Network resource scheduling method, device, equipment and medium

Abstract

The embodiment of the invention discloses a network resource scheduling method and device, equipment and a medium, and relates to the technical field of network resource scheduling. The method comprises the following steps: training a network resource scheduling model of a current network node according to historical service data; sending model parameters of the trained network resource scheduling model to a central server, so that the central server performs integration processing on the model parameters of the plurality of network nodes to obtain a global optimization model; and if the global optimization model sent by the central server is received, performing local fine tuning on the global optimization model according to the service data of the current network node to obtain a fine tuning model. According to the technical scheme, the global optimization model is constructed through the model parameters, the effect of integrating the characteristics of the network nodes under the condition of protecting the original data is achieved, and the network nodes can consider global and meet local requirements during network resource scheduling through local fine tuning.

Description

Technical Field

[0001] This invention relates to the field of network resource scheduling technology, and in particular to a method, apparatus, device and medium for scheduling network resources. Background Technology

[0002] In scenarios where multiple network nodes handle hybrid network services, each network node deploys basic data acquisition devices or monitoring modules to periodically obtain metrics such as bandwidth usage, core network transmission latency, and service request characteristics. The processed information is then centrally reported to the management platform. The platform automatically switches between network channels for high-bandwidth or low-latency services using pre-defined load balancing rules to meet the comprehensive performance requirements of critical applications in different regions.

[0003] When handling mixed services across multiple domains and scenarios, the lack of in-depth local data mining often makes it difficult to capture the differentiated needs between nodes in a timely manner. This architecture may gradually face problems such as inflexible resource scheduling and long model update cycles when expanded to a wider range or involving higher bandwidth and lower latency 6G networks. Furthermore, centralized processing often requires the transmission of large amounts of detailed business information across nodes, posing a risk of leakage if sensitive data is involved. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for scheduling network resources, which can protect data privacy and achieve collaborative network resource scheduling among multiple nodes when handling mixed services across domains and multiple scenarios.

[0005] According to one aspect of the present invention, a method for scheduling network resources is provided, the method being executed by a network node, the method comprising:

[0006] Train a network resource scheduling model for the current network node based on historical business data;

[0007] The model parameters of the trained network resource scheduling model are sent to the central server so that the central server can integrate and process the model parameters of multiple network nodes to obtain a global optimization model; the historical service data of the multiple network nodes have at least two service types.

[0008] If a global optimization model is received from the central server, the global optimization model is fine-tuned locally based on the current network node's service data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests.

[0009] According to another aspect of the present invention, a method for scheduling network resources is provided, the method being executed by a central server, the method comprising:

[0010] If the model parameters of the trained network resource scheduling model are received from each network node, the model parameters are integrated to obtain a global optimization model; the service types of the service data of the multiple network nodes are at least two; the network resource scheduling model is trained by each network node based on its local historical service data.

[0011] The parameters of the global optimization model are sent to each network node so that each network node can fine-tune the global optimization model locally based on local business data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for business requests.

[0012] According to another aspect of the present invention, a network resource scheduling apparatus is provided, applied to a network node, the apparatus comprising:

[0013] The model training module is used to train the network resource scheduling model of the current network node based on historical business data.

[0014] The model parameter integration module is used to send the model parameters of the trained network resource scheduling model to the central server, so that the central server can integrate the model parameters of multiple network nodes to obtain a global optimization model; the business data of the multiple network nodes have at least two business types.

[0015] The model fine-tuning module is used to fine-tune the global optimization model locally based on the current network node's service data if it receives the global optimization model sent by the central server, thereby obtaining a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests.

[0016] According to another aspect of the present invention, a network resource scheduling apparatus is provided, applied to a central server, the apparatus comprising:

[0017] The global optimization model determination module is used to integrate the model parameters of the trained network resource scheduling model sent by each network node to obtain a global optimization model; the service types of the service data of the multiple network nodes are at least two; the network resource scheduling model is trained by each network node based on its local historical service data.

[0018] The model parameter sending module is used to send the parameters of the global optimization model to each network node, so that each network node can fine-tune the global optimization model locally according to local business data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for business requests.

[0019] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0020] At least one processor; and

[0021] A memory communicatively connected to the at least one processor; wherein,

[0022] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute a network resource scheduling method executed by the network node in any embodiment of the present invention, or to execute a network resource scheduling method executed by the central server in any embodiment of the present invention.

[0023] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute, implement, a network resource scheduling method executed by the network node in any embodiment of the present invention, or to implement a network resource scheduling method executed by the central server in any embodiment of the present invention.

[0024] In the technical solution of this application embodiment, the method is executed by a network node. The method includes: training a network resource scheduling model for the current network node based on historical service data; sending the model parameters of the trained network resource scheduling model to a central server, so that the central server integrates the model parameters of multiple network nodes to obtain a global optimization model; the service types of the historical service data of the multiple network nodes are at least two; if the global optimization model sent by the central server is received, the global optimization model is locally fine-tuned based on the service data of the current network node to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests. This technical solution achieves the effect of integrating the characteristics of each network node while protecting the original data through the global optimization model constructed by the model parameters, and enables each network node to consider both global and local needs when scheduling network resources through local fine-tuning.

[0025] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of a network resource scheduling method according to Embodiment 1 of this application;

[0028] Figure 2 This is a flowchart illustrating a vertical large-scale collaborative system for 6G government and enterprise hybrid services, provided according to Embodiment 1 of this application.

[0029] Figure 3 This is a schematic diagram of the continuous iteration process of updating 6G network nodes combined with vertical large model labels according to Embodiment 1 of this application;

[0030] Figure 4 This is a flowchart of a network resource scheduling method according to Embodiment 2 of this application;

[0031] Figure 5 This is a schematic diagram of a network resource scheduling device according to Embodiment 3 of this application;

[0032] Figure 6 This is a schematic diagram of a network resource scheduling device according to Embodiment 4 of this application;

[0033] Figure 7 This is a schematic diagram of the structure of an electronic device that implements a network resource scheduling method according to an embodiment of this application. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0035] It should be noted that the terms "first," "second," "target," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0036] Example 1

[0037] Figure 1 This application provides a flowchart of a network resource scheduling method according to Embodiment 1. This embodiment is applicable to network scheduling of mixed services. The method can be executed by a network resource scheduling device, which can be implemented in hardware and / or software and configured in a network node. Figure 1 As shown, the method includes:

[0038] S110, train the network resource scheduling model of the current network node based on historical business data.

[0039] Historical business data refers to business data previously processed by the current node. In this embodiment, the business data includes, but is not limited to, business request types, bandwidth usage, network slicing information, and end-to-end latency. This embodiment allows multiple network nodes to process their respective business requests, and these requests may cross domains, enabling collaborative network resource scheduling among multiple network nodes while protecting business data privacy.

[0040] For example, the network node is a 6G network node. The service data is mixed government and enterprise service data.

[0041] Optionally, in this embodiment of the application, the process of acquiring the historical service data includes: acquiring a time-series historical dataset; the historical dataset includes at least one of the following: service request type, bandwidth or spectrum usage, network slicing information, and end-to-end latency and jitter data; and performing feature extraction on the preprocessed historical dataset to obtain the historical service data.

[0042] For example, acquiring historical time-series datasets includes: extracting historical datasets of mixed government and enterprise services from multiple 6G network nodes, including service request types, bandwidth usage, network slicing information, and end-to-end latency. Each 6G network node independently acquires and saves the corresponding historical data to ensure data privacy and security.

[0043] Specifically, this could involve selecting multiple network nodes Nᵢ (i=1,2,…,n) within a 6G network, covering different administrative agencies, enterprise departments, or cross-domain collaborative scenarios, including government offices, public services, smart parks, and enterprise production lines. A business scenario description would be established for each network node Nᵢ, specifying its geographical location, network topology information, supported access methods (such as microwave, millimeter wave, and terahertz bands), and its association with mixed government and enterprise services (such as real-time video surveillance, ultra-high-definition conferencing, automated remote control, and industrial IoT data backhaul). Further data categories and field planning would then be implemented.

[0044] The business request types are further subdivided into government office data requests, enterprise internal shared resource requests, AR / VR collaborative application requests, ultra-low latency control command requests, etc., reflecting the diverse needs of government and enterprise businesses in the 6G environment from multiple dimensions.

[0045] Bandwidth or spectrum usage, recording the bandwidth occupancy, peak traffic, and mobile hotspot data of each node in different frequency bands and time slots, characterizing the ultra-high speed and ultra-high capacity characteristics of 6G;

[0046] Network slicing information includes traditional eMBB, URLLC, and other slices, as well as 6G-specific holographic interactive slices, industrial IoT slices, cross-domain private network slices, etc., to support the wide range of needs of government and enterprise mixed business.

[0047] End-to-end latency and jitter data, covering the transmission latency of multiple network segments from edge nodes to the core network and from government departments to enterprises, is the data on how to ensure the stable operation of critical government and enterprise businesses in ultra-low latency scenarios.

[0048] Furthermore, a unified data collection cycle and time period T are set to ensure that each node Nᵢ collects the aforementioned field data within the same time slice, facilitating subsequent cross-analysis of time-series changes and inter-node collaboration. The geographical location and network domain of each node are clearly defined, and the adjacency / cross-domain relationships between nodes (such as the interconnection between government / enterprise private networks and public networks) are recorded to provide a reference for subsequent slice collaboration and load balancing strategies. Furthermore, each node locally... The data is then integrated to form a time-series historical dataset. Through multiple iterations, historical data with time intervals from each 6G network node were gradually collected to form a complete historical dataset for government and enterprise services.

[0049] .

[0050] For example, feature extraction is performed on the preprocessed historical dataset to obtain the historical business data, including: local preprocessing of the historical dataset of mixed government and enterprise business obtained from each 6G network node, including data denoising, data cleaning, data standardization and feature extraction, to provide a high-quality data foundation for subsequent vertical large model collaborative optimization.

[0051] Specifically, this could be for 6G network nodes. The obtained historical dataset of government and enterprise services (historical dataset) Perform data denoising and set a noise level threshold. For data signals exceeding the noise threshold, an adaptive filter is used for filtering. By removing random noise from the historical dataset of government and enterprise services, we obtain the denoised historical dataset of government and enterprise services. ; for the denoised historical dataset of government and enterprise services Perform data cleaning to remove outliers and missing values, and establish criteria for judging outliers. The missing values ​​were then interpolated to obtain the cleaned historical dataset of government and enterprise services. : Cleaned historical dataset of government and enterprise services Perform data standardization processing and set standardization factors. and standard deviation Each feature value in the historical dataset of government and enterprise services is normalized to obtain a standardized historical dataset of government and enterprise services. ; for the standardized historical dataset of government and enterprise services Feature extraction is performed based on relevant features of network resource allocation, network slicing configuration, and latency data for 6G network enterprise services.

[0052] ;

[0053] in, This represents the historical dataset of government and enterprise services after feature extraction, i.e., historical business data. Indicates 6G network node The bandwidth resource allocation matrix for government and enterprise services reflects the bandwidth allocation situation as time t changes. The feature correlation matrix represents the relationships between different service features in the historical dataset of government and enterprise services. It captures the dependencies between network slice configuration and delayed data by fitting historical data. This is a network slice weight matrix, used to represent the impact of each network slice on the allocation of 6G network node resources, dynamically adjusted based on the actual needs of government and enterprise services. For feature extraction activation function, This is a feature weight matrix, representing the weights of different features in the historical data of government and enterprise services. Historical data that has undergone standardization processing. For feature regularization terms, For regularization parameters, and These are feature vectors and data feature matrices, respectively, describing the relationship between historical behavior patterns of government and enterprise services and network performance.

[0054] In this embodiment of the application, optionally, training a network resource scheduling model for the current network node based on historical service data includes: initializing a self-supervised learning model; performing comparative learning on positive and negative samples in the service data by combining the self-supervised learning model and a vertical large model, and generating optimized labels based on the self-supervised learning model trained by comparative learning; the optimized labels reflect network scheduling information; and after the iteration of the optimized labels is completed, the trained self-supervised learning model is obtained.

[0055] The vertical large-scale model and the self-supervised learning model can be models used for network resource scheduling, and the optimized labels can be network resource scheduling schemes. For example, the self-supervised learning model and the vertical large-scale model are combined to perform comparative learning on positive and negative samples in the business data. Since the historical business data can be time-series data, comparative learning training can be continuously performed, i.e., the optimized labels generated by the self-supervised learning model are iterated. After meeting the iteration requirements, the trained self-supervised learning model is obtained, and the parameters of the final self-supervised learning model can be sent to the central server in subsequent steps.

[0056] For example, on each 6G network node, a self-supervised learning model is trained using a large vertical model. By deeply mining the patterns and relationships within the historical dataset of mixed government and enterprise services, optimization labels for network resource allocation, network slicing configuration, and load balancing are automatically generated. This process is completed locally without disclosing the specific data content.

[0057] Specifically, this could be: at each 6G network node The above is based on the preprocessed historical dataset of government and enterprise services (i.e., historical business data). Initialize the self-supervised learning model by setting the initial parameters as follows: , 6G network node The parameter vector of the self-supervised learning model; in each 6G network node Above, a contrastive learning method is used on a historical dataset of government and enterprise services. In the middle, a set of positive samples is randomly sampled. and negative samples Positive samples Negative samples represent historical data of government and enterprise services in similar service scenarios. Represents historical data of government and enterprise services in different service scenarios; defines the contrastive loss function. The comparison loss function minimizes the loss of the same 6G network node. To maximize the distance between historical government and enterprise service data in similar service scenarios and to maximize the distance between historical government and enterprise service data in different service scenarios:

[0058] ;

[0059] in, Represents the model parameter matrix. For nonlinear mapping of the differences in features of sample data, These are preset marginal hyperparameters used to control the difference between positive and negative samples.

[0060] In each Network Nodes Above, collaborative vertical large model With local self-supervised learning models , on data Perform deep feature mining and local pseudo-label generation.

[0061] set up Indicates the local model's response to samples The expression, Indicates the vertical large model for samples The expression, This can be a similarity function (such as cosine similarity or dot product). Let be a hyperparameter (temperature coefficient). Let the set of positive samples be denoted as . The negative sample set is Then it can be done on the local node. Construct the following contrastive loss function The core idea is minimization. and The distance between them, while maintaining sufficient discriminability with negative samples, allows for the learning of deep semantic representations in collaboration with large vertical models and the automatic generation of pseudo-labels or auxiliary features:

[0062] ;

[0063] in This is the regularization coefficient used to prevent overfitting. The loss function is trained locally on the local node, without disclosing the original data, thus ensuring data security and privacy.

[0064] A self-supervised learning model trained based on contrastive learning is used through 6G network nodes. Optimize label generation by analyzing patterns and relationships within historical datasets of government and enterprise services. :

[0065] ;

[0066] in, This represents the historical data x after processing by a nonlinear feature extraction function. For activation function, Represents 6G network nodes The generated optimization tags are used to adjust network resource allocation, network slicing configuration, and load balancing parameters. Used to adjust 6G network nodes The network resource allocation, network slicing configuration, and load balancing parameters are configured; through multiple iterative training, each 6G network node... The self-supervised learning model continuously generates updated optimized labels at different time steps t. The updated optimization tags continuously adjust network resource allocation, network slicing configuration, and load balancing parameters in 6G network nodes to achieve dynamic and automated optimization of government and enterprise services.

[0067] ;

[0068] in, The optimized label generated for the k-th iteration. The learning rate represents the step size for parameter adjustment in each iteration;

[0069] By calculating the loss function Relative to the parameters of a self-supervised learning model The gradient is used to gradually update and optimize the labels; the self-supervised learning model of each 6G network node continuously learns new historical data of government and enterprise services, and adjusts the parameters in the contrastive loss function. Optimize service performance in real time and generate the final optimization strategy:

[0070] ;

[0071] in, The global learning rate, This is the regularization coefficient, used to avoid overfitting. This indicates the real-time load status of the 6G network.

[0072] S120, the model parameters of the trained network resource scheduling model are sent to the central server so that the central server can integrate and process the model parameters of multiple network nodes to obtain a global optimization model; the historical service data of the multiple network nodes have at least two service types.

[0073] The model parameters of the trained network resource scheduling model can be the parameters of a self-supervised learning model.

[0074] Specifically, the current network node sends the model parameters of the trained network resource scheduling model to the central server, and other network nodes do the same, sending the model parameters of their locally trained network resource scheduling models to the central server. After receiving the model parameters from multiple network nodes, the central server integrates and processes them to obtain a globally optimized model.

[0075] S130, if a global optimization model is received from the central server, the global optimization model is fine-tuned locally based on the current network node's service data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests.

[0076] Specifically, the global optimization model is obtained by integrating the model parameters uploaded by each network node. It can comprehensively reflect the network scheduling information between each network node, absorb the differentiated characteristics of each network node, and form a global optimization model that is suitable for the needs of multiple fields such as government and enterprise services.

[0077] After obtaining the global optimization model, the central server can send the global optimization model to each network node. Each network node can then fine-tune it based on its local business data, thereby obtaining a fine-tuned model that is both globally applicable and suited to local needs.

[0078] In this embodiment of the application, optionally, the global optimization model is locally fine-tuned based on the service data of the current network node to obtain a fine-tuned model, including: in each round of training, the global optimization model is locally fine-tuned based on the service data of the current network node to obtain updated global optimization model parameters; the updated global optimization model parameters are sent to a central server so that the central server integrates the updated global optimization model parameters of multiple network nodes to obtain a new global optimization model; if the parameters of the new global optimization model sent by the central server are received, the new global optimization model is locally fine-tuned again until the training termination condition is met to obtain the fine-tuned model.

[0079] Specifically, a fine-tuned model can be obtained through multiple iterations: In each training round, the global optimization model is locally fine-tuned based on the business data of the current network node (the business data of each network node is different, and the domain and business type of the business data of different network nodes can also be different), resulting in updated global optimization model parameters; then the updated global optimization model parameters are sent to the central server, which integrates the updated global optimization model parameters of multiple network nodes again to obtain a new global optimization model; if the parameters of the new global optimization model sent by the central server are received, the new global optimization model is fine-tuned locally again. Since the local business data is time-series, that is, continuously updated, the above iterative process can be continued until the training termination condition is met, and the fine-tuned model is obtained.

[0080] For example, the central server distributes the updated global optimization model to the corresponding 6G network nodes, enabling each node to perform local fine-tuning training on the distributed model based on the latest government and enterprise business data it has collected. Each node quantifies local network slice resources, peak bandwidth utilization, and end-to-end latency as feature inputs, and then generates new pseudo-labels based on the contrastive loss function mechanism of self-supervised learning, which are used to perform fine-tuning based on the local business operation status.

[0081] When node scenarios change, such as recently starting to support more intensive low-latency applications, the fine-tuning process adaptively adjusts relevant weights such as latency sensitivity and bandwidth priority to ensure that the scheduling strategy output by the model meets local real-time requirements. Since fine-tuning is completed independently within each node and does not require large-scale data transmission, it maintains the security of critical business information on each node while also providing the model with immediacy and flexibility, laying the foundation for further training and global model optimization in subsequent iterations.

[0082] In this embodiment, S130 can specifically be:

[0083] S71. In each round of training, each 6G network node Adjusted global optimization model parameters Using historical datasets Further local optimizations were performed, and the global optimization model parameters were updated to... :

[0084] ;

[0085] in, For learning rate, The loss function for the local node;

[0086] After S72 and local model optimization are completed, each 6G network node Update the global optimization model parameters The data is sent to the central server, which integrates the global optimization model parameters from all nodes to generate the global optimization model parameters. :

[0087] ;

[0088] in, is the weight coefficient of the i-th 6G network node, and the weight is dynamically adjusted according to the node's data volume or communication quality;

[0089] S73, Global Optimization Model Parameters Generated by the Central Server The data is distributed to each 6G network node. After receiving the global optimization model, the 6G network node combines it with the historical dataset of government and enterprise services locally. Further fine-tuning was performed, and the adjusted global optimization model parameters were obtained. Represented as:

[0090] ;

[0091] in, The global optimization model learning rate represents the step size adjusted in the local model after global optimization.

[0092] S74. After each training round, calculate the global loss function. Evaluate the global optimization model parameters for all nodes:

[0093] ;

[0094] S75. Through iterative steps S71 to S74, multiple 6G network nodes continuously optimize the local and global optimization models in different training rounds based on the dynamic changes of government and enterprise services, ultimately generating a stable global optimization model and obtaining a local fine-tuning model.

[0095] In the technical solution of this application embodiment, the method is executed by a network node. The method includes: training a network resource scheduling model for the current network node based on historical service data; sending the model parameters of the trained network resource scheduling model to a central server, so that the central server integrates the model parameters of multiple network nodes to obtain a global optimization model; the service types of the historical service data of the multiple network nodes are at least two; if the global optimization model sent by the central server is received, the global optimization model is locally fine-tuned based on the service data of the current network node to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests. This technical solution achieves the effect of integrating the characteristics of each network node while protecting the original data through the global optimization model constructed by the model parameters, and enables each network node to consider both global and local needs when scheduling network resources through local fine-tuning.

[0096] For example, Figure 2 This is a flowchart illustrating a vertical large-scale collaborative system suitable for 6G government and enterprise mixed business. Figure 3 A schematic diagram illustrating the continuous iteration process of updating vertical large model labels for 6G network nodes.

[0097] The technical solution of this application, when facing complex scenarios of distributed government and enterprise business, can, without directly sharing raw data, enable each node to generate more accurate optimized labels for its own business slices and resource allocation through a deeper self-supervised algorithm and multiple rounds of iterative updates. Furthermore, it integrates differentiated features in a global model, thereby providing a more flexible and comprehensive automated management and control method for mixed government and enterprise business. Solving this technical problem will help improve network resource allocation efficiency and achieve a higher level of integration and compatibility between cross-domain collaboration and security management.

[0098] In traditional centralized training or single-node optimization, multi-source data is typically collected at a central location, and then algorithm iterations are performed using a single model structure. This approach often struggles to balance privacy protection, diverse scenario requirements, and real-time load changes when dealing with mixed 6G government and enterprise services. Furthermore, when network nodes exhibit significant differences in characteristics, pure algorithm fine-tuning cannot adequately adapt to their unique business models. This application's solution departs from the approach of improving only algorithm parameters, instead collaborating with self-supervised learning, large vertical models, and federated learning frameworks to build a multi-stage, distributed, end-to-end optimization pathway. In this process, each node first performs noise filtering, feature extraction, and contrastive learning training on its local data, while simultaneously collaborating with the large vertical model for deep semantic mapping, thereby obtaining effective feature representations and optimized labels without leaking the original data.

[0099] Compared to existing technologies that generally rely on a single center for centralized modeling of multi-source data, this proposal deploys a collaborative mechanism of self-supervised learning and large-scale vertical models at the multi-node level. This allows each node to independently complete data preprocessing, feature extraction, and comparative learning training. This distributed structure avoids large-scale cross-domain transmission of raw data from mixed government and enterprise businesses, reducing pressure on privacy and network bandwidth, and allowing the diverse business scenarios and operating environments of each node to be fully considered locally. Compared to existing solutions that only iterate algorithms at a central location, the node-side deep feature mining and pseudo-label generation process provided in this proposal can obtain more targeted optimized labels in the initial stage. Then, using a federated learning framework, the trained model parameters are comprehensively integrated at the central server, thereby incorporating the differentiated features of each node to form a global model suitable for the multi-domain needs of government and enterprise services.

[0100] Example 2

[0101] Figure 4This is a flowchart of a network resource scheduling method provided in Embodiment 2 of this application. This embodiment of the application can be applied to the situation of network scheduling for mixed services. The method can be executed by a network resource scheduling device, which can be implemented in hardware and / or software and can be configured in a central server.

[0102] like Figure 4 As shown, the method in this embodiment of the application specifically includes the following steps:

[0103] S210, if the model parameters of the trained network resource scheduling model sent by each network node are received, the model parameters are integrated to obtain a global optimization model.

[0104] The service types of the service data of the multiple network nodes are at least two; the network resource scheduling model is trained by each network node based on its local historical service data.

[0105] Specifically, each network node trains a network resource scheduling model based on its local historical service data and sends the trained model parameters to the central server. After receiving the model parameters from each network node, the central server integrates the model parameters to obtain a global optimization model.

[0106] Optionally, in this embodiment of the application, the model parameters are integrated to obtain a global optimization model, including: performing a weighted average of the model parameters of each network node to obtain the parameters of the global optimization model.

[0107] For example, in one feasible approach, weights can be set for the model parameters of each network node, and then the weighted average of the model parameters can be calculated to obtain the parameters of the globally optimized model.

[0108] For example, a federated learning framework is established between a central server and multiple 6G network nodes, enabling each node to collaboratively optimize a self-supervised learning model under a vertical large-scale model without directly sharing historical data from mixed government and enterprise services. After each 6G network node completes the training of its self-supervised learning model locally, it sends its model parameters to the central server. The central server receives the self-supervised learning model parameters from each 6G network node, integrates and optimizes the model parameters through aggregation, and obtains a global model that can be collaboratively optimized for all 6G network nodes. This global model, while absorbing the differentiated characteristics of each node, improves its adaptability to the needs of mixed government and enterprise services.

[0109] Specifically, this could involve constructing a federated learning framework and configuring multiple 6G network nodes within the 6G network. Participation in federated learning, the federated learning framework allows individual 6G network nodes to participate in federated learning. Collaboratively optimize the self-supervised learning model without sharing historical data on government and enterprise services; at each 6G network node The self-supervised learning model is trained locally, and after training, each 6G network node... The parameters of the self-supervised learning model obtained from local training The data is sent to a central server, which integrates the self-supervised learning model parameters uploaded by each 6G network node using a weighted averaging algorithm. Through matrix operations, it combines the resource configuration and network slicing configuration of each 6G network node, dynamically considering the impact of latency data, to generate globally optimized model parameters. :

[0110] ;

[0111] in, is the weighting coefficient of the i-th 6G network node at time step t. The weighting coefficient is dynamically adjusted according to the data volume or communication quality of each 6G network node. and 6G network nodes Resource allocation and slice configuration matrix, Indicates 6G network node Delayed data weights, To control the adjustment coefficient of weights under the influence of delayed data.

[0112] The central server generates global optimization model parameters. Then, it is distributed to each 6G network node. After receiving the global optimization model parameters, each 6G network node uses the local government and enterprise service historical dataset. Fine-tuning of the global optimization model; the adjusted global optimization model parameters. Represented as:

[0113] ;

[0114] in, The learning rate; The weight matrix for network resource allocation represents the resource allocation of 6G network nodes at different historical stages; Activation matrix of historical data for government and enterprise services; extract key service patterns of 6G network nodes at specific times. For regularization parameters, and These are feature matrices related to latency and service quality, respectively.

[0115] The central server receives self-supervised learning model parameters from each 6G network node and integrates the feature mapping vectors, contrastive learning weights, and pseudo-label derived matrices uploaded by multiple nodes. The integration process filters and weights the parameters based on the weight factors of each node, ensuring that the parameter distributions of the respective supervised models complement each other, thus obtaining a global model that balances differentiated features with common requirements.

[0116] During the model parameter aggregation phase, the central server considers the data scale, real-time load, bandwidth usage, and latency information of each node, and makes appropriate corrections or discards for overfitted or noisy local parameters, ultimately ensuring that the global model can cover the core patterns of multiple government and enterprise scenarios without causing overall deviation due to local anomalies.

[0117] S220, the parameters of the global optimization model are sent to each network node so that each network node can fine-tune the global optimization model locally based on local service data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests.

[0118] For example, the central server distributes the updated global model to the corresponding 6G network nodes, enabling each node to perform local fine-tuning training on the distributed model based on the latest government and enterprise business data it has collected. Each node quantifies local network slice resources, peak bandwidth utilization, and end-to-end latency as feature inputs, and then generates new pseudo-labels based on the contrastive loss function mechanism of self-supervised learning, which are used to fine-tune the local business operation status.

[0119] When node scenarios change, such as recently starting to support more intensive low-latency applications, the fine-tuning process adaptively adjusts relevant weights such as latency sensitivity and bandwidth priority to ensure that the scheduling strategy output by the model meets local real-time requirements. Since fine-tuning is completed independently within each node and does not require large-scale data transmission, it maintains the security of critical business information on each node while also providing the model with immediacy and flexibility, laying the foundation for further training and global model updates in subsequent iterations.

[0120] This technical solution, on the central server side, constructs a federated learning stage that not only simply aggregates the model parameters uploaded by nodes, but also dynamically weights them based on the slicing requirements, bandwidth usage, and latency characteristics of different nodes' business environments. Traditional algorithms typically focus only on minimizing the model's loss on the global dataset, while this solution incorporates the unique network configuration and temporal information of each node into the overall optimization considerations. This ensures that the updated global model can accurately adapt to the local scenario after being distributed to each node. This collaborative approach, beyond the algorithm itself, integrates systemic factors such as network structure and privacy protection requirements. Ultimately, through multiple rounds of interactive iteration, it continuously calibrates the differences between the local and global models in each training round, and continuously integrates new features and labels generated at the node end into the next self-supervised learning structure, forming a distributed, multi-layered collaborative optimization system that differs from existing algorithms.

[0121] In contrast to existing solutions that often struggle to detail inter-node differences and overlook local bandwidth and slice configuration requirements, this proposal introduces a dynamically weighted parameter aggregation mechanism. This allows the system to comprehensively consider node-level network configuration, service request patterns, and end-to-end latency during multiple iterations. This approach goes beyond simply improving algorithmic accuracy; it also embeds actual network resource allocation and load balancing needs into the federated learning update logic, enhancing the model's adaptability to different administrative departments, enterprise parks, and cross-domain collaborative scenarios. Through this structured design combining self-supervised learning and distributed collaboration, this proposal significantly strengthens data security, node diversity support, and localized response capabilities to real-time service demands while retaining existing technically feasible approaches. This highlights the overall application value of this solution in 6G government and enterprise hybrid services.

[0122] Example 3

[0123] Figure 5 This is a schematic diagram of a network resource scheduling device provided in Embodiment 3 of this application. This device can execute the network resource scheduling method performed by the network node in any embodiment of the present invention, and possesses the corresponding functional modules and beneficial effects of the method. For example... Figure 5 As shown, the device includes:

[0124] Model training module 310 is used to train the network resource scheduling model of the current network node based on historical business data;

[0125] The model parameter integration module 320 is used to send the model parameters of the trained network resource scheduling model to the central server, so that the central server can integrate the model parameters of multiple network nodes to obtain a global optimization model; the business types of the business data of the multiple network nodes are at least two.

[0126] The model fine-tuning module 330 is used to fine-tune the global optimization model locally based on the current network node's service data if it receives the global optimization model sent by the central server, so as to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests.

[0127] The technical solution of this application embodiment includes: a model training module 310, used to train a network resource scheduling model for the current network node based on historical service data; a model parameter integration module 320, used to send the model parameters of the trained network resource scheduling model to a central server, so that the central server can integrate the model parameters of multiple network nodes to obtain a global optimization model; the service types of the service data of the multiple network nodes are at least two; a model fine-tuning module 330, used to, if receiving the global optimization model sent by the central server, locally fine-tune the global optimization model based on the service data of the current network node to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests. This technical solution, through the global optimization model constructed by model parameters, achieves the effect of integrating the characteristics of each network node while protecting the original data, and through local fine-tuning, enables each network node to consider both global and local needs when scheduling network resources.

[0128] Optionally, in this embodiment of the application, the device further includes: a historical business data acquisition module, specifically used for:

[0129] Obtain a historical dataset of time series data; the historical dataset includes at least one of the following: service request type, bandwidth or spectrum usage, network slicing information, and end-to-end latency and jitter data;

[0130] Feature extraction is performed on the preprocessed historical dataset to obtain the historical business data.

[0131] Optionally, in this embodiment of the application, the model training module 310 includes:

[0132] The model initialization unit is used to initialize the self-supervised learning model;

[0133] The contrastive learning unit is used to perform contrastive learning on positive and negative samples in the business data by combining a self-supervised learning model and a vertical large model, and to generate optimized labels based on the self-supervised learning model trained by the contrastive learning; the optimized labels reflect network scheduling information.

[0134] An optimized label iteration unit is used to obtain the trained self-supervised learning model after the optimized label iteration is completed.

[0135] Optionally, in this embodiment of the application, the model fine-tuning module 330 includes:

[0136] The model parameter update unit is used to fine-tune the global optimization model locally based on the business data of the current network node during each round of training to obtain the updated global optimization model parameters.

[0137] The parameter sending unit is used to send the updated global optimization model parameters to the central server, so that the central server can integrate the updated global optimization model parameters of multiple network nodes to obtain a new global optimization model.

[0138] The model parameter update unit is used to fine-tune the new global optimization model locally again if it receives new global optimization model parameters sent by the central server, until the training termination condition is met, and the fine-tuned model is obtained.

[0139] The network resource scheduling device provided in this application embodiment can execute a network resource scheduling method performed by the network node provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0140] Example 4

[0141] Figure 6 This is a schematic diagram of a network resource scheduling device provided in Embodiment 4 of this application. This device can execute the network resource scheduling method executed by the central server in any embodiment of this invention, and possesses the corresponding functional modules and beneficial effects of the execution method. For example... Figure 6 As shown, the device includes:

[0142] The global optimization model determination module 410 is used to integrate the model parameters of the trained network resource scheduling model sent by each network node to obtain a global optimization model; the service types of the service data of the multiple network nodes are at least two; the network resource scheduling model is trained by each network node based on its local historical service data.

[0143] The model parameter sending module 420 is used to send the parameters of the global optimization model to each network node, so that each network node can fine-tune the global optimization model locally according to local business data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for business requests.

[0144] Optionally, in this embodiment of the application, the global optimization model determination module 410 includes:

[0145] The global optimization model determination unit is used to perform weighted averaging of the model parameters of each network node to obtain the parameters of the global optimization model.

[0146] The network resource scheduling device provided in this application embodiment can execute a network resource scheduling method executed by the central server provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0147] Example 4

[0148] Figure 7 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0149] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0150] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0151] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as network resource scheduling methods.

[0152] In some embodiments, the network resource scheduling method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the network resource scheduling method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the network resource scheduling method by any other suitable means (e.g., by means of firmware).

[0153] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0154] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0155] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0156] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0157] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0158] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0159] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0160] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for scheduling network resources, characterized in that, The method is executed by a network node, and the method includes: Train a network resource scheduling model for the current network node based on historical business data; The model parameters of the trained network resource scheduling model are sent to the central server so that the central server can integrate and process the model parameters of multiple network nodes to obtain a global optimization model; the historical service data of the multiple network nodes have at least two service types. If a global optimization model is received from the central server, the global optimization model is fine-tuned locally based on the current network node's service data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests.

2. The method according to claim 1, characterized in that, The process of acquiring the historical business data includes: Obtain a historical dataset of time series data; the historical dataset includes at least one of the following: service request type, bandwidth or spectrum usage, network slicing information, and end-to-end latency and jitter data; Feature extraction is performed on the preprocessed historical dataset to obtain the historical business data.

3. The method according to claim 1, characterized in that, Training a network resource scheduling model for the current network nodes based on historical business data includes: Initialize the self-supervised learning model; The positive and negative samples in the business data are compared and learned by combining a self-supervised learning model and a vertical large model, and optimized labels are generated based on the self-supervised learning model trained by the comparative learning; the optimized labels reflect network scheduling information. After the optimization label iteration is completed, the trained self-supervised learning model is obtained.

4. The method according to claim 1, characterized in that, The global optimization model is fine-tuned locally based on the current network node's service data to obtain a fine-tuned model, including: In each round of training, the global optimization model is fine-tuned locally based on the business data of the current network node to obtain the updated global optimization model parameters; The updated global optimization model parameters are sent to the central server so that the central server can integrate the updated global optimization model parameters of multiple network nodes to obtain a new global optimization model. If the parameters of the new global optimization model are received from the central server, the new global optimization model is fine-tuned locally again until the training termination condition is met, and the fine-tuned model is obtained.

5. A method for scheduling network resources, characterized in that, The method is executed by a central server, and the method includes: If the model parameters of the trained network resource scheduling model are received from each network node, the model parameters are integrated to obtain a global optimization model; the service types of the service data of the multiple network nodes are at least two; the network resource scheduling model is trained by each network node based on its local historical service data. The parameters of the global optimization model are sent to each network node so that each network node can fine-tune the global optimization model locally based on local business data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for business requests.

6. The method according to claim 5, characterized in that, The model parameters are integrated to obtain a global optimization model, including: The parameters of the model for each network node are weighted and averaged to obtain the parameters of the global optimization model.

7. A network resource scheduling device, characterized in that, Applied to network nodes, the device includes: The model training module is used to train the network resource scheduling model of the current network node based on historical business data. The model parameter integration module is used to send the model parameters of the trained network resource scheduling model to the central server, so that the central server can integrate the model parameters of multiple network nodes to obtain a global optimization model; the business data of the multiple network nodes have at least two business types. The model fine-tuning module is used to fine-tune the global optimization model locally based on the current network node's service data if it receives the global optimization model sent by the central server, thereby obtaining a fine-tuned model; the fine-tuned model is used to schedule network resources for service requests.

8. A network resource scheduling device, characterized in that, The device, applied to a central server, includes: The global optimization model determination module is used to integrate the model parameters of the trained network resource scheduling model sent by each network node to obtain a global optimization model; the service types of the service data of the multiple network nodes are at least two; the network resource scheduling model is trained by each network node based on its local historical service data. The model parameter sending module is used to send the parameters of the global optimization model to each network node, so that each network node can fine-tune the global optimization model locally according to local business data to obtain a fine-tuned model; the fine-tuned model is used to schedule network resources for business requests.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the network resource scheduling method executed by the network node according to any one of claims 1-4, or to perform the network resource scheduling method executed by the central server according to any one of claims 5-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the network resource scheduling method executed by the network node according to any one of claims 1-4, or to execute the network resource scheduling method executed by the central server according to any one of claims 5-6.

Images