Transform-based satellite network service routing deployment method combined with swarm intelligence

Abstract

The invention discloses a satellite network service routing deployment method based on Transform combined with swarm intelligence, and the key points of the technical scheme are as follows: the method realizes optimized routing path selection of a service function chain (SFC) request and embedded deployment of a virtual network function (VNF) by combining a Transform encoder, a decoder module and a swarm intelligence algorithm; comprising the following steps of S1, satellite network state construction, S2, service function chain request modeling, S3, initial global resource state coding, S4, SFC request feature coding, S5, heuristic information generation, S6, path search based on ant colony optimization, S7, VNF embedding deployment decision making, S8, path resource state updating and S9, multi-round iterative optimization. According to the method, routing and deployment optimization of the SFC request can be efficiently completed in the dynamic satellite network, network delay and resource consumption are remarkably reduced, load balancing and efficient resource utilization are achieved, routing and deployment decisions of the SFC request can be effectively optimized, network loads are balanced, and the resource utilization rate is increased.

Description

Technical Field

[0001] The present invention relates to the field of satellite communication technology, and in particular to a satellite network service routing deployment method based on Transformer combined with swarm intelligence. Background Art

[0002] With the development of communication technology, low-orbit satellite (LEO) networks, as part of non-terrestrial networks (NTNs), have gradually become a research hotspot because they can provide global coverage and high-reliability connections. LEO satellite networks can effectively solve the problem of network access in remote areas, especially in 5G and future 6G networks, and their application prospects are becoming more and more extensive. However, in these complex network services, service function chaining (SFC) technology needs to arrange multiple service functions (such as firewalls, load balancing, etc.) in sequence to achieve customized network behaviors. In LEO satellite networks, due to limited resources, frequent dynamic changes in links, and limited node computing power, SFC routing and virtual network function (VNF) deployment become very challenging. Existing deployment methods (such as integer linear programming, greedy algorithms, deep reinforcement learning, etc.) are effective in static scenarios, but in large-scale dynamic networks, they have high computational complexity, poor flexibility, and low resource utilization.

[0003] In recent years, swarm intelligence algorithms (such as ant colony optimization) have been mentioned as a solution, but their applications in dynamic environments are still rare. At the same time, the Transformer model has great potential in sequence modeling and resource reasoning. How to combine it with swarm intelligence algorithms to improve SFC deployment efficiency and resource utilization is still in the exploration stage. To solve the above problems, we propose a satellite network service routing deployment method based on Transformer combined with swarm intelligence. It aims to solve the SFC routing and VNF embedding problems in dynamic satellite networks by integrating the modeling capabilities of Transformer and the global search capabilities of ant colony optimization, and provide an efficient and flexible network service deployment solution. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a satellite network service routing deployment method based on Transformer combined with swarm intelligence to solve the problems raised in the background technology.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions:

[0006] A satellite network service routing deployment method based on Transformer combined with swarm intelligence, which realizes the optimized routing path selection of service function chain (SFC) requests and the embedded deployment of virtual network functions (VNF) by combining Transformer encoder and decoder modules and swarm intelligence algorithms, and includes the following steps:

[0007] S1. Satellite network status construction: Construct the topology of the satellite network and define the resource status of network nodes and links. The characteristics of nodes include available computing resource capacity, storage resource capacity, energy reserve, and whether they are the source node or target node of the service function chain; the characteristics of links include bandwidth resources, delay, and the remaining effective time of the link;

[0008] S2, Service Function Chain Request Modeling: The resource requirements of each virtual network function (VNF) in the SFC request are represented as a multi-dimensional resource vector, including computing resource requirements, storage resource requirements, bandwidth requirements, energy consumption requirements and duration;

[0009] S3, initial global resource state encoding: using graph neural network (GNN) to feature encode the node and link resource states of the satellite network and generate a global network resource state representation;

[0010] S4, SFC request feature encoding: The dependency relationship between virtual network functions (VNFs) in the service function chain request is modeled through the Transformer encoder module to generate the service request feature encoding;

[0011] S5, heuristic information generation: through the resource node matching attention mechanism, the global network state encoding is matched with the service request feature encoding to generate heuristic information for ant colony optimization;

[0012] S6. Path search based on ant colony optimization: Using the ant colony optimization algorithm, combined with the generated heuristic information, search for the optimized routing path from the service function chain source node to the target node, and calculate the resource distribution on the path;

[0013] S7, VNF embedding deployment decision: Using the Transformer decoder module, embedding deployment decisions are made for each virtual network function (VNF) in the service function chain in turn in an autoregressive manner, and the optimal embedding node is selected based on the remaining resources of the current path node and the VNF demand;

[0014] S8, path resource status update: after each virtual network function is embedded, the resource status of the path is updated to provide a dynamic reference for subsequent embedding decisions;

[0015] S9, multi-round iterative optimization: through multiple rounds of ant colony optimization iterations, the pheromone level is updated, and finally the routing path and VNF embedding solution with the lowest total cost function are selected.

[0016] Preferably, the heuristic information generation module is generated through a graph neural network and a resource node matching attention mechanism, wherein node characteristics include computing resource capacity, storage resource capacity, energy reserves, and a flag indicating whether it is a source or target node, and link characteristics include bandwidth, delay, and the remaining effective time of the link.

[0017] Preferably, the Transformer encoder module adopts a multi-head self-attention mechanism to model the dependencies between virtual network functions in the service function chain and generate a feature matrix containing resource requirements.

[0018] Preferably, the Transformer decoder module dynamically refers to the feature encoding of the service request and the current path resource status through the encoder-decoder attention mechanism when embedding each virtual network function, thereby optimizing the embedding decision.

[0019] Preferably, the ant colony optimization module simulates the foraging behavior of ants and uses the state transition probability function to search and select paths. The quality of the path is determined by the heuristic function and the pheromone level. The formula is as follows:

[0020]

[0021] Among them, τ i，j represents the path pheromone concentration, η i，j is the heuristic function value, α and β are adjustment parameters, N i is the set of neighbor nodes of node i. The transition probability is jointly affected by the heuristic information of the path (such as bandwidth, delay) and the historical pheromone concentration. This module generates multiple candidate paths by simulating the path selection behavior of multiple ants and updates the pheromone concentration on the path according to the path cost (such as resource consumption and delay):

[0022]

[0023] Where cost n The total cost of the path generated for the nth ant. Finally, the candidate path output by this module provides a decision basis for subsequent modules.

[0024] Preferably, the heuristic function generation module is used to generate accurate heuristic function values ​​through a neural network to guide the ant colony optimization module to quickly converge to a more optimal path. The module uses a graph neural network (GNN) to encode the satellite network topology, resource status, and service requests to capture the complex relationship between nodes in the network;

[0025] The resource characteristics of node i are expressed as:

[0026]

[0027] in, and Represent the remaining CPU, memory and energy resources of the node, respectively. i and Dest i It is a binary flag (indicating whether the node is the source or destination node of the SFC request);

[0028] The link characteristic connecting nodes i and j is expressed as:

[0029] y i，j =[B i，j , D i，j , Dur i，j ],

[0030] Among them B i，j is the link bandwidth, D i，j is the link delay, Dur i，j is the available duration of the link;

[0031] Through the propagation mechanism of the graph neural network, the features of nodes and edges are updated in multiple layers, capturing the global characteristics of the network:

[0032]

[0033] Among them, l is the number of network layers, σ is the nonlinear activation function, and W l , U l 、V l and Q l is the weight matrix. Through multiple layers of iteration, the heuristic function value η is finally generated. i，j It can characterize the quality of the path and thus guide the path selection of the ant colony optimization module.

[0034] Preferably, the SFC encoding module is used to encode the complex dependencies between the service function chain (SFC) request and its internal virtual network functions (VNFs), so as to provide accurate representation for subsequent deployment decisions. The module is implemented based on the Transformer model and uses its self-attention mechanism to process the global characteristics in the SFC request;

[0035] Input representation:

[0036] For each VNF in the SFC request, its resource requirement is expressed as:

[0037]

[0038] The items represent the required CPU, memory, bandwidth, energy, and execution time, respectively. The requirements of all VNFs are combined into a matrix X:

[0039]

[0040] Through the Transformer encoder, the query (Q), key (K) and value (V) are calculated using the multi-head self-attention mechanism:

[0041] Q=W Q X, K = W K X, V = W V X,

[0042] And calculate the output through the attention mechanism:

[0043]

[0044] The multi-layer Transformer encoder can capture the complex dependencies between VNFs, and the encoding results are used for subsequent deployment decisions.

[0045] Preferably, the VNF deployment decision module is used to gradually deploy each VNF in the SFC to the optimal node on the path according to the generated routing path. The module uses the autoregressive characteristics of the Transformer decoder to dynamically update the path resource status in each decision. The following is the VNF deployment decision process:

[0046] Input the encoded SFC request and path resource status;

[0047] Generate the deployment probability distribution of the current VNF ​​on the path node through the multi-head attention mechanism of the decoder;

[0048] Select the optimal node for deployment through the Softmax function:

[0049]

[0050] where z i is the score of node i;

[0051] After the deployment is completed, the path resource status is updated to provide a reference for the next deployment;

[0052] This module iterates step by step in an autoregressive manner. Each time a VNF is deployed, the resource representation in the path is updated to provide instructions for the next VNF ​​deployment decision, thus completing the deployment of all VNFs.

[0053] Preferably, the resource node matching attention mechanism is used to calculate the node's fitness based on the node's remaining resources and the SFC's requirements, thereby providing more accurate guidance for path selection and deployment;

[0054] The matching weight is calculated by the following formula:

[0055]

[0056] Among them, y i，j is the resource status representation of the path, X k is the resource requirement of the kth VNF in the SFC. The higher the weight, the more suitable the node is for the deployment of the current VNF.

[0057] In summary, the present invention mainly has the following beneficial effects:

[0058] 1. The present invention searches for the optimal routing path of SFC requests through the ant colony optimization module through state transition probability and pheromone update rules. The heuristic information generation module generates heuristic information using a dual graph neural network (GNN) layer and a resource-node matching attention mechanism to guide the ant colony to converge quickly. The SFC encoder module is based on the Transformer encoder to capture the complex dependencies between the VNFs in the SFC request. Through the collaborative work of the above five modules, the present invention can efficiently complete the routing and deployment optimization of SFC requests in dynamic satellite networks, significantly reduce network latency and resource consumption, and achieve load balancing and efficient resource utilization.

[0059] 2. Based on the Transformer decoder, the deployment decision of the VNF is completed autoregressively. The data matrix module is used to store and update the network resource status and related information of the SFC request. Through the above training and execution process, the present invention can effectively optimize the routing and deployment decision of the SFC request, balance the network load, and improve resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a neural network framework diagram of a satellite network service routing deployment method based on Transformer combined with swarm intelligence of the present invention;

[0061] Figure 2 It is a VNF embedding flowchart based on the Transformer autoregressive form of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0063] The following examples are used to illustrate the present invention, but they cannot be used to limit the scope of protection of the present invention. The conditions in the examples can be further adjusted according to specific conditions. Simple improvements to the method of the present invention under the premise of the concept of the present invention belong to the scope of protection claimed in the present invention.

[0064] Example 1

[0065] refer to Figure 1-Figure 2 A satellite network service routing deployment method based on Transformer combined with swarm intelligence. The present invention optimizes the routing and deployment decisions of the satellite network service function chain (SFC) by training the SFC encoder module, the virtual network function (VNF) deployment decision module and the heuristic information generation module. The specific steps of the training phase are as follows:

[0066] In the training stage, the present invention optimizes the routing and deployment decision of the satellite network service function chain (SFC) by training the SFC encoder module, the virtual network function (VNF) deployment decision module and the heuristic information generation module. Specifically, the parameters are initialized first, including the SFC request instance ρ, the initial routing path solution s, the module parameter θ2, the benchmark sampling number T b , epsilon-greedy parameter ò and number of training iterations T tr ; Then, the initial ant colony optimization algorithm is used to generate the initial routing path solution $s$ for a given SFC request instance. In each training iteration, the SFC request instance and routing resource distribution are input into the SFC encoder and VNF deployment decision module. The epsilon-greedy strategy is used to select nodes suitable for the current VNF ​​deployment and update the resource distribution of the routing path. After all VNFs are deployed, the benchmark function b(v|s) is calculated and the resource distribution variance function V is evaluated, where the resource distribution variance function is defined as:

[0067] V=max(0,V after -V ini )+α·V after ,

[0068] Among them, V ini represents the initial resource distribution variance of the routing path, V after represents the resource distribution variance of the path after VNF deployment, α is the weight factor, and the gradient of the network parameters is further calculated according to the resource distribution variance function. The gradient formula is:

[0069]

[0070] Among them, b(v|s) is the benchmark function, which is obtained by calculating the average variance value under the current network parameters. Finally, the above gradient is used to update the module parameter θ2. At the same time, in the training of the heuristic information generation module, the trained SFC encoder module and VNF deployment decision module are used to generate the heuristic information η θ , the path search process is simulated by virtual ant colonies, and the network parameters θ of the heuristic information generation module are gradually optimized net , its gradient formula is:

[0071]

[0072] Where f(s, v) is the cost function calculated based on the routing solution s and the VNF deployment solution v, b(r) is the average cost value of the solution searched by the ants, is the distribution of path solutions.

[0073] Example 2

[0074] refer to Figure 1-Figure 2 ,A satellite network service routing and deployment method based on Transformer combined with swarm intelligence,In the execution phase, the trained model is used to route and deploy SFC requests in the satellite network.,First, the parameters are initialized, including the SFC request instance ρ, the number of ant colonies T a , ant colony iteration times T c And pheromone level, then, in each ant colony iteration, the trained SFC encoder module and heuristic information generation module are used to generate heuristic information η θ , each ant moves according to the current node information and heuristic information η θ , combined with the pheromone level, calculate the next hop state transition probability;

[0075] The state transition probability formula is:

[0076]

[0077] After the path search is completed, the SFC encoder module and the VNF deployment decision module are used to generate the VNF deployment solution through autoregression. k , and evaluate the cost function value cost of the routing solution and deployment solution generated by the current ant k , according to the solutions of all ants and the cost function value, the pheromone distribution is updated, and the update formula is:

[0078]

[0079] When all ant colonies have completed iterations, the routing solution with the best overall cost is selected. * and deployment solution * As the final result.

[0080] Working principle: Please refer to Figure 1-Figure 2 As shown, the present invention searches for the optimal routing path of the SFC request through the state transition probability and pheromone update rule through the ant colony optimization module, the heuristic information generation module generates heuristic information using the dual graph neural network (GNN) layer and the resource-node matching attention mechanism to guide the ant colony to converge quickly, and the SFC encoder module is based on the Transformer encoder to capture the complex dependencies between the VNFs in the SFC request. Through the collaborative work of the above five modules, the present invention can efficiently complete the routing and deployment optimization of the SFC request in the dynamic satellite network, significantly reduce network latency and resource consumption, and achieve load balancing and efficient resource utilization. Based on the Transformer decoder, the deployment decision of the VNF is completed autoregressively. The data matrix module is used to store and update the network resource status and related information of the SFC request. Through the above training and execution process, the present invention can effectively optimize the routing and deployment decisions of the SFC request, balance the network load, and improve resource utilization.

[0081] Although the embodiments of the present invention have been shown and described, it is understood by those skilled in the art that, unless otherwise defined, the technical terms or scientific terms used in the present invention should have the usual meanings understood by persons with ordinary skills in the field to which the present invention belongs, and the words "include" or "comprise" and the like used in the present invention mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents.

[0082] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A satellite network service routing deployment method based on Transformer combined with swarm intelligence, characterized in that: The method combines the Transformer encoder and decoder modules and the swarm intelligence algorithm to achieve the optimal routing path selection of service function chain (SFC) requests and the embedded deployment of virtual network functions (VNFs), including the following steps: S1. Satellite network status construction: Construct the topology of the satellite network and define the resource status of network nodes and links. The characteristics of nodes include available computing resource capacity, storage resource capacity, energy reserve, and whether they are the source node or target node of the service function chain; the characteristics of links include bandwidth resources, delay, and the remaining effective time of the link; S2, Service Function Chain Request Modeling: The resource requirements of each virtual network function (VNF) in the SFC request are represented as a multi-dimensional resource vector, including computing resource requirements, storage resource requirements, bandwidth requirements, energy consumption requirements and duration; S3, initial global resource state encoding: using graph neural network (GNN) to feature encode the node and link resource states of the satellite network and generate a global network resource state representation; S4, SFC request feature encoding: The dependency relationship between virtual network functions (VNFs) in the service function chain request is modeled through the Transformer encoder module to generate the service request feature encoding; S5, heuristic information generation: through the resource node matching attention mechanism, the global network state encoding is matched with the service request feature encoding to generate heuristic information for ant colony optimization; S6. Path search based on ant colony optimization: Using the ant colony optimization algorithm, combined with the generated heuristic information, search for the optimized routing path from the service function chain source node to the target node, and calculate the resource distribution on the path; S7, VNF embedding deployment decision: Using the Transformer decoder module, embedding deployment decisions are made for each virtual network function (VNF) in the service function chain in turn in an autoregressive manner, and the optimal embedding node is selected based on the remaining resources of the current path node and the VNF demand; S8, path resource status update: after each virtual network function is embedded, the resource status of the path is updated to provide a dynamic reference for subsequent embedding decisions; S9, multi-round iterative optimization: through multiple rounds of ant colony optimization iterations, the pheromone level is updated, and finally the routing path and VNF embedding solution with the lowest total cost function are selected.

2. According to claim 1, a satellite network service routing deployment method based on Transformer combined with swarm intelligence is characterized in that: The heuristic information generation module is generated through a graph neural network and a resource node matching attention mechanism, wherein node features include computing resource capacity, storage resource capacity, energy reserves, and a flag indicating whether it is a source or target node, and link features include bandwidth, latency, and the remaining effective time of the link.

3. According to claim 1, a satellite network service routing deployment method based on Transformer combined with swarm intelligence is characterized in that: The Transformer encoder module adopts a multi-head self-attention mechanism to model the dependencies between virtual network functions in the service function chain and generate a feature matrix containing resource requirements.

4. According to claim 1, a satellite network service routing deployment method based on Transformer combined with swarm intelligence is characterized in that: The Transformer decoder module dynamically refers to the feature encoding of the service request and the current path resource status through the encoder-decoder attention mechanism when embedding each virtual network function, thereby optimizing the embedding decision.

5. According to claim 1, a satellite network service routing deployment method based on Transformer combined with swarm intelligence is characterized in that: The ant colony optimization module simulates the foraging behavior of ants and uses the state transition probability function to search and select paths. The quality of the path is determined by the heuristic function and the pheromone level. The formula is as follows: Among them, τ i,j represents the path pheromone concentration, η i,j is the heuristic function value, α and β are adjustment parameters, N i is the set of neighbor nodes of node i. The transition probability is jointly affected by the heuristic information of the path (such as bandwidth, delay) and the historical pheromone concentration. This module generates multiple candidate paths by simulating the path selection behavior of multiple ants and updates the pheromone concentration on the path according to the path cost (such as resource consumption and delay): Where cost n The total cost of the path generated for the nth ant. Finally, the candidate path output by this module provides a decision basis for subsequent modules.

6. According to claim 1, a satellite network service routing deployment method based on Transformer combined with swarm intelligence is characterized in that: The heuristic function generation module is used to generate accurate heuristic function values ​​through a neural network to guide the ant colony optimization module to quickly converge to a better path. The module uses a graph neural network (GNN) to encode the satellite network topology, resource status, and service requests to capture the complex relationships between nodes in the network.

7. The satellite network service routing deployment method based on Transformer combined with swarm intelligence according to claim 1, characterized in that: The SFC encoding module is used to encode the complex dependencies between the service function chain (SFC) request and its internal virtual network functions (VNFs), so as to provide accurate representation for subsequent deployment decisions. This module is implemented based on the Transformer model and uses its self-attention mechanism to process the global characteristics in the SFC request.

8. According to claim 1, a satellite network service routing deployment method based on Transformer combined with swarm intelligence is characterized in that: The function of the VNF deployment decision module is to gradually deploy each VNF in the SFC to the optimal node on the path according to the generated routing path. The module uses the autoregressive characteristics of the Transformer decoder to dynamically update the path resource status in each decision.

9. The satellite network service routing deployment method based on Transformer combined with swarm intelligence according to claim 1, characterized in that: The role of the resource node matching attention mechanism is to calculate the node's fitness based on the node's remaining resources and the SFC's requirements, thereby providing more precise guidance for path selection and deployment.

Images