Traffic load balancing method and device, electronic equipment and storage medium

Abstract

The invention provides a traffic load balancing method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining traffic data of a preset period of a base station, carrying out the feature analysis of the traffic data through employing a pre-training network traffic prediction model, and obtaining the spatial-temporal feature representation of the traffic data, and obtaining predicted flow data of the base station at the next moment according to the spatial-temporal characteristic representation, calculating to obtain pheromones between the base station and the adjacent base station by adopting the actual flow data, the predicted flow data and the adjacency matrix, and calculating to obtain heuristic information between the base station and the adjacent base station by adopting the residual flow capacity and the adjacency matrix. And calculating a path selection probability of traffic transmission between the base station and the adjacent base station according to the pheromones and the heuristic information, determining a target transmission path, and performing traffic distribution on the base stations on the target transmission path. According to the method, high performance is ensured to be kept under high-flow load through network flow prediction, meanwhile, overload of a network port is avoided through load balancing, and the stability and high efficiency of the network are ensured.

Description

Technical Field

[0001] The present application belongs to the field of communication technology, and specifically relates to a traffic load balancing method, a traffic load balancing device, an electronic device, and a computer-readable storage medium. Background Art

[0002] In today's rapidly developing communication technology, the growth of network scale and the massive access needs of end users have led to an exponential growth in communication network service traffic. In this case, measuring and monitoring network traffic to ensure that the network and server maintain high performance under high traffic loads while avoiding network port overload has become the key to improving network performance and user experience.

[0003] At present, communication network traffic prediction uses real-time analysis and deep learning models to predict future traffic trends, thereby optimizing network load distribution. Since network traffic has the characteristics of self-similarity, periodicity and chaos, and is affected by the spatial node topology structure, network traffic is essentially a nonlinear spatiotemporal series data. However, traditional network traffic prediction is mainly based on time series analysis, ignoring the spatiotemporal characteristics of network traffic, affecting the accuracy of traffic prediction results, and making it difficult to cope with load fluctuations in complex network environments. It is easy to cause uneven network load distribution, further affecting network performance. Summary of the invention

[0004] In view of this, the present invention aims to propose a traffic load balancing method, device, electronic device and storage medium to solve the problem that the current network traffic prediction ignores the temporal and spatial characteristics of network traffic, the traffic prediction results are inaccurate, and the network load distribution is uneven, which affects the network performance.

[0005] According to a first aspect of the present invention, a traffic load balancing method is provided, the method comprising:

[0006] Obtain the traffic data of the base station at a preset period;

[0007] Using a pre-trained network traffic prediction model to perform feature analysis on the traffic data, obtain a spatiotemporal feature representation of the traffic data, and obtain predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation;

[0008] Obtain the adjacency matrix of the base station, the actual traffic data at the current moment, and the remaining traffic capacity;

[0009] The actual traffic data, the predicted traffic data and the adjacency matrix are used to calculate the pheromone between the base station and the adjacent base stations;

[0010] The heuristic information between the base station and the adjacent base stations is calculated using the remaining flow capacity and the adjacency matrix;

[0011] Calculating the path selection probability of traffic transmission between the base station and the adjacent base stations according to the pheromone and the heuristic information, and determining the target transmission path;

[0012] Traffic is distributed to base stations on the target transmission path.

[0013] Optionally, the adopting a pre-trained network traffic prediction model to perform feature analysis on the traffic data to obtain a spatiotemporal feature representation of the traffic data, and obtaining predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation, includes:

[0014] Using a pre-trained network traffic prediction model to perform feature analysis on the traffic data to obtain a temporal feature representation and a spatial feature representation of the traffic data;

[0015] The temporal feature representation and the spatial feature representation are integrated to obtain the spatiotemporal feature representation of the traffic data;

[0016] The spatiotemporal feature representation is mapped to obtain the predicted traffic data of the base station at the next moment.

[0017] Optionally, the adopting a pre-trained network traffic prediction model to perform feature analysis on the traffic data to obtain a temporal feature representation and a spatial feature representation of the traffic data includes:

[0018] Convolution processing is performed on the flow data and a preset time embedding matrix to obtain a time series feature representation of the flow data;

[0019] Performing static adaptive graph learning and dynamic adaptive graph learning on the traffic data to obtain static spatial relationships and dynamic spatial relationships;

[0020] The static spatial relationship and the dynamic spatial relationship are fused to obtain a spatial feature representation of the traffic data.

[0021] Optionally, the using the actual traffic data, the predicted traffic data and the adjacency matrix to calculate pheromones between a base station and an adjacent base station includes:

[0022] Calculating the flow difference between the actual flow data and the predicted flow data, and determining the adjacent base stations of the base station according to the adjacency matrix of the base station;

[0023] Calculating the pheromone increment of the adjacent base station by using the flow difference and a preset increment parameter;

[0024] The pheromone increment is used to adjust the current pheromone of the adjacent base station to obtain the pheromone between the base station and the adjacent base station.

[0025] Optionally, the using the remaining traffic capacity and the adjacency matrix to calculate the heuristic information between the base station and the adjacent base stations includes:

[0026] The remaining flow capacity of the base station is calculated with the adjacency matrix, and non-adjacent matrices in the adjacency matrix are adjusted to zero to obtain heuristic information between the base station and adjacent base stations.

[0027] Optionally, calculating the path selection probability of traffic transmission between the base station and the adjacent base station according to the pheromone and the heuristic information to determine the target transmission path includes:

[0028] The pheromone and the heuristic information are used to calculate the selection probability of each path connecting the base station to the base station, so as to obtain the path selection probability of traffic transmission between the base station and the adjacent base station;

[0029] A target transmission path is selected from each path connected to the base station according to the path selection probability.

[0030] Optionally, after allocating traffic to the base stations on the target transmission path, the method further includes:

[0031] Storing the target transmission path into an initial solution space matrix;

[0032] Monitoring the flow distribution of the base station and updating the remaining flow capacity of the target transmission path;

[0033] Calculating the resource utilization rate of the target transmission path and the variance of the resource utilization rate using the remaining traffic capacity;

[0034] The target transmission path whose variance is less than a preset threshold is determined as the optimal transmission path.

[0035] According to a second aspect of the present invention, a traffic load balancing device is provided, the device comprising:

[0036] A first acquisition module, used to acquire flow data of a preset period of a base station;

[0037] A traffic prediction module, used to perform feature analysis on the traffic data using a pre-trained network traffic prediction model to obtain a spatiotemporal feature representation of the traffic data, and obtain predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation;

[0038] The second acquisition module is used to acquire the adjacency matrix of the base station, the actual flow data at the current moment, and the remaining flow capacity;

[0039] A pheromone calculation module, used to calculate the pheromone between the base station and the adjacent base stations using the actual traffic data, the predicted traffic data and the adjacency matrix;

[0040] A heuristic information calculation module, used to calculate the heuristic information between the base station and the adjacent base stations using the remaining flow capacity and the adjacency matrix;

[0041] A path determination module, used to calculate the path selection probability of traffic transmission between the base station and the adjacent base stations according to the pheromone and the heuristic information, and determine the target transmission path;

[0042] The traffic distribution module is used to distribute traffic to the base stations on the target transmission path.

[0043] According to another aspect of the present invention, there is also provided an electronic device, comprising:

[0044] processor;

[0045] a memory for storing instructions executable by the processor;

[0046] The processor is configured to execute the instructions to implement the traffic load balancing method as described above.

[0047] According to another aspect of the present invention, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the traffic load balancing method described above are implemented.

[0048] The traffic load balancing method provided by the embodiment of the present invention obtains the traffic data of a preset period of the base station, uses a pre-trained network traffic prediction model to perform feature analysis on the traffic data, obtains the spatiotemporal feature representation of the traffic data, obtains the predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation, obtains the adjacency matrix of the base station, the actual traffic data at the current moment and the remaining traffic capacity, uses the actual traffic data, the predicted traffic data and the adjacency matrix to calculate the pheromone between the base station and the adjacent base stations, uses the remaining traffic capacity and the adjacency matrix to calculate the heuristic information between the base station and the adjacent base stations, calculates the path selection probability of the traffic transmission between the base station and the adjacent base stations according to the pheromone and the heuristic information, determines the target transmission path, and distributes traffic to the base stations on the target transmission path. The embodiment of the present invention realizes the fusion of the spatiotemporal characteristics of traffic data through a pre-trained network traffic prediction model, realizes traffic prediction that conforms to the spatiotemporal characteristics of network traffic, improves the accuracy of business traffic prediction in complex and changeable cellular network environments, provides reliable data support for load balancing, calculates pheromones and heuristic information based on traffic prediction results, obtains the changing trends of current and future loads, dynamically adjusts the load distribution between base stations, and more flexibly allocates network resources to base stations with expected high loads to cope with future traffic fluctuations. Through the prediction and monitoring of network traffic, high performance is ensured under high traffic loads, and load balancing is used to avoid network port overload, thereby ensuring the stability and efficiency of the network.

[0049] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0051] Figure 1 It is a flow chart of the steps of a traffic load balancing method provided by an embodiment of the present invention;

[0052] Figure 2 yes Figure 1 Flow chart of step 102 in the traffic load balancing method provided by an embodiment of the present invention;

[0053] Figure 3 yes Figure 1 Flow chart of step 104 in the traffic load balancing method provided by an embodiment of the present invention;

[0054] Figure 4 yes Figure 1 Flow chart of step 106 in the traffic load balancing method provided by an embodiment of the present invention;

[0055] Figure 5 is a flow chart of the steps of another traffic load balancing method provided by an embodiment of the present invention;

[0056] Figure 6 It is a scenario schematic diagram of the traffic load balancing method provided by an embodiment of the present invention;

[0057] Figure 7 It is a structural schematic diagram of a traffic load balancing device provided by an embodiment of the present invention;

[0058] Figure 8 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] To make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in the embodiments of the present invention, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can be implemented. The division of the following embodiments is for the convenience of description, and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined and referenced with each other without contradiction.

[0060] Reference Figure 1 , shows a flow chart of the steps of a traffic load balancing method provided by an embodiment of the present invention, the method may include:

[0061] Step 101, obtaining the flow data of a preset period of a base station.

[0062] In the embodiment of the present invention, in the field of network traffic management, traffic prediction is a key task, which aims to accurately predict network traffic in future time periods. However, with the increase in network scale and application complexity, traditional network prediction models have been unable to meet the needs of traffic prediction in modern networks. The embodiment of the present invention aims at the problem of cellular network traffic prediction and proposes a traffic prediction model based on an adaptive hybrid spatiotemporal graph neural network. By inputting a time embedding matrix to dynamically consider time information, the spatiotemporal characteristics are integrated, and the accuracy and real-time performance of traffic prediction are improved; secondly, based on the business traffic prediction results, the network resource allocation is dynamically adjusted to cope with future traffic peaks and load imbalances, thereby improving network performance and resource utilization efficiency.

[0063] It should be noted that this embodiment is explained by taking a cellular network as an example. The cellular network provides wide coverage by dividing the geographical area into multiple small "cellular" areas (commonly referred to as "cells"). Each cell is managed by a base station or a "base transceiver station". This embodiment adopts an improved adaptive hybrid graph neural network (Adaptive HybridSpatial-Temporal Graph Neural Network, AHSTGNN), which comprehensively considers the spatial correlation, nonlinear time dependency and heterogeneity between cellular towers through hybrid graph learning, temporal convolution and spatiotemporal adaptive modules, thereby achieving better network traffic prediction effects.

[0064] Specifically, the traffic data of the preset period of the base station is obtained. The traffic data includes three periodic spatiotemporal data, namely, the traffic of the recent, daily and weekly cellular tower base stations. Each periodic traffic data has a different length, namely, T R , T D , T W , X∈R T×N×F , where N is the number of cell towers and F is the number of features.

[0065] Step 102, using a pre-trained network traffic prediction model to perform feature analysis on the traffic data, obtain a spatiotemporal feature representation of the traffic data, and obtain the predicted traffic data of the base station at the next moment based on the spatiotemporal feature representation.

[0066] The embodiment of the present invention adopts a pre-trained network traffic prediction model to perform feature analysis on traffic data to obtain the spatiotemporal feature representation of the traffic data, wherein the pre-trained network traffic prediction model is an improved adaptive hybrid graph neural network model, and the pre-trained network traffic prediction model includes an input layer, an adaptive hybrid spatiotemporal learning module and an output layer. The traffic data of the preset period of the base station obtained is input into the input layer of the model, and the traffic data is feature analyzed by the adaptive hybrid spatiotemporal learning module to obtain the spatiotemporal feature representation of the traffic data. The spatiotemporal feature representation obtained by the adaptive hybrid spatiotemporal learning module is connected to the output layer to output the predicted traffic data of the base station at the next moment.

[0067] It should be noted that this embodiment uses an improved adaptive hybrid graph neural network model for traffic prediction, specifically using the convolution operation of GCN to capture the spatial relationship between base stations combined with the time embedding matrix to fully consider the space-time interaction. Specifically, the time embedding matrix is ​​placed on the traffic data so that the model can capture the relationship in the time dimension through learning and updating. In this case, the weights of the time embedding matrix will be trained as part of the model so that it can adapt to the time-related patterns in the task. The time information is dynamically considered through the input time embedding matrix, which can effectively integrate spatial and temporal information. By combining the graph convolution network and the sequence model, the modeling of complex space-time dependencies is realized, thereby improving the accuracy and robustness of the prediction, and providing a more reliable reference for network planning and resource allocation.

[0068] Specifically, refer to Figure 6 , shows the architecture of the adaptive hybrid spatio-temporal graph neural network model of the traffic load balancing method provided by an embodiment of the present invention. The adaptive hybrid spatio-temporal learning module includes three important sub-modules: Temporal Convolution Module (TCM), Adaptive Hybrid Graph Learning Module (AHGLM) and Spatio-Temporal Adaptive Module (STAM). Among them, the temporal convolution module TCM is used to process traffic data, and a gated temporal convolution network can be used to capture the complex relationship and periodicity of time series data to obtain the time series feature representation of traffic data; the adaptive hybrid graph learning module AHGLM is used to capture the spatial relationship between cellular towers, among which the static adaptive graph learning (SAGL) captures relatively stable spatial relationships, and the dynamic adaptive graph learning (STAM) captures relatively stable spatial relationships. The spatial feature representation of traffic data is obtained by fusing the temporal feature representation and spatial feature representation obtained by TCM and AHGLM respectively.

[0069] In an embodiment of the present invention, after obtaining the spatiotemporal feature representation of the traffic data that integrates the spatiotemporal information, the predicted traffic data of the base station at the next moment is obtained according to the spatiotemporal feature representation. Specifically, the output layer performs a jump connection on the output of each adaptive hybrid spatiotemporal learning module and directly connects them to the output layer. With the stacking of multiple adaptive hybrid spatiotemporal learning modules, the time perception of the adaptive hybrid spatiotemporal learning module also increases. The bottom-level blocks pay more attention to the temporally adjacent traffic features, while the high-level blocks focus on long-term time information. The jump connection is used to solve the spatial dependency modeling problem at different time levels. After the jump connection, the outputs of all adaptive hybrid spatiotemporal learning modules are fused. The output layer consists of two fully connected layers to generate the final multi-step prediction, i.e., the predicted traffic data of the base station at the next moment.

[0070]

[0071] It should be noted that this embodiment adds a time embedding matrix in the initialization part of the adaptive hybrid graph neural network model and updates it in each iteration to represent the relationship in the time dimension, wherein the time embedding matrix is ​​generated by sine and cosine functions to dynamically represent time information. Specifically, during the forward propagation process of the static adaptive graph learning SAGL module in the adaptive hybrid graph learning module, the time embedding matrix is ​​applied to the traffic data so that the model can dynamically consider the time information.

[0072] Step 103, obtaining the adjacency matrix of the base station, the actual flow data at the current moment, and the remaining flow capacity.

[0073] In an embodiment of the present invention, in a cellular network, a traditional load balancing strategy allocates resources only based on the network load situation at the current moment, without considering the dynamic nature of the network load, and is difficult to cope with load fluctuations in a complex network environment, resulting in uneven load distribution. This embodiment takes into account the traffic situation at future moments, and adjusts the resource allocation strategy by predicting traffic data to cope with possible load changes in the future. Specifically, by establishing a model, calculating pheromone increments, updating pheromones, and redefining path selection strategies, the difference between actual traffic and predicted traffic is used as the basis for calculating pheromone increments, and the path selection strategy is redefined in combination with heuristic information to achieve more intelligent and efficient load balancing.

[0074] Specifically, first, obtain the adjacency matrix of the base station, the actual traffic data at the current moment, and the remaining traffic capacity. The adjacency matrix of the base station is a square matrix, in which each element indicates whether there is a direct adjacency relationship between the two base stations. Specifically, if base station i and base station j are adjacent, then the element in the i-th row and j-th column of the adjacency matrix is ​​1, otherwise it is 0; the actual traffic data of the base station at the current moment refers to the actual data traffic that the base station is processing at a certain moment. The actual data traffic may include uplink traffic (from user equipment to base station) and downlink traffic (from base station to user equipment), which is not specifically limited here; the remaining traffic capacity of the base station refers to the number of additional users or traffic that the base station can support under the current load.

[0075] In this embodiment, Y t ={y t,1 ,y t,2 ,...,y t,N} is the real traffic data of N base stations at time t, where y t,n represents the actual traffic situation of the nth base station at time t, Y t+1 ={y t+1,1 ,y t+1,2 ,...,y t+1,n}, represents the predicted traffic data of each base station at time t+1, where y t+1,n A represents the predicted traffic situation of the nth base station at time t+1. ij ∈R N×N , represents the adjacency matrix of each base station in the cellular network, where N represents the number of base stations. ij ∈R N×N It represents a load balancing scheduling scheme, and represents the traffic that base station i needs to transfer to base station j in the scheme. This embodiment requires a load balancing decision strategy D, through which the load balancing scheduling scheme is implemented, the user traffic allocation scheme is adjusted, and the base stations to which the user traffic is allocated are determined, so that the load difference of each base station is minimized or the overall network resource utilization is maximized, thereby improving network performance.

[0076] Step 104: Calculate the pheromone between the base station and the adjacent base stations using the actual traffic data, the predicted traffic data and the adjacency matrix.

[0077] In the embodiment of the present invention, the pheromone between the base station and the adjacent base stations is calculated using the actual traffic data, the predicted traffic data and the adjacency matrix. Specifically, the difference between the actual traffic and the predicted traffic is first calculated to determine the preferred direction of path selection. If the load increases in the next time step, the difference value is positive, and the pheromone increment of the transfer path of the traffic outflow from the base station is less than 0. Conversely, the pheromone increment of the transfer path of the traffic outflow from the base station is greater than 0. Secondly, the traffic difference is multiplied by the set maximum increment parameter to calculate the increment of the pheromone. The maximum increment parameter is used to control the adjustment range of the pheromone. When the base station is an expected high-load base station, the corresponding maximum increment parameter will increase, greatly reducing its pheromone concentration. Finally, for each pair of adjacent base stations, the pheromone value of the corresponding path is adjusted according to the pheromone increment.

[0078] It should be noted that pheromone is a mechanism for simulating biological behavior, which is usually used in optimization algorithms, such as Ant Colony Optimization (ACO). In ACO, pheromones are used to simulate the chemical substances left by ants on the path to guide other ants to choose the path. The concentration of pheromones reflects the quality or quality of the path. Among them, the ant colony optimization algorithm is a heuristic optimization algorithm inspired by the behavior of ants in finding food paths. Its core idea is to find the optimal solution by simulating the process of ants releasing pheromones when searching for food. In ant colony optimization, ants search for solutions in the solution space and guide ants to choose paths through the deposition and volatilization of pheromones, thereby achieving a balance between global search and local search and finding the optimal solution. In a cellular network, it is necessary to balance the load between multiple base stations, which means that the load conditions between multiple base stations need to be considered at the same time. The distributed search capability of the ant colony algorithm enables it to effectively handle such multi-variable and multi-objective problems. In this embodiment, pheromones are used as a mechanism to adjust and optimize path selection in the network.

[0079] Specifically, for each pair of adjacent base stations, the pheromone value of the corresponding path is adjusted according to the pheromone increment. For each base station i, its adjacent base station list is obtained, and the adjacent base station list is traversed. For each adjacent base station j, the pheromone value on the path from base station i to base station j is updated. The update rule is: add the increment to the original pheromone value and ensure that the pheromone value is not less than 0. It will not be repeated here.

[0080] Step 105: Use the remaining flow capacity and the adjacency matrix to calculate and obtain the heuristic information between the base station and the adjacent base stations.

[0081] In an embodiment of the present invention, the residual flow capacity and the adjacency matrix are used to calculate the heuristic information between the base station and the adjacent base stations. The heuristic information is calculated based on the residual flow capacity and the adjacency matrix of the current base station. For each base station, the residual flow capacity is multiplied by the adjacency matrix, and the residual capacity of the non-adjacent base stations is set to zero. A heuristic information matrix corresponding to the adjacency matrix can be obtained. The heuristic information is used to measure the pros and cons of each path in the path selection process. The heuristic information is mainly used to determine the probability of selecting a path, thereby affecting the decision of the base station selection for the base station traffic transfer.

[0082] Step 106, calculating the path selection probability of traffic transmission between the base station and the adjacent base stations according to the pheromone and heuristic information, and determining the target transmission path.

[0083] In an embodiment of the present invention, for each base station, the path selection probability of traffic transmission between the base station and the adjacent base stations is calculated based on pheromones and heuristic information, that is, the selection probability of each path connected to the base station is calculated based on the pheromone concentration and the heuristic information. Specifically, the selection probability of the path is calculated based on the heuristic information value of the path combined with the pheromone value of the path. The heuristic information value can be the estimated cost, distance, delay, etc. of the path. The pheromone value reflects the quality or merits of the path.

[0084] Step 107: distribute traffic to base stations on the target transmission path.

[0085] In the embodiment of the present invention, traffic is distributed to the base stations on the target transmission path. For each base station, the selection probability of each path connected to the base station is calculated according to the pheromone concentration and heuristic information, so that the traffic distribution can be optimized. In the load balancing process of each time step, the pheromone concentration will be updated according to the traffic prediction data of the next moment to adapt to the changes in the network, help the network to better cope with traffic fluctuations, and improve the effect of load balancing. The load balancing in this embodiment combines the strategy based on ant colony optimization, and uses the traffic prediction results to realize the intelligent scheduling and optimal allocation of cellular network resources, which can more intelligently schedule network resources, improve the utilization rate of cellular network resources, and provide a solution for the sustainable development of mobile communications.

[0086] The traffic load balancing method provided by the embodiment of the present invention obtains the traffic data of a preset period of the base station, uses a pre-trained network traffic prediction model to perform feature analysis on the traffic data, obtains the spatiotemporal feature representation of the traffic data, obtains the predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation, obtains the adjacency matrix of the base station, the actual traffic data at the current moment and the remaining traffic capacity, uses the actual traffic data, the predicted traffic data and the adjacency matrix to calculate the pheromone between the base station and the adjacent base stations, uses the remaining traffic capacity and the adjacency matrix to calculate the heuristic information between the base station and the adjacent base stations, calculates the path selection probability of the traffic transmission between the base station and the adjacent base stations according to the pheromone and the heuristic information, determines the target transmission path, and distributes traffic to the base stations on the target transmission path. The embodiment of the present invention realizes the fusion of the spatiotemporal characteristics of traffic data through a pre-trained network traffic prediction model, realizes traffic prediction that conforms to the spatiotemporal characteristics of network traffic, improves the accuracy of business traffic prediction in complex and changeable cellular network environments, provides reliable data support for load balancing, calculates pheromones and heuristic information based on traffic prediction results, obtains the changing trends of current and future loads, dynamically adjusts the load distribution between base stations, and more flexibly allocates network resources to base stations with expected high loads to cope with future traffic fluctuations. Through the prediction and monitoring of network traffic, high performance is ensured under high traffic loads, and load balancing is used to avoid network port overload, thereby ensuring the stability and efficiency of the network.

[0087] Further, see Figure 2 , showing Figure 1 A flow chart of step 102 in a traffic load balancing method is provided. The method is basically the same as the traffic load balancing method provided in the first embodiment of the present invention. Step 102 may include:

[0088] Step 1021, use the pre-trained network traffic prediction model to perform feature analysis on the traffic data to obtain the time series feature representation and spatial feature representation of the traffic data.

[0089] Step 1022, the temporal feature representation and the spatial feature representation are integrated to obtain the spatiotemporal feature representation of the traffic data.

[0090] Step 1023, mapping the spatiotemporal feature representation to obtain the predicted traffic data of the base station at the next moment.

[0091] It should be noted that in the embodiment of the present invention, the pre-trained network traffic prediction model is an adaptive hybrid spatiotemporal graph neural network model, and the adaptive hybrid spatiotemporal learning module in the adaptive hybrid spatiotemporal graph neural network model is used to process periodic traffic data, including the most recent, daily and weekly base station traffic. The time convolution module TCM uses a gated time convolutional network or a graph convolutional neural network (Graph Convolutional Network, GCN) to capture the complex relationship and periodicity of time series data. The output of each time convolution TCN module is a time series feature representation of the traffic data. Where D represents the depth or dimension of the feature.

[0092] Specifically, the adaptive hybrid graph learning module AHGLM captures the spatial relationship between cellular towers. It mainly uses static adaptive graph learning (SAGL) to capture relatively stable spatial relationships, and dynamic adaptive graph learning (DAGL) to capture the dynamic influence between neighbors and process dynamic spatial features. The air control gate fusion is used to fuse the outputs of SAGL and DGL to adaptively control the flow of static and dynamic spatial dependencies. The input of this module is the spatial relationship between cellular towers, and the output is the learned spatial feature representation. The data dimensions are all T×N×D.

[0093] In this embodiment, the spatiotemporal adaptation module STAM is used to capture the spatiotemporal adaptation trend at the node level. The input is the output of TCM and AHGLM, which are the temporal feature representation and spatial feature representation respectively, and the output is the spatiotemporal feature representation obtained after STAM processing that integrates spatiotemporal information, and the data dimensions are all T×N×D.

[0094] This embodiment uses the output layer to perform skip connections on the outputs of each adaptive hybrid graph learning module, directly connecting them to the output layer. As multiple adaptive hybrid graph learning modules are stacked, the outputs of all adaptive hybrid graph learning modules are fused using summation. The result is expressed as H out ∈R T×N×D Finally, the output layer generates the final multi-step forecast

[0095] Specifically, the formula for the predicted traffic data of the base station at the next moment output by the output layer is as follows:

[0096]

[0097] in, and is a learnable parameter, It is the final output of the entire AHSTGNN.

[0098] In some embodiments, the mean absolute error between the predicted value and the true value can be used as a loss function in the multi-step base station traffic prediction task, and it can be minimized through back propagation to improve the accuracy of the predicted traffic data output by the model, where the formula of the loss function is:

[0099]

[0100] Among them, θ represents all learnable parameters of the model, represents the model's predicted value for all base stations at time step i, Y i is the actual value.

[0101] The service flow prediction method based on the adaptive hybrid spatiotemporal graph neural network model in the embodiment of the present invention makes the model more sensitive to time changes, improves the accuracy of service flow prediction in complex and changeable cellular network environments, and provides reliable data support for load balancing.

[0102] Specifically, step 1021 uses a pre-trained network traffic prediction model to perform feature analysis on the traffic data to obtain a temporal feature representation and a spatial feature representation of the traffic data, which may specifically include the following steps:

[0103] First, the flow data is convolved with a preset time embedding matrix to obtain a time series feature representation of the flow data;

[0104] Secondly, static adaptive graph learning and dynamic adaptive graph learning are performed on the traffic data to obtain static spatial relationships and dynamic spatial relationships;

[0105] Secondly, the static spatial relationship and the dynamic spatial relationship are fused to obtain a spatial feature representation of the traffic data.

[0106] It should be noted that in the above steps, the flow data is convolved with the preset time embedding matrix to obtain the time series feature representation of the flow data, and the static spatial relationship and dynamic spatial relationship are obtained by static adaptive graph learning and dynamic adaptive graph learning respectively on the flow data, and then the static spatial relationship and the dynamic spatial relationship are fused to obtain the spatial feature representation of the flow data. Among them, the temporal embedding matrix is ​​a matrix used to represent the time information in time series data. The time embedding is generated using sine and cosine functions to dynamically represent the time information. It is usually used in deep learning models, especially when processing time-dependent data. The time embedding matrix can encode the time information into a high-dimensional vector, so that the model can better capture the patterns and trends in the time series.

[0107] Specifically, a time embedding matrix is ​​added to the model initialization part and updated in each iteration to represent the relationship in the time dimension. In the forward propagation process of the SAGL module in AHGLM, the time embedding matrix is ​​applied to the input data and combined with the graph convolutional neural network GCN calculation. In this embodiment, the GCN calculation formula is modified to include the time embedding matrix, so that the model can dynamically consider the time information. The modified GCN calculation formula is:

[0108]

[0109] Where D is the degree diagonal matrix determined by the adjacency matrix A, X is the input, θ is the neural network parameter, and t emb is the time embedding matrix.

[0110] The embodiment of the present invention introduces a time embedding matrix into traffic data, so that the model can dynamically learn and update the relationship in the time dimension, thereby adapting to the time-related patterns in the task. The weight of the time embedding matrix is ​​trained as part of the model, making the model more sensitive to time changes and improving the accuracy of traffic prediction.

[0111] Further, see Figure 3 , showing Figure 1 A flow chart of step 104 in a traffic load balancing method is provided. The method is basically the same as the traffic load balancing method provided in the first embodiment of the present invention. Step 104 may include:

[0112] Step 1041, calculate the flow difference between the actual flow data and the predicted flow data, and determine the adjacent base stations of the base station according to the adjacency matrix of the base station.

[0113] Step 1042: Calculate the pheromone increment of the adjacent base station using the flow difference and the preset increment parameter.

[0114] Step 1043: Use the pheromone increment to adjust the current pheromone of the adjacent base station to obtain the pheromone between the base station and the adjacent base station.

[0115] It should be noted that, in the embodiment of the present invention, the entire network topology of the cellular network, the base station load at the current moment and the base station load predicted at the next moment are considered, and the adjacency matrix A of the cellular network is converted into ij ∈R N×N , the initial traffic data Y of the base station at time t t ={y t,1 ,y t,2 ,...,y t,N} and the predicted traffic data Y of the base station at time t+1 t+1 ={y t+1,1 ,y t+1,2,...,y t+1,N} as the input of the model.

[0116] Specifically, the adjacent base stations of the base station are determined according to the adjacency matrix of the base station, and the actual flow data and predicted flow data of the base station are calculated to obtain the flow difference of the base station. The flow difference is multiplied by the set maximum increment parameter to calculate the increment of pheromone. When the base station is an expected high-load base station, the corresponding maximum increment parameter will increase, greatly reducing its pheromone concentration. The current pheromone of the adjacent base station is adjusted using the pheromone increment to obtain the pheromone between the base station and the adjacent base station. Specifically, the flow difference is used to determine the preferred direction of path selection. The flow difference can be expressed as Y t -Y t+1 If the load increases in the next time step, the value is positive, and the pheromone increment of the transfer path of the outflow traffic from the base station is less than 0. Otherwise, the pheromone increment of the transfer path of the outflow traffic from the base station is greater than 0. For each pair of adjacent base stations, the pheromone value of the corresponding path is adjusted according to the pheromone increment.

[0117] For example, for each pair of adjacent base stations, adjust the pheromone value of the corresponding path according to the pheromone increment. The specific steps are: for each base station, obtain its adjacent base station list, traverse the adjacent base station list, and for each adjacent base station, update the pheromone value on the path. The update rule is: add the increment to the original pheromone value and ensure that the pheromone value is not less than 0. The expression is:

[0118] m ij =m ij +d i

[0119] m ij =max(0,m ij )

[0120] Among them, m ij is the pheromone value on the path from base station i to j, d i is the pheromone increment of base station i.

[0121] The embodiment of the present invention is based on the ant colony optimization algorithm, uses the difference between the actual flow and the predicted flow to calculate the pheromone increment, and dynamically updates the pheromone to ensure that the pheromone value reflects the changing trend of the current and future loads.

[0122] Specifically, step 105 uses the remaining traffic capacity and the adjacency matrix to calculate the heuristic information between the base station and the adjacent base stations, which may specifically include: calculating the remaining traffic capacity of the base station and the adjacency matrix, and adjusting the non-adjacent matrix in the adjacency matrix to zero, to obtain the heuristic information between the base station and the adjacent base stations.

[0123] In this embodiment, the heuristic information is calculated based on the remaining traffic capacity of the current base station and the adjacency matrix. For each base station, the remaining capacity is multiplied by the adjacency matrix, and the remaining capacity of the non-adjacent base stations is set to zero, so that a heuristic information matrix corresponding to the adjacency matrix can be obtained. The heuristic information expression is:

[0124] h ij =A ij ×L j

[0125] Among them, h ij is the heuristic information, indicating the expected value from node i to node j, A ij is the connection relationship between node i and node j in the adjacency matrix. If there is a connection, it is 1, otherwise it is 0. L j Indicates the remaining traffic capacity of the base station.

[0126] Further, see Figure 4 , showing Figure 1 A flow chart of step 106 in a traffic load balancing method is provided. The method is basically the same as the traffic load balancing method provided in the first embodiment of the present invention. Step 106 may include:

[0127] Step 1061, using pheromone and heuristic information to calculate the selection probability of each path connecting the base station to the base station, and obtain the path selection probability of traffic transmission between the base station and the adjacent base station.

[0128] Step 1062: Filter out a target transmission path from each path connected to the base station according to the path selection probability.

[0129] It should be noted that, in the embodiment of the present invention, pheromones and heuristic information are used to calculate the selection probability of each path connecting the base station and the base station, and the path selection probability of the traffic transmission between the base station and the adjacent base station is obtained. The path selection probability is the probability value of selecting the path. According to the path selection probability, the target transmission path is screened out from each path connected to the base station. The target transmission path is a path randomly selected according to the probability as the next node for transmitting traffic.

[0130] Specifically, for each base station, the selection probability of each path connected to the base station is calculated based on the pheromone concentration and heuristic information. For the load balancing at each time step, the pheromone concentration matrix will change according to the traffic prediction data at the next moment. The selection probability formula for each path is:

[0131]

[0132] Among them, α is a parameter that controls the influence of pheromone concentration on path selection, β is a parameter that controls the influence of heuristic information on path selection, v is a vector representing the nodes that have been visited, and m ij is the pheromone value on the path from base station i to j, h ij is the heuristic information, which indicates the expected value from node i to node j.

[0133] The embodiment of the present invention selects a target transmission path from each path connected to the base station according to the path selection probability, dynamically adjusts the load distribution between the base stations, and more flexibly allocates network resources to base stations with expected high loads to cope with future traffic fluctuations.

[0134] Reference Figure 5 , shows a flow chart of steps of another traffic load balancing method provided by an embodiment of the present invention, which is basically the same as the traffic load balancing method provided by the first embodiment of the present invention, except that the method may further include:

[0135] Step 101, obtaining the flow data of a preset period of a base station.

[0136] Step 102, using a pre-trained network traffic prediction model to perform feature analysis on the traffic data, obtain a spatiotemporal feature representation of the traffic data, and obtain the predicted traffic data of the base station at the next moment based on the spatiotemporal feature representation.

[0137] Step 103, obtaining the adjacency matrix of the base station, the actual flow data at the current moment, and the remaining flow capacity.

[0138] Step 104: Calculate the pheromone between the base station and the adjacent base stations using the actual traffic data, the predicted traffic data and the adjacency matrix.

[0139] Step 105: Use the remaining flow capacity and the adjacency matrix to calculate and obtain the heuristic information between the base station and the adjacent base stations.

[0140] Step 106, calculating the path selection probability of traffic transmission between the base station and the adjacent base stations according to the pheromone and heuristic information, and determining the target transmission path.

[0141] Step 107: distribute traffic to base stations on the target transmission path.

[0142] The above steps 101 to 107 are described in the foregoing description and will not be repeated here.

[0143] Step 108: Store the target transmission path into the initial solution space matrix.

[0144] Step 109: monitor the traffic allocation of the base station and update the remaining traffic capacity of the target transmission path.

[0145] Step 110: Calculate the resource utilization rate and the variance of the resource utilization rate of the target transmission path using the remaining traffic capacity.

[0146] Step 111: determine the target transmission path whose variance is less than a preset threshold as the optimal transmission path.

[0147] It should be noted that, in the above steps 108 to 111, an empty solution space matrix ant_solutions is initialized, the solution space matrix is ​​used to store each path selection scheme, the target transmission path is stored in the initial solution space matrix, the traffic distribution of the base station is monitored, the remaining traffic capacity of the target transmission path is updated, and the remaining traffic capacity is calculated, that is, the maximum capacity is subtracted from the initial traffic, the remaining traffic capacity represents the available resources of each node, and the remaining traffic capacity is used to calculate the resource utilization of the target transmission path and the variance of the resource utilization.

[0148] Specifically, for a resource reallocation, a starting node is selected, and path selection is performed without visiting all nodes: first, the path selection probability is used to select the next node, and resources are allocated to the selected node. If the next node cannot be selected (all paths have been visited or are not feasible), the current resource reallocation is terminated, and the solution space matrix and the remaining traffic capacity are updated according to the path selection result. Then, an iteration of the resource reallocation operation is performed until the number of iterations reaches the set parameter. After all iterations are completed, the optimal value of all target transmission paths is calculated, and the resource utilization variance is used as a load balancing evaluation indicator to measure the degree of balance of resource utilization in the network. The algorithm is designed with the goal of minimizing the resource utilization variance, and the target transmission path with a variance less than a preset threshold is determined as the optimal transmission path. At this time, the network traffic transfer strategy can be obtained by the solution space matrix ant_solutions.

[0149] In this embodiment, the cellular network load balancing algorithm is designed with reference to the ant colony optimization algorithm. Based on the simulation of ant foraging behavior, traffic scheduling between base stations is implemented to optimize network resource utilization. The resource utilization rate and variance of the target transmission path of the base station are expressed as:

[0150]

[0151] Among them, x i is the actual resource allocation of the ith base station, c i is the total capacity of the ith base station, and N is the number of base stations.

[0152] Compared with the prior art, the implementation mode of the present invention, on the basis of achieving the beneficial effects brought by the first implementation mode, uses the resource utilization variance as a load balancing evaluation indicator to measure the degree of balance of resource utilization in the network, designs an algorithm with the goal of minimizing the resource utilization variance, and determines the target transmission path with a variance less than a preset threshold as the optimal transmission path, thereby achieving load balancing optimization to cope with future traffic fluctuations.

[0153] Reference Figure 7 , shows a schematic diagram of the structure of a traffic load balancing device provided by an embodiment of the present invention, the device comprising:

[0154] The first acquisition module 201 is used to acquire the flow data of the base station in a preset period;

[0155] The traffic prediction module 202 is used to perform feature analysis on the traffic data using a pre-trained network traffic prediction model to obtain a spatiotemporal feature representation of the traffic data, and obtain the predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation;

[0156] The second acquisition module 203 is used to acquire the adjacency matrix of the base station, the actual flow data at the current moment, and the remaining flow capacity;

[0157] A pheromone calculation module 204, configured to calculate pheromones between a base station and adjacent base stations using the actual traffic data, the predicted traffic data and the adjacency matrix;

[0158] A heuristic information calculation module 205, configured to calculate heuristic information between a base station and adjacent base stations using the remaining traffic capacity and the adjacency matrix;

[0159] A path determination module 206, configured to calculate a path selection probability for traffic transmission between a base station and an adjacent base station based on the pheromone and the heuristic information, and determine a target transmission path;

[0160] The traffic allocation module 207 is used to allocate traffic to the base stations on the target transmission path.

[0161] Furthermore, the traffic prediction module 202 includes:

[0162] A feature analysis submodule, used to perform feature analysis on the traffic data using a pre-trained network traffic prediction model to obtain a temporal feature representation and a spatial feature representation of the traffic data;

[0163] A feature fusion submodule, used for fusing the temporal feature representation and the spatial feature representation to obtain the spatiotemporal feature representation of the traffic data;

[0164] The mapping submodule is used to map the spatiotemporal feature representation to obtain the predicted traffic data of the base station at the next moment.

[0165] Furthermore, the feature analysis submodule includes:

[0166] A convolution processing unit, used for performing convolution processing on the flow data and a preset time embedding matrix to obtain a time series feature representation of the flow data;

[0167] An adaptive learning unit, used for performing static adaptive graph learning and dynamic adaptive graph learning on the traffic data to obtain static spatial relationships and dynamic spatial relationships;

[0168] A fusion unit is used to fuse the static spatial relationship and the dynamic spatial relationship to obtain a spatial feature representation of the traffic data.

[0169] Furthermore, the pheromone calculation module 204 includes:

[0170] A first calculation submodule, configured to calculate a flow difference between the actual flow data and the predicted flow data, and determine an adjacent base station of the base station according to an adjacency matrix of the base station;

[0171] A second calculation submodule, configured to calculate the pheromone increment of the adjacent base station by using the flow difference and a preset increment parameter;

[0172] The adjustment submodule is used to adjust the current pheromone of the adjacent base station by using the pheromone increment to obtain the pheromone between the base station and the adjacent base station.

[0173] Furthermore, the heuristic information calculation module 205 includes:

[0174] The third calculation submodule is used to calculate the remaining flow capacity of the base station and the adjacency matrix, and adjust the non-adjacent matrix in the adjacency matrix to zero, so as to obtain the heuristic information between the base station and the adjacent base stations.

[0175] Furthermore, the path determination module 206 includes:

[0176] A fourth calculation submodule, configured to calculate the selection probability of each path connecting a base station to another base station by using the pheromone and the heuristic information, and obtain the path selection probability of traffic transmission between a base station and an adjacent base station;

[0177] The screening submodule is used to screen out a target transmission path from each path connected to the base station according to the path selection probability.

[0178] Furthermore, the device also includes:

[0179] A storage path module, used for storing the target transmission path into an initial solution space matrix;

[0180] An updating capacity module, used to monitor the flow distribution of the base station and update the remaining flow capacity of the target transmission path;

[0181] A resource utilization calculation module, used to calculate the resource utilization of the target transmission path and the variance of the resource utilization using the remaining traffic capacity;

[0182] The optimal path determination module is used to determine the target transmission path whose variance is less than a preset threshold as the optimal transmission path.

[0183] The traffic load balancing device provided by the embodiment of the present invention obtains the traffic data of a preset period of the base station, uses a pre-trained network traffic prediction model to perform feature analysis on the traffic data, obtains the spatiotemporal feature representation of the traffic data, obtains the predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation, obtains the adjacency matrix of the base station, the actual traffic data at the current moment and the remaining traffic capacity, uses the actual traffic data, the predicted traffic data and the adjacency matrix to calculate the pheromone between the base station and the adjacent base stations, uses the remaining traffic capacity and the adjacency matrix to calculate the heuristic information between the base station and the adjacent base stations, calculates the path selection probability of the traffic transmission between the base station and the adjacent base stations according to the pheromone and the heuristic information, determines the target transmission path, and distributes traffic to the base stations on the target transmission path. The embodiment of the present invention realizes the fusion of the spatiotemporal characteristics of traffic data through a pre-trained network traffic prediction model, realizes traffic prediction that conforms to the spatiotemporal characteristics of network traffic, improves the accuracy of business traffic prediction in complex and changeable cellular network environments, provides reliable data support for load balancing, calculates pheromones and heuristic information based on traffic prediction results, obtains the changing trends of current and future loads, dynamically adjusts the load distribution between base stations, and more flexibly allocates network resources to base stations with expected high loads to cope with future traffic fluctuations. Through the prediction and monitoring of network traffic, high performance is ensured under high traffic loads, and load balancing is used to avoid network port overload, thereby ensuring the stability and efficiency of the network.

[0184] Reference Figure 8 , an embodiment of the present invention further provides an electronic device, such as Figure 8 As shown, it includes a processor 301, a communication interface 302, a memory 303 and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304.

[0185] A processor 301, a memory 303 for storing processor executable instructions;

[0186] The processor 301 is configured to execute the instructions to implement the following traffic load balancing method:

[0187] Obtain the traffic data of the base station at a preset period;

[0188] Using a pre-trained network traffic prediction model to perform feature analysis on the traffic data, obtain a spatiotemporal feature representation of the traffic data, and obtain predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation;

[0189] Obtain the adjacency matrix of the base station, the actual traffic data at the current moment, and the remaining traffic capacity;

[0190] The actual traffic data, the predicted traffic data and the adjacency matrix are used to calculate the pheromone between the base station and the adjacent base stations;

[0191] The heuristic information between the base station and the adjacent base stations is calculated using the remaining flow capacity and the adjacency matrix;

[0192] Calculating the path selection probability of traffic transmission between the base station and the adjacent base stations according to the pheromone and the heuristic information, and determining the target transmission path;

[0193] Traffic is distributed to base stations on the target transmission path.

[0194] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0195] The communication interface is used for communication between the above terminal and other devices.

[0196] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0197] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0198] In another embodiment provided by the present invention, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, the traffic load balancing method described in any of the above embodiments is implemented.

[0199] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0200] 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 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 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 device including the elements.

[0201] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0202] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A traffic load balancing method, characterized in that: The method comprises: Obtain the traffic data of the base station at a preset period; Using a pre-trained network traffic prediction model to perform feature analysis on the traffic data, obtain a spatiotemporal feature representation of the traffic data, and obtain predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation; Obtain the adjacency matrix of the base station, the actual traffic data at the current moment, and the remaining traffic capacity; The actual traffic data, the predicted traffic data and the adjacency matrix are used to calculate the pheromone between the base station and the adjacent base stations; The heuristic information between the base station and the adjacent base stations is calculated using the remaining flow capacity and the adjacency matrix; Calculating the path selection probability of traffic transmission between the base station and the adjacent base stations according to the pheromone and the heuristic information, and determining the target transmission path; Traffic is distributed to base stations on the target transmission path.

2. The method according to claim 1, characterized in that: The method of using a pre-trained network traffic prediction model to perform feature analysis on the traffic data to obtain a spatiotemporal feature representation of the traffic data, and obtaining predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation, includes: Using a pre-trained network traffic prediction model to perform feature analysis on the traffic data to obtain a temporal feature representation and a spatial feature representation of the traffic data; The temporal feature representation and the spatial feature representation are integrated to obtain the spatiotemporal feature representation of the traffic data; The spatiotemporal feature representation is mapped to obtain the predicted traffic data of the base station at the next moment.

3. The method according to claim 2, characterized in that: The method of using a pre-trained network traffic prediction model to perform feature analysis on the traffic data to obtain a temporal feature representation and a spatial feature representation of the traffic data includes: Convolution processing is performed on the flow data and a preset time embedding matrix to obtain a time series feature representation of the flow data; Performing static adaptive graph learning and dynamic adaptive graph learning on the traffic data to obtain static spatial relationships and dynamic spatial relationships; The static spatial relationship and the dynamic spatial relationship are fused to obtain a spatial feature representation of the traffic data.

4. The method according to claim 1, characterized in that: The step of using the actual traffic data, the predicted traffic data, and the adjacency matrix to calculate pheromones between a base station and adjacent base stations includes: Calculating the flow difference between the actual flow data and the predicted flow data, and determining the adjacent base stations of the base station according to the adjacency matrix of the base station; Calculating the pheromone increment of the adjacent base station by using the flow difference and a preset increment parameter; The pheromone increment is used to adjust the current pheromone of the adjacent base station to obtain the pheromone between the base station and the adjacent base station.

5. The method according to claim 1, characterized in that: The using the remaining flow capacity and the adjacency matrix to calculate the heuristic information between the base station and the adjacent base stations includes: The remaining flow capacity of the base station is calculated with the adjacency matrix, and non-adjacent matrices in the adjacency matrix are adjusted to zero to obtain heuristic information between the base station and adjacent base stations.

6. The method according to claim 1, characterized in that: The calculating the path selection probability of traffic transmission between the base station and the adjacent base station according to the pheromone and the heuristic information, and determining the target transmission path, includes: The pheromone and the heuristic information are used to calculate the selection probability of each path connecting the base station to the base station, so as to obtain the path selection probability of traffic transmission between the base station and the adjacent base station; A target transmission path is selected from each path connected to the base station according to the path selection probability.

7. The method according to claim 1, characterized in that: After allocating traffic to the base stations on the target transmission path, the method further includes: Storing the target transmission path into an initial solution space matrix; Monitoring the flow distribution of the base station and updating the remaining flow capacity of the target transmission path; Calculating the resource utilization rate of the target transmission path and the variance of the resource utilization rate using the remaining traffic capacity; The target transmission path whose variance is less than a preset threshold is determined as the optimal transmission path.

8. A traffic load balancing device, characterized in that: The device comprises: A first acquisition module, used to acquire flow data of a preset period of a base station; A traffic prediction module, used to perform feature analysis on the traffic data using a pre-trained network traffic prediction model to obtain a spatiotemporal feature representation of the traffic data, and obtain predicted traffic data of the base station at the next moment according to the spatiotemporal feature representation; The second acquisition module is used to acquire the adjacency matrix of the base station, the actual flow data at the current moment, and the remaining flow capacity; A pheromone calculation module, used to calculate the pheromone between the base station and the adjacent base stations using the actual traffic data, the predicted traffic data and the adjacency matrix; A heuristic information calculation module, used to calculate the heuristic information between the base station and the adjacent base stations using the remaining flow capacity and the adjacency matrix; A path determination module, used to calculate the path selection probability of traffic transmission between the base station and the adjacent base stations according to the pheromone and the heuristic information, and determine the target transmission path; The traffic distribution module is used to distribute traffic to the base stations on the target transmission path.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; The memory is used to store computer programs; The processor is used to implement the method according to any one of claims 1 to 7 when executing the program stored in the memory.

10. A computer-readable storage medium having instructions stored thereon, which, when executed by one or more processors, cause the processors to perform the method according to any one of claims 1 to 7.

Images