A robust virtual network embedding method and system for ultra-dense low earth orbit satellite dynamic space environment

Abstract

The application discloses a kind of robust virtual network embedding methods and systems for ultra-dense low-orbit satellite dynamic space environment, method includes: constructing the virtual network embedding optimization problem with the goal of maximizing long-term system utility, wherein system utility is composed of virtual network request acceptance rate and revenue cost ratio;The optimization problem is decomposed into two sub-problems of node mapping and link mapping, when node mapping, virtual network request is mapped to the cluster with the highest average available resources, based on reinforcement learning method, adaptive noise mask deep deterministic policy gradient algorithm is used to map satellite and virtual node within the cluster;Link mapping is carried out using bandwidth priority shortest path algorithm after node mapping.Simulation results show that the application can more effectively cope with the challenges brought by dynamic changes of link resources in ultra-dense low-orbit satellite network, still maintain good robustness under dynamic network topology and resource damage environment, enhance the flexibility and stability of satellite network resource management.

Description

Technical Field

[0001] This invention relates to the field of resource management based on ultra-dense low-Earth orbit satellite constellations, and mainly to a robust virtual network embedding method and system for the dynamic space environment of ultra-dense low-Earth orbit satellites. Background Technology

[0002] The sixth-generation (6G) communication networks aim to provide faster, more reliable, and lower-latency services to meet the growing demand for high bandwidth, high capacity, and multiple connections. However, traditional terrestrial networks have limitations, such as limited coverage and network congestion, which pose challenges to fully meeting the requirements of high-performance communication. Satellite networks have emerged as an important technological solution to address these issues. Compared to geostationary orbit (GEO) and medium Earth orbit (MEO) satellites, low Earth orbit (LEO) satellites operating at lower altitudes offer several advantages, including reduced signal latency, faster data transmission speeds, and lower deployment costs. Therefore, companies like SpaceX and OneWeb plan to establish ultra-dense satellite constellations by launching thousands of LEO satellites. These ultra-dense low Earth orbit satellite networks (UD-LSNs) can provide seamless global coverage and high-performance services, effectively compensating for the limitations of terrestrial networks.

[0003] UD-LSN offers services with different service level agreements, such as Enhanced Mobile Broadband (eMBB), Ultra-Reliable Low-Latency Communications (URLLC), and Massive Machine-Type Communications (mMTC). This wide variety of service demands requires satellite network operators to provide tailored solutions for their customers, primarily through network slicing technology. Network slicing, combined with Network Functions Virtualization (NFV) and Software-Defined Networking (SDN) technologies, enables satellite network operators to build dedicated, virtualized logical networks on a common physical infrastructure to meet diverse customer communication needs. Each transport network slice is represented by a virtual network containing a set of Virtual Network Functions (VNFs) and logical links connecting them. The core of effectively implementing network slicing in UD-LSN lies in Virtual Network Embedding (VNE). This process involves mapping virtual network resources onto physical underlying network resources such as satellite nodes and communication paths. Designing efficient VNE algorithms can maximize the utilization of underlying resources, accommodate more Virtual Network Requests (VNRs), increase operator revenue, and minimize embedding costs.

[0004] Because the Virtual Network Optimization (VNE) problem is a nondeterministic multinomial-time problem, existing research has largely focused on heuristic virtual network connectivity algorithms, which use a set of hand-defined rules and constraints. With the development of artificial intelligence (AI) technology, some papers have transformed the original problem into a Markov Decision Process (MDP) and used deep reinforcement learning (DRL) algorithms to solve these problems. However, these works primarily focus on VNEs in static terrestrial networks. However, the movement of satellites causes dynamic changes in the topology of satellite networks, meaning that VNE algorithms designed for terrestrial networks cannot be directly applied to satellite networks.

[0005] While some existing studies have considered the dynamic characteristics of satellite network topologies, these studies consider relatively small-scale satellite networks, and the proposed solutions cannot be directly applied to large-scale satellite network scenarios. Furthermore, since UD-LSNs operate in an open space environment, electromagnetic, thermal, and other space environmental effects can affect the stability of inter-satellite links, leading to random damage to inter-satellite link resources. Therefore, a VNE (Video Network Adapter) solution capable of handling such resource fluctuations is needed. Summary of the Invention

[0006] Purpose of the invention: To address the shortcomings of existing technologies, the present invention aims to provide a robust virtual network embedding method and system for the dynamic space environment of ultra-dense low-Earth orbit satellites. This method is applicable to scenarios where satellite mobility and space environment influences dynamic changes in satellite network topology and resource availability. It ensures the reception rate and cost-benefit ratio of virtual network requests in dynamic space environments, thereby improving service quality.

[0007] Technical Solution: To achieve the above objectives, the technical solution adopted by this invention is: a robust virtual network embedding method for ultra-dense low-Earth orbit satellite dynamic space environments, comprising the following steps:

[0008] Construct a virtual network embedding optimization problem with the goal of maximizing long-term system utility, where system utility consists of virtual network request reception rate and revenue-cost ratio; constrain each virtual node in the virtual network to be mapped to only one satellite node, the computing resources required by the virtual node must be within the available computing resources of the satellite node to be mapped, each virtual link is mapped to at least one satellite link, and the bandwidth allocated to the virtual link is always within the available bandwidth of the satellite link; the actual available bandwidth resources of the satellite link take into account the damage caused by the space environment to the inter-satellite link bandwidth resources.

[0009] The optimization problem is decomposed into two sub-problems: node mapping and link mapping. During node mapping, virtual network requests are mapped to the cluster with the highest average available resources. Based on reinforcement learning, an adaptive noise masking deep deterministic policy gradient algorithm is used to map satellites within a cluster to virtual nodes. The current environmental state is observed, and the current mapping decision is output through the policy network. If the mapped node resources are insufficient or unmappable, the noise is adjusted, and a new action is selected.

[0010] After node mapping, a bandwidth-first shortest path algorithm is used for link mapping. Among the shortest paths that meet the bandwidth requirements, we prioritize those that provide the most available bandwidth resources.

[0011] Furthermore, satellite links in The actual available bandwidth resources within the window are:

[0012]

[0013] in, Indicates the first The bandwidth resources owned by each satellite link This indicates the damage caused by the space environment to inter-satellite link bandwidth resources.

[0014] Furthermore, the virtual network request reception rate is the ratio of successfully embedded virtual network requests to the total number of virtual network requests arriving within a given time window; in the virtual network embedding benefit-cost ratio, the embedding benefit is measured by the total amount of virtual computing and bandwidth resources of the virtual network requests, and the embedding cost is measured by the amount of physical satellite network resources used to accommodate the virtual network.

[0015] Furthermore, a virtual network mapping management system architecture based on grouping and clustering is adopted. The MEO satellite assesses the remaining resources of each LEO cluster and maps incoming virtual network requests to the LEO cluster with the highest average available resources.

[0016] Furthermore, the reinforcement learning action is the selected satellite for node mapping, the state is the available computing and bandwidth resources of each satellite node in the cluster, the set of satellite node indexes that have been mapped in the current virtual network request, the computing and bandwidth resources of the virtual node request at the current time, and the number of unmapped virtual nodes in the current virtual network request; the reward is determined by the success rate of virtual network request embedding and the benefit-cost ratio.

[0017] Furthermore, time step Rewards Among them, rewards ,award ; This represents the number of unmapped nodes in the current VNR. This represents the total number of nodes in the current VNR. and They represent the actions to be performed. The corresponding benefits and costs.

[0018] Furthermore, in the adaptive noise mask depth deterministic policy gradient algorithm, actual actions are generated by adding noise to the output of the actor network. If the selected satellite already carries other virtual nodes requested by the same virtual network or there is a lack of sufficient computing resources, the noise is regenerated. The number of attempts to select a valid node mapping operation is recorded as follows. Adaptive noise is represented as:

[0019]

[0020] in, It is the maximum value of the noise, a constant. This determines the rate at which noise increases with the number of attempts.

[0021] Furthermore, in the bandwidth-first shortest path algorithm, only paths that meet the bandwidth requirements are considered when calculating the shortest path; in the set of shortest paths, the path that provides the most available bandwidth resources is selected first; the minimum bandwidth of each path in the set is calculated, and the path with the maximum minimum bandwidth is selected as the final link mapping path; if no shortest path meets the resource requirements, an reachable path that meets the bandwidth requirements is selected.

[0022] The present invention also provides a computer system, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites.

[0023] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites.

[0024] Beneficial Effects: This invention provides a robust virtual network embedding method for ultra-dense low-Earth orbit (LEO) satellites in dynamic space environments. It achieves a robust virtual network mapping scheme for ultra-dense LEO satellite networks, ensuring service quality and improving network profitability. Compared with existing technologies, the main advantages are as follows: 1. To address the problem of a large number of nodes in ultra-large-scale LEO satellite networks, this invention adopts a grouping and clustering architecture for management. The system calculates the average available resource level of nodes within each cluster and prioritizes mapping virtual network requests to the cluster with the highest average available resource, thereby maximizing resource utilization and dynamically balancing network load. 2. Considering the inevitable impact of the space environment on satellites during on-orbit operation, this invention introduces resource impairment to model inter-satellite resources and incorporates this feature into the VNE decision-making process, significantly improving the accuracy and robustness of resource allocation in uncertain environments. 3. This invention innovatively introduces an adaptive noise masking mechanism in the node mapping stage, combined with a deep reinforcement learning strategy. When a selected node cannot be mapped, the noise is adaptively adjusted, aiming to achieve dynamic avoidance of mapping constraints and simultaneous improvement of strategy robustness. Attached Figure Description

[0025] Figure 1 This is a virtual network mapping scenario diagram of an ultra-dense low-Earth orbit satellite network according to an embodiment of the present invention;

[0026] Figure 2 This is a schematic diagram of the ANMD-BP algorithm according to an embodiment of the present invention;

[0027] Figure 3 This is a schematic diagram comparing the VNR receive rate of the ANMD-BP algorithm with other algorithms according to an embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram comparing the benefit-cost ratio of the ANMD-BP algorithm with other algorithms in this embodiment of the invention;

[0029] Figure 5 This is a schematic diagram comparing the overall efficiency of the ANMD-BP algorithm with other algorithms in this embodiment of the invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention are described in detail below with reference to the accompanying drawings. These embodiments are implemented based on the technical solutions of this invention, providing detailed implementation methods and specific operating procedures. It should be understood that the specific examples described herein are merely illustrative of this invention, but the scope of protection of this invention is not limited to the following embodiments.

[0031] This invention discloses a robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites, the main steps of which include:

[0032] Step 1: Construct a virtual network embedding optimization problem with the goal of maximizing long-term system utility.

[0033] This step first defines a virtual network mapping model in an ultra-dense low-Earth orbit satellite network, including dynamic satellite network modeling, virtual network modeling, node mapping modeling, and link mapping modeling in an open space environment. Then, considering the dynamics of inter-satellite links and the uncertainty of resources, a virtual network embedding optimization problem is constructed to maximize long-term system utility, where system utility consists of the virtual network request reception rate and the revenue-cost ratio. Detailed steps include:

[0034] Step 1-1: Define the network model, considering a large-scale low-Earth orbit satellite network architecture based on packet clustering. Satellites are modeled as an undirected graph that changes over time windows. Satellite nodes have computational resources, and satellite links have bandwidth resources. Due to the effects of the space environment, the bandwidth resources of satellite links are subject to random degradation.

[0035] Steps 1-2: Virtual network requests are modeled as an undirected graph. Virtual network nodes request a certain amount of computing resources, and virtual network links request a certain amount of bandwidth resources. Virtual network mapping includes node mapping and link mapping.

[0036] Steps 1-3: Define virtual node mappings. Each virtual node in the virtual network must be mapped to a different satellite node. The computing resources required by the virtual node must be within the range of the available computing resources of the mapped satellite node.

[0037] Steps 1-4: Define virtual link mapping. Each virtual link in the virtual network can be mapped to multiple satellite links. Considering the potential damage to satellite link resources caused by the space environment, the bandwidth allocated to the virtual link cannot exceed the available bandwidth of the satellite link.

[0038] Steps 1-5: Define the revenue-cost ratio of virtual network embedding, which represents the revenue generated per unit of cost and is a key indicator reflecting the efficiency of the VNR embedding process.

[0039] Steps 1-6: Define the virtual network request reception rate: the ratio of successfully embedded virtual network requests to the total number of virtual network requests arriving within a given time window.

[0040] Steps 1-7: Propose a virtual network embedding optimization problem aimed at maximizing long-term system utility.

[0041] Step 2: Decompose the optimization problem into two sub-problems: node mapping and link mapping. The main steps include:

[0042] Step 2-1: Due to the complex network state and large-scale nature of satellite networks, direct interaction with the network environment makes it difficult to generate a globally optimal node mapping scheme. A virtual network mapping management architecture based on grouping and clustering is adopted. MEO satellites assess the remaining resources of each cluster and map incoming VNRs to the cluster with the highest average available resources.

[0043] Step 2-2: Define the actions, states, and reward functions in reinforcement learning. The cluster head satellite agent collects state data and makes VNR mapping decisions, assigning virtual nodes to satellite nodes. Then, the environment returns a reward based on the action taken, and the system enters the next state.

[0044] Steps 2-3: Adaptive noise masking depth-deterministic policy gradient algorithm is used for node mapping. The current environment state is observed, and the current mapping decision is output through the policy network. If the mapped node resources are insufficient or unmappable, the noise is adjusted and a new action is selected. The policy network and evaluation network are then updated.

[0045] Step 3: After node mapping, further link mapping is performed using the bandwidth-first shortest path algorithm. The main steps include:

[0046] Step 3-1: Select the set of shortest paths that meet the bandwidth requirements. When calculating the shortest path, a bandwidth constraint is added to ensure that only paths with sufficient bandwidth are considered.

[0047] Step 3-2: In the shortest path set, we prioritize paths that provide the most available bandwidth resources. We calculate the minimum bandwidth for each path in the path set, and the path with the largest minimum bandwidth resource is selected as the final link mapping path.

[0048] Step 3-3: If no shortest path satisfies the resource requirements, we select an reachable path that meets the minimum bandwidth requirements to ensure that the mapping process can continue even in resource-constrained environments.

[0049] The following is combined with Figure 1 The large-scale low-Earth orbit satellite network scenario based on grouping and clustering shown in the illustration provides a detailed description of a robust virtual network embedding method for ultra-dense low-Earth orbit satellite dynamic space environments disclosed in this invention.

[0050] First, a virtual network mapping model is defined in an ultra-dense low-Earth orbit satellite network, including dynamic satellite network modeling, virtual network modeling, node mapping modeling, and link mapping modeling in an open space environment.

[0051] Satellite network modeling as a time window Changing undirected graph ,in Represents a set of satellite nodes. This represents a set of satellite links. Each satellite node is represented as , possess Computing resources. Each satellite link is represented as ,have Bandwidth resources. Within each time window, the bandwidth resources of the satellite link are subject to random degradation due to the space environment. Therefore, the satellite link... The actual available bandwidth resources within the window are:

[0052]

[0053] in, This indicates the damage caused by the space environment to inter-satellite link bandwidth resources.

[0054] In satellite network slicing, multiple VNRs are embedded in the satellite network to share resources; this process is called VNE. Indicates time window The set of VNRs arriving in the middle. Each VNR is modeled as an undirected graph. ,in It is a set of virtual nodes. It is a virtual link set. Each virtual node is represented as ,need Computing resources. Each virtual link is represented as ,need Bandwidth resources.

[0055] Virtual network mapping consists of two key parts: node mapping and link mapping. We use variables... Indicates node mapping decision:

[0056]

[0057] Each virtual node in the virtual network can only be mapped to one satellite node. Therefore, we have:

[0058]

[0059] The computing resources required by the virtual node must be within the range of the available computing resources of the satellite node to be mapped. Therefore, we have:

[0060]

[0061] Link mapping embeds each virtual link into a loop-free path within the satellite network. We use variables... To represent link mapping decisions:

[0062]

[0063] Virtual nodes at both ends of a virtual link may be mapped to non-adjacent satellite nodes, so a virtual link may be mapped to multiple satellite links.

[0064]

[0065] This ensures that every virtual link At least one link in the satellite network is mapped, thus ensuring connectivity. Similarly, the bandwidth allocated to the virtual link is always within the available bandwidth of the satellite link, while also taking into account potential damage from the space environment. This can be expressed as...

[0066]

[0067] Then, a virtual network embedding optimization problem is proposed to maximize long-term system utility, where system utility consists of the virtual network request reception rate and the revenue-cost ratio. Embedding revenue refers to the benefit gained from successfully embedding a VNR, measured by the total virtual computation and bandwidth resources required for the VNR request. The profit can be expressed as:

[0068]

[0069] Embedding cost quantifies the resource consumption required to embed VNR into a satellite network, and it is measured by the amount of physical satellite network resources used to house the virtual network. The cost can be described as:

[0070]

[0071] in Indicates virtual link The number of hops in the path. The benefit-cost ratio is a key performance indicator reflecting the efficiency of the virtual node embedding process; it represents the benefit generated per unit of cost. The benefit-cost ratio is calculated as follows:

[0072]

[0073] in It is a time window The total number of VNRs arriving within the country.

[0074] VNR reception rate is a metric that measures the success rate of VNE algorithm in embedding VNRs. It is defined as the ratio of the number of successfully embedded VNRs to the total number of VNRs arriving within a given time window. (Time window) The VNR receiver rate can be expressed as:

[0075]

[0076] therefore The VNE problem within a window can be described as follows:

[0077]

[0078] This problem is a long-running robust combinatorial optimization problem. The coupled decisions regarding node and link mappings jointly determine the performance of the VNE. In ultra-large-scale LEO satellite networks, solving this problem involves high computational complexity, and the uncertainty of inter-satellite link bandwidth resources further increases the complexity. Therefore, we propose an Adaptive Noise Mask Deep deterministic policy gradient and Bandwidth Prioritized shortest path (ANMD-BP) algorithm to solve this optimization problem. The algorithm flow is as follows: Figure 2 As shown.

[0079] The Virtual Network Request (VNR) problem is decomposed into two sub-problems: node mapping and link mapping. The node mapping sub-problem is solved using an adaptive noise mask depth-deterministic policy gradient algorithm. We combine a grouping and clustering management architecture for VNR mapping. MEO satellites assess the remaining resources of each group and map incoming VNRs to the cluster with the highest average available resources. The cluster head satellite agent collects state data and makes VNR mapping decisions, assigning virtual nodes to satellite nodes. Then, the environment returns a reward based on the action taken, and the system enters the next state.

[0080] Action: The set of satellites within the cluster to be mapped is represented as At each time step ,action Used to select a satellite node This maps the virtual node to this node. Therefore, we have:

[0081]

[0082] Status: Cluster head satellites at each time step Acquire environmental status data, time step The time window of time is The environmental status data includes the following key parameters: each satellite node in the cluster. Available computing resources and bandwidth resources The current VNR contains a set of satellite node indexes that have already mapped virtual nodes. Current time Computing resources requested by virtual nodes and bandwidth resources And the number of unmapped virtual nodes in the current VNR. A satellite node's bandwidth resources are the sum of the bandwidth resources of its connected satellite links. Therefore, we can consider time... The state at time is defined as:

[0083]

[0084] Rewards: Rewards are determined by the success rate of VNR embedding and the cost-benefit ratio. Specifically, the rewards... Reflecting the result of the mapping decision, indicating whether the virtual node mapping was successful, it can be described as:

[0085]

[0086] in This represents the number of unmapped nodes in the current VNR. This represents the total number of nodes in the current VNR. This is because later nodes typically have fewer embedding options and available resources compared to earlier nodes, making them more critical. Therefore, they should be given greater weight to reflect their importance in the mapping process. Reward Rewards related to the benefit-cost ratio can be described as:

[0087]

[0088] in and They represent the actions to be performed. The corresponding revenue and costs. Since the revenue-cost ratio is primarily related to link mapping, it is not necessary to calculate the revenue-cost ratio when node mapping fails; therefore, the total revenue... This can be expressed as:

[0089]

[0090] To optimize the node mapping problem and efficiently map virtual nodes to satellite nodes, we employ the Deep Deterministic Policy Gradient (DDPG) algorithm, further incorporating an adaptive noise masking operation to better suit our problem. DDPG uses two neural networks: an actor and a critic. The actor is responsible for selecting actions based on the state, while the critic evaluates the quality of the selected actions. The goal of the actor network is to maximize the following expectation:

[0091]

[0092] Indicates actor In a given state Expected action of the next choice value. It is the policy function of the actor network. These are parameters of the actor network. These are the parameters of the critic network. Gradient update term. In essence, it helps actors move towards increasing expectations. The value direction update strategy corresponds to choosing the action that leads to a higher reward. It can be described as:

[0093]

[0094] The parameters of the actor network are updated using the gradient ascent method:

[0095]

[0096] in It is the learning rate of the actor network.

[0097] DDPG uses a critique network to estimate The value function, the role of the critique network, is to guide actor updates by providing value estimates. In DDPG, the critique network updates by minimizing the temporal difference (TD) error, with the goal of making... The difference between the current value and the target value should be as small as possible. Given the current state... ,action ,award and the next state We can obtain the target Q value:

[0098]

[0099] in It is a discount factor. It is the output of the target critic network. It is the output of the target critic network. The goal is to make the current output of the critic network... Get as close to the target as possible Therefore, the following loss function is used to train the critic network:

[0100]

[0101] in These are the parameters of the critique network. Gradient descent is used to update the critique network parameters.

[0102]

[0103] in It is the learning rate of the critic network.

[0104] In the early stages of training, DDPG typically introduces noise into its actions to explore the environment. This noise helps the agent explore various actions in the state space, preventing it from prematurely converging to local optima. As training progresses, the need for exploration gradually decreases, and the agent should rely more and more on its learned policy for decision-making. The noise value is usually gradually reduced through a decay factor, while also being subject to a minimum threshold lower bound. This ensures that the agent retains a degree of randomness after training, thus avoiding over-reliance on the learned policy and maintaining flexibility to adapt to new situations.

[0105] According to the constraints, different virtual nodes within the same VNR must be mapped to different satellites. If the remaining computing resources of the selected satellite are insufficient to meet the computational requirements of the virtual nodes, the mapping will fail. Therefore, we implement a noise masking technique to eliminate these invalid actions. Specifically, we first employ a noise mechanism to generate actual actions by adding noise to the output of the actor network. Simultaneously, if the selected satellite already hosts other virtual nodes within the same VNR or lacks the necessary computing resources, the noise is regenerated. This ensures that the selected node mapping decision satisfies all constraints. We denote the number of attempts to select a valid node mapping operation as . Adaptive noise can be expressed as:

[0106]

[0107] in This is the maximum value of the noise. This ensures that the noise does not become excessive and disrupt the action selection process. As the number of attempts increases, the noise value gradually increases, thereby expanding the range of possible actions. (Constant) This determines the rate at which noise grows with increasing attempts. Therefore, the noise is regenerated and adjusted based on the number of attempts and other constants. This balance between exploration and exploitation allows the agent to strike a balance between exploring new possibilities and utilizing known information.

[0108] Finally, link mapping is performed after node mapping, and the link mapping subproblem is solved using the bandwidth-first shortest path algorithm. We select the shortest path that meets the bandwidth requirements. When calculating the shortest path, we incorporate bandwidth constraints to ensure that only paths with sufficient bandwidth are considered. Next, in the set of shortest paths, we prioritize those paths that provide the most available bandwidth resources. To determine the path with the most sufficient resources, we calculate the minimum bandwidth for each path in the set. The path with the maximum minimum bandwidth is selected as the final link mapping path. This can be represented as...

[0109]

[0110] in Representing a path The minimum bandwidth is selected. This ensures that the chosen path not only meets the bandwidth requirements but also maximizes resource availability on the path, thereby improving the overall performance of the virtual network. Finally, if no shortest path meets the resource requirements, we select an reachable path that meets the minimum bandwidth requirement. This ensures that the mapping process can continue even in resource-constrained environments.

[0111] In this embodiment of the invention, we used the 2493 satellites at the lowest level of Starlink, which were divided into 12 groups, each containing a maximum of 35 satellites. We simulated 10 time windows, each lasting 60 seconds. The main simulation parameter settings are shown in Table 1.

[0112] Table 1 Simulation Parameter Settings

[0113]

[0114] To demonstrate the effectiveness of the proposed ANMD-BP algorithm, we compare it with other benchmark algorithms. DBP: Node mapping uses the DDPG algorithm, only masking satellites already mapped to the current VNR virtual node; link mapping uses the shortest path algorithm. NS: Node mapping uses a node sorting algorithm. In this method, satellite nodes are sorted according to remaining resources, and the node with the most available resources is selected for mapping. Then, the shortest path algorithm is used to perform link mapping. Random: Both node mapping and link mapping are performed randomly.

[0115] We simulated the VNR reception rate, revenue-cost ratio, and total return for several algorithms under different resource impairment rates. As the resource impairment rate increases, the impact on the space environment also becomes more significant, leading to the destruction of more satellite link bandwidth resources. Therefore, in Figure 3 The VNR reception rate decreased for all algorithms. Among all three robustness scenarios, the proposed ANMD-BP algorithm consistently achieved the highest VNR reception rate, demonstrating its superior performance in handling different levels of robustness. While the NS algorithm outperformed the random algorithm, its reception rate was still lower than ANMD-BP. This is because the NS algorithm only focuses on mapping nodes to satellite nodes with the most remaining resources, thus reducing the probability of node mapping failure. In contrast, the random algorithm, due to its arbitrary node and link mapping without considering resource availability or topology, had the significantly lowest reception rate.

[0116] In terms of cost-benefit ratio, the proposed ANMD-BP algorithm... Figure 4It outperforms all other algorithms in all three robustness scenarios. The benefit-cost ratio is closely related to the link mapping algorithm. Specifically, the benefit-cost ratio is often higher when virtual links are mapped to fewer satellite links with fewer hops. The NS algorithm only considers the remaining resources of nodes when mapping, often resulting in virtual links spanning multiple satellite links, leading to a relatively low benefit-cost ratio. On the other hand, these two AI-based algorithms consider the network topology when mapping virtual nodes, achieving a higher benefit-cost ratio. Figure 5 As shown, the ANMD-BP algorithm proposed in this invention achieved the highest total reward in all robust scenarios, demonstrating its adaptability to different robust environments. The total reward is jointly determined by the VNR reception rate and the benefit-cost ratio. As the resource contamination rate increases, the VNR reception rate decreases because the damage to bandwidth resources is greater, leading to a reduction in the total reward for all algorithms.

[0117] This invention also discloses a computer system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites.

[0118] This invention also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites.

[0119] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A robust virtual network embedding method for ultra-dense low-Earth orbit satellite dynamic space environments, characterized in that, Includes the following steps: This paper proposes a virtual network embedding optimization problem aimed at maximizing long-term system utility. System utility is composed of the virtual network request reception rate and the revenue-cost ratio. Constraints include: each virtual node in the virtual network can only be mapped to one satellite node; the computational resources required by a virtual node must be within the available computational resources of the satellite node to be mapped; each virtual link must be mapped to at least one satellite link; and the bandwidth allocated to virtual links must always be within the available bandwidth of the satellite links. The actual available bandwidth resources of the satellite links take into account the damage caused by the space environment to the inter-satellite link bandwidth resources. The optimization problem is decomposed into two sub-problems: node mapping and link mapping. In node mapping, virtual network requests are mapped to the cluster with the highest average available resources. Based on reinforcement learning, an adaptive noise mask depth deterministic policy gradient algorithm is used to map satellites and virtual nodes within a cluster. After node mapping, the bandwidth-first shortest path algorithm is used for link mapping.

2. The robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to claim 1, characterized in that: Satellite link in The actual available bandwidth resources within the window are: ； in, Indicates the first The bandwidth resources owned by each satellite link This indicates the damage caused by the space environment to inter-satellite link bandwidth resources.

3. The robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to claim 1, characterized in that: The virtual network request reception rate is the ratio of successfully embedded virtual network requests to the total number of virtual network requests arriving within a given time window. In the virtual network embedding benefit-cost ratio, the embedding benefit is measured by the total amount of virtual computing and bandwidth resources for virtual network requests, while the embedding cost is measured by the amount of physical satellite network resources used to accommodate the virtual network.

4. The robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to claim 1, characterized in that: The MEO satellite adopts a virtual network mapping management architecture based on grouping and clustering. It assesses the remaining resources of each LEO cluster and maps incoming virtual network requests to the LEO cluster with the highest average available resources.

5. A robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to claim 1, characterized in that: The reinforcement learning action is the selected satellite for node mapping, the state is the available computing and bandwidth resources of each satellite node in the cluster, the set of satellite node indexes that have been mapped in the current virtual network request, the computing and bandwidth resources of the virtual node request at the current time, and the number of unmapped virtual nodes in the current virtual network request. The reward is determined by the success rate of virtual network request embedding and the benefit-cost ratio.

6. The robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to claim 1, characterized in that: Time step Rewards Among them, rewards ,award ; This represents the number of unmapped nodes in the current VNR. This represents the total number of nodes in the current VNR. and They represent the actions to be performed. The corresponding benefits and costs.

7. A robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to claim 1, characterized in that: In the adaptive noise mask depth deterministic policy gradient algorithm, actual actions are generated by adding noise to the output of the actor network. If the selected satellite already carries other virtual nodes requested by the same virtual network or there is a lack of sufficient computing resources, the noise is regenerated. The number of attempts to select a valid node mapping operation is denoted as... Adaptive noise is represented as: ； in, It is the maximum value of the noise, a constant. This determines the rate at which noise increases with the number of attempts.

8. A robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to claim 1, characterized in that: In the bandwidth-first shortest path algorithm, only paths that meet the bandwidth requirements are considered when calculating the shortest path; in the set of shortest paths, the path that provides the most available bandwidth resources is selected first; the minimum bandwidth of each path in the set is calculated, and the path with the maximum minimum bandwidth is selected as the final link mapping path. If no shortest path satisfies the resource requirements, select an reachable path that meets the bandwidth requirements.

9. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of a robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to any one of claims 1-8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of a robust virtual network embedding method for a dynamic space environment of ultra-dense low-Earth orbit satellites according to any one of claims 1-8.

Images