Method, device and equipment for training network diagnosis model and storage medium

Abstract

According to embodiments of the present disclosure, a method, an apparatus, a device and a storage medium for training a network diagnosis model are provided. The method comprises: simulating a set of network states by using a network simulator; constructing a plurality of training samples respectively corresponding to the plurality of network states based on the simulation of the set of network states, wherein each training sample in the plurality of training samples comprises at least device information related to a network state corresponding to the training sample, and the plurality of network states are obtained based on the set of network states; for each training sample in the plurality of training samples, generating a model diagnosis result for the network state corresponding to the training sample by using the network diagnosis model based on the device information related to the network state, wherein the network diagnosis model is configured to perform diagnosis on the network state; and updating the network diagnosis model based on the model diagnosis results respectively generated for the plurality of training samples and diagnosis reference information respectively corresponding to the plurality of training samples.

Description

Technical Field

[0001] The exemplary embodiments disclosed herein generally relate to the field of computers, and more particularly to methods, apparatus, devices, and computer-readable storage media for training network diagnostic models. Background Technology

[0002] With the rapid development of computer technology, network fault diagnosis is a key component in ensuring stable network operation and efficient maintenance. In the rapidly evolving digital age, the network is one of the fundamental infrastructures of an organization's operations. Any network failure can quickly lead to decreased productivity, data loss, or even business interruption. Therefore, the rapid identification, diagnosis, and resolution of network faults are crucial. Summary of the Invention

[0003] In a first aspect of this disclosure, a method for training a network diagnostic model is provided, comprising: simulating a set of network states using a network simulator, the network simulator being configured to simulate multiple component devices of a network and connections between the multiple component devices, and simulating different network states; constructing multiple training samples corresponding to the multiple network states based on the simulation of the set of network states, each training sample including at least device information related to the network state corresponding to the training sample, the multiple network states being derived from the set of network states; generating a model diagnostic result for the corresponding network state using a network diagnostic model for each training sample based on the device information related to the corresponding network state, the network diagnostic model being configured to perform network state diagnosis; and updating the network diagnostic model based on the model diagnostic results generated for the multiple training samples and diagnostic reference information corresponding to the multiple training samples respectively; wherein the diagnostic reference information indicates the correct diagnostic result corresponding to the training sample.

[0004] In a second aspect of this disclosure, an apparatus for training a network diagnostic model is provided, comprising: a network state simulation module configured to simulate a set of network states using a network simulator, the network simulator being configured to simulate multiple component devices of a network and the connections between the multiple component devices, and to simulate different network states; a training sample construction module configured to construct multiple training samples corresponding to the multiple network states based on the simulation of the set of network states, wherein each training sample includes at least device information related to the network state corresponding to the training sample, and the multiple network states are obtained based on a set of network states; a diagnostic result generation module configured to generate a model diagnostic result for each of the multiple training samples based on the device information related to the corresponding network state using the network diagnostic model, the network diagnostic model being configured to perform network state diagnosis; and a network diagnostic model update module configured to update the network diagnostic model based on the model diagnostic results generated for each of the multiple training samples and diagnostic reference information corresponding to each of the multiple training samples; wherein the diagnostic reference information indicates the correct diagnostic result corresponding to the training sample.

[0005] In a third aspect of this disclosure, an electronic device is provided. The device includes at least one processor; and at least one memory coupled to the at least one processor and storing instructions for execution by the at least one processor. When executed by the at least one processor, the instructions cause the device to perform the method of the first aspect.

[0006] In a fourth aspect of this disclosure, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program that can be executed by a processor to implement the method of the first aspect.

[0007] According to a fifth aspect of this disclosure, a computer program product is provided, including computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the method of the first aspect.

[0008] It should be understood that the content described in this content section is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0009] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0010] Figure 1 A schematic diagram of an example environment in which embodiments of the present disclosure can be implemented is shown;

[0011] Figure 2 A flowchart is shown illustrating a process for training a network diagnostic model according to some embodiments of the present disclosure;

[0012] Figure 3 A schematic diagram of an example architecture for updating a network diagnostic model according to some embodiments of the present disclosure is shown;

[0013] Figure 4 A schematic diagram of an example architecture for determining training samples for updating a network diagnostic model according to some embodiments of the present disclosure is shown;

[0014] Figure 5 A block diagram of an apparatus for training a network diagnostic model according to some embodiments of the present disclosure is shown; and

[0015] Figure 6 A block diagram of an apparatus capable of implementing one or more embodiments of the present disclosure is shown. Detailed Implementation

[0016] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0017] In the description of embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". Other explicit and implicit definitions may also be included below.

[0018] In this document, unless explicitly stated otherwise, performing a step in response to A does not mean that the step is performed immediately after A, but may include one or more intermediate steps.

[0019] As used in this paper, the term "model" refers to a system that learns the relationship between inputs and outputs from training data, enabling it to generate corresponding outputs for a given input after training. Model generation can be based on machine learning techniques. Deep learning is a machine learning algorithm that uses multiple layers of processing units to process inputs and provide corresponding outputs. In this paper, "model" may also be referred to as a "machine learning model," a "machine learning network," or simply a "network," and these terms are used interchangeably. A model can also include different types of processing units or networks.

[0020] As used herein, a “unit,” “operation unit,” or “subunit” can consist of any suitable machine learning model or network. As used herein, a set of elements or similar expressions can include one or more such elements. For example, “a set of convolutional units” can include one or more convolutional units.

[0021] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0022] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure through appropriate means in accordance with relevant laws and regulations, and user authorization should be obtained.

[0023] For example, in response to receiving a user's active request, a prompt message is sent to the user to clearly inform the user that the requested operation will require the acquisition and use of the user's personal information, thereby enabling the user to choose whether to provide personal information to the software or hardware such as electronic devices, applications, servers or storage media that perform the operation of the technical solution disclosed herein, based on the prompt message.

[0024] As an optional but non-restrictive implementation, in response to a user's active request, a prompt message can be sent to the user, such as a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0025] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0026] As briefly mentioned earlier, any network failure can quickly lead to decreased productivity, data loss, and even business interruption. Conventionally, network failures, their causes, and solutions can be troubleshooted using Standard Operating Procedures (SOPs). SOPs are a set of written instructions designed to help managers execute complex operational processes, ensuring consistency, efficiency, and quality. However, this method is inefficient, labor-intensive, and costly. Correspondingly, machine learning models (e.g., Large Language Models, LLM) can be used to determine network failures, their causes, and solutions based on SOPs and prompts. However, because SOPs involve a lot of content, the prompts are often long, significantly reducing the model's comprehension and planning capabilities.

[0027] Furthermore, machine learning models for handling network failures can be tuned by collecting data from real-world environments. However, this approach only allows data collection from actual failures, increasing the cost of data gathering. Consequently, the limited coverage of failure types by the data collected from real-world failures restricts the machine learning model's ability to handle network failures.

[0028] In view of this, this disclosure proposes an improved scheme for training a network diagnostic model. According to the scheme of an embodiment of this disclosure, a network simulator is used to simulate a set of network states. The network simulator is configured to simulate multiple component devices of the network and the connections between the component devices, as well as different network states. Further, based on the simulation of the set of network states, multiple training samples corresponding to the multiple network states are constructed. Each training sample includes at least device information related to the network state corresponding to that training sample, and the multiple network states are obtained based on the set of network states. Accordingly, for each training sample, based on the device information related to the corresponding network state, a model diagnostic result for the corresponding network state is generated using a network diagnostic model, which is configured to perform network state diagnosis. Based on the model diagnostic results generated for each of the multiple training samples and diagnostic reference information corresponding to each of the multiple training samples, the network diagnostic model is updated. The diagnostic reference information indicates the correct diagnostic result corresponding to the training sample.

[0029] In this way, a large number of network states can be simulated using a network simulator, such as various types of faulty and normal network states. Furthermore, various types of training samples can be constructed based on these network states, thus covering a wide range of training data to update the network diagnostic model and reducing the cost of data collection. In this way, by utilizing these training samples, a network diagnostic model with fault diagnosis and repair capabilities can be trained efficiently.

[0030] Example embodiments of this disclosure are described below with reference to the accompanying drawings.

[0031] Figure 1 A schematic diagram of an example environment 100 in which embodiments of the present disclosure can be implemented is shown. Figure 1 In environment 100, it is desired to train and use a network diagnostic model 130 configured for various application environments. For example, network diagnostic model 130 may output a sample set 120 of diagnostic results for a sample set 112 of network states, based on an acquired sample set 112 of network states. In some embodiments, a set of network states may include different network states corresponding to multiple simulated networks with different network structures. For example, normal network states under normal network conditions and faulty network states under faulty network conditions. In some embodiments, the sample set 112 of network states may be obtained using a network simulator 170. For example, for any simulated network among multiple simulated networks, the network simulator 170 may be used to obtain network states generated based on at least one fault operation, thereby obtaining a sample set 112 of network states. In some embodiments, at least one fault operation may be preset by a user (e.g., an administrator).

[0032] like Figure 1 As shown, environment 100 includes model update system 150. Figure 1 The upper part illustrates the model training phase (i.e., the model update phase), and the lower part illustrates the model application phase. Before the update, the parameter values ​​of the network diagnostic model 130 can have the values ​​before the update, or they can have pre-trained parameter values ​​obtained through a pre-training process. The network diagnostic model 130 can be trained via a sample set 112 based on a set of network states and a sample set 120 of diagnostic results. During the update process, the parameter values ​​of the network diagnostic model 130 can be updated and adjusted. After the network diagnostic model 130 is updated, the network diagnostic model 130 is obtained. ′ At this point, network diagnostic model 130 ′The parameter values ​​have been updated, and based on the updated parameter values, the network diagnostic model 130 can be used to perform various types of network diagnostic or recovery tasks during the model application phase. For example, the network diagnostic model 130 can be used to troubleshoot network faults, determine the cause of the network fault, and resolve the network fault, etc.

[0033] During the model update phase, the network diagnostic model 130 can be trained using the model update system 150, based on a sample set 110 that includes a set of network states 112 used to update the network diagnostic model 130 and a sample set 120 of model diagnostic results. Here, the network diagnostic model 130 can generate a set of sample sets 120 of model diagnostic results corresponding to each set of network states 112, based on the sample set 112. Subsequently, the model update system 150 can also update the network diagnostic model 130 based on the sample set 112 of network states included in the sample set 110 and the sample set 120 of model diagnostic results. Specifically, the update process can be performed iteratively using a large number of training samples. After the update is complete, the network diagnostic model 130 can include knowledge about the task to be processed. During the model application phase, the network diagnostic model 130... ′ (Network diagnostic model 130 at this time) ′ (With updated parameter values) can be used to perform the corresponding task. For example, in a network diagnostic task, network diagnostic model 130 ′ It can receive network state 142 and output the corresponding target model diagnostic result 144. If the current network state 142 is a faulty network state, the target model diagnostic result 144 can indicate that network state 142 is faulty. In the task of determining the cause of network faults, network diagnostic model 130 ′ It can receive network state 142 and output the corresponding target model diagnostic result 144. In this scenario, the target model diagnostic result 144 can indicate the cause of the failure in network state 142. For example, in the task of resolving network faults, network diagnostic model 130... ′ It can receive network status 142 and output the corresponding target model diagnostic result 144. In this scenario, the target model diagnostic result 144 can indicate the steps to resolve the fault corresponding to network status 142. In some embodiments, the network diagnostic model 130 ′ The troubleshooting steps for network state 142 can be output in a step-by-step manner. Network diagnostic model 130 ′ It can also output the troubleshooting steps for network status 142 in a multi-step format.

[0034] exist Figure 1In this context, the model update system 150 may include any computing system with computing capabilities, such as various computing devices / systems, terminal devices, servers, etc. Terminal devices may involve any type of mobile terminal, fixed terminal, or portable terminal, including mobile phones, desktop computers, laptop computers, notebook computers, netbook computers, tablet computers, media computers, multimedia tablets, or any combination thereof, including accessories and peripherals of these devices or any combination thereof. Servers include, but are not limited to, mainframes, edge computing nodes, computing devices in cloud environments, etc.

[0035] It should be understood that Figure 1 The components and arrangements shown in environment 100 are merely examples, and a computing system suitable for implementing the exemplary implementations described in this disclosure may include one or more different components, other components, and / or different arrangements. Implementations of this disclosure are not limited in this respect.

[0036] The following description will continue with reference to the accompanying drawings, outlining some exemplary embodiments of this disclosure. In the hereinafter, the exemplary embodiments will be described primarily with respect to the model update system 150.

[0037] The following will refer to Figure 2 and Figure 3 This disclosure describes a scheme for network status diagnosis and repair. Figure 2 A flowchart of a process 200 for training a network diagnostic model according to some embodiments of the present disclosure is shown. Figure 3 A schematic diagram of an example architecture 300 for updating a network diagnostic model according to some embodiments of the present disclosure is shown. In embodiments of the present disclosure, it is desirable to update such a network diagnostic model 130, which is configured to determine multiple network states using a network simulator 170, and then update it based on a sample set of multiple training samples corresponding to the multiple network states and a sample set of model diagnostic results corresponding to the multiple network states. That is, after being updated by the generated sample set of multiple training samples corresponding to the multiple network states and a sample set of model diagnostic results corresponding to the multiple network states, the model diagnostic result corresponding to the input target network state is accurately output.

[0038] In process 200, at block 210, model update system 150 uses network simulator 170 to simulate a set of network states. In some embodiments, the network simulator is configured to simulate multiple component devices of the network and the connections between these component devices, as well as to simulate different network states. Network simulator 170 is configured to simulate different network states through the individual component devices and the connections between them. In some examples, network simulator 170 can utilize various types of component device simulators to build various types of network structures. For example, the component devices required to build the network simulator may include, but are not limited to, wireless access points (APs), routers, switches, terminal devices, and so on.

[0039] In some examples, the parameters of the multiple component devices included in the network device simulator 170 can be configured. The multiple component devices included in the network device simulator 170 can be in a powered-off or powered-on state according to their respective configurations. The connection between the multiple component devices included in the network device simulator 170 can also be configured to be connected or disconnected. Additionally, the network device simulator 170 may also include additional component devices for providing interference factors. Accordingly, the power of the component devices providing interference factors can be configured to simulate interference received by the access point (AP). Thus, various types of network conditions can be simulated using a network device simulator with multiple capabilities. However, the above are merely exemplary and this disclosure is not intended to limit the scope of the invention.

[0040] The following describes in detail how the model update system 150 simulates a set of network states.

[0041] In some embodiments, the model update system 150 can construct multiple simulated networks using multiple component devices simulated by a network device simulator. These simulated networks (also referred to as virtual networks) have different network structures from each other. The model update system 150 can select a simulated network from the multiple simulated networks, and then obtain a fault operation sequence by sequentially applying at least one fault operation to the simulated network. Subsequently, the model update system 150 can determine the network state (also referred to as virtual network state) of the selected simulated network after the fault operation sequence has been applied as one of a set of network states.

[0042] As shown in the example architecture 300, the model update system 150 can utilize the network simulator 170 to obtain a set of network states 112 for multiple simulated networks. The model update system 150 can use the network simulator 170 to obtain normal state samples 311 for multiple simulated networks in a normal state and multiple network states 310 (also referred to as fault samples) for multiple simulated networks in a fault state. In some embodiments, the multiple network states 310 obtained by the model update system 150 may include single-step fault samples 312 and multi-step fault samples 313. In some embodiments, the model update system 150 can obtain single-step fault samples 312 for multiple simulated networks in a single-step manner. A single-step fault sample 312 indicates that a user (e.g., an administrator) performs a single-step fault-inducing operation for a simulated network to cause the current network state of that network device to be in a fault state. Correspondingly, the model update system 150 can also obtain multi-step fault samples 313 in a multi-step manner. A multi-step fault sample 313 indicates that a user (e.g., an administrator) performs multiple fault-inducing operations for a simulated network at once to cause the current network state of that network device to be in a fault state.

[0043] In other words, the model update system 150 can construct multiple simulated networks with various network structures based on multiple virtual devices. Furthermore, the model update system 150 can obtain the network states of these simulated networks in a normal state (i.e., normal state sample 311), and thus incorporate these normal state network states as part of a set of network states 112. Further, the model update system 150 can randomly select any one simulated network (e.g., simulated network A) from the multiple simulated networks to obtain the network states for that simulated network in a fault state (e.g., single-step fault sample 312, multi-step fault sample 313). For ease of understanding, the following description uses simulated network A as an example to illustrate how the model update system 150 obtains the network states for simulated network A in a fault state.

[0044] In some embodiments, the model update system 150 may apply a first fault operation to the simulated network A. For example, it may randomly select a setting (including but not limited to parameters, on / off states, and inter-device link states) of a device from among the multiple virtual devices included in the simulated network A, and modify its value from a normal value (e.g., represented by good_val) to an abnormal value (e.g., represented by bad_val). Accordingly, the model update system 150 may also add the first fault operation to a fault operation sequence (e.g., represented by destroy_list). At this time, the network state of the simulated network A to which the first fault operation is applied (i.e., single-step fault sample 312) may be one of multiple network states 310.

[0045] In some embodiments, the model update system 150 can sequentially apply a second fault operation, a third fault operation, and a fourth fault operation to the A simulated network. For example, the fault operations applied to the A simulated network may include disconnecting the connection between the AA component device and the BB component device included in the A simulated network, modifying the parameter values ​​of the AA component device, adding interference devices to the A simulated network, and so on. Accordingly, the model update system 150 can add the second fault operation, the third fault operation, and the fourth fault operation to the fault operation sequence. At this time, the network state of the A simulated network to which the second fault operation, the third fault operation, and the fourth fault operation have been applied (i.e., the multi-step fault sample 313) can be used as one of multiple network states 310.

[0046] In some embodiments, the model update system 150 may store a sequence of fault operations and a sequence of truth operations corresponding to the fault operation sequences. The sequence of truth operations corresponding to the fault operation sequences indicates the inverse operation of each fault operation in the fault operation sequence; that is, the inverse operation of each fault operation can indicate a recovery operation for the fault operation. For example, if a fault operation changes the value of a certain operating parameter of a device from a normal value A to a fault value B, then the inverse operation or recovery operation can change the value of that operating parameter from the fault value B to the normal value A. In other words, the model update system 150 can repair the fault operation through the inverse operation of the fault operation. The model update system 150 can update the network diagnostic model 130 based on its stored sequence of fault operations and the sequence of truth operations. The following will describe in detail how the model update system 150 updates the network diagnostic model 130 according to its stored sequence of fault operations and the sequence of truth operations.

[0047] In some embodiments, the model update system 150 can also acquire training samples for updating the network diagnostic model 130 based on the real network. The model update system 150 can acquire training samples under normal conditions and training samples under fault conditions. In some examples, for training samples under fault conditions, the model update system 150 can assist in updating the network diagnostic model 130 based on pre-configured expert configuration information. In this case, when the network diagnostic model 130 is processing a real network in a fault condition, the expected operations included in the information related to the real network provided by the model update system 150 to the network diagnostic model 130 may include repair tools and normal condition information.

[0048] Therefore, a large number of network states can be simulated by using a network simulator, such as various types of faulty network states and normal network states.

[0049] The following will continue to refer to Figure 2 and Figure 3 This describes how to construct multiple training samples based on a set of network states 112.

[0050] Continuing process 200, in box 220, the model update system 150 constructs multiple training samples, each corresponding to a set of network states, based on simulations of a set of network states. In embodiments of this disclosure, the multiple network states are obtained based on a set of network states. Figure 3 In the example architecture 300 shown, the model update system 150 can transform a set of network states 112 into corresponding training samples 315. The training samples 315 include training samples 316 simulating a normal network state, and multiple training samples corresponding to multiple network states 310 (e.g., single-step samples 317, single-step samples 318, and multi-step samples 319). That is, the model update system 150 can construct normal state training samples 316 corresponding to normal state samples 311 in a set of network states 112. Correspondingly, the model update system 150 can construct single-step samples 317 corresponding to single-step fault samples 312 in a set of network states 112. The model update system 150 can also construct single-step samples 318 and / or multi-step samples 319 corresponding to multi-step fault samples 313 in a set of network states 112. The following will describe in detail how to construct some training samples corresponding to a set of network states.

[0051] In embodiments of this disclosure, each training sample includes at least device information related to the network state corresponding to that training sample. In some embodiments, the device information related to the network state corresponding to a particular training sample may include, but is not limited to, the state information of the constituent devices included in the simulated network, the state information of the current terminal device, etc. The network state information corresponding to a particular training sample can be provided by the model update system 130 to the network diagnostic model 130 as a prompt, thereby enabling the network model 130 to diagnose and repair the simulated network corresponding to that training sample.

[0052] In some embodiments, the prompts provided by the model update system 130 to the network diagnostic model 130 may also include information about a simulated network, tool operation information, log information, a list of provided tools, sample construction configuration information (e.g., pre-defined user input), and so on. The prompts will be described below with reference to Table 1, which is an example of a prompt.

[0053] Table 1

[0054]

[0055]

[0056] In some embodiments, the component devices and their status information included in the prompt message include, but are not limited to, the configuration parameters, operating status, and communication status between connected devices in the simulated network. In some examples, the large number of terminal devices (e.g., personal terminal devices) increases the load on the simulated network. Therefore, in scenarios where a network failure is caused by a terminal device, the administrator of that terminal device can initiate fault diagnosis, and the information of that terminal device can be included in its parameters and status information. Accordingly, the parameters and status information of the terminal devices included in the prompt message are optional. For example, if a terminal device fails, it may be due to a failure in the simulated network or a failure in the corresponding terminal device. Therefore, in scenarios where the model update system 150 does not provide terminal device information, it is necessary to focus on diagnosing the cause of the simulated network failure. Additionally, in scenarios where the model update system 150 provides terminal device information, it is necessary to diagnose the cause of the failure in both the simulated network and the corresponding terminal device.

[0057] In some embodiments, in single-step output mode, the tool operation information included in the prompt message consists of the solution obtained from the operation of the corresponding tool in the previous step. Typically, diagnostic tools provide rich output information when invoked, while repair tools may not provide output information when invoked. Because repair tools can change the state of the simulated network when invoked, the impact of the repair tool can be reflected in the "network device parameters and status information," thereby affecting the tool prediction in the next stage.

[0058] In some embodiments, the log information included in the prompt message may indicate the corresponding log information of each device acquired by the model update system 150. In some examples, the model update system 150 may filter log information according to log level (e.g., fatal > error > warning >, or any other appropriate log level). The model update system 150 may filter log information according to log time (e.g., from most recent to oldest, or any other appropriate time). The model update system 150 may use a machine learning model to filter log information from the above-mentioned log information for fault diagnosis and repair. In some embodiments, the length of the log information acquired by the model update system 150 may be greater than or equal to a length threshold, which may be pre-configured by a user (e.g., an administrator).

[0059] In some embodiments, the tool list included in the prompt message may include diagnostic tools for diagnosing the cause of a fault in the simulated network, such as ping, route tracing, and other diagnostic tools. In some embodiments, the tool list may include repair tools for repairing faults in the simulated network. For example, this may include, but is not limited to, modifying parameter settings of specific devices, the on / off state of specific devices, the connectivity status between devices, etc. A repair tool may be abstracted as `change(prev_val, cur_val)`, representing modifying the relevant settings of one or more settings from previous values ​​(e.g., represented by `prev_val`) to current values ​​(e.g., represented by `cur_val`). All tools used can be abstracted as functions `func(params)`, i.e., functions that accept zero to multiple parameters, are used to solve specific problems, and return results. In some embodiments, the tool list may include special tools for indicating that the simulated network is in a normal state, such as represented by `normal`. That is, if the network diagnostic model outputs `normal`, it indicates that the current simulated network is in a normal state, or that the simulated network has returned to normal after a series of repair operations. In some embodiments, the tool list may include the function name and description of each tool, a parameter list, and detailed descriptions of the corresponding parameters.

[0060] In some embodiments, the sample build configuration information (i.e., the pre-defined user input) included in the prompt message may include information such as the time of the failure and the device that failed. In scenarios where the prompt message does not include sample build configuration information, a time period from the current time to several minutes ago can be selected as the failure time to facilitate filtering logs within that time period, thereby helping to narrow down the scope of the failure. Correspondingly, if the time period covers the current time, the network status information can be the current status information. If the time period does not cover the current time, the network status information can be the periodically dumped network status information within the corresponding time period.

[0061] In some embodiments, the output type included in the prompt information can be set to a first type (e.g., single-step diagnosis), a second type (e.g., single-step repair), or multi-step. That is, the updated network diagnostic model 130 ′ Different results can be output according to different output types. In some embodiments, the network diagnostic model 130 can be updated based on training samples of various output types. Thus, the updated network diagnostic model 130 ′ It can flexibly switch between multiple output types to better meet the needs of various business scenarios.

[0062] In some embodiments, the training samples of various output types constructed may include single-step samples, which can be represented as follows:<prompt,func> `prompt` indicates a hint message. `Func` indicates the expected operation function (e.g., it can indicate the tool to be invoked next). For example, for a normal sample where the network is in a normal state, it can be represented as `func = normal`. For a faulty sample where the network is in a faulty state, `func` can indicate a tool from the tool list, which could be a diagnostic tool or a repair tool. In some embodiments, the network diagnostic model updated based on single-step samples can output the tool to be invoked next based on the current network state, and can directly invoke that tool to diagnose or repair the fault. In some embodiments, after the network diagnostic model updated based on single-step samples outputs the tool to be invoked next based on the current network state, it can invoke the tool after receiving confirmation from the user (e.g., an administrator). Then, the corresponding return results of the tool, as well as the current network state after executing the tool, are collected, and the next tool to be invoked is predicted, in this way until the fault is resolved. Thus, through a single-step execution method of "prediction => execution => feedback => prediction...", better decisions can be made dynamically based on the current network state, thereby improving the efficiency of resolving network faults.

[0063] In some embodiments, the training samples of various output types constructed may include multi-step samples, which can be represented as follows:<prompt,func_list> `prompt` indicates a hint message. The function list (`Func_list`) indicates a list of expected operation functions (e.g., it can indicate the tools to be called at each step). For example, a normal sample in a normal network state can be represented as `func_list = [normal]`. A faulty sample in a faulty network state can be represented as `func = [repair tool 1, ..., repair tool n, normal]`, consisting of a series of repair tools and normal states. In some embodiments, a network diagnostic model updated based on multi-step samples can output all steps for diagnosing or repairing faults according to the current network state. Therefore, by outputting all steps for diagnosing or repairing faults at once, the efficiency of resolving network faults can be improved. The following section describes in detail how to construct single-step or multi-step samples corresponding to multiple network states.

[0064] The following is for reference first. Figure 3This describes how to construct a single-step sample based on a set of network states 112. It should be understood that any single-step fault sample (e.g., single-step fault sample 312) among the multiple single-step fault samples included in the set of network states 112 can be constructed in the following manner. Similarly, any multi-step fault sample (e.g., multi-step fault sample 313) among the multiple multi-step fault samples included in the set of network states 112 can be constructed in the following manner. For ease of understanding, the following description will use single-step fault sample 312 or multi-step fault sample 313 as examples. However, this is merely exemplary and is not intended to limit the scope of the description.

[0065] In some embodiments, the model update system 150 determines a first network state from a set of network states. The first network state is a fault state of the simulated first network (also referred to as the "first simulated network") and is obtained by applying a single fault operation to the first network. The model update system 150 determines the first network state from a set of network states 112 (i.e., single-step fault sample 312). The single-step fault sample 312 can indicate that the first network is in a fault state after a single fault operation has been applied.

[0066] The model update system 150 can construct a first single-step sample for the first network state based on device information related to the first network state and a single fault operation, as one of multiple training samples. In other words, the model update system 150 can construct a single-step sample 317 for the first network state (i.e., single-step fault sample 312) based on the parameters and state information of the first network, the parameters and state information of the terminal device, and a single fault operation.

[0067] In some examples, the model update system 130 can construct a single-step sample 317 by modifying the output type in the prompt message corresponding to the current network state (e.g., single-step fault sample 312) to a first type (e.g., single-step - diagnosis). The model update system 130 can also construct a single-step sample 317 by modifying the output type in the prompt message corresponding to the current network state (e.g., single-step fault sample 312) to a second type (e.g., single-step - repair). In some embodiments, for fault samples collected from a real network, the first element in the function list (Func_list) can be used to construct a first single-step sample for a first network state.

[0068] In some embodiments, the model update system 150 determines a second network state from a set of network states. The second network state is a fault state of the simulated second network (also referred to as the "second simulated network") and is obtained by applying multiple fault operations to the second network. The model update system 150 determines the second network state (i.e., multi-step fault sample 313) from a set of network states 112. The multi-step fault sample 313 may indicate that the second network is in a fault state after multiple fault operations have been applied. For example, the multi-step fault sample 313 may be generated based on the application of a first fault operation, a second fault operation, and a third fault operation to the second network. In some embodiments, a first sub-network state may be generated via the second network to which the first fault operation has been applied. A second sub-network state may be generated via the second network to which the second fault operation has been applied. Correspondingly, a third sub-network state may be generated via the second network to which the third fault operation has been applied. However, these are merely exemplary and are not intended to limit the scope of this disclosure.

[0069] In some embodiments, a single-step sample can be constructed for any one of the multiple sub-network states (e.g., the first sub-network state) included in the second network state (i.e., the multi-step fault sample 313) in the following manner: The model update system 150 can determine the first sub-network state from the second network state. The model update system 150 can update the output type used to configure the network diagnostic model output in the device information related to the second network state based on the sample construction configuration information. In some embodiments, the sample construction configuration information can be pre-configured or pre-defined user input. Pre-defined user input can indicate historical user input collected in advance by the model update system 150. In some examples, the model update system 150 can update the output type used to configure the network diagnostic model output in the device information (i.e., the prompt information) related to the second network state based on whether the sample construction configuration information indicates whether the first sub-network state requires diagnostic tools.

[0070] Furthermore, the model update system 150 constructs a second single-step sample for the first sub-network state as one of multiple training samples based on the updated device information and the first fault operation. The model update system 150 constructs the second single-step sample (e.g., single-step sample 318) for the first sub-network state based on the updated prompt information. In some embodiments, if the model update system 150 determines that the sample construction configuration information indicates that the first sub-network state requires diagnosis, it can update the output type to a first type indicating diagnosis. In some embodiments, if the model update system 150 determines that the sample construction configuration information indicates that the first sub-network state requires diagnosis, it can update the output type in the device information related to the second network state used to configure the network diagnostic model output to the first type indicating diagnosis. That is, if the model update system 150 determines that the sample construction configuration information (e.g., predetermined user input) indicates that the first sub-network state requires diagnosis, it can update the output type in the prompt information to single-step-diagnosis. Subsequently, a single-step sample (e.g., single-step sample 318) for diagnosis corresponding to the first sub-network state is constructed. Accordingly, if the model update system 150 determines that the sample construction configuration information (e.g., pre-defined user input) indicates that the first sub-network state needs repair, it can update the output type in the prompt information to single-step-repair. Then, it constructs a single-step sample (e.g., single-step sample 318) for repair corresponding to the first sub-network state.

[0071] In some embodiments, if the model update system 150 determines that the sample construction configuration information does not indicate that the first sub-network state requires diagnosis, it can update the output type to a second type indicating repair. In some embodiments, if the model update system 150 determines that the sample construction configuration information does not indicate that the first sub-network state requires diagnosis, it can update the output type in the prompt message used to configure the network diagnostic model output to a first type indicating diagnosis. Subsequently, the model update system 150 can also update the output type in the information used to configure the network diagnostic model output to the second type. That is, if the model update system 150 determines that the sample construction configuration information (e.g., predetermined user input) does not indicate that the first sub-network state requires diagnosis, it can update the output type in the prompt message to single-step-diagnosis. Then, a single-step sample for diagnosis corresponding to the first sub-network state (e.g., single-step sample 318) is constructed. Following this, the model update system 150 can also update the output type in the prompt message to single-step-repair to construct a single-step sample for repair corresponding to the first sub-network state (e.g., single-step sample 318).

[0072] Understandably, if the sample construction configuration information does not indicate that the first sub-network state requires diagnosis, the network diagnostic model 130 can be updated to first diagnose the current network state and then repair it. Correspondingly, this method enables the updated network diagnostic model 130 to... ′ First, a diagnosis is performed based on the current network condition, and then the network is repaired. In some embodiments, when user input is received, the updated network diagnostic model 130... ′ The target prompt information shown in Table 2 can be used to determine whether the user input indicates that the current network status requires diagnostics. Table 2 is an example of target prompt information.

[0073] Table 2

[0074]

[0075] The following is for reference only. Figure 3 This describes how to construct multi-step samples based on a set of network states 112. It should be understood that any multi-step fault sample (e.g., multi-step fault sample 313) among the multiple multi-step fault samples included in the set of network states 112 can be constructed in the following manner.

[0076] In some embodiments, for a second network state (e.g., multi-step fault sample 313) in a set of network states 112, the model update system 150 can construct at least one multi-step sample for the second network state (e.g., multi-step fault sample 313) as part of a plurality of training samples, based on device information related to the second network state and multiple fault operations. In other words, the model update system 150 can construct multi-step samples (e.g., multi-step sample 319) for the second network state (i.e., single-step fault sample 312) based on the parameters and state information of the second network, the parameters and state information of the terminal device, and multiple fault operations.

[0077] In some examples, the model update system 130 can construct multi-step samples (e.g., multi-step sample 319) by modifying the output type in the prompt information corresponding to the current network state (e.g., multi-step fault sample 313) to multi-step. For example, for fault A formed by 5 fault operations: one 5-step multi-step sample, one 4-step multi-step sample, one 3-step multi-step sample, and so on, can be generated. See below for further details. Figure 3 This describes how to construct a sample of normal states based on a set of network states 112. It should be understood that any one of the multiple normal state samples included in the set of network states 112 (e.g., normal state sample 311) can be constructed in the following manner.

[0078] In some embodiments, for a third network state (e.g., normal state sample 311) in a set of network states 112 that indicates the network state is in a normal state, the model update system 150 may update the device information of the third network state related to the third network state based on the output type used to configure the network diagnostic model output. Further, the model update system 150 constructs samples for the third network state based on the updated device information related to the third network state, as part of a plurality of training samples.

[0079] In other words, the model update system 150 can construct samples of the normal state (e.g., normal state sample 316) for the third network state (i.e., normal state sample 311) based on the updated parameters and state information of the third network (also referred to as the "third simulation network") and the parameters and state information of the terminal device. In some examples, the model update system 130 can construct multi-step samples by modifying the output type in the prompt information corresponding to the current network state (e.g., normal state sample 311) to multi-step. The model update system 130 can construct single-step samples for diagnosis by modifying the output type in the prompt information corresponding to the current network state (e.g., normal state sample 311) to single-step-diagnosis. The model update system 130 can construct single-step samples for repair by modifying the output type in the prompt information corresponding to the current network state (e.g., normal state sample 311) to single-step-repair.

[0080] In summary, various types of training samples can be constructed based on some network states, thereby covering various types of training data to update network diagnostic models and reduce the cost of data collection.

[0081] The following text continues for reference. Figure 2 and Figure 3 This describes the fine-tuning and enhancement of the network diagnostic model.

[0082] Continuing process 200, in block 230, for each of the plurality of training samples, the model update system 150 generates a model diagnostic result for the corresponding network state based on device information related to the network state corresponding to that training sample, using a network diagnostic model. In embodiments of this disclosure, the network diagnostic model is configured to perform network state diagnosis. In some embodiments, the network diagnostic model can be configured to diagnose and repair network states. In some embodiments, for each of the plurality of training samples (e.g., normal state sample 316, single-step sample 317, single-step sample 318, or multi-step sample 319), the model update system 150 can generate a model diagnostic result 122 for the corresponding network state based on device information related to the network state corresponding to that training sample and using the network diagnostic model 130.

[0083] In box 240, the model update system 150 updates the network diagnostic model based on model diagnostic results generated for multiple training samples and diagnostic reference information corresponding to each of the multiple training samples. In some embodiments, the model update system 150 may update the network diagnostic model 130 based on the model diagnostic result 122 corresponding to each training sample. In some embodiments, the model update system 150 may also update the network diagnostic model 130 based on the diagnostic reference information corresponding to each training sample. The diagnostic reference information indicates the correct diagnostic result corresponding to the training sample. For training samples corresponding to a normal network state, the correct diagnostic result may be a judgment that the network state is normal. For training samples corresponding to a faulty network state, the correct diagnostic result may be an effective processing method for recovering the corresponding network fault. In some embodiments, the model update system 150 may fine-tune the network diagnostic model 130. The diagnostic reference information used for fine-tuning may include a sequence of truth operations for a faulty network state, which can restore the network from a faulty state to a normal state. In some embodiments, the model update system 150 may perform reinforcement learning on the network diagnostic model 130. The diagnostic reference information used for reinforcement learning may include feedback information on the diagnostic results generated by the network diagnostic model 130. Feedback information can indicate whether the diagnosis is correct.

[0084] The process of fine-tuning the network diagnostic model 130 is described below. It should be understood that the network diagnostic model 130 can be updated for any one of the multiple training samples (e.g., the first training sample) in the following manner. For ease of understanding, the following description uses the first training sample as an example. The first training sample corresponds to the fourth network state among multiple network states. In some embodiments, the fourth network state is obtained by applying a first fault sequence to the simulated fourth network (also referred to as the "fourth simulated network"), the first fault sequence including one or more fault operations. In some examples, the fourth network state is obtained after the fourth network is applied a first fault operation sequence including one or more fault operations. That is, the first training sample is constructed based on the fourth network after the first fault operation sequence is applied.

[0085] In some embodiments, the model update system 150 can determine a sequence of truth operations based on a first fault sequence. One or more truth operations in the truth operation sequence are recovery operations for one or more fault operations in the first fault operation sequence. That is, the model update system 150 can determine the sequence of truth operations corresponding to the first fault operation sequence based on a pre-stored sequence of first fault operations corresponding to the fourth network state. In some examples, the recovery operation for each fault operation can indicate the inverse operation of the fault operation. In some embodiments, the model update system 150 can repair the network state of the fourth network by performing the inverse operation of the fault operation.

[0086] Furthermore, the model update system 150 updates the network diagnostic model based on the differences between the first model diagnostic result generated for the fourth network state and one or more truth operations in the truth operation sequence. The model update system 150 can input the fourth network state into the network diagnostic model 130 to obtain a model diagnostic result for the fourth network. Subsequently, the model update system 150 can update the network diagnostic model 130 based on the differences between the model diagnostic result and the truth operation sequence corresponding to the fourth network. This improves the diagnostic and repair capabilities of the network diagnostic model.

[0087] The following is a reference. Figure 4 This disclosure describes a method for enhancing network diagnostic models. Figure 4A schematic diagram of an example architecture 400 for determining training samples for updating a network diagnostic model according to some embodiments of the present disclosure is shown. It should be understood that the network diagnostic model 130 can be updated for any one of a plurality of training samples (e.g., a second training sample). For ease of understanding, the following description uses a second training sample as an example. The second training sample corresponds to a fifth network state among a plurality of network states. In some embodiments, the fifth network state may indicate a state generated after at least one fault operation is applied to an initialized simulated network.

[0088] The fifth network state can also indicate the network state when the simulated network is in a normal state after initialization.

[0089] In some embodiments, the model update system 150 determines whether the second model diagnostic result is valid by executing the second model diagnostic result for the fifth network state using a network simulator. A valid model diagnostic result may mean that the diagnostic result can restore the network to a normal state. In some embodiments, the second model diagnostic result is generated based on the fifth network state via a network diagnostic model. In some examples, the model update system 150 may use a simulator to execute the model diagnostic result for the fifth network state to determine whether the model diagnostic result can diagnose and / or repair the fifth network state.

[0090] In some embodiments, the model update system 150 can generate feedback information indicating the validity of the second model's diagnostic result, serving as diagnostic reference information corresponding to the second training sample. That is, if the model update system 150 determines that the second model's diagnostic result is valid, the feedback information can be determined as positive feedback. If the model update system 150 determines that the second model's diagnostic result is invalid, the feedback information can be determined as negative feedback. Then, the model update system 150 can update the network diagnostic model based on the fifth network state, the second model's diagnostic result, and the feedback information.

[0091] In some examples, if the model update system 150 determines that the second model diagnostic result is valid, it can add the fifth network state and the model diagnostic result to the positive feedback sample set used for the network diagnostic model. Correspondingly, if the model update system 150 determines that the second model diagnostic result is invalid, it can add the fifth network state and the model diagnostic result to the negative feedback sample set used for the network diagnostic model. The model update system 150 then updates the network diagnostic model based on the positive and negative feedback sample sets.

[0092] In some embodiments, the model update system 150 can input the fifth network state multiple times into the network diagnostic model 130 to obtain multiple model diagnostic results corresponding to the fifth network state. Then, the model update system 150 uses the network simulator 170 to execute these model diagnostic results for the fifth network state, thereby determining which of these model diagnostic results are valid and which are invalid. Furthermore, the valid model diagnostic results and their corresponding fifth network states are added to the positive feedback sample set, and the invalid model diagnostic results and their corresponding fifth network states are added to the negative feedback sample set, thereby updating the network diagnostic model 130. In some embodiments, if at least two of the multiple model diagnostic results for the fifth network state are valid, a model diagnostic result with fewer than a step count threshold can be selected from the at least two valid model diagnostic results as one of the positive feedback sample sets.

[0093] like Figure 4 The example architecture 400 shown inputs the B network state (i.e., the fifth network state) multiple times into the network diagnostic model 130 to obtain multiple model diagnostic results (e.g., model diagnostic result 411, model diagnostic result 412, model diagnostic result 413, and model diagnostic result 414). Further, the model update system 150, using the network simulator 170, can determine that model diagnostic result 412 (i.e., the problem of the B network state can be resolved) includes invalid training samples 421, and thus can use model diagnostic result 412 and the B network state as one of the negative feedback sample sets. Similarly, the model update system 150, using the network simulator 170, can determine that model diagnostic result 414 (i.e., the problem of the B network state can be resolved) includes invalid training samples 422, and thus can use model diagnostic result 414 and the B network state as one of the negative feedback sample sets. The model update system 150, using the network simulator 170, can determine that model diagnostic results 411 and 413 are valid (i.e., they can resolve the B network state problem). Therefore, model diagnostic results 411 and the B network state, and model diagnostic results 413 and the B network state, can each be used as part of a positive feedback sample set. Additionally, since the operation steps for model diagnostic result 413 are shorter than those for model diagnostic result 411, model diagnostic result 413 can be used as one of the positive feedback sample sets compared to model diagnostic result 411.

[0094] In some embodiments, the model update system 150 determines that the second model diagnostic result is valid if it determines that the simulated network recovers to a normal state after the second model diagnostic result is executed. In some examples, the simulated network can be restored to a normal state by utilizing the second model diagnostic result. In some embodiments, the model update system 150 determines that the second model diagnostic result is valid if it determines that the second model diagnostic result conforms to a predetermined format. In some examples, the format of the second model diagnostic result for the network state output by the network diagnostic model 130 is correct. Additionally, the model update system 150 determines that the second model diagnostic result is valid if it determines that predetermined sub-model diagnostic results included in the second model diagnostic result indicate a normal state. In some examples, the second model diagnostic result is valid if the last sub-model diagnostic result included in the second model diagnostic result is a normal value. In other words, the second model diagnostic result is valid if the network state recovers to normal after the second model diagnostic result is executed.

[0095] In some embodiments, if the model update system 150 determines that the second model diagnostic result is invalid, it may do so if it determines that the simulated network has not recovered to a normal state after the second model diagnostic result has been executed. In some examples, the simulated network cannot be recovered to a normal state by utilizing the second model diagnostic result. In some embodiments, if the model update system 150 determines that the second model diagnostic result does not conform to a predetermined format, it may do so if it determines that the second model diagnostic result is invalid. In some examples, this is due to an incorrect format of the second model diagnostic result for the network state output by the network diagnostic model 130.

[0096] In some embodiments, the model update system 150 may determine that the second model diagnostic result is invalid if it determines that a predetermined sub-model diagnostic result included in the second model diagnostic result does not indicate a normal state. In some examples, the second model diagnostic result is invalid if the last sub-model diagnostic result included in the second model diagnostic result is not a normal value. In other words, the second model diagnostic result is invalid if the network state does not return to normal after the second model diagnostic result is executed. Alternatively, the model update system 150 may determine that the second model diagnostic result is invalid if it determines that the second model diagnostic result includes a sub-diagnostic result that is not output according to a predetermined output type. In some examples, it is assumed that the sub-network state corresponding to the sub-model diagnostic result needs to be diagnosed, but the sub-model diagnostic result is a repair; in this case, the second model diagnostic result is invalid. In some embodiments, the model update system 150 may use a single-step mode and / or a multi-step mode to collect feedback information based on the above-described manner.

[0097] In this way, by using a network simulator 170, the effectiveness of the model's diagnostic results regarding the network state is determined, thereby collecting feedback information. Based on this feedback, the capabilities of the network diagnostic model can be continuously strengthened, making the updated network model more intelligent.

[0098] In some embodiments, the network diagnostic model 130 updated in the manner described above ′ It can be applied to diagnose and repair faulty networks. In some embodiments, the network diagnostic model 130 updated via the scheme of this disclosure... ′ The system can flexibly switch between single-step and multi-step prediction based on user permissions and business needs. For example, if a user has network setting modification permissions (e.g., an administrator), then the network diagnostic model 130 can be used. ′ Perform single-step predictions, predicting the next tool to be executed each time, until the network recovers. In some embodiments, in network diagnostic model 130 ′ In scenarios configured to diagnose the network in a single-step mode, if the network diagnosis model is 130... ′ If the output repair tool fails to perform the repair, it can automatically switch to multi-step mode and output a suggested sequence of operations for the user's reference. In some embodiments, the sequence of operations can be presented to the user in the form of natural language, including but not limited to, prompts or rules.

[0099] In summary, network simulators can generate a large number of network states, such as various types of faulty and normal network states. Furthermore, based on these network states, various types of training samples can be constructed, thus covering diverse training data to update the network diagnostic model and reducing data collection costs. In this way, these training samples can be used to efficiently train a network diagnostic model with fault diagnosis and repair capabilities.

[0100] Embodiments of this disclosure also provide corresponding apparatus for implementing the above methods or processes.

[0101] Figure 5 A schematic structural block diagram of an apparatus 500 for network status diagnosis and repair according to certain embodiments of the present disclosure is shown. The apparatus 500 may be implemented as or included in a model update system 150. The various modules / components in the apparatus 500 may be implemented by hardware, software, firmware, or any combination thereof.

[0102] like Figure 5As shown, the device 500 includes a network state generation module 510, configured to generate a set of network states using a network simulator. The network simulator is configured to simulate multiple component devices of a network and the connections between these components, as well as to simulate different network states. The device 500 also includes a training sample construction module 520, configured to construct multiple training samples corresponding to the multiple network states based on the simulation of the set of network states. Each training sample includes at least device information related to the network state corresponding to that training sample, and the multiple network states are obtained based on the set of network states. The device 500 also includes a model diagnostic result generation module 530, configured to generate a model diagnostic result for each training sample based on the device information related to the corresponding network state, using a network diagnostic model. The network diagnostic model is configured to perform network state diagnosis. The device 500 also includes a network diagnostic model update module 540, configured to update the network diagnostic model based on the model diagnostic results generated for each of the multiple training samples and diagnostic reference information corresponding to each of the multiple training samples; wherein the diagnostic reference information indicates the correct diagnostic result corresponding to the training sample.

[0103] In some embodiments, the network state generation module 510 is further configured to construct multiple simulated networks using multiple component devices simulated by a network simulator, the multiple simulated networks having different network structures from each other; select a simulated network from the multiple simulated networks; obtain a fault operation sequence by sequentially applying at least one fault operation to the selected simulated network; and determine the network state of the selected simulated network after the fault operation sequence has been applied as one of a set of network states.

[0104] In some embodiments, the training sample construction module 520 is further configured to determine a first network state from a set of network states, the first network state being a fault state of the simulated first network and the first network state being obtained by applying a single fault operation to the first network; and to construct a first single-step sample for the first network state as one of a plurality of training samples based on device information of the first network related to the first network state and the single fault operation.

[0105] In some embodiments, the training sample construction module 520 is further configured to determine a second network state from a set of network states, the second network state being a fault state of the simulated second network and the second network state being obtained by applying multiple fault operations to the second network; and to construct at least one multi-step sample for the second network state as part of a plurality of training samples based on device information of the second network related to the second network state and the multiple fault operations.

[0106] In some embodiments, the training sample construction module 520 is further configured to determine a first sub-network state from a second network state, the first sub-network state being obtained by applying a first fault operation among multiple fault operations to the second network; update the output type used to configure the network diagnostic model output in device information related to the second network state based on sample construction configuration information; and construct a second single-step sample for the first sub-network state as one of multiple training samples based on the updated information and the first fault operation.

[0107] In some embodiments, the training sample construction module 520 is further configured to, in response to the sample construction configuration information indicating that the first sub-network state requires diagnosis, update the output type in the device information related to the second network state used to configure the output of the network diagnostic model to a first type indicating diagnosis, or in response to the sample construction configuration information not indicating that the first sub-network state requires diagnosis, update the output type in the information used to configure the output of the network diagnostic model to a second type indicating repair.

[0108] In some embodiments, the training sample construction module 520 is further configured to determine a third network state from a set of network states, the third network state being the normal state of the simulated third network; update device information of the third network state related to the third network state based on the output type used to configure the output of the network diagnostic model; and construct samples for the third network state as part of a plurality of training samples based on the updated device information of the third network state related to the third network state.

[0109] In some embodiments, updating the network diagnostic model includes fine-tuning the network diagnostic model, and a first training sample among multiple training samples corresponds to a fourth network state among multiple network states. The fourth network state is obtained by applying a first fault operation sequence to the simulated fourth network. The first fault operation sequence includes one or more fault operations, and the diagnostic reference information corresponding to the first training sample is obtained by: determining a truth operation sequence based on the first fault operation sequence, wherein one or more truth operations in the truth operation sequence are recovery operations for one or more fault operations in the first fault operation sequence; and the network diagnostic model update module 540 is further configured to update the network diagnostic model based on the difference between the first model diagnostic result generated by the network diagnostic model for the fourth network state and one or more truth operations in the truth operation sequence.

[0110] In some embodiments, updating the network diagnostic model includes reinforcement learning of the network diagnostic model, and a second training sample among multiple training samples corresponds to a fifth network state among multiple network states, and diagnostic reference information corresponding to the first training sample is obtained by: determining whether the second model diagnostic result is valid by performing a second model diagnostic result for the fifth network state using a network simulator, the second model diagnostic result being generated by the network diagnostic model based on the fifth network state; and generating feedback information indicating the validity of the second model diagnostic result as reference information corresponding to the second training sample, wherein the feedback information is determined to be positive feedback in response to the validity of the second model diagnostic result, and negative feedback in response to the invalidity of the second model diagnostic result; and the network diagnostic model update module 540 is further configured to update the network diagnostic model based on the fifth virtual network state, the second model diagnostic result, and the feedback information.

[0111] In some embodiments, the validity of the second model diagnostic result is determined by at least one of the following: the simulated network recovers to a normal state after the second model diagnostic result is executed; the second model diagnostic result conforms to a predetermined format; or the predetermined sub-model diagnostic result included in the second model diagnostic result indicates a normal state.

[0112] In some embodiments, the second model diagnostic result is determined to be invalid by at least one of the following: the simulated network is in an abnormal state after the second model diagnostic result is executed; the second model diagnostic result does not conform to a predetermined format; the predetermined sub-model diagnostic result included in the second model diagnostic result does not indicate a normal state; or the sub-diagnostic result included in the second model diagnostic result is not output according to a predetermined output type.

[0113] The units and / or modules included in device 500 can be implemented in various ways, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units and / or modules can be implemented using software and / or firmware, such as machine-executable instructions stored on a storage medium. In addition to or as an alternative to machine-executable instructions, some or all of the units and / or modules in device 500 can be implemented at least partially by one or more hardware logic components. By way of example and not limitation, exemplary types of hardware logic components that can be used include field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chips (SoCs), complex programmable logic devices (CPLDs), and so on.

[0114] It should be understood that one or more steps in the above methods can be performed by suitable electronic devices or combinations of electronic devices. Such electronic devices or combinations of electronic devices may include, for example, […]. Figure 1 The model update system 150 in the middle.

[0115] Figure 6 A block diagram of an electronic device 600 in which one or more embodiments of the present disclosure may be implemented is shown. It should be understood that... Figure 6 The electronic device 600 shown is merely exemplary and should not be construed as limiting the functionality and scope of the embodiments described herein. Figure 6 The electronic device 600 shown can be used to achieve Figure 1 The model update system 150.

[0116] like Figure 6 As shown, electronic device 600 is in the form of a general-purpose electronic device. Components of electronic device 600 may include, but are not limited to, one or more processors 610 or processing units, memory 620, storage device 630, one or more communication units 640, one or more input devices 650, and one or more output devices 660. Processor 6106 may be a physical or virtual processor and is capable of performing various processes according to programs stored in memory 620. In a multiprocessor system, multiple processors execute computer-executable instructions in parallel to improve the parallel processing capability of electronic device 600.

[0117] Electronic device 600 typically includes multiple computer storage media. Such media can be any available media accessible to electronic device 600, including but not limited to volatile and non-volatile media, removable and non-removable media. Memory 620 can be volatile memory (e.g., registers, cache, random access memory (RAM)), non-volatile memory (e.g., read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. Storage device 630 can be removable or non-removable media and can include machine-readable media, such as flash drives, disks, or any other media capable of storing information and / or data and accessible within electronic device 600.

[0118] Electronic device 600 may further include additional removable / non-removable, volatile / non-volatile storage media. Although not explicitly stated... Figure 6 As shown, disk drives for reading from or writing to removable, non-volatile disks (e.g., "floppy disks") and optical disk drives for reading from or writing to removable, non-volatile optical disks can be provided. In these cases, each drive can be connected to a bus (not shown) via one or more data media interfaces. Memory 620 may include computer program product 625 having one or more program modules configured to perform various methods or actions of various embodiments of this disclosure.

[0119] The communication unit 640 enables communication with other electronic devices via a communication medium. Additionally, the functionality of the components of the electronic device 600 can be implemented using a single computing cluster or multiple computing machines capable of communicating via communication connections. Therefore, the electronic device 600 can operate in a networked environment using logical connections to one or more other servers, network personal computers (PCs), or another network node.

[0120] Input device 650 can be one or more input devices, such as a mouse, keyboard, trackball, etc. Output device 660 can be one or more output devices, such as a monitor, speaker, printer, etc. Electronic device 600 can also communicate with one or more external devices (not shown) via communication unit 640 as needed. These external devices include storage devices, display devices, etc., and can communicate with one or more devices that enable user interaction with electronic device 600, or with any device that enables electronic device 600 to communicate with one or more other electronic devices (e.g., network card, modem, etc.). Such communication can be performed via input / output (I / O) interface (not shown).

[0121] According to an exemplary implementation of this disclosure, a computer-readable storage medium is provided that stores computer-executable instructions thereon, wherein the computer-executable instructions are executed by a processor to implement the methods described above. According to an exemplary implementation of this disclosure, a computer program product is also provided, which is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, which are executed by a processor to implement the methods described above.

[0122] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, devices, and computer program products implemented according to this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0123] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0124] Computer-readable program instructions can be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions that execute on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0125] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0126] Various implementations of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed implementations. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described implementations. The terminology used herein is chosen to best explain the principles, practical applications, or improvements to technology in the market, or to enable others skilled in the art to understand the various implementations disclosed herein.

Claims

1. A method for training a network diagnostic model, comprising: Use a network simulator to simulate a set of network states; The network simulator is configured to simulate multiple component devices of a network and the connections between the multiple component devices, as well as to simulate different network states; Based on the simulation of the set of network states, multiple training samples are constructed, each corresponding to a multiple network state. Each training sample includes at least device information related to the network state corresponding to the training sample. The multiple training samples include at least single-step samples and multi-step samples. The single-step samples are constructed based at least on a single fault operation applied to the corresponding network state. The multi-step samples are constructed based at least on multiple fault operations applied to the corresponding network state. The multiple network states are obtained based on the set of network states. For each of the plurality of training samples, based on device information related to the corresponding network state, a model diagnostic result for the corresponding network state is generated using a network diagnostic model, wherein the network diagnostic model is configured to perform network state diagnosis. as well as The network diagnostic model is updated based on the model diagnostic results generated for the multiple training samples and the diagnostic reference information corresponding to the multiple training samples. The diagnostic reference information indicates the correct diagnostic result corresponding to the training sample. The diagnostic reference information includes a truth operation sequence for the faulty network state, and the truth operation sequence indicates how to restore the faulty network state to a normal network state.

2. The method according to claim 1, wherein simulating the set of network states comprises: Using the network simulator to simulate the multiple component devices, multiple simulated networks are constructed, and the multiple simulated networks have different network structures from each other; Select a simulation network from the plurality of simulation networks; By sequentially applying at least one fault operation to the selected simulation network, a fault operation sequence is obtained; and The network state of the selected simulated network after the fault operation sequence is applied is determined as one of the set of network states.

3. The method according to claim 1, wherein constructing multiple training samples corresponding to multiple network states comprises: A first network state is determined from the set of network states. The first network state is the fault state of the simulated first network and is obtained by applying a single fault operation to the first network. as well as Based on the device information related to the state of the first network and the single fault operation, a first single-step sample for the state of the first network is constructed as one of the plurality of training samples.

4. The method according to claim 1, wherein constructing multiple training samples corresponding to multiple network states comprises: A second network state is determined from the set of network states, the second network state being a fault state of the simulated second network, and the second network state is obtained by applying multiple fault operations to the second network; as well as Based on the device information related to the second network state and the multiple fault operations of the second network, at least one multi-step sample for the second network state is constructed as part of the multiple training samples.

5. The method according to claim 4, wherein constructing multiple training samples corresponding to multiple network states further comprises: A first sub-network state is determined from the second network state, the first sub-network state being obtained by applying a first fault operation among the plurality of fault operations to the second network; Based on the sample-based configuration information, update the output type used to configure the network diagnostic model output in the device information related to the second network state; as well as Based on the updated information and the first fault operation, a second single-step sample for the first sub-network state is constructed as one of the plurality of training samples.

6. The method of claim 5, wherein updating the output type used to configure the network diagnostic model output in the device information related to the second network state includes: In response to the sample construction configuration information indicating that the first sub-network state requires diagnosis, the output type is updated to a first type indicating diagnosis, or In response to the fact that the sample construction configuration information does not indicate that the first sub-network state needs diagnosis, the output type is updated to a second type indicating repair.

7. The method of claim 1, wherein constructing a plurality of training samples corresponding to a plurality of network states comprises: A third network state is determined from the set of network states, wherein the third network state is the normal state of the simulated third network; Based on the output type used to configure the output of the network diagnostic model, update the device information of the third network related to the state of the third network; as well as Based on the updated device information related to the third network state, samples for the third network state are constructed as part of the plurality of training samples.

8. The method of claim 1, wherein updating the network diagnostic model includes fine-tuning the network diagnostic model, and a first training sample among the plurality of training samples corresponds to a fourth network state among the plurality of network states, the fourth network state being obtained by applying a first fault operation sequence to a simulated fourth network, the first fault operation sequence including one or more fault operations, and The diagnostic reference information corresponding to the first training sample was obtained in the following way: Based on the first fault operation sequence, a truth value operation sequence is determined, wherein one or more truth value operations in the truth value operation sequence are recovery operations for one or more fault operations in the first fault operation sequence; and Updating the network diagnostic model includes: The network diagnostic model is updated based on the difference between the first model diagnostic result generated by the network diagnostic model for the fourth network state and one or more truth operations in the truth operation sequence.

9. The method of claim 1, wherein updating the network diagnostic model includes reinforcement learning of the network diagnostic model, and the second training sample among the plurality of training samples corresponds to the fifth network state among the plurality of network states, and the diagnostic reference information corresponding to the second training sample is obtained by: The validity of the second model diagnostic result is determined by executing the second model diagnostic result for the fifth network state using the network simulator; the second model diagnostic result is generated by the network diagnostic model for the fifth network state. Feedback information indicating the validity of the second model's diagnostic result is generated as diagnostic reference information corresponding to the second training sample. The feedback information is determined to be positive feedback in response to a valid second model diagnostic result, and negative feedback in response to an invalid second model diagnostic result. Updating the network diagnostic model includes: The network diagnostic model is updated based on the fifth network state, the second model diagnostic results, and the feedback information.

10. The method of claim 9, wherein the validity of the second model diagnostic result is determined by at least one of the following: The simulated network recovered to a normal state after the second model's diagnostic results were applied. The diagnostic results of the second model conform to a predetermined format, or The predefined sub-diagnostic results included in the second model's diagnostic results indicate a normal state.

11. The method of claim 9, wherein the invalidity of the second model diagnostic result is determined by at least one of the following: The simulated network was in an abnormal state after the second model's diagnostic results were applied. The diagnostic results from the second model do not conform to the predetermined format. The predetermined sub-diagnostic results included in the second model's diagnostic results do not indicate a normal state, or The second model diagnostic results include sub-diagnostic results that are not output according to the predetermined output type.

12. An apparatus for network status diagnosis and repair, comprising: The network state simulation module is configured to use a network simulator to simulate a set of network states, wherein the network simulator is configured to simulate multiple component devices of the network and the connections between the multiple component devices, as well as simulate different network states; The training sample construction module is configured to construct multiple training samples corresponding to multiple network states based on the simulation of the set of network states. The training samples in the multiple training samples include at least device information related to the network state corresponding to the training sample. The multiple training samples include at least single-step samples and multi-step samples. The single-step samples are constructed based at least on a single fault operation applied to the corresponding network state. The multi-step samples are constructed based at least on multiple fault operations applied to the corresponding network state. The multiple network states are obtained based on the set of network states. The model diagnostic result generation module is configured to generate a model diagnostic result for each of the plurality of training samples based on device information related to the corresponding network state using a network diagnostic model, wherein the network diagnostic model is configured to perform network state diagnosis. as well as The network diagnostic model update module is configured to update the network diagnostic model based on the model diagnostic results generated for the plurality of training samples and the diagnostic reference information corresponding to the plurality of training samples respectively. The diagnostic reference information indicates the correct diagnostic result corresponding to the training sample. The diagnostic reference information includes a truth operation sequence for the faulty network state, and the truth operation sequence indicates how to restore the faulty network state to a normal network state.

13. An electronic device, comprising: At least one processor; as well as At least one memory coupled to the at least one processor and storing instructions for execution by the at least one processor, the instructions causing the electronic device to perform the method according to any one of claims 1 to 11 when executed by the at least one processor.

14. A computer-readable storage medium having stored thereon computer-executable instructions that can be executed by a processor to implement the method according to any one of claims 1 to 11.

15. A computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the method according to any one of claims 1 to 11.

Images