Method, system and equipment for preventing and repairing cross-regional network fault and medium

Abstract

The invention discloses a cross-regional network fault prevention and repair method, system and device and a medium, and the method specifically comprises the steps: collecting the multi-modal data of a cross-regional network according to distributed monitoring nodes, and generating a multi-modal AI model through the training of the multi-modal data based on a quantum computing platform; deploying the multi-modal AI model to each edge node of a cross-regional network, and performing fault detection on each edge node according to the multi-modal AI model to obtain a fault detection result; according to the fault detection result and the real-time network state of the cross-regional network, dynamically adjusting a traffic scheduling strategy of the cross-regional network through a reinforcement learning algorithm; and determining a fault self-healing strategy according to the fault detection result, and executing a fault self-healing operation on the cross-regional network through the fault self-healing strategy. According to the method, the technical bottleneck faced by a cross-regional network fault prediction and self-healing system is effectively solved, and the real-time performance, the accuracy and the self-adaptive capability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of network fault repair, and in particular to a method, system, device and medium for preventing and repairing cross-regional network faults. Background Art

[0002] With the rapid development of information technology, cross-regional networks have become an indispensable information infrastructure in modern society, carrying data transmission and communication tasks around the world. However, with the continuous increase in network scale and complexity, cross-regional network fault prediction and self-healing systems face unprecedented challenges. In particular, in terms of real-time processing of massive multimodal data and efficient training of complex AI models, traditional computing methods have been unable to meet real-time requirements, resulting in delayed fault prediction and self-healing measures, which seriously affects the quality of network services.

[0003] First, traditional computing methods are limited in processing speed and computing efficiency. Cross-regional network fault prediction and self-healing systems need to process a large amount of data from different sources and types in real time, including traffic data, device status data, user behavior data, etc. These data have multimodal characteristics, that is, they contain information in multiple dimensions such as time, space, and attributes. When processing these massive amounts of data, traditional computing methods often take too long to process due to limitations in computing speed and efficiency, and cannot meet real-time requirements. This not only causes a lag in fault prediction results, but may also cause self-healing measures to fail to be triggered in a timely manner, thereby extending the fault recovery time and reducing the quality of network services.

[0004] Secondly, the existing traffic scheduling algorithms and fault self-healing mechanisms are insufficient. With the increase in network scale and complexity, the existing traffic scheduling algorithms and fault self-healing mechanisms have been unable to adapt to the dynamically changing network environment. Traditional traffic scheduling algorithms are often based on static rules or simple thresholds for judgment, and are unable to accurately identify potential risks and adjust strategies in a timely manner. At the same time, the existing fault self-healing mechanism also lacks sufficient flexibility and intelligence, which results in the system being unable to respond quickly when a network failure occurs, thereby increasing the impact of the failure on business operations.

[0005] In addition, the analysis and mining of massive historical data poses a challenge to the system optimization capability. The cross-regional network fault prediction and self-healing system needs to make full use of massive historical data for analysis and mining to optimize model performance and improve prediction accuracy. However, when processing these historical data, traditional computing methods are often limited to simple statistical analysis or rule-based reasoning due to computing power limitations. This makes it impossible for the system to deeply explore the potential laws and correlations in the data, thus limiting the improvement of model performance and prediction accuracy. Summary of the invention

[0006] The purpose of the present invention is to provide a method, system, device and medium for preventing and repairing cross-regional network failures, which effectively solves the technical bottlenecks faced by cross-regional network failure prediction and self-healing systems, improves the real-time, accuracy and adaptability of the system, and provides a strong guarantee for the stable operation of cross-regional networks, so as to solve at least one of the above-mentioned existing technical problems.

[0007] In a first aspect, the present invention provides a method for preventing and repairing cross-regional network failures, the method specifically comprising:

[0008] Collect multimodal data across regional networks according to distributed monitoring nodes, and generate a multimodal AI model through training of the multimodal data based on the quantum computing platform;

[0009] Deploy the multimodal AI model to each edge node of the cross-regional network, perform fault detection on each edge node according to the multimodal AI model, and obtain a fault detection result;

[0010] According to the fault detection result and the real-time network status of the cross-regional network, dynamically adjust the traffic scheduling strategy of the cross-regional network through a reinforcement learning algorithm;

[0011] Determine a fault self-healing strategy according to the fault detection result, perform a fault self-healing operation on the cross-regional network through the fault self-healing strategy, and obtain fault self-healing execution status data;

[0012] Determine whether the fault self-healing operation is successful according to the fault self-healing execution status data; if the fault self-healing operation is not successful, trigger an alarm process.

[0013] In a second aspect, the present invention provides a system for preventing and repairing cross-regional network failures, the system specifically comprising:

[0014] A first prevention and repair module is used to collect multimodal data across regional networks according to distributed monitoring nodes, and generate a multimodal AI model through training of the multimodal data based on the quantum computing platform;

[0015] A second prevention and repair module is used to deploy the multimodal AI model to each edge node of the cross-regional network, perform fault detection on each edge node according to the multimodal AI model, and obtain a fault detection result;

[0016] A third prevention and repair module is used to dynamically adjust the traffic scheduling strategy of the cross-regional network through a reinforcement learning algorithm according to the fault detection result and the real-time network status of the cross-regional network;

[0017] A fourth prevention and repair module is used to determine a fault self-healing strategy according to the fault detection result, perform a fault self-healing operation on the cross-regional network through the fault self-healing strategy, and obtain fault self-healing execution status data;

[0018] The fifth preventive repair module is used to determine whether the fault self-healing operation is successful according to the fault self-healing execution status data, and if the fault self-healing operation is not successful, trigger an alarm process.

[0019] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor and a computer program stored in the memory, and when the computer program is executed on the processor, a method for preventing and repairing cross-regional network failures as described in any one of the above methods is implemented.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for preventing and repairing cross-regional network failures as described in any one of the above methods is implemented.

[0021] Compared with the prior art, the present invention has at least one of the following technical effects:

[0022] 1. The present invention effectively solves the technical bottlenecks faced by cross-regional network fault prediction and self-healing systems, improves the real-time, accuracy and adaptability of the system, and provides a strong guarantee for the stable operation of cross-regional networks.

[0023] 2. The present invention utilizes a quantum computing platform to process massive multimodal data, which can significantly improve processing speed and computing efficiency, thereby meeting real-time requirements and making fault prediction and self-healing measures more timely and effective.

[0024] 3. The present invention dynamically adjusts the traffic scheduling strategy through a reinforcement learning algorithm, optimizes network resource allocation, reduces the impact of faults on network performance, and improves network stability and reliability.

[0025] 4. By correlating and analyzing historical data and building a multivariate time series prediction model, the present invention can deeply explore the potential patterns and correlations in the data, thereby optimizing model performance and improving prediction accuracy.

[0026] 5. Through the analysis of the graph neural network model, the present invention can accurately identify key nodes and links in the network, so that monitoring resources can be more effectively deployed at these key locations, thereby improving the accuracy and efficiency of fault detection.

[0027] 6. The present invention deeply mines the time trends and spatial distribution characteristics in the data to form a comprehensive fault data training set, and then trains an AI model that can predict the temporal and spatial evolution of faults, thereby improving the accuracy and timeliness of fault prediction.

[0028] 7. The present invention optimizes the multimodal AI model through technical means such as model pruning, quantization, and knowledge distillation, so that it can run efficiently on edge nodes while maintaining high prediction accuracy, thereby ensuring the stability and reliability of the model in practical applications.

[0029] 8. The present invention accurately identifies the fault type and severity, quickly screens out suitable candidate strategies from a pre-built fault self-healing strategy library, and obtains the best self-healing strategy through optimization combination, thereby improving the efficiency and success rate of fault self-healing.

[0030] 9. The present invention uses historical data to build a prediction model to predict the changing trend of future network traffic and self-healing status, so as to discover potential network failure risks in advance and provide a scientific basis for the formulation of failure prevention and self-healing strategies. At the same time, according to the prediction results of the self-healing status trend, the self-healing strategy can be adjusted in time to ensure that the failure can be handled promptly and effectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 It is a flowchart of a method for preventing and repairing cross-regional network failures provided by the first embodiment of the present invention;

[0033] Figure 2 It is a flowchart of a method for preventing and repairing cross-regional network failures provided by the second embodiment of the present invention;

[0034] Figure 3 It is a flowchart of a method for preventing and repairing cross-regional network failures provided by the third embodiment of the present invention;

[0035] Figure 4 It is a flowchart of a method for preventing and repairing cross-regional network failures provided by a fourth embodiment of the present invention;

[0036] Figure 5 It is a flowchart of a method for preventing and repairing cross-regional network failures provided by a fifth embodiment of the present invention;

[0037] Figure 6 It is a flowchart of a method for preventing and repairing cross-regional network failures provided by a sixth embodiment of the present invention;

[0038] Figure 7It is a flowchart of a method for preventing and repairing cross-regional network failures provided by the seventh embodiment of the present invention;

[0039] Figure 8 It is a structural diagram of a system for preventing and repairing cross-regional network failures provided by an embodiment of the present invention;

[0040] Fig. 9 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0042] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0043] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0044] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0045] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0046] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0047] In the embodiment of the present application, the execution subject of the process includes a terminal device, which includes but is not limited to: a server, a computer, a smart phone, a tablet computer, and other devices capable of executing the method disclosed in the present application. Figure 1 A schematic flow chart of a method for preventing and repairing cross-regional network failures disclosed in the first embodiment of the present invention is shown, and is described in detail as follows:

[0048] S101, collecting multimodal data across regional networks according to distributed monitoring nodes, and generating a multimodal AI model through training of the multimodal data based on a quantum computing platform.

[0049] In this embodiment, distributed monitoring nodes are deployed to collect multimodal data of the network, including network traffic data, device status data, user behavior data, etc. The collected multimodal data is input into the quantum computing platform, and quantum algorithms are used for data processing and feature extraction. At the same time, a structure of a multimodal AI model is designed to accelerate the training process of the model by using the efficient parallel processing capabilities of quantum computing. During the training process, the parameters and structure of the model are adjusted to gradually adapt it to the distribution and characteristics of the data, thereby improving the accuracy of the prediction. After the training is completed, the model needs to be verified and optimized. A portion of the data can be used as a verification set to evaluate the performance of the model. Based on the verification results, the model can be further adjusted and optimized to improve its generalization ability and robustness.

[0050] In this embodiment, by utilizing multimodal data and the efficient processing capabilities of the quantum computing platform, an accurate multimodal AI model can be trained, which can capture multiple fault modes and characteristics in the network, thereby improving the accuracy of fault prediction. Through automated fault prediction and monitoring, manual intervention and operation and maintenance costs can be reduced. At the same time, timely fault warning and location can also help operation and maintenance personnel solve problems faster and improve the stability and reliability of the network.

[0051] S102, deploying the multimodal AI model to each edge node of the cross-regional network, performing fault detection on each edge node according to the multimodal AI model, and obtaining a fault detection result.

[0052] In this embodiment, before deploying the multimodal AI model to the edge node, the model must first be optimized and pruned to reduce the complexity and computational complexity of the model, making it more suitable for running on resource-constrained edge devices. Technical means such as model pruning and weight quantization can be used to reduce the size and computational requirements of the model while maintaining the prediction accuracy of the model as much as possible. The optimized multimodal AI model is deployed to each edge node of the cross-regional network. This can be achieved by uploading the model file to the storage system of the edge device and configuring the corresponding operating environment on the device. During the deployment process, it is necessary to ensure that the model file is compatible with the hardware and software environment of the edge device to ensure that the model can run normally. Run the multimodal AI model on the edge node and perform real-time analysis of the collected network data to detect potential faults. The model can output fault prediction results, including fault type, fault probability and other information based on the characteristics of the input data. When the model detects a potential fault, it can trigger the corresponding alarm mechanism so that the operation and maintenance personnel can take timely measures to deal with it.

[0053] In this embodiment, by deploying the multimodal AI model to the edge node, real-time fault detection can be achieved. This can greatly shorten the time of fault discovery and improve the efficiency of fault handling. Real-time fault detection helps operation and maintenance personnel to promptly discover and handle potential problems in the network, thereby ensuring the stability and reliability of the network. By deploying the multimodal AI model to the edge node, it is possible to avoid transmitting data to the cloud for processing, thereby reducing the delay in data transmission. This is especially important for fault detection scenarios that require rapid response, and can ensure that measures can be taken quickly to handle the fault when it occurs.

[0054] S103, dynamically adjusting the traffic scheduling strategy of the cross-regional network through a reinforcement learning algorithm according to the fault detection result and the real-time network status of the cross-regional network.

[0055] In this embodiment, traffic data, device status data, user behavior data, etc. of cross-regional networks, as well as the fault detection results of multimodal AI models are collected in real time, and the collected data is preprocessed, including data cleaning, formatting, feature extraction, etc., for subsequent analysis and decision-making. A reinforcement learning model is designed, which takes the real-time network status of cross-regional networks and the fault detection results of multimodal AI models as input, and outputs a dynamically adjusted traffic scheduling strategy. The reinforcement learning model needs to define the state space, action space, and reward function. The state space may include network traffic, device load, fault type, etc.; the action space may include the adjustment direction of the traffic scheduling strategy (such as increasing or decreasing traffic in a certain direction); the reward function may be designed based on indicators such as network performance and user satisfaction. The reinforcement learning model is trained using historical data and a simulation environment so that it gradually learns the optimal traffic scheduling strategy. During the training process, the parameters and structure of the model are continuously adjusted to improve its prediction accuracy and generalization ability. You can use deep learning frameworks (such as TensorFlow or PyTorch) and reinforcement learning libraries (such as OpenAI Gym or TensorTrade) to build and train models. Deploy the trained reinforcement learning model to the traffic scheduling system of the cross-regional network, and dynamically adjust the traffic scheduling strategy based on the real-time network status and the fault detection results of the multimodal AI model. Monitor the implementation effect of the traffic scheduling strategy and continuously optimize and adjust the model based on feedback.

[0056] In this embodiment, by dynamically adjusting the traffic scheduling strategy, the distribution of network traffic can be optimized, network congestion and delay can be reduced, and the network throughput and response speed can be improved, which helps to improve user experience and meet user needs for high-quality network services. When a network failure occurs, the reinforcement learning model can quickly adjust the traffic scheduling strategy based on the fault detection results and real-time network status to bypass the fault point or reduce the impact of the fault on the network, which helps to enhance the fault tolerance and reliability of the network and ensure business continuity and stability. Through automated traffic scheduling and fault recovery mechanisms, manual intervention and operation and maintenance costs can be reduced. At the same time, real-time network monitoring and early warning systems can help operation and maintenance personnel to promptly discover and deal with potential problems to avoid the expansion of faults and increased losses.

[0057] S104, determining a fault self-healing strategy according to the fault detection result, performing a fault self-healing operation on the cross-regional network according to the fault self-healing strategy, and obtaining fault self-healing execution status data.

[0058] In this embodiment, when the multimodal AI model detects a fault in a cross-regional network, it outputs key information such as the fault type and fault location. Based on this information, we can determine the corresponding fault self-healing strategy. For example, if a network device failure is detected, you can choose to restart the device or replace the backup device; if it is a network congestion problem, you can adjust the traffic scheduling strategy or increase bandwidth resources. According to the determined fault self-healing strategy, the automation system will perform corresponding self-healing operations on the cross-regional network, including sending restart commands to the faulty device, adjusting the network device configuration, and reallocating network resources. In the process of performing the self-healing operation, the system will monitor the operation results and network status in real time to ensure the effectiveness and safety of the self-healing operation. After performing the self-healing operation, the system will collect fault self-healing execution status data, including the success or failure of the self-healing operation, the operation time, the operation results, etc. These data will be used for subsequent network status analysis and fault prediction in order to continuously optimize and adjust the fault self-healing strategy.

[0059] In this embodiment, through the execution of the fault self-healing strategy, the fault problem in the network can be quickly solved, the stability and reliability of the network can be improved, the network interruption time can be reduced, and the continuity of business and user satisfaction can be guaranteed. The automated fault self-healing mechanism can reduce manual intervention and operation and maintenance costs, and the operation and maintenance personnel can devote more energy to network optimization and business development instead of frequently handling network failures. Fast fault recovery and self-healing operations can reduce the network delay and interruption time felt by users, help improve user experience and satisfaction, and enhance users' trust and dependence on the network.

[0060] S105: determining whether the fault self-healing operation is successful according to the fault self-healing execution status data; if the fault self-healing operation is not successful, triggering an alarm process.

[0061] In this embodiment, after executing the fault self-healing operation, the system will collect and analyze the fault self-healing execution status data, including the operation results, operation time, operation log, etc. The system will compare these data with the expected fault self-healing results to determine whether the fault self-healing operation is successful. If the fault self-healing operation is successful, that is, the network status returns to normal, the system will record the success information and update the network status database. If the fault self-healing operation is not successful, that is, the network status is still abnormal or the problem is not solved, the system will immediately trigger the alarm process. When the fault self-healing operation fails, the system will automatically generate alarm information, including the fault type, fault location, and the reason for the failure of the self-healing operation. These alarm information will be sent to the designated operation and maintenance personnel or team through emails, text messages, instant messaging tools, etc. At the same time, the system will also display the alarm information on the monitoring interface so that the operation and maintenance personnel can find and handle it in time. After the alarm process is triggered, the operation and maintenance personnel will receive the alarm information and immediately troubleshoot and handle the fault. The operation and maintenance personnel will quickly locate the cause of the problem according to the fault type and location in the alarm information, and take corresponding solutions. After resolving the fault, the operation and maintenance personnel will update the network status database and close the alarm information.

[0062] In this embodiment, by automatically judging whether the self-healing operation is successful and immediately triggering the alarm process when the operation fails, the problem can be quickly notified to the operation and maintenance personnel, which helps to shorten the fault handling time, improve the fault handling efficiency, and reduce the service interruption time caused by the fault. Timely alarms and fault handling can ensure that network faults are quickly resolved, thereby enhancing the stability and reliability of the network, helping to improve user experience and satisfaction, and ensuring business continuity and stability. By analyzing the self-healing execution status data and alarm information, the operation and maintenance personnel can continuously summarize experience and lessons and improve fault detection and handling capabilities.

[0063] In some embodiments, Figure 2 As shown, in the above step S101, the multimodal data across the regional network is collected according to the distributed monitoring nodes, and the following is also included before:

[0064] S201, acquiring network device attribute data and link connection attribute data of a cross-regional network, constructing a node feature vector according to the network device data, and constructing an edge feature vector according to the link connection attribute data;

[0065] S202, constructing a network topology graph using a graph neural network model according to the node feature vector and the edge feature vector, calculating the network topology graph through a graph convolution operation, and determining an importance score of each node feature vector;

[0066] S203, determining a key node set by comparing the importance score of each node feature vector with a preset score threshold;

[0067] S204, detecting whether there is a link connection between every two key nodes in the key node set, and if there is a link connection, determining the link between the two key nodes as a key link;

[0068] S205: monitor each key node and each key link of the cross-regional network according to the distributed monitoring nodes.

[0069] In this embodiment, the attribute data of network devices and link connections in the cross-regional network are obtained, and a node feature vector representing the characteristics of each network device is constructed for the obtained network device attribute data. For the obtained link connection attribute data, an edge feature vector representing the characteristics of each link connection is constructed. According to the constructed node feature vector and edge feature vector, a graph neural network model is used to construct a network topology diagram of the cross-regional network, wherein the network devices are nodes of the network topology diagram and the link connections are edges of the network topology diagram. A graph convolution operation is performed on the constructed network topology diagram, and the node feature vector is updated by aggregating the neighborhood node information of each node, and multiple rounds of graph convolution operations are iterated until the node feature vector converges. According to the feature vector of each node obtained by the graph convolution operation, the importance score of each node feature vector is calculated using an attention mechanism, and the higher the score, the more important the node is in the network topology. The importance score of the node is compared with a preset threshold, and if it is greater than the threshold, the node is added to the key node set. The key node set is traversed, and any two key nodes are obtained to determine whether there is a link connection between them. If there is a link connection between the two key nodes, the link is determined as a key link and added to the key link set. According to the network topology, clustering algorithms are used to group key nodes to obtain several key node subsets. Distributed monitoring nodes are deployed in multiple areas of the network, and each monitoring node is responsible for monitoring a key node subset. The monitoring node collects the performance indicator data of the key nodes and key links in real time, triggers an alarm when an abnormality is found, and reports the abnormal information to the central management node for processing by the operation and maintenance personnel.

[0070] Exemplarily, network device attribute data may include information such as the type, model, operating system, processing power, memory size, storage capacity, etc. of the device. Link connection attribute data may include information such as the bandwidth, delay, packet loss rate, connection type, etc. of the link. These data may be obtained through a network management system or a network detection tool. For example, suppose there is a cross-regional network that connects three data centers in City A, City B, and City C. Through the network management system, it can be obtained that the model of router A in City A data center is XYZ-123, the operating system is Linux, the processing power is 10Gbps, the memory is 16GB, and the storage capacity is 1TB. At the same time, it can also be obtained that the bandwidth of the link between City A data center and City B data center is 10Gbps, the delay is 10ms, and the packet loss rate is 0.1%. Similarly, the attribute data of other devices and links may be obtained. Next, it is necessary to construct a node feature vector based on the network device data and to construct an edge feature vector based on the link connection attribute data. The node feature vector can be used to represent the attribute characteristics of the network device node, and the edge feature vector can be used to represent the attribute characteristics of the link connection. Some feature encoding methods, such as one-hot encoding and embedded encoding, can be used to convert the attribute data of the device and the attribute data of the link into a numerical vector representation. In the above example, the model XYZ-123 of router A can be represented by one-hot encoding as [1,0,0], the operating system Linux can be represented by one-hot encoding as [0,1], the processing capacity of 10Gbps can be represented by a value of 10, the memory of 16GB can be represented by a value of 16, and the storage capacity of 1TB can be represented by a value of 1. Combining these values, we can get the node feature vector of router A [1,0,0,0,1,10,16,1]. Similarly, the bandwidth of the link 10Gbps can be represented by a value of 10, the delay of 10ms can be represented by a value of 10, and the packet loss rate of 0.1% can be represented by a value of 0.1. Combining these values, we can get the edge feature vector of the link [10,10,0.1]. Then, we need to use the graph neural network model to construct the network topology based on the node feature vector and the edge feature vector. The graph neural network model is a neural network model specifically designed to process graph structured data. It can effectively learn the feature representation of nodes and edges and perform reasoning on the graph structure. In this process, network devices are used as nodes and link connections are used as edges to construct a graph-structured network topology. Based on the above example, router A can be used as a node in the graph, and the link between the data center in city A and the data center in city B can be used as an edge in the graph. Then, based on the node feature vector and the edge feature vector, a complete network topology graph can be constructed using the graph neural network model. Next, the network topology graph needs to be calculated through a graph convolution operation to determine the importance score of each node feature vector.Graph convolution operation is a common operation in graph neural network models. It can propagate and aggregate information on the graph structure to obtain the feature representation of the node. The importance score of a node can reflect the importance of the node in the network, such as the degree centrality and betweenness centrality of the node. In the above example, the importance score of router A can be calculated through graph convolution operation. If router A is connected to routers in multiple other data centers, its importance score will be relatively high because it is a key node for communication between multiple data centers. Then, it is necessary to determine the key node set by comparing the importance score of each node feature vector with the preset score threshold. The preset score threshold can be set according to actual conditions, for example, it can be determined according to factors such as the scale of the network and the importance of the business. If the importance score of the node exceeds the preset score threshold, then the node is considered to be a key node and is added to the key node set. Assume that the preset score threshold is 0.8. In the above example, if the importance score of router A is calculated to be 0.9, which is greater than 0.8, then router A can be added to the key node set. Next, it is necessary to detect whether there is a link connection between each two key nodes in the key node set. If there is a link connection, the link between the two key nodes is determined as a key link. Critical links are links connecting critical nodes, which are crucial to the stability and performance of the network. If router A and router B in the data center of city B are both critical nodes, and there is a direct link connection between them, then this link is determined to be a critical link. Finally, it is necessary to monitor each critical node and each critical link of the cross-regional network based on distributed monitoring nodes. Distributed monitoring nodes can be deployed in different geographical locations to monitor critical nodes and critical links in real time, collect performance data, alarm information, etc., so as to timely discover and handle network failures. For example, monitoring nodes can be deployed in cities A, B, and C respectively to monitor the critical nodes router A and router B, as well as the critical links between them. Through monitoring, problems such as increased link delay and increased packet loss rate can be discovered in a timely manner, and corresponding measures can be taken to deal with them to ensure the stability and performance of the network. In this way, a more robust and efficient network management system can be built to improve the reliability and service quality of the network.

[0071] In some embodiments, Figure 3 As shown, in the above step S101, the multimodal AI model is generated by the multimodal data training, specifically including:

[0072] S1011, extracting time dimension features from the multimodal data, and obtaining first fault data by performing trend analysis, seasonal decomposition, and autocorrelation analysis on the time dimension features, where the first fault data includes a high-incidence period of faults and fault frequency features;

[0073] S1012, extracting spatial dimension features from the multimodal data, and performing spatial clustering, hot spot analysis, and spatial difference on the spatial dimension features to obtain second fault data, where the second fault data includes a high-incidence area of ​​faults and a spatial propagation path of faults;

[0074] S1013, forming a fault data training set by performing correlation analysis on the first fault data and the second fault data;

[0075] S1014, taking the fault data training set as input, adopting the LSTM model for modeling training to form a fault spatiotemporal evolution model, so that the fault spatiotemporal evolution model is used to predict the evolution process of the fault problem of the cross-regional network in time and space;

[0076] S1015, based on the fault spatiotemporal evolution model, using a Bayesian model to identify the root cause and key influencing factors of the fault, and constructing a directed acyclic graph between the fault cause and the fault manifestation;

[0077] S1016: Form a multimodal AI model based on the fault spatiotemporal evolution model and the directed acyclic graph.

[0078] In this embodiment, according to the timestamp information of the multimodal data, the time dimension characteristics of the data are extracted, including time point data and time period data accurate to the second level. The time series analysis algorithm is used to perform trend analysis, seasonal decomposition and autocorrelation analysis on the time dimension feature data to obtain the high-incidence time period distribution and frequency distribution characteristics of the fault event. According to the geographical location information of the multimodal data, the spatial dimension characteristics of the data are extracted, including longitude and latitude coordinate data, regional division data, etc. The spatial clustering algorithm is used to perform cluster analysis on the spatial dimension feature data to obtain the high-incidence area distribution of the fault event; at the same time, hot spot analysis methods such as kernel density analysis are used to determine the distribution of fault hot spots in each area. Through spatial interpolation analysis, the spatial distribution of the fault event is calculated by difference calculation to obtain the propagation path and impact range of the fault in space. The high-incidence time period and frequency characteristics of the fault obtained in the time dimension are associated with the high-incidence area and propagation path obtained in the spatial dimension to construct a fault spatiotemporal feature association matrix. According to the association matrix, data mining algorithms such as association rule mining and spatiotemporal sequence pattern mining are used to obtain the spatiotemporal evolution law of the fault occurrence, and form a complete fault data training set for subsequent fault prediction and diagnosis analysis.

[0079] The LSTM model is used to model the fault data training set. By learning the time series characteristics of the fault data, a model that can describe the temporal and spatial evolution of the fault is formed. The trained LSTM model is used to predict new network faults, determine the development trend and impact range of the fault in the future, and provide early warning for fault handling. On the basis of the LSTM model, the Bayesian model is introduced to identify the root cause and key influencing factors of the fault through probabilistic reasoning. According to the reasoning results of the Bayesian model, a directed acyclic graph between the fault cause and the fault manifestation is constructed to describe the causal relationship and mutual influence between the various factors. The LSTM model is integrated with the Bayesian model to form a multimodal AI model, which comprehensively utilizes the temporal and spatial characteristics and causal relationship of the fault to improve the accuracy of fault prediction and diagnosis. The multimodal AI model is used to monitor cross-regional networks in real time. When signs of faults are found, timely warnings are issued and possible causes are inferred to guide operation and maintenance personnel to conduct targeted fault troubleshooting and handling.

[0080] For example, to extract time dimension features from multimodal data, the data needs to be preprocessed first, and the data of different modes, such as logs, sensor readings, and user feedback, need to be time-aligned. Suppose there is a data set of a cross-regional network, including server logs, network traffic records, and user complaint records. By extracting the timestamps in each data modality, these data can be arranged in chronological order to form a unified time series data set. Trend analysis of time dimension features can be performed through time series analysis methods such as moving average and exponential smoothing to identify long-term trends in the data. For example, by calculating the average response time per hour in the server log, it can be found that the response time shows an upward trend in a specific period (such as weekday mornings), indicating that the network load is high during this period. Seasonal decomposition is used to identify periodic changes in the data. Suppose that by analyzing network traffic records, it is found that the traffic increases significantly every Friday afternoon. This may be due to seasonal fluctuations caused by employees intensively uploading and downloading files before the weekend. Autocorrelation analysis is used to detect the time dependency between data points. For example, by calculating the autocorrelation coefficient of user complaint records, it is found that the number of complaints is still significantly correlated after a one-day lag, indicating that the fault has a certain degree of persistence. Through the above analysis, the first fault data is obtained, such as the high-incidence period of faults is every Friday afternoon, and the fault frequency characteristic is an average of 3 times per hour.

[0081] To extract spatial dimension features from multimodal data, we first need to spatially locate the data and map each data point to a specific geographic location. Assuming that the servers are distributed in multiple data centers in different cities, each record in the server log can be associated with a specific data center through IP address resolution. Spatial clustering is used to identify high-incidence areas of faults. For example, by clustering network fault records using the K-means clustering algorithm, it is found that data center A in a certain city frequently fails, indicating that the area is a high-incidence area of ​​faults. Hot spot analysis further refines the spatial distribution of faults. Assuming that through hot spot map analysis, it is found that a specific rack in data center A frequently fails, which may be due to problems with the power module of the rack. Spatial difference is used to infer the spatial propagation path of the fault. For example, by calculating the time difference between the occurrence of faults in different data centers, it is found that the fault gradually spreads from data center A to the adjacent data center B, indicating that the fault has a certain degree of spatial propagation. Through the above analysis, the second fault data is obtained. For example, if the high-incidence area of ​​faults is data center A, the spatial propagation path of the fault is from data center A to data center B.

[0082] The first fault data and the second fault data are correlated and analyzed to form a fault data training set. For example, by correlating the high-incidence time period of faults in the time dimension with the high-incidence area of ​​faults in the space dimension, it is found that the fault rate of data center A is significantly higher than that of other time periods and areas every Friday afternoon, indicating that the faults in this time period and area have a strong correlation. The fault data training set is used as input, and the LSTM (long short-term memory) model is used for modeling training. The LSTM model is suitable for processing time series data and can capture long-term dependencies in the data. For example, by training the LSTM model, the probability of failure of data center A in a certain period in the future can be predicted, so that preventive measures can be taken in advance. Based on the fault spatiotemporal evolution model, the Bayesian model is used to identify the root cause and key influencing factors of the fault. The Bayesian model can infer the probability distribution of fault occurrence based on existing data. For example, through Bayesian network analysis, it is found that the main reason for the frequent failures of data center A is the aging of the power module, and the secondary reason is the excessive network load. Constructing a directed acyclic graph (DAG) between the cause of the fault and the manifestation of the fault can intuitively display the cause-effect relationship of the fault. For example, the DAG shows that the aging of the power module directly causes the server to crash, while the excessive network load indirectly affects the server response time, which ultimately leads to an increase in user complaints. A multimodal AI model is formed based on the fault spatiotemporal evolution model and the directed acyclic graph. The model combines the fault characteristics of the time dimension and the space dimension, and can comprehensively predict and diagnose fault problems in cross-regional networks. For example, through the multimodal AI model, the power module status and network load of data center A can be monitored in real time. Once an abnormality is found, an early warning is immediately issued, thereby effectively reducing the fault incidence and impact range. Through the above steps, not only can the accurate prediction and diagnosis of cross-regional network faults be achieved, but also targeted optimization suggestions can be provided to network operation and maintenance personnel to improve the overall stability and reliability of the network. Compared with single-dimensional fault detection, this multi-dimensional and multi-level analysis method has higher accuracy and practicality, and can significantly improve the efficiency and effectiveness of network management.

[0083] In some embodiments, Figure 4 As shown, in the above step S102, the multimodal AI model is deployed to each edge node of the cross-regional network, specifically including:

[0084] S1021, using a deep learning framework to obtain a model structure and parameter information of the multimodal AI model, and determining a pruning strategy according to the model structure and the parameter information;

[0085] S1022, quantizing the parameters and activation values ​​of the multimodal AI model using a uniform quantization method to obtain a quantization scheme;

[0086] S1023, based on the pruning strategy and the quantization scheme, construct a student model for the multimodal AI model through knowledge distillation technology, and fine-tune the student model to form an optimized multimodal AI model;

[0087] S1024, deploying the optimized multimodal AI model to each edge node of the cross-regional network, and evaluating the model performance of the optimized multimodal AI model to obtain an evaluation result;

[0088] S1025: Readjust the pruning strategy and the quantization scheme according to the evaluation result.

[0089] In this embodiment, the structural information and parameter information of the multimodal AI model in the deep learning framework are obtained, and a suitable pruning strategy is determined based on the obtained information; the parameters and activation values ​​of the multimodal AI model are quantized using a uniform quantization method to obtain a corresponding quantization scheme; according to the determined pruning strategy and quantization scheme, a corresponding student model is constructed for the multimodal AI model through knowledge distillation technology; the constructed student model is obtained, and the parameters of the student model are adjusted and optimized using a model fine-tuning method; according to the fine-tuned student model, it is determined whether its performance reaches the preset optimization goal, if so, it is used as the optimized multimodal AI model, otherwise it returns to the previous step to continue model fine-tuning; for the optimized multimodal AI model, a test data set is used to evaluate its performance to obtain indicators such as model accuracy and speed; according to the performance evaluation indicators, it is determined whether the optimized multimodal AI model meets the actual application requirements, if so, it is determined as the final optimized model and deployed for application, otherwise it returns to the pruning and quantization steps, and the model is optimized again after adjusting the strategy.

[0090] Obtain the optimized multimodal AI model and deploy it to each edge node of the cross-regional network; design a model performance evaluation scheme for the multimodal AI model deployed on the edge node, use appropriate evaluation indicators to evaluate the model's reasoning efficiency and performance, and obtain evaluation results; based on the evaluation results, determine whether the performance of the model on the edge node meets the preset performance threshold. If not, trigger the adjustment of the pruning strategy and quantization scheme; by analyzing the evaluation results, determine the specific parameters of the pruning strategy and quantization scheme that need to be adjusted, such as pruning ratio, quantization bit width, etc., and generate the adjusted pruning strategy and quantization scheme; apply the adjusted pruning strategy to the multimodal AI model, trim the network structure of the model, remove redundant and unimportant connections, and obtain the pruned model; apply the adjusted quantization scheme to the pruned model, quantize the model's weights and activation values, reduce the model's numerical accuracy, and obtain the quantized model; redeploy the pruned and quantized multimodal AI model to the edge nodes of the cross-regional network, conduct a new round of performance evaluation, and further optimize the model based on the evaluation results until the preset performance requirements are met.

[0091] Exemplarily, using a deep learning framework to obtain the model structure and parameter information of a multimodal AI model is the first step in optimizing the model. Assuming that the TensorFlow framework is used, the layer structure of the model, the number of parameters of each layer, and the specific values ​​of the parameters can be easily obtained through the API provided by it. For example, for a multimodal AI model containing a convolutional layer, a fully connected layer, and an LSTM layer, information such as the number of filters in the convolutional layer, the number of neurons in the fully connected layer, and the hidden state dimension of the LSTM layer can be obtained. This information provides a basis for subsequent pruning strategies. Determining the pruning strategy based on the model structure and parameter information is to reduce the redundant parts of the model and improve computational efficiency. Suppose that it is found that the weight values ​​of some filters in the convolutional layer are very small and close to zero, which indicates that these filters do not contribute much to the model. Therefore, a weight size pruning strategy can be adopted to directly remove filters with weights less than a certain threshold. For example, set the threshold to 0.01, and prune all filters with an absolute value of weight less than 0.01, thereby reducing the number of parameters of the model. The uniform quantization method is used to quantize the parameters and activation values ​​of the multimodal AI model in order to further compress the model size and make it more suitable for deployment on resource-constrained edge nodes. Uniform quantization maps floating-point parameters and activation values ​​to a limited integer range. For example, quantizing 32-bit floating-point numbers to 8-bit integers can significantly reduce the storage and computation requirements of the model. Through quantization, the parameters and activation values ​​of the model are limited to 256 possible values, thereby reducing the complexity of the model. Based on the pruning strategy and quantization scheme, the student model is constructed for the multimodal AI model through knowledge distillation technology in order to further reduce the model size while maintaining the model performance. Knowledge distillation is a technique that transfers the knowledge of a large and complex model (teacher model) to a small and simple model (student model). For example, a large multimodal AI model is first trained as a teacher model, and then its output probability distribution is used as a soft label to train a smaller student model. Assuming that the output probability distribution of the teacher model is [0.1, 0.6, 0.3], the student model can capture the decision logic of the teacher model by learning this distribution. Fine-tuning the student model is to ensure its performance in practical applications. Fine-tuning is usually achieved by continuing to train the student model on a small amount of labeled data. For example, the student model can be fine-tuned on a small amount of historical fault data across a regional network to better adapt it to the actual environment. Assume that during the fine-tuning process, the accuracy of the student model increases from 90% to 95%, indicating that fine-tuning effectively improves the performance of the model. The optimized multimodal AI model is deployed to each edge node of the cross-regional network to achieve real-time fault detection and prediction. Edge nodes usually have limited computing resources, so the optimized model needs to be lightweight. For example, the optimized model is deployed to the edge servers of each data center, which are responsible for real-time monitoring of the network status and issuing warnings when anomalies are detected.The purpose of evaluating the model performance of the optimized multimodal AI model is to verify its performance in the actual environment. Evaluation indicators can include accuracy, recall, F1 score, etc. The purpose of re-adjusting the pruning strategy and quantization scheme based on the evaluation results is to further improve the performance and efficiency of the model. Assuming that the evaluation results show that the model has poor detection effect on certain specific fault types, the pruning strategy can be adjusted to retain more filters related to these fault types. At the same time, if it is found that the performance of the quantized model has decreased significantly, the number of quantization bits can be appropriately increased, for example, from 8 bits to 16 bits, to balance performance and model size. Through the above steps, the optimized multimodal AI model can not only run efficiently on resource-constrained edge nodes, but also maintain a high level of fault detection and prediction accuracy. This comprehensive optimization strategy not only improves the practicality of the model, but also reduces the deployment and maintenance costs, providing a strong guarantee for the stable operation of cross-regional networks. Pruning and quantization reduce the redundancy and computational burden of the model, and knowledge distillation and fine-tuning ensure the performance of the model. The resulting optimized model performs well in practical applications and can effectively improve the efficiency and effectiveness of network management.

[0092] In some embodiments, Figure 5 As shown, in the above step S103, the traffic scheduling strategy of the cross-regional network is dynamically adjusted through a reinforcement learning algorithm according to the fault detection result and the real-time network status of the cross-regional network, specifically including:

[0093] S1031, obtaining real-time network status data of the cross-regional network, using the fault detection result and the implementation network status data as the state space, using the traffic scheduling strategy as the action space, and using the weighted combination of the delay rate, the packet loss rate and the load balancing degree as the reward function;

[0094] S1032, based on the state space, the action space and the reward function, a reinforcement learning algorithm is used to perform modeling training to obtain a traffic scheduling model, where the traffic scheduling model is used to dynamically adjust the traffic scheduling strategy of the cross-regional network.

[0095] In this embodiment, real-time status data of the cross-regional network is obtained, including indicators such as network delay, packet loss rate, and bandwidth utilization, to form a network status feature vector. Through the preset fault detection rules and thresholds, it is determined whether the network has a fault. If there is a fault, the fault information is merged with the network status feature vector as the state space of the reinforcement learning algorithm. Different traffic scheduling strategies are abstracted into action spaces, and each action corresponds to a traffic scheduling method, such as link load balancing, traffic speed limit, etc. According to indicators such as delay rate, packet loss rate, and load balancing, a reward function is designed to evaluate and score different traffic scheduling actions, and the weight of the reward function can be adjusted according to actual needs. A deep reinforcement learning algorithm, such as DQN or A3C, is used to perform modeling training based on state space, action space, and reward function, and the optimal traffic scheduling strategy is obtained through continuous trial and error and learning. The trained traffic scheduling model is deployed in the cross-regional network, and the network status data is collected in real time, input into the model, and the traffic scheduling strategy is dynamically generated. According to the generated traffic scheduling strategy, the flow table is issued through the SDN controller to dynamically schedule and optimize the traffic across regional networks to improve network performance and service quality.

[0096] Exemplarily, real-time network status data includes, but is not limited to, key indicators such as network delay, packet loss rate, and bandwidth utilization. For example, through the monitoring tools deployed at each edge node, the delay data of each link can be collected in real time, such as the delay of link A is 10 milliseconds and the delay of link B is 15 milliseconds; packet loss rate data, such as the packet loss rate of link C is 2%, and the packet loss rate of link D is 1%; and bandwidth utilization data, such as the bandwidth utilization of link E is 80%, and the bandwidth utilization of link F is 90%. The fault detection results and real-time network status data are used as state space to fully reflect the health status and operation status of the network. The fault detection results include information such as fault type, fault location, and fault severity. For example, a certain detection result shows that a server in data center X has a hardware failure with a fault severity of "high", while the network equipment in data center Y has a slight configuration error with a fault severity of "low". Combining this information with real-time network status data to form a multi-dimensional state space helps to more accurately evaluate the overall status of the network. Traffic scheduling strategies, as action spaces, cover a variety of possible traffic adjustment schemes. For example, the action space may include strategies such as transferring traffic from high-latency links to low-latency links, transferring traffic from high-packet-loss rate links to low-packet-loss rate links, and dynamically adjusting traffic allocation according to bandwidth utilization. Specifically, if the delay of link A suddenly increases to 20 milliseconds, while the delay of link B is still 15 milliseconds, the scheduling strategy may choose to transfer part of the traffic from link A to link B to reduce the overall network delay. The weighted combination of delay rate, packet loss rate, and load balancing degree is used as a reward function to comprehensively evaluate the effect of traffic scheduling strategies. Delay rate and packet loss rate directly reflect the service quality of the network, while load balancing degree reflects the utilization efficiency of network resources. For example, suppose the reward function is defined as: reward = -(delay rate × weight 1 + packet loss rate × weight 2 + load balancing degree × weight 3), where weight 1, weight 2, and weight 3 are 0.5, 0.3, and 0.2, respectively. If after a certain scheduling, the delay rate is reduced from 10% to 5%, the packet loss rate is reduced from 2% to 1%, and the load balancing degree is improved from 0.8 to 0.9, the calculated reward value will be significantly improved, indicating that the scheduling strategy is effective. Based on the state space, action space and reward function, the reinforcement learning algorithm is used for modeling and training in order to find the optimal traffic scheduling strategy through continuous trial and error and learning. Reinforcement learning algorithms such as Q-learning or DQN (DeepQ-Network) in deep reinforcement learning can be used for such tasks. During the training process, the model selects an action based on the current state at each time step, observes the new state and the reward obtained after execution, and gradually optimizes the strategy by continuously updating the Q value table or neural network parameters. After a large number of iterative training, the model can learn to select the optimal traffic scheduling strategy under different network conditions.The traffic scheduling model obtained is used to dynamically adjust the traffic scheduling strategy of the cross-regional network to cope with the real-time changing network conditions. For example, when a sudden failure occurs on a link and the delay increases sharply, the model can respond quickly and transfer the traffic to other healthy links, thereby minimizing the impact of the failure on network performance. For another example, during peak hours, the model can dynamically adjust the traffic distribution according to the bandwidth utilization of each link to avoid excessive congestion of a certain link and ensure the stable operation of the overall network. In this way, the traffic scheduling model can not only improve the response speed and service quality of the network, but also optimize resource utilization and reduce operating costs. For example, in actual applications, after adopting the optimized traffic scheduling strategy, the average network delay is reduced by 20%, the packet loss rate is reduced by 15%, and the bandwidth utilization is increased by 10%, which significantly improves the user experience and the overall performance of the network. In addition, this traffic scheduling model based on reinforcement learning is also characterized by strong adaptability and good robustness. Faced with complex network environments and diverse types of failures, the model can maintain an efficient scheduling strategy through continuous learning and adjustment to ensure the stability and reliability of the cross-regional network. When faced with new network attacks or large-scale sudden traffic, the model can quickly identify anomalies, adjust scheduling strategies, effectively respond to emergencies, and ensure the security and stable operation of the network. By obtaining real-time network status data, building a multi-dimensional state space, defining a comprehensive reward function, and using reinforcement learning algorithms for modeling and training, the traffic scheduling model finally obtained can dynamically and intelligently adjust the traffic scheduling strategy of cross-regional networks, comprehensively improve the performance and efficiency of the network, and provide users with high-quality network services.

[0097] In some embodiments, Figure 6 As shown, in the above step S103, the fault self-healing strategy is determined according to the fault detection result, specifically including:

[0098] S1041, determining the fault type and fault severity of the fault event according to the fault detection result;

[0099] S1042, determining matching candidate fault self-healing strategies from a pre-built fault self-healing strategy library according to the fault type and the fault severity;

[0100] S1043, according to the strategy evaluation index, use the heuristic algorithm to optimize and combine the candidate fault self-healing strategies to obtain the target fault self-healing strategy, wherein the strategy evaluation index includes the historical success rate, the execution time, and the degree of influence on the network performance.

[0101] In this embodiment, fault event information is obtained according to the fault detection result, and key features of the fault event are extracted; a rule-based method is used to determine the type and severity of the fault event according to the extracted key features of the fault event; a fault feature vector is constructed according to the fault type and severity; a pre-constructed fault self-healing strategy library is obtained, and a strategy feature vector is extracted for each fault self-healing strategy in the strategy library; a cosine similarity algorithm is used to calculate the similarity between the fault feature vector and each strategy feature vector, and the fault self-healing strategy corresponding to the strategy feature vector whose similarity is greater than a preset similarity threshold is used as a candidate fault self-healing strategy. For each candidate strategy, its score on each evaluation indicator is calculated based on historical data; according to the preset evaluation indicator weight parameters, the scores of each indicator of each candidate strategy are weighted and summed to obtain a comprehensive score; a simulated annealing algorithm is used, with the candidate strategy as the state space and the comprehensive score as the energy function, to search for the optimal strategy combination through state transition; at each state transition, whether to accept the new strategy combination state is determined according to the current temperature parameter; as the temperature parameter decreases, the probability of accepting an inferior solution is gradually reduced, and finally converges to the globally optimal target strategy combination; output the target fault self-healing strategy combination, and apply it to the fault handling process to achieve intelligent fault self-healing.

[0102] For example, suppose that during a network monitoring, the system detects that a server in data center X has a response timeout phenomenon. Through further analysis, it is determined that the server has a hardware failure, which is specifically manifested as a CPU overheating causing the system to crash. The severity of the failure is evaluated as "high" because the server carries critical services and its failure directly affects the access of a large number of users. According to the determined fault type and severity, the system selects matching candidate policies from the pre-built fault self-healing policy library. For example, for hardware failures, the policy library may contain the following candidate policies: 1. Restart the server: Applicable to temporary hardware failures, and restore services by restarting. 2. Switch to the backup server: Applicable to serious hardware failures, switch traffic to the backup server to maintain service continuity. 3. Trigger the hardware maintenance process: Applicable to hardware failures that require physical repair, notify the maintenance team to handle on-site. Next, use the heuristic algorithm to optimize the candidate combination. Assume that the policy evaluation indicators include historical success rate, execution time, and the degree of impact on network performance. The specific analysis is as follows: Historical success rate: The historical success rate of restarting the server is 80%, the success rate of switching to the backup server is 95%, and the success rate of triggering the hardware maintenance process is 70%. Execution time: Restarting the server takes 5 minutes, switching to the backup server takes 2 minutes, and triggering the hardware maintenance process takes 10 minutes. Impact on network performance: Restarting the server will cause a brief service interruption, which has a higher impact; switching to the backup server has a smaller impact and is almost seamless; triggering the hardware maintenance process requires suspending the service during the maintenance period, which has the highest impact. Through the heuristic algorithm, the system comprehensively considers the above indicators and optimizes the strategy combination. For example, the algorithm may give priority to strategies with a high historical success rate and a short execution time. In this case, switching to the backup server is selected as the target fault self-healing strategy because of its high success rate and short execution time. This strategy selection process is based not only on historical data, but also on real-time network conditions. If the current network load is light, restarting the server may not have much impact on the user experience, and the system may choose the restart strategy; in the case of high load, switching to the backup server is more secure.

[0103] For further example, suppose that in another scenario, a configuration error occurs in the network equipment of data center Y, and the fault severity is "low". At this time, the policy library may contain the following candidate policies: 1. Automatically reset configuration: Applicable to minor configuration errors, and the correct configuration is automatically restored through scripts. 2. Manual inspection and repair: Applicable to complex configuration errors, which require engineer intervention. 3. Temporarily isolate faulty devices: Applicable to configuration errors that cannot be repaired immediately to avoid affecting other devices. Analysis through evaluation indicators: Historical success rate: The success rate of automatic resetting configuration is 90%, the success rate of manual inspection and repair is 95%, and the success rate of temporarily isolating faulty devices is 80%. Execution time: Automatically resetting configuration takes 1 minute, manual inspection and repair takes 15 minutes, and temporarily isolating faulty devices takes 5 minutes. Impact on network performance: Automatically resetting configuration has a small impact, while manual inspection and repair has a large impact. Temporarily isolating faulty devices will cause some services to be interrupted. In this scenario, the system may choose automatic resetting configuration as the target policy because it has a high success rate and short execution time, and has a small impact on network performance. In this way, the system can not only respond to faults quickly, but also select the optimal self-healing strategy according to the actual situation to ensure the stability and reliability of network services. For example, during peak hours, the system will tend to choose strategies that have less impact on network performance to minimize the impact on users; while during off-peak hours, the system may choose strategies with longer execution time to completely solve the problem. In addition, the historical success rate reflects the reliability of the strategy, the execution time reflects the response speed, and the degree of impact on network performance is directly related to the user experience. Taking these indicators into consideration, it can ensure that the self-healing strategy strikes a balance between efficiency and stability. For example, in an actual application, the system detected high latency in a certain link, and after analysis, it was determined that the network device configuration was wrong, with a severity of "medium". The system screened out two candidate strategies from the policy library: automatic reset configuration and manual check and repair. Through evaluation, the success rate of automatic reset configuration was 85%, the execution time was 2 minutes, and the impact on network performance was small; the success rate of manual check and repair was 90%, the execution time was 10 minutes, and the impact on network performance was large. In the end, the system selected the automatic reset configuration strategy, successfully restored the link performance in a short time, and avoided large-scale service interruption. This strategy selection method based on multi-dimensional evaluation indicators not only improves the efficiency and success rate of fault self-healing, but also significantly improves the overall quality of network services and user satisfaction. By continuously optimizing and updating the policy library, the system can better cope with various complex network failures and ensure the stable operation of the network.

[0104] In some embodiments, Figure 7 As shown, the method also includes:

[0105] S301, obtaining historical network traffic data and historical self-healing execution status data, performing correlation analysis on the historical network traffic data and the historical self-healing execution status data, and obtaining a correlation feature set;

[0106] S302, based on the associated feature set, using a machine learning algorithm to construct a multivariate time series prediction model for network traffic and self-healing status;

[0107] S303, obtaining a network traffic trend and a self-healing status trend within a future preset time period according to the multivariate time series prediction model;

[0108] S304, determining whether a network failure may occur within a future preset time period by comparing the network traffic trend with a preset network traffic threshold;

[0109] S305, determining the fault recovery time after the fault self-healing strategy is executed according to the self-healing state trend, and determining whether a new fault self-healing strategy needs to be replaced by comparing the fault recovery time with a preset fault recovery time threshold.

[0110] In this embodiment, historical network traffic data and historical self-healing execution status data within a specified time period are obtained, and the two types of data are aligned by timestamp to form time series data. The aligned time series data is preprocessed, including missing value processing, outlier processing and data normalization, to obtain preprocessed time series data. A time series association analysis algorithm, such as mutual information method, Granger causality test, etc., is used to analyze the correlation between the preprocessed network traffic time series and the self-healing state time series, and obtain a significantly related set of associated features. According to the associated feature set, a multivariate time series prediction model for network traffic and self-healing state is constructed, and the machine learning algorithms that can be used include long short-term memory network (LSTM), gated recurrent unit network (GRU), etc. The constructed prediction model is trained using historical data, and the model is optimized using methods such as cross-validation to obtain the optimal multivariate time series prediction model. Based on the trained and optimized prediction model, the network traffic data and self-healing state data for a recent period of time are input to predict the network traffic trend and self-healing state trend in the future preset time period. Determine whether the network traffic trend exceeds a preset network traffic threshold; if so, predict that a network failure may occur within a preset time period in the future; based on the self-healing status trend, use a machine learning algorithm to predict the fault recovery time after executing the current fault self-healing strategy; determine whether the fault recovery time exceeds a preset fault recovery time threshold; if so, determine that a new fault self-healing strategy needs to be replaced; obtain a new fault self-healing strategy with the highest matching degree with the current network fault type from a pre-built fault self-healing strategy library; send the new fault self-healing strategy to the network management system, and execute the new fault self-healing strategy through the system to achieve rapid self-healing of network faults.

[0111] Exemplarily, historical network traffic data includes but is not limited to network bandwidth usage, packet transmission rate, delay, packet loss rate and other indicators; while historical self-healing execution status data covers the triggering time, execution time, recovery effect and other aspects of the self-healing strategy. Correlation analysis of these data can reveal the intrinsic connection between network traffic changes and self-healing strategy execution, forming a set of associated features. For example, in the historical data of a data center, hourly network traffic data and self-healing strategy execution are recorded in the past year. Through analysis, it is found that whenever the network traffic exceeds a certain threshold (such as 1000Mbps per second), the probability of triggering the self-healing strategy of hardware failure (such as restarting the server) increases significantly. Further analysis also found that after the self-healing strategy is executed, the network traffic will fluctuate in a short period of time, but then gradually return to stability. These findings constitute a set of associated features, such as "high traffic triggers hardware failure self-healing" and "traffic fluctuations after self-healing". Based on these sets of associated features, a multivariate time series prediction model is constructed using a machine learning algorithm. The model not only considers network traffic data, but also incorporates self-healing status data to predict the trend of network traffic and self-healing status in a preset time period in the future. For example, by using the long short-term memory network (LSTM) algorithm and inputting the network traffic and self-healing status data of the past week, the model can predict the traffic and self-healing status in the next 24 hours. Assume that the model prediction results show that the network traffic will peak in the next 24 hours, reaching 1200Mbps per second, and the probability of triggering the self-healing strategy is high. By comparing this predicted traffic trend with the preset traffic threshold (such as 1100Mbps per second), it can be determined that network failures may occur in the future. Next, the fault recovery time after the self-healing strategy is executed is determined based on the self-healing status trend. For example, the model predicts that during the peak traffic period, the self-healing strategy of restarting the server will take 8 minutes to restore service. By comparing this predicted recovery time with the preset fault recovery time threshold (such as 10 minutes), it can be determined whether the current strategy is effective. If the predicted recovery time exceeds the threshold, the system will consider replacing a new self-healing strategy. For example, during the peak traffic period, restarting the server may not be enough to quickly restore service. The system may choose to switch to the backup server strategy, which is expected to restore service in only 3 minutes. The application of this multivariate time series prediction model not only improves the accuracy of fault prediction, but also optimizes the selection of self-healing strategies. Through correlation analysis, the system can identify the strong correlation between high traffic and hardware fault self-healing strategies, so that preventive measures can be taken in advance during traffic peaks to reduce the probability of failures. In addition, the prediction results of the model can also be used to dynamically adjust network resource configuration. For example, when a traffic peak is predicted, the system can increase bandwidth or optimize routing strategies in advance to reduce network load and reduce the risk of failure. In this way, the system can not only respond to failures quickly, but also select the optimal self-healing strategy based on actual conditions to ensure the stability and reliability of network services.For example, during peak hours, the system will tend to choose strategies that have less impact on network performance to minimize the impact on users; during off-peak hours, the system may choose strategies that take longer to execute in order to completely solve the problem. The construction of a set of associated features and the application of machine learning models enable the system to extract valuable information from historical data and form an accurate prediction of future network status. This predictive capability not only improves the efficiency and success rate of fault self-healing, but also significantly improves the overall quality of network services and user satisfaction. By continuously optimizing and updating the policy library, the system can better cope with various complex network failures and ensure the stable operation of the network.

[0112] Reference Figure 8 An embodiment of the present invention provides a cross-region network failure prevention and repair system 8, and the system 8 specifically includes:

[0113] The first preventive repair module 801 is used to collect multimodal data across regional networks according to distributed monitoring nodes, and generate a multimodal AI model through training of the multimodal data based on the quantum computing platform;

[0114] The second preventive repair module 802 is used to deploy the multimodal AI model to each edge node of the cross-regional network, perform fault detection on each edge node according to the multimodal AI model, and obtain a fault detection result;

[0115] The third preventive repair module 803 is used to dynamically adjust the traffic scheduling strategy of the cross-regional network through a reinforcement learning algorithm according to the fault detection result and the real-time network status of the cross-regional network;

[0116] The fourth preventive repair module 804 is used to determine a fault self-healing strategy according to the fault detection result, perform a fault self-healing operation on the cross-regional network according to the fault self-healing strategy, and obtain fault self-healing execution status data;

[0117] The fifth preventive repair module 805 is used to determine whether the fault self-healing operation is successful according to the fault self-healing execution status data, and trigger an alarm process if the fault self-healing operation is not successful.

[0118] It is understandable that if Figure 1 The contents of the cross-regional network failure prevention and repair method embodiment shown in the figure are applicable to the cross-regional network failure prevention and repair system embodiment. The functions specifically implemented by the cross-regional network failure prevention and repair system embodiment are similar to those in the embodiment shown in the figure. Figure 1 The embodiment of the method for preventing and repairing cross-regional network failures shown in FIG. 1 is the same as that shown in FIG. 1 , and the beneficial effects achieved are the same as those of FIG. Figure 1 The beneficial effects achieved by the embodiment of the cross-regional network failure prevention and repair method shown are also the same.

[0119] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0120] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0121] Reference Fig. 9 The embodiment of the present invention further provides a computer device 9, comprising: a memory 902 and a processor 901 and a computer program 903 stored in the memory 902. When the computer program 903 is executed on the processor 901, a method for preventing and repairing cross-regional network failures as described in any one of the above methods is implemented.

[0122] The computer device 9 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 9 may include, but is not limited to, a processor 901 and a memory 902. Those skilled in the art will appreciate that Fig. 9 It is only an example of the computer device 9 and does not constitute a limitation on the computer device 9. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0123] The processor 901 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0124] In some embodiments, the memory 902 may be an internal storage unit of the computer device 9, such as a hard disk or memory of the computer device 9. In other embodiments, the memory 902 may also be an external storage device of the computer device 9, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 9. Further, the memory 902 may also include both an internal storage unit of the computer device 9 and an external storage device. The memory 902 is used to store an operating system, an application program, a boot loader (Boot Loader), data, and other programs, such as the program code of the computer program, etc. The memory 902 may also be used to temporarily store data that has been output or is to be output.

[0125] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method for preventing and repairing cross-regional network failures as described in any one of the above methods is implemented.

[0126] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0127] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0128] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0129] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0130] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Claims

1. A method for preventing and repairing cross-regional network failures, characterized in that: The method specifically comprises: Collect multimodal data across regional networks according to distributed monitoring nodes, and generate a multimodal AI model through training of the multimodal data based on the quantum computing platform; Deploy the multimodal AI model to each edge node of the cross-regional network, perform fault detection on each edge node according to the multimodal AI model, and obtain a fault detection result; According to the fault detection result and the real-time network status of the cross-regional network, dynamically adjust the traffic scheduling strategy of the cross-regional network through a reinforcement learning algorithm; Determine a fault self-healing strategy according to the fault detection result, perform a fault self-healing operation on the cross-regional network through the fault self-healing strategy, and obtain fault self-healing execution status data; Determine whether the fault self-healing operation is successful according to the fault self-healing execution status data; if the fault self-healing operation is not successful, trigger an alarm process.

2. The method according to claim 1, characterized in that The method of collecting multimodal data across regional networks according to distributed monitoring nodes also includes: Acquire network device attribute data and link connection attribute data of the cross-regional network, construct a node feature vector according to the network device data, and construct an edge feature vector according to the link connection attribute data; According to the node feature vector and the edge feature vector, a network topology graph is constructed using a graph neural network model, and the network topology graph is calculated through a graph convolution operation to determine the importance score of each node feature vector; By comparing the importance score of each node feature vector with a preset score threshold, a set of key nodes is determined; Detecting whether there is a link connection between every two key nodes in the key node set, and if there is a link connection, determining the link between the two key nodes as a key link; Monitor each key node and each key link of the cross-regional network based on distributed monitoring nodes.

3. The method according to claim 1, characterized in that The generating a multimodal AI model through the multimodal data training specifically includes: Extracting time dimension features from the multimodal data, and obtaining first fault data by performing trend analysis, seasonal decomposition, and autocorrelation analysis on the time dimension features, wherein the first fault data includes a high-incidence period of faults and fault frequency features; Extracting spatial dimension features from the multimodal data, and obtaining second fault data by performing spatial clustering, hot spot analysis, and spatial difference on the spatial dimension features, wherein the second fault data includes a high-incidence area of ​​faults and a spatial propagation path of faults; Forming a fault data training set by performing correlation analysis on the first fault data and the second fault data; The fault data training set is used as input, and the LSTM model is used for modeling training to form a fault spatiotemporal evolution model. The fault spatiotemporal evolution model is used to predict the evolution of fault problems in cross-regional networks in time and space. Based on the fault spatiotemporal evolution model, a Bayesian model is used to identify the root cause and key influencing factors of the fault, and a directed acyclic graph between the fault cause and the fault manifestation is constructed; A multimodal AI model is formed according to the fault spatiotemporal evolution model and the directed acyclic graph.

4. The method according to claim 1, characterized in that: Deploying the multimodal AI model to each edge node of the cross-regional network specifically includes: Using a deep learning framework to obtain the model structure and parameter information of the multimodal AI model, and determining a pruning strategy according to the model structure and the parameter information; Quantizing the parameters and activation values ​​of the multimodal AI model using a uniform quantization method to obtain a quantization scheme; Based on the pruning strategy and the quantization scheme, a student model is constructed for the multimodal AI model through knowledge distillation technology, and the student model is fine-tuned to form an optimized multimodal AI model; Deploy the optimized multimodal AI model to each edge node of the cross-regional network, and evaluate the model performance of the optimized multimodal AI model to obtain an evaluation result; The pruning strategy and the quantization scheme are readjusted according to the evaluation result.

5. The method according to claim 1, characterized in that The method of dynamically adjusting the traffic scheduling strategy of the cross-regional network by a reinforcement learning algorithm according to the fault detection result and the real-time network status of the cross-regional network specifically includes: Acquire real-time network status data of the cross-regional network, use the fault detection result and the implemented network status data as the state space, use the traffic scheduling strategy as the action space, and use the weighted combination of delay rate, packet loss rate and load balancing degree as the reward function; Based on the state space, the action space and the reward function, a reinforcement learning algorithm is used to perform modeling training to obtain a traffic scheduling model, and the traffic scheduling model is used to dynamically adjust the traffic scheduling strategy of the cross-regional network.

6. The method according to claim 1, characterized in that The determining of the fault self-healing strategy according to the fault detection result specifically includes: Determining the fault type and fault severity of the fault event according to the fault detection result; According to the fault type and the fault severity, determining matching candidate fault self-healing strategies from a pre-built fault self-healing strategy library; According to the strategy evaluation index, a heuristic algorithm is used to optimize and combine the candidate fault self-healing strategies to obtain the target fault self-healing strategy. The strategy evaluation index includes the historical success rate, the execution time, and the degree of influence on the network performance.

7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: Acquire historical network traffic data and historical self-healing execution status data, perform correlation analysis on the historical network traffic data and the historical self-healing execution status data, and acquire a correlation feature set; Based on the associated feature set, a multivariate time series prediction model of network traffic and self-healing status is constructed using a machine learning algorithm; According to the multivariate time series prediction model, the network traffic trend and self-healing status trend within a future preset time period are obtained; By comparing the network traffic trend with a preset network traffic threshold, determining whether a network failure may occur within a preset time period in the future; The fault recovery time after the fault self-healing strategy is executed is determined according to the self-healing state trend, and whether a new fault self-healing strategy needs to be replaced is determined by comparing the fault recovery time with a preset fault recovery time threshold.

8. A system for preventing and repairing cross-regional network failures, characterized in that: The system specifically comprises: A first prevention and repair module is used to collect multimodal data across regional networks according to distributed monitoring nodes, and generate a multimodal AI model through training of the multimodal data based on the quantum computing platform; A second prevention and repair module is used to deploy the multimodal AI model to each edge node of the cross-regional network, perform fault detection on each edge node according to the multimodal AI model, and obtain a fault detection result; A third prevention and repair module is used to dynamically adjust the traffic scheduling strategy of the cross-regional network through a reinforcement learning algorithm according to the fault detection result and the real-time network status of the cross-regional network; A fourth prevention and repair module is used to determine a fault self-healing strategy according to the fault detection result, perform a fault self-healing operation on the cross-regional network through the fault self-healing strategy, and obtain fault self-healing execution status data; The fifth preventive repair module is used to determine whether the fault self-healing operation is successful according to the fault self-healing execution status data, and if the fault self-healing operation is not successful, trigger an alarm process.

9. A computer device, characterized in that: include: A memory, a processor and a computer program stored in the memory, when the computer program is executed on the processor, implements the method for preventing and repairing cross-regional network failures as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the method for preventing and repairing cross-regional network failures as described in any one of claims 1 to 7 is implemented.

Images