Network slice resource optimization method, device, system and product

Abstract

The invention relates to the technical field of communication, and discloses a network slice resource optimization method, device, system and product. The method comprises the following steps: collecting flow data of each edge node in a target network in a current time step according to the size of the time step; the traffic data comprises short-time traffic data and long-time traffic data of all network slices in the edge node; predicting predicted flow data of each edge node in the next time step based on the flow data of each current edge node; determining a target distribution strategy based on the predicted traffic data of each edge node in the next time step; determining target resource configuration in each edge node in the next time step according to a target allocation strategy and the predicted traffic data of each edge node in the next time step; and allocating network resources to each edge node according to the target resource configuration. By adopting the method, the accurate and efficient dynamic network resource allocation of the whole target network can be realized, and the overall performance of the network is improved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a network slicing resource optimization method, device, system and product. Background Art

[0002] In recent years, some traffic prediction and resource optimization solutions based on machine learning have begun to emerge, trying to overcome the limitations of traditional methods. These solutions analyze historical traffic data and try to predict future traffic patterns, thereby guiding the dynamic adjustment of resources. Although these methods have improved resource utilization and network performance to a certain extent, there are still several challenges. For example, traffic prediction on a single time scale may not accurately capture the complex changes in traffic, and centralized resource management mechanisms may bring high computational burdens and latency issues.

[0003] Network slicing is one of the key technologies of 5G communication. This technology allows the creation of multiple virtual networks on the same physical network infrastructure to meet the needs of different business scenarios. In a dynamically changing traffic environment, how to efficiently manage and optimize the resources of network slices has become an urgent problem to be solved. The current dynamic optimization solution usually simply predicts the next traffic change based on the traffic data of the current time period. However, the traffic change patterns of different business scenarios are different. Only by predicting through a unified time period, it is impossible to accurately capture the traffic change patterns of all business scenarios, resulting in inaccurate actual predicted traffic, low resource utilization, and difficulty in ensuring the efficient operation of the business system. Summary of the invention

[0004] In view of this, the present application aims to propose a network slicing resource optimization method, device, system and product to achieve accurate dynamic network resource allocation and improve the overall performance of the network.

[0005] To achieve the above purpose, the technical solution of this application is as follows:

[0006] A first aspect of an embodiment of the present application provides a network slice resource optimization method, the method comprising:

[0007] According to the size of the time step, the flow data of each edge node in the target network in the current time step is collected; the flow data includes: short-term flow data and long-term flow data of all network slices in the edge node;

[0008] Based on the current traffic data of each edge node, predict the predicted traffic data of each edge node in the next time step;

[0009] Determine a target allocation strategy based on the predicted traffic data of each edge node in the next time step; determine a target resource configuration in each edge node in the next time step according to the target allocation strategy and the predicted traffic data of each edge node in the next time step;

[0010] Allocate network resources to each edge node according to the target resource configuration.

[0011] Optionally, based on the current traffic data of each edge node, predicting the predicted traffic data of each edge node in the next time step includes:

[0012] Constructing a first prediction model based on reinforcement learning, and using a Q-learning algorithm to train the first prediction model;

[0013] The trained first prediction model is used to process the traffic data of each edge node to obtain the predicted traffic data of each edge node in the next time step.

[0014] Optionally, based on the current traffic data of each edge node, predicting the predicted traffic data of each edge node in the next time step includes:

[0015] Constructing and training a second prediction model based on a neural network, wherein the second prediction model includes sub-neural networks deployed on each edge node; in the trained second prediction model, the parameters of each sub-neural network are the same;

[0016] The local traffic data is processed using each sub-neural network of the second prediction model to obtain the local predicted traffic data in the next time step.

[0017] Optionally, training a second prediction model based on a neural network comprises:

[0018] Iteratively train the local sub-neural network using the traffic data of each edge node. In each round of training, based on the local actual traffic data and the predicted traffic data of the sub-neural network, calculate the loss value and update the parameters of the sub-neural network.

[0019] Calculate the comprehensive parameters based on the parameters of all sub-neural networks;

[0020] Based on the comprehensive parameters, the parameters of all sub-neural networks are updated and the next round of training is carried out.

[0021] Optionally, predict the predicted traffic data of each edge node in the next time step, including:

[0022] Obtaining a first prediction value obtained by the edge node based on long-term traffic data and a second prediction value obtained based on short-term traffic data;

[0023] A long-term weight and a short-term weight are set based on the device type of the edge node; the sum of the long-term weight and the short-term weight is 1;

[0024] The predicted traffic data of the edge node is calculated according to the first predicted value, the second predicted value, the long-term weight and the short-term weight of each edge node.

[0025] Optionally, determining the target resource configuration in each edge node in the next time step according to the target allocation strategy and the predicted traffic data of each edge node in the next time step includes:

[0026] Obtaining an initial allocation strategy for the target network;

[0027] According to the initial allocation strategy, the current network resource configuration of each edge node and the predicted traffic data, the network resource configuration in the next time step is calculated as the target resource configuration.

[0028] Optionally, after calculating the initial resource configuration in the next time step, it also includes:

[0029] Obtain the current resource allocation of each network slice in each edge node, and calculate the overall utility and fairness of the target network based on the current resource allocation;

[0030] Based on the predicted traffic data of each edge node, the initial allocation strategy is adjusted using a gradient descent method to maximize the overall utility and optimize the fairness, thereby obtaining a first allocation strategy;

[0031] Based on the first allocation strategy, the current network resource configuration of each edge node and the predicted traffic data, the network resource configuration in each edge node in the next time step is calculated, and the target resource configuration is updated.

[0032] Optionally, after collecting the flow data of each edge node in the target network in the current time step, it also includes:

[0033] The flow data is preprocessed, specifically including: removing invalid data and erroneous data in the flow data; detecting abnormal data in the flow data, and replacing the abnormal data with valid data in the previous time step; and compressing the flow data using a principal component analysis method to obtain compressed data.

[0034] Optionally, the network slice resource optimization method further includes:

[0035] Obtain the current network resource configuration of multiple target networks and calculate the comprehensive utility and comprehensive fairness;

[0036] According to the optimization preference, the utility weight and the fairness weight are set; the sum of the utility weight and the fairness weight is 1; the optimization preference is utility priority or fairness priority;

[0037] Calculating comprehensive performance based on the utility weight and the fairness weight;

[0038] Obtaining a comprehensive allocation strategy of the multiple target networks, and adjusting the comprehensive allocation strategy using a gradient descent method to maximize the comprehensive performance, thereby obtaining a second allocation strategy;

[0039] Calculate the network resource configuration of each target network in the next time step as the comprehensive resource configuration according to the second allocation strategy, the current network resource configuration of each target network and the predicted traffic data;

[0040] Based on the comprehensive resource configuration, network resources are allocated to each target network.

[0041] According to a second aspect of an embodiment of the present application, a network slice resource optimization device is provided, which is used to implement the steps in the method provided in the first aspect of the embodiment of the present application, and the device includes:

[0042] The perception module is configured to collect traffic data of each edge node in the target network in the current time step according to the size of the time step; the traffic data includes: short-term traffic data and long-term traffic data of all network slices in the edge node;

[0043] The prediction module is configured to predict the predicted traffic data of each edge node in the next time step based on the current traffic data of each edge node;

[0044] The optimization module is configured to determine a target allocation strategy based on the predicted traffic data of each edge node in the next time step; determine a target resource configuration in each edge node in the next time step according to the target allocation strategy and the predicted traffic data of each edge node in the next time step;

[0045] The allocation module is configured to allocate network resources to each edge node according to the target resource configuration.

[0046] According to a third aspect of an embodiment of the present application, a network slice resource optimization system is provided, including:

[0047] Multiple target networks, the multiple target networks include: an access network, a transmission network and a core network; wherein the access network is deployed with:

[0048] A plurality of edge nodes in the method provided in the first aspect of the embodiment of the present application; wherein each edge node is deployed with a network slice resource optimization device provided in the second aspect of the embodiment of the present application;

[0049] The data processing center is used to aggregate the traffic data or sub-neural network parameters sent by each edge node.

[0050] According to a fourth aspect of an embodiment of the present application, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps in the method provided in the first aspect of the embodiment of the present application.

[0051] According to a fifth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the method provided in the first aspect of the embodiment of the present application are implemented.

[0052] According to the sixth aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps in the method provided in the first aspect of an embodiment of the present application are implemented.

[0053] The network slice resource optimization method provided by the present application is adopted to collect the current traffic data of each edge node based on different time scales, and predict the predicted traffic data of the edge node in the next time step based on the current multi-scale traffic data, so as to determine the target allocation strategy of each edge node in the next time step. Determine the specific network resource configuration in the next time step according to the target allocation strategy, and make dynamic adjustments based on the current network resource configuration.

[0054] In view of the large differences in traffic change patterns in business scenarios of different network slices, traffic data based on a single time scale cannot accurately capture the traffic change patterns of all business scenarios. Therefore, this application introduces multi-time scale traffic data when collecting traffic data of edge nodes. By collecting short-term traffic data to capture the instantaneous changes and short-term trends of node traffic, and by collecting long-term traffic data to capture the long-term trends and periodic changes of node traffic, all business scenarios are covered to more comprehensively obtain the laws of node traffic changes, and accurate prediction of node traffic in the next time step is achieved. Furthermore, accurate dynamic network resource allocation is achieved based on the prediction results, which improves the overall resource utilization and overall performance of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0056] Figure 1 is a flow chart of a network slice resource optimization method proposed in an embodiment of the present application;

[0057] Figure 2 is a flow chart of adjusting the target allocation strategy in one embodiment of the present application;

[0058] Figure 3 is a flow chart of adjusting a cross-layer target allocation strategy in an embodiment of the present application;

[0059] Figure 4 is a schematic diagram of a network slice resource optimization device proposed in an embodiment of the present application;

[0060] Figure 5 is a schematic diagram of a network slicing resource optimization system proposed in an embodiment of the present application;

[0061] Figure 6 It is a schematic diagram of an electronic device proposed in an embodiment of the present application. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0063] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present application. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0064] In the various embodiments of the present application, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0065] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects as detailed in the present application.

[0066] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other.

[0067] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0068] Figure 1 : is a flow chart of a network slice resource optimization method proposed in an embodiment of the present application. Figure 1 As shown, the method includes:

[0069] S1: Collect the traffic data of each edge node in the target network in the current time step according to the size of the time step; the traffic data includes: short-term traffic data and long-term traffic data of all network slices in the edge node;

[0070] S2: Based on the current traffic data of each edge node, predict the predicted traffic data of each edge node in the next time step;

[0071] S3: Determine a target allocation strategy based on the predicted traffic data of each edge node in the next time step; determine the target resource configuration in each edge node in the next time step according to the target allocation strategy and the predicted traffic data of each edge node in the next time step;

[0072] S4: Allocate network resources to each edge node according to the target resource configuration.

[0073] In this embodiment, the target network is a 5G network. A perception module is deployed at each edge node in the target network to monitor and record the traffic-related information of each node in real time, including inbound traffic, outbound traffic, traffic rate, traffic direction, etc. Through the perception module, the multi-time scale traffic data of the local edge node is collected according to the size of the pre-set time step. Among them, the traffic data includes short-term traffic data and long-term traffic data. Based on the traffic data collected by each edge node, a prediction is made to obtain the predicted traffic data of the edge node in the next time step, and then, based on the predicted traffic data of each edge node, the target allocation strategy for the network resources of the edge node is determined. The target resource configuration of each edge node in the next time step is calculated according to the target allocation strategy. According to the target resource configuration, the current network resource configuration (for example: bandwidth allocation) is dynamically adjusted.

[0074] In this embodiment, by deploying a perception module at the edge node, the timeliness of the collected traffic data is ensured. By collecting traffic data of multiple time scales of edge nodes in real time, the traffic change rules of all business types on the edge nodes can be fully captured. Specifically, short-term traffic data is collected to capture the recent traffic change trend of the node, and long-term traffic data is collected to capture the long-term trend and historical pattern of the node traffic. In turn, the traffic data of the edge node in the next time step is predicted more accurately. Based on the predicted traffic data, dynamic network resource allocation of network slices in each edge node is realized, improving the overall resource utilization and performance of the network.

[0075] As an implementation mode of the present application, after collecting the flow data of each edge node in the target network in the current time step, it also includes:

[0076] The flow data is preprocessed, specifically including: removing invalid data and erroneous data in the flow data; detecting abnormal data in the flow data, and replacing the abnormal data with valid data in the previous time step; and compressing the flow data using a principal component analysis method to obtain compressed data.

[0077] In one embodiment, each edge node collects traffic data through a perception module and then sends it to a central data processing center for preprocessing. Finally, the short-term traffic data and the long-term traffic data are processed into compressed data for subsequent traffic data prediction.

[0078] Specifically, the collection period is defined as Δt, and the traffic data x of the i-th edge node recorded in each collection period is i (t) is expressed in vector form:

[0079] x i (t) = [x i1 (t),x i2 (t),…,x im (t)];

[0080] Among them, x ij (t) represents the j-th type of traffic data of the ith edge node at the current time step.

[0081] Furthermore, the perception module calculates the current total traffic data X of the local node i (t):

[0082]

[0083] A data processing center is also deployed in the target network and is connected to each edge node. In this embodiment, each edge node sends the local total flow data to the data processing center, and the data processing center performs preprocessing operations.

[0084] Specifically, the preprocessing includes the following steps:

[0085] (1) Data cleaning, including removing invalid flow data and erroneous data. For invalid data, invalid data is defined as data points with flow values ​​of 0 or negative. The following expression is used to remove invalid data and obtain valid flow data x i ′ (t):

[0086]

[0087] Error data usually include abnormal peaks and abnormal valleys. Abnormal peaks and abnormal valleys are defined as data that exceeds the mean of historical data by 3-5 times the standard deviation. Then, the following expression is used to remove the error data to obtain the flow data x″ i (t):

[0088]

[0089] Among them, μ j is the mean value of the j-th type of traffic data, σ j is the standard deviation of the j-th type of traffic data.

[0090] (2) Detect outliers in the traffic data and replace them with normal values. Specifically, use the Z-score method to detect outliers and calculate x ij (t) Z-score value of traffic data, defining data points exceeding 3-5 times the standard deviation as outliers, and detecting them using the following expression:

[0091]

[0092] Among them, μ j is the mean value of the j-th type of traffic data, σ j is the standard deviation of the j-th type of flow data. The way to deal with outliers is: use the valid flow data x″ in the previous time step (t-1) i (t) is replaced to obtain the flow data x″′ i (t):

[0093]

[0094] (3) Data compression. In this embodiment, the principal component analysis method is used to reduce the dimension of the flow data to reduce the data transmission overhead, calculate the covariance matrix C of the flow data, solve the eigenvalues ​​and eigenvectors, select the eigenvector corresponding to the largest eigenvalue in the previous time step (t-1) as the principal component, and perform data compression using the following expression:

[0095] Y i(t) = W T x′′ i (t)

[0096] Among them, W is a matrix composed of the feature vectors of the previous traffic data, Y i (t) is the compressed traffic data of the i-th edge node at the current time step.

[0097] In this embodiment, by removing invalid data, erroneous data and processing outliers, unnecessary data calculations are reduced for the subsequent prediction stage, thereby improving prediction efficiency. In addition, the collected traffic data is compressed to reduce the transmission overhead of edge nodes communicating with the data processing center, thereby saving network resources and improving the overall efficiency and accuracy of the network.

[0098] As an implementation mode of the present application, based on the current traffic data of each edge node, predicting the predicted traffic data of each edge node in the next time step includes:

[0099] Constructing a first prediction model based on reinforcement learning, and using a Q-learning algorithm to train the first prediction model;

[0100] The trained first prediction model is used to process the traffic data of each edge node to obtain the predicted traffic data of each edge node in the next time step.

[0101] In one embodiment, a first prediction model based on reinforcement learning is constructed, the model is trained using a Q-learning algorithm, and the trained first prediction model is used to optimize network resources for each edge node.

[0102] Specifically, define the state space S: state s t Contains the compressed flow data Y for the current time step i (t) and network resource configuration R t ;

[0103] Define action space A: action a t Including adjusting the network resource allocation strategy g;

[0104] Based on the actual flow data and predicted flow data in the next time step, calculate the error reward r(s t ,a t ):

[0105]

[0106] in, is the predicted traffic data for the next time step, X i (t+1) is the actual flow data in the next time step.

[0107] Update the Q value table by the following expression:

[0108]

[0109] Among them, α is the learning rate, γ is the discount factor; a ′ Represents the state s at the next time step t+1 The maximum Q value is obtained by:

[0110]

[0111] In the process of training the first prediction model, by defining appropriate state space s, action space a and reward function r, the Q value is iteratively updated by continuous interaction with the environment to minimize the prediction error.

[0112] In this embodiment, the characteristic of reinforcement learning that no prior knowledge is required when facing unknown and complex environments and learning is performed only through interaction with the environment is utilized, and a first prediction model based on reinforcement learning is deployed in each edge node. The first prediction model is used to autonomously learn the dynamic traffic change patterns at multiple time scales in the edge nodes, thereby providing a more accurate basis for the dynamic adjustment of network resources.

[0113] As an implementation mode of the present application, based on the current traffic data of each edge node, predicting the predicted traffic data of each edge node in the next time step includes:

[0114] Constructing and training a second prediction model based on a neural network, wherein the second prediction model includes sub-neural networks deployed on each edge node; in the trained second prediction model, the parameters of each sub-neural network are the same;

[0115] The local traffic data is processed using each sub-neural network of the second prediction model to obtain the local predicted traffic data in the next time step.

[0116] In one embodiment, a second prediction model is constructed using federated learning technology to achieve comprehensive optimization of all edge nodes in the network. The second prediction model includes multiple sub-neural networks, and the same sub-neural network is deployed on each edge node. The model uses deep learning for resource allocation, and the aggregation of global model parameters is achieved through the interaction between each edge node and the data processing center. After the second prediction model is trained, the parameters of the sub-neural networks in all edge nodes are the same. The local multi-time scale traffic data is predicted by each sub-neural network, thereby achieving network resource optimization of each edge node.

[0117] In this embodiment, the traffic data collected by each edge node does not need to be uploaded to the data processing center, thus protecting user privacy. In addition, compared with the traditional central computing method, this solution distributes the model calculation to each edge node, and the edge node and the data processing center only need to transmit a small amount of model parameter data, which reduces the computing burden of a single node and improves the efficiency of resource allocation and the overall performance of the network.

[0118] In addition, each edge node sends the sub-neural network parameters to the data processing center, which aggregates the parameters based on global node considerations, achieving efficient global network resource optimization and making the dynamic adjustment of network resources at each edge node in the network more flexible and efficient.

[0119] As an implementation of the present application, training a second prediction model based on a neural network includes:

[0120] Iteratively train the local sub-neural network using the traffic data of each edge node. In each round of training, based on the local actual traffic data and the predicted traffic data of the sub-neural network, calculate the loss value and update the parameters of the sub-neural network.

[0121] Calculate the comprehensive parameters based on the parameters of all sub-neural networks;

[0122] Based on the comprehensive parameters, the parameters of all sub-neural networks are updated and the next round of training is carried out.

[0123] In one embodiment, the local sub-neural network is iteratively trained using the traffic data of each edge node, and parameter aggregation is performed in each round of training.

[0124] Specifically, in each round of training, the predicted traffic data of the sub-neural network is calculated through the prediction function, and the loss value is calculated based on the local actual traffic data and the predicted traffic data, and the parameters θ of the sub-neural network are updated. i ; Among them, the loss value L i (θ i ) is calculated as follows:

[0125]

[0126] Among them, N i is the amount of data of the ith edge node, l is the loss function, f(Y i (t j );θ i ) is the prediction function of the sub-neural network, Y i (t j ) is the j-th type of compressed traffic data of the i-th edge node in the current time step.

[0127] For each sub-neural network parameter θ i Perform aggregation and calculate the federation parameter θglobal:

[0128]

[0129] Wherein, M is the total number of edge nodes, and N is the total amount of compressed traffic data of all edge nodes.

[0130] Based on the aggregated federated parameter θglobal, update the parameters θ of each sub-neural network i , and continue with the next round of training until the model converges.

[0131] This embodiment uses federated learning technology to construct the second prediction model, so that each edge node can jointly train the model while protecting local data privacy, thereby improving the generalization ability and prediction accuracy of the second prediction model, and realizing efficient global network slice resource optimization in large-scale, distributed 5G networks.

[0132] As an implementation mode of the present application, predicting the predicted traffic data of each edge node in the next time step includes:

[0133] Obtaining a first prediction value obtained by the edge node based on long-term traffic data and a second prediction value obtained based on short-term traffic data;

[0134] A long-term weight and a short-term weight are set based on the device type of the edge node; the sum of the long-term weight and the short-term weight is 1;

[0135] The predicted traffic data of the edge node is calculated according to the first predicted value, the second predicted value, the long-term weight and the short-term weight of each edge node.

[0136] In one embodiment, the short-term traffic data and long-term traffic data of each edge node are predicted separately through a prediction model (a first prediction model or a second prediction model), and the prediction results of multiple time scales are comprehensively analyzed to obtain the predicted traffic data of the edge node in the next time step, thereby improving the accuracy of the prediction results.

[0137] Short-term traffic data reflects the traffic change trend in the recent period of time. To capture short-term traffic fluctuations, it is defined as Specifically, the most recent T s The flow data of time steps is expressed as:

[0138]

[0139] Among them, Y i (t) is the compressed traffic data of the i-th edge node at the current time step.

[0140] Long-term traffic data reflects long-term traffic trends and periodic characteristics, and is defined as Long-term traffic data is the past T l The flow data of time steps is expressed as:

[0141]

[0142] Among them, Y i (t) is the compressed traffic data of the i-th edge node at the current time step.

[0143] Combining short-term traffic data with long-term traffic data can provide a more comprehensive understanding of network traffic patterns, while building more accurate multi-time scale prediction models to improve prediction accuracy.

[0144] Specifically, long-term traffic data Input the prediction model to obtain the first predicted traffic data: Short-term traffic data Input the prediction model to obtain the second predicted traffic data:

[0145] According to the device configuration type or business deployment of the edge node, set the corresponding long-term weight w l and the short-term weight w s To indicate the node's preference for long-term and short-term traffic. The long-term weight and short-term weight satisfy the following relationship: s +w l =1.

[0146] Based on the first prediction value and the second prediction value, calculate the predicted flow data

[0147]

[0148] In this embodiment, the prediction results of short-term traffic data and long-term traffic data are weighted and fused to generate the final predicted load value, which comprehensively considers the traffic characteristics under different time scales and obtains more accurate and stable prediction results. This multi-time scale prediction method not only improves the accuracy of the prediction, but also can better adapt to the different dynamic changes of traffic. Compared with the single time scale prediction method in the related art, the method of the present application can more comprehensively capture the traffic change law, thereby providing a more accurate basis for the dynamic adjustment of network resources.

[0149] As an implementation mode of the present application, according to the target allocation strategy and the predicted traffic data of each edge node in the next time step, the target resource configuration in each edge node in the next time step is determined, including:

[0150] Obtaining an initial allocation strategy for the target network;

[0151] According to the initial allocation strategy, the current network resource configuration of each edge node and the predicted traffic data, the network resource configuration in the next time step is calculated as the target resource configuration.

[0152] In one embodiment, the initial allocation strategy in the target network (i.e., the allocation strategy in the current time step) is used as the target allocation strategy, and the target resource configuration in the next time step is calculated based on the predicted traffic data of each edge node. Specifically, the local target resource configuration R in the next time step is slice,t+1 for:

[0153]

[0154] in, is the predicted traffic data of the ith edge node, g is the initial allocation strategy, R slice,t It is the current network resource configuration of the target network.

[0155] As an implementation mode of the present application, after calculating the initial resource configuration in the next time step, it also includes:

[0156] Obtain the current resource allocation of each network slice in each edge node, and calculate the overall utility and fairness of the target network based on the current resource allocation;

[0157] Based on the predicted traffic data of each edge node, the initial allocation strategy is adjusted using a gradient descent method to maximize the overall utility and optimize the fairness, thereby obtaining a first allocation strategy;

[0158] Based on the first allocation strategy, the current network resource configuration of each edge node and the predicted traffic data, the network resource configuration in each edge node in the next time step is calculated, and the target resource configuration is updated.

[0159] In one embodiment, when dynamically optimizing network resources, the overall utility of the network and the fairness of each edge node also need to be considered. Therefore, in this embodiment, the initial allocation strategy of the network is dynamically optimized based on the overall utility and fairness to further improve the overall resource utilization and performance of the network.

[0160] Figure 2 1 is a flow chart of adjusting the target allocation strategy in one embodiment of the present application. Figure 2As shown in the figure, the initial target resource configuration is calculated based on the current network resource configuration of each node in the target network and the initial allocation strategy. Then, according to the current overall network resource configuration, the overall utility and fairness are calculated. Taking both factors into consideration, the initial allocation strategy is adjusted using the gradient descent method to maximize the overall utility and optimize fairness, thus obtaining the first allocation strategy.

[0161] First, the network resources of each edge node are configured to calculate the overall utility U(R) and fairness F(R) of the target network:

[0162]

[0163] Where N is the number of edge nodes, R i represents the network resource configuration of the ith edge node, u i is the utility function of the ith edge node.

[0164] Then, the gradient descent method is used to adjust the initial allocation strategy g so that the overall utility is maximized and the fairness is optimized, and the optimization result is obtained, that is, the first allocation strategy g opt The specific optimization strategies are as follows:

[0165] (Maximizing overall utility)

[0166] F(R)≥θ; (fairness is not less than the fairness threshold)

[0167] (The allocated network resources cannot exceed the total network resources)

[0168] (Allocated network resources are not negative)

[0169] Among them, R i represents the network resource configuration of the i-th edge node, θ is the fairness threshold, R total is the total allocatable network resources.

[0170] Further, the network resource configuration of the target network is dynamically adjusted according to the optimization result (the first allocation strategy), and a new target resource configuration R is generated based on the following expression: slice-opt,t+1 :

[0171]

[0172] Among them, g opt is the first allocation strategy, is the predicted traffic data of the ith edge node, R slice,t It is the current network resource configuration of the target network.

[0173] Furthermore, based on the target resource configuration Rslice-opt,t+1 , allocate corresponding network resources to the network slice of each edge node.

[0174] In this embodiment, in the process of optimizing resource allocation, the overall utility of the target network and the fairness of each edge node are taken into consideration, so as to achieve fairness in resource allocation of each edge node while maximizing the overall utility of the network, thereby ensuring the service quality of each edge node.

[0175] As an implementation manner of the present application, the network slice resource optimization method further includes:

[0176] Obtain the current network resource configuration of multiple target networks and calculate the comprehensive utility and comprehensive fairness;

[0177] According to the optimization preference, the utility weight and the fairness weight are set; the sum of the utility weight and the fairness weight is 1; the optimization preference is utility priority or fairness priority;

[0178] Calculating comprehensive performance based on the utility weight and the fairness weight;

[0179] Obtaining a comprehensive allocation strategy of the multiple target networks, and adjusting the comprehensive allocation strategy using a gradient descent method to maximize the comprehensive performance, thereby obtaining a second allocation strategy;

[0180] Calculate the network resource configuration of each target network in the next time step as the comprehensive resource configuration according to the second allocation strategy, the current network resource configuration of each target network and the predicted traffic data;

[0181] Based on the comprehensive resource configuration, network resources are allocated to each target network.

[0182] In this embodiment, network resource configurations of multiple target networks are combined to dynamically allocate network resources across network layers, thereby achieving end-to-end resource optimization across network layers. Figure 3 1 is a flow chart of adjusting the cross-layer target allocation strategy in one embodiment of the present application. Figure 3 As shown in the figure, first obtain the network resource configuration of each target network, and calculate the comprehensive utility and comprehensive fairness. Set the utility weight and fairness weight according to the optimization preference of the dynamic adjustment of network resources to indicate the preference of this optimization for the overall utility and fairness, and calculate the comprehensive performance based on the set weights. Based on the comprehensive performance, the gradient descent method is used to adjust the comprehensive allocation strategy of all current target networks to maximize the comprehensive performance and obtain the optimized second allocation strategy.

[0183] In one embodiment, multiple target networks take access network, transmission network and core network as examples, first obtain the current network resource status of each target network, and then determine the current network resource configuration. Specifically, the access network resource status A(t), the transmission network resource status T(t) and the core network resource status C(t). The network resource status includes: bandwidth, delay, packet loss rate, available resources, etc.

[0184] Define the resource state vector R layer :

[0185] R layer =[a(t);T(t);c(t)].

[0186] Based on the network resource configuration of each target network, the overall comprehensive utility U(R(t)) and comprehensive fairness F(R(t)) of the network are calculated:

[0187]

[0188] Where N is the number of target networks, U(R) is the overall utility of the i-th network, and R(t) represents the network resource configuration of the i-th target network.

[0189] The purpose of cross-layer optimization is to maximize the overall performance of the network and ensure the resource utilization and service quality of each network layer. Based on this, the comprehensive performance of cross-layer optimization J(R(t)) is defined as:

[0190] J(R(t))=αU(R(t))+βF(R(t));

[0191] Among them, U(R(t)) is the overall utility function across layers, F(R(t)) is the fairness function across layers, α and β are weight coefficients, representing the relative importance of utility and fairness, respectively, and satisfying α+β=1.

[0192] Obtain the comprehensive allocation strategy for all target networks, and use the gradient descent method as the optimization algorithm to adjust the comprehensive allocation strategy to maximize the overall utility J(R(t)) and obtain the second allocation strategy. The specific optimization rules are as follows:

[0193] (Maximizing overall utility)

[0194] (The allocated network resources cannot exceed the total network resources)

[0195] (Allocated network resources are not negative)

[0196] Among them, R i is the network resource configuration of the i-th target network, R totalis the total allocatable network resources.

[0197] Further, the comprehensive resource configuration is calculated according to the optimization result (second allocation strategy), and the network resource configuration of each target network is dynamically adjusted based on the comprehensive resource configuration. Specifically, a new comprehensive resource configuration R is generated based on the following expression: layer,t+1 :

[0198] R layer,t+1 =h(R(t),R layer );

[0199] Among them, h is the optimization result (second allocation strategy), R layer is the current resource state vector.

[0200] In this embodiment, the multi-layer characteristics of the network are taken into account. By obtaining the network resource configurations of multiple target networks (network layers) and performing cross-layer optimization, dynamic resource allocation optimization across network layers based on overall utility and fairness is achieved. This enables more refined and efficient network resource management for multi-layer networks, improves the overall performance of the network, and ensures the stability and reliability of services in each network layer.

[0201] Based on the same inventive concept, an embodiment of the present application provides a network slice resource optimization device. Figure 4 , Figure 4 Schematic diagram of a network slice resource optimization device 100 proposed in an embodiment of the present application. Figure 4 As shown, the device comprises:

[0202] The perception module 101 is configured to collect traffic data of each edge node in the target network in the current time step according to the size of the time step; the traffic data includes: short-term traffic data and long-term traffic data of all network slices in the edge node;

[0203] The prediction module 102 is configured to predict the predicted traffic data of each edge node in the next time step based on the current traffic data of each edge node;

[0204] The optimization module 103 is configured to determine a target allocation strategy based on the predicted traffic data of each edge node in the next time step; determine a target resource configuration in each edge node in the next time step according to the target allocation strategy and the predicted traffic data of each edge node in the next time step;

[0205] The allocation module 104 is configured to allocate network resources to each edge node according to the target resource configuration.

[0206] As an implementation manner of the present application, the device further includes a first training module, configured to construct a first prediction model based on reinforcement learning, and train the first prediction model using a Q learning algorithm;

[0207] The prediction module 102 is specifically configured to use the trained first prediction model to process the traffic data of each edge node to obtain the predicted traffic data of each edge node in the next time step.

[0208] As an implementation manner of the present application, the device further includes a second training module, which is configured to construct and train a second prediction model based on a neural network, wherein the second prediction model includes sub-neural networks deployed on each edge node; in the trained second prediction model, the parameters of each sub-neural network are the same;

[0209] The prediction module 102 is specifically configured to use each sub-neural network of the second prediction model to process the local traffic data to obtain the local predicted traffic data in the next time step.

[0210] As an implementation manner of the present application, the second training module is configured to construct and train a second prediction model based on a neural network, specifically including:

[0211] Iteratively train the local sub-neural network using the traffic data of each edge node. In each round of training, based on the local actual traffic data and the predicted traffic data of the sub-neural network, calculate the loss value and update the parameters of the sub-neural network.

[0212] Calculate the comprehensive parameters based on the parameters of all sub-neural networks;

[0213] Based on the comprehensive parameters, the parameters of all sub-neural networks are updated and the next round of training is carried out.

[0214] As an implementation manner of the present application, the prediction module 102 is configured to predict the predicted traffic data of each edge node in the next time step, specifically including:

[0215] Obtaining a first prediction value obtained by the edge node based on long-term traffic data and a second prediction value obtained based on short-term traffic data;

[0216] A long-term weight and a short-term weight are set based on the device type of the edge node; the sum of the long-term weight and the short-term weight is 1;

[0217] The predicted traffic data of the edge node is calculated according to the first predicted value, the second predicted value, the long-term weight and the short-term weight of each edge node.

[0218] As an implementation mode of the present application, the optimization module 103 is configured to determine the target resource configuration in each edge node in the next time step according to the target allocation strategy and the predicted traffic data of each edge node in the next time step, specifically including:

[0219] Obtaining an initial allocation strategy for the target network;

[0220] According to the initial allocation strategy, the current network resource configuration of each edge node and the predicted traffic data, the network resource configuration in the next time step is calculated as the target resource configuration.

[0221] As an implementation manner of the present application, the optimization module 103 is further configured to perform the following operations after calculating the initial resource configuration in the next time step:

[0222] Obtain the current resource allocation of each network slice in each edge node, and calculate the overall utility and fairness of the target network based on the current resource allocation;

[0223] Based on the predicted traffic data of each edge node, the initial allocation strategy is adjusted using a gradient descent method to maximize the overall utility and optimize the fairness, thereby obtaining a first allocation strategy;

[0224] Based on the first allocation strategy, the current network resource configuration of each edge node and the predicted traffic data, the network resource configuration in each edge node in the next time step is calculated, and the target resource configuration is updated.

[0225] As an implementation mode of the present application, the perception module 101 is configured to perform the following steps after collecting the flow data of each edge node in the target network in the current time step:

[0226] The flow data is preprocessed, specifically including: removing invalid data and erroneous data in the flow data; detecting abnormal data in the flow data, and replacing the abnormal data with valid data in the previous time step; and compressing the flow data using a principal component analysis method to obtain compressed data.

[0227] As an implementation manner of the present application, the optimization module 103 is further configured to perform the following steps:

[0228] Obtain the current network resource configuration of multiple target networks and calculate the comprehensive utility and comprehensive fairness;

[0229] According to the optimization preference, the utility weight and the fairness weight are set; the sum of the utility weight and the fairness weight is 1; the optimization preference is utility priority or fairness priority;

[0230] Calculating comprehensive performance based on the utility weight and the fairness weight;

[0231] Obtaining a comprehensive allocation strategy of the multiple target networks, and adjusting the comprehensive allocation strategy using a gradient descent method to maximize the comprehensive performance, thereby obtaining a second allocation strategy;

[0232] Calculate the network resource configuration of each target network in the next time step as the comprehensive resource configuration according to the second allocation strategy, the current network resource configuration of each target network and the predicted traffic data;

[0233] The allocation module 104 is further configured to allocate network resources to each target network based on the comprehensive resource configuration.

[0234] Based on the same inventive concept, an embodiment of the present application provides a network slicing resource optimization system. Figure 5 Schematic diagram of a network slice resource optimization system proposed in an embodiment of the present application. Figure 5 As shown, the system includes:

[0235] Multiple target networks, the multiple target networks include: an access network, a transmission network and a core network; wherein the access network is deployed with:

[0236] A plurality of edge nodes as described in any of the above embodiments; wherein each edge node is deployed with a network slice resource optimization device as described in any of the above embodiments;

[0237] The data processing center is used to aggregate the traffic data or sub-neural network parameters sent by each edge node.

[0238] Based on the same inventive concept, an embodiment of the present application provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the steps in the network slice resource optimization method described in any of the above embodiments of the present application are implemented.

[0239] Based on the same inventive concept, an embodiment of the present application provides a readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the network slice resource optimization method described in any of the above embodiments of the present application are implemented.

[0240] Based on the same inventive concept, an embodiment of the present application provides an electronic device, referring to Figure 6 . Figure 6 : is a schematic diagram of an electronic device proposed in one embodiment of the present application. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the computer program is executed by the processor, the steps in the network slice resource optimization method described in any of the above embodiments of the present application are implemented.

[0241] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0242] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0243] For the method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and components involved are not necessarily required by the present application.

[0244] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the embodiments of the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0245] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0246] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0247] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0248] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic creative concepts. Therefore, the present application is interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0249] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0250] The above is a detailed introduction to the network slicing resource optimization method, device, system and product provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A network slice resource optimization method, characterized in that: include: According to the size of the time step, the traffic data of each edge node in the target network in the current time step is collected; The traffic data includes: short-term traffic data and long-term traffic data of all network slices in the edge node; Based on the current traffic data of each edge node, predict the predicted traffic data of each edge node in the next time step; Determine a target allocation strategy based on the predicted traffic data of each edge node in the next time step; determine a target resource configuration in each edge node in the next time step according to the target allocation strategy and the predicted traffic data of each edge node in the next time step; Allocate network resources to each edge node according to the target resource configuration.

2. The network slice resource optimization method according to claim 1, characterized in that: Based on the current traffic data of each edge node, the predicted traffic data of each edge node in the next time step is predicted, including: Constructing a first prediction model based on reinforcement learning, and using a Q-learning algorithm to train the first prediction model; The trained first prediction model is used to process the traffic data of each edge node to obtain the predicted traffic data of each edge node in the next time step.

3. The network slice resource optimization method according to claim 1, characterized in that: Based on the current traffic data of each edge node, the predicted traffic data of each edge node in the next time step is predicted, including: Constructing and training a second prediction model based on a neural network, wherein the second prediction model includes sub-neural networks deployed on each edge node; in the trained second prediction model, the parameters of each sub-neural network are the same; The local traffic data is processed using each sub-neural network of the second prediction model to obtain the local predicted traffic data in the next time step.

4. The network slice resource optimization method according to claim 3, characterized in that: Training a second prediction model based on a neural network includes: Iteratively train the local sub-neural network using the traffic data of each edge node. In each round of training, based on the local actual traffic data and the predicted traffic data of the sub-neural network, calculate the loss value and update the parameters of the sub-neural network. Calculate the comprehensive parameters based on the parameters of all sub-neural networks; Based on the comprehensive parameters, the parameters of all sub-neural networks are updated and the next round of training is carried out.

5. The network slice resource optimization method according to claim 2 or 3, characterized in that: Predict the predicted traffic data of each edge node in the next time step, including: Obtaining a first prediction value obtained by the edge node based on long-term traffic data and a second prediction value obtained based on short-term traffic data; A long-term weight and a short-term weight are set based on the device type of the edge node; the sum of the long-term weight and the short-term weight is 1; The predicted traffic data of the edge node is calculated according to the first predicted value, the second predicted value, the long-term weight and the short-term weight of each edge node.

6. The network slice resource optimization method according to claim 1, characterized in that: According to the target allocation strategy and the predicted traffic data of each edge node in the next time step, the target resource configuration in each edge node in the next time step is determined, including: Obtaining an initial allocation strategy for the target network; According to the initial allocation strategy, the current network resource configuration of each edge node and the predicted traffic data, the network resource configuration in the next time step is calculated as the target resource configuration.

7. The network slice resource optimization method according to claim 6, characterized in that: After calculating the initial resource configuration for the next time step, it also includes: Obtain the current resource allocation of each network slice in each edge node, and calculate the overall utility and fairness of the target network based on the current resource allocation; Based on the predicted traffic data of each edge node, the initial allocation strategy is adjusted using a gradient descent method to maximize the overall utility and optimize the fairness, thereby obtaining a first allocation strategy; Based on the first allocation strategy, the current network resource configuration of each edge node and the predicted traffic data, the network resource configuration in each edge node in the next time step is calculated, and the target resource configuration is updated.

8. The network slice resource optimization method according to claim 1, characterized in that: After collecting the traffic data of each edge node in the target network in the current time step, it also includes: The flow data is preprocessed, specifically including: removing invalid data and erroneous data in the flow data; detecting abnormal data in the flow data, and replacing the abnormal data with valid data in the previous time step; and compressing the flow data using a principal component analysis method to obtain compressed data.

9. The network slice resource optimization method according to claim 1, characterized in that: Also includes: Obtain the current network resource configuration of multiple target networks and calculate the comprehensive utility and comprehensive fairness; According to the optimization preference, the utility weight and the fairness weight are set; the sum of the utility weight and the fairness weight is 1; the optimization preference is utility priority or fairness priority; Calculating comprehensive performance based on the utility weight and the fairness weight; Obtaining a comprehensive allocation strategy of the multiple target networks, and adjusting the comprehensive allocation strategy using a gradient descent method to maximize the comprehensive performance, thereby obtaining a second allocation strategy; Calculate the network resource configuration of each target network in the next time step as the comprehensive resource configuration according to the second allocation strategy, the current network resource configuration of each target network and the predicted traffic data; Based on the comprehensive resource configuration, network resources are allocated to each target network.

10. A network slice resource optimization device, characterized in that: Used to implement the method according to any one of claims 1 to 9, comprising: The perception module is configured to collect traffic data of each edge node in the target network in the current time step according to the size of the time step; the traffic data includes: short-term traffic data and long-term traffic data of all network slices in the edge node; The prediction module is configured to predict the predicted traffic data of each edge node in the next time step based on the current traffic data of each edge node; The optimization module is configured to determine a target allocation strategy based on the predicted traffic data of each edge node in the next time step; determine a target resource configuration in each edge node in the next time step according to the target allocation strategy and the predicted traffic data of each edge node in the next time step; The allocation module is configured to allocate network resources to each edge node according to the target resource configuration.

11. A network slice resource optimization system, characterized in that: include: Multiple target networks, the multiple target networks include: an access network, a transmission network and a core network; wherein the access network is deployed with: A plurality of edge nodes according to any one of the methods of claims 1 to 9; wherein each edge node is deployed with a network slice resource optimization device according to claim 10; The data processing center is used to aggregate the traffic data or sub-neural network parameters sent by each edge node.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps in the method of any one of claims 1 to 9 are implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 9 are implemented.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps in the method according to any one of claims 1 to 9 are implemented.

Images