Cross-domain network quality detection method, device, equipment and computer program product

Abstract

The embodiments of the present invention relate to the field of network technology, and specifically provide a method, device, equipment and computer program product for quality detection of a cross-domain network. The method includes: obtaining a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network; determining time series data and spatial topology data corresponding to the cross-domain network, the time series data including network data of the cross-domain network that changes over time, and the spatial topology data for identifying the structural characteristics of the cross-domain network; inputting the time series data and the spatial topology data into the neural network model to obtain the quality detection information of the cross-domain network output by the neural network model. This embodiment can efficiently and accurately perform quality detection operations on the cross-domain network, and the obtained quality detection information helps to ensure the timeliness and reliability of responding to the cross-domain network.

Description

Technical Field

[0001] The present invention relates to the field of network technology, and in particular to a method, device, equipment and computer program product for quality detection of a cross-domain network. Background Art

[0002] Network quality analysis (Network Quality Analysis) refers to the process of evaluating and testing network performance. Specifically, network performance evaluation and testing operations can be achieved through key indicators such as network latency, throughput, and packet loss rate. Effective network quality analysis helps to promptly discover and resolve network bottlenecks and faults.

[0003] At present, related technologies mostly rely on static data to implement network quality detection operations. Specifically, fixed detection points are set in the network to be detected, and predefined performance indicator thresholds (for example, delay indicator thresholds, packet loss rate indicator thresholds, etc.) are set. The performance indicator thresholds are then used to perform network quality analysis operations. When the detection indicator exceeds the predefined performance indicator threshold, it can be determined that there is a fault in the network.

[0004] However, the above network quality detection operation relies on predefined performance indicator thresholds, which causes the network to have a response delay defect. Specifically, when the user perceives that the network quality has deteriorated, the network quality problem may have occurred for a long time, which may easily cause the network to fail to respond in time. Summary of the invention

[0005] The embodiments of the present invention provide a method, apparatus, device and computer program product for detecting quality of a cross-domain network, which can realize the detection operation of network quality and help ensure the timeliness and reliability of the cross-domain network response.

[0006] In a first aspect, an embodiment of the present invention provides a method for detecting quality of a cross-domain network, including:

[0007] Acquire a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network, wherein the neural network model is obtained by training based on preset network sample data;

[0008] Determine time series data and spatial topology data corresponding to the cross-domain network, the time series data is used to identify state characteristics of the cross-domain network that change over time, and the spatial topology data is used to identify structural characteristics of the cross-domain network;

[0009] The time series data and the spatial topology data are input into the neural network model to obtain the quality detection information of the cross-domain network output by the neural network model.

[0010] In a second aspect, an embodiment of the present invention provides a quality detection device for a cross-domain network, including:

[0011] A first acquisition module, used to acquire a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network, wherein the neural network model is obtained by training based on preset network sample data;

[0012] A first determination module is used to determine time series data and spatial topology data corresponding to the cross-domain network, the time series data is used to identify state characteristics of the cross-domain network that change over time, and the spatial topology data is used to identify structural characteristics of the cross-domain network;

[0013] The first processing module is used to input the time series data and the spatial topology data into the neural network model to obtain the quality detection information of the cross-domain network output by the neural network model.

[0014] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein when the one or more computer instructions are executed by the processor, the cross-domain network quality detection method in the above-mentioned first aspect is implemented.

[0015] In a fourth aspect, an embodiment of the present invention provides a computer storage medium for storing a computer program, wherein the computer program enables a computer to implement the cross-domain network quality detection method in the first aspect when executed.

[0016] In a fifth aspect, an embodiment of the present invention provides a computer program product, comprising: a computer-readable storage medium storing computer instructions, which, when executed by one or more processors, causes the one or more processors to execute the steps in the cross-domain network quality detection method shown in the first aspect above.

[0017] The cross-domain network quality detection method, device, equipment and computer program product provided by the embodiments of the present invention obtain the cross-domain network to be detected and the neural network model used to implement the quality detection operation of the cross-domain network, determine the time series data and spatial topology data corresponding to the cross-domain network, and then use the neural network model to perform network detection operations on the time series data and spatial topology data, so that the quality detection information of the cross-domain network can be stably obtained. In addition, since the quality detection operation of the cross-domain network is implemented by combining time characteristics and spatial characteristics, the quality and efficiency of the network quality detection operation can be effectively guaranteed. The quality detection information of the cross-domain network obtained can timely and accurately reflect the operating status of the cross-domain network. When the quality detection information identifies that there is a fault in the cross-domain network, the timeliness and reliability of the network response can be guaranteed through the quality detection information, further ensuring the practicality of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A schematic diagram of a scenario of a cross-domain network quality detection method provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of a flow chart of a method for detecting quality of a cross-domain network provided by an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of a process for obtaining a neural network model for implementing a quality detection operation across a cross-domain network provided by an embodiment of the present invention;

[0022] Figure 4 A schematic diagram of performing a fault configuration operation in the network topology diagram based on the fault configuration information to obtain training sample data provided by an embodiment of the present invention;

[0023] Figure 5 A schematic diagram of a process for performing network detection on the time series data and spatial topology data using the neural network model provided in an embodiment of the present invention to obtain quality detection information of a cross-domain network;

[0024] Figure 6 A flow chart of another cross-domain network quality detection method provided by an embodiment of the present invention;

[0025] Figure 7A principle block diagram of training a spatiotemporal graph convolutional neural network model provided by an application embodiment of the present invention;

[0026] Figure 8 A schematic diagram of the principle of a cross-domain network quality detection method provided by an application embodiment of the present invention;

[0027] Fig. 9 A schematic diagram of a system for quality detection of a cross-domain network provided by an application embodiment of the present invention;

[0028] Fig.10 A schematic diagram of a data processing method provided by an embodiment of the present invention;

[0029] Fig.11 A schematic diagram of the structure of a cross-domain network quality detection device provided by an embodiment of the present invention;

[0030] Fig.12 For Fig.11 A schematic diagram of the structure of an electronic device corresponding to the cross-domain network quality detection device provided in an embodiment is shown. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "an", and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two, but does not exclude the situation of including at least one.

[0033] It should be understood that the term "and / or" used in this article is only a way to describe the association relationship of associated devices, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated devices before and after are in an "or" relationship.

[0034] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0035] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a product or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such a product or system. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the product or system including the elements.

[0036] In addition, the step sequence in the following method embodiments is only an example and not a strict limitation.

[0037] Terminology explanation:

[0038] Spatio-Temporal Graph Convolutional Network is a convolutional neural network model that combines spatiotemporal data and graph structure. It can be used to process dynamic data with temporal and spatial characteristics, such as network traffic, node status, etc.; the model models the adjacency relationship of the graph through convolution operations, so that it can stably extract spatiotemporal features in the data.

[0039] PingMesh Detection Technology: PingMesh Detection Technology is a technology based on distributed network detection. It achieves real-time measurement of network latency and quality by selecting multiple detection points and optimizing detection paths. This technology can effectively improve the flexibility and accuracy of performance analysis of cross-domain networks.

[0040] Network Quality Analysis: Network Quality Analysis refers to the process of evaluating and testing network performance, mainly covering key indicators such as latency, throughput, and packet loss rate; effective network quality analysis helps to promptly discover and resolve network bottlenecks and faults.

[0041] Data fusion: Data fusion refers to the process of integrating data from different sources to obtain more comprehensive and accurate information; data fusion operations can improve the accuracy and decision-making ability of network quality assessment.

[0042] Backbone Network Topology Data: Backbone Network Topology Data, including data used to describe sites, physical distances, and connection relationships in the backbone network.

[0043] GraphSAGE Model: GraphSAGE Model is a graph neural network model used to generate node embeddings and supports neighbor sampling.

[0044] Edge GAT model: Edge Graph Attention Network is a graph neural network model that processes side information through the attention mechanism to achieve information transmission.

[0045] In order to facilitate understanding of the specific implementation process and implementation effect of the cross-domain network quality detection method, device, equipment and computer program product in this embodiment, the relevant technology is briefly described below:

[0046] With the rapid development of the digital economy and the continuous evolution of information technology, cross-domain networks have become the infrastructure for various application scenarios, especially in large enterprises and cloud computing services. In today's era of rapid development of digitalization and information networks, the reliability and stability of network quality have become key factors in various application scenarios. For cloud network products, they can provide enterprises with a convenient and efficient cross-cloud network management solution. However, although cloud network products have significant advantages in resource scheduling and cross-domain connections, real-time detection and management of network quality still face many challenges.

[0047] Specifically, the following two methods are provided in the related art to implement network detection operations:

[0048] Implementation method 1: Generate a network detection tool through PingMesh technology. Since the PingMesh detection data obtained by PingMesh technology is an important indicator of network quality, the network detection tool can detect the delay and packet loss in the network environment in real time, and then detect the abnormalities in the time dimension through the delay and packet loss data obtained by the detection, that is, perform comprehensive network performance detection and fault diagnosis functions based on delay and packet loss, and be able to aggregate and converge from device granularity (granularity of a single device) to coarse granularity (AZ, Region), where coarse granularity can include AZ granularity (computer room granularity) and Region granularity (regional granularity). When there is a coarse-grained fault in the network, a high-level telephone alarm can be issued, which can not only provide users with powerful network performance analysis capabilities, but also ensure high availability and performance of the network.

[0049] However, when there are large-scale faults in the network, the root cause troubleshooting of the network fault may be misleading due to unclear convergence of detection data in the network. At this time, not only are there many alarms, but even when alarms exist on multiple links at the same time, convergence alarms that are not the source of the fault may be generated.

[0050] Implementation method 2: Rely on static data and point detection technologies to implement network quality detection operations. Specifically, set fixed detection points in the network and set predefined performance indicator thresholds (such as latency, packet loss rate, etc.) to analyze network quality. When the detected performance indicator exceeds the predefined performance indicator threshold, an alarm will be triggered or an event will be recorded.

[0051] However, existing network quality detection mostly relies on static data analysis, which cannot quickly respond to complex traffic changes in cross-domain networks, and is prone to network bottlenecks and quality degradation, affecting user experience and data processing continuity. Especially in high-demand application scenarios, especially online games and live video broadcasts, network problems such as network delays and packet loss can directly affect the operation of the scenario and customer satisfaction; specifically, the above implementation has the following disadvantages:

[0052] (1) Static: The detection method with a fixed threshold cannot adapt to the dynamic changes in the network environment. In addition, it may not be reasonable to use the same threshold in different time periods and traffic patterns, which may easily lead to false positives or false negatives.

[0053] (2) Lack of intelligent analysis: The inability to use historical data for learning and model optimization leads to insufficient understanding of complex network behaviors such as cross-domain networks and inability to make accurate performance predictions;

[0054] (3) Response delay: Since the above network detection operation is triggered based on the threshold, the network may fail to respond to instantaneous network problems in a timely manner. When users perceive that the network quality has deteriorated, the problem may have occurred for a long time.

[0055] (4) Single indicator reliance: Usually focusing on a certain performance indicator, ignoring the comprehensive analysis of multi-dimensional data, which may lead to an incomplete assessment of network status.

[0056] In order to solve the above technical problems, the embodiments of the present invention provide a method, device, equipment and computer program product for quality detection of a cross-domain network, wherein, with reference to the attached Figure 1As shown, the execution subject of the quality detection method of the cross-domain network is the quality detection device 200 of the cross-domain network, and the quality detection device 200 of the cross-domain network is communicatively connected with the client 100. Among them, when the quality detection device 200 of the cross-domain network is implemented as a server in the cloud, the quality detection method of the cross-domain network can be executed in the cloud, and several computing nodes (cloud servers) can be deployed in the cloud, and each computing node has computing, storage and other processing resources. In the cloud, multiple computing nodes can be organized to provide a certain service. Of course, a computing node can also provide one or more services. The way to provide the service in the cloud can be to provide a service interface to the outside, and the user calls the service interface to use the corresponding service. The service interface includes software development kits (Software Development Kit, referred to as SDK), application programming interfaces (Application Programming Interface, referred to as API), etc.

[0057] Among them, the cross-domain network quality detection device 200 refers to a device that can provide a cross-domain network quality detection operation in a network virtual environment, and generally refers to a device that uses the network to perform information planning and cross-domain network quality detection operations. In physical implementation, the cross-domain network quality detection device 200 can be any device that can provide computing services, respond to cross-domain network quality detection requests, and perform cross-domain network quality detection operations, for example: it can be a cluster server, a conventional server, a cloud server, a cloud host, a virtual center, etc. The cross-domain network quality detection device 200 mainly includes a processor, a hard disk, a memory, a system bus, etc., which is similar to a general computer architecture.

[0058] The client 100 can be implemented as a mobile phone, a personal computer PC, a tablet computer, a setting application, etc. In addition, the basic structure of the client 100 may include: at least one processor. The number of processors depends on the configuration and type of the client 100. The client 100 may also include a memory, which may be volatile, such as: random access memory (Random Access Memory, referred to as RAM), or non-volatile, such as: read-only memory (Read-Only Memory, referred to as ROM), flash memory, etc., or may include both types. The memory usually stores an operating system (Operating System, referred to as OS), one or more application programs, and may also store program data, etc. In addition to the processing unit and the memory, the client 100 also includes some basic configurations, such as: a network card chip, an IO bus, a display component, and some peripheral devices, etc. Optionally, some peripheral devices may include, for example: a keyboard, a mouse, an input pen, a printer, etc. Other peripheral devices are well known in the art and will not be described in detail here.

[0059] In the above embodiment, the client 100 is connected to the cross-domain network quality detection device 200 through a network connection, and the network connection can be a wireless or wired network connection. If the client 100 is connected to the cross-domain network quality detection device 200 through a communication connection, the network standard of the mobile network can be any one of 2G (GSM), 2.5G (GPRS), 3G (WCDMA, TD-SCDMA, CDMA2000, UTMS), 4G (LTE), 4G+ (LTE+), WiMax, 5G, 6G, etc.

[0060] The client 100 is used for application by the user, so that when the user has a quality detection requirement for the cross-domain network, the quality detection operation of the cross-domain network can be triggered. Specifically, when the user has a quality detection requirement for the cross-domain network to be detected, the client 100 can obtain the quality detection request of the cross-domain network to be detected. In some examples, the client 100 can display a human-computer interaction interface, obtain the execution operation for the cross-domain network through the human-computer interaction interface, and obtain the quality detection request for the cross-domain network to be detected based on the execution operation; or, the client 100 can be configured with a voice recognition module, and then the voice recognition module can be used to obtain the voice interaction operation, and then the quality detection request of the cross-domain network to be detected can be obtained based on the voice interaction operation. After obtaining the quality detection request, the quality detection request can be sent to the quality detection device 200 of the cross-domain network, so that the quality detection operation of the corresponding cross-domain network is implemented based on the quality detection request by the quality detection device 200 of the cross-domain network.

[0061] The quality detection device 200 of the cross-domain network is connected to the client 100 for communication, and is used to obtain a quality detection request sent by the client 100, and then can implement the quality detection operation of the cross-domain network based on the quality detection request. Specifically, the cross-domain network to be detected and the neural network model for implementing the quality detection operation of the cross-domain network can be obtained based on the quality detection request, wherein the neural network model is obtained by training based on preset network sample data, and the network sample data may include time series sample data and spatial topology sample data; then determine the time series data and spatial topology data corresponding to the cross-domain network, and the time series data is used to identify the state characteristics of the cross-domain network that change over time. In some instances, the time series data may include at least: network delay and packet loss rate; the spatial topology data is used to identify the structural characteristics of the cross-domain network; in some instances, the spatial topology data may include at least: the physical distance between each regional network, the number of optical cables, and the network connection topology relationship; and a neural network model is used to perform network detection operations on the time series data and the spatial topology data, that is, the time series data and the spatial topology data are input into the neural network model to obtain quality detection information of the cross-domain network output by the neural network model, and the quality detection information is used to identify whether there is fault information in the cross-domain network, thereby realizing efficient quality detection operations on the cross-domain network.

[0062] In this embodiment, when it is necessary to perform quality detection operations on the cross-domain network, the obtained neural network model is used to perform network detection operations on the time series data and spatial topology data of the cross-domain network, so that the quality detection information of the cross-domain network can be stably obtained. Since the time characteristics and spatial characteristics are combined to realize the quality detection operation of the cross-domain network, the quality and efficiency of the network quality detection operation can be effectively guaranteed. In addition, the obtained quality detection information of the cross-domain network can timely and accurately reflect the operating status of the cross-domain network. When the quality detection information identifies a fault in the cross-domain network, the timeliness and reliability of the cross-domain network response can be guaranteed through the quality detection information, further ensuring the practicality of the method.

[0063] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the case where there is no conflict between the embodiments, the following embodiments and the features in the embodiments can be combined with each other. In addition, the step sequence in the following method embodiments is only an example and not a strict limitation.

[0064] Figure 2 A flow chart of a method for detecting quality of a cross-domain network provided by an embodiment of the present invention; Figure 2As shown, this embodiment provides a quality detection method for a cross-domain network. The execution subject of the method is a quality detection device for a cross-domain network. The quality detection device for a cross-domain network can be implemented as software, or a combination of software and hardware. When the quality detection device for a cross-domain network is implemented as hardware, it can be various electronic devices with cross-domain network quality detection capabilities, including but not limited to personal computers, servers, etc. When the quality detection device for a cross-domain network is implemented as software, it can be installed in the electronic devices listed above. Based on the above-mentioned quality detection device for a cross-domain network, the method in this embodiment may include:

[0065] Step S201: Acquire a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network.

[0066] Step S202: determining time series data and spatial topology data corresponding to the cross-domain network, the time series data being used to identify state characteristics of the cross-domain network that change over time, and the spatial topology data being used to identify structural characteristics of the cross-domain network.

[0067] Step S203: input the time series data and the spatial topology data into the neural network model to obtain the quality detection information of the cross-domain network output by the neural network model.

[0068] The specific implementation process and effects of each of the above steps are described in detail below:

[0069] Step S201: Acquire a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network.

[0070] Among them, a cross-domain network refers to a network system composed of multiple heterogeneous networks, which may be geographically dispersed, with different protocols or managed by different vendors; for example, a cross-domain network may include local area networks located in multiple different regions, that is, a cross-domain network may include a local area network in region A (or city a), a local area network in region B (or city b), a local area network in region C (or city c), and so on.

[0071] When a user has a demand for quality detection of a cross-domain network, a quality detection device for the cross-domain network (hereinafter referred to as the "quality detection device") can obtain the cross-domain network to be detected and a neural network model for implementing the quality detection operation of the cross-domain network. In some instances, the cross-domain network can be obtained through human-computer interaction operations. At this time, obtaining the cross-domain network to be detected may include: displaying a human-computer interaction interface; obtaining user input to execute operations in the human-computer interaction interface; and obtaining the cross-domain network to be detected based on the execution operations.

[0072] In some other instances, the cross-domain network can be obtained not only through human-computer interaction operations, but also through a preset device (for example, a client). In this case, obtaining the cross-domain network to be detected may include: determining a preset device that is communicatively connected to the quality detection device, wherein the preset device stores the cross-domain network to be detected; and actively or passively obtaining the cross-domain network to be detected through the preset device.

[0073] Since the quality detection operation of a cross-domain network often needs to process multiple data sources and evaluation indicators, in order to improve the quality and efficiency of the quality detection of the cross-domain network, a neural network model for implementing the quality detection operation of the cross-domain network can be obtained, wherein the neural network model can be obtained by training based on preset network sample data. Furthermore, the preset network sample data may include: preset time series sample data and spatial sample data. At this time, the neural network model can be stored in a preset area or a preset device, and the neural network model can be obtained by accessing the preset area or the preset device.

[0074] Step S202: determining time series data and spatial topology data corresponding to the cross-domain network, the time series data being used to identify state characteristics of the cross-domain network that change over time, and the spatial topology data being used to identify structural characteristics of the cross-domain network.

[0075] In order to combine time dimension data and space dimension data to efficiently perform quality detection operations on cross-domain networks, after obtaining the cross-domain network, time series data corresponding to the cross-domain network can be determined, wherein the time series data is used to identify the state characteristics of the cross-domain network that change over time. In some instances, the time series data may at least include: network delay and packet loss rate corresponding to each node in the cross-domain network, and the time series data can be obtained by detecting the cross-domain network through PingMesh detection technology. Specifically, PingMesh detection technology is used to generate a tool for detecting and probing network performance, and then the detection tool is used to detect the cross-domain network, so as to record the round-trip delay (RTT) of the data packets in the cross-domain network. The obtained round-trip delay can determine the number of data packets lost during data transmission and calculate the number of routers passed, so as to identify the complexity of the network path.

[0076] For time series data, it can not only include the network delay and packet loss rate corresponding to each node in the cross-domain network, but also include other indicators for evaluating the health of the network. For example, time series data can also include the number of hops, bandwidth information and jitter information corresponding to each node in the cross-domain network. In practical applications, the PingMesh detection tool can be used to measure the actual data transmission rate and evaluate the rate of change of network delay (i.e. jitter). The obtained data transmission rate and change rate can provide comprehensive network performance data, which is helpful to help optimize network conditions and solve potential problems in cross-domain networks.

[0077] For a cross-domain network, in order to efficiently and accurately perform quality detection operations on the cross-domain network, it is necessary not only to determine the time series data corresponding to the cross-domain network, but also to determine the spatial topology data corresponding to the cross-domain network, wherein the spatial topology data is used to identify the structural characteristics of the cross-domain network, which may refer to the real site data of the cross-domain network. In some instances, the spatial topology data may at least include: the physical distance between each network node, the number of optical cables, and the network connection topology relationship. The physical distance between the above-mentioned network nodes may include at least one of the following: the physical distance between different regional networks, the physical distance between different computer rooms, the physical distance between different network nodes (devices), etc., that is, the network nodes in different application scenarios may correspond to different implementation methods, the network node may be a regional network, or the network node may be implemented as a certain computer room in a preset regional network, or the network node may be implemented as a single device in a computer room; the above-mentioned spatial topology data is helpful for constructing a network topology map for the cross-domain network, so that the structural information corresponding to the cross-domain network in the spatial dimension can be obtained.

[0078] Step S203: input the time series data and the spatial topology data into the neural network model to obtain the quality detection information of the cross-domain network output by the neural network model.

[0079] Since time series data can reflect the time dimension characteristics of the cross-domain network, and spatial topology data can reflect the spatial dimension characteristics of the cross-domain network, after obtaining the time series data and spatial topology data, the time series data and spatial topology data can be input into the neural network model. The neural network model can perform network detection operations on the time series data and spatial topology data, so as to obtain the quality detection information of the cross-domain network output by the neural network model. The quality detection information is used to identify whether there is a fault in the cross-domain network, thereby completing the quality detection operation on the cross-domain network.

[0080] In some instances, after obtaining quality detection information of a cross-domain network, it can be determined whether alarm information needs to be generated based on the quality detection information. At this time, the method in this embodiment may also include: when the quality detection information is used to identify a fault in the cross-domain network, generating alarm information corresponding to the quality detection information; and displaying the quality detection information and the alarm information in association with each other.

[0081] After obtaining the quality detection information, it can be determined whether there is a fault in the cross-domain network based on the quality detection information. Specifically, when the quality detection information is used to identify that there is no fault in the cross-domain network, the quality detection information can be stored, or the quality detection information can be ignored; when the quality detection information is used to identify that there is a fault in the cross-domain network, in order to enable users of the cross-domain network to understand the operating status of the cross-domain network in a timely manner, an alarm information corresponding to the quality detection information can be generated. After the alarm information corresponding to the quality detection information is generated, the quality detection information and the alarm information can be displayed in association, so that the quality detection device can realize the point-by-point alarm capability and the second-level alarm capability for the cross-domain network, further improving the practicality of the method.

[0082] Among them, the alarm information can be generated based on the fault type included in the quality detection information. At this time, generating the alarm information corresponding to the quality detection information may include: determining the fault type existing in the cross-domain network based on the quality detection information; different fault types determine different levels of alarm information, for example: when there is a fault in a network node in the cross-domain network, that is, the fault type is a node fault type, then a first level of alarm information corresponding to the node fault type can be generated, for example: the first level of alarm information can be implemented as an alarm prompt message; when there is a fault on an edge in a preset area in the cross-domain network, that is, the fault type is an edge fault type, then a second level of alarm information corresponding to the edge fault type can be generated The second level of alarm information is higher than the first level of alarm information, for example, the second level of alarm information can be realized as a sound alarm message and an alarm prompt message; when there is a fault on the edge across regions in a cross-domain network, that is, the fault type is a cross-region edge fault type, a third level of alarm information corresponding to the cross-region edge fault type can be generated, wherein the third level of alarm information is higher than the second level of alarm information, for example, the third level of alarm information can be realized as a telephone type of alarm information. After the above operation, it is possible to generate different levels of alarm information for different types of fault information, thereby ensuring the accuracy and reliability of the generation of alarm information.

[0083] In some other instances, after obtaining the quality detection information of the cross-domain network, in order to facilitate users to view and call the quality detection information, a network quality report can also be generated based on the quality detection information. At this time, the method in this embodiment can also include: when the quality detection information is used to identify the existence of a fault in the cross-domain network, determining the fault generation cause and fault repair suggestions corresponding to the quality detection information; generating a network quality report based on the quality detection information, the fault generation cause and the fault repair suggestions.

[0084] After the quality detection information is obtained, it can be determined whether there is a fault in the cross-domain network based on the quality detection information. Specifically, when the quality detection information is used to identify that there is no fault in the cross-domain network, the quality detection information can be stored; when the quality detection information is used to identify that there is a fault in the cross-domain network, in order to enable users of the cross-domain network to promptly understand the fault status of the cross-domain network and to help users perform fault repair operations, the fault generation cause and fault repair suggestion corresponding to the quality detection information can be determined. In some instances, the fault generation cause and the fault repair suggestion can be determined based on a preset mapping relationship. At this time, determining the fault generation cause and the fault repair suggestion corresponding to the quality detection information can include: determining the fault type based on the quality detection information; determining the fault generation cause and the fault repair suggestion corresponding to the fault type using the preset mapping relationship. Alternatively, the fault generation cause and the fault repair suggestion can be implemented based on a pre-trained neural network model. At this time, determining the fault generation cause and the fault repair suggestion corresponding to the quality detection information can include: obtaining a pre-trained neural network model; inputting the quality detection information and the cross-domain network into the neural network model, and obtaining the fault generation cause and the fault repair suggestion corresponding to the quality detection information output by the neural network model.

[0085] After determining the fault generation cause and fault repair suggestion corresponding to the quality detection information, the quality detection information, the fault generation cause and the fault repair suggestion can be analyzed and processed to generate a network quality report. Specifically, based on the quality detection information, the fault generation cause and the fault repair suggestion, generating a network quality report may include: obtaining a report template for generating a network quality report; filling the quality detection information, the fault generation cause and the fault repair suggestion in a preset position in the report template to generate a network quality report. Alternatively, based on the quality detection information, the fault generation cause and the fault repair suggestion, generating a network quality report may include: obtaining a report generation format for the network quality report; organizing the quality detection information, the fault generation cause and the fault repair suggestion based on the report generation format to generate a network quality report.

[0086] It should be noted that the network quality report can be generated based on real-time quality detection operations for the cross-domain network, or it can also be generated based on quality detection operations for the cross-domain network at any time. This allows users to view or call the above-mentioned cross-domain network quality report generated in real time or at any time according to application requirements, further improving the practicality of the method.

[0087] The quality detection method for a cross-domain network provided in this embodiment obtains a cross-domain network to be detected and a neural network model for implementing a quality detection operation for the cross-domain network, determines time series data and spatial topology data corresponding to the cross-domain network, and then uses the neural network model to perform a network detection operation on the time series data and the spatial topology data, thereby stably obtaining quality detection information of the cross-domain network. Since the quality detection operation of the cross-domain network is implemented by combining time features and spatial features, the quality and efficiency of the network quality detection operation can be efficiently guaranteed. Moreover, the quality detection information of the cross-domain network obtained can timely and accurately reflect the operating status of the cross-domain network. When the quality detection information identifies a fault in the cross-domain network, the timeliness and reliability of the network response can be guaranteed through the quality detection information, further ensuring the practicality of the method.

[0088] Figure 3 A flowchart of a neural network model for implementing a quality detection operation of a cross-domain network is provided in an embodiment of the present invention; based on the above embodiment, refer to the attached Figure 3 As shown, the neural network model in this embodiment can be implemented as a spatiotemporal graph convolutional neural network model; at this time, the neural network model can be obtained by training with historical fault data. Specifically, obtaining a neural network model for implementing a quality detection operation of a cross-domain network may include:

[0089] Step S301: Acquire historical fault data of a cross-domain network.

[0090] For a cross-domain network, in the process of using the cross-domain network for information transmission, the operation data of the cross-domain network can be stored in a preset server. In order to be able to train a neural network model for implementing quality inspection operations of the cross-domain network, so that the neural network model can learn the distribution and regularity of historical faults of the cross-domain network, the historical fault data of the cross-domain network can be obtained, wherein the historical fault data may include at least one of the following: fault information, fault type, fault distribution, etc. existing in the cross-domain network within a historical time period. Specifically, the historical fault data can be obtained by accessing the preset server; or, the historical fault database is obtained by analyzing and extracting the historical operation data of the cross-domain network.

[0091] Step S302: determine a network topology diagram of the cross-domain network, where the network topology diagram includes a plurality of network nodes, edges between any two network nodes, and network operation information corresponding to each network node.

[0092] For historical fault data, the real fault data included therein is relatively sparse, which makes the positive and negative samples in the historical fault data unbalanced, that is, there are more positive sample data and fewer negative sample data. In order to ensure the training quality and effect of the neural network model, it is necessary to balance the positive and negative samples. At this time, after obtaining the cross-domain network, the cross-domain network can be analyzed and processed to determine the network topology of the cross-domain network, wherein the network topology can include multiple network nodes, the edges after any two network nodes, and the network operation information corresponding to each network node, and the network operation information can include at least one of the following: the number of packets sent, the packet sending rate, the number of packet loss, the packet loss rate, bandwidth information, etc.

[0093] As for the network topology map, this embodiment does not limit the method for determining the network topology map. In some instances, the network topology map of the cross-domain network can be determined by the spatial topology data of the cross-domain network. In this case, determining the network topology map of the cross-domain network may include: obtaining the network connection topology relationship in the spatial topology data corresponding to the cross-domain network; and determining the network topology map of the cross-domain network based on the network connection topology relationship. Alternatively, the network topology map of the cross-domain network can be obtained by scanning the cross-domain network. In this case, determining the network topology map of the cross-domain network may include: performing a structural scanning operation on the cross-domain network to obtain the network topology map of the cross-domain network, which effectively ensures the accuracy and reliability of determining the network topology map.

[0094] Step S303: performing a fault addition operation in the network topology map based on the historical fault data to obtain training sample data, wherein the training sample data includes balanced positive sample data and negative sample data.

[0095] After obtaining the historical fault data and the network topology map, in order to be able to balance the positive sample data and the negative sample data in the historical fault data, the fault addition operation can be performed in the network topology map based on the historical fault data, that is, the fault addition configuration operation is performed in the network topology map based on the historical fault data, the fault sample data is increased, and then the training sample data including the balanced positive sample data and negative sample data can be obtained.

[0096] In some instances, the fault addition operation can be implemented through a machine learning model. At this time, the fault addition operation is performed in the network topology map based on historical fault data, and obtaining training sample data may include: obtaining a pre-trained machine learning model; using the machine learning model and historical fault data to perform the fault addition operation in the network topology map, so as to obtain the training sample data output by the machine learning model, and the training sample data may include balanced positive sample data and negative sample data.

[0097] In other instances, the fault addition operation can not only be implemented through a machine learning model, but also can be performed according to historical fault data. At this time, performing a fault addition operation in a network topology map based on historical fault data and obtaining training sample data may include: determining fault configuration information corresponding to a cross-domain network based on historical fault data, the fault configuration information including at least: fault type, fault occurrence probability corresponding to each fault type, and fault adjustment parameters corresponding to each fault type; performing a fault addition operation in a network topology map based on the fault configuration information to obtain training sample data.

[0098] After obtaining the historical fault data, the historical fault data can be analyzed and processed to determine the fault configuration information corresponding to the cross-domain network, which at least includes: fault type, fault occurrence probability corresponding to each fault type, and fault adjustment parameters corresponding to each fault type. Specifically, the fault type may include: node fault type, edge fault type, high packet loss rate type under high packet transmission link, low packet loss rate type under high packet transmission link, high packet loss rate type under low packet transmission link, and low packet loss rate type under low packet transmission link; the fault adjustment parameters corresponding to each fault type can be implemented as at least one of the following: fault adjustment packet loss number, fault adjustment packet loss rate, etc. After obtaining the fault configuration information, the fault configuration operation can be performed in the network topology diagram based on the fault configuration information to obtain training sample data.

[0099] In some instances, performing an operation of adding faults in a network topology map based on fault configuration information to obtain training sample data may include: performing an operation of adding node fault types in the network topology map based on the fault configuration information to obtain first training sample data; performing an operation of adding edge fault types in the network topology map based on the fault configuration information to obtain second training sample data; and determining training sample data for implementing model training operations based on the first training sample data and the second training sample data, thereby effectively ensuring the accuracy and reliability of obtaining the training sample data.

[0100] Among them, since the edges in the cross-domain network can be divided into high-packet transmission links and low-packet transmission links according to the number of packets sent, the additional operation of the fault type may include an additional operation of a high packet loss rate type and an additional operation of a low packet loss rate type. At this time, the additional operation of the edge fault type in the network topology diagram based on the fault configuration information may include: based on the fault configuration information, performing in the network topology diagram an additional operation of a high packet loss rate type under a high packet transmission link, an additional operation of a low packet loss rate type under a high packet transmission link, an additional operation of a high packet loss rate type under a low packet transmission link, and an additional operation of a low packet loss rate type under a low packet transmission link, so that the second training sample data can be stably obtained.

[0101] It should be noted that, for training sample data, according to the data type, it can include two types of data: time series sample data and spatial sample data. In this way, when the model is trained based on the training sample data, the neural network model can have the ability to analyze and process time series data and spatial topology data, which is conducive to improving the reasoning accuracy of the neural network model, and thus improving the accuracy and reliability of network quality detection.

[0102] Step S304: Perform model training operations using the training sample data to obtain a spatiotemporal graph convolutional neural network model for implementing cross-domain network quality detection operations.

[0103] After obtaining the training sample data, model training operations can be performed based on the training sample data, so as to obtain a spatiotemporal graph convolutional neural network model for implementing quality detection operations on cross-domain networks. Since the training sample data includes balanced positive sample data and negative sample data, the quality and effect of training the spatiotemporal graph convolutional neural network model can be guaranteed. The obtained spatiotemporal convolutional neural network model can be used to perform quality detection operations on cross-domain networks, thereby improving the accuracy and reliability of the quality detection operations.

[0104] In this embodiment, the network topology of the cross-domain network is determined by acquiring historical fault data of the cross-domain network, and fault addition operations are performed in the network topology map based on the historical fault data to obtain training sample data. In this way, the problem of imbalance between positive and negative samples is solved by fault addition, so that the training sample data can include balanced positive sample data and negative sample data. Then, the model training operation is performed using the training sample data, which effectively guarantees the quality and effect of training the spatiotemporal graph convolutional neural network model. The obtained spatiotemporal graph convolutional neural network model can be used to implement quality detection operations on cross-domain networks, thereby effectively improving the accuracy and reliability of quality detection operations on cross-domain networks, and further improving the practicability of the method.

[0105] Figure 4The present invention provides a schematic diagram of performing a fault configuration operation in a network topology diagram based on fault configuration information to obtain training sample data; based on the above embodiment, refer to the attached Figure 4 As shown, the fault configuration operation can be implemented from the node fault dimension and the edge fault dimension. At this time, the fault configuration operation is performed in the network topology diagram based on the fault configuration information in this embodiment, and the training sample data is obtained, which may include:

[0106] Step S401: Performing an additional configuration operation of a node failure in a network topology diagram based on the fault configuration information to obtain node failure sample data. And / or,

[0107] Since the fault types in the cross-domain network may include: node fault type and edge fault type, therefore, in the process of performing fault additional configuration operations, the fault additional configuration operations can be implemented from at least one of the node fault type and the edge fault type. After obtaining the fault configuration information, the node fault additional configuration operations can be performed in the network topology map based on the fault configuration information to obtain node fault sample data.

[0108] In some instances, the node failure additional configuration operation can be configured in the network topology diagram through the fault type included in the fault configuration information, the fault occurrence probability corresponding to each fault type, and the fault adjustment parameters corresponding to each fault type. Specifically, the node failure data additional configuration operation can be performed for each network node or part of the network nodes in the network topology diagram based on the fault configuration information, so that node failure sample data can be obtained.

[0109] In other instances, the operation of adding node failures can be achieved by performing a random sampling operation on the network nodes in the network topology map. At this time, the operation of adding node failures in the network topology map based on the fault configuration information in this embodiment to obtain node failure sample data may include: randomly sampling the network nodes in the network topology map to obtain failure sampling points; determining at least one associated edge with the failure sampling point as the destination node; performing an operation of adding packet loss rate on at least one associated edge based on the fault configuration information to obtain node failure sample data.

[0110] When it is necessary to perform additional configuration operations for node failures in the network topology diagram, in order to make the additional fault conditions conform to the actual scenario and ensure that the fault configuration operations are random, after obtaining the network topology diagram, the network nodes in the network topology diagram can be randomly sampled to obtain fault sampling points, wherein the number of fault sampling points obtained by random sampling can be one or more.

[0111] For a network node in a network topology diagram, if a network node fails, the edge with the network node as the target node will most likely not be able to perform normal data transmission operations. Therefore, in order to make the added node failure conform to the actual situation, when performing the node failure addition operation, it is necessary to perform a synchronous fault addition operation on the corresponding edge. At this time, after obtaining the fault sampling point, at least one associated edge with the fault sampling point as the destination node can be determined, wherein the at least one associated edge can be determined by the structure of the network topology diagram and the fault sampling point.

[0112] In addition, for cross-domain networks, since the packet loss rate is an important indicator to measure network performance, it can reflect the loss of data packets in the network during data transmission. Therefore, in order to realize the fault allocation operation, the fault allocation operation can be realized in the dimension of packet loss rate. Specifically, after determining the fault sampling point and at least one associated edge, the packet loss rate allocation operation can be performed on the fault sampling point and / or at least one associated edge based on the fault configuration information, so that the node fault sample data can be obtained.

[0113] In some instances, when it is necessary to perform an operation of increasing the packet loss rate on at least one associated edge, the at least one associated edge can be first screened, and then a fault-adding operation can be performed based on the screened associated edge. At this time, a packet loss rate-adding operation is performed on at least one sampling edge based on the fault configuration information, and obtaining node fault sample data can include: determining at least one fault-adding edge for performing the packet loss rate-adding operation in at least one associated edge; determining the packet loss rate configuration information corresponding to each fault-adding edge based on the fault configuration information, the packet loss rate configuration information including at least: the probability of occurrence of the packet loss rate, and the numerical range of the packet loss rate corresponding to each occurrence probability; performing a packet loss rate-adding operation on at least one fault-adding edge based on the packet loss rate configuration information to obtain the node fault sample data.

[0114] Specifically, after obtaining at least one associated edge, a screening operation can be performed on the at least one associated edge to determine at least one fault-additional edge for performing an operation on increasing the packet loss rate. In some instances, the fault-additional edge can be obtained by randomly screening at least one associated edge, or the fault-additional edge can be obtained by screening at least one associated edge using historical fault data. At this time, determining at least one fault-additional edge for performing an operation on increasing the packet loss rate may include: determining edge ratio information for implementing an operation on increasing the packet loss rate based on historical fault data; and determining at least one fault-additional edge for performing an operation on increasing the packet loss rate in at least one associated edge based on the edge ratio information.

[0115] Since historical fault data can reflect the historical fault distribution of cross-domain networks, in order to conform to real scenarios, the edge proportion information used to implement the packet loss rate allocation operation can be determined based on the historical fault data. Specifically, the edge proportion information can be determined by the packet loss rate distribution in the historical fault data. For example, the edge proportion information can be implemented as 70% of the associated edges as fault allocation edges, and 30% of the associated edges are not used as fault allocation edges; or, the edge proportion information can be implemented as 60% of the associated edges as fault allocation edges, and 40% of the associated edges are not used as fault allocation edges, and so on.

[0116] After obtaining the edge ratio information for implementing the packet loss rate increase operation, at least one faulty additional configuration edge for performing the packet loss rate increase operation can be determined in at least one associated edge based on the edge ratio information, that is, in at least one associated edge, a randomly selected associated edge that satisfies the edge ratio information is selected, and then the randomly selected associated edge that satisfies the edge ratio information is determined as at least one faulty additional configuration edge for performing the packet loss rate increase operation, thereby effectively ensuring the accuracy and reliability of the determination of at least one faulty additional configuration edge.

[0117] After obtaining the fault-added edges, the packet loss rate configuration information corresponding to each fault-added edge can be determined based on the fault configuration information. The obtained packet loss rate configuration information at least includes: the probability of occurrence of packet loss rate, and the numerical range of packet loss rate corresponding to each occurrence probability. For example, the packet loss rate configuration information can be implemented as follows: randomly configure a packet loss rate of 0.3%-1% with a probability of 75%; randomly configure a packet loss rate of 1%-10% with a probability of 20%; randomly configure a packet loss rate of 10%-90% with a probability of 5%; and then, based on the packet loss rate configuration information, the packet loss rate of at least one fault-added edge can be increased. This allows the node fault sample data to be stably obtained.

[0118] In other instances, the packet loss rate allocation operation can not only be performed by screening at least one associated edge and performing the fault allocation operation based on the screened associated edge, but also the fault allocation operation can be implemented directly based on the packet loss probability information. At this time, the packet loss rate allocation operation is performed on at least one associated edge based on the fault configuration information, and obtaining the node fault sample data may include: determining the packet loss probability information corresponding to each associated edge based on historical fault data; determining the fault configuration information corresponding to each associated edge, the fault configuration information at least including: the probability of occurrence of the packet loss rate, and the numerical range of the packet loss rate corresponding to each occurrence probability; performing the packet loss rate allocation operation on each associated edge based on the packet loss probability information and the fault configuration information to obtain the node fault sample data.

[0119] Specifically, since the historical fault data can reflect the distribution of historical faults in the cross-domain network, after obtaining the historical fault data, the historical fault data can be analyzed and processed to determine the packet loss probability information corresponding to each associated edge, and the packet loss probability information can be implemented as: 70% probability of generating a packet loss fault; 30% probability of generating a packet loss fault; or, the packet loss probability information can be implemented as: 60% probability of generating a packet loss fault; 40% probability of generating a packet loss fault, etc. In order to accurately perform the fault configuration operation, the fault configuration information corresponding to each associated edge can be determined, and the fault configuration information at least includes: the probability of occurrence of the packet loss rate, and the numerical range of the packet loss rate corresponding to each occurrence probability; for example, the fault configuration information can include at least one of the following: randomly configuring a packet loss rate of 0.3%-1% with a probability of 65%; randomly configuring a packet loss rate of 1%-10% with a probability of 30%; and randomly configuring a packet loss rate of 10%-90% with a probability of 5%. After obtaining the packet loss probability information and the fault configuration information, the packet loss rate of each associated edge can be configured based on the packet loss probability information and the fault configuration information, so that the node fault sample data can be stably obtained.

[0120] Step S402: performing an additional configuration operation of edge faults in the network topology diagram based on the fault configuration information to obtain edge fault sample data.

[0121] After the fault configuration information is obtained, not only can additional node faults be performed in the network topology map based on the fault configuration information, but also additional edge faults can be performed in the network topology map based on the fault configuration information, so that edge fault sample data can be obtained. In some instances, the additional edge fault operation can be implemented by a pre-trained machine learning model. In this case, the additional edge fault operation in the network topology map based on the fault configuration information can include: obtaining a pre-trained machine learning model for implementing the additional edge fault operation, inputting the fault configuration information and the network topology map into the machine learning model to perform the additional edge fault operation, and obtaining the edge fault sample data output by the machine learning model, thereby completing the additional edge fault operation.

[0122] In other instances, the edge fault configuration operation can not only be implemented through a pre-trained machine learning model, but also can be implemented in combination with historical fault data. At this time, performing the edge fault configuration operation in the network topology map based on the fault configuration information, and obtaining the edge fault sample data may include: randomly sampling the edges in the network topology map to obtain fault sampling edges, and the fault sampling edges include: first-type sampling edges located in the same area and second-type sampling edges located between different areas; based on the historical fault data, determining the target packet loss rate for performing the edge fault configuration operation, the target packet loss rate at least includes: the probability of occurrence of the packet loss rate, and the packet loss rate occurrence range corresponding to each occurrence probability; performing the edge fault configuration operation on the fault sampling edges based on the target packet loss rate to obtain edge fault sample data, and the edge fault sample data includes: first edge fault sample data corresponding to the first-type sampling edges, and second edge fault sample data corresponding to the second-type sampling edges.

[0123] After obtaining the network topology map, a random sampling operation can be performed on the network topology map to obtain fault sampling edges. Since the edges in the cross-domain network include edges in the same area and edges between different areas, when randomly sampling the edges in the network topology map, the obtained fault sampling edges can include first-class sampling edges located in the same area and second-class sampling edges located between different areas, wherein the number of the first-class sampling edges and the second-class sampling edges is one or more, and the number of the first-class sampling edges and the number of the second-class sampling edges can be the same or different, for example: the number of the first-class sampling edges is 50, the number of the second-class sampling edges is 100, and so on; or, the number of the first-class sampling edges is 20, the number of the second-class sampling edges is 50, and so on.

[0124] In order to make the edge failure of fault addition conform to the rules and distribution of real scenarios, after obtaining the historical failure data, the historical failure data can be analyzed and processed to determine the target packet loss rate for the edge failure addition operation. The target packet loss rate can at least include: the probability of occurrence of the packet loss rate, and the packet loss rate occurrence range corresponding to each occurrence probability.

[0125] In other instances, the target packet loss rate for performing the additional configuration operation for edge failures can be determined in combination with the edge loss information. In this case, determining the target packet loss rate for performing the additional configuration operation for edge failures based on historical fault data in this embodiment can include: obtaining the edge loss information corresponding to each fault sampling edge; determining the initial fault packet loss rate of each fault sampling edge based on the historical fault data, the initial fault packet loss rate including at least one of the following: the probability of occurrence of the packet loss rate, and the numerical range of the packet loss rate corresponding to each occurrence probability; determining the target packet loss rate for performing the additional configuration operation for edge failures based on the edge loss information and the initial fault packet loss rate.

[0126] Specifically, in the process of using a cross-domain network for data transmission, edge loss information can affect the packet loss rate. Therefore, in order to accurately determine the target packet loss rate, the edge loss information corresponding to each fault sampling edge can be first obtained, wherein the edge loss information corresponding to each fault sampling edge can be determined by multiple physical transmission links (generally including at least 4 physical transmission links) included in the cross-domain network. At this time, obtaining the edge loss information corresponding to each fault sampling edge may include: obtaining multiple physical transmission links corresponding to each fault sampling edge; determining the number of packets sent by each physical transmission link, the total number of packets sent by all physical transmission links, the number of packet losses of each physical transmission link, the total number of packet losses of all physical transmission links, and the degree of influence of each physical transmission link on the fault sampling edge; determining the edge loss information corresponding to each fault sampling edge based on the number of packets sent by each physical transmission link, the total number of packets sent by all physical transmission links, the number of packet losses of each physical transmission link, the total number of packet losses of all physical transmission links, and the degree of influence of each physical transmission link on the fault sampling edge.

[0127] After obtaining the historical fault data, the historical fault data can be analyzed and processed to determine the initial fault packet loss rate of each fault sampling edge. The initial fault packet loss rate can be packet loss rate information that conforms to the actual situation and is determined based on the historical fault data. The initial fault packet loss rate may include at least one of the following: the probability of occurrence of the packet loss rate, and the numerical range of the packet loss rate corresponding to each occurrence probability. For example, the initial fault packet loss rate can be implemented as follows: a 70% probability fluctuates between 1% and 5% (i.e., the numerical range of the packet loss rate corresponding to each occurrence probability); a 25% probability fluctuates between 5% and 20%; and a 5% probability fluctuates between 20% and 90%.

[0128] After obtaining the edge loss information and the initial fault packet loss rate, the edge loss information and the initial fault packet loss rate can be analyzed and processed to determine a target packet loss rate for performing an edge failure configuration operation. In some instances, based on the edge loss information and the initial fault packet loss rate, determining the target packet loss rate for performing an edge failure configuration operation can include: taking a logarithm operation on the edge loss information to obtain processed information; and determining the ratio of the initial fault packet loss rate to the processed information as the target packet loss rate.

[0129] For example, the edge loss information is expressed as , the initial fault packet loss rate is expressed as , and then we can first perform a logarithmic operation on the edge loss information to obtain the processed information ; Then the ratio of the initial fault packet loss rate to the processed information can be determined as the target packet loss rate ,Right now , which effectively ensures the accuracy and reliability of determining the target packet loss rate.

[0130] After determining the target packet loss rate for performing the edge fault allocation operation, the edge fault allocation operation can be performed on the faulty sampling edge based on the target packet loss rate to obtain edge fault sample data. Since the faulty sampling edge includes a first type of sampling edge and a second type of sampling edge, when the edge fault allocation operation is performed on the faulty sampling edge, the edge fault sample data obtained may include: first edge fault sample data corresponding to the first type of sampling edge, and second edge fault sample data corresponding to the second type of sampling edge.

[0131] In order to further make the edge fault addition operation conform to the real scenario, the edge fault addition operation can be performed on the second type of sampling edge in combination with the fault diffusion method. At this time, the fault sampling edge can be the second type of sampling edge; the edge fault addition operation is performed on the fault sampling edge based on the target packet loss rate, and obtaining the edge fault sample data may include: obtaining the adjacent edges and second-degree edges of the second type of sampling edges; determining a preset decay ratio for performing the fault diffusion operation on the adjacent edges and the second-degree edges; performing the edge fault addition operation on the second type of sampling edges based on the target packet loss rate, and performing the fault diffusion operation on the adjacent edges and the second-degree edges according to the preset decay ratio to obtain the second edge fault sample data.

[0132] Specifically, when the fault sampling edge is a second-type sampling edge, in order to make the additional configuration operation of the edge fault conform to the actual scenario, the adjacent edges and second-degree edges of the second-type sampling edge can be obtained, wherein the adjacent edges and second-degree edges can be determined based on the network topology diagram corresponding to the cross-domain network, and then the preset decay ratio for performing fault operations on the adjacent edges and second-degree edges can be determined. The preset decay ratio can be pre-configured by the user according to the application scenario or application requirement. In some embodiments, the preset decay ratio can be implemented as a ratio range of 0-0.5, that is, each time a fault diffusion operation is performed, it can be decayed at a ratio of 0-0.5.

[0133] After obtaining the preset decay ratio, target packet loss rate, adjacent edges and second-degree edges, the edge fault addition operation can be performed on the second type of sampling edges based on the target packet loss rate, and the fault diffusion operation can be performed on the adjacent edges and second-degree edges according to the preset decay ratio, so that the second edge fault sample data can be obtained stably.

[0134] In some other instances, when an edge failure configuration operation is performed on a first type of sampling edge or a second type of sampling edge based on a target packet loss rate, the edge failure configuration operation can be performed in combination with white noise. At this time, after determining the target packet loss rate for performing the edge failure configuration operation, the method in this embodiment may include: obtaining white noise information for performing the edge failure configuration operation; adjusting the target packet loss rate based on the white noise information to obtain an adjusted packet loss rate.

[0135] Specifically, white noise information for implementing the edge fault configuration operation is pre-configured. When the edge fault configuration operation needs to be performed, the white noise information for performing the edge fault configuration operation can be obtained, and then the target packet loss rate can be adjusted based on the white noise information, so as to obtain the adjusted packet loss rate. For example, the white noise information can be implemented as a packet loss rate less than or equal to 0.5. When the white noise information is a packet loss rate of 0.3, after the target packet loss rate is obtained, the target packet loss rate can be adjusted based on the white noise information to obtain the adjusted packet loss rate. The lower limit value of the adjusted packet loss rate can be the target packet loss rate -0.3, and the upper limit value of the adjusted packet loss rate can be the target packet loss rate +0.3. Then, the edge fault configuration operation can be performed based on the range corresponding to the adjusted packet loss rate (target packet loss rate -0.3, target packet loss rate +0.3), so that the accuracy and reliability of the edge fault configuration operation can be ensured.

[0136] Step S403: Determine training sample data based on the node fault sample data and / or the edge fault sample data.

[0137] After acquiring the node fault sample data and / or edge fault sample data, the node fault sample data and / or edge fault sample data can be analyzed and processed to determine the training sample data. At this time, the training sample data may include the node fault sample data; or, the training sample data may include the edge fault sample data; or, the training sample data may include the node fault sample data and the edge fault sample data, thereby effectively ensuring the flexibility and reliability of determining the training sample data.

[0138] In this embodiment, based on the fault configuration information, a node fault addition operation is performed in the network topology map to obtain node fault sample data. At the same time, based on the fault configuration information, an edge fault addition operation is performed in the network topology map to obtain edge fault sample data. Then, the training sample data can be determined based on the node fault sample data and / or the edge fault sample data, thereby effectively ensuring the flexibility and reliability of determining the training sample data.

[0139] Figure 5 A schematic diagram of a flow chart of using a neural network model to perform network detection on time series data and spatial topology data to obtain quality detection information of a cross-domain network according to an embodiment of the present invention; based on any of the above embodiments, refer to the attached Figure 5 As shown, the neural network model may include a first sub-network model, a second sub-network model, and a third sub-network model for implementing network detection operations. At this time, using the neural network model to perform network detection on time series data and spatial topology data, obtaining quality detection information of the cross-domain network may include:

[0140] Step S501: Use the first sub-network model in the neural network model to perform node fault detection operations on time series data and spatial topology data to obtain node fault quality detection information.

[0141] After acquiring the time series data and spatial topology data, the first sub-network model in the neural network model can be used to perform node fault detection operations on the time series data and spatial topology data. In some instances, the time series data and spatial topology data can be directly input into the first sub-network model for analysis and processing to obtain node fault quality detection information output by the first sub-network model.

[0142] In other instances, for a certain network node in a cross-domain network, a quality fault detection operation of the node fault can be implemented in combination with the neighbor nodes corresponding to the network node. At this time, the first sub-network model in the neural network model is used to perform a node fault detection operation on the time series data and the spatial topology data, and obtaining the node fault quality detection information may include: based on the spatial topology data, determining at least one network node in the cross-domain network and the neighbor nodes corresponding to each network node; based on the time series data, determining the first node representation of each network node and the second node representation of the corresponding neighbor node; aggregating the first node representation and the second node representation of the corresponding neighbor node to obtain an aggregated node representation; using the first sub-network model in the neural network model to perform a node fault detection operation on the aggregated node representation to obtain node fault quality detection information.

[0143] Specifically, after obtaining the spatial topology data, the spatial topology data can be analyzed and processed to determine at least one network node in the cross-domain network and the neighbor nodes corresponding to each network node, and then the first node representation of each network node and the second node representation of the corresponding neighbor node can be determined based on the time series data, the first node representation can represent the operation characteristics of the network node, and the second node representation can represent the operation characteristics of the neighbor node; after obtaining the first node representation and the second node representation, the first node representation and the second node representation of the corresponding neighbor node can be aggregated. In some examples, the aggregation operation can be implemented as a splicing operation on the first node representation and the second node representation; or, the aggregation operation can be implemented as a weighted fusion operation on the first node representation and the second node representation, etc. Then, the first sub-network model in the neural network model is used to perform a node fault detection operation on the aggregated node representation, so that the node fault quality detection information can be stably obtained, and the node fault detection operation is completed.

[0144] Step S502: using the second sub-network model in the neural network model to perform edge fault detection operations within a preset area on the time series data and the spatial topology data to obtain first edge fault quality detection information.

[0145] Among them, in order to accurately perform fault detection operations on edges in a preset area in a cross-domain network, the neural network model includes a second sub-network model for implementing edge fault detection operations. The second sub-network model can then be used to perform edge fault detection operations on time series data and spatial topology data in a preset area to obtain first edge fault quality detection information.

[0146] In some instances, using the second sub-network model in the neural network model to perform edge fault detection operations within a preset area on the time series data and the spatial topology data, obtaining first edge fault quality detection information may include: determining the node embedding representation of the network nodes and the edge embedding representation of the edges in the cross-domain network located in the preset area based on the time series data and the spatial topology data; using the second sub-network model to perform edge fault detection operations within the preset area on the node embedding representation and the edge embedding representation, thereby obtaining the first edge fault quality detection information output by the second sub-network model, and the obtained first edge fault quality detection information is used to identify whether there is a fault on the edge within the preset area.

[0147] Step S503: using the third sub-network model in the neural network model to perform cross-region edge fault detection operations on the time series data and the spatial topology data to obtain second edge fault quality detection information.

[0148] Among them, in order to accurately perform fault detection operations on edges across regions in a cross-domain network, the neural network model may include a third sub-network model for implementing edge fault detection operations, and then the third sub-network model can be used to perform edge fault detection operations on time series data and spatial topology data across regions to obtain second edge fault quality detection information.

[0149] In some instances, using the third sub-network model in the neural network model to perform edge fault detection operations within a preset area on the time series data and the spatial topology data, obtaining second edge fault quality detection information may include: determining the node embedding representation of the network nodes located in the cross-domain area and the edge embedding representation of the edges in the cross-domain network based on the time series data and the spatial topology data; using the third sub-network model to perform edge fault detection operations within the region on the node embedding representation and the edge embedding representation, thereby obtaining the second edge fault quality detection information output by the third sub-network model, and the obtained second edge fault quality detection information is used to identify whether there is a fault in the edge within the cross-region.

[0150] Step S504: Determine the quality detection information of the cross-domain network based on the node failure quality detection information, the first edge failure quality detection information and the second edge failure quality detection information.

[0151] After obtaining the node fault quality detection information, the first-side fault quality detection information, and the second-side fault quality detection information, the node fault quality detection information, the first-side fault quality detection information, and the second-side fault quality detection information can be analyzed and processed. In some cases, the node fault quality detection information, the first-side fault quality detection information, and the second-side fault quality detection information can be summarized and processed to determine the quality detection information of the cross-domain network, and the quality detection information is used to identify whether there is a fault in the cross-domain network.

[0152] In this embodiment, the first sub-network model, the second sub-network model and the third sub-network model in the neural network model are respectively used to perform node fault detection operations, edge fault detection operations within a preset area and edge fault detection operations across areas, to obtain node fault quality detection information, first edge fault quality detection information and second edge fault quality detection information, and then the quality detection information of the cross-domain network is determined based on the node fault quality detection information, the first edge fault quality detection information and the second edge fault quality detection information, thereby effectively realizing the quality detection operation on the cross-domain network and further improving the practicality of the method.

[0153] Figure 6 A flow chart of another cross-domain network quality detection method provided by an embodiment of the present invention; based on any one of the above embodiments, refer to the attached Figure 6 As shown, after obtaining the quality detection information of the cross-domain network, when the quality detection information of the cross-domain network is used to identify the presence of a fault in the cross-domain network, the fault impact range of the cross-domain network can be further determined. At this time, the method in this embodiment can also include:

[0154] Step S601: when the quality detection information is used to identify the existence of a fault in the cross-domain network, determine the faulty node or faulty edge corresponding to the quality detection information.

[0155] After the quality detection information is obtained, it can be determined whether there is a fault in the cross-domain network based on the quality detection information. Specifically, when the quality detection information is used to identify that there is no fault in the cross-domain network, it is not necessary to perform any processing operation on the quality detection information, and the quality detection information is stored. When the quality detection information is used to identify that there is a fault in the cross-domain network, the corresponding faulty node or faulty edge can be determined based on the quality detection information.

[0156] Step S602: Determine the fault impact range corresponding to the faulty node or faulty edge.

[0157] For a cross-domain network, when a node or edge in a certain regional network fails, other nodes or edges with the regional network as the destination cannot perform normal data transmission operations through the nodes or edges in the above regional network. For example, the cross-domain network includes nodes A, B, C, D and E. Node A is connected to nodes C and D, and node B is connected to nodes D and E. When node B fails, nodes C and D cannot normally perform data transmission operations to node B. At this time, nodes C and D will also fail, that is, nodes C and D fail due to the influence of node B, that is, different nodes will affect each other. Similarly, different edges in a cross-domain network will also affect each other, and different regional networks in a cross-domain network will also affect each other.

[0158] When performing fault detection operations on a cross-domain network, in order to accurately understand the degree of impact of the fault, after obtaining the quality detection information of the cross-domain network, the fault impact range corresponding to the fault node or fault edge can be determined. In some instances, the fault impact range can be determined by the mutual influence information between different domain networks in the cross-domain network, wherein the mutual influence information can be determined by a pre-trained neural network model. At this time, obtaining the mutual influence information between different domain networks in the cross-domain network may include: obtaining a pre-trained neural network model; determining a network topology diagram of the cross-domain network; using the neural network model to analyze and process the network topology diagram to obtain the mutual influence information output in the neural network model, which effectively ensures the accuracy and reliability of determining the mutual influence information. Specifically, after obtaining the fault node or fault edge, the fault impact range corresponding to the fault node or fault edge can be determined based on the mutual influence information, which effectively ensures the accuracy and reliability of determining the fault impact range.

[0159] Step S603: associate and display the faulty node or faulty edge and the corresponding fault impact range.

[0160] After the faulty node, faulty edge, and faulty impact range are obtained, in order to enable the user to quickly and accurately understand the fault situation of the cross-domain network, the faulty node or faulty edge and the corresponding faulty impact range may be displayed in association.

[0161] In this embodiment, when the quality detection information is used to identify the existence of a fault in a cross-domain network, the faulty node or faulty edge corresponding to the quality detection information is determined, and then the fault impact range corresponding to the faulty node or faulty edge is determined, and the faulty node or faulty edge and the corresponding fault impact range are displayed in association. This not only provides network quality analysis from a global perspective for the cross-domain network, but also accurately predicts cross-domain impacts by learning the interactions between different domains, further improving the accuracy and reliability of quality detection for the cross-domain network.

[0162] In specific applications, this application embodiment provides a quality analysis method for a cross-domain network. The method can perform efficient detection and abnormal analysis of network quality for cross-domain networks based on a spatiotemporal graph convolutional neural network model and PingMesh detection technology. First, the spatiotemporal graph convolutional neural network can use multi-dimensional network traffic data to learn and model network behavior, so as to be able to perform real-time network prediction and network performance evaluation operations; secondly, by introducing PingMesh detection technology, intelligent detection path optimization is implemented between multiple detection points, and key performance indicators such as delay and packet loss of cross-domain connections are obtained in real time, which is conducive to promoting efficient network quality detection, optimizing resource allocation, and improving service quality, ensuring that users can still enjoy a stable and smooth online experience under high load conditions.

[0163] In addition, the execution subject of the above-mentioned quality analysis method can be implemented as a quality analysis device, which can be integrated into a network management tool. Through the network management tool, the enterprise can be provided with efficient network quality analysis operations, which can not only reduce the user churn rate, but also improve user satisfaction, and also lay a solid foundation for the development of future cross-domain network technology, thereby significantly improving the practical value of the method. Specifically, the method may include: a training phase of a spatiotemporal graph convolutional neural network model and an application phase of a spatiotemporal graph convolutional neural network model, wherein, with reference to the attached Figure 7 As shown in Figure 2, the training phase of the spatiotemporal graph convolutional neural network model includes:

[0164] Step 1: Obtain training sample data corresponding to the cross-domain network.

[0165] The quality analysis device may include: a data layer, a processing layer, an enhancement layer, a model layer and a deployment layer. The data layer may be used to obtain training sample data corresponding to the cross-domain network. The training sample data includes two parts of data: time series data and spatial topology data.

[0166] Time series data is also called PingMesh detection data, which can be obtained by using the PingMesh detection tool to detect cross-domain networks. The above-mentioned PingMesh detection tool is a tool for realizing network performance detection and detection. It mainly evaluates the health of the network by measuring indicators such as network delay, packet loss rate, number of hops, bandwidth and jitter. The PingMesh detection tool can record the round-trip time (RTT) of data packets, determine the number of data packets lost during data transmission, and calculate the number of routers passed, thereby identifying the complexity of the network path. In addition, the PingMesh detection tool can also measure the actual data transmission rate and evaluate the rate of change of delay (jitter), providing network managers with comprehensive network performance data to help optimize network conditions and solve potential problems.

[0167] Spatial topology data is also called backbone network topology data. It is the real site data of the backbone network corresponding to the cross-domain network, including: the physical distance between each site, the number of optical cables, and the network connection topology relationship. Spatial topology data helps to model the topology of the network and form spatial structural information.

[0168] Step 2: Perform feature extraction operations on the training sample data to obtain network topology features, node features, and edge features, wherein the network topology features correspond to a network topology graph including multiple nodes and edges between the nodes.

[0169] The processing layer is used to perform feature extraction operations on the time series data in the training sample data to obtain node features and edge features; the processing layer is used to perform feature extraction operations on the spatial topology data in the training sample data to obtain network topology features.

[0170] Step 3: Perform a fault injection operation (corresponding to the fault addition operation in the above embodiment) based on the network topology features, node features and edge features to obtain processed sample data, which includes balanced positive sample data and negative sample data.

[0171] For the acquired training sample data, since the sample size of real faults is relatively sparse, in order to solve the imbalance between positive and negative samples, fault injection is used to enhance negative samples. Among them, fault injection mainly abides by two principles: 1) it is necessary to refer to historical fault data and conform to the real scene; 2) the injected faults need to be random and comprehensive.

[0172] Considering the two principles mentioned above, the design operation of fault injection is carried out from the following four dimensions:

[0173] A: Node failures and link failures (corresponding to edge failures in the above embodiment) occur randomly with a preset equilibrium probability (e.g., 50%) or a preset true distribution probability;

[0174] For cross-domain networks, there are mainly two types of failures: node failure and link failure. In order to ensure the balance of sample data, node failure and link failure can be made to occur randomly with a probability of 50%. At this time, the node failure data and link failure data in the obtained fault sample data are balanced.

[0175] In other instances, the historical distribution ratio between node failures and link failures (for example, 4:6 or 3:7) can be determined based on historical failure data, and then the random occurrence of node failures and link failures can be controlled based on the above historical distribution ratio, so that the obtained fault sample data is more real and reliable.

[0176] B: Adjust the number and rate of packet loss to control the severity of the fault;

[0177] The number of packet losses and the packet loss rate in the cross-domain network can be adjusted by writing software code. Specifically, the adjustment can be made with reference to historical fault data. For example, the fault parameters in fault scenarios of different severity can be learned through historical fault data. The number of packet losses and the packet loss rate can be adjusted based on the above fault parameters to control the severity of the fault.

[0178] C: Various fault types are injected, including high packet loss and low packet loss rate problems in high packet transmission links, as well as low packet loss and high packet loss rate problems in low packet transmission links.

[0179] For links in a cross-domain network, the links can be divided into high-packet transmission links and low-packet transmission links according to the number of packets that can be sent by the links. Among them, the high-packet transmission link can refer to a link that can send data packets exceeding a preset number; the low-packet transmission link can refer to a link that can send data packets exceeding but less than a preset number. In order to ensure the comprehensive reliability of fault data, different fault types can be injected into the above-mentioned different types of links, for example: injecting a high packet loss rate fault into a high packet transmission link; injecting a low packet loss rate fault into a high packet transmission link; injecting a high packet loss rate fault into a low packet transmission link; injecting a low packet loss rate fault into a low packet transmission link, etc. The above-mentioned high packet loss rate may refer to a packet loss rate exceeding a preset threshold, and a low packet loss rate may refer to a packet loss rate less than or equal to the preset threshold.

[0180] In addition, for different types of fault injection operations on the above different links, the distribution parameters of different types of faults can be learned through historical fault data, and fault injection operations can be performed on different links based on the above distribution parameters, so that the distribution degree of fault type injection can be controlled.

[0181] D: Artificially inject link packet loss noise to improve the stability of model training and reduce the false alarm rate.

[0182] The noise injection operation is implemented by writing software code, that is, a white noise will be configured for the cross-domain network. The white noise will allow the faults injected into the cross-domain network to vary within an allowable range. This can improve the stability of fault injection, make the fault input operation more consistent with the actual situation, and reduce false alarm operations of faults.

[0183] With reference to the above objectives, combined with a specific link failure scenario, the cross-domain network fault injection operation is configured to include: a node failure injection operation and a link failure injection operation, wherein the node failure injection operation may include:

[0184] Step 3.11: Determine the network topology map corresponding to the cross-domain network based on the network topology characteristics.

[0185] Step 3.12: Randomly select a node in the network topology as the failure point.

[0186] Step 3.13: Based on the network topology, determine the edge with the fault point as the destination (dst), and perform fault injection on the edge in a way that increases the packet loss rate.

[0187] In some instances, performing a fault injection operation by increasing the packet loss rate on an edge may include: obtaining all edges with the fault point as the destination; determining edge occupancy information for performing the fault injection operation (for example, 70%, 80%, or 85%, etc.); determining the fault injection edge for performing the fault injection operation based on the edge occupancy information and all edges, and determining other edges (for example, edges corresponding to a 30% occupancy, edges corresponding to a 20% occupancy, and edges corresponding to a 15% occupancy) as edges that do not require a fault injection operation.

[0188] In other instances, performing a fault injection operation on an edge by increasing the packet loss rate may include: determining probability information for performing a fault injection operation on an edge in a network topology graph, the probability information may include: a 70% probability of packet loss and a 30% probability of no packet loss; or, an 80% probability of packet loss and a 20% probability of no packet loss; or, an 85% probability of packet loss and a 15% probability of no packet loss, etc., and then based on the probability information, a fault injection operation may be performed on the edge by increasing the packet loss rate.

[0189] Step 3.14: For points with packet loss, fault injection operations can be performed according to the following information: randomly inject a packet loss rate of 0.3%-1% with a probability of 75%, randomly inject a packet loss rate of 1%-10% with a probability of 20%, and randomly inject a packet loss rate of 10%-90% with a probability of 5%. The above information can be determined based on the actual fault distribution.

[0190] In addition, link fault injection operations may include:

[0191] Step 3.21: Determine the network topology map corresponding to the cross-domain network based on the network topology characteristics.

[0192] Step 3.22: Randomly sample 50 same-region edges and 100 cross-region edges in the network topology graph.

[0193] Among them, the 50 same-region edges may refer to edges located in the same computer room, and the 100 cross-region edges may refer to edges in different computer rooms, edges in different regional networks, and so on.

[0194] Step 3.23: Perform a white noise injection operation on the 50 same-region edges and the 100 cross-region edges to obtain white noise corresponding to the 50 region edges and the 100 cross-region edges.

[0195] The white noise may include injecting 1 to 5 packet loss numbers into a single side, or injecting a packet loss rate less than or equal to 0.5% (or 0.3%, 0.4%, etc.) into a single side.

[0196] Step 3.23: Determine the fault injection probability of the same-region edge and the fault injection probability of the cross-region edge.

[0197] In some instances, in order to ensure the balance of sample data, the fault injection probability of the same-region edge and the fault injection probability of the cross-region edge can be determined through preset operations. Specifically, the fault injection probability of the same-region edge and the fault injection probability of the cross-region edge can be made the same, that is, the fault injection probability of the same-region edge and the fault injection probability of the cross-region edge can both be 50%.

[0198] In other instances, in order to ensure that the sample data conforms to the actual situation, the fault injection probability of the same-region edge and the fault injection probability of the cross-region edge can be determined based on the historical fault data. Specifically, the relative distribution information between the faults of the same-region edge and the faults of the cross-region edge can be determined based on the historical fault data; the fault injection probability of the same-region edge and the fault injection probability of the cross-region edge can be determined based on the relative distribution information.

[0199] Step 3.24: Set the initial failure packet loss rate for the same-region edge and the cross-region edge .

[0200] Among them, the initial failure packet loss rate It can be configured as follows: 70% probability fluctuates between 1%-5%, 25% probability fluctuates between 5%-20%, and 5% probability fluctuates between 20%-90%.

[0201] Step 3.25: Determine the adjusted packet loss rate based on the initial failure packet loss rate.

[0202] Determining the adjusted packet loss rate based on the initial fault packet loss rate may include: determining loss information of each edge, and determining the adjusted packet loss rate based on the edge loss information and the initial fault packet loss rate. In some examples, the adjusted packet loss rate may be implemented as , the edge loss information can be realized as , and then the following formula can be used This effectively ensures the accuracy and reliability of determining the target packet loss rate.

[0203] Step 3.25: For the same-area edge, perform fault injection according to the adjusted packet loss rate.

[0204] Step 3.26: For cross-region edges, perform fault injection operations according to the adjusted packet loss rate, and perform fault diffusion operations on the adjacent edges and second-degree edges corresponding to the cross-region edges with a preset probability. The packet loss rate of each diffusion can decay at a ratio of 0-0.5.

[0205] Specifically, the decay ratio of the diffused packet loss rate can be determined by the following formula:

[0206]

[0207] in, can be the packet loss rate after the fault diffusion operation, It can be the packet loss rate corresponding to the fault diffusion operation before. is the decay ratio for fault spreading operation.

[0208] Step 4: Perform model training operations based on the processed sample data to obtain the spatiotemporal graph convolutional neural network model.

[0209] Specifically, the GraphSAGE-Based model can be trained based on the positive sample node failure data and the negative sample node failure data included in the processed sample data to obtain a first sub-network model for implementing node fault detection operations; the EageGAT-Based model can be trained based on the positive sample edge failure data and the negative sample edge failure data corresponding to the same-region edges included in the processed sample data to obtain a second sub-network model for performing fault detection operations on the same-region edges; the LSTM-EageGAT-Based model can be trained based on the positive sample edge failure data and the negative sample edge failure data corresponding to the cross-region edges included in the processed sample data to obtain a third sub-network model for performing fault detection operations on cross-region edges; based on the first sub-network model, the second sub-network model and the third sub-network model, a spatiotemporal graph convolutional neural network model for analyzing and detecting the quality of the cross-domain network is determined.

[0210] In addition, the application phase of the spatiotemporal graph convolutional neural network model can include the following steps:

[0211] Step 11: Obtain the cross-domain network to be detected and a spatiotemporal graph convolutional neural network model for implementing the quality detection operation of the cross-domain network, wherein the spatiotemporal graph convolutional neural network model can be obtained by training based on the method in the above embodiment.

[0212] Step 12: Determine the time series data and spatial topology data corresponding to the cross-domain network. The time series data includes at least: network delay and packet loss rate, and the spatial topology data includes at least: the physical distance between each network node, the number of optical cables, and the network connection topology relationship.

[0213] Step 13: Use the spatiotemporal graph convolutional neural network model to perform network detection on time series data and spatial topology data to obtain quality detection information of cross-domain networks.

[0214] Among them, refer to the attached Figure 8 As shown, the quality detection operation of the cross-domain network can be divided into: fault detection operation on the node (or called the node problem to be solved), fault detection operation on the edge in the same area (or called the link problem in the same area to be solved), and fault detection operation on the edge between regions (or called the link problem between regions to be solved). For the above-mentioned link problem between regions, the spatiotemporal graph convolutional neural network model can be divided into a first sub-network model (GraphSage model) for implementing the fault detection operation on the node, a second sub-network model for implementing the fault detection operation on the edge in the same area, and a third sub-network model for implementing the fault detection operation on the edge between regions.

[0215] In some instances, using the first sub-network model to perform network detection on time series data and spatial topology data to obtain node quality detection information may include: extracting features from the time series data and spatial topology data to obtain node feature information; using the first sub-network model to perform node fault detection operations on the node feature information to obtain node fault detection information.

[0216] Among them, the node feature information and the edge feature information can be realized based on the preset Edge Graph Attention Network (Edge GAT for short) model and the multi-layer perceptron (MLP). Specifically, the Edge GAT model and the MLP model can obtain the edge feature information based on time series data and spatial topology data, transmit the edge feature information on the network topology graph and aggregate it to the node to form the node feature information (i.e., node embedded representation Embedding). Specifically, the attention operation can be performed on the network nodes and the corresponding links, thereby realizing the transmission operation between the node feature information and the edge feature information. The obtained node feature information and edge feature information are used to solve the node problem and link problem proposed above, so that the obtained node feature information and edge feature information have the ability to be reused.

[0217] After obtaining the node feature information and edge feature information, the node feature information may include feature data of multiple dimensions, for example: feature data of network delay dimension, feature data of packet loss rate dimension, feature data of hop count dimension, feature data of bandwidth dimension, feature data of jitter dimension, etc.; or, the feature data may include: feature data of node dimension, feature data of edge dimension in the same region, feature data of edge dimension across regions, etc., and then the feature data of multiple dimensions may be spliced ​​to obtain the node feature information, and then the first sub-network model may be used to perform node fault detection operation on the node feature information to obtain node fault detection information, and the node fault detection information is used to identify whether there is a fault in a network node in a cross-domain network. When a network node fails, all links pointing to the network node will most likely cause packet loss.

[0218] In some other instances, using the second sub-network model to perform network detection on time series data and spatial topology data to obtain edge quality detection information may include: extracting features from the time series data and spatial topology data to obtain node feature information and edge feature information; using the second sub-network model to perform fault detection operations on edges in the same area based on the node feature information and edge feature information to obtain edge fault detection information.

[0219] In other instances, using the third sub-network model to perform network detection on time series data and spatial topology data to obtain edge quality detection information may include: extracting features from the time series data and spatial topology data to obtain node feature information and edge feature information; using the third sub-network model to perform fault detection operations on edges across regions based on the node feature information and edge feature information to obtain edge fault detection information.

[0220] For the spatiotemporal graph convolutional neural network model, the spatiotemporal graph convolutional neural network can be deployed on an operation and maintenance system that can be continuously iterated through the deployment layer. Fig. 9 As shown, the operation and maintenance system combines the spatiotemporal graph convolutional neural network model and the analysis capability of PingMesh to realize cross-domain network detection operations, cross-domain network change operations, and cross-domain network diagnosis operations. When performing detection operations on cross-domain networks, the spatiotemporal graph convolutional neural network model is used to realize point-by-point alarm capabilities and second-level alarm operations; when performing diagnostic operations on cross-domain networks, a real-time / anytime cross-domain network quality analysis report can be obtained according to needs. The cross-domain network quality analysis report can include information such as fault causes, fault repair suggestions, and fault-related data. The obtained cross-domain network quality analysis report is helpful for locating and analyzing fault problems; when there are changes in the cross-domain network (network data updates, network version upgrades, etc.), the spatiotemporal graph convolutional neural network and PingMesh technology are used to realize the changes, so that the full observability of unattended changes can be realized, ensuring the white-box and observability of the changes.

[0221] It should be noted that the spatiotemporal graph convolutional neural network model can have automatic learning capabilities. Specifically, it can obtain real-time feedback from operation and maintenance / emergency personnel when a fault occurs, and can perform model upgrades based on iterative data and at regular intervals, thereby realizing self-learning operations of the neural network model and ensuring the iterative nature of the system.

[0222] The above technical solution can significantly improve the performance of abnormal analysis compared with the traditional method. The specific advantages are:

[0223] (1) Application of spatiotemporal graph convolutional neural network model: For the first time, spatiotemporal graph convolutional neural network technology is applied to cross-domain network quality analysis in cloud computing environment. By training the spatiotemporal graph convolutional neural network model and conducting deep learning on the network topology structure, more accurate prediction and analysis of network status can be achieved. Specifically, the spatiotemporal graph neural network model can be constructed based on GraphSAGE and Edge GAT. In order to ensure the training quality and effect of the spatiotemporal graph neural network model, for the model training sample data, the fault injection mechanism can be used to balance the positive and negative sample data, that is, to enhance the negative samples through fault injection, and to design four dimensions based on real scenarios to improve the robustness of the model and reduce the false alarm rate. The obtained spatiotemporal graph neural network model can solve node problems, link problems in the same area, and cross-regional link problems, and can analyze data from different levels (such as nodes, links, and areas), supporting multi-granular analysis operations such as fine-grained and coarse-grained. Compared with traditional methods, it can better capture the complex correlations and dynamic changes in the network and improve the accuracy of analysis.

[0224] (2) Integrated analysis of cross-domain network topology: Through the spatiotemporal graph convolutional neural network model, we deeply explored how different domains are connected to each other through network topology, and realized a new integrated analysis framework of cross-domain network topology, which can not only provide network quality analysis from a global perspective, but also accurately predict cross-domain impact by learning the interactions between different domains;

[0225] (3) Dynamic anomaly detection and location mechanism: Advanced anomaly detection algorithms can dynamically detect network operation status and promptly identify potential problems. More importantly, with the guidance of the spatiotemporal graph neural network model, the specific source of the problem (such as a specific node or link) can be quickly located, greatly improving the efficiency and accuracy of fault handling, which is of great significance to actual network operation and maintenance;

[0226] (4) Integration of multiple data sources and indicators: PingMesh detection data is combined with backbone network topology data to form a comprehensive analysis of time series and spatial topology, providing a more comprehensive perspective on network quality. This not only takes into account the physical topology of the network, but also integrates multiple data sources including network traffic, device status, and service quality indicators. The temporal and spatial characteristics are used to perform deep learning on these data, realizing a multi-dimensional analysis of network quality, making the network analysis more comprehensive and able to adapt to the diversity and complexity of network behavior.

[0227] The technical solution provided by this application embodiment, combined with the spatiotemporal graph convolutional neural network model and PingMesh detection technology, can stably implement quality detection operations across domain networks. The spatiotemporal graph convolutional neural network model has the ability of automatic learning and iterative updating. Specifically, it can perform self-learning and regular updates through real-time feedback from operation and maintenance personnel, thereby improving the adaptability and accuracy of the system and ensuring long-term effective link quality analysis. Secondly, it can integrate multi-dimensional data (for example: delay, packet loss rate, and bandwidth utilization) to achieve comprehensive network quality assessment, thereby reducing failures caused by errors in single indicator judgment. In addition, the use of the spatiotemporal graph convolutional neural network model for intelligent fault prediction can provide early warning before the problem occurs, and can specifically achieve point-by-point alarm capabilities in seconds. This not only ensures timely response to changes in network quality, but also improves the timeliness and accuracy of fault detection, avoiding the proliferation of alarms under large-scale faults; finally, the spatiotemporal graph convolutional neural network model can achieve all-round observability, specifically in the three aspects of detection, change and diagnosis, to ensure the efficiency and transparency of network operations, and by setting up detection points in multiple areas, all-round real-time detection and full coverage of detection points can be achieved, thereby further improving the accuracy and response capabilities of detection, which can significantly enhance the capabilities and user experience of enterprises in cross-domain network management, and further ensure the practicality of the solution.

[0228] Fig.10 A flowchart of a data processing method provided by an embodiment of the present invention; see attached Fig.10 As shown, this embodiment provides a data processing method, the execution subject of the method is a data processing device, and the data processing device can be implemented as software, or a combination of software and hardware. When the data processing device is implemented as hardware, it can be various electronic devices with data processing capabilities, including but not limited to personal computers, servers, etc. When the data processing device is implemented as software, it can be installed in the above-mentioned electronic devices. Based on the above-mentioned data processing device, the data processing method in this embodiment may include:

[0229] Step S1001: Acquire a network topology diagram and historical fault data of a cross-domain network, wherein the network topology diagram includes a plurality of network nodes, edges between any two network nodes, and network operation information corresponding to each network node.

[0230] Step S1002: Based on historical fault data, fault configuration information for implementing fault configuration operations is determined, where the fault configuration information at least includes: fault type, fault occurrence probability corresponding to each fault type, and fault adjustment parameter corresponding to each fault type.

[0231] Step S1003: performing a fault addition operation in the network topology map based on the historical fault data to obtain training sample data, wherein the training sample data includes balanced positive sample data and negative sample data.

[0232] Among them, the specific implementation process and implementation principle of each step in this embodiment are similar to the specific implementation process and implementation principle of "performing fault additional configuration operation in the network topology diagram" in the above embodiment. Please refer to the above statements for details and will not be repeated here.

[0233] In addition, the method in this embodiment may also include the above Figure 2-Figure 9 For other method steps shown in the embodiment not described in detail, please refer to Figure 2-Figure 9 The implementation process and technical effects of this technical solution refer to Figure 2-Figure 9 The description in the illustrated embodiment will not be repeated here.

[0234] Fig.11 A schematic diagram of a cross-domain network quality detection device provided by an embodiment of the present invention; Fig.11 As shown, this embodiment provides a cross-domain network quality detection device, which is used to perform the above Figure 2 The quality detection method of the cross-domain network shown is specifically; the device may include:

[0235] A first acquisition module 11 is used to acquire a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network;

[0236] A first determination module 12 is used to determine time series data and spatial topology data corresponding to the cross-domain network, the time series data is used to identify the state characteristics of the cross-domain network that change over time, and the spatial topology data is used to identify the structural characteristics of the cross-domain network;

[0237] The first processing module 13 is used to input the time series data and the spatial topology data into the neural network model to obtain the quality detection information of the cross-domain network output by the neural network model.

[0238] In some instances, the neural network model is a spatiotemporal graph convolutional neural network model; when the first acquisition module 11 acquires the neural network model for implementing quality inspection operations on cross-domain networks, the first acquisition module 11 is used to execute: acquiring historical fault data of the cross-domain network; determining a network topology graph of the cross-domain network, the network topology graph including multiple network nodes, edges after any two network nodes, and network operation information corresponding to each network node; performing a fault addition operation in the network topology graph based on the historical fault data to obtain training sample data, the training sample data including balanced positive sample data and negative sample data; performing a model training operation using the training sample data to obtain a spatiotemporal graph convolutional neural network model for implementing quality inspection operations on cross-domain networks.

[0239] In some instances, when the first acquisition module 11 performs a fault adding operation in the network topology map based on historical fault data to obtain training sample data, the first acquisition module 11 is used to execute: based on the historical fault data, determine the fault configuration information corresponding to the cross-domain network, the fault configuration information includes at least: the fault type, the fault occurrence probability corresponding to each fault type, and the fault adjustment parameter corresponding to each fault type; based on the fault configuration information, perform a fault adding operation in the network topology map to obtain training sample data.

[0240] In some instances, when the first acquisition module 11 performs a fault adding operation in the network topology map based on the fault configuration information to obtain training sample data, the first acquisition module 11 is used to execute: performing a node fault adding operation in the network topology map based on the fault configuration information to obtain node fault sample data; performing an edge fault adding operation in the network topology map based on the fault configuration information to obtain edge fault sample data; and determining training sample data based on the node fault sample data and the edge fault sample data.

[0241] In some instances, when the first acquisition module 11 performs an additional operation of node failure in the network topology map based on the fault configuration information to obtain node failure sample data, the first acquisition module 11 is used to execute: randomly sampling the network nodes in the network topology map to obtain fault sampling points; determining at least one associated edge with the fault sampling point as the destination node; and performing an additional operation of packet loss rate on at least one associated edge based on the fault configuration information to obtain node failure sample data.

[0242] In some instances, when the first acquisition module 11 performs a packet loss rate allocation operation on at least one sampling edge based on the fault configuration information to obtain node fault sample data, the first acquisition module 11 is used to execute: determining at least one fault allocation edge for performing the packet loss rate allocation operation in at least one associated edge; determining the packet loss rate configuration information corresponding to each fault allocation edge based on the fault configuration information, the packet loss rate configuration information including at least: the probability of occurrence of the packet loss rate, and the numerical range of the packet loss rate corresponding to each occurrence probability; performing a packet loss rate allocation operation on at least one fault allocation edge based on the packet loss rate configuration information to obtain node fault sample data.

[0243] In some instances, when the first acquisition module 11 determines at least one fault-added edge for performing an operation of increasing the packet loss rate in at least one associated edge, the first acquisition module 11 is used to execute: determining edge proportion information for implementing the operation of increasing the packet loss rate based on historical fault data; and determining at least one fault-added edge for performing an operation of increasing the packet loss rate in at least one associated edge based on the edge proportion information.

[0244] In some instances, when the first acquisition module 11 performs an operation of adding a packet loss rate on at least one associated edge based on the fault configuration information to obtain node fault sample data, the first acquisition module 11 is used to execute: based on historical fault data, determining the packet loss probability information corresponding to each associated edge; determining the fault configuration information corresponding to each associated edge, the fault configuration information including at least: the probability of occurrence of the packet loss rate, and the numerical range of the packet loss rate corresponding to each occurrence probability; performing an operation of adding a packet loss rate on each associated edge based on the packet loss probability information and the fault configuration information to obtain node fault sample data.

[0245] In some instances, when the first acquisition module 11 performs an edge fault configuration operation in the network topology map based on the fault configuration information to obtain edge fault sample data, the first acquisition module 11 is used to execute: randomly sampling the edges in the network topology map to obtain fault sampling edges, the fault sampling edges include: first type of sampling edges located in the same area and second type of sampling edges located between different areas; based on historical fault data, determining a target packet loss rate for performing the edge fault configuration operation, the target packet loss rate includes at least: the probability of occurrence of the packet loss rate, and the packet loss rate occurrence range corresponding to each occurrence probability; based on the target packet loss rate, performing an edge fault configuration operation on the fault sampling edges to obtain edge fault sample data, the edge fault sample data include: first edge fault sample data corresponding to the first type of sampling edges, and second edge fault sample data corresponding to the second type of sampling edges.

[0246] In some instances, when the first acquisition module 11 determines a target packet loss rate for performing an additional configuration operation for an edge failure based on historical fault data, the first acquisition module 11 is used to execute: obtaining edge loss information corresponding to each fault sampling edge; determining an initial fault packet loss rate for each fault sampling edge based on historical fault data, the initial fault packet loss rate including at least one of the following: an occurrence probability of the packet loss rate, a numerical range of the packet loss rate corresponding to each occurrence probability; determining a target packet loss rate for performing an additional configuration operation for an edge failure based on the edge loss information and the initial fault packet loss rate.

[0247] In some instances, the fault sampling edge is a second-type sampling edge; when the first acquisition module 11 performs an edge fault adding operation on the fault sampling edge based on the target packet loss rate to obtain edge fault sample data, the first acquisition module 11 is used to execute: obtaining adjacent edges and second-degree edges of the second-type sampling edges; determining a preset decay ratio for performing a fault diffusion operation on the adjacent edges and second-degree edges; performing an edge fault adding operation on the second-type sampling edge based on the target packet loss rate, and performing a fault diffusion operation on the adjacent edges and second-degree edges according to the preset decay ratio to obtain second edge fault sample data.

[0248] In some examples, after determining the target packet loss rate for performing the edge failure provisioning operation, the first acquisition module 11 and the first processing module 13 in this embodiment are used to perform the following steps:

[0249] A first acquisition module 11 is used to acquire white noise information for performing an additional configuration operation for edge failures;

[0250] The first processing module 13 is used to adjust the target packet loss rate based on the white noise information to obtain an adjusted packet loss rate.

[0251] In some instances, when the first processing module 13 uses a neural network model to perform network detection on time series data and spatial topology data to obtain quality detection information of a cross-domain network, the first processing module 13 is used to perform the following steps: use the first sub-network model in the neural network model to perform node fault detection operations on the time series data and spatial topology data to obtain node fault quality detection information; use the second sub-network model in the neural network model to perform edge fault detection operations within a preset area on the time series data and spatial topology data to obtain first edge fault quality detection information; use the third sub-network model in the neural network model to perform edge fault detection operations on the time series data and spatial topology data across regions to obtain second edge fault quality detection information; determine the quality detection information of the cross-domain network based on the node fault quality detection information, the first edge fault quality detection information, and the second edge fault quality detection information.

[0252] In some instances, when the first processing module 13 uses the first sub-network model in the neural network model to perform node fault detection operations on time series data and spatial topology data to obtain node fault quality detection information, the first processing module 13 is used to perform the following steps: based on the spatial topology data, determine at least one network node in the cross-domain network and the neighbor nodes corresponding to each network node; based on the time series data, determine the first node representation of each network node and the second node representation of the corresponding neighbor node; aggregate the first node representation and the second node representation of the corresponding neighbor node to obtain an aggregated node representation; use the first sub-network model in the neural network model to perform node fault detection operations on the aggregated node representation to obtain node fault quality detection information.

[0253] In some instances, after obtaining the quality detection information of the cross-domain network, the first processing module 13 in this embodiment is also used to execute: when the quality detection information is used to identify the existence of a fault in the cross-domain network, determine the fault node or fault edge corresponding to the quality detection information; determine the fault impact range corresponding to the fault node or fault edge; and associate and display the fault node or fault edge and the corresponding fault impact range.

[0254] Fig.11 The device shown can perform Figure 2-Figure 9 For the method of the embodiment shown in the figure, the part not described in detail in this embodiment can be referred to Figure 2-Figure 9 The implementation process and technical effects of this technical solution refer to Figure 2-Figure 9 The description in the illustrated embodiment will not be repeated here.

[0255] In one possible design, Fig.11 The structure of the cross-domain network quality detection device shown can be implemented as an electronic device, which can be a controller, a personal computer, a server, and other devices, such as Fig.12 As shown, the electronic device may include: a first processor 21 and a first memory 22. The first memory 22 is used to store the corresponding electronic device to execute the above Figure 2-Figure 9 In the illustrated embodiment, the first processor 21 is configured to execute the program stored in the first memory 22 .

[0256] The program includes one or more computer instructions, wherein when the one or more computer instructions are executed by the first processor 21, the following steps can be implemented: obtaining a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network; determining time series data and spatial topology data corresponding to the cross-domain network, the time series data being used to identify state characteristics of the cross-domain network that change over time, and the spatial topology data being used to identify structural characteristics of the cross-domain network; inputting the time series data and the spatial topology data into the neural network model to obtain quality detection information of the cross-domain network output by the neural network model.

[0257] Furthermore, the first processor 21 is also used to execute the aforementioned Figure 2 All or part of the steps in the illustrated embodiments.

[0258] The structure of the electronic device may further include a first communication interface 23 for the electronic device to communicate with other devices or a communication network.

[0259] In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions used by electronic devices, which includes instructions for executing the above Figure 2 The program involved in the cross-domain network quality detection method in the illustrated embodiment.

[0260] In addition, an embodiment of the present invention provides a computer program product, including: a computer-readable storage medium storing computer instructions, when the computer instructions are executed by one or more processors, causing the one or more processors to execute the above Figure 2 The steps in the cross-domain network quality detection method in the method embodiment are shown.

[0261] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0262] The device embodiments described above are merely illustrative, wherein 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, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0263] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by adding a necessary general hardware platform, and of course can also be implemented by combining hardware and software. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a computer product, and the present invention can be implemented in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0264] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable device to generate a machine, so that the instructions executed by the processor of the computer or other programmable device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0265] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable device to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide the functions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps of the functions specified in the blocks or multiple blocks. In a typical configuration, the computing device includes one or more processors (CPU), input / output interfaces, network interfaces and memory. The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0266] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0267] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting quality of a cross-domain network, characterized in that: include: Acquire a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network; Determine time series data and spatial topology data corresponding to the cross-domain network, the time series data is used to identify state characteristics of the cross-domain network that change over time, and the spatial topology data is used to identify structural characteristics of the cross-domain network; The time series data and the spatial topology data are input into the neural network model to obtain quality detection information of the cross-domain network output by the neural network model, wherein the neural network model includes: a first sub-network model for implementing node fault detection operations, a second sub-network model for implementing edge fault detection operations within a preset area, and a third sub-network model for implementing edge fault detection operations across regions, and the quality detection information includes node fault quality detection information obtained by the first sub-network model, first edge fault quality detection information obtained by the second sub-network model, and second edge fault quality detection information obtained by the third sub-network model.

2. The method according to claim 1, characterized in that The neural network model is a spatiotemporal graph convolutional neural network model; obtaining a neural network model for implementing a quality detection operation of a cross-domain network includes: Acquiring historical fault data of the cross-domain network; Determine a network topology diagram of the cross-domain network, wherein the network topology diagram includes a plurality of network nodes, edges between any two network nodes, and network operation information corresponding to each network node; Performing a fault addition operation in the network topology map based on the historical fault data to obtain training sample data, wherein the training sample data includes balanced positive sample data and negative sample data; The training sample data is used to perform a model training operation to obtain a spatiotemporal graph convolutional neural network model for implementing a quality detection operation of a cross-domain network.

3. The method according to claim 2, characterized in that Performing a fault addition operation in the network topology diagram based on the historical fault data to obtain training sample data includes: Based on the historical fault data, determine fault configuration information corresponding to the cross-domain network, the fault configuration information at least including: fault type, fault occurrence probability corresponding to each fault type, and fault adjustment parameter corresponding to each fault type; Based on the fault configuration information, a fault configuration operation is performed in the network topology diagram to obtain training sample data.

4. The method according to claim 3, characterized in that: Performing a fault configuration operation in the network topology map based on the fault configuration information to obtain training sample data includes: Performing an additional configuration operation of a node failure in the network topology diagram based on the fault configuration information to obtain node failure sample data; Performing an additional configuration operation of an edge fault in the network topology diagram based on the fault configuration information to obtain edge fault sample data; The training sample data is determined based on the node fault sample data and the edge fault sample data.

5. The method according to claim 4, characterized in that Performing an additional configuration operation of a node failure in the network topology map based on the fault configuration information to obtain node failure sample data includes: Randomly sampling network nodes in the network topology diagram to obtain fault sampling points; Determine at least one associated edge having the fault sampling point as a destination node; Based on the fault configuration information, a packet loss rate configuration operation is performed on the at least one associated edge to obtain node fault sample data.

6. The method according to claim 5, characterized in that Performing a packet loss rate configuration operation on the at least one sampling edge based on the fault configuration information to obtain node fault sample data includes: Determine at least one fault-addition edge for performing an addition operation of the packet loss rate in the at least one associated edge; Based on the fault configuration information, determine the packet loss rate configuration information corresponding to each fault configuration edge, wherein the packet loss rate configuration information at least includes: the probability of occurrence of the packet loss rate, and the value range of the packet loss rate corresponding to each occurrence probability; Based on the packet loss rate configuration information, a packet loss rate configuration operation is performed on the at least one fault configuration edge to obtain the node fault sample data.

7. The method according to claim 6, characterized in that Determining at least one fault-addition edge for performing an addition operation of a packet loss rate in the at least one associated edge includes: Based on the historical fault data, determining edge ratio information for implementing an additional configuration operation for a packet loss rate; At least one faulty additional edge for performing an additional operation of the packet loss rate is determined in the at least one associated edge based on the edge proportion information.

8. The method according to claim 5, characterized in that Performing a packet loss rate configuration operation on the at least one associated edge based on the fault configuration information to obtain node fault sample data includes: Based on the historical fault data, determining packet loss probability information corresponding to each associated edge; Determine the fault configuration information corresponding to each associated edge, wherein the fault configuration information includes at least: the probability of occurrence of the packet loss rate, and the numerical range of the packet loss rate corresponding to each occurrence probability; Based on the packet loss probability information and the fault configuration information, a packet loss rate configuration operation is performed on each associated edge to obtain the node fault sample data.

9. The method according to claim 4, characterized in that Performing an additional configuration operation of an edge fault in the network topology map based on the fault configuration information to obtain edge fault sample data includes: Randomly sampling edges in the network topology graph to obtain fault sampling edges, wherein the fault sampling edges include: first-type sampling edges located in the same area and second-type sampling edges located between different areas; Based on the historical fault data, determine a target packet loss rate for performing an edge fault reconfiguration operation, wherein the target packet loss rate includes at least: an occurrence probability of the packet loss rate and a packet loss rate occurrence range corresponding to each occurrence probability; An edge fault addition operation is performed on the fault sampling edge based on the target packet loss rate to obtain the edge fault sample data, wherein the edge fault sample data includes: first edge fault sample data corresponding to the first type of sampling edge and second edge fault sample data corresponding to the second type of sampling edge.

10. The method according to claim 9, characterized in that Determining a target packet loss rate for performing an edge failure provisioning operation based on the historical failure data includes: Obtain edge loss information corresponding to each fault sampling edge; Based on the historical fault data, determine the initial fault packet loss rate of each fault sampling edge, wherein the initial fault packet loss rate includes at least one of the following: an occurrence probability of the packet loss rate, and a numerical range of the packet loss rate corresponding to each occurrence probability; Based on the edge loss information and the initial failure packet loss rate, a target packet loss rate for performing an edge failure provisioning operation is determined.

11. The method according to claim 9, characterized in that The fault sampling edge is a second type of sampling edge; performing an edge fault configuration operation on the fault sampling edge based on the target packet loss rate to obtain the edge fault sample data, including: Obtaining adjacent edges and second-degree edges of the second-type sampling edges; determining a preset decay ratio for performing a fault spreading operation on the adjacent edges and the second-degree edges; An edge fault addition operation is performed on the second type of sampling edge based on the target packet loss rate, and a fault diffusion operation is performed on the adjacent edge and the second-degree edge according to the preset decay ratio to obtain the second edge fault sample data.

12. The method according to any one of claims 1 to 11, characterized in that: Inputting the time series data and the spatial topology data into the neural network model to obtain the quality detection information of the cross-domain network output by the neural network model includes: Using the first sub-network model in the neural network model to perform node fault detection operations on the time series data and the spatial topology data to obtain node fault quality detection information; Using the second sub-network model in the neural network model, performing an edge fault detection operation in a preset area on the time series data and the spatial topology data to obtain first edge fault quality detection information; Using a third sub-network model in the neural network model, performing an edge fault detection operation across regions on the time series data and the spatial topology data to obtain second edge fault quality detection information; Based on the node failure quality detection information, the first edge failure quality detection information and the second edge failure quality detection information, quality detection information of the cross-domain network is determined.

13. The method according to claim 12, characterized in that Using the first sub-network model in the neural network model to perform a node fault detection operation on the time series data and the spatial topology data to obtain node fault quality detection information, including: Based on the spatial topology data, determining at least one network node in the cross-domain network and neighbor nodes corresponding to each network node; Based on the time series data, determining a first node representation of each network node and a second node representation of a corresponding neighboring node; Aggregating the first node representation and the second node representation of the corresponding neighboring node to obtain an aggregated node representation; The first sub-network model in the neural network model is used to perform a node fault detection operation on the aggregation node representation to obtain the node fault quality detection information.

14. The method according to any one of claims 1 to 11, characterized in that: After obtaining the quality detection information of the cross-domain network, the method further includes: When the quality detection information is used to identify a fault in the cross-domain network, determining a faulty node or a faulty edge corresponding to the quality detection information; Determine the fault impact range corresponding to the fault node or fault edge; The fault node or the fault edge and the corresponding fault impact range are displayed in association.

15. A quality detection device for a cross-domain network, characterized in that: include: A first acquisition module, used to acquire a cross-domain network to be detected and a neural network model for implementing a quality detection operation of the cross-domain network, wherein the neural network model is obtained by training based on preset network sample data; A first determination module is used to determine time series data and spatial topology data corresponding to the cross-domain network, the time series data is used to identify state characteristics of the cross-domain network that change over time, and the spatial topology data is used to identify structural characteristics of the cross-domain network; A first processing module is used to input the time series data and the spatial topology data into the neural network model to obtain quality detection information of the cross-domain network output by the neural network model, wherein the neural network model includes: a first sub-network model for implementing node fault detection operations, a second sub-network model for implementing edge fault detection operations within a preset area, and a third sub-network model for implementing edge fault detection operations across regions, and the quality detection information includes node fault quality detection information obtained by the first sub-network model, first edge fault quality detection information obtained by the second sub-network model, and second edge fault quality detection information obtained by the third sub-network model.

16. An electronic device, characterized in that: include: A memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the method of any one of claims 1 to 14.

17. A computer storage medium, characterized in that: Used to store a computer program, which enables a computer to implement the method of any one of claims 1 to 14 when executed.

18. A computer program product, characterized in that include: A computer program, when executed by a processor of an electronic device, causes the processor to execute the steps of any one of the methods of claims 1 to 14.

Images