is used as a participant, and vertical federated learning is performed according to each of the communication subtasks.
2. The communication network control method based on vertical federated learning according to claim 1 is characterized in that: The communication network further includes a third network element and a fourth network element, and the first network element acquiring the communication task includes: The third network element obtains the communication task from an external application; The third network element queries the fourth network element for available second network elements according to the communication task; When the available second network element is found, the third network element sends the communication task to the first network element.
3. The communication network control method based on vertical federated learning according to claim 1 is characterized in that: The first network element divides the communication task into multiple communication subtasks, including: Obtaining data types corresponding to each of the trainable models; The first network element divides the communication task according to each of the data types to obtain a plurality of the communication subtasks.
4. The communication network control method based on vertical federated learning according to claim 3 is characterized in that: The first network element divides the communication task according to each of the data types to obtain a plurality of the communication subtasks, including: For any of the data types, determining a portion matching the data type from the communication tasks; Slice out the determined part of the communication task to obtain a corresponding communication subtask; The communication subtask is assigned to the second network element corresponding to the data type.
5. The communication network control method based on vertical federated learning according to claim 4 is characterized in that: The first network element divides the communication task according to each of the data types to obtain a plurality of the communication subtasks, further comprising: Acquire the remaining part of the communication task; the remaining part is the part that does not match any of the data types; obtaining the urgency of the remaining portion; When the urgency of the remaining part is lower than a threshold, performing data obfuscation processing on the remaining part to obtain a corresponding communication subtask; The communication subtask obtained by processing the remaining part is randomly assigned to one of the second network elements.
6. The communication network control method based on vertical federated learning according to claim 5 is characterized in that: The first network element divides the communication task according to each of the data types to obtain a plurality of the communication subtasks, further comprising: When the urgency of the remaining part is higher than a threshold, the remaining part is allocated to the first network element for processing.
7. The communication network control method based on vertical federated learning according to claim 4 is characterized in that: The allocating the communication subtask to the second network element corresponding to the data type includes: For any of the communication subtasks, when the data type corresponding to the communication subtask corresponds to a second network element, assigning the communication subtask to the second network element; When the data type corresponding to the communication subtask corresponds to a plurality of the second network elements, the priority of each of the second network elements is determined, and the communication subtask is allocated according to each priority.
8. The communication network control method based on vertical federated learning according to any one of claims 1 to 7, characterized in that: The first network element is used as a collaborator, each of the second network elements is used as a participant, and vertical federated learning is performed according to each of the communication subtasks, including: The first network element distributes respective key information to each of the second network elements; For any second network element, the second network element inputs the data from the communication subtask into the trainable model running by itself for processing, obtains intermediate information, and encrypts the intermediate information using the key information allocated to it, to obtain encrypted intermediate information; The second network elements exchange the respective encrypted intermediate information; For any second network element, the second network element uses all the obtained encrypted intermediate information to train the trainable model, obtains gradient information, and uploads the gradient information to the first network element; The first network element decrypts the gradient information using the key information to obtain decrypted gradient information, and sends the decrypted gradient information to the second network element; For any of the second network elements, the second network element uses the received decrypted gradient information to update parameters of the trainable model running on itself.
9. A computer device, characterized in that: It includes a memory and a processor, the memory is used to store at least one program, and the processor is used to load at least one program to execute the communication network control method based on vertical federated learning as described in any one of claims 1-8.
10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to execute the communication network control method based on vertical federated learning as described in any one of claims 1-8 when executed by the processor.