Network topology updating method, device and communication equipment

Abstract

The present application provides a network topology updating method, apparatus, and communication device. The method includes: iteratively updating the adjacency matrix of a first network topology based on a genetic algorithm to obtain the adjacency matrices of multiple second network topologies; using the network traffic time series of each node in the first network topology and the adjacency matrices of multiple second network topologies as inputs to a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction to obtain the predicted network traffic time series of each node in each second network topology; obtaining the load balancing coefficient of each second network topology based on the predicted network traffic time series of each node in each second network topology; and updating the first network topology using a target network topology, where the target network topology is the network topology with the smallest load balancing coefficient among the multiple second network topologies. The present application can improve the optimization effect of the network topology.

Description

Technical Field

[0001] The present application relates to the field of network optimization, and in particular to a network topology updating method, apparatus, and communication equipment. Background Art

[0002] The exponential growth of internet traffic is placing higher demands on network carrying capacity. Current communication network planning is increasingly unable to meet the growing demand for internet services, especially during peak hours, when network carrying capacity is low. This necessitates optimization of network topology. Existing technologies use equal-cost multi-routing algorithms to split data streams and transmit them along multiple equal-cost paths. However, these algorithms only consider static network topology information, resulting in poor optimization results. Summary of the Invention

[0003] The present application provides a network topology updating method, apparatus, and communication equipment to solve the problem of poor optimization effect of network topology.

[0004] In a first aspect, an embodiment of the present application provides a network topology updating method, including:

[0005] Obtaining an adjacency matrix of a first network topology and a network traffic time series of each node in the first network topology;

[0006] Iteratively updating the adjacency matrix of the first network topology based on a genetic algorithm to obtain multiple adjacency matrices of the second network topology;

[0007] Using the network traffic time series of each node in the first network topology and the adjacency matrices of the plurality of second network topologies as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction, so as to obtain a predicted network traffic time series for each node in each of the second network topologies;

[0008] Obtaining a load balancing coefficient of each second network topology based on a predicted network traffic time series of each node in each second network topology;

[0009] The first network topology is updated using a target network topology, where the target network topology is a network topology with the smallest load balancing coefficient among the plurality of second network topologies.

[0010] In a second aspect, an embodiment of the present application further provides a network topology updating device, comprising:

[0011] A first acquisition module is configured to acquire an adjacency matrix of a first network topology and a network traffic time series of each node in the first network topology;

[0012] A first updating module, configured to iteratively update the adjacency matrix of the first network topology based on a genetic algorithm to obtain adjacency matrices of multiple second network topologies;

[0013] a prediction module, configured to use the network traffic time series of each node in the first network topology and the adjacency matrices of the plurality of second network topologies as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction, so as to obtain a predicted network traffic time series for each node in each of the second network topologies;

[0014] A second acquisition module, configured to acquire a load balancing coefficient of each second network topology based on a predicted network traffic time series of each node in each second network topology;

[0015] The second updating module is configured to update the first network topology using a target network topology, where the target network topology is a network topology with the smallest load balancing coefficient among the plurality of second network topologies.

[0016] In the third aspect, an embodiment of the present application also provides a communication device, including: a transceiver, a memory, a processor, and a program stored on the memory and runnable on the processor; the processor is used to read the program in the memory to implement the steps in the method described in the first aspect of the embodiment of the present application.

[0017] In a fourth aspect, an embodiment of the present application further provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the method described in the first aspect of the embodiment of the present application.

[0018] In an embodiment of the present application, the adjacency matrix of the first network topology is iteratively updated based on a genetic algorithm to obtain the adjacency matrices of multiple second network topologies, and a network topology can be selected from the multiple second network topologies to update the first network topology; the network traffic time series of each node in the first network topology and the adjacency matrix of the multiple second network topologies are used as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction to obtain the predicted network traffic time series of each node in each second network topology; based on the predicted network traffic time series of each node in each second network topology, the load balancing coefficient of each second network topology is obtained; the spatiotemporal traffic prediction model can perform predictions based on spatial and temporal relationships, and obtain the target network topology with the smallest load balancing coefficient among the multiple second network topologies, and by using the target network topology to update the first network topology, the load balancing degree of the updated network topology is improved, thereby improving the optimization effect of the first network topology. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the embodiments or descriptions of the prior art. 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 any creative work.

[0020] Figure 1 This is a flow chart of a network topology updating method provided in an embodiment of the present application;

[0021] Figure 2 This is a flow chart of a network topology optimization method provided by an embodiment of the present application;

[0022] Figure 3 This is a schematic diagram of the structure of a network topology updating device provided in an embodiment of the present application;

[0023] Figure 4 It is a structural diagram of a communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] The terms "first", "second" etc. in the embodiments of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. In addition, the terms "comprise" and "have" and any deformation thereof are intended to cover non-exclusive inclusions, such as, the process, method, system, product or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or that are intrinsic to these processes, methods, products or equipment. In addition, "and / or" is used in the present application to represent at least one of connected objects, such as A and / or B and / or C, and represents comprising independent A, independent B, independent C, and A and B all exist, B and C all exist, A and C all exist, and 7 situations that A, B and C all exist.

[0026] See also Figure 1 , Figure 1 This is a flow chart of a network topology updating method provided by an embodiment of the present application. Figure 1 As shown, the following steps are included:

[0027] Step 101: Obtain an adjacency matrix of a first network topology and a network traffic time series of each node in the first network topology.

[0028] It can be understood that the above-mentioned first network topology is the current network topology, and the adjacency matrix obtained by the above-mentioned first network topology can characterize the adjacent relationship between each node in the above-mentioned first network topology, that is, the spatial relationship of the network topology can be represented by the above-mentioned adjacency matrix.

[0029] Among them, the above-mentioned network traffic time series can be a time series composed of the network traffic of each of the above-mentioned nodes at multiple historical times. For example, the network traffic of each node before the current time 1 minute, 2 minutes,..., 10 minutes can be obtained as the above-mentioned network traffic time series, and the network traffic time series can be used to predict the network traffic at a certain time in the future.

[0030] Step 102: Iteratively update the adjacency matrix of the first network topology based on a genetic algorithm to obtain multiple adjacency matrices of second network topologies.

[0031] It will be understood that the genetic algorithm simulates a natural evolutionary process to search for an optimal solution. During the iterative update of the adjacency matrix of the first network topology, each update yields an updated adjacency matrix, and each updated adjacency matrix corresponds to an updated second network topology. The adjacency matrices of the multiple second network topologies obtained through iterative updates based on the genetic algorithm are, thus, candidate network topologies obtained based on the genetic algorithm that can be used to update the first network topology.

[0032] Step 103: Use the network traffic time series of each node in the first network topology and the adjacency matrix of the multiple second network topologies as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction to obtain a predicted network traffic time series for each node in each of the second network topologies.

[0033] Among them, the above-mentioned spatiotemporal traffic prediction model can be pre-trained. During the training process, multiple training samples can be used to train the above-mentioned spatiotemporal traffic prediction model. It can be understood that the above-mentioned multiple training samples can also include the adjacency matrix of the above-mentioned first network topology and the historical network traffic time series of each node in the above-mentioned first network topology.

[0034] Optionally, the spatiotemporal traffic prediction model includes a graph neural network model and a gated recurrent unit model;

[0035] The spatiotemporal flow prediction model is trained in the following way:

[0036] The graph neural network model and the gated recurrent unit model are trained using training samples of the adjacency matrix of the network topology and the network traffic time series, where the output of the graph neural network model serves as the input of the gated recurrent unit model;

[0037] Obtain a loss value of a prediction result output by the spatiotemporal traffic prediction model, and update parameters of the spatiotemporal traffic prediction model based on the loss value until the spatiotemporal traffic prediction model converges.

[0038] It can be understood that the above-mentioned spatiotemporal traffic prediction model includes a graph neural network model and a gated recurrent unit model. The input of the above-mentioned spatiotemporal traffic prediction model is the input of the above-mentioned graph neural network model, that is, the graph neural network model is trained using the adjacency matrix of the network topology and the training samples of the network traffic time series, and the output of the above-mentioned graph neural network model is used as the input of the above-mentioned gated recurrent unit model, thereby obtaining the predicted network traffic time series of each node in the network topology. The above-mentioned gated recurrent unit can also better capture the dependency relationship with a large time step distance in the time series, thereby improving the prediction accuracy of the above-mentioned spatiotemporal traffic prediction model. In addition, the loss value can be calculated based on the prediction result output by the above-mentioned spatiotemporal traffic prediction model and the actual traffic in the training sample, and the loss value can be back-propagated to update the parameters of the above-mentioned spatiotemporal traffic prediction model until the above-mentioned spatiotemporal traffic prediction model converges.

[0039] Step 104: Obtain a load balancing coefficient of each second network topology based on the predicted network traffic time series of each node in each second network topology.

[0040] Among them, the above-mentioned load balancing coefficient can be used to represent the load balancing degree of each of the above-mentioned second network topologies. The smaller the above-mentioned load balancing coefficient, the more balanced the load of the corresponding second network topology. The above-mentioned first network topology is updated based on the above-mentioned load balancing coefficient, thereby maximizing the utilization of each node in the above-mentioned second network topology and improving the optimization effect of the above-mentioned first network topology.

[0041] Step 105: Use a target network topology to update the first network topology, where the target network topology is a network topology with the smallest load balancing coefficient among the plurality of second network topologies.

[0042] It can be understood that the above-mentioned target network topology is the network topology with the smallest load balancing coefficient among multiple second network topologies. Therefore, by using the above-mentioned target network topology to update the above-mentioned first network topology, the above-mentioned first network topology can be updated according to the above-mentioned predicted network traffic time series and the above-mentioned load balancing coefficient. The above-mentioned target network topology is used to update the first network topology, so that the update optimization of the network topology takes into account both the predicted network traffic time series and the load balancing degree of the updated network topology.

[0043] Optionally, the updating of the first network topology using the target network topology in step 105 may specifically include the following steps:

[0044] Determine a target network topology having a minimum load balancing coefficient among the plurality of second network topologies;

[0045] Obtaining an adjacency matrix of the target network topology;

[0046] Acquire the target network topology based on the adjacency matrix of the target network topology;

[0047] The first network topology is updated using the target network topology.

[0048] It can be understood that the above-mentioned multiple second network topologies are network topologies that can update the above-mentioned first network topology. The nodes of the above-mentioned multiple second network topologies are consistent with the nodes in the above-mentioned first network topology, but the connection relationship between the nodes is different. By determining the target network topology with the smallest load balancing coefficient among the above-mentioned multiple second network topologies, the load of the network topology obtained after the above-mentioned first network topology is updated by the above-mentioned target network topology can be more balanced.

[0049] In an embodiment of the present application, the adjacency matrix of the first network topology is iteratively updated based on a genetic algorithm to obtain the adjacency matrices of multiple second network topologies, and the multiple second network topologies are candidate network topologies for updating the first network topology; the network traffic time series of each node in the first network topology and the adjacency matrices of the multiple second network topologies are used as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction to obtain the predicted network traffic time series of each node in each of the second network topologies; based on the predicted network traffic time series of each node in each of the second network topologies, the load balancing coefficient of each of the second network topologies is obtained; the spatiotemporal traffic prediction model can perform predictions based on spatial and temporal relationships, and obtain the target network topology with the smallest load balancing coefficient among the multiple second network topologies, and by using the target network topology to update the first network topology, the load balancing degree of the updated network topology is improved, thereby improving the optimization effect of the first network topology.

[0050] In addition, the target network topology is the network topology with the smallest load balancing coefficient among multiple second network topologies. The load balancing coefficient is obtained based on the predicted network traffic time series of each node in each second network topology. The global load balancing degree can be evaluated, and the network resource information is fully considered. The optimal topology result that meets the global load balancing conditions can be obtained, thereby improving the network stability of the topology optimization result.

[0051] Optionally, the iterative updating of the adjacency matrix of the first network topology based on the genetic algorithm in step 102 to obtain the adjacency matrices of the plurality of second network topologies includes:

[0052] Obtaining a first population corresponding to the first network topology;

[0053] Iteratively updating the first population based on a genetic algorithm until a preset number of iterations is reached to obtain a second population obtained by each update;

[0054] Acquire adjacency matrices of a plurality of second network topologies corresponding to the plurality of second populations.

[0055] Among them, the above-mentioned first population can include all nodes in the above-mentioned first network topology, that is, each node corresponds to an individual in the above-mentioned first population, and the above-mentioned first population is the initial population in the above-mentioned genetic algorithm. The above-mentioned initial population is iteratively updated by the above-mentioned genetic algorithm. For example: the above-mentioned first population is subjected to selection, crossover and mutation operations to obtain a second population after the first update, and then the second population is subjected to selection, crossover and mutation operations to obtain a second population after the second update, and so on. By performing iterative updates for a preset number of times, the second population obtained each time can be obtained, and then the network topology corresponding to each population can be restored based on the multiple second populations obtained.

[0056] It will be appreciated that in the aforementioned genetic algorithm, each update to the population produces an optimal individual, i.e., the second network topology obtained after each update. If the population evolves endlessly, the optimal solution can be obtained during iterative updates. By iteratively updating for a predetermined number of iterations (which can be set based on empirical values ​​or the first network topology to be updated), the number of iterative updates can be reduced while simultaneously achieving a more optimal second population. By determining the aforementioned multiple second network topologies from multiple second populations generated within the predetermined number of iterations, and then determining the target network topology within these multiple second network topologies, the number of iterative updates can be reduced, thereby improving the efficiency of obtaining multiple second network topologies.

[0057] In this embodiment, a first population corresponding to the first network topology is obtained; the first population is iteratively updated based on a genetic algorithm until a preset number of iterations is reached to obtain a second population obtained by each update; and the adjacency matrices of multiple second network topologies corresponding to the multiple second populations are obtained, thereby iteratively updating the adjacency matrix of the first network topology based on the genetic algorithm to obtain the adjacency matrices of multiple second network topologies.

[0058] Optionally, the acquiring the load balancing coefficient of each second network topology based on the predicted network traffic time series of each node in each second network topology includes:

[0059] Acquire a link set in each second network topology based on a depth-first search algorithm;

[0060] determining a link bandwidth utilization variance in each of the second network topologies based on a predicted network traffic time series of each node in each of the second network topologies and the link set;

[0061] The load balancing coefficient of each second network topology is determined based on the link bandwidth utilization variance in each second network topology.

[0062] Specifically, when using the depth-first search algorithm, an initial node can be arbitrarily selected, and other nodes in the second network topology can be searched starting from the initial node, thereby obtaining a link set in the second network topology, which includes all links in the second network topology. Furthermore, based on the predicted network traffic time series of each node on each link, the bandwidth utilization of each link can be obtained, thereby obtaining the link bandwidth utilization variance in each second network topology. It can be understood that the link bandwidth utilization variance can be used to indicate whether the load in each second network topology is balanced. The smaller the link bandwidth utilization variance, the more balanced the link load in the corresponding second network topology.

[0063] Among them, the above-mentioned link bandwidth utilization variance can be directly used as the above-mentioned load balancing coefficient, or the above-mentioned load balancing coefficient can be further calculated through the above-mentioned link bandwidth utilization variance to improve the differentiation of the load balancing of different network topologies represented by the above-mentioned load balancing coefficient. It can be understood that the above-mentioned load balancing coefficient is proportional to the above-mentioned link bandwidth utilization variance.

[0064] In this embodiment, a depth-first search algorithm is used to obtain a link set in each second network topology; the link bandwidth utilization variance in each second network topology is determined based on the predicted network traffic time series of each node in each second network topology and the link set; thereby, the link bandwidth utilization variance in each second network topology can be used to determine the load balancing coefficient of each second network topology.

[0065] Optionally, acquiring the link set in each second network topology based on a depth-first search algorithm includes:

[0066] Traversing and searching each node in each second network topology based on the depth-first search algorithm to obtain a plurality of links in each second network topology;

[0067] A link set of each second network topology is obtained based on the multiple links.

[0068] In this embodiment, each node in each second network topology is traversed and searched based on the depth-first search algorithm to obtain multiple links of each second network topology; based on the multiple links, the link set of each second network topology is obtained, so that the link set of each second network topology can be quickly obtained.

[0069] The various optional implementation methods introduced in the embodiments of the present application can be implemented in combination with each other or separately if they do not conflict with each other, and the embodiments of the present application do not limit this.

[0070] For ease of understanding, the specific implementation is as follows:

[0071] Internet TV traffic, including video requests and video transmission, is transmitted over the Internet Protocol (IP) network. Each router in the IP network can be considered a network node, and the traffic at each network node is characterized by high frequency and continuity. With the increasing number of Internet TV users, Internet TV traffic is growing exponentially, placing higher demands on the carrying capacity of IP networks. While simply expanding network bandwidth can alleviate the problem of insufficient network capacity to a certain extent, it is extremely costly and slow. Traditional IP routing systems, while capable of traffic planning, still cannot fully address the problem of insufficient network capacity. For example, Open Shortest Path First (OSPF) distributes traffic data only along the shortest path in the network topology, disregarding real-time network conditions and link loads, which can easily lead to link congestion. Specifically, there are two lines between two routers, one with 1G bandwidth and the other with 300M bandwidth. The 1G line has a shorter path, so Internet TV traffic will only use the 1G line. Even if the 1G line fails, the 300M line will be unavailable, meaning that load balancing is completely ineffective, leading to unresolved peak network congestion. Improper network topology planning can easily lead to link load imbalance, where some links are overloaded, causing congestion, while many links are underloaded and underutilized.

[0072] like Figure 2 As shown, the embodiment of the present application also provides a network topology optimization method, which can specifically include two steps: establishing a spatiotemporal traffic prediction model and optimizing the network topology;

[0073] Among them, establishing a spatiotemporal traffic prediction model: using network topology information, establishing a spatiotemporal traffic prediction model to predict network traffic, which can specifically include the following process:

[0074] Obtain network topology G and network traffic time series X;

[0075] According to the network topology G, the corresponding adjacency matrix A is calculated, where:

[0076]

[0077] The GNN (Graph Neural Network) model for spatiotemporal traffic prediction is established based on the adjacency matrix A and the network traffic time series X: the network structure of the GNN layer l+1 is Among them, Z (l) represents the features of the lth layer, Z (l+1) represents the features of the l+1th layer, σ represents the nonlinear transformation, D represents the degree matrix corresponding to the adjacency matrix A, and W (l) and b (l) Represents the neural network parameters; the features of the last layer of GNN are sent to the time series model GRU (Gate Recurrent Unit, gated recurrent unit) to predict the network traffic time series, then: [Y t+1 ,...Y t+T ]=f GNN+GRU ((X t-h ,...,X t ), where h represents the historical window size, T represents the prediction window size, and Y represents the predicted traffic;

[0078] Calculate the mean absolute error between the actual flow and the predicted flow based on the prediction results Where n represents the number of samples, Y represents the predicted traffic, X represents the actual traffic, and the mean absolute error is back-propagated to update the model parameters W and b until the model converges.

[0079] Among them, network topology optimization: use the existing spatiotemporal traffic prediction model to predict network traffic load, and optimize the network topology based on genetic algorithm. First, define the complex network topology optimization objective function to measure the network load balancing performance: the link bandwidth utilization of each link at time t is Where N is the number of nodes, y is the predicted traffic value of the node on the link, and U is the maximum capacity value of the node on the link. The average link bandwidth utilization of all links at time t is Where R is the number of links; the variance of link bandwidth utilization at time t is The smaller the index is, the more balanced the load is. The average link bandwidth utilization variance at T moments is defined as the objective function of complex network topology optimization: The smaller this indicator is, the more balanced the link load will be at the next time T. The goal of the genetic algorithm is to minimize the average link bandwidth utilization variance Var. The network topology optimization steps can specifically include the following:

[0080] (1) Obtain the network topology G and initialize the genetic algorithm population P = |I n*1 , S n*1 , E n*1 |, where S n*1 represents n starting nodes, E n*1 represents n terminal nodes, I n*1 Indicates whether two nodes are connected (connected is 1, not connected is 0), and each row of |i, s, e| represents the connection relationship between node i and node s; the adjacency matrix A is obtained based on P.

[0081] (2) Based on the network traffic data X from time th to t t-h ,...,X t And the adjacency matrix A predicts the network traffic of all nodes in the future T moments and obtains Y t+1 ,...Y t+T .

[0082] (3) Perform a depth-first search on the network topology G to obtain all links between any two nodes:

[0083] A. Initialize node V as the first traversed node; initialize the result link table res to an empty list;

[0084] B. Perform a depth-first search on the network topology G starting from the initial node V, that is, traverse each node N of the network topology G starting from the initial node V. The specific process is as follows:

[0085] a. Initialize the result link list n_res starting from node N to an empty list; initialize the visited list tmp_visited to an empty list; initialize the temporary link table tmp_res to an empty list.

[0086] b. If the current node N is not in the visited list tmp_visited, then add the current node N to the temporary link table tmp_res, add the temporary link table tmp_res to n_res, and add the current node N to the visited list tmp_visited:

[0087] (a) If the visited list tmp_visited contains all nodes of the network topology G, or there are no new nodes that can be deeply searched, then the search at this level is skipped.

[0088] (b) If condition a) is not met, find the neighbor nodes of the current node N in the network topology G, perform a next-level depth search for each neighbor node, and repeat operation b.

[0089] c. Add n_res to the result link table res.

[0090] C. Return the result link table res, which stores all links of the network topology G.

[0091] (4) Based on the result link table res and the network prediction traffic Y at the next T time t+1 ,...Y t+T Calculate the average link bandwidth utilization variance Var, which is the complex network topology optimization objective function; record the average link bandwidth utilization variance Var and the corresponding population P.

[0092] (5) Use genetic algorithm to perform selection, crossover and mutation operations on population P:

[0093] A. Selection: Repeat the selection operation on the population P n times, randomly select m individuals from P each time and select the individual with the smallest individual fitness, and send the selected n individuals to the next generation as one parent for crossover operation.

[0094] B. Crossover: Randomly select two parents P1 and P2 from n parents for crossover operation, randomly select a certain number of intersection points in the two parents, exchange |i, s, e| of these intersection points, and finally obtain n offspring.

[0095] C. Mutation: Randomly select a mutation point from n offspring and regenerate the |i, s, e| of the mutation point.

[0096] (6) Recalculate the adjacency matrix A based on P obtained in (5), and repeat operations (2) to (6) until the number of iterations is reached.

[0097] (7) The corresponding population P* obtained based on the global optimal value Var* of the average link bandwidth utilization variance is used to restore the network topology G, which is the final network topology optimization result.

[0098] In the embodiment of the present application, the optimization target of the network topology is not only the transmission delay, path length and other indicators that can be directly calculated based on the topological structure, but also takes into account the network space information and network time information, that is, the network topological structure and the network traffic information. The spatiotemporal traffic prediction algorithm can predict future network traffic information based on real-time network traffic information, capturing the spatiotemporal correlation in complex network environments. In addition, the network topology optimization method of the present application can avoid the local topology optimization caused by only considering local link information during topology optimization. When searching in the topological space, the global load balancing degree of the topology optimization result is evaluated based on the predicted traffic, fully considering the network resource information, and can obtain the optimal topology result that meets the global load balancing conditions, thereby improving the network stability of the topology optimization result. In addition, it can make full use of existing network resources to improve load balancing performance, avoid rough expansion, and maintain a good user experience.

[0099] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of a network topology updating device provided in an embodiment of the present application. Figure 3 As shown, the network topology updating device 300 includes:

[0100] A first acquisition module 301 is configured to acquire an adjacency matrix of a first network topology and a network traffic time series of each node in the first network topology;

[0101] A first updating module 302 is configured to iteratively update the adjacency matrix of the first network topology based on a genetic algorithm to obtain adjacency matrices of multiple second network topologies;

[0102] A prediction module 303 is configured to use the network traffic time series of each node in the first network topology and the adjacency matrices of the plurality of second network topologies as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction, thereby obtaining a predicted network traffic time series for each node in each of the second network topologies;

[0103] A second acquisition module 304 is configured to acquire a load balancing coefficient of each second network topology based on a predicted network traffic time series of each node in each second network topology;

[0104] The second updating module 305 is configured to update the first network topology using a target network topology, where the target network topology is a network topology with the smallest load balancing coefficient among the plurality of second network topologies.

[0105] Optionally, the spatiotemporal traffic prediction model includes a graph neural network model and a gated recurrent unit model;

[0106] The spatiotemporal flow prediction model is trained in the following way:

[0107] The graph neural network model and the gated recurrent unit model are trained using training samples of the adjacency matrix of the network topology and the network traffic time series, where the output of the graph neural network model serves as the input of the gated recurrent unit model;

[0108] Obtain a loss value of a prediction result output by the spatiotemporal traffic prediction model, and update parameters of the spatiotemporal traffic prediction model based on the loss value until the spatiotemporal traffic prediction model converges.

[0109] Optionally, the first updating module 302 may specifically include:

[0110] A first acquiring unit, configured to acquire a first population corresponding to the first network topology;

[0111] an updating unit, configured to iteratively update the first population based on a genetic algorithm until a preset number of iterations is reached, to obtain a second population obtained by each update;

[0112] The second acquisition unit is configured to acquire adjacency matrices of a plurality of second network topologies corresponding to a plurality of the second populations.

[0113] Optionally, the second obtaining module 304 may specifically include:

[0114] A third acquisition unit is configured to acquire a link set in each second network topology based on a depth-first search algorithm;

[0115] a first determining unit, configured to determine a link bandwidth utilization variance in each second network topology based on a predicted network traffic time series of each node in each second network topology and the link set;

[0116] The second determining unit is configured to determine the load balancing coefficient of each second network topology based on the link bandwidth utilization variance in each second network topology.

[0117] Optionally, the third obtaining unit may specifically include:

[0118] Traversing and searching each node in each second network topology based on the depth-first search algorithm to obtain a plurality of links in each second network topology;

[0119] A link set of each second network topology is obtained based on the multiple links.

[0120] Optionally, the second updating module 305 may specifically include:

[0121] Determine a target network topology having a minimum load balancing coefficient among the plurality of second network topologies;

[0122] Obtaining an adjacency matrix of the target network topology;

[0123] Acquire the target network topology based on the adjacency matrix of the target network topology;

[0124] The first network topology is updated using the target network topology.

[0125] The network topology updating device 300 can realize the embodiment of the present application Figure 1 The various processes of the method embodiment and the achievement of the same beneficial effects are not described again here to avoid repetition.

[0126] The embodiment of the present application also provides a communication device. Figure 1 The network topology updating method shown is similar, so the implementation of the communication device can refer to the implementation of the method, and the repeated parts will not be repeated. Figure 4 As shown, the communication device of the embodiment of the present application includes: a processor 400, which is used to read the program in the memory 410 and perform the following process:

[0127] Obtaining an adjacency matrix of a first network topology and a network traffic time series of each node in the first network topology;

[0128] Iteratively updating the adjacency matrix of the first network topology based on a genetic algorithm to obtain multiple adjacency matrices of the second network topology;

[0129] Using the network traffic time series of each node in the first network topology and the adjacency matrices of the plurality of second network topologies as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction, so as to obtain a predicted network traffic time series for each node in each of the second network topologies;

[0130] Obtaining a load balancing coefficient of each second network topology based on a predicted network traffic time series of each node in each second network topology;

[0131] The first network topology is updated using a target network topology, where the target network topology is a network topology with the smallest load balancing coefficient among the plurality of second network topologies.

[0132] Among them, Figure 4In the present disclosure, the bus architecture may include any number of interconnected buses and bridges, specifically linking various circuits such as one or more processors represented by processor 400 and memory represented by memory 410. The bus architecture may also link various other circuits such as peripherals, voltage regulators, and power management circuits, all of which are well known in the art and are not further described herein. The bus interface provides an interface. Processor 400 is responsible for managing the bus architecture and general processing, while memory 410 may store data used by processor 400 when performing operations.

[0133] Optionally, the spatiotemporal traffic prediction model includes a graph neural network model and a gated recurrent unit model;

[0134] The spatiotemporal flow prediction model is trained in the following way:

[0135] The graph neural network model and the gated recurrent unit model are trained using training samples of the adjacency matrix of the network topology and the network traffic time series, where the output of the graph neural network model serves as the input of the gated recurrent unit model;

[0136] Obtain a loss value of a prediction result output by the spatiotemporal traffic prediction model, and update parameters of the spatiotemporal traffic prediction model based on the loss value until the spatiotemporal traffic prediction model converges.

[0137] Optionally, the processor 400 is further configured to read a program in the memory 410 and execute the following steps:

[0138] Obtaining a first population corresponding to the first network topology;

[0139] Iteratively updating the first population based on a genetic algorithm until a preset number of iterations is reached to obtain a second population obtained by each update;

[0140] Acquire adjacency matrices of a plurality of second network topologies corresponding to the plurality of second populations.

[0141] Optionally, the processor 400 is further configured to read a program in the memory 410 and execute the following steps:

[0142] Acquire a link set in each second network topology based on a depth-first search algorithm;

[0143] determining a link bandwidth utilization variance in each of the second network topologies based on a predicted network traffic time series of each node in each of the second network topologies and the link set;

[0144] The load balancing coefficient of each second network topology is determined based on the link bandwidth utilization variance in each second network topology.

[0145] Optionally, the processor 400 is further configured to read a program in the memory 410 and execute the following steps:

[0146] Traversing and searching each node in each second network topology based on the depth-first search algorithm to obtain a plurality of links in each second network topology;

[0147] A link set of each second network topology is obtained based on the multiple links.

[0148] Optionally, the processor 400 is further configured to read a program in the memory 410 and execute the following steps:

[0149] Determine a target network topology having a minimum load balancing coefficient among the plurality of second network topologies;

[0150] Obtaining an adjacency matrix of the target network topology;

[0151] Acquire the target network topology based on the adjacency matrix of the target network topology;

[0152] The first network topology is updated using the target network topology.

[0153] The communication device provided in the embodiment of the present application can perform the above Figure 1 The implementation principle and technical effects of the method embodiment shown are similar, and will not be described in detail in this embodiment.

[0154] The present application also provides a readable storage medium for storing a program, which is executed by a processor to implement the following Figure 1 The various processes of the method embodiments are similar and can achieve the same technical effects. To avoid repetition, they will not be described here.

[0155] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection of some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0156] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may be physically included separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0157] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute some steps of the sending and receiving methods described in various embodiments of the present application. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program code.

[0158] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A network topology updating method, characterized in that: include: Obtaining an adjacency matrix of a first network topology and a network traffic time series of each node in the first network topology; Iteratively updating the adjacency matrix of the first network topology based on a genetic algorithm to obtain multiple adjacency matrices of the second network topology; Using the network traffic time series of each node in the first network topology and the adjacency matrices of the plurality of second network topologies as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction, so as to obtain a predicted network traffic time series for each node in each of the second network topologies; Obtaining a load balancing coefficient for each second network topology based on a predicted network traffic time series for each node in each second network topology; the load balancing coefficient is used to represent a degree of load balancing for each second network topology, where a smaller load balancing coefficient indicates a more balanced load in the corresponding second network topology; Updating the first network topology using a target network topology, where the target network topology is a network topology with the smallest load balancing coefficient among the plurality of second network topologies; The iterative updating of the adjacency matrix of the first network topology based on the genetic algorithm to obtain multiple adjacency matrices of the second network topology includes: Obtain a first population corresponding to the first network topology; iteratively update the first population based on a genetic algorithm until a preset number of iterations is reached to obtain a second population obtained by each update; and obtain adjacency matrices of multiple second network topologies corresponding to multiple second populations.

2. The method according to claim 1, wherein The spatiotemporal traffic prediction model includes a graph neural network model and a gated recurrent unit model; The spatiotemporal flow prediction model is trained in the following way: The graph neural network model and the gated recurrent unit model are trained using training samples of the adjacency matrix of the network topology and the network traffic time series, where the output of the graph neural network model serves as the input of the gated recurrent unit model; Obtain a loss value of a prediction result output by the spatiotemporal traffic prediction model, and update parameters of the spatiotemporal traffic prediction model based on the loss value until the spatiotemporal traffic prediction model converges.

3. The method according to claim 1, wherein The acquiring of the load balancing coefficient of each second network topology based on the predicted network traffic time series of each node in each second network topology includes: Acquire a link set in each second network topology based on a depth-first search algorithm; determining a link bandwidth utilization variance in each of the second network topologies based on a predicted network traffic time series of each node in each of the second network topologies and the link set; The load balancing coefficient of each second network topology is determined based on the link bandwidth utilization variance in each second network topology.

4. The method according to claim 3, wherein The acquiring the link set in each second network topology based on the depth-first search algorithm includes: Traversing and searching each node in each second network topology based on the depth-first search algorithm to obtain a plurality of links in each second network topology; A link set of each second network topology is obtained based on the multiple links.

5. The method according to any one of claims 1 to 4, characterized in that The updating of the first network topology using the target network topology includes: Determine a target network topology having a minimum load balancing coefficient among the plurality of second network topologies; Obtaining an adjacency matrix of the target network topology; Acquire the target network topology based on the adjacency matrix of the target network topology; The first network topology is updated using the target network topology.

6. A network topology updating device, characterized in that: include: A first acquisition module is configured to acquire an adjacency matrix of a first network topology and a network traffic time series of each node in the first network topology; A first updating module, configured to iteratively update the adjacency matrix of the first network topology based on a genetic algorithm to obtain adjacency matrices of multiple second network topologies; a prediction module, configured to use the network traffic time series of each node in the first network topology and the adjacency matrices of the plurality of second network topologies as inputs of a pre-trained spatiotemporal traffic prediction model to perform network traffic time series prediction, so as to obtain a predicted network traffic time series for each node in each of the second network topologies; a second acquisition module, configured to acquire a load balancing coefficient for each second network topology based on a predicted network traffic time series of each node in each second network topology; the load balancing coefficient being used to represent a degree of load balancing for each second network topology, wherein a smaller load balancing coefficient indicates a more balanced load in the corresponding second network topology; A second updating module is configured to update the first network topology using a target network topology, where the target network topology is a network topology with the smallest load balancing coefficient among the plurality of second network topologies; The first update module may specifically include: A first acquiring unit, configured to acquire a first population corresponding to the first network topology; an updating unit, configured to iteratively update the first population based on a genetic algorithm until a preset number of iterations is reached, to obtain a second population obtained by each update; The second acquisition unit is configured to acquire adjacency matrices of a plurality of second network topologies corresponding to a plurality of the second populations.

7. The device according to claim 6, characterized in that The spatiotemporal traffic prediction model includes a graph neural network model and a gated recurrent unit model; The spatiotemporal flow prediction model is trained in the following way: The graph neural network model and the gated recurrent unit model are trained using training samples of the adjacency matrix of the network topology and the network traffic time series, where the output of the graph neural network model serves as the input of the gated recurrent unit model; Obtain a loss value of a prediction result output by the spatiotemporal traffic prediction model, and update parameters of the spatiotemporal traffic prediction model based on the loss value until the spatiotemporal traffic prediction model converges.

8. A communication device comprising: A transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor; characterized in that: The processor is configured to read a program in a memory to implement the steps of the method according to any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that The computer readable storage medium is used to store a computer program. The computer program is stored on the computer readable storage medium. When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 5 are implemented.

Images