Sdn link fault locating method, system and electronic device

Abstract

The purpose of this invention is to provide an SDN link fault location method, system, and electronic device. The method includes: acquiring link status data and service traffic data of each SDN link within the current data plane; extracting features from the link status data and the service traffic data to obtain state feature vectors and traffic feature vectors; constructing network topology data based on the network topology information of the current data plane, using devices within the current data plane as graph nodes and the state feature vectors and traffic feature vectors as edge attributes; inputting the network topology data into a pre-trained graph convolutional network fault location model to obtain the fault probability of each SDN link; and identifying SDN links whose fault probabilities meet a threshold condition as faulty links. This solution improves the adaptability and accuracy of fault link location.

Description

Technical Field

[0001] This invention relates to the field of industrial inspection technology, and in particular to an SDN link fault location method, system and electronic equipment. Background Technology

[0002] Software-defined networking (SDN), with its core architecture of separating control and forwarding, breaks the limitations of traditional network hardware, enabling programmable scheduling, centralized management and control, and elastic scaling of network resources. It has been deeply integrated into key areas such as cloud computing, big data, and the industrial internet, and is widely used in scenarios such as data center interconnection, carrier backbone network upgrades, and enterprise-level network transformation. With the rapid development of the digital economy, the types of services carried by SDN networks are becoming increasingly complex, and data transmission volume is growing exponentially. As the core channel for network data flow, the operational stability of the link directly determines the quality of service. The causes of link failures are becoming increasingly diverse. In addition to traditional fiber optic cable interruptions, port hardware failures, and link congestion, they also include complex factors such as malicious attacks, misconfigurations, and environmental interference. Such failures not only cause data transmission interruptions on a single link but can also spread through topological relationships, triggering a chain reaction and causing widespread service paralysis. This can lead to incalculable economic losses and security risks for critical businesses such as financial transactions, telemedicine, and industrial control. Therefore, building efficient and accurate link failure location and self-healing mechanisms has become a core requirement for ensuring the high-quality operation of SDN networks.

[0003] While SDN link fault location and self-healing technologies have made some progress, they still face many bottlenecks in complex network environments. Traditional fault location methods generally rely on static threshold judgments or simple association rules, which cannot accurately capture the dynamic topological dependencies and traffic relationships between links and nodes, and service flows in SDN networks. They are also difficult to deal with complex scenarios such as concurrent faults and fault propagation, resulting in significant false alarms and missed alarms in fault location.

[0004] Therefore, there is an urgent need for a solution that can adapt to complex scenarios and quickly locate faulty links. Summary of the Invention

[0005] The purpose of this invention is to provide an SDN link fault location method and system, which effectively solves the problem that traditional solutions are difficult to deal with complex scenarios such as concurrent faults and fault propagation of multiple links, resulting in obvious false alarms and missed alarms in fault location, and improves the adaptability and accuracy of fault link location.

[0006] To address the above problems, embodiments of the present invention provide an SDN link fault location method, the method comprising: The system acquires link status data and service traffic data for each SDN link within the current data plane. It then extracts features from the link status data and service traffic data to obtain state feature vectors and traffic feature vectors. Based on the network topology information of the current data plane, it constructs network topology data by using devices within the current data plane as graph nodes and the state feature vectors and traffic feature vectors as edge attributes. Finally, it inputs the network topology data into a pre-trained graph convolutional network fault location model to obtain the fault probability of each SDN link. SDN links whose fault probabilities meet a threshold condition are identified as faulty links.

[0007] This invention also proposes an SDN link fault location system, comprising: a fault perception module, deployed on the SDN controller side, used to acquire link status data and service traffic data of each SDN link in the current data plane through the southbound interface of the SDN controller, and extract features from the link status data and the service traffic data to obtain a state feature vector and a traffic feature vector; a network topology construction module, used to construct network topology structure data based on the network topology information of the current data plane, taking the devices in the current data plane as graph nodes and the state feature vector and the traffic feature vector as edge attributes; and a fault location module, used to input the network topology structure data into a pre-trained graph convolutional network fault location model to obtain the fault probability of each SDN link, and to determine the SDN link whose fault probability meets a threshold condition as a faulty link.

[0008] In addition, in some embodiments, before performing feature extraction on the link status data and the service traffic data as described above, the method further includes: performing data cleaning on the link status data and the service traffic data using a three-standard-deviation criterion; and performing standardization processing on the link status data and the service traffic data using a min-max standardization method.

[0009] In addition, in some embodiments, as described above, feature extraction is performed on the link state data and the service traffic data to obtain a state feature vector and a traffic feature vector, including: extracting latency, packet loss rate, bandwidth utilization, port transmit / receive rate, bit error rate, and signal strength features from the link state data, and concatenating them after standardization to form the state feature vector; extracting service traffic size, QoS level, flow duration, and average data packet length features from the service traffic data, and concatenating them after standardization to form the traffic feature vector.

[0010] In addition, in some embodiments, as described above, based on the network topology information of the current data plane, devices within the current data plane are treated as graph nodes, and the state feature vector and the traffic feature vector are treated as edge attributes to construct network topology data. This includes: abstracting switches and hosts in the network topology information as graph nodes, and abstracting physical links between adjacent nodes in the network topology information as graph edges; determining the edge attribute corresponding to each graph edge based on the state feature vector and the traffic feature vector corresponding to each graph edge; characterizing the edge attribute through a comprehensive feature vector formed by concatenating the state feature vector and the traffic feature vector; and obtaining the network topology data based on the set of graph nodes, the set of graph edges, and the feature matrix formed by each comprehensive feature vector.

[0011] In addition, in some embodiments, the method described above further includes: when the network topology information changes, determining the graph nodes and / or graph edges corresponding to the change; modifying the graph nodes and / or graph edges in the network topology data, and updating the feature matrix to obtain new network topology data.

[0012] In addition, in some embodiments, the graph convolutional network fault location model described above includes an input layer, two edge-attention-enhanced graph convolutional layers, a fully connected layer, and an output layer. The step of obtaining the pre-trained graph convolutional network fault location model includes: collecting historical data, dividing the historical data into a training set and a validation set, wherein the historical data includes normal link data and faulty link data; training the graph convolutional network fault location model using the training set, validating the trained graph convolutional network fault location model using the validation set, and obtaining the pre-trained graph convolutional network fault location model when the validation result meets preset conditions.

[0013] In addition, in some embodiments, the method described above further includes: obtaining the location, fault level, and service information affected by the faulty link based on the unique identifier information of the faulty link; obtaining real-time network status data; constructing a multi-objective optimization function based on the service information and the real-time network status data; constructing constraints based on the location of the faulty link and the real-time network status data; and solving the multi-objective optimization function using a preset genetic algorithm to obtain a self-healing strategy for the faulty link.

[0014] In addition, in some embodiments, the method described above further includes: converting the self-healing strategy into a control command and sending it to the corresponding device in the current data plane to perform the self-healing operation; after the self-healing operation is completed, collecting the link status data and the service traffic data in the current data plane again and comparing them with the preset self-healing success criteria; if unsuccessful, restarting the fault link location; if successful, recording the fault handling log.

[0015] This embodiment takes the current data plane as the spatiotemporal object of analysis, combines the real-time collected link status and service traffic data with the dynamic network topology, and encapsulates them into graph structure data. It also uses graph convolutional networks to perform deep modeling of the complex topology and traffic dependencies between links, thereby achieving fault location. This significantly improves the accuracy and robustness of fault location and effectively reduces the risk of false alarms and missed alarms in a dynamically changing network environment. Attached Figure Description

[0016] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0017] Figure 1 This is a flowchart illustrating the SDN link fault location method according to the first embodiment of the present invention. Figure 2 This is a flowchart illustrating step 102 in the second embodiment of the present invention. Figure 3 The following is a detailed flowchart of the SDN link fault location method according to the fourth embodiment of the present invention. Figure 1 ; Figure 4 The following is a detailed flowchart of the SDN link fault location method according to the fourth embodiment of the present invention. Figure 2 ; Figure 5 This is a schematic diagram of the SDN link fault location system according to the fifth embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of the electronic device according to the sixth embodiment of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] The first embodiment of the present invention relates to an SDN link fault location method, such as... Figure 1 As shown, the method in this embodiment specifically includes the following steps 101-103: Step 101: Obtain the link status data and service traffic data of each SDN link in the current data plane, and extract features from the link status data and service traffic data to obtain the status feature vector and traffic feature vector.

[0021] In this context, the current data plane refers to the real-time set of entities constituting the SDN network data forwarding entities when a fault location process is initiated. It is not a static hardware set, but rather a dynamic connection framework maintained by the SDN controller, possessing the current topology connections and real-time operational status. Specifically, this data plane can include currently networked switches, hosts, and other device nodes, as well as the physical or logical links between them that actually carry service traffic. Link status data refers to various indicators reflecting the transmission quality of physical links, such as link latency, packet loss rate, bandwidth utilization, and port transmit / receive rates. Service traffic data refers to the characteristics of various service flows carried on the links, such as service traffic size, Quality of Service (QoS) level identifier, flow duration, and average packet length. This embodiment does not limit the specific indicator data included in either. It should be noted that the QoS level identifier in this field typically refers to different levels of service requirements for network performance such as latency, jitter, and packet loss rate. In this embodiment, the QoS level identifier is a preset digital mapping value that reflects priority ranking.

[0022] Specifically, in this embodiment, the southbound interface of the SDN controller can periodically send status query commands to each switch in the data plane. The switches obtain raw measurement values ​​and transmit them back through embedded sensor modules or connected external acquisition devices, such as traffic collectors deployed on mirror ports or status sensors deployed at both ends of the link. From these values, the required link status data and service traffic data can be extracted. The acquisition cycle can be dynamically configured according to the service sensitivity requirements. For example, for highly latency-sensitive services, a shorter acquisition interval can be set to ensure the data's ability to capture sudden changes. Subsequently, after necessary cleaning and standardization of the acquired data, features are extracted from the cleaned data using manual design or automatic encoding. That is, fixed-dimensional link status feature vectors and traffic feature vectors are constructed, with each dimension of the vector corresponding to a data feature.

[0023] Step 102: Based on the network topology information of the current data plane, construct network topology data by taking the devices in the current data plane as graph nodes and the state feature vector and traffic feature vector as edge attributes.

[0024] Specifically, considering that the feature vectors of isolated links alone are insufficient to reflect the propagation correlation of faults across the topology, this embodiment introduces network topology information. This network topology information refers to a complete dataset describing the unique identifiers of all nodes in the current data plane and the physical connections between nodes. This information is maintained and stored by the SDN controller.

[0025] In practical implementation, the network topology information maintained by the SDN controller is synchronized in real time through the northbound interface of the SDN controller. This includes the identification information (ID, IP address, device model) of nodes (switches, hosts), the connection relationship of links (source node ID, destination node ID, link identifier), and the basic attributes of links (link type, maximum bandwidth). The synchronization period is consistent with the data acquisition period in step 101 to ensure the spatiotemporal consistency of link load status and topology connection relationship. After obtaining the topology information, graph structure construction is performed: each switch and host node is abstracted as a graph node and assigned a unique identifier; for each pair of adjacent nodes that have a physical connection and are confirmed to be active, an edge connecting the corresponding nodes is abstracted; then, the state feature vector and traffic feature vector generated for the link in step 101 are used as the edge attributes of the edge, thereby constructing the network topology structure data.

[0026] Step 103: Input the network topology data into the pre-trained graph convolutional network fault location model to obtain the fault probability of each SDN link, and identify the SDN links whose fault probabilities meet the threshold conditions as fault links.

[0027] Specifically, the pre-trained graph convolutional network fault location model in this embodiment is a fully trained graph neural network model. For the constructed graph structure data, after inputting into the pre-trained graph convolutional network fault location model, the model, through its internal multiple graph convolutional layers, performs weighted fusion and non-linear activation on the feature vectors of each edge and their neighboring edge features according to learned weights, ultimately outputting a fault probability value between zero and one for each edge. By comparing this probability value with a pre-set threshold, SDN links with probabilities greater than or equal to the threshold are determined as faulty links, thus completing the fault location. In this embodiment, the pre-set threshold can be determined by statistical analysis of historical fault data or operational experience; this embodiment does not impose specific limitations on it.

[0028] Compared with related technologies, this embodiment takes the current data plane as the spatiotemporal object of analysis, combines the real-time collected link status and service traffic data with the dynamic network topology, encapsulates them into graph structure data, and uses graph convolutional networks to perform deep modeling of the complex topology and traffic dependencies between links, thereby achieving fault location. This significantly improves the accuracy and robustness of fault location and effectively reduces the risk of false alarms and missed alarms in a dynamically changing network environment.

[0029] The second embodiment of the present invention relates to an SDN link fault location method. This embodiment is a supplement to the foregoing embodiment. The supplement is that, before performing feature extraction on the link status data and service traffic data respectively, the method of this embodiment further includes the following steps.

[0030] The link status data and service traffic data are cleaned using the three-standard-deviation criterion. The link status data and service traffic data are standardized using the min-max standardization method. Specifically, the three-standard-deviation criterion (3σ criterion) is used to remove abnormal data caused by sensor failures and transmission interference. The specific steps and formulas are as follows: Step 1: Calculate the sample mean μ and standard deviation σ for a certain data collection indicator; Step 2: Determine whether the sample data (any link status data or service traffic data) x satisfies |x -μ|>3σ. If it does, it is judged as abnormal data and removed. Mean: Where i = 1, 2, ..., n, n is the number of samples collected, and x i Let i be the i-th sample data. Standard deviation: ; Subsequently, the min-max normalization method was used to map the cleaned data to the [0,1] interval to eliminate the influence of different metrics: ; Where x is the sample data before standardization, min x The maximum value is the minimum value of the sample for this indicator. x x' represents the maximum value of the sample for this indicator, and x' represents the standardized sample data.

[0031] In another example, step 101 involves feature extraction from the link state data and service traffic data to obtain state feature vectors and traffic feature vectors, including: Delay, packet loss rate, bandwidth utilization, port transmit / receive rate, bit error rate, and signal strength features are extracted from the link state data, and then concatenated into a state feature vector after standardization. Features such as traffic size, QoS level, flow duration, and average packet length are extracted from traffic data and then concatenated into a traffic feature vector after standardization.

[0032] Specifically, six core features are extracted: link latency, packet loss rate, bandwidth utilization, port transmit / receive rate, bit error rate, and signal strength. After standardization, these features are concatenated to generate a 6-dimensional vector [latency, packet loss rate, bandwidth utilization, receive rate, transmit rate, bit error rate, signal strength]. Simultaneously, four features are extracted: service traffic size, QoS level (mapped to values ​​from 1 to 5), flow duration, and average packet length. After standardization, these features are concatenated to generate a 4-dimensional vector [traffic size, QoS level, duration, average packet length]. It should be noted that this embodiment does not limit the specific dimensions of the state feature vector and the traffic feature vector. In practical applications, the data and dimensions of the extracted state feature vector and traffic feature vector can be adaptively modified according to the actual situation.

[0033] In another example, such as Figure 2 As shown, step 102 specifically includes the following: Step 201: Abstract the switches and hosts in the network topology information as graph nodes, and abstract the physical links between adjacent nodes in the network topology information as graph edges.

[0034] Specifically, switches and hosts in the network topology information are uniformly abstracted into a graph node set V. To distinguish node types, each node can be assigned a type label t∈{0,1}, where 0 represents a switch node and 1 represents a host node. The mathematical expression of the node set is: Where M is the total number of nodes in the network, and ID is... k t is the unique identifier of the k-th node. k This is the type label for the k-th node.

[0035] Next, the physical links between adjacent nodes in the network topology information are abstracted as a set of graph edges E. Here, port transmit and receive rates can be pre-filtered to retain only valid links with data transmission capabilities. The mathematical expression of the edges is: ; Among them, e ij Indicates the connection node v i With v j The edge, L ij This serves as a unique identifier for the corresponding physical link, consistent with the link identifier in the data collected in S1, ensuring the association between the edge and the feature data in S1. Step 202: Determine the edge attributes corresponding to each graph edge based on the state feature vector and flow feature vector; the edge attributes are represented by a comprehensive feature vector formed by concatenating the state feature vector and the flow feature vector.

[0036] Specifically, the state feature vector and flow feature vector corresponding to each graph edge are concatenated. The concatenated composite feature vector is then used as the edge attribute corresponding to the graph edge.

[0037] For example, let the first The link state feature vector of each link is The vector has 6 dimensions, corresponding to latency, packet loss rate, bandwidth utilization, receive rate, transmit rate, bit error rate, and signal strength, respectively. These are the standardized values, and the traffic feature vector is... The dimension is 4, corresponding to the normalized values ​​of traffic size, QoS level, duration, and average packet length, respectively. The corresponding edge e is... ij The eigenvector is the concatenated vector of the two: The total dimension is 10.

[0038] Step 203: Based on the set of graph nodes, the set of graph edges, and the feature matrix composed of each comprehensive feature vector, obtain the network topology data.

[0039] Specifically, a feature matrix X is constructed based on the edge attributes corresponding to each graph edge. This feature matrix is ​​an edge feature matrix, where each row corresponds to an edge in the graph edge set E, and each column corresponds to an edge attribute. In the same example of step 202 above, the formula for constructing the feature matrix X is: ; Where N is the number of edges in the edge set E, i.e. the number of effective links, and each row corresponds to a 10-dimensional feature vector of an edge. This yields the network topology data G=(V,E,X).

[0040] Based on this, when the network topology information changes, the corresponding graph nodes and / or graph edges are determined; the graph nodes and / or graph edges are modified in the network topology data, and the feature matrix is ​​updated to obtain new network topology data.

[0041] Specifically, when a topology change notification pushed in real time by the SDN controller is detected, such as the addition of a link, node going online / offline, or link disconnection, the following update operations can be performed: If a new node is added, the new element is added to the node set V. If a node is offline, then delete the corresponding node from V. k Update M-1 to the current total number of nodes; If a new valid link is added, then add it to E. At the same time, the link collected in step 101 will be... and spliced ​​as , added to the end of the feature matrix X to update the feature matrix; If the link is broken, the corresponding e is deleted from E. ij Then delete the corresponding row in X and update N to the current total number of edges. After the update is complete, output the new network topology data G'=(V',E',X').

[0042] Compared with related technologies, this embodiment explicitly abstracts switches and hosts in network topology information as graph nodes, explicitly abstracts physical links between adjacent nodes as graph edges, and concatenates the link state characteristics and service traffic characteristics of each link into a comprehensive feature vector as edge attributes. This solution realizes the standardized transformation of multi-source heterogeneous network data into a unified graph data model, enabling subsequent graph convolutional networks to distinguish the functional differences of different types of nodes and capture the pattern of fault propagation along the topology path, thereby improving the comprehensiveness and accuracy of fault location.

[0043] The third embodiment of the present invention relates to an SDN link fault location method. This embodiment is a supplement to the foregoing embodiments, and the supplement is that the graph convolutional network fault location model in step 103 is explained.

[0044] In one example, the graph convolutional network fault localization model of this embodiment includes an input layer, two edge-attention-enhanced graph convolutional layers, a fully connected layer, and an output layer. At this point, the steps to obtain a pre-trained graph convolutional network fault localization model include: Collect historical data, and divide the training set and validation set based on the historical data. The historical data includes normal link data and faulty link data. The graph convolutional network fault location model is trained using a training set and validated using a validation set. When the validation results meet the preset conditions, the pre-trained graph convolutional network fault location model is obtained.

[0045] Specifically, the historical data may include 6000 normal link data entries, 2000 minor fault data entries, and 2000 severe fault data entries, divided into training and validation sets in a 7:3 ratio. The Adam optimizer is used with a learning rate of 0.001 and a decay coefficient of 0.9, employing the cross-entropy loss function for training. The training iterations are 50 rounds, with a batch size of 32. After training, validation is performed. When the validation results meet preset conditions, the pre-trained graph convolutional network fault localization model is obtained. In this embodiment, the preset conditions can be accuracy, judgment time, etc., and can be set according to actual needs; this embodiment does not impose specific limitations on these.

[0046] A well-trained graph convolutional network fault localization model can obtain fault probability outputs based on the input network topology data. The specific process is as follows: The input layer receives network topology data G, where the edge feature matrix X∈R (N is the number of effective links, and the 10-dimensional features correspond to the preprocessing result of S1), and the adjacency matrix A∈R of the network topology is also input (A=1 indicates that link i and link j are adjacent, and A=0 indicates that they are not adjacent, which is derived from the topology structure of S2). The first layer is an edge attention-enhanced graph convolutional layer with an input dimension of 10 and an output dimension of 64. Its core function is to integrate edge attention weights to achieve differentiated feature aggregation. The second layer is an edge attention-enhanced graph convolutional layer with an input dimension of 64 and an output dimension of 128, which further strengthens the deep correlation features across links. Fully connected layer: Input dimension 128, output dimension 1, completes feature dimension compression; Output layer: The sigmoid activation function is used to map the output to the [0,1] interval to obtain the failure probability of each link. Among them, the first and second edge attention-enhanced graph convolutional layers introduce an edge attention mechanism. To address the shortcomings of equal neighbor weights in traditional GCN, the edge attention weights are adaptively calculated through the following three steps to ensure that fault-related link features are given priority.

[0047] Step 1: Adjacent edge feature matching. For any link i (corresponding to graph edge e), filter its adjacent link set N(i) (links that share nodes with e in the network topology data), and extract the comprehensive feature vector corresponding to the preprocessed edge. (10-dimensional) and adjacent link feature vectors , j∈N(i), (10 dimensions); Step 2: Attention score calculation. The attention score of link j to link i is calculated using feature similarity and learnable parameters. The formula is: ; in, For attention score weight vectors, For the feature fusion weight matrix, express and Concatenate (dimension 20), tanh is the activation function; Step 3: Softmax normalization. Normalize the attention scores of all adjacent links of link i to obtain the final attention weights. The formula is: ; in, ,and A larger weight indicates that the adjacent link j contributes more to the fault characteristics of i.

[0048] Subsequently, attention weights will be incorporated into the graph convolution aggregation process. The feature aggregation formulas for the first and second edge-attention-enhanced graph convolutional layers are as follows: ; ; in, , These are the output features of the first and second edge-attention-enhanced convolutional layers, respectively. W0 represents the feature vector learned by adjacent link j after the first convolutional layer computation; 10×64 W1∈R 10×64 W2∈R 64×128 W3∈R 64×128 The weight matrix is ​​a learnable weight matrix optimized through training; ReLU is the activation function to alleviate gradient vanishing. This is a normalization factor to avoid eigenvalue overflow.

[0049] The output of the second layer The failure probability is obtained from a (128-dimensional) fully connected layer using the following formula: ; Where W4∈R 128×1 Here is the weight matrix of the fully connected layer, b is the bias term, and P is the weight matrix of the fully connected layer. i Let be the failure probability of link i.

[0050] In setting the threshold conditions for the probability of failure, the preset threshold is... =0.8, when P i A value ≥0.8 indicates a faulty link.

[0051] In addition, the original packet loss rate data collected in step 101 can be used to classify the levels; Minor fault: Raw packet loss rate ∈ [5%, 10%] (corresponding to standardized value ∈ [0.05, 0.1]); Critical fault: If the original packet loss rate is >10% or the port transmit / receive rate collected in step 101 is 0 for 3 consecutive collection cycles, it is determined to be a link interruption. The fourth embodiment of the present invention relates to an SDN link fault location method, such as... Figure 3 As shown, this embodiment also includes the following steps: Step 301: Based on the unique identifier of the faulty link, obtain the location, fault level, and service information affected by the faulty link, as well as obtain real-time network status data.

[0052] Based on the unique identifier of the located faulty link (L), the precise location of the faulty link in the network topology, its fault level (L∈{minor fault, severe fault}), such as minor or severe fault, and all service information affected by the faulty link interruption can be directly obtained from the link location results. The list of affected service information may specifically include the source / destination nodes, QoS requirements (latency ≤ T, packet loss rate ≤ P), and traffic volume for each service. This data originates from the service traffic data continuously collected in step 101. Simultaneously, to assess the resources required for self-healing and construct feasible solutions, real-time network status data of the entire network needs to be acquired synchronously. This data is directly taken from the latest output after preprocessing in step 101, specifically including but not limited to the remaining bandwidth of the entire link (B=B). max -B used B max B is the maximum bandwidth of the link. used For: Bandwidth utilization rate collected in step 101 * B max The topology constraints are derived from the network topology data G=(V,E,X) from step 102, including the currently available link set (E0 = E - {L}) and node connection relationships. It should be noted that obtaining the above information provides a data foundation for subsequent fault self-healing, clarifying under what network conditions, and to meet the needs of which services, the problem boundary at which fault location must be avoided.

[0053] Step 302: Construct a multi-objective optimization function based on business information and real-time network status data, and construct constraints based on the location of the faulty link and real-time network status data.

[0054] Specifically, addressing the shortcomings of traditional self-healing strategies—namely, their singular objectives and fixed weights—a multi-objective optimization function is constructed. The essence of this function is to find a self-healing solution that minimizes a comprehensive cost, which is a weighted sum of three factors: service interruption time, the complement of the average bandwidth utilization, and the complement of service quality satisfaction. The estimation of service interruption time includes both strategy calculation time and instruction issuance and execution time, the latter derived from the communication latency between the controller and switches in real-time network status data. The calculation of service quality satisfaction relies entirely on the specific service quality requirement thresholds for each service in the affected service information, determined by comparing the expected performance of each service after self-healing to ensure it meets the standards. Crucially, the weight coefficients in the multi-objective optimization function are not fixed but dynamically determined by the highest service quality level among the affected service information. When high-priority services are included in the affected services, the function automatically assigns a higher weight to service quality satisfaction to protect high-value services. This adaptive weighting allows the optimization function itself to be parameterized by the service information, enabling it to autonomously adjust its optimization bias based on the severity of the fault's impact.

[0055] While constructing the multi-objective optimization function, the constraint conditions are constructed simultaneously to limit the feasible space of the optimization solution. On the one hand, hard infeasibility constraints at the topology level are directly generated from the location information of the faulty link, meaning that the list of links in any candidate self-healing path scheme must absolutely not contain the faulty link, ensuring that the scheme is physically feasible. On the other hand, hard resource constraints at the capacity level are generated from the remaining available bandwidth of each link in the real-time network status data, meaning that the additional traffic bandwidth planned to be allocated to any backup link in the self-healing scheme must not exceed the current remaining available bandwidth limit of that link. In addition, additional optional constraints can be defined, such as requiring that the expected final load rate of all links on the path be controlled below a safe threshold to avoid overloading of new links after self-healing, thereby causing secondary failures.

[0056] In one example, the formula for constructing the multi-objective optimization function is as follows: ; Service interruption time T down : Refers to the business downtime from the occurrence of the fault to the completion of self-healing, calculated as T. down =T1+T2, where T1 is the strategy calculation time (based on the solution time of the preset genetic algorithm) and T2 is the flow table update / bandwidth configuration time (derived based on the controller-switch communication delay collected in step 101). Average bandwidth utilization The specific formula for measuring the balance of network resource utilization is as follows: ; Where, N available E represents the total number of available links in the self-healing network. available B represents the set of available links in the self-healing network. used,i Let ΔB be the bandwidth of link i before the failure. i After the bandwidth already used by link i has self-healed, the newly allocated bandwidth for link i is B. max,i This represents the maximum physical bandwidth of link i. S QoS QoS satisfaction is used to quantify the degree to which service QoS requirements are met for the affected service set. ; in This is an indicator function (1 if the condition is met, 0 otherwise). S represents the total number of services affected by the failure. affect For the set of services affected by the fault, delay s T represents the actual transmission delay of service s. qos,s For the maximum acceptable latency required by business s, loss s P represents the actual packet loss rate of service s. qos,s The maximum acceptable packet loss rate required by business s; The dynamic weights ω1 and ω2 can be dynamically adjusted according to the service priority. The priority is mapped by the QoS level collected in step 101, increasing from level 1 to level 5. The formula is ω1=0.3×(6-Q) / 5, ω2=0.7×Q / 5, to ensure that high-priority services have higher QoS weights (e.g., level 5 services ω2=0.7, ω1=0.06).

[0057] Step 303: Use a preset genetic algorithm to solve the multi-objective optimization function to obtain a self-healing strategy for the faulty link.

[0058] Specifically, to improve positioning efficiency, a pre-defined genetic algorithm is used to solve the multi-objective optimization function. In the implementation, the first step is to design an encoding scheme, mapping a complete self-healing strategy—containing the forwarding path selected for each affected service and the bandwidth allocation on each path link—as a structured string of numbers, i.e., a chromosome. Then, based on the feasible region defined by the constraints, a population containing several individuals is initialized using the shortest path algorithm and random bandwidth allocation. In each iteration, each individual in the population is substituted into the objective optimization function, and the reciprocal of the objective function value is used as the individual's fitness. Higher fitness indicates a lower overall cost for the self-healing scheme. Based on fitness, roulette wheel selection and elite retention are performed to select high-performing individuals for the next generation. New individuals are generated through specially designed crossover and mutation operators, continuously evolving the population while satisfying the constraints. When a pre-defined convergence condition is met, such as minimal fluctuation in the optimal fitness across several generations or reaching the maximum number of iterations, evolution stops, and the individual with the highest fitness in the population is output. This individual is then decoded to represent a specific path switching scheme and bandwidth adjustment command, thus obtaining the optimal self-healing strategy for the current fault.

[0059] In one example, the default genetic algorithm solution process is as follows: 1. The coding design adopts a hybrid coding method of path coding + bandwidth allocation coding. Chromosome length = number of affected services × (number of path nodes + bandwidth allocation value). Example: For service s (source A - destination B), the code "0101|300" indicates that the path is A→S1→S2→B (0101 corresponds to the node index), and the allocated bandwidth is 300Mbps. 2. Based on the set of available links ,use The shortest path algorithm generates the initial path, and the initial bandwidth allocation scheme is generated by combining the remaining bandwidth constraint △B≤B in step 101. The population size is set to 50. 3. The multi-objective optimization function Take the reciprocal as the fitness function, f=1 / (F+ɛ), ɛ=10 to avoid the denominator being 0. The higher the fitness value, the better the strategy. 4. The selection operation employs a roulette wheel and elite retention strategy, keeping the top 10% of the best individuals directly into the next generation to avoid the loss of superior genes; the crossover operation uses single-point crossover (the crossover point is selected at the node index) for path encoding and arithmetic crossover for bandwidth encoding, with a crossover probability P = 0.7 + 0.2 * (f max -f / f max At this point, the higher the fitness, the lower the crossover probability, thus retaining superior individuals; Mutation operation: Path mutation randomly replaces one node with an adjacent available node; bandwidth mutation is randomly adjusted within the remaining bandwidth constraint, with a mutation probability P = 0.05 - 0.03 * (f) max -f / f max Higher fitness results in a lower mutation probability, thus avoiding the destruction of high-quality individuals.

[0060] 5. Stop when the fitness value fluctuates by ≤1% for 5 consecutive generations, or when the number of iterations reaches 100 generations, and output the optimal individual as the self-healing strategy.

[0061] After acquiring self-healing strategies, such as Figure 4 As shown, the method in this embodiment further includes: Step 401: Convert the self-healing strategy into control commands and send them to the corresponding devices in the current data plane to execute the self-healing operation; Step 402: After the self-healing operation is completed, collect the link status data and service traffic data in the current data plane again and compare them with the preset self-healing success criteria. Step 403: If unsuccessful, restart the fault link location process; if successful, record the fault handling log.

[0062] Specifically, the standardized policy documents corresponding to the self-healing strategies are parsed. Based on the policy type (e.g., bandwidth adjustment, fault isolation, traffic rerouting) and the target device type, the policy parameters are converted into standardized control commands that the SDN controller can directly recognize. Specifically, bandwidth adjustment policies are converted into OpenFlow queue configuration commands, fault isolation policies into switch port shutdown commands, and traffic rerouting policies into OpenFlow flow table entries containing matching fields (source IP, destination IP, service priority) and action fields (forwarding port, bandwidth limit). After conversion, a command validity verification mechanism (checking if the target device ID exists and if the command parameters exceed device hardware thresholds, such as bandwidth adjustment values ​​not exceeding the maximum link bandwidth) is used to prevent device anomalies caused by invalid commands. If the verification fails, the policy is immediately regenerated.

[0063] During the control command issuance process, a hierarchical issuance and sequential execution mechanism is adopted to ensure the effective implementation of commands. First, fault isolation commands are prioritized and issued to the switches at both ends of the faulty link via the southbound interface of the SDN controller, ensuring rapid isolation of the faulty link from the network and preventing fault propagation. After the isolation command is executed, flow table commands and bandwidth adaptation commands corresponding to traffic rerouting are issued to ensure a smooth switchover of traffic on the new path. During execution, the execution feedback module monitors the command issuance status in real time and obtains the command execution results through the controller's northbound interface. If a command fails to be issued or times out, a command retransmission mechanism is triggered, with a maximum of three retransmissions. If the command still fails, it is immediately reported and the self-healing strategy is readjusted.

[0064] After the command is executed, feedback data collection is initiated using the existing deployed link status sensors, traffic collectors, and data collection agents. The collected data is consistent with that in step 101 and includes two parts: first, real-time status data of each link after self-healing (latency, packet loss rate, bandwidth utilization, port transmit / receive rate, etc.), with a focus on collecting the status of new path links and links surrounding the original faulty link; second, QoS data of affected services (service latency, packet loss rate, transmission rate, etc.) to verify whether the service has returned to normal.

[0065] The preprocessed feedback data is compared with the preset self-healing success criteria to verify the effect: If the feedback data shows that the faulty link has been completely isolated (port transmit / receive rate remains at 0), the load rate of the new path link is ≤80% and there are no signs of congestion, and the QoS indicators of the affected services all meet the preset requirements, and the data is stable for three consecutive collection cycles, then the self-healing is considered successful, the closed loop ends, and the feedback module records the fault handling log (including fault location, level, processing time, policy type, etc.) for subsequent model optimization. If the feedback data does not meet the above criteria (e.g., service latency still exceeds the standard, the new path link is congested, or the faulty link is not completely isolated), then the self-healing is considered unsuccessful, and the feedback module immediately triggers a secondary fault location and self-healing process based on the feedback data until self-healing is successful.

[0066] It should be noted that throughout the entire execution and feedback process of this solution, a full-process log is recorded in real time, including instruction conversion details, issuance time, execution results, feedback data collection time and content, self-healing effect judgment criteria and results, etc. The logs are stored in a standardized format on the log server on the SDN controller side and retained for no less than 90 days to facilitate subsequent fault tracing, policy optimization and operation and maintenance auditing. At the same time, it can provide historical data for fine-tuning the graph convolutional network fault location model.

[0067] Compared with related technologies, this embodiment utilizes service information and real-time network status data to construct a target optimization function, and transforms the fault location and real-time network status data into constraints, achieving multiple mutually restrictive optimization objectives. Furthermore, the weights in the target function can be adjusted based on the priority of affected services, demonstrating adaptive adjustment capabilities. This solution can dynamically generate the optimal self-healing strategy based on the severity of the fault, the differentiated needs of affected services, and the real-time distribution of network resources. While prioritizing the service quality of high-priority services, it balances the utilization of network bandwidth resources, effectively avoiding the risk of secondary faults caused by local link overload after recovery, and significantly improving the network's self-healing capability and resource utilization efficiency.

[0068] The fifth embodiment of the present invention relates to an SDN link fault location system, such as... Figure 5As shown, the system in this embodiment includes: The fault perception module 501 is deployed on the SDN controller side. It is used to obtain the link status data and service traffic data of each SDN link in the current data plane through the southbound interface of the SDN controller, and to extract features from the link status data and service traffic data to obtain the status feature vector and traffic feature vector respectively. The network topology construction module 502 is used to construct network topology structure data based on the network topology information of the current data plane, taking devices in the current data plane as graph nodes and state feature vectors and traffic feature vectors as edge attributes; The fault location module 503 is used to input network topology data into a pre-trained graph convolutional network fault location model to obtain the fault probability of each SDN link, and to identify the SDN link whose fault probability meets the threshold condition as the fault link.

[0069] In addition, this embodiment also includes: a topology update module, used to determine the graph nodes and / or graph edges corresponding to the changes when the network topology information changes; to modify the graph nodes and / or graph edges in the network topology structure data, and to update the feature matrix at the same time to obtain new network topology structure data.

[0070] In addition, this embodiment also includes: a fault self-healing module, used to obtain the location, fault level and service information affected by the faulty link based on the unique identification information of the faulty link, and to obtain real-time network status data; to construct a target optimization function based on the service information and real-time network status data, and to construct constraints based on the location of the faulty link and real-time network status data; and to solve the target optimization function using a preset genetic algorithm to obtain a self-healing strategy for the faulty link. The fault self-healing module in this embodiment is also used to: convert the self-healing strategy into control commands and send them to the corresponding devices in the current data plane to perform the self-healing operation; after the self-healing operation is completed, collect the link status data and service traffic data in the current data plane again and compare them with the preset self-healing success criteria; if it is unsuccessful, start fault link location again; if it is successful, record the fault handling log.

[0071] In summary, it is not difficult to see that this embodiment is a system embodiment corresponding to the aforementioned method embodiments. The implementation details described in the above method embodiments are still applicable and effective in this embodiment, and will not be repeated here.

[0072] The sixth embodiment of the present invention relates to an electronic device, such as... Figure 6As shown, it includes at least one processor 602; and a memory 601 communicatively connected to at least one processor 602; wherein the memory 601 stores instructions executable by at least one processor 602, the instructions being executed by at least one processor 602 to enable at least one processor 602 to perform any of the above method embodiments.

[0073] The memory 601 and processor 602 are connected via a bus, which may include any number of interconnecting buses and bridges. The bus connects various circuits of one or more processors 602 and memory 601 together. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 602 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 602.

[0074] Processor 602 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 601 can be used to store data used by processor 602 during operation.

[0075] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements any of the above-described method embodiments.

[0076] Another embodiment of the present invention relates to a computer program product, including computer instructions that, when executed by a processor, implement any of the above-described method embodiments.

[0077] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0078] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0079] While the embodiments disclosed herein are as described above, the foregoing content is merely for the purpose of facilitating understanding of this disclosure and is not intended to limit this disclosure. Any person skilled in the art to which this disclosure pertains may make any modifications and changes in form and detail of the implementation without departing from the spirit and scope of this disclosure; however, the scope of patent protection of this disclosure shall still be determined by the scope defined in the appended claims.

Claims

1. A method for locating SDN link faults, characterized in that, include: Obtain link status data and service traffic data from each SDN link in the current data plane, and extract features from the link status data and the service traffic data to obtain status feature vectors and traffic feature vectors. Based on the network topology information of the current data plane, the devices in the current data plane are used as graph nodes, and the state feature vector and the traffic feature vector are used as edge attributes to construct network topology data. The network topology data is input into a pre-trained graph convolutional network fault location model to obtain the fault probability of each SDN link. The SDN links whose fault probabilities meet the threshold condition are identified as faulty links.

2. The method according to claim 1, characterized in that, Before performing feature extraction on the link status data and the service traffic data respectively, the method further includes: The link status data and the service traffic data are cleaned using a three-standard-deviation criterion. The link status data and the service traffic data are standardized using the min-max standardization method.

3. The method according to claim 2, characterized in that, The step of extracting features from the link status data and the service traffic data to obtain a status feature vector and a traffic feature vector includes: The latency, packet loss rate, bandwidth utilization, port transmit / receive rate, bit error rate, and signal strength features are extracted from the link status data, and then standardized and concatenated to form the status feature vector. The traffic volume, QoS level, flow duration, and average packet length features are extracted from the traffic data and then concatenated after standardization to form the traffic feature vector.

4. The method according to claim 1, characterized in that, Based on the network topology information of the current data plane, devices within the current data plane are used as graph nodes, and the state feature vector and the traffic feature vector are used as edge attributes to construct network topology structure data, including: The switches and hosts in the network topology information are abstracted as graph nodes, and the physical links between adjacent nodes in the network topology information are abstracted as graph edges; The edge attribute corresponding to each graph edge is determined based on the state feature vector and the flow feature vector; the edge attribute is characterized by a comprehensive feature vector formed by concatenating the state feature vector and the flow feature vector. The network topology data is obtained based on the set of graph nodes, the set of graph edges, and the feature matrix formed by the comprehensive feature vectors.

5. The method according to claim 4, characterized in that, The method further includes: When the network topology information changes, determine the graph nodes and / or graph edges corresponding to the change; Modify the graph nodes and / or graph edges in the network topology data, and update the feature matrix to obtain new network topology data.

6. The method according to claim 1, characterized in that, The graph convolutional network fault localization model includes an input layer, two edge-attention-enhanced graph convolutional layers, a fully connected layer, and an output layer. The steps for obtaining the pre-trained graph convolutional network fault localization model include: Collect historical data, and divide the historical data into training set and validation set. The historical data includes normal link data and faulty link data. The graph convolutional network fault location model is trained using the training set and validated using the validation set. When the validation results meet the preset conditions, the pre-trained graph convolutional network fault location model is obtained.

7. The method according to claim 1, characterized in that, The method further includes: Based on the unique identifier of the faulty link, the location, fault level, and service information affected by the faulty link are obtained, as well as real-time network status data are obtained. A multi-objective optimization function is constructed based on the business information and the real-time network status data, and constraints are constructed based on the location of the faulty link and the real-time network status data. A preset genetic algorithm is used to solve the multi-objective optimization function to obtain a self-healing strategy for the faulty link.

8. The method according to claim 7, characterized in that, The method further includes: The self-healing strategy is converted into control commands and sent to the corresponding device in the current data plane to execute the self-healing operation; After the self-healing operation is completed, the link status data and service traffic data in the current data plane are collected again and compared with the preset self-healing success criteria. If unsuccessful, restart the fault link location process; if successful, record the fault handling log.

9. An SDN link fault location system, characterized in that, include: The fault perception module, deployed on the SDN controller side, is used to obtain link status data and service traffic data of each SDN link in the current data plane through the southbound interface of the SDN controller, and to extract features from the link status data and the service traffic data to obtain status feature vectors and traffic feature vectors respectively. The network topology construction module is used to construct network topology structure data based on the network topology information of the current data plane, taking the devices in the current data plane as graph nodes and the state feature vector and the traffic feature vector as edge attributes. The fault location module is used to input the network topology data into a pre-trained graph convolutional network fault location model to obtain the fault probability of each SDN link, and to identify the SDN link whose fault probability meets the threshold condition as a fault link.

10. An electronic device, characterized in that, include: At least one processor; as well as, A memory that is communicatively connected to at least one processor; wherein, The memory stores instructions that are executed by at least one processor to enable the at least one processor to perform the SDN link fault location method as described in any one of claims 1-8.

Images