Network operation and maintenance method and related equipment

Abstract

The embodiment of the invention provides a network operation and maintenance method and related equipment, and relates to the field of AI. In the method, a query text, a network topology of a first network and network state data at multiple moments are obtained; the network state data includes operating state data of a plurality of nodes in the first network. The feature vector extraction process of network topology and nodes is guided based on the query text, and the representation capability of node feature vectors and topological feature vectors can be improved; and the response text is generated based on the query text, the node feature vector and the topological feature vector, so that the accurate response text can be provided. The response text can be used to assist the operation and maintenance personnel in network operation and maintenance, and the operation and maintenance efficiency is improved. In addition, the method does not depend on the experience of operation and maintenance personnel, and the application threshold is low.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of artificial intelligence, and in particular to a network operation and maintenance method and related equipment. Background Art

[0002] The telecommunications network is a system composed of interconnected communication devices (or nodes). Efficient network operation and maintenance is the prerequisite for business stability, operational security, and efficiency and cost reduction.

[0003] refer to Figure 1A , Figure 1A It is a schematic diagram of a network operation and maintenance model based on question and answer provided in an embodiment of the present application. The network operation and maintenance model based on question and answer mainly includes three stages: the first stage is that the operation and maintenance engineer uses auxiliary tools and combines his own operation and maintenance experience to give a preliminary abstraction of the problem based on the problem phenomenon and network status, that is, abstracts the problem into a general problem. The second stage is to obtain answers to general questions through an artificial intelligence (AI) question and answer system. The third stage is that the operation and maintenance engineer uses his own experience to correspond the answers to general questions back to specific cases. This type of method can improve the work efficiency of operation and maintenance engineers to a certain extent. However, this operation and maintenance method relies too much on the experience of operation and maintenance personnel, and the application threshold of the operation and maintenance method is high.

[0004] On the other hand, data-driven intelligent operation and maintenance has shown good results in various operation and maintenance tasks. Figure 1B , Figure 1B This is a schematic diagram of a single-task single-model operation and maintenance solution provided by an embodiment of the present application. For a specific task scenario, a specific scientific problem is abstracted, and the collected data (such as data 1, data 2, ..., data n) is used to train a model dedicated to the task scenario (such as model 1, model 2, ..., model n). However, this operation and maintenance method requires the customization of different models for different problems, and the versatility of the model is poor, resulting in a huge workload for the operation and maintenance personnel and low network operation and maintenance efficiency. Summary of the invention

[0005] The present application provides a network operation and maintenance method and related equipment, which has a low application threshold and can improve network operation and maintenance efficiency.

[0006] In a first aspect, a network operation and maintenance method is provided. The method is applied to a network operation and maintenance device. The network operation and maintenance method can be executed by the network operation and maintenance device itself or by a chip in the network operation and maintenance device.

[0007] The network operation and maintenance method comprises the following steps: obtaining a query text, a network topology of a first network, and network status data at multiple times. The query text is used to characterize the task of querying the first network, and the network status data includes the operation status data of multiple nodes in the first network. Based on the query text, feature extraction is performed on the network topology and the network status data at multiple times to obtain a node feature vector of each of the multiple nodes and a topology feature vector of the network topology. A response text is generated based on the query text, the node feature vector, and the topology feature vector, and the response text is used to characterize the query result.

[0008] The first network is a network interconnected by multiple devices, and the interconnection may be an actual physical connection or a logical connection. The multiple nodes in the first network may refer to all devices in the first network or to some devices in the first network. The query text, the network topology of the first network, and the network status data at multiple times may be obtained separately or simultaneously.

[0009] In this solution, the feature vector extraction process of network topology and nodes is guided by the query text, which can improve the representation ability of node feature vectors and topology feature vectors; and the response text is generated based on the query text, node feature vectors and topology feature vectors, which can provide accurate response text. The response text can assist operation and maintenance personnel in network operation and maintenance, and improve operation and maintenance efficiency. In addition, this solution does not depend on the experience of operation and maintenance personnel, and has a low application threshold.

[0010] In a possible implementation of the first aspect, the above-mentioned extracting features of the network topology and the network status data at multiple times based on the query text to obtain a node feature vector of each of the multiple nodes and a topology feature vector of the network topology specifically includes the following steps:

[0011] Feature extraction is performed on the query text to obtain a text feature vector. Feature extraction is performed on the network status data at multiple times to determine the initial node feature vector of each of the multiple nodes at multiple times. Based on the text feature vector and the network topology, the initial node feature vector of each node at multiple times is updated to obtain an updated initial node feature vector. Based on the text feature vector and the updated initial node feature vector of each node at multiple times, a node feature vector of each node is determined.

[0012] It can be seen that in this application, the initial node feature vector is updated based on the text feature vector and the network topology, which takes into account the influence of the query text itself and the influence of the interaction characteristics between nodes in the first network, and can improve the representation ability of the updated initial node feature vector, which helps to improve the representation ability of the node feature vector. In addition, the node feature vector of each node is determined based on the text feature vector and the updated initial node feature vector of each node at multiple times, which takes into account the influence of the query text itself and the influence of the running status data of the node at different times, further improving the representation ability of the node feature vector.

[0013] In a possible implementation of the first aspect, the updating of the initial node feature vector of each node at multiple times based on the text feature vector and the network topology specifically includes the following steps:

[0014] Determine the neighbor nodes of node A according to the network topology, where node A is any one of the above multiple nodes. According to the text feature vector and the initial node feature vector of the neighbor nodes of node A at time t, the initial node feature vector of node A at time t is Update; time t is any one of multiple time moments.

[0015] In this application, when updating the initial node feature vector, both the influence of the query text itself and the influence of the neighboring nodes of node A are taken into consideration, which can improve the representation ability of the updated initial node feature vector and help improve the representation ability of the node feature vector.

[0016] In a possible implementation of the first aspect, the above-mentioned feature extraction of the network topology and the network status data at multiple times based on the query text to obtain the node feature vector of each node in the multiple nodes and the topology feature vector of the network topology also includes the following steps: determining the topology feature vector based on the node feature vectors of the multiple nodes.

[0017] In the present application, node feature vectors of multiple nodes are used to obtain a topological feature vector, so that the topological feature vector can represent the characteristics of the first network as a whole.

[0018] In a possible implementation of the first aspect, the above-mentioned determination of the topological feature vector based on the node feature vectors of multiple nodes specifically includes the following steps: determining the average value of the node feature vectors of multiple nodes as the topological feature vector. Alternatively, determining the feature vector obtained by weighted aggregation of the node feature vectors of multiple nodes and the weights corresponding to the multiple nodes as the topological feature vector. Alternatively, selecting the top-k node feature vectors of multiple nodes according to their importance and concatenating them to obtain the topological feature vector, where k is an integer.

[0019] In a possible implementation of the first aspect, the generation of the response text based on the query text, the node feature vector and the topological feature vector specifically includes the following steps: when generating the i-th word of the response text, the first feature vector is updated according to the feature vector set to obtain an updated feature vector. The feature vector set includes node feature vectors of multiple nodes, topological feature vectors and multiple word segmentation feature vectors, and the multiple word segmentation feature vectors are feature vectors of word segmentations in the query text. The probability of each word in the vocabulary is determined according to the updated feature vector, the word with the highest probability is used as the i-th word in the response text, and the feature vector of the i-th word is added to the feature vector set, and the above steps are repeated until the i-th word is a terminator. Wherein, when i is greater than 1, the first feature vector is the feature vector of the i-1-th word in the response text. When i is 1, the first feature vector is the feature vector of the last word segmentation in the query text.

[0020] In this application, when generating a response text, the first feature vector is updated based on the node feature vector, the topology feature vector, and the feature vector of the existing text (such as the query text, the words of the generated response text), and then the i-th word in the response text is predicted based on the updated feature vector. It can be seen that when generating a response text, both the existing text is taken into consideration and the relevant information of the first network is fully utilized, so that the response text is more accurate.

[0021] In a possible implementation of the first aspect, the above-mentioned generating a response text based on the query text, the node feature vector and the topological feature vector specifically includes the following steps: determining M candidate texts based on the query text, the node feature vectors of multiple nodes and the topological feature vector, where M is greater than or equal to 2. A candidate text with the largest average probability among the M candidate texts is used as the response text, and the average probability of the candidate text is the average of the occurrence probabilities of all words in the candidate text.

[0022] In this application, when generating a response text, M candidate texts can be determined first, and then the candidate text with the largest average probability among the M candidate texts can be determined as the response text. The method of determining the response text by the average probability of the M candidate texts can effectively improve the accuracy of the response text.

[0023] In a possible implementation of the first aspect, the network operation and maintenance method further includes the following steps: outputting a response text.

[0024] In this solution, outputting the response text may refer to display output, voice output, or output to other modules / devices, so that the user can know the response text, and other modules / devices can further use the response text for other processing.

[0025] In a possible implementation of the first aspect, the network operation and maintenance method further includes the following steps: determining a control instruction according to the response text, where the control instruction is used to control the first node in the first network to perform the first operation. Sending the control instruction to the first node.

[0026] The first node is one or more nodes in the first network. A control instruction for controlling the first node is determined according to the response text, and after the first node performs the first operation, the problem existing in the first network indicated in the response text can be improved.

[0027] In a second aspect, the present application also provides a network operation and maintenance device, which includes an acquisition module, an extraction module and a generation module.

[0028] The acquisition module is used to acquire the query text, the network topology of the first network and the network status data at multiple times. The query text is used to represent the task of querying the first network, and the network status data includes the operation status data of multiple nodes in the first network.

[0029] The extraction module is used to extract features of the network topology and network status data at multiple times based on the query text, and obtain a node feature vector of each of the multiple nodes and a topology feature vector of the network topology.

[0030] The above-mentioned generation module is used to generate a response text based on the query text, the node feature vector and the topology feature vector, and the response text is used to represent the query result.

[0031] In this solution, the network operation and maintenance device guides the feature vector extraction process of the network topology and nodes based on the query text, which can improve the representation ability of the node feature vector and the topology feature vector; and then generates a response text based on the query text, the node feature vector and the topology feature vector, which can provide an accurate response text. The response text can assist the operation and maintenance personnel in network operation and maintenance and improve the operation and maintenance efficiency. In addition, the use of the network operation and maintenance device does not depend on the experience of the operation and maintenance personnel, and the threshold for use is low.

[0032] In a possible implementation of the second aspect, the network operation and maintenance device further includes an output module. The output module is configured to output a response text.

[0033] In a possible implementation of the second aspect, the network operation and maintenance device further includes a determination module and a sending module.

[0034] The determination module is used to determine a control instruction according to the response text, where the control instruction is used to control the first node in the first network to perform a first operation.

[0035] The sending module is used to send a control instruction to the first node.

[0036] In a third aspect, the present application also provides a network operation and maintenance device, including a processor and a memory, wherein the processor and the memory are connected, wherein the memory is used to store program code, and the processor is used to call the program code to execute the network operation and maintenance method as described in the first aspect.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the network operation and maintenance method as described in the first aspect.

[0038] In a fifth aspect, the present application also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the network operation and maintenance method described in the first aspect.

[0039] In a sixth aspect, the present application further provides a chip, comprising a processor and a data interface, wherein the processor reads instructions stored in a memory through the data interface to execute the network operation and maintenance method described in the first aspect.

[0040] Optionally, as an implementation method, the chip may also include a memory, in which instructions are stored, and the processor is used to execute the instructions stored in the memory. When the instructions are executed, the processor is used to execute the network operation and maintenance method described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The following is an introduction to the drawings used in the embodiments of the present application.

[0042] Figure 1A This is a schematic diagram of a network operation and maintenance mode based on question and answer provided in an embodiment of the present application;

[0043] Figure 1B This is a schematic diagram of an operation and maintenance solution for a single task and a single model provided in an embodiment of the present application;

[0044] Figure 2 is a schematic diagram of the attention layer provided in an embodiment of the present application;

[0045] Figure 3 It is a system architecture provided by an embodiment of the present application;

[0046] Figure 4A This is a schematic diagram of an application scenario provided by an embodiment of the present application;

[0047] Figure 4B is a schematic diagram of a response text provided in an embodiment of the present application;

[0048] Figure 4C This is another application scenario schematic diagram provided by an embodiment of the present application;

[0049] Figure 4D is a network diagram of a specific application scenario provided by an embodiment of the present application;

[0050] Figure 4E This is a schematic diagram of locating a root cause failure node provided by an embodiment of the present application;

[0051] Figure 5 It is a flowchart of a network operation and maintenance method provided in an embodiment of the present application;

[0052] Figure 6 is a schematic diagram of a multimodal model provided in an embodiment of the present application;

[0053] Figure 7 It is a schematic diagram of a model architecture provided in an embodiment of the present application;

[0054] Figure 8 It is a structural diagram of a network operation and maintenance device provided by an embodiment of the present invention;

[0055] Fig. 9 It is a structural diagram of a network operation and maintenance device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The technical solution in this application will be described below in conjunction with the accompanying drawings.

[0057] Due to the applications involved in the embodiments of the present application, in order to facilitate understanding, the relevant terms and other related concepts involved in the embodiments of the present application are first introduced below.

[0058] In the embodiments of the present application, the words "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0059] The "at least one" mentioned in the embodiments of the present application refers to one or more, and "multiple" refers to two or more. "At least one of the following" or its similar expression refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, (a and b), (a and c), (b and c), or (a and b and c), where a, b, c can be single or multiple. "And / or" describes the association relationship of the associated objects, indicating that three relationships can exist. For example, A and / or B can be represented by: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are an "or" relationship. The serial numbers of the steps of the embodiments of the present application (such as step S1, step S21, etc.) are only for distinguishing different steps, and do not limit the order of execution between the steps.

[0060] Furthermore, unless otherwise specified, the ordinal numbers such as "first" and "second" used in the embodiments of the present application are used to distinguish multiple objects, and are not used to limit the order, timing, priority or importance of multiple objects. For example, the first device and the second device are only for the convenience of description, and do not represent the difference in structure, importance, etc. between the first device and the second device. In some embodiments, the first device and the second device can also be the same device.

[0061] In the above embodiments, the term "when..." can be interpreted as meaning "if..." or "after..." or "in response to determining..." or "in response to detecting...", depending on the context. The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modification, equivalent substitution, improvement, etc. made within the concept and principle of the present application shall be included in the protection scope of the present application.

[0062] (1) Artificial Intelligence

[0063] Artificial intelligence is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is also the study of the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making. Research in the field of artificial intelligence includes robotics, natural language processing, computer vision, decision-making and reasoning, human-computer interaction, recommendation and search, basic AI theory, etc.

[0064] (2) Attention

[0065] The attention operation extracts the weights used to describe the importance of features from the large amount of information contained in the data, and combines the weights with the features to make the network focus on the information with greater weights. The network layer that performs the attention operation is the attention layer. Specifically, refer to Figure 2 , Figure 2 is a schematic diagram of the attention layer; the attention layer includes the first query feature transformation module (i.e. Figure 2 Q module in), the first key feature transformation module (i.e. Figure 2 K module in the first value feature transformation module (i.e. Figure 2 V module in the attention layer), the input x Q 、x K 、x V After being transformed by the Q module, K module, and V module, it becomes Q out , K out , V out According to K out and Q out Generate relevant weights, and then adjust V according to the weight coefficients out Perform weighted summation and finally output the weighted feature out. Among them, the self-attention operation (Self-Attention) is a special attention operation, that is, when the attention layer x Q =x K =x V =x, the attention layer at this time is the self-attention layer.

[0066] refer to Figure 2 The following is an introduction to its basic process. The input of the self-attention layer is x. First, three feature transformations are performed on x to obtain Q out , K out and V out :

[0067] Q out =xW Q , K out =xW K , V out =xWV (1)

[0068] In formula (1), x is the input of the self-attention layer, W Q , W K , W v is the feature transformation matrix, W Q , W K , W v In convolutional neural networks, these are fully connected layers, and their weights are learnable parameters.Q , W K , W v Multiply them and we get Q out , K out and V out The purpose is to map the x feature to another dimension so that the mapped feature can satisfy the requirements of Q in the Attention operation. out , K out and V out The requirements of W Q , W K , W v The function will gradually manifest itself as the deep neural network training process proceeds. K After transformation, we get K out It can describe the content characteristics of the input and represent what the input characteristics are like; after W Q Q after feature transformation out Can contain guiding features about the input, representing what information the model needs to process; and after W v V after feature transformation out is a vector representing the input features.

[0069] Then Q out With K out Perform matrix multiplication (MatMul), and then use the numerical scaling (Scale) layer to perform numerical scaling processing; optionally, in the field of text processing, in the attention operation, the mask (Mask) layer is also used to process the numerically scaled data; finally, after the Softmax normalization operation, the Attention Map describing the autocorrelation between a certain dimension of x is obtained, that is, the Attention Map is:

[0070]

[0071] In formula (2), d is a constant, which is a custom value and is set according to different needs of the model. The matrix multiplication operation is equivalent to finding the guiding features in the content features.

[0072] Finally, use Attention Map to out Weighted, V out It is matrix multiplied with the Attention Map and then transformed as the output of the attention layer. out is the feature transformation matrix.

[0073] out=AV out W out (3)

[0074] Based on the above introduction to the Attention operation, the Attention operation is recorded as:

[0075] Attention(x Q , x K , x V )

[0076] The Self-Attention (SA) operation is based on the Attention operation. The inputs of the Q module, K module, and V module are set to the same input x. The Self-Attention operation is recorded as:

[0077] SA(x)=Attention(x,x,x)

[0078] (3) Network topology

[0079] Network topology refers to the physical layout of various devices interconnected by transmission media. It refers to the specific physical (real) or logical (virtual) arrangement of members that constitute the network. A network topology includes nodes corresponding to various devices in the network and edges between nodes. Edges are used to represent the physical or logical connection relationship between nodes.

[0080] (4) Telecommunications network

[0081] A telecommunications network is a system consisting of interconnected communication devices (or nodes or network elements), such as mobile communication networks and the Internet of Things (IoT).

[0082] Exemplarily, when the telecommunication network is a mobile communication network, the communication equipment in the mobile communication network includes a mobility management entity function (Mobility Management Entity function, MME), a serving gateway (ServingGateWay, SGW), a router and a base station.

[0083] The Internet of Things refers to the use of various devices and technologies such as information sensors, radio frequency identification technology, global positioning systems, infrared sensors, laser scanners, etc. to collect real-time information on any object or process that needs to be monitored, connected, and interacted with, and to collect various required information such as sound, light, heat, electricity, mechanics, chemistry, biology, and location. Through various possible network access, it realizes ubiquitous connection between objects and objects, and between objects and people, and realizes intelligent perception, identification, and management of objects and processes.

[0084] (5) Large Language Model (LLM)

[0085] The Big Language Model is a model based on machine learning and natural language processing technology. It learns to serve the ability of human language understanding and generation by training a large amount of text data. The Big Language Model can handle a variety of natural language tasks, such as text classification, question answering, and dialogue, and is an important path to artificial intelligence.

[0086] (6) Bidirectional Encoder Representation from Transformers (BERT) model

[0087] The BERT model is a multi-layer bidirectional transformer encoder based on fine-tuning. The BERT model aims to pre-train deep bidirectional representations by jointly adjusting the context in all layers. Specifically, the BERT model can extract semantic information from the text through the language representation learned by the pre-trained model. The BERT model uses a transformer model, which can effectively capture the contextual information in the text through a multi-layer self-attention mechanism, thereby obtaining a more accurate language representation. This language representation can be applied to various natural language processing tasks, such as text classification, named entity recognition, sentiment analysis, etc.

[0088] (7) Beam Search Algorithm

[0089] The beam search algorithm has a hyperparameter: beam size, which is set to M. In the first time step, the M words with the highest conditional probability are selected as the first word of the candidate output sequence. In each subsequent time step, based on the output sequence of the previous step, the M combinations with the highest conditional probability are selected as the candidate output sequence for that time step. M candidate output sequences are always maintained. Finally, the best sequence (such as the one with the highest average probability) is selected from the M candidate output sequences as the final output sequence.

[0090] refer to Figure 1A and Figure 1B , the network operation and maintenance method has the problems of over-reliance on the operation and maintenance experience of the operation and maintenance engineers, high application threshold of the method, and low operation and maintenance efficiency. To this end, the present application proposes a network operation and maintenance method, which guides the feature vector extraction process of the network topology and nodes of the first network based on the query text, and can improve the characterization ability of the node feature vector and the topology feature vector of the first network; and then generates a response text based on the query text, the node feature vector and the topology feature vector, and can provide an accurate response text. The method has a low application threshold and can improve the network operation and maintenance efficiency of the first network. Simply put, the embodiment of the present application can provide corresponding response texts for different query texts, realize intelligent question and answer, and assist operation and maintenance personnel to quickly perform network operation and maintenance.

[0091] The first network is a network composed of interconnected devices (or apparatuses or nodes), and the interconnection may be an actual physical connection or a logical connection, such as a telecommunications network or other network from which the network topology can be abstracted.

[0092] The above-mentioned network operation and maintenance method can be applied to a network operation and maintenance device, and the network operation and maintenance method can be executed by the network operation and maintenance device itself, or by a chip in the network operation and maintenance device.

[0093] The above-mentioned network operation and maintenance device can be a server or a control device in the first network, etc. The above-mentioned control device refers to a device in the first network that has information processing and communication capabilities, and can be any node in the first network.

[0094] The above network operation and maintenance methods can also be applied to software services such as cloud services or applications.

[0095] The following introduces the system architecture provided by the embodiments of the present application.

[0096] See attached Figure 3 , an embodiment of the present invention provides a system architecture 300. As shown in the system architecture 300, a data acquisition device 360 ​​is used to collect training data, and the training data includes the network topology and network status data of the network, and the training data varies with different networks. The training data is stored in a database 330, and the training device 320 can train a first model 301 based on the training data maintained in the database 330, and the first model 301 is used to implement the network operation and maintenance method described in the embodiment of the present application. Among them, when the training data includes the network topology and network status data of a network, the trained first model 301 serves the network operation and maintenance questions and answers of this network. When the training data includes the network topology and network status data of multiple (greater than or equal to two) networks, the trained first model 301 can serve the network operation and maintenance questions and answers of multiple networks.

[0097] The training device 320 can be a terminal, such as a mobile phone terminal, a tablet computer, a laptop computer, AR / VR, a vehicle-mounted terminal, a monitoring device, a vehicle-mounted automatic driving system, an unmanned vending machine, etc., and can also be a server or the cloud, etc. It should be noted that in actual applications, the training data maintained in the database 330 does not necessarily all come from the data acquisition device 360, but may also be received from other devices, such as cameras, etc. It should also be noted that the training device 320 does not necessarily perform model training entirely based on the training data maintained by the database 330, but may also obtain training data from the cloud or other places for model training. The above description should not be used as a limitation on the embodiments of the present application.

[0098] The first model 301 trained by the training device 320 can be applied to different systems or devices, such as Figure 3 The terminal device 310 shown in the figure may be a terminal such as a mobile phone terminal, a tablet computer, a laptop computer, AR / VR, a vehicle terminal, a monitoring device, a vehicle automatic driving system, an unmanned vending machine, etc. It may also be a server or a cloud. Figure 3 In the embodiment, the terminal device 310 is configured with an I / O interface 312 for data interaction with an external device. The user can input data to the I / O interface 312 through the client device 340. The input data includes query text, the network topology of the first network, and network status data at multiple times in the embodiment of the application.

[0099] The preprocessing module 313 is used to preprocess the input data (such as network status data) received by the I / O interface 312, and the preprocessed data enters the calculation module 311. In the embodiment of the present application, the input data takes the network status data as an example. When the network status data includes a picture, the preprocessing module 313 can be used to perform at least one of filtering, enhancement, and denoising on the picture to obtain a picture that meets the requirements.

[0100] When the terminal device 310 preprocesses the input data, or when the computing module 311 of the terminal device 310 performs calculations and other related processing, the terminal device 310 can call the data, code, etc. in the data storage system 350 for corresponding processing, and can also store the data, instructions, etc. obtained from the corresponding processing in the data storage system 350.

[0101] Finally, the I / O interface 312 returns the model processing result (such as response text) of the input data to the client device 340 to provide it to the user. At this time, the client device 340 can be a display.

[0102] exist Figure 3 In the case shown, the user can manually input data, which can be operated through the interface provided by the I / O interface 312. In another case, the client device 340 can automatically send input data to the I / O interface 312. If the client device 340 needs to obtain the user's authorization to automatically send the input data, the user can set the corresponding permissions in the client device 340. The user can view the results output by the terminal device 310 on the client device 340, and the specific presentation form can be display, sound, action, etc. The client device 340 can also be used as a data collection terminal to collect data such as Figure 3The input data of the input I / O interface 312 and the output result of the output I / O interface 312 are taken as new sample data and stored in the database 330. Of course, it is also possible to collect data without going through the client device 340, but directly by the I / O interface 312. Figure 3 The input data of the input I / O interface 312 and the output result of the output I / O interface 312 are stored in the database 330 as new sample data.

[0103] It is worth noting that Figure 3 This is only a schematic diagram of a system architecture provided by an embodiment of the present invention. The positional relationship between the devices, components, modules, etc. shown in the figure does not constitute any limitation. Figure 3 In the embodiment, the data storage system 350 is an external memory relative to the terminal device 310. In other cases, the data storage system 350 may also be placed in the terminal device 310. In addition, the training device 320 and the terminal device 310 may be the same device.

[0104] like Figure 3 As shown, the first model 301 is obtained by training the training device 320. In the embodiment of the present application, the first module 301 may include: a feature extractor and a predictor, wherein the feature extractor is used to extract feature vectors of query text, network topology, and network status data, and the predictor is used to process the feature vectors extracted by the feature extractor to obtain corresponding processing results. In the embodiment of the present application, an attention mechanism can be applied in the feature extractor to extract features that the model task pays more attention to.

[0105] In order to more clearly introduce the network operation and maintenance method of the embodiment of the present application, the following takes the execution subject as an example of a server to introduce possible application scenarios of the embodiment of the present application:

[0106] The first application scenario, refer to Figure 4A , Figure 4A is a schematic diagram of an application scenario provided by an embodiment of the present application; in this embodiment, the server can obtain the network topology of the first network a and the network status data of the first network a, the network topology of the first network b and the network status data of the first network b in real time, user A is the operation and maintenance personnel of the first network a, and user B is the operation and maintenance personnel of the first network b. The terminal (such as the first terminal and the second terminal described below) can remotely access the server to realize intelligent operation and maintenance Q&A.

[0107] During the operation and maintenance process, user B edits a query text on the first terminal (such as a computer) and sends the query text to the server through the network. For example, the query text is "node status query", "change risk estimation" and "fault root cause node inference". The server processes the received query text, the network topology of the first network b and the network status data of the first network b, and generates a corresponding response text. The server sends the response text to the first terminal through the network. The response text is as follows: Figure 4B As shown, user B can obtain relevant operation and maintenance information of the first network b according to the response text displayed on the first terminal. In daily operation and maintenance work, the method of the embodiment of the present application can be used to easily obtain the relevant status of the first network b; and when a failure occurs in the first network b, based on the response text, the operation and maintenance personnel can quickly carry out the operation and maintenance work, effectively improving the operation and maintenance efficiency.

[0108] Similarly, during the operation and maintenance process, user A can edit the query text on the second terminal (such as a mobile phone) and send the query text to the server through the network. After generating the response text, the server sends the response text to the second terminal through the network so that user A can obtain relevant information of the response text.

[0109] The second application scenario, refer to Figure 4C , Figure 4C This is another application scenario diagram provided by an embodiment of the present application; in this embodiment, the server includes a host and a display, and the host and the display are connected. Similarly, the server can obtain the network topology of the first network and the network status data of the first network in real time, and the user can directly access the server on the display, that is, enter the query text on the display, and the host processes the query text, the network topology of the first network, and the network status data of the first network to generate a response text. The response text can be directly displayed on the display to be output to the user to know, so as to realize local intelligent operation and maintenance Q&A.

[0110] In the above two application scenarios, the server can obtain the network status data of the first network in real time. When generating the response text, the server can obtain the network status data of the first network at multiple times for processing according to actual needs.

[0111] The following is an example of a real application scenario: locating the root cause of a fault node. Figure 4D , Figure 4D is a network diagram of a specific application scenario provided by an embodiment of the present application. Figure 4D A network topology of a first network is shown. The first network includes a core network, a base station, and a transmission node. The transmission node is such as an optical fiber or a transmission device. Figure 4DThe legend shows the devices represented by symbols of different shapes, and the letters above the symbols further show the manufacturer of the equipment.

[0112] When the first network fails, the operation and maintenance personnel of the first network can input the network topology corresponding to the first network, the logs reflecting the operating status of each node in the first network before and after the failure, KPI and other information, and the query text into the server (taking the second application scenario mentioned above as an example), and then obtain the corresponding response text. For example, the query text is "the root cause of the failure node is inferred, and the scope of impact caused by the failure is determined", that is, the topological boundary and time boundary affected by the failure are confirmed. The corresponding response text can be "the root cause of the failure node is the reset of the first device and the second device, resulting in abnormal status of the upstream and downstream nodes.".

[0113] For example, in the embodiment of the present application, in addition to outputting the response text, a schematic diagram related to the response text may also be shown, such as a schematic diagram of the location of the root cause failure node, or a schematic diagram of the impact range of the root cause failure node. Figure 4E , Figure 4E It is a schematic diagram of locating a root cause fault node provided in an embodiment of the present application; wherein nodes with a dark background are root cause fault nodes, and light-colored areas mark devices that are affected by the relevant root cause fault nodes and generate relevant alarms, that is, the influence range of the root cause fault nodes.

[0114] The embodiment of the present application proposes a new paradigm of end-to-end question-answer network operation and maintenance that integrates network spatiotemporal information. The dynamic changes of network topology and node status can be directly input into the model, and corresponding response texts can be given according to different query texts. Among them, the network spatiotemporal information includes network status data at multiple moments in the time dimension and the network topology of the first network in the space dimension.

[0115] The network operation and maintenance method provided in the embodiment of the present application is specifically described below:

[0116] Take the execution subject as a network operation and maintenance device as an example.

[0117] refer to Figure 5 , Figure 5 5 is a flow chart of a network operation and maintenance method provided in an embodiment of the present application. The network operation and maintenance method 500 includes the following steps:

[0118] 501. A network operation and maintenance device obtains a query text, a network topology of a first network, and network status data at multiple times.

[0119] The query text is used to represent the task of querying the first network, and the query task may include at least one of the following: network status query, fault demarcation and location, fault self-healing strategy recommendation, or change risk assessment. Change risk assessment refers to the risk brought about by changes in network-related parameters, such as the risk brought about by changes in network channels, or the risk brought about by restarting a node in the network. The query text may exist in the form of natural language.

[0120] The network status data includes the operation status data of multiple nodes in the first network; the multiple nodes may refer to all devices in the first network or to some devices in the first network. The query text, the network topology of the first network, and the network status data at multiple times may be obtained separately, successively, or simultaneously. For example, when the network operation and maintenance device obtains the query text, it simultaneously obtains the network topology of the first network and the network status data at multiple times.

[0121] 502. The network operation and maintenance device extracts features of the network topology and network status data at multiple times based on the query text to obtain a node feature vector of each of the multiple nodes and a topology feature vector of the network topology.

[0122] Specifically, the node feature vector is a feature representation of the operation status data of the node, and the topology feature vector is a feature representation of the network topology of the first network.

[0123] 503. The network operation and maintenance device generates a response text based on the query text, the node feature vector, and the topology feature vector.

[0124] The above response text is used to represent the query result, that is, the query result corresponding to the query text. The response text can exist in the form of natural language.

[0125] In the embodiment of the present application, the feature vector extraction process of the network topology and nodes is guided based on the query text, which can improve the representation ability of the extracted node feature vector and topology feature vector; and then the response text is generated based on the query text, the node feature vector and the topology feature vector, which can provide an accurate response text. The response text can assist the operation and maintenance personnel in network operation and maintenance and improve the operation and maintenance efficiency. In addition, this solution does not depend on the experience of the operation and maintenance personnel, and the application threshold is low.

[0126] In one possible implementation, reference Figure 5 The above network operation and maintenance method 500 further includes the following steps:

[0127] 504. The network operation and maintenance device outputs a response text.

[0128] In an embodiment of the present application, the above-mentioned output response text may refer to output means such as display output, voice output or output to other modules / devices, so that the user can obtain the response text, and other modules / devices can further use the response text for other processing. For example, other modules / devices determine the control instructions of related devices based on the response text, and send the control instructions to the related devices. The related devices perform related operations according to the control instructions. The above-mentioned control instructions can improve the problems existing in the first network pointed out in the response text.

[0129] Exemplarily, the network operation and maintenance method 500 may further include outputting a related schematic diagram of the response text, such as a schematic diagram of a root cause failure node, a schematic diagram of the scope of influence of the root cause failure node, etc. When the response text indicates the root cause failure node, the schematic diagram of the output root cause failure node may be displayed. When the response text indicates the scope of influence of the root cause failure node, the schematic diagram of the scope of influence of the output root cause failure node may be displayed. When the response text indicates the state parameter of the node, the node operation state data table of the node may be displayed.

[0130] In a possible implementation manner, the network operation and maintenance method 500 further includes the following steps:

[0131] 505. The network operation and maintenance device determines a control instruction according to the response text, where the control instruction is used to control a first node in the first network to perform a first operation. The first node is one or more nodes in the first network.

[0132] Specifically, when determining the control instruction based on the response text, it can be determined by means such as an instruction mapping table or AI model prediction, and there is no limitation on the method of determining the control instruction based on the response text. The instruction mapping table stores the correspondence between the control instruction and the text. For example, if the text is that the device is powered off, then the control instruction is to control the device to power on. If the text is that the temperature of the device is too high, then the control instruction is to control the device to power off for a certain period of time (the specific time can be set according to the actual situation). By inputting the response text into the corresponding AI model, the corresponding control instruction can be output.

[0133] 506. The network operation and maintenance device sends a control instruction to the first node.

[0134] Specifically, a control instruction for controlling the first node is determined based on the response text, and after the first node performs the first operation, the problem of the first network indicated in the response text can be improved. When the network operation and maintenance device is any node in the first network, the response text can be mapped to a specific operation and maintenance action, that is, the network operation and maintenance device can control the first node in the first network to perform relevant operation and maintenance actions after the response text, so as to realize the automatic driving of the first network.

[0135] For example, the response text indicates that the switch in the first network is powered off, and the control instruction is to control the switch to power on. The network operation and maintenance device sends the control instruction to the node that controls power on, and the node responds to the control instruction to power on the switch again.

[0136] In the above step 501, the operation status data of the node may include the key performance indicator (KPI) of the node, the log of the node, the alarm information of the node, the description document of the first network, the received instructions of the node, etc., which are used to describe the operation status of the entire node. Further, as an example, if the node is a base station, the KPI of the base station may include the throughput of the base station, the wireless call drop rate of the base station, the number of physical resource blocks used by the base station, etc. The log of the node may be the transaction log, operation log or other types of logs of the node, etc., which are not exhaustive here.

[0137] The data form of the node operation status data is numerical value, text, etc. Taking the network operation status of node i in the telecommunication network at time t as an example, the KPI, alarm and log data generated in the time interval (t-τ, t] can be used. express, The data in has numerical and textual categories, which is a multimodal organization form. τ is the time interval between different moments. It is assumed that the multiple moments in step 501 are K moments. The network topology of the first network describes the connection relationship between nodes in the first network, which can be represented by the adjacency matrix A. A ij =1, indicating that there is a connection between node i and node j.

[0138] In a possible implementation, the above step 502 specifically includes the following steps:

[0139] 521. The network operation and maintenance device extracts features from the query text to obtain a text feature vector. Feature extraction is performed on network status data at multiple times to determine an initial node feature vector of each of the multiple nodes at multiple times.

[0140] Specifically, assuming that the plurality of time moments is K time moments, and the plurality of nodes in the first network is N nodes, and K and N are positive integers, then in step 521, K*N initial node feature vectors may be obtained.

[0141] 522. The network operation and maintenance device updates the initial node feature vector of each node at multiple times based on the text feature vector and the network topology to obtain an updated initial node feature vector.

[0142] The initial node feature vector is updated based on the text feature vector and the network topology, which takes into account both the influence of the query text itself and the influence of the interaction characteristics between nodes in the first network. This can improve the representation ability of the updated initial node feature vector and help improve the representation ability of the node feature vector.

[0143] 523. The network operation and maintenance device determines a node feature vector of each node based on the text feature vector and the updated initial node feature vectors of each node at multiple times.

[0144] The node feature vector of each node is determined based on the text feature vector and the updated initial node feature vectors of each node at multiple times. This takes into account both the influence of the query text itself and the influence of the node's operating status data at different times, which can further enhance the representation capability of the node feature vector.

[0145] Exemplarily, in the above step 521, the feature extraction model is used to obtain the initial node feature vectors of the node at different times. The above feature extraction model can be a multimodal model, which is a model that supports multimodal data input, analyzes each modal information describing the node operation status, and finally fuses it to obtain a unified feature representation. Figure 6 , Figure 6 An example of a multimodal model is given, which includes a Transformer model with self-attention, a BERT model and a fusion module. The node's operating status data (KPI, logs, alarm information, instruction documents, etc.) is input into the multimodal model, and the initial node feature vector corresponding to each node at each moment can be obtained.

[0146] As another example, in step 521, another feature extraction model is used to obtain the text feature vector of the query text. The other feature extraction model may be an LLM model, which is a large model supporting multiple languages. By inputting the query text into the LLM, the corresponding semantic vector, i.e., the text feature vector, can be obtained.

[0147] In a possible implementation, the above step 522 specifically includes the following steps:

[0148] 5221. The network operation and maintenance device determines a neighbor node of node A according to the network topology, wherein node A is any one of the above-mentioned multiple nodes.

[0149] 5222. The network operation and maintenance device performs an operation on the initial node feature vector of node A at time t according to the text feature vector and the initial node feature vector of the neighboring node of node A at time t. The above time t is any one of multiple time.

[0150] In an embodiment of the present application, when updating the initial node feature vector, both the influence of the query text itself and the influence of the neighboring nodes of node A are taken into consideration, which can improve the representation ability of the updated initial node feature vector and help improve the representation ability of the node feature vector.

[0151] Specifically, in step 5222, based on the attention mechanism, the initial node feature vector of each node at different times is updated by combining the query text and the network topology information. Represents the initial node feature vector of node i, with N i Represents the neighbor nodes of node i shown in matrix A, then the initial node feature vector of node i at time t can be updated as:

[0152]

[0153] In the above formula (4), σ is the activation function (e.g., Rectified Linear Unit, ReLU), is the learnable parameter matrix, is the attention coefficient, which is calculated as follows:

[0154]

[0155]

[0156] in, is the text feature vector, The above four matrices are all learnable parameter matrices. Concat means concatenating vectors according to their dimensions, and exp represents an exponential function with the natural constant e as the base.

[0157] It can be seen that when updating the initial node feature vector of node i at time t, the influence of its neighboring nodes is taken into account through the attention mechanism; at the same time, the text feature vector is introduced in the calculation of the attention coefficient. These operations make the updated feature It takes into account not only the impact of the question itself, but also the impact of the interaction characteristics between nodes, thereby providing sufficient content support for the subsequent determination of the response text.

[0158] In the above step 523, the query text and the updated initial node feature vector of the node at different times are combined to obtain the final node feature vector of the node and the topological feature vector of the entire graph. Assuming that the multiple times are K times, taking node i as an example, the updated initial node feature vectors of node i at different times are obtained through step 5222. These updated initial node feature vectors are recorded as Then the final node feature vector of node i is

[0159]

[0160] In the above formula (5), σ is the activation function, is the learnable parameter matrix, β j is the attention coefficient, which is calculated as follows:

[0161]

[0162] and are all learnable parameter matrices.

[0163] It can be seen that in the process of obtaining the final node feature vector of the node, the contribution of the initial node feature vector at different times to the final feature vector is taken into account through the attention mechanism, and the text feature vector is introduced in the calculation of the attention coefficient. These operations make the final node feature vector take into account both the impact of the problem itself and the impact of the node's running status at different times.

[0164] In a possible implementation, the above step 502 further includes the following steps:

[0165] 524. The network operation and maintenance device determines a topology feature vector based on node feature vectors of multiple nodes.

[0166] In the embodiment of the present application, the node feature vectors of multiple nodes are used to obtain the topological feature vector, so that the topological feature vector can represent the characteristics of the first network as a whole.

[0167] Exemplarily, the network operation and maintenance device determines the average value of the node feature vectors of multiple nodes as the topological feature vector. Alternatively, the network operation and maintenance device determines the feature vector obtained by weighted aggregation of the node feature vectors of multiple nodes and the weights corresponding to the multiple nodes as the topological feature vector. Alternatively, the network operation and maintenance device selects the top-k node feature vectors of multiple nodes according to their importance and concatenates them to obtain the topological feature vector, where k is an integer.

[0168] The sum of the weights of the node feature vectors of multiple nodes is 1. Exemplarily, the specific size of the weight corresponding to the node feature vector of each node can be determined based on experience. Also exemplarily, a trainable attention mechanism can be used to learn and obtain the weight of the node feature vector of each node.

[0169] Wherein, k can be 1, 2, 3 or other integers. When k is 1, the node feature vector is concatenated with the empty vector, and the concatenated result is the node feature vector. The concatenation can be horizontal concatenation or vertical concatenation.

[0170] Taking the average value of node feature vectors of multiple nodes as the topological feature vector as an example, assuming that the multiple nodes in the first network are N nodes, after obtaining the node feature vectors {R1, R2, …, R N}After that, the topological feature vector is obtained through the following function:

[0171] R G =Average({R1,R2,…,R N})

[0172] For the convenience of subsequent description, R G As a special node representing R N+1 .

[0173] In an embodiment of the present application, in the feature extraction process of nodes and the entire graph, the changing characteristics of the nodes themselves and the interaction characteristics between nodes can be fully considered on the basis of referring to the specific query text, thereby providing sufficient content support for the subsequent determination of the response text.

[0174] In a possible implementation, the above step 503 specifically includes the following steps:

[0175] 531. When generating the i-th word of the response text, the network operation and maintenance device updates the first feature vector according to the feature vector set to obtain an updated feature vector.

[0176] The feature vector set includes node feature vectors of multiple nodes, a topological feature vector, and multiple word segmentation feature vectors. The multiple word segmentation feature vectors are feature vectors of word segments in the query text.

[0177] 532. The network operation and maintenance device determines the probability of each word in the vocabulary according to the updated feature vector, takes the word with the highest probability as the i-th word in the response text, and adds the feature vector of the i-th word to the feature vector set, and repeats the above steps until the i-th word is a terminator.

[0178] Among them, probability refers to the probability of a word appearing in the response text. When i is greater than 1, the first feature vector is the feature vector of the i-1th word in the response text. When i is 1, the first feature vector is the feature vector of the last word segmentation in the query text. The vocabulary can be set according to the actual situation.

[0179] In the embodiment of the present application, when generating a response text, the first feature vector is updated based on the node feature vector, the topology feature vector, and the feature vector of the existing text (such as the query text, the words of the generated response text), and then the i-th word in the response text is predicted based on the updated feature vector. It can be seen that when generating a response text, both the existing text is taken into consideration and the relevant information of the first network is fully utilized, so that the response text is more fluent and accurate.

[0180] Assuming that the number of nodes in the first network is N nodes and the number of time moments is K time moments, when generating a response text based on the query text, the node feature vector of each node and the topology feature vector, the specific process is as follows:

[0181] For the query text, after segmenting it, use LLM to obtain the representation of each segmentation

[0182] Based on the attention mechanism, the last word segmentation is updated based on the feature vector set The eigenvectors of are:

[0183]

[0184] In the above formula (6), σ is the activation function, is the learnable parameter matrix, γ j is the attention coefficient, which is calculated as follows:

[0185]

[0186]

[0187] are all learnable parameter matrices.

[0188] according to The probability of predicting the next word to appear in the vocabulary is as follows:

[0189]

[0190] is the learnable parameter matrix, and d4 is the total number of word categories. The above formula can be used to obtain the corresponding occurrence probability of each word in the vocabulary, and the word with the highest occurrence probability is taken as the first word in the predicted response text.

[0191] When predicting the second and subsequent words in the response text, taking the prediction of the second word as an example, the feature vector of the predicted first word is added to the above feature vector set, and the above steps are repeated, that is, the feature vector of the first word is updated according to the feature vector set, and the second word is predicted based on the updated feature vector of the first word. This is repeated until the predicted word is a terminator, and the complete response text can be obtained.

[0192] Exemplarily, the above step 503 specifically includes the following steps:

[0193] 533. The network operation and maintenance device determines M candidate texts based on the query text, node feature vectors of multiple nodes, and topology feature vectors, where M is greater than or equal to two.

[0194] Exemplarily, a beam search algorithm may be used to determine the M candidate texts, or other methods may be used to determine the M candidate texts. The number of words in the M candidate texts may be different or the same.

[0195] 534. The network operation and maintenance device uses a candidate text with the largest average probability among the M candidate texts as a response text, where the average probability of the candidate text is the average of the occurrence probabilities of all words in the candidate text.

[0196] In the embodiment of the present application, when generating a response text, M candidate texts may be determined first, and then a candidate text with the largest average probability among the M candidate texts may be determined as the response text. The method of determining the response text by using the average probability of the M candidate texts can effectively improve the accuracy of the response text compared to the above-mentioned methods of step 531 and step 532.

[0197] In this document, the numbering of the steps does not represent the execution order of the steps, and there is no special limitation on the execution order between the steps.

[0198] In order to more clearly illustrate the network operation and maintenance method of the embodiment of the present application, the network operation and maintenance method of the embodiment of the present application is described below based on the model architecture:

[0199] refer to Figure 7 , Figure 7 : is a schematic diagram of a model architecture provided in an embodiment of the present application; the model architecture corresponding to the network operation and maintenance method in an embodiment of the present application includes an input module, an encoder based on query text, a sentence decoder, and an output module, wherein:

[0200] The input module is used to receive the query text, the network topology of the first network and the network status data at multiple moments. Figure 7In the example, the network topology of the first network includes three nodes, namely, node A, node B, and node C, and the multiple time points are three time points t, t+1, and t+2. The input module receives node operation status data of the three nodes at three time points. Exemplarily, the query texts are "node status query", "change risk estimation", and "fault root cause node inference".

[0201] The encoder introduces the query text into the node encoding process to guide each node to extract features based on the query text; it also fully explores the interaction between each node and its neighboring nodes and the node's own change characteristics to improve the representation ability of the extracted feature vector.

[0202] Specifically, the text feature vector corresponding to the query text is R P , based on network topology and R P The updating process of the initial node feature vector of the guiding node is performed to obtain the updated initial node feature vector. P The final node feature vector of the node is determined by the updated initial node feature vector of the node at different times (such as R A , R B and R C ), determine the topological feature vector R according to the node feature vector G .

[0203] The sentence decoder is used to determine the response text based on the query text, the node feature vector and the topology feature vector. Exemplarily, the sentence decoder is an LLM model.

[0204] The output module outputs the response text. For example, the response text is "The current display is the XXX network service layer operation status diagram. The root cause of the fault is the reset of device B, which causes a large ratio of communication packet loss, thus affecting the abnormal status of the upstream and downstream A and C nodes."

[0205] refer to Figure 7 The embodiment of the present application adopts a new paradigm of end-to-end question-answer network operation and maintenance that integrates network spatiotemporal information. The dynamic changes of network topology and node status are directly input into the model, and then corresponding response texts can be given according to different query texts. The method of the embodiment of the present application can be used as a tool to assist expert operation and maintenance and improve efficiency.

[0206] The method of the embodiment of the present application is described in detail above, and the device provided by the embodiment of the present application is introduced below.

[0207] Figure 8A schematic diagram of the structure of a possible network operation and maintenance device provided in an embodiment of the present application. The network operation and maintenance device can be used to implement the functions of the above method embodiment, and thus can also achieve the beneficial effects possessed by the above method embodiment. In an embodiment of the present application, the network operation and maintenance device can be an electronic device, and can also be a module (such as a chip) applied to an electronic device.

[0208] like Figure 8 As shown, the network operation and maintenance device 800 includes an acquisition module 801, an extraction module 802 and a generation module 803. The network operation and maintenance device 800 is used to implement the above Figure 5 Alternatively, the network operation and maintenance device 800 may include a device for implementing the above Figure 5 Any function or operation module in the method embodiment shown in the figure may be implemented in whole or in part by software, hardware, firmware or any combination thereof.

[0209] When the network operation and maintenance device 800 is used to implement Figure 5 When the functions in the method embodiment are shown, the acquisition module 801 is used to obtain the query text, the network topology of the first network and the network status data at multiple times. The above query text is used to characterize the task of querying the first network, and the above network status data includes the operating status data of multiple nodes in the first network. The extraction module 802 is used to extract features of the network topology and the network status data at multiple times based on the query text, and obtain the node feature vector of each node in the multiple nodes and the topology feature vector of the network topology. The generation module 803 is used to generate a response text based on the query text, the node feature vector and the topology feature vector, and the response text is used to characterize the query result.

[0210] In the embodiment of the present application, the network operation and maintenance device 800 guides the feature vector extraction process of the network topology and nodes based on the query text, which can improve the representation ability of the node feature vector and the topology feature vector; and then generates a response text based on the query text, the node feature vector and the topology feature vector, which can provide an accurate response text. The response text can assist the operation and maintenance personnel in network operation and maintenance and improve the operation and maintenance efficiency. In addition, the use of the network operation and maintenance device 800 does not depend on the experience of the operation and maintenance personnel, and the threshold for use is low.

[0211] In one possible implementation, reference Figure 8 The network operation and maintenance device 800 further includes an output module 804. The output module 804 is configured to output a response text.

[0212] In a possible implementation manner, the network operation and maintenance device 800 further includes a determination module and a sending module.

[0213] The above-mentioned determination module is used to determine a control instruction according to the response text, and the control instruction is used to control the first node in the first network to perform the first operation.

[0214] The sending module is used to send a control instruction to the first node.

[0215] This application also provides a network operation and maintenance device, reference Fig. 9 , Fig. 9 900 is a schematic diagram of a network operation and maintenance device provided by an embodiment of the present invention. The network operation and maintenance device 900 includes a memory 901, a processor 902, a communication interface 904 and a bus 903. The memory 901, the processor 902 and the communication interface 904 are connected to each other through the bus 903.

[0216] The memory 901 may be a read-only memory (ROM), a static storage device, a dynamic storage device or a random access memory (RAM). The memory 901 may store a program. When the program stored in the memory 901 is executed by the processor 902, the processor 902 and the communication interface 904 are used to execute each step of the network operation and maintenance method of the embodiment of the present application.

[0217] The processor 902 can adopt a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), a graphics processing unit (GPU) or one or more integrated circuits to execute relevant programs to implement the functions that need to be performed by the units in the network operation and maintenance device of the embodiment of the present application, or to execute the network operation and maintenance method of the embodiment of the present application.

[0218] The processor 902 may also be an integrated circuit chip having signal processing capabilities. In the implementation process, each step of the network operation and maintenance method of the embodiment of the present application may be completed by an integrated logic circuit of hardware or software instructions in the processor 902. The above-mentioned processor 902 may also be a general-purpose processor, a digital signal processor (Digital Signal Processing, DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present application may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the network operation and maintenance method in combination with the embodiment of the present application may be directly embodied as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 901, and the processor 902 reads the information in the memory 901, and combines its hardware to complete the functions required to be performed by the units included in the network operation and maintenance device of the embodiment of the present application, or executes the network operation and maintenance method of the embodiment of the present application.

[0219] The communication interface 904 uses a transceiver device such as, but not limited to, a transceiver to implement communication between the network operation and maintenance device 900 and other devices or communication networks. For example, query texts can be obtained through the communication interface 904.

[0220] The bus 903 may include a path for transmitting information between various components of the network operation and maintenance device 900 (eg, the memory 901 , the processor 902 , and the communication interface 904 ).

[0221] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0222] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0223] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0224] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media integration. The available medium can be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk), etc.

[0225] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A network operation and maintenance method, characterized in that: The method comprises: Acquire a query text, a network topology of the first network, and network status data at multiple times; the query text is used to represent a task of querying the first network, and the network status data includes operation status data of multiple nodes in the first network; Based on the query text, feature extraction is performed on the network topology and the network status data at the multiple moments to obtain a node feature vector of each of the multiple nodes and a topology feature vector of the network topology; A response text is generated based on the query text, the node feature vector and the topology feature vector, and the response text is used to represent the query result.

2. The method according to claim 1, characterized in that The extracting features of the network topology and the network status data at the multiple times based on the query text to obtain a node feature vector of each of the multiple nodes and a topology feature vector of the network topology includes: Performing feature extraction on the query text to obtain a text feature vector; performing feature extraction on the network status data at the multiple times to determine an initial node feature vector of each of the multiple nodes at the multiple times; Based on the text feature vector and the network topology, updating the initial node feature vector of each node at the multiple time points; Based on the text feature vector and the updated initial node feature vectors of each node at multiple time points, a node feature vector of each node is determined.

3. The method according to claim 2, characterized in that The updating of the initial node feature vector of each node at the multiple time points based on the text feature vector and the network topology includes: Determine a neighbor node of node A according to the network topology, wherein node A is any one of the multiple nodes; According to the text feature vector and the initial node feature vector of the neighboring node of node A at time t, the initial node feature vector of node A at time t is Update; the time t is any one of the multiple time moments.

4. The method according to claim 2 or 3, characterized in that: The method of extracting features from the network topology and the network status data at the multiple times based on the query text to obtain a node feature vector of each of the multiple nodes and a topology feature vector of the network topology also includes: The topology feature vector is determined according to the node feature vectors of the multiple nodes.

5. The method according to claim 4, characterized in that The determining the topological feature vector according to the node feature vectors of the plurality of nodes comprises: Determine an average value of node feature vectors of the plurality of nodes as the topological feature vector; or, Determine a feature vector obtained by weighted aggregation of node feature vectors of the multiple nodes and weights corresponding to the multiple nodes as the topological feature vector; or, The node feature vectors of the multiple nodes are selected according to their importance and the top-k nodes are concatenated to obtain the topological feature vector, where k is an integer.

6. The method according to any one of claims 1 to 5, characterized in that: The generating a response text based on the query text, the node feature vector and the topology feature vector comprises: When generating the i-th word of the response text, updating the first feature vector according to the feature vector set to obtain an updated feature vector, wherein the feature vector set includes node feature vectors of the multiple nodes, the topological feature vector, and multiple word segmentation feature vectors, and the multiple word segmentation feature vectors are feature vectors of the word segments in the query text; Determine the probability of each word in the vocabulary according to the updated feature vector, take the word with the highest probability as the i-th word in the response text, add the feature vector of the i-th word to the feature vector set, and repeat the above steps until the i-th word is a terminator; When i is greater than 1, the first feature vector is the feature vector of the i-1th word in the response text; when i is 1, the first feature vector is the feature vector of the last word in the query text.

7. The method according to any one of claims 1 to 5, characterized in that: The generating a response text based on the query text, the node feature vector and the topology feature vector comprises: Determine M candidate texts based on the query text, the node feature vectors of the plurality of nodes, and the topological feature vector, where M is greater than or equal to two; A candidate text with the largest average probability among the M candidate texts is used as the response text, and the average probability of the candidate text is the average of the occurrence probabilities of all words in the candidate text.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: Output the response text.

9. The method according to any one of claims 1 to 8, characterized in that: The method further comprises: Determine a control instruction according to the response text, where the control instruction is used to control a first node in the first network to perform a first operation; Send the control instruction to the first node.

10. A network operation and maintenance device, characterized in that: include: An acquisition module, used to acquire the query text, the network topology of the first network, and network status data at multiple times; The query text is used to represent a task of querying the first network, and the network status data includes operation status data of a plurality of nodes in the first network; An extraction module, configured to perform feature extraction on the network topology and the network status data at the multiple moments based on the query text, to obtain a node feature vector of each of the multiple nodes and a topology feature vector of the network topology; A generating module is used to generate a response text based on the query text, the node feature vector and the topology feature vector, wherein the response text is used to represent the query result.

11. The device according to claim 10, characterized in that The device also includes: An output module is used to output the response text.

12. The device according to claim 10 or 11, characterized in that The device also includes: a determination module, configured to determine a control instruction according to the response text, wherein the control instruction is used to control a first node in the first network to perform a first operation; A sending module is used to send the control instruction to the first node.

13. A network operation and maintenance device, characterized in that: It comprises a processor and a memory, wherein the processor and the memory are connected, wherein the memory is used to store program code, and the processor is used to call the program code to execute the network operation and maintenance method according to any one of claims 1 to 9.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the network operation and maintenance method according to any one of claims 1 to 9.

Images