Satellite network service path dynamic recovery method based on damaged state evaluation

Abstract

The invention discloses a satellite network service path dynamic recovery method based on damaged state evaluation, and mainly solves the problem of low task transmission success rate of the existing large-scale satellite network in a complex scene that multiple nodes and multiple links simultaneously fail. The method comprises the following implementation steps of: 1) acquiring node and link state information and service characteristics of a satellite network, and constructing a network topology mapping model; 2) constructing a network damage state evaluation model, and obtaining a network performance degradation index; 3) building a path priority decision model and training to obtain an intelligent agent; 4) outputting the priority probability of the alternative paths in combination with the network state and the service information, sorting the alternative paths according to an agent decision result, and configuring a forwarding table item; and 5) automatically switching to the optimal path when the link fails, and triggering the centralized controller to recompute the path if the switching fails. Efficient service path recovery can be realized, the task transmission success rate is remarkably improved, and the high-performance requirement of the network is ensured.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, and further relates to network service path recovery technology. Specifically, it is a dynamic recovery method for satellite network service paths based on damage status assessment, which can be used to efficiently restore service paths in large-scale satellite networks and improve the success rate of mission transmission. Background Technology

[0002] Large-scale satellite networks are a crucial component of sixth-generation mobile communications, and the complex effects of the space environment are increasingly exacerbating the problem of simultaneous failures of multiple nodes and links. When a satellite link fails, recovery requires consideration of the network's status to improve mission transmission success rates and shorten recovery time. Therefore, a pressing issue is how to quickly and accurately combine network status and service information to dynamically plan and prioritize service paths between different source-destination pairs, enabling rapid switching to available paths in the event of a failure for better recovery quality.

[0003] In patent application CN103986512A, entitled "A Method for Inter-Satellite Link Fault Recovery Based on Region Division in Satellite Networks," the link failure probability is added to the path cost formula to avoid links with high failure rates. When a link failure occurs in the satellite network, rerouting is performed within a small area. If no suitable alternative path is available, a flooding method with constraints is used to find a suitable forwarding path for the service. While this method improves recovery quality to some extent, rerouting is required when a failure occurs in each area. In cross-domain systems, due to the large number of nodes, the simultaneous failure of multiple nodes and links can easily lead to high computational overhead and low timeliness in rerouting. Zhejiang University, in its patent application CN103428031A, proposed a fast inter-domain link fault recovery method based on software-defined networking. This method selects a path to bypass the failed link by storing an adjacency node table locally on the OpenFlow switch, then finds the path with the shortest hop count in the adjacency node table as the optimal path. When the hop counts are the same, a path is randomly selected as the optimal path. However, this scheme only considers the number of hops and does not comprehensively consider the current network status and service requirements. In the cross-domain collaborative satellite constellation system, when multiple nodes and multiple links fail simultaneously, it will cause load imbalance and network congestion, resulting in a low success rate of task transmission. Patent application document CN117394904A proposes a link failure recovery method and system for large-scale domain-division satellite networks. It involves an improved link failure recovery method based on the SRv6 mechanism. It configures inter-domain backup paths according to the initial path and uses a preventive method to configure intra-domain backup paths. When a link fails, the failed node notifies the local controller to plan a new path for forwarding and restoring service data. However, this method is not applicable to the recovery of multiple nodes and multiple links failing simultaneously, which may lead to overload of the local controller and failure to calculate a new path in time to complete the recovery process.

[0004] Existing service recovery methods, such as traditional rerouting or alternative path methods, often lead to network congestion in large-scale satellite networks when multiple nodes and links fail simultaneously. This results in long recovery times and low mission transmission success rates, failing to guarantee the high-performance requirements of large-scale satellite networks. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a dynamic recovery method for satellite network service paths based on damage status assessment. This method aims to solve the problem of low mission transmission success rates in complex scenarios where multiple nodes and links fail simultaneously in large-scale satellite networks. First, the node and link status information and service characteristics of the satellite network are acquired to construct a network topology mapping model. A network damage status assessment model is then built to quantify the Network Performance Degradation Index (NPDI). Next, a path priority decision model is constructed, and a reinforcement learning agent is used to construct service recovery as a sequential decision problem with the objective of minimizing NPDI. The priority probabilities of alternative paths are output based on network status and service information. Then, the alternative paths are ranked and forwarding table entries are configured according to the agent's decision results. In the event of a link failure, the system automatically switches to the optimal path. If the autonomous decision fails, a centralized controller is triggered to recalculate the path. This invention can achieve efficient service path recovery in complex failure scenarios involving multiple nodes and links, significantly improving mission transmission success rates and ensuring the high-performance requirements of the network.

[0006] To achieve the above objectives, the technical solution of the present invention includes the following:

[0007] (1) Obtain the node and link status information and service characteristics of the satellite network, extract the network topology map, the set of service requests to be restored and the routing table of alternative transmission paths pre-planned for each service, and construct the network topology mapping model;

[0008] (2) The three indicators of end-to-end throughput, end-to-end average latency and network topology connectivity are normalized and weighted and fused together to construct a network damage status assessment model and obtain a network performance degradation index that quantifies the degree of network damage.

[0009] (3) Build a path priority decision model, construct the service recovery as a Markov decision process, and train the model with the goal of minimizing the network performance degradation index to obtain the service recovery agent;

[0010] (4) Using the service recovery agent, the probability of each alternative path being selected and the probability of executing the route reconstruction RESET action are calculated based on the real-time network status and service information; and the path with the highest probability of being selected among the alternative paths is taken as the optimal alternative path.

[0011] (5) In each service's alternative paths, delete the alternative paths whose probability is less than the probability of RESET action, and then sort the remaining alternative paths in descending order according to their selection probability; and send priority forwarding table entries to the switches through which the path passes according to the sorting results, and add RESET action at the end of the table entry.

[0012] (6) When a link fails, the network automatically switches to the optimal alternative path according to the priority of the forwarding table entries to achieve autonomous decision-making recovery; if the recovery fails, a RESET action is triggered, and the controller centrally recalculates the path and sends it out, ultimately completing the dynamic recovery of the satellite network service path.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] First, this invention constructs a network damage status assessment model. By normalizing and weighting three key indicators—end-to-end throughput, end-to-end average latency, and network topology connectivity—the goal of service recovery is no longer limited to a single hop count or link status. Instead, it can comprehensively and quantitatively assess the overall performance degradation and health status of the network. This allows for a balance between load balancing and network connectivity during path planning, effectively avoiding secondary congestion caused by recovery and ensuring that the network still has high throughput and low latency after fault recovery.

[0015] Secondly, because this invention adopts a path priority decision model based on reinforcement learning, it constructs service recovery as a sequential decision problem and uses an agent to output the priority probability of alternative paths based on the network damage status assessment model. This enables the recovery decision to consider the correlation between the preceding and following steps and dynamically perceive the impact of the current service path selection on other subsequent services and the global network status. This overcomes the shortcomings of traditional methods that cannot dynamically plan in combination with the actual network status. Thus, in the complex scenario of multiple nodes and multiple links failing simultaneously in large-scale satellite networks, it can avoid local optima and significantly reduce the number of service recovery failures. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the application process of the present invention;

[0017] Figure 2 This is a flowchart illustrating the overall implementation of the method of the present invention;

[0018] Figure 3 A flowchart illustrating the service recovery process performed by the service recovery agent in this invention;

[0019] Figure 4 This is a graph showing the comparison of the number of service recovery failures between the present invention and existing methods. Detailed Implementation

[0020] The present invention will now be further described with reference to the accompanying drawings.

[0021] Example 1: Refer to Appendix Figures 1-2 The present invention proposes a method for dynamic recovery of satellite network service paths based on damage status assessment, which specifically includes the following steps:

[0022] Step 1) Obtain the node and link status information and service characteristics of the satellite network, extract the network topology map, the set of service requests to be restored, and the routing table of alternative transmission paths pre-planned for each service, and construct the network topology mapping model.

[0023] This embodiment targets large-scale satellite networks (including multi-layered constellations such as low Earth orbit, medium Earth orbit, and high Earth orbit), acquiring real-time network node and link status information, and extracting and abstracting feature parameters for modeling. The construction process of the network topology mapping model includes: assuming the large-scale satellite network consists of... It consists of several satellite nodes, which are connected by inter-satellite laser links; the entire satellite network is modeled as a topology graph. ;in Represents a set of satellite nodes. It is a collection of inter-satellite laser links. Indicates the starting node of the connection. and the endpoint Communication links, This indicates the physical available capacity of the link; it also involves real-time monitoring and extraction of service requests awaiting recovery within the network, resulting in a set of such requests. ,in , , They represent the first The source node, destination node, and bandwidth requirements of each service; for each satellite node pair in the network, obtain a pre-planned set of alternative transmission paths. This refers to the routing table containing the set of alternative transmission paths, where each path consists of several links connected in series. During subsequent recovery, the system will make dynamic planning decisions based on the extracted G, D, and P feature data, adhering to the constraint that the service load on any link does not exceed its remaining capacity.

[0024] Step 2) Normalize the three metrics of end-to-end throughput, end-to-end average latency, and network topology connectivity, and then perform weighted fusion based on their respective weights to construct a network damage assessment model, obtaining a network performance degradation index that quantifies the degree of network damage, as follows:

[0025] (2.1) Select the network's end-to-end throughput, end-to-end average latency, and topological connectivity as evaluation indicators and normalize them to obtain their normalized results. , and ;

[0026] (2.2) Define the network performance degradation index to characterize the degree of network damage. :

[0027] ,

[0028] in , , These represent the weights of end-to-end throughput, end-to-end average latency, and network topology connectivity, respectively.

[0029] Step 3) Build a path priority decision model, construct service recovery as a Markov decision process, and train the model to obtain the service recovery agent with the objective of minimizing the network performance degradation exponent. Specifically, this is achieved through the following steps:

[0030] (3.1) Construct a path priority decision model, and construct the business recovery as a Markov decision process, setting the action space, state space and reward function; the construction process of the path priority decision model in this embodiment specifically includes: establishing a Markov decision process quintuple. ,in , , These represent the action space, state space, and reward function, respectively. As a discount factor, Represent the transition matrix; represent the state space Reconstructed into a throughput correlation matrix End-to-end average delay correlation matrix and network connectivity correlation matrix The combination of the three correlation matrices, where The non-zero elements are composed of the product of the remaining link capacity and the normalized throughput. The zero element in the middle and non-zero elements is composed of time delay. The zero element in the middle represents the link connectivity state; the action space is defined. for 3D discrete vector ,in The number of alternative paths, to This represents the probability that each alternative path will be selected. Indicate the probability that the "RESET" action is selected; design the reward function. as follows:

[0031] .

[0032] in Indicates the reward weight. Indicates end-to-end throughput weight. Indicates the end-to-end average delay weight. This represents the topological flux weight of the network. Indicate alternative paths The corresponding path has been successfully restored. Indicate alternative paths The corresponding path recovery failed.

[0033] (3.2) The path priority decision model is trained by using the near-end policy optimization algorithm, so that it learns the mapping strategy from the state space to the action space to maximize the cumulative reward. The trained model is used as the business recovery agent.

[0034] Step 4) Using the service recovery agent, the probability of each alternative path being selected and the probability of executing the route reconstruction RESET action are calculated based on the real-time network status and service information. The path with the highest probability of being selected among the alternative paths is taken as the optimal alternative path. In this embodiment, the route reconstruction RESET action in this step specifically refers to generating a routing table of the set of alternative transmission paths pre-planned for each service as described in step 1) based on the current network status.

[0035] In this embodiment, the above-mentioned service recovery agent calculates the probability of each alternative path for each service being selected and the probability of executing the route reconstruction (RESET) action based on the real-time network status and service information. The implementation steps include:

[0036] (4.1) Initialize the queue of pending business processes ;

[0037] (4.2) Determine the current queue If the value is not empty, then for the current service request in the queue, the real-time monitored network status information and the flow characteristic information of the current service in the queue will be used as input to the service recovery agent.

[0038] (4.3) The agent outputs a probability distribution vector corresponding to the action space based on the input information. This vector contains the probability distribution vector of the action space selected by the current service. The probability of alternative paths And the probability of choosing the RESET action. And the sum of the probabilities of all actions is 1.

[0039] Step 5) In each service's alternative paths, delete the alternative paths whose probability is less than the probability of a RESET action, and then sort the remaining alternative paths in descending order of their selection probability; in this embodiment, this is specifically done by comparing the selection probability of each alternative path. The probability of the RESET action The conditions will be met. The candidate paths are removed from the set, and the remaining candidate paths are then sorted in descending order according to their respective selection probabilities to obtain the set of paths to be processed. This refers to the sorting result. Then, based on the sorting result, priority-bearing forwarding entries are sent to the switches along the path, and a RESET action is added to the end of each entry. In this embodiment, this is specifically achieved by traversing the set of paths to be processed. For each path in the routing table, forwarding entries with corresponding priorities are sent to all switches along that path, forming a set of alternative transmission paths. Simultaneously, for the current service... Configure RESET flow table entries with priority 0 for all switches that need to be traversed.

[0040] Step 6) When a link fails, the network automatically switches to the optimal alternative path based on the priority of the forwarding table entries to achieve autonomous decision-making recovery; if the recovery fails, a RESET action is triggered, and the controller centrally recalculates and distributes the path, ultimately completing the dynamic recovery of the satellite network service path.

[0041] In this embodiment, the aforementioned autonomous decision-making recovery refers to the automatic switching of service traffic to the currently available optimal alternative path for transmission based on the priority of forwarding table entries when a link failure occurs in the satellite network. If the recovery fails, a RESET action is triggered. Specifically, when all alternative paths for the current service fail, a RESET flow table entry with a priority of 0 is matched, and a RESET message is sent to the controller. After receiving the message, the controller determines that the autonomous decision-making recovery has failed, deletes the old flow table entry corresponding to the service, recalculates the set of available paths for the service, and then adds the service back to the queue of pending services, returning to step 4) until the service path recovery is completed.

[0042] Example 2: Refer to Appendix Figures 1-2 The satellite network service path dynamic recovery method provided in this embodiment is implemented in the same way as in Embodiment 1. The implementation process of constructing a network damage status assessment model to obtain the network performance degradation index is described in further detail below:

[0043] To quantify the extent of network damage, this invention constructs a network damage assessment model to obtain the Network Performance Degradation Index (NPDI). By normalizing and weighting throughput, latency, and connectivity, the degree of network performance degradation can be assessed, providing a core decision-making basis for subsequent service recovery aimed at minimizing NPDI.

[0044] First, the three metrics of network end-to-end throughput, network end-to-end average latency, and network topology connectivity are normalized to resolve the issue of dimensionality.

[0045] network end-to-end throughput The larger the value, the healthier the network (positive indicator).

[0046] ,

[0047] in, This represents the end-to-end throughput of the network after the damage. This represents the initial end-to-end throughput.

[0048] Network end-to-end average latency The smaller the value, the healthier the network (a negative indicator that needs to be positively correlated).

[0049] ,

[0050] in, This represents the average end-to-end latency of the network after the damage. This represents the initial end-to-end average delay.

[0051] Network topological connectivity The larger the value, the healthier the network (positive indicator).

[0052] ,

[0053] in, This represents the network topology connectivity after damage. This represents the initial network topology connectivity; after normalization, all metrics are within... Interval.

[0054] The Network Performance Degradation Index (NPDI) is negatively correlated with all three normalized metrics. A higher NPDI value indicates a more severe loss of network performance and worse overall performance.

[0055] ,

[0056] in, , The smaller the value, the healthier the network. The larger the value, the worse the network performance. Indicates end-to-end throughput weight. Indicates the end-to-end average delay weight. This represents the topological throughput weight of the network.

[0057] It provides a powerful tool for unified quantification of throughput, latency, and connectivity, clearly and intuitively reflecting the overall impairment status of satellite networks. It serves as a key comprehensive indicator for network monitoring, alarming, diagnosis, and resource scheduling. Through the network impairment status assessment model, the degree of performance degradation of satellite networks can be evaluated in real time, minimizing [the impact of network degradation]. To restore business operations to the target.

[0058] Example 3: Refer to Appendix Figures 1-2 The satellite network service path dynamic recovery method provided in this embodiment is implemented in the same way as in Embodiment 1. The implementation process of constructing the path priority decision model and training the service recovery agent is described in further detail below:

[0059] First, the problem that service recovery solves is essentially a network resource allocation optimization problem: by controlling the priority of paths, network resources (link capacity) are efficiently allocated to different services when a fault occurs, thereby maximizing the training objectives (recovery success rate, recovery latency, recovery fault severity, etc.).

[0060] In the path priority decision model constructed in this invention, the main strategy to be learned is to complete the mapping from network state and service information to path priority. Therefore, the problem is described as a complete Markov Decision Process (MDP), which consists of a quintuple. Decision, among which , , These represent the action space, state space, and reward function, respectively. This embodiment uses the discount factor... Set the value to 0.99, and the state transition matrix is... In model-free reinforcement learning, the state space, action space, and reward function are generally learned by the agent itself (without explicit explanation). The following sections will introduce the specific design details of these elements:

[0061] (3a) State space:

[0062] The state that the agent can observe from the environment contains two key pieces of information that directly influence NPDI decisions: first, the global network health status directly related to NPDI, including topology connectivity, capacity status, and latency status; and second, the characteristics of the service to be decided, including required bandwidth, latency and capacity attributes of all alternative paths, and the potential impact of path selection on topology connectivity. Leveraging the centralized control capabilities of the control center, this information is acquired and updated in real time through periodic probing. To adapt to deep neural network processing, this invention reconstructs the state space into a combination of three correlation matrices. The three-dimensional features of NPDI are simultaneously embedded through a bivariate (business-link) mapping relationship, where The incidence matrix representing throughput, where non-zero elements are defined as... The potential contribution of the quantification link to global throughput. The correlation matrix represents the average end-to-end delay, with non-zero elements representing the transmission delay after link load weighting. The correlation matrix represents network connectivity, with non-zero elements indicating that a link is connected. This processing aggregates the relationships between links and alternative paths, encompassing various network states, and has been shown to be efficiently processed by deep neural networks, thus providing a prerequisite for reinforcement learning decision-making.

[0063] (3b) Action space:

[0064] The agent needs to prioritize all paths for each task sequentially to maximize the cumulative reward. The action space is defined as a k+1 dimensional discrete space, where k represents the number of candidate paths. Therefore, the action space takes the form of a vector: After each decision, the agent outputs the probability of different actions being selected, with the sum of all action probabilities equal to 1. The reinforcement learning output action probability is based on the action's contribution to the cumulative reward. If the goal of this invention is to improve the recovery success rate, then actions with higher probabilities indicate that using the path corresponding to that action to complete the service transmission maximizes the overall recovery success rate of all services, and so on, with the action with the lowest probability corresponding to the worst effect. Therefore, when determining the priority of candidate paths for each service, this invention only needs to sort them based on the action probability output values ​​of the reinforcement learning agent, thus enabling the selection of available paths with a high contribution to the overall goal to continue service transmission during failover.

[0065] The k+1 dimension action represents "RESET". When all alternative paths fail to complete fault recovery, the service will be handed over to the controller to complete centralized fault recovery. This action is generally the lowest priority action because it is the last resort recovery method. However, there may be some cases where the quality of alternative paths is too poor, or even the effect of not going through the control center to perform "RESET". Therefore, the intelligent system will select the centralized recovery method at the appropriate time to complete the backup work of fault recovery.

[0066] (3c) Reward function:

[0067] The design of the reward function is a key factor in the effectiveness of reinforcement learning algorithms and whether their learning performance meets expectations. It also affects the convergence speed of the algorithm. A suitable reward function should reflect the characteristics of the task, guide the agent to learn the correct policy quickly and effectively, and also have a relatively flat distribution to accelerate the convergence speed of the reinforcement learning algorithm. Reward value It is a numerical signal fed back to the agent from the environment, which reflects the value of the action chosen by the agent.

[0068] ;

[0069] Meanwhile, different reward function designs correspond to different fault recovery optimization objectives. This invention designs a reward function with the objective of effectively minimizing the key factors of NPDI. If the agent selects a path that cannot be successfully transmitted for the current service, the reward value for this action is a negative number. Conversely, when the agent's decision enables the service to be successfully transmitted, a positive value is always returned, the magnitude of which depends on the impact of the path chosen by the agent on the performance of the network after recovery. The "RESET" action also always yields a negative value because centralized control recovery, as a backup scheme for self-decision recovery, has a lower priority. If "RESET" can successfully calculate an available path, then a reward value is returned according to the path quality index.

[0070] This invention aims to minimize NPDI (Network Point Indicator) and configures different priorities for alternative paths based on network status and service information. This achieves the effect of selecting the optimal path for each service, thereby completing service transmission by quickly switching to a suitable path when a link fails, reducing the number of task transmission failures and improving the service recovery success rate.

[0071] Based on the network damage assessment model, this invention focuses on the network quality after fault recovery, including the restored end-to-end throughput, restored end-to-end average latency, restored connectivity, and restoration success rate. All labels are shown in Table 1.

[0072] Table 1: Service Recovery Symbol Marking Table

[0073]

[0074] Given a set of nodes, links with available capacity, and the amount of traffic between nodes, a traffic allocation is feasible if it satisfies the constraints of demand and link capacity. The training objective is to find a feasible traffic allocation that optimizes the given objective. In real-world large-scale satellite network environments, traffic is confined to specified paths. Using the notation in Table 1, in the traffic allocation optimization problem on a pre-configured set of paths, this invention redefines the optimization objective as minimizing the network performance degradation index, which comprehensively characterizes the overall health of the satellite network. The optimization model is defined as follows:

[0075] ,

[0076] ,

[0077] ,

[0078] ,

[0079] ,

[0080] In the above formula, The variable represents whether path p is selected for the k-th service request. This constraint ensures that each service request selects only one path from the candidate path set for transmission. This represents the bandwidth allocated to request k during fault recovery, which must be no less than the required bandwidth. This is to ensure the basic transmission needs of the business are met. Indicates link The total traffic carried on the link must not exceed the link's physical capacity. This is to prevent link congestion. The optimization objective NPDI is a comprehensive index with a value range of [0,1]. The smaller the value, the better the network health. This index is composed of the normalized values ​​of three key performance indicators of the fault assessment model. This model not only considers link load balancing, but also optimizes the three key dimensions of throughput, latency, and connectivity, making the assessment of network damage status after fault recovery more comprehensive.

[0081] By formalizing the description of this problem, we can conclude that the service recovery method based on real-time dynamic planning of path priorities according to network status can pre-configure the priority of the main path for different services. Furthermore, the configuration of primary and backup priorities can be combined with network information and service characteristics to minimize the degree of failure, thereby optimizing the overall network service quality after failure recovery.

[0082] Example 4: Refer to Appendix Figures 3-4 The overall implementation steps of the satellite network service path dynamic recovery method proposed in this invention are the same as those in Embodiment 1, and are now referred to the appendix. Figure 3 A further detailed description of the specific process steps for recovery via the business recovery agent, including:

[0083] Step 1. Define the state G, the service matrix T, and the number of candidate paths k; then complete the initialization operation, including initializing the service queue F, and generating a set of candidate paths for all node pairs in the network based on the shortest path. ;

[0084] Step 2. Determine the current business queue Is it empty? If it is empty, proceed to step 3; otherwise, proceed to step 7.

[0085] Step 3. Based on the intelligent agent, combined with the flow information and network status, calculate the probability of each alternative path of the service being selected and the probability of the "RESET" action;

[0086] Step 4. Delete paths with a selection probability less than "RESET" from the candidate paths (these paths contribute less to the recovery than recalculated paths), and then sort the remaining paths in descending order of selection probability to obtain the set of paths to be processed. , ;

[0087] Step 5. Traverse each path and send priority-assigned forwarding entries to all switches along that path; simultaneously, for Configure a "RESET" flow table entry with a priority of 0 for all switches that may forward the data (to mark a scenario where the entire path fails).

[0088] Step 6. Remove the current service from queue F and return to step 2 to continue processing the next service;

[0089] Step 7. Determine if a "RESET" message has been received. If no "RESET" message has been received, it means that no service needs to be restored, and proceed to step 9. If a "RESET" message has been received, it means that all current service forwarding paths have failed, further triggering the centralized recovery method, and proceed to step 8.

[0090] Step 8. The controller centrally processes the data, deletes old entries, recalculates the path set, and adds it to the business queue. Return to step 2.

[0091] Step 9: The process ends automatically.

[0092] In large-scale satellite networks, this invention utilizes a well-trained service recovery agent based on real-time network status assessment results to perform service recovery, thereby addressing the low success rate of fault recovery caused by simultaneous failures of multiple nodes and links. By pre-training an agent that prioritizes alternative paths based on network status and service requirements, and by setting intervention times for a centralized control recovery strategy, the reliability of fault recovery is ensured. Through this fault recovery method, since the path priority determined by the agent incorporates network status information, it effectively improves the success rate of service recovery even when multiple nodes and links fail simultaneously.

[0093] In practical applications, a business recovery agent is pre-trained. Self-decision-making is the primary method for fault recovery, while centralized control calculates the recovery path through the controller when the self-decision-making method fails. This intelligent agent is first utilized... Based on actual network characteristics and service information, the system outputs the priorities of alternative paths for all services in the network, as well as the priority of "RESET" relative to the alternative paths. Then, it configures all paths in the data plane and sets the priority fields corresponding to the completed flow table entries. When a fault occurs and autonomous recovery fails, the controller performs fault recovery. In this case, it can quickly calculate a suitable path to meet the rerouting requirements of the service, with a short processing time. It can also recalculate a new set of alternative paths, utilize the agent to determine the priorities, and then proceed with recovery.

[0094] The effects of the present invention will be further explained below with reference to simulation experiments.

[0095] Existing service recovery methods, such as traditional rerouting or alternative path methods, can lead to network congestion in large-scale satellite networks when multiple nodes and links fail simultaneously. This can result in excessively long response times and low mission transmission success rates, failing to guarantee the communication needs of large-scale satellite networks.

[0096] To compare the efficiency of the method of the present invention, the simulation parameters are set as shown in Table 2 below.

[0097] Table 2 Simulation Parameters

[0098]

[0099] To verify the effectiveness of the service recovery method of this invention compared with traditional service recovery methods (OF-FF, CASA), a simulation comparison experiment was conducted to simulate link failures and service recovery for the above parameters. Figure 4 As shown, the method of the present invention has significant effects in reducing the number of services that fail to recover after a fault and shortening the recovery time. The X-axis represents the number of links that have failed, and the Y-axis represents the number of services that fail to recover after a fault, i.e., the lower the better. Figure 4 The study neglects cases with fewer than 20 link failures, because all three recovery methods can achieve good recovery for such cases, with almost zero service failures during recovery, making them unsuitable for comparison. Figure 4 As can be seen, the OF-FF method, which only switches alternative paths, recovers the most services that failed. This is because the OF-FF method cannot dynamically plan according to the actual network conditions. The CASA method has a relatively better recovery effect. However, the method proposed in this invention has a significantly improved recovery effect compared to other methods. The main reason is that this invention can combine the actual network status and service information for real-time dynamic planning. Furthermore, during the planning process, it can comprehensively consider the impact of different service path selections on the overall network status, aiming to minimize the degree of network failure. This is something that the CASA algorithm cannot achieve.

[0100] In summary, the "Dynamic Recovery Method for Satellite Network Service Path Based on Damage Status Assessment" proposed in this invention achieves the function of dynamically planning paths to complete service recovery by combining the real-time status of the network and service requirements. Compared with traditional service recovery methods, it can achieve a lower number of service recovery failures.

[0101] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0102] The above simulation analysis proves the correctness and effectiveness of the method proposed in this invention.

[0103] The parts of this invention not described in detail are common knowledge to those skilled in the art.

[0104] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Obviously, those skilled in the art, after understanding the content and principles of the present invention, may make various modifications and changes in form and detail without departing from the principles and structure of the present invention. These modifications and changes include: constructing the path priority decision problem as a Markov model and, through a training algorithm, enabling the pre-training of an agent that combines network state and business requirement information to make priority decisions on alternative paths; constructing the fault assessment model as a multi-dimensional evaluation index and converting the damage judgment condition into a reward function for calculating priority decision. These modifications and changes based on the ideas of the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A method for dynamic recovery of satellite network service paths based on damaged state assessment, characterized in that, Includes the following steps: (1) Obtain the node and link status information and service characteristics of the satellite network, extract the network topology map, the set of service requests to be restored and the routing table of alternative transmission paths pre-planned for each service, and construct the network topology mapping model; (2) The three indicators of end-to-end throughput, end-to-end average latency and network topology connectivity are normalized and weighted and fused together to construct a network damage status assessment model and obtain a network performance degradation index that quantifies the degree of network damage. (3) Build a path priority decision model, construct the service recovery as a Markov decision process, and train the model with the goal of minimizing the network performance degradation index to obtain the service recovery agent; (4) Using the service recovery agent, the probability of each alternative path being selected and the probability of executing the route reconstruction RESET action are calculated based on the real-time network status and service information. The path with the highest probability of being selected from the alternative paths is taken as the optimal alternative path. (5) In each service's alternative paths, delete the alternative paths whose probability is less than the probability of RESET action, and then sort the remaining alternative paths in descending order according to their selection probability; and send priority forwarding table entries to the switches through which the path passes according to the sorting results, and add RESET action at the end of the table entry. (6) When a link fails, the network automatically switches to the optimal alternative path according to the priority of the forwarding table entries to achieve autonomous decision-making recovery; if the recovery fails, a RESET action is triggered, and the controller centrally recalculates the path and sends it out, ultimately completing the dynamic recovery of the satellite network service path.

2. The method according to claim 1, characterized in that: The network topology mapping model described in step (1) is constructed by: assuming that a large-scale satellite network is composed of... It consists of several satellite nodes, which are connected by inter-satellite laser links; the entire satellite network is modeled as a topology graph. ;in Represents a set of satellite nodes. It is a collection of inter-satellite laser links. Indicates the starting node of the connection. and the endpoint Communication links, Indicates the physical available capacity of the link; let This represents the set of service requests that need to be restored within the network. ,in , , They represent the first The source node, destination node, and bandwidth requirements of each service; This represents a pre-planned set of alternative transmission paths in the routing table, where each path... It consists of multiple links connected in series.

3. The method according to claim 1, characterized in that: The network damage assessment model described in step (2) is constructed according to the following steps: (2.1) Select the network's end-to-end throughput, end-to-end average latency, and topological connectivity as evaluation indicators and normalize them to obtain their normalized results. , and : ， ， ； in, This represents the end-to-end throughput of the network after the damage. This represents the initial end-to-end throughput; This represents the average end-to-end latency of the network after the damage. This represents the initial end-to-end average delay; This represents the network topology connectivity after damage. This represents the initial network topology connectivity. (2.2) Define the network performance degradation index to characterize the degree of network damage. : ， in , , These represent the weights of end-to-end throughput, end-to-end average latency, and network topology connectivity, respectively.

4. The method according to claim 1, characterized in that: The service recovery agent mentioned in step (3) is obtained according to the following steps: (3.1) Build a path priority decision model, construct the business recovery as a Markov decision process, and set the action space, state space and reward function; (3.2) The path priority decision model is trained by using the near-end policy optimization algorithm, so that it learns the mapping strategy from the state space to the action space to maximize the cumulative reward. The trained model is used as the business recovery agent.

5. The method according to claim 4, characterized in that: The path priority decision model described in step (3.1) is constructed as follows: This includes: establishing a Markov decision process quintuple. ,in , , Let these represent the action space, state space, and reward function, respectively. As a discount factor, Represent the transition matrix; represent the state space Reconstructed into a throughput correlation matrix End-to-end average delay correlation matrix and network connectivity correlation matrix The combination of the three correlation matrices, where The non-zero elements are composed of the product of the remaining link capacity and the normalized throughput. The zero element in the middle and non-zero elements is composed of time delay. The zero element in the middle represents the link connectivity state; the action space is defined. for 3D discrete vector ,in The number of alternative paths, to This represents the probability that each alternative path will be selected. Represent the probability of the "RESET" action being selected; design the reward function. as follows: ； in Indicates the reward weight. Indicates end-to-end throughput weight. Indicates the end-to-end average delay weight. This represents the topological flux weight of the network. Indicate alternative paths The corresponding path has been successfully restored. Indicate alternative paths The corresponding path recovery failed.

6. The method according to claim 1, characterized in that: The RESET action in step (4) specifically refers to generating a routing table of alternative transmission paths pre-planned for each service in step (1) based on the current network status.

7. The method according to claim 1 or 6, characterized in that: Step (4) utilizes a service recovery agent to calculate the probability of each alternative path for each service being selected and the probability of executing a route reconstruction (RESET) action based on real-time network status and service information. The implementation steps include: (4.1) Initialize the queue of pending business processes ; (4.2) Determine the current queue If the value is not empty, then for the current service request in the queue, the real-time monitored network status information and the flow characteristic information of the current service in the queue will be used as input to the service recovery agent. (4.3) The agent outputs a probability distribution vector corresponding to the action space based on the input information. This vector contains the probability distribution vector of the action space selected by the current service. The probability of alternative paths And the probability of choosing the RESET action. And the sum of the probabilities of all actions is 1.

8. The method according to claim 1, characterized in that: In step (5), among the candidate paths for each service, candidate paths with a probability less than the probability of a RESET action are deleted. Then, the remaining candidate paths are sorted in descending order according to their selection probability. Specifically, this is done by comparing the selection probability of each candidate path. The probability of the RESET action The conditions will be met. The candidate paths are removed from the set, and the remaining candidate paths are then sorted in descending order according to their respective selection probabilities to obtain the set of paths to be processed. That is, the sorting result.

9. The method according to claim 8, characterized in that: Step (5) involves sending priority forwarding entries to the switches along the path based on the sorting results, and adding a RESET action at the end of each entry. Specifically, this is done by traversing the set of paths to be processed. For each path in the routing table, forwarding entries with corresponding priorities are sent to all switches along that path, forming a set of alternative transmission paths. Simultaneously, for the current service... Configure RESET flow table entries with priority 0 for all switches that need to be traversed.

10. The method according to claim 9, characterized in that: The autonomous decision recovery mentioned in step (6) refers to the automatic switching of service traffic to the currently available optimal alternative path for transmission based on the priority of the forwarding table entries when a link failure occurs in the satellite network. If the recovery fails, a RESET action is triggered. Specifically, when all alternative paths for the current service fail, a RESET flow table entry with a priority of 0 is matched, and a RESET message is sent to the controller. After receiving the message, the controller determines that the autonomous decision-making recovery has failed, deletes the old flow table entry corresponding to the service, recalculates the set of available paths for the service, adds the service back to the queue of pending services, and returns to step (4) until the recovery of the service path is completed.

Images