Power operation and maintenance method, device, equipment, readable storage medium and program product

Abstract

The invention relates to a power operation and maintenance method, device and equipment, a readable storage medium and a program product. The method comprises the following steps: fusing a scene vector of a scene image of a power operation and maintenance scene, a text vector of an operation and maintenance text and a state vector of equipment state data to obtain a scene state vector; querying a matched task template from a corresponding task template library based on the scene state vector, determining an operation and maintenance intention vector corresponding to the queried task template, determining at least one power operation and maintenance sub-task based on the task template, and determining task attribute information corresponding to the at least one power operation and maintenance sub-task based on the operation and maintenance intention vector; for each electric power operation and maintenance sub-task, based on the corresponding task attribute information, querying a matched agent, and executing the electric power operation and maintenance sub-task through the queried agent to obtain a corresponding operation and maintenance sub-result; and the operation and maintenance result is determined based on the operation and maintenance sub-result of each power operation and maintenance sub-task, so that the power operation and maintenance accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a power operation and maintenance method, device, equipment, readable storage medium and program product. BACKGROUND

[0002] With the development of computer technology, the patrol task of power production environment continues to increase in complexity, task diversity and safety constraints.

[0003] In related technologies, the power scene is usually patrolled and maintained in a way mainly based on manual observation, manual interpretation and manual scheduling. However, this way cannot accurately identify the actual operation and maintenance task in the current power scene, and there is a problem of low power operation and maintenance accuracy. SUMMARY

[0004] Therefore, it is necessary to provide a power operation and maintenance method, device, equipment, readable storage medium and program product capable of improving the accuracy of power operation and maintenance to solve the above technical problems.

[0005] In a first aspect, the present application provides a power operation and maintenance method, comprising:

[0006] obtaining a scene image of a power operation and maintenance scene, operation and maintenance text, and device state data of a device in the power operation and maintenance scene;

[0007] fusing a scene vector of the scene image, a text vector of the operation and maintenance text, and a state vector of the device state data to obtain a scene state vector of the power operation and maintenance scene;

[0008] based on the scene state vector, querying a matched task template from a task template library corresponding to the power operation and maintenance scene, determining an operation and maintenance intention vector corresponding to the queried task template, and based on the task template, determining at least one power operation and maintenance subtask, and based on the operation and maintenance intention vector, determining task attribute information corresponding to the at least one power operation and maintenance subtask;

[0009] for each power operation and maintenance subtask, based on the task attribute information corresponding to the power operation and maintenance subtask, querying a matched agent from a plurality of agents to execute the power operation and maintenance subtask through the queried agent to obtain a corresponding operation and maintenance subresult;

[0010] based on the operation and maintenance subresult of each power operation and maintenance subtask, determining an operation and maintenance result about the power operation and maintenance scene.

[0011] In a second aspect, the present application further provides a power operation and maintenance device, comprising:

[0012] The data acquisition module is used to acquire scene images, maintenance text, and equipment status data of devices in the power operation and maintenance scenario.

[0013] The vector fusion module is used to fuse the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data to obtain the scene state vector of the power operation and maintenance scenario.

[0014] The task determination module is used to query a matching task template from the task template library corresponding to the power operation and maintenance scenario based on the scenario state vector, determine the operation and maintenance intent vector corresponding to the queried task template, determine at least one power operation and maintenance sub-task based on the task template, and determine the task attribute information corresponding to at least one power operation and maintenance sub-task based on the operation and maintenance intent vector.

[0015] The task execution module is used to query a matching intelligent agent from multiple intelligent agents for each power operation and maintenance sub-task based on the task attribute information corresponding to the power operation and maintenance sub-task, so as to execute the power operation and maintenance sub-task through the queried intelligent agent and obtain the corresponding operation and maintenance sub-result;

[0016] The operation and maintenance result determination module is used to determine the operation and maintenance result for the power operation and maintenance scenario based on the operation and maintenance sub-results of each power operation and maintenance sub-task.

[0017] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0018] Acquire scene images, maintenance text, and device status data of equipment in the power operation and maintenance scenario;

[0019] By fusing the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data, the scene state vector of the power operation and maintenance scenario is obtained.

[0020] Based on the scenario state vector, a matching task template is queried from the task template library corresponding to the power operation and maintenance scenario, and the operation and maintenance intent vector corresponding to the queried task template is determined. Based on the task template, at least one power operation and maintenance sub-task is determined, and based on the operation and maintenance intent vector, the task attribute information corresponding to at least one power operation and maintenance sub-task is determined.

[0021] For each power operation and maintenance sub-task, based on the task attribute information corresponding to the power operation and maintenance sub-task, a matching intelligent agent is queried from multiple intelligent agents, and the power operation and maintenance sub-task is executed by the queried intelligent agent to obtain the corresponding operation and maintenance sub-result;

[0022] Based on the operation and maintenance sub-results of each power operation and maintenance sub-task, the operation and maintenance results for the power operation and maintenance scenario are determined.

[0023] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0024] Acquire scene images, maintenance text, and device status data of equipment in the power operation and maintenance scenario;

[0025] By fusing the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data, the scene state vector of the power operation and maintenance scenario is obtained.

[0026] Based on the scenario state vector, a matching task template is queried from the task template library corresponding to the power operation and maintenance scenario, and the operation and maintenance intent vector corresponding to the queried task template is determined. Based on the task template, at least one power operation and maintenance sub-task is determined, and based on the operation and maintenance intent vector, the task attribute information corresponding to at least one power operation and maintenance sub-task is determined.

[0027] For each power operation and maintenance sub-task, based on the task attribute information corresponding to the power operation and maintenance sub-task, a matching intelligent agent is queried from multiple intelligent agents, and the power operation and maintenance sub-task is executed by the queried intelligent agent to obtain the corresponding operation and maintenance sub-result;

[0028] Based on the operation and maintenance sub-results of each power operation and maintenance sub-task, the operation and maintenance results for the power operation and maintenance scenario are determined.

[0029] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0030] Acquire scene images, maintenance text, and device status data of equipment in the power operation and maintenance scenario;

[0031] By fusing the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data, the scene state vector of the power operation and maintenance scenario is obtained.

[0032] Based on the scenario state vector, a matching task template is queried from the task template library corresponding to the power operation and maintenance scenario, and the operation and maintenance intent vector corresponding to the queried task template is determined. Based on the task template, at least one power operation and maintenance sub-task is determined, and based on the operation and maintenance intent vector, the task attribute information corresponding to at least one power operation and maintenance sub-task is determined.

[0033] For each power operation and maintenance sub-task, based on the task attribute information corresponding to the power operation and maintenance sub-task, a matching intelligent agent is queried from multiple intelligent agents, and the power operation and maintenance sub-task is executed by the queried intelligent agent to obtain the corresponding operation and maintenance sub-result;

[0034] Based on the operation and maintenance sub-results of each power operation and maintenance sub-task, the operation and maintenance results for the power operation and maintenance scenario are determined.

[0035] The aforementioned power operation and maintenance methods, devices, equipment, readable storage media, and program products acquire scene images, operation and maintenance text, and equipment status data of devices in the power operation and maintenance scenario. They then fuse scene vectors from the scene images, text vectors from the operation and maintenance text, and status vectors from the equipment status data to obtain a scene state vector for the power operation and maintenance scenario. In other words, by fusing vectors of different dimensions, the scene state vector encompasses multimodal details such as images, text, and status data, ensuring the effectiveness of power operation and maintenance. Subsequently, based on the scene state vector, a matching task template is retrieved from a task template library corresponding to the power operation and maintenance scenario. This automatically determines the operation and maintenance intent vector corresponding to the retrieved task template and identifies at least one power operation and maintenance sub-task based on the task template. Based on the operation and maintenance intent vector, the task attribute information corresponding to at least one power operation and maintenance sub-task is determined, eliminating the need for manual confirmation of the task details of the power operation and maintenance sub-task to be executed. For each power operation and maintenance sub-task, based on the task attribute information corresponding to the sub-task, a matching agent is accurately queried from multiple agents. This matched agent then automatically executes the sub-task, yielding the corresponding sub-result. Based on the sub-result of each sub-task, the overall operation and maintenance outcome for the power operation and maintenance scenario is determined. Throughout this process, without human intervention, the operation and maintenance intent vector, reflecting the operation and maintenance intent, is accurately identified based on the scenario state vector containing multimodal details. This allows for the accurate and automatic identification of the task attribute information of the required sub-tasks. After assigning a corresponding agent to each sub-task, the agent automatically executes the sub-task according to its task attribute information, thus improving the accuracy of power operation and maintenance. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1This is an application environment diagram of a power operation and maintenance method in one embodiment;

[0038] Figure 2 This is a flowchart illustrating a power operation and maintenance method in one embodiment;

[0039] Figure 3 This is a structural block diagram of a power operation and maintenance device in one embodiment;

[0040] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0042] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0043] The power operation and maintenance method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0044] In some embodiments, after obtaining the scene image, maintenance text, and device status data of the devices in the power maintenance scenario sent by the terminal 102, the server 104 fuses the scene vector of the scene image, the text vector of the maintenance text, and the status vector of the device status data to obtain the scene status vector of the power maintenance scenario. Based on the scene status vector, the server 104 queries a matching task template from the task template library corresponding to the power maintenance scenario, determines the maintenance intent vector corresponding to the queried task template, and determines at least one power maintenance sub-task based on the task template. Based on the maintenance intent vector, the server 104 determines the task attribute information corresponding to at least one power maintenance sub-task. For each power maintenance sub-task, the server 104 queries a matching intelligent agent from multiple intelligent agents based on the task attribute information corresponding to the power maintenance sub-task, and executes the power maintenance sub-task through the queried intelligent agent to obtain the corresponding maintenance sub-result. Based on the maintenance sub-result of each power maintenance sub-task, the server 104 determines the maintenance result for the power maintenance scenario.

[0045] For example, terminal 102 acquires scene images sent by a camera device deployed on terminal 102. After acquiring device status data sent by various devices in the power operation and maintenance scenario, in response to an input operation of operation and maintenance text, terminal 102 sends the input operation and maintenance text, the acquired scene images, and the device status data to server 104. Alternatively, the camera device can be deployed on a drone. During drone patrol, the camera device captures images of the power operation and maintenance scenario it encounters and directly sends the captured scene images to server 104. Furthermore, the devices in the power operation and maintenance scenario also communicate directly with server 104, sending their own device status data directly to server 104. Therefore, server 104 performs subsequent vector fusion based on the acquired scene images, operation and maintenance text, and device status data.

[0046] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, etc. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0047] In one exemplary embodiment, such as Figure 2 As shown, a power operation and maintenance method is provided, which can be applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S202 to S210. Wherein:

[0048] Step S202: Obtain scene images, maintenance text, and device status data of equipment in the power operation and maintenance scenario.

[0049] The scene images are real-time images captured of the power operation and maintenance (O&M) scenario. The O&M text is text describing the power O&M process within that scenario. The equipment status data includes the real-time operating status of the corresponding equipment within the O&M scenario. For example, equipment status data includes the voltage, current, active power, temperature, cable potential, and switch operation status of the corresponding equipment.

[0050] Optionally, the image acquisition device acquires images of the power operation and maintenance scene, resizes the acquired images uniformly (e.g., uniformly resize to 224×224), maintains the RGB (Red, Green, Blue) three-channel structure, and uses the resized images as the scene images. The image acquisition device then sends the scene images to the server. The image acquisition device can be a field operation and maintenance terminal or a high-definition camera mounted on a drone.

[0051] Optionally, maintenance personnel can input maintenance text through the inspection task client app (application software) on the maintenance terminal. For example, the maintenance text can be in the form of instructions such as "Check if the wiring of transformer No. 10 is abnormal" or "Review the current status of switch No. 3". Then, the inspection task client app sends the obtained maintenance text to the server.

[0052] Optionally, for each device in the power operation and maintenance scenario, the device collects its own device status data and sends the collected device status data to the server.

[0053] Step S204: Fuse the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data to obtain the scene state vector of the power operation and maintenance scenario.

[0054] Optionally, the server determines the scene vector of the scene image, the text vector of the maintenance text, and the state vector of the device status data. Then, according to a fusion method matching the power maintenance scenario, the scene vector of the scene image, the text vector of the maintenance text, and the state vector of the device status data are fused to obtain the scene state vector of the power maintenance scenario. The fusion method can be either splicing according to a preset splicing order or superimposing according to a preset superimposition method.

[0055] In some embodiments, the scene vector determination step of the scene image includes: acquiring a scene image sent by an image acquisition device; and extracting features from the region image of the key inspection area in the scene image according to an image feature extraction model to obtain the scene vector of the scene image.

[0056] The image feature extraction model can be a neural network-based model. For example, it could include a lightweight ResNet18 network (18 layers deep) and pooling layers. The lightweight ResNet18 network contains five residual blocks, each consisting of two 3×3 convolutional layers, BatchNorm (batch normalization), and ReLU (Remote Luminary Unit) activations. The region image is part of the scene image and represents the key inspection areas; these key inspection areas include equipment nameplates, cable heads, and high-voltage switches.

[0057] For example, after the server acquires the scene image sent by the image acquisition device, the server calls the trained image feature extraction model, inputs the scene image into the image feature extraction model, identifies the key inspection area in the scene image through the image feature extraction model, extracts features from the area image, and outputs the scene vector of the scene image.

[0058] For example, to improve the image feature extraction model's attention to key inspection areas in the scene image, an attention mechanism is introduced on the last residual output feature map. Convolution plus softmax operation generates spatial attention weights The feature maps are then weighted and converged, as detailed in the following formula (1):

[0059] (1)

[0060] in, It is the image semantic vector obtained after attention-weighted convergence of scene images. Indicates the first Location-based attention weights It is the first output of the ResNet18 network. The feature response vector of the location, As the normalization factor, Through a A two-dimensional weight distribution generated by a softmax function is applied after the convolutional layer to ensure that the image feature extraction model focuses more on key inspection regions, improving the semantic concentration of the representation. Then, the image semantic vector can be globally averaged using the pooling layer of the image feature extraction model to obtain the scene vector.

[0061] In the above embodiments, the image feature extraction model can extract regional images of key inspection areas from scene images, focusing on scene details of key inspection areas, thereby ensuring the accuracy and effectiveness of power operation and maintenance.

[0062] In some embodiments, the text vector determination step of the operation and maintenance text includes: acquiring the operation and maintenance text sent by the inspection task client, performing word segmentation on the operation and maintenance text to obtain the word segmentation result; and determining the text vector of the operation and maintenance text based on the word segmentation result through a text feature extraction model.

[0063] The text feature extraction model can be built on a neural network. For example, it can be a pre-trained BERT (Bidirectional Encoder Representation Model). For instance, the text feature extraction model includes a 12-layer Transformer structure, with each layer outputting a hidden state dimension of 768. Finally, the output at the CLS (Classifier Label) position is taken as the text vector. , used to express the core semantics of task instructions in the input operation and maintenance text.

[0064] For example, after receiving the maintenance text sent by the inspection task client, the server uses a word segmenter to map the maintenance text into a token (word) sequence, which is the word segmentation result. The server inputs the word segmentation result into a text feature extraction model to extract text features and obtain the text vector of the maintenance text.

[0065] In the above embodiments, by segmenting the operation and maintenance text and extracting features, the details at the operation and maintenance text level can be identified in a key way, ensuring the accuracy and effectiveness of power operation and maintenance.

[0066] In some embodiments, the step of determining the state vector of device status data includes: acquiring device status data reported by the device in the power operation and maintenance scenario, wherein the device status data includes the state values ​​of multiple state dimensions; and extracting features from the device status data through a state feature extraction model to obtain the state vector of the device status data.

[0067] Among them, equipment status data reflects the real-time operating status of the corresponding equipment. This equipment status data includes status values ​​for various status dimensions, such as voltage, current, active power, temperature, cable potential, and switch operation status of the corresponding equipment. The format is integer or floating-point, and the units include... Etc. The data acquisition cycle is generally 5-10 seconds, and the data is actively reported by the device.

[0068] Optionally, the server identifies the currently inspected device from multiple devices in the power operation and maintenance scenario and sends a device data collection command to the inspected device, enabling the device to collect its own device status data and send it to the server. The server then obtains the device status data. Subsequently, the device status data includes status values ​​across multiple status dimensions. For each status dimension, the server acquires the status value for that dimension and inputs the status values ​​of each status dimension into the status feature extraction model to obtain the status vector of the device status data. For example, each status value is first min-max normalized by the status feature extraction model, and then mapped to a vector representation in a unified semantic space through a fully connected layer, as shown in formula (2):

[0069] (2)

[0070] in, Represents the state vector. The first The minimum and maximum values ​​of each status value in the historical device status data. The linear mapping weight matrix, As the bias term, the final output .

[0071] In the above embodiments, by extracting features from the equipment status data, a state vector representing detailed information about the equipment status can be extracted, ensuring the accuracy and effectiveness of power operation and maintenance.

[0072] In some embodiments, the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data are fused to obtain the scene state vector of the power operation and maintenance scenario, including: determining the concatenation order of the scene vector, text vector, and state vector respectively; and concatenating the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data according to the concatenation order of the scene vector, text vector, and state vector to obtain the scene state vector of the power operation and maintenance scenario.

[0073] For example, the concatenation order of the scene vector, text vector, and state vector can be pre-set. For instance, the scene vector can be concatenated first, the text vector second, and the state vector third. In this case, the text vector is concatenated after the scene vector, and the state vector is concatenated after the state vector to obtain the scene-state vector.

[0074] For example, when the server verifies that the clarity of the scene image reaches a preset level, it determines that the scene vector is the first concatenation order; then, after the scene vector, the state vector is concatenated, and after the state vector, the text vector is concatenated to obtain the scene state vector.

[0075] For example, the server maps the scene vector, text vector, and state vector to a unified dimension d=256, and then concatenates the scene vector, text vector, and state vector mapped to the unified dimension in the concatenation order of scene vector, text vector, and state vector. The concatenated vector is then processed by a two-layer multi-sensor network (MLP) and then processed by an activation function and LayerNorm (layer normalization) to obtain the scene state vector.

[0076] In the above embodiments, the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data are concatenated in the order of scene vector, text vector, and state vector to obtain the scene state vector of the power operation and maintenance scenario. This allows for the orderly concatenation of vectors from the scene, text, and state dimensions to obtain multimodal detailed information, ensuring the effectiveness of power operation and maintenance.

[0077] Step S206: Based on the scenario state vector, query the matching task template from the task template library corresponding to the power operation and maintenance scenario, determine the operation and maintenance intent vector corresponding to the queried task template, determine at least one power operation and maintenance sub-task based on the task template, and determine the task attribute information corresponding to at least one power operation and maintenance sub-task based on the operation and maintenance intent vector.

[0078] The task template library stores preset task templates, and each task template has a corresponding structured template vector, which is used to describe a specific task. For example, the task template library... Each task template A standardized task structure, namely a structured template vector, is defined. This structured template vector includes specific values ​​for dimensions such as task type encoding, target equipment type, execution action slots, and constraint structure. The task template library is constructed based on typical work orders and regulations in the power inspection field. For example, "checking circuit breaker status," "reading cable temperature," and "determining whether the voltage exceeds the upper limit" are all basic types in the prototype. The operation and maintenance intent vector reflects the true intent of power operation and maintenance. Decomposing the operation and maintenance intent vector, that is, decomposing the true intent, yields each sub-intent. This sub-intent can be regarded as the task attribute information required for each power operation and maintenance sub-task. The task attribute information can be understood as the task representation of the corresponding power operation and maintenance sub-task, reflecting the task details of the specific power operation and maintenance sub-task.

[0079] For example, the server generates template selection prompt text based on the scenario state vector and the library identifier of the task template library corresponding to the power operation and maintenance scenario, and calls a large language model to perform semantic understanding on the template selection prompt text in order to select a task template that matches the scenario state vector from the task templates of the task template library corresponding to the library identifier.

[0080] For example, after retrieving a matching task template, the server determines the corresponding operational intent vector based on the scene state vector and the structured vector of the retrieved task template. For instance, based on the scene state vector and the structured vector of the retrieved task template, it generates intent vector recognition prompt text, calls a large language model to perform semantic understanding on the intent vector recognition prompt text, and outputs the operational intent vector.

[0081] For example, the server determines a task graph that matches the queried task template. The task graph includes at least one power operation and maintenance sub-task. Based on the operation and maintenance intent vector and the task graph, an operation and maintenance task generation prompt text is generated. The large language model is called again to perform semantic understanding on the operation and maintenance task generation prompt text to identify the true operation and maintenance intent of the operation and maintenance intent vector and decompose the true operation and maintenance intent. Based on the decomposed sub-intents, the task attribute information of the corresponding power operation and maintenance sub-tasks is determined respectively.

[0082] In some embodiments, based on the scenario state vector, a matching task template is retrieved from the task template library corresponding to the power operation and maintenance scenario, including: obtaining the structured template vector of each task template in the task template library; constructing a task template matching function based on the scenario state vector and the structured template vector of each task template in the task template library; and retrieving a matching task template from the task template library corresponding to the power operation and maintenance scenario with the goal of maximizing the task template matching function.

[0083] For example, the server obtains the scene state vector C and the structured template vector of each task template in the task template library. Then, construct the task template matching function as shown in formula (3) below:

[0084] (3)

[0085] Formula (3) introduces a task scenario adaptation term. Adjustable hyperparameters Task scenario adaptation items Adjustable hyperparameters This is to improve the structural sensitivity and task background matching ability during the matching process. This calculates the degree of matching between the scene state vector and the corresponding structured template vector in the same dimension. It should be noted that the dimensions of both the scene state vector and the structured template vector are adjusted to be the same. The calculation can be found in the following formula (4):

[0086] (4)

[0087] For any dimension, It is the vector value in this dimension of the scene state vector; It is the vector value in this dimension of the structured template vector. This can be understood as a reference value for the corresponding task template in this dimension, such as the normal voltage range or temperature threshold. The introduction of this item ensures that the task template is not only semantically close to the input, but also matches the actual scenario in terms of electrical state rationality, effectively avoiding task ambiguity caused by mismatch.

[0088] Then, the task template corresponding to the maximum value of the task template matching function is determined by the following formula (5):

[0089] (5)

[0090] in, The template number is the template number of the task template corresponding to the maximum value of the task template matching function. Thus, the matching task template can be determined based on the template number that satisfies the maximization of the task template matching function.

[0091] In the above embodiments, a task template matching function is constructed using the scene state vector and the structured template vector of each task template in the task template library. This function evaluates the degree of matching between the scene and the task template, and the goal is to maximize the matching function to retrieve the task template with the highest degree of matching. Thus, matching task templates can be retrieved from the dimension of the scene state vector.

[0092] In some embodiments, determining the operation and maintenance intent vector corresponding to the queried task template includes: concatenating the scenario state vector and the structured template vector of the queried task template to obtain the concatenated vector; and performing a linear transformation on the concatenated vector to obtain the operation and maintenance intent vector corresponding to the queried task template.

[0093] For example, after retrieving the matching task template, the operation and maintenance intent vector can be determined by referring to the processing method of formula (6). :

[0094] (6)

[0095] in, It is a structured template vector of the matching task template. For linear weights, As a dimension, For bias terms, This indicates a splicing operation. ,in This is used for subsequent action graph generation. Linear weights and bias terms are used to perform linear transformations on the concatenated vectors. The final output maintenance intent vector Q is a task description vector that reflects the actual maintenance intent. The maintenance intent vector contains information such as task type encoding (e.g., "inspection," "detection," "record"), target device number or type (determined by image recognition results and text prompts), execution parameters (e.g., sampling frequency, type of read index), trigger condition vector (whether it exceeds limits, whether it is a fault), and safety boundary description (e.g., engineering rules such as "remote power outage prohibited," "no high-voltage closing involved"). For example, when the scene state vector C is generated based on an image of a scene containing a transformer area, maintenance text containing "check temperature," and device state data containing temperature data above the upper limit, the matched task template is "detection task template," the task type is "read device temperature," and the condition is "current temperature exceeds limit."

[0096] In the above embodiments, since the scene state vector provides detailed information about the specific environment and the task template provides action boxes that can be done, the linear transformation after splicing is essentially to perform "weighted selection" and "parameter adaptation" of standardized actions in the corresponding scene, thereby extracting the real operation and maintenance intention, that is, the operation and maintenance intention vector of "what operation and maintenance should be performed now".

[0097] In some embodiments, determining at least one power operation and maintenance sub-task based on a task template includes: obtaining the template number of the queried task template; and querying a matching task map from a preset task map library based on the template number, wherein the matching task map indicates at least one power operation and maintenance sub-task.

[0098] The task graph library includes multiple task graphs, which are acyclic graphs. Each task graph contains multiple nodes connected as needed, and each node can be considered a power operation and maintenance sub-task. Nodes in this task graph can be considered empty, meaning the specific details of the corresponding power operation and maintenance sub-task are unknown. Therefore, it is necessary to determine the task attribute information of each power operation and maintenance sub-task based on the operation and maintenance intent vector. This task graph illustrates the specific execution order of each power operation and maintenance sub-task.

[0099] For example, the server obtains the template number of the retrieved task template. Based on the template number, it retrieves a matching task image from a pre-defined task image library. .in, A collection of nodes of atomic action types. This is a set of dependent edges. Parse the task graph to identify at least one power operation and maintenance path, each path comprising at least one ordered power operation and maintenance sub-task.

[0100] In the above embodiments, by querying the template number of the task template, the appropriate task diagram can be retrieved from the task diagram library, thereby directly determining the various power operation and maintenance sub-tasks that need to be executed at present.

[0101] Since the operation and maintenance intent vector includes parameters such as the target, threshold, and sampling period required by each node, the server performs parameter binding and representation alignment on the nodes of the queried task graph to determine the task attribute information. That is, after determining at least one power operation and maintenance sub-task matching the queried task template, in some embodiments, the step of determining the task attribute information of the power operation and maintenance sub-task includes: for each node in the queried task graph... Based on the field mapping rules of action type, extract the corresponding parameter slices from the operation and maintenance intent vector. The node type embedding is then concatenated with the minimum context and fed into a two-layer perceptron. (Hidden dimension is) The activation uses ReLU (an activation function) with LayerNorm (layer normalization) to obtain the representation of the node. For details, please refer to formula (7):

[0102] (7)

[0103] in, Embed the corresponding vocabulary; This is a slice of the parameters required for this node extracted from the operational intent vector; This is a vector representing the minimum context (including task template category labels and stage labels). The above mapping only changes the attribute representation and dimension of the nodes; it does not change the edge set of the template graph. The instantiated task graph is denoted as ,in .Should This is considered as the task attribute information of the corresponding power operation and maintenance sub-task. In the above process, to avoid numerical instability caused by differences in scale between fields from different sources, the numerical fields involved in the concatenation are range-scaled according to empirical upper bounds in the template mapping table, while discrete fields are embedded using a vocabulary, and the resulting composite representation is processed through the same... Projected onto a unified dimension.

[0104] Therefore, after determining the node representation of each node, the instantiated task graph is obtained, and the node representation of the node can be regarded as the task attribute information of the corresponding power operation and maintenance sub-task.

[0105] Step S208: For each power operation and maintenance sub-task, based on the task attribute information corresponding to the power operation and maintenance sub-task, a matching intelligent agent is queried from multiple intelligent agents, so as to execute the power operation and maintenance sub-task through the queried intelligent agent and obtain the corresponding operation and maintenance sub-result.

[0106] Optionally, after determining the task attribute information of each power operation and maintenance sub-task, the server obtains the capability vector of the currently available agents, and determines the agent corresponding to each power operation and maintenance sub-task based on the task attribute information of each power operation and maintenance sub-task and the capability vector of the agent. The capability vector illustrates the capabilities configured by the corresponding agent. For example, based on the task attribute information of each power operation and maintenance sub-task, an action requirement vector matching the task attribute information is queried from the task action requirement vector library. This action requirement vector reflects the capabilities required to execute the corresponding power operation and maintenance sub-task.

[0107] For example, suppose the number of currently available agents is . , No. The capability vector of each agent is (Action capability markers registered during platform deployment), after each node is determined The action requirement vector is Next, define the assignment matrix. , Indicates that the node Assigned to intelligent agents Considering the concurrency limitations and operational safety boundaries of power line inspections, the following task allocation function is constructed, referring to formula (8):

[0108] (8)

[0109] Among them, the first term in the task allocation function punishes capability mismatch; the second term inhibits the same agent from undertaking multiple tasks on potentially congested pairs simultaneously; and the third term restricts managed agents from undertaking higher-risk actions. For a set of potential congested pairs, the congestion indicator Calculated using rules based on the same region or classmate; node risk scalar. These are pre-set parameters; agent limitation indication. Provided by platform permission configuration. In formula (8) These are the weighting coefficients, default. Recommended range is (The value can be increased appropriately when there are many congested pairs or long concurrent chains.) (Adjustments can be made upwards when the proportion of high-risk actions increases). The adjustment logic is: if the number of feasible solutions is low, adjustments will be made downwards first. If the assignment involves restricted agents and high-risk combinations, the allocation should be adjusted upwards appropriately. The constraints include each node being assigned to exactly one agent (row sum equal to 1) and an agent concurrency limit (column sum does not exceed its concurrency cap). This problem can be solved online using a 0–1 integer programming solver or a heuristic approximation to obtain the assignment matrix. Then by dependency edge The topological order is used to generate execution queues for each agent.

[0110] Therefore, by solving the task allocation function described above, the allocation matrix M can be determined, and based on the allocation matrix M, the power operation and maintenance sub-tasks that each agent needs to execute can be determined.

[0111] For example, after determining the intelligent agent matched to each power operation and maintenance sub-task, the corresponding power operation and maintenance sub-task is executed by the intelligent agent to obtain the corresponding operation and maintenance sub-result.

[0112] Step S210: Based on the operation and maintenance sub-results of each power operation and maintenance sub-task, determine the operation and maintenance results for the power operation and maintenance scenario.

[0113] For example, the server aggregates the operation and maintenance sub-results of each power operation and maintenance sub-task to obtain the operation and maintenance results for the power operation and maintenance scenario.

[0114] In the aforementioned power operation and maintenance method, scene images, operation and maintenance text, and equipment status data of devices in the power operation and maintenance scenario are acquired. The scene vectors from the scene images, the text vectors from the operation and maintenance text, and the status vectors from the equipment status data are then fused to obtain the scene state vector of the power operation and maintenance scenario. In other words, by fusing vectors of different dimensions, the scene state vector can encompass multimodal details such as images, text, and status data, ensuring the effectiveness of power operation and maintenance. Then, based on the scene state vector, a matching task template is retrieved from the task template library corresponding to the power operation and maintenance scenario. This automatically determines the operation and maintenance intent vector corresponding to the retrieved task template, and at least one power operation and maintenance sub-task is determined based on the task template. Based on the operation and maintenance intent vector, the task attribute information corresponding to at least one power operation and maintenance sub-task is determined, eliminating the need for manual confirmation of the task details of the power operation and maintenance sub-task to be executed. For each power operation and maintenance sub-task, based on the task attribute information corresponding to the sub-task, a matching agent is accurately queried from multiple agents. This matched agent then automatically executes the sub-task, yielding the corresponding sub-result. Based on the sub-result of each sub-task, the overall operation and maintenance outcome for the power operation and maintenance scenario is determined. Throughout this process, without human intervention, the operation and maintenance intent vector, reflecting the operation and maintenance intent, is accurately identified based on the scenario state vector containing multimodal details. This allows for the accurate and automatic identification of the task attribute information of the required sub-tasks. After assigning a corresponding agent to each sub-task, the agent automatically executes the sub-task according to its task attribute information, thus improving the accuracy of power operation and maintenance.

[0115] In some embodiments, after determining the power operation and maintenance sub-tasks to be executed by each agent, the method further includes: for each agent, generating a corresponding set of executable nodes based on the power operation and maintenance sub-tasks to be executed, wherein each executable node in the set represents the power operation and maintenance sub-task to be executed by that agent. The server converts the set of executable nodes for each agent and the set of dependency edges in the instantiated task graph into an actual execution queue and executes it. At the same time, the entire process is archived as a node-level structured entry sequence, which includes a node identifier, executing agent, timestamp, and observation summary.

[0116] For example, in the instantiated task graph obtained above With the dispatch matrix Then, the set of executable nodes and dependencies for each agent are transformed into an actual execution queue and executed, while the entire process is archived in a structured trajectory format. The node set in the instantiated task graph. For parameters that have been bound and aligned. ; Used to specify the dependency topology; This is used to assign each node (power operation and maintenance subtask) to a specific agent. Then, for each agent... ,according to Extract its set of responsible nodes This involves obtaining the set of executable nodes for agent j. To avoid stacking in the same area, risk concentration, and electrical interlocking conflicts, while also considering the topology depth of the template graph, when selecting the next node from the executable set each time, the remaining nodes in the executable set (i.e., candidate nodes) are considered. Calculate priority score Those with higher scores join the team first, as detailed in formula (9):

[0117] (9)

[0118] in, For the reason Computational topology hierarchy (source to) (length of the longest acyclic path). For intelligent agents Recently joined the team and Overlap score in work area / segment / tower grouping; Risk level of the node; For intelligent agents Recent high-risk node counts in the queue; This is an interlocking suppression term, indicating that if and When there are electrical interlocks or permission exclusivity relationships between nodes that have joined the team but have not yet been completed, and The intensity of the conflict between them (by The interlocking tag in the node attributes and (The order of events is calculated and scaled to obtain the result). The time window relaxation penalty is used to represent... The earliest executable time (by all its predecessors) The time window is the ratio of the estimated end times accumulated above to the allowed upper bound of the time window (the time window is obtained by summing the estimated end times above). The constraint fields are given, and the relevant quantities are range-scaled before use. (Hyperparameters) For dimensionless weights, the default values ​​are taken separately. Recommended range is , , , , For intelligent agents ,according to Iteratively generate topologically feasible and security-constraint-sensitive execution sequences. Then, agent j can sequentially execute the required power operation and maintenance sub-tasks based on the execution sequence.

[0119] Then, when the agent executes each power operation and maintenance subtask, the instantaneous state of the execution of the power operation and maintenance subtask can be archived, and can be formalized according to the following formula (10):

[0120] (10)

[0121] in, For nodes Read-only fingerprints, that is, for The node identifier is hashed; j is the agent number that executes the node; and These are the timestamps generated by the scheduler when issuing and receiving the response, respectively. The node's execution status; This is for observation summaries (e.g., visual nodes store image fingerprints, state nodes store summaries of read values, and a concise representation of the rule comparison results). Among them, These can be considered as the instantaneous states of the power operation and maintenance sub-tasks when executed by the intelligent agent j.

[0122] In a specific embodiment, the specific execution steps of the power operation and maintenance method are as follows:

[0123] First, the server acquires scene images sent by the image acquisition device. Based on an image feature extraction model, it extracts features from the key inspection areas within the scene images to obtain a scene vector. Next, the server acquires maintenance text sent by the inspection task client, performs word segmentation on the text, and obtains the segmentation results. Based on the word segmentation results, it determines the text vector of the maintenance text using a text feature extraction model. Finally, the server acquires equipment status data reported by devices in the power maintenance scenario. This equipment status data includes the status values ​​of multiple status dimensions. Using a status feature extraction model, it extracts features from the equipment status data to obtain a status vector.

[0124] Secondly, the server determines the concatenation order of the scene vector, text vector, and state vector respectively; according to the concatenation order of the scene vector, text vector, and state vector, the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data are concatenated to obtain the scene state vector of the power operation and maintenance scenario.

[0125] Next, the server obtains the structured template vector of each task template in the task template library; based on the scenario state vector and the structured template vector of each task template in the task template library, a task template matching function is constructed; with the goal of maximizing the task template matching function, the matching task template is retrieved from the task template library corresponding to the power operation and maintenance scenario.

[0126] Then, the server concatenates the scenario state vector and the structured template vector of the queried task template to obtain the concatenated vector; the concatenated vector is then linearly transformed to obtain the operation and maintenance intent vector corresponding to the queried task template.

[0127] Then, the template number of the retrieved task template is obtained; based on the template number, a matching task graph is retrieved from a preset task graph library, and the matching task graph indicates at least one power operation and maintenance sub-task. Based on the operation and maintenance intent vector, the task attribute information corresponding to at least one power operation and maintenance sub-task is determined.

[0128] Finally, for each power operation and maintenance sub-task, the server queries a matching agent from multiple agents based on the task attribute information corresponding to the power operation and maintenance sub-task, so as to execute the power operation and maintenance sub-task through the queried agent and obtain the corresponding operation and maintenance sub-result; based on the operation and maintenance sub-result of each power operation and maintenance sub-task, the operation and maintenance result for the power operation and maintenance scenario is determined.

[0129] In the above embodiments, scene images, maintenance text, and device status data of equipment in the power operation and maintenance scenario are acquired. The scene vector of the scene image, the text vector of the maintenance text, and the status vector of the device status data are fused to obtain the scene status vector of the power operation and maintenance scenario. That is, by fusing vectors of different dimensions, the scene status vector can cover multimodal details such as images, text, and status data, ensuring the effectiveness of power operation and maintenance. Then, based on the scene status vector, a matching task template is queried from the task template library corresponding to the power operation and maintenance scenario. This automatically determines the maintenance intent vector corresponding to the queried task template, and at least one power operation and maintenance sub-task is determined based on the task template. Based on the maintenance intent vector, the task attribute information corresponding to at least one power operation and maintenance sub-task is determined, without requiring manual confirmation of the task details of the power operation and maintenance sub-task to be executed. For each power operation and maintenance sub-task, based on the task attribute information corresponding to the sub-task, a matching agent is accurately queried from multiple agents. This matched agent then automatically executes the sub-task, yielding the corresponding sub-result. Based on the sub-result of each sub-task, the overall operation and maintenance outcome for the power operation and maintenance scenario is determined. Throughout this process, without human intervention, the operation and maintenance intent vector, reflecting the operation and maintenance intent, is accurately identified based on the scenario state vector containing multimodal details. This allows for the accurate and automatic identification of the task attribute information of the required sub-tasks. After assigning a corresponding agent to each sub-task, the agent automatically executes the sub-task according to its task attribute information, thus improving the accuracy of power operation and maintenance.

[0130] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0131] Based on the same inventive concept, this application also provides a power operation and maintenance device for implementing the power operation and maintenance method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more power operation and maintenance device embodiments provided below can be found in the limitations of the power operation and maintenance method described above, and will not be repeated here.

[0132] In one exemplary embodiment, such as Figure 3 As shown, a power operation and maintenance device 300 is provided, including: a data acquisition module 302, a vector fusion module 304, a task determination module 306, a task execution module 308, and an operation and maintenance result determination module 310, wherein:

[0133] The data acquisition module 302 is used to acquire scene images, maintenance text, and equipment status data of devices in the power operation and maintenance scenario.

[0134] The vector fusion module 304 is used to fuse the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data to obtain the scene state vector of the power operation and maintenance scenario.

[0135] The task determination module 306 is used to query a matching task template from the task template library corresponding to the power operation and maintenance scenario based on the scenario state vector, determine the operation and maintenance intent vector corresponding to the queried task template, determine at least one power operation and maintenance sub-task based on the task template, and determine the task attribute information corresponding to at least one power operation and maintenance sub-task based on the operation and maintenance intent vector.

[0136] The task execution module 308 is used to query a matching intelligent agent from multiple intelligent agents for each power operation and maintenance sub-task based on the task attribute information corresponding to the power operation and maintenance sub-task, so as to execute the power operation and maintenance sub-task through the queried intelligent agent and obtain the corresponding operation and maintenance sub-result.

[0137] The operation and maintenance result determination module 310 is used to determine the operation and maintenance result for the power operation and maintenance scenario based on the operation and maintenance sub-results of each power operation and maintenance sub-task.

[0138] In some embodiments, the data acquisition module 302 is used to acquire scene images sent by an image acquisition device; extract features from the region images of key inspection areas in the scene images according to an image feature extraction model to obtain a scene vector of the scene images; the data acquisition module 302 is used to acquire maintenance text sent by an inspection task client, perform word segmentation on the maintenance text to obtain word segmentation results; based on the word segmentation results, determine the text vector of the maintenance text through a text feature extraction model; the data acquisition module 302 is used to acquire equipment status data reported by equipment in a power maintenance scenario, the equipment status data including the status values ​​of multiple status dimensions; extract features from the equipment status data through a status feature extraction model to obtain a status vector of the equipment status data.

[0139] In some embodiments, the vector fusion module 304 is used to determine the concatenation order of the scene vector, text vector, and state vector respectively; and to concatenate the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data according to the concatenation order of the scene vector, text vector, and state vector to obtain the scene state vector of the power operation and maintenance scenario.

[0140] In some embodiments, the task determination module 306 is used to obtain the structured template vector of each task template in the task template library; construct a task template matching function based on the scenario state vector and the structured template vector of each task template in the task template library; and query the matching task template from the task template library corresponding to the power operation and maintenance scenario with the goal of maximizing the task template matching function.

[0141] In some embodiments, the task determination module 306 is used to concatenate the scene state vector and the structured template vector of the queried task template to obtain a concatenated vector; and to perform a linear transformation on the concatenated vector to obtain the operation and maintenance intent vector corresponding to the queried task template.

[0142] In some embodiments, the task determination module 306 is used to obtain the template number of the queried task template; based on the template number, to query a matching task map from a preset task map library, wherein the matching task map indicates at least one power operation and maintenance sub-task.

[0143] Each module in the aforementioned power operation and maintenance device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0144] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows.Figure 4 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media to run. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a power operation and maintenance method.

[0145] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0146] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: acquiring a scene image of a power operation and maintenance scenario, operation and maintenance text, and device status data of devices in the power operation and maintenance scenario; fusing the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data to obtain a scene state vector of the power operation and maintenance scenario; based on the scene state vector, querying a matching task template from a task template library corresponding to the power operation and maintenance scenario, determining an operation and maintenance intent vector corresponding to the queried task template, and determining at least one power operation and maintenance sub-task based on the task template, and determining task attribute information corresponding to at least one power operation and maintenance sub-task based on the operation and maintenance intent vector; for each power operation and maintenance sub-task, querying a matching intelligent agent from multiple intelligent agents based on the task attribute information corresponding to the power operation and maintenance sub-task, executing the power operation and maintenance sub-task through the queried intelligent agent, and obtaining a corresponding operation and maintenance sub-result; and determining the operation and maintenance result for the power operation and maintenance scenario based on the operation and maintenance sub-result of each power operation and maintenance sub-task.

[0147] In one embodiment, when the processor executes the computer program, it further performs the following steps: acquiring a scene image sent by an image acquisition device; extracting features from the region image of the key inspection area in the scene image according to an image feature extraction model to obtain a scene vector of the scene image; acquiring maintenance text sent by an inspection task client, performing word segmentation on the maintenance text to obtain a word segmentation result; determining the text vector of the maintenance text based on the word segmentation result through a text feature extraction model; acquiring equipment status data reported by equipment in the power maintenance scenario, the equipment status data including the status values ​​of multiple status dimensions; and extracting features from the equipment status data through a status feature extraction model to obtain a status vector of the equipment status data.

[0148] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining the concatenation order of the scene vector, text vector, and state vector respectively; concatenating the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data according to the concatenation order of the scene vector, text vector, and state vector to obtain the scene state vector of the power operation and maintenance scene.

[0149] In one embodiment, when the processor executes the computer program, it further performs the following steps: obtaining the structured template vector of each task template in the task template library; constructing a task template matching function based on the scenario state vector and the structured template vector of each task template in the task template library; and querying the matching task template from the task template library corresponding to the power operation and maintenance scenario with the goal of maximizing the task template matching function.

[0150] In one embodiment, when the processor executes the computer program, it further performs the following steps: concatenating the scene state vector and the structured template vector of the queried task template to obtain a concatenated vector; and performing a linear transformation on the concatenated vector to obtain the operation and maintenance intent vector corresponding to the queried task template.

[0151] In one embodiment, when the processor executes the computer program, it further performs the following steps: obtaining the template number of the queried task template; and based on the template number, querying a matching task graph from a preset task graph library, wherein the matching task graph indicates at least one power operation and maintenance sub-task.

[0152] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it performs the following steps: acquiring scene images, maintenance text, and device status data of equipment in a power operation and maintenance scenario; fusing scene vectors from the scene images, text vectors from the maintenance text, and state vectors from the device status data to obtain a scene state vector for the power operation and maintenance scenario; based on the scene state vector, querying a matching task template from a task template library corresponding to the power operation and maintenance scenario, determining the maintenance intent vector corresponding to the queried task template, and determining at least one power operation and maintenance sub-task based on the task template; and determining task attribute information corresponding to at least one power operation and maintenance sub-task based on the maintenance intent vector; for each power operation and maintenance sub-task, querying a matching intelligent agent from multiple intelligent agents based on the task attribute information corresponding to the power operation and maintenance sub-task, executing the power operation and maintenance sub-task through the queried intelligent agent, and obtaining the corresponding maintenance sub-result; and determining the operation and maintenance result for the power operation and maintenance scenario based on the maintenance sub-result of each power operation and maintenance sub-task.

[0153] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: acquiring a scene image sent by an image acquisition device; extracting features from the region image of the key inspection area in the scene image according to an image feature extraction model to obtain a scene vector of the scene image; acquiring maintenance text sent by an inspection task client, performing word segmentation on the maintenance text to obtain a word segmentation result; determining the text vector of the maintenance text based on the word segmentation result through a text feature extraction model; acquiring equipment status data reported by equipment in the power maintenance scenario, the equipment status data including the status values ​​of multiple status dimensions; and extracting features from the equipment status data through a status feature extraction model to obtain a status vector of the equipment status data.

[0154] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the concatenation order of the scene vector, text vector, and state vector respectively; concatenating the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data according to the concatenation order of the scene vector, text vector, and state vector to obtain the scene state vector of the power operation and maintenance scene.

[0155] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining the structured template vector of each task template in the task template library; constructing a task template matching function based on the scenario state vector and the structured template vector of each task template in the task template library; and querying the matching task template from the task template library corresponding to the power operation and maintenance scenario with the goal of maximizing the task template matching function.

[0156] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: concatenating the scene state vector and the structured template vector of the queried task template to obtain a concatenated vector; and performing a linear transformation on the concatenated vector to obtain the operation and maintenance intent vector corresponding to the queried task template.

[0157] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining the template number of the queried task template; and based on the template number, querying a matching task graph from a preset task graph library, wherein the matching task graph indicates at least one power operation and maintenance sub-task.

[0158] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps: acquiring scene images, maintenance text, and device status data of equipment in a power operation and maintenance scenario; fusing scene vectors from the scene images, text vectors from the maintenance text, and state vectors from the device status data to obtain a scene state vector for the power operation and maintenance scenario; based on the scene state vector, querying a matching task template from a task template library corresponding to the power operation and maintenance scenario, determining a maintenance intent vector corresponding to the queried task template, and determining at least one power operation and maintenance sub-task based on the task template; and determining task attribute information corresponding to at least one power operation and maintenance sub-task based on the maintenance intent vector; for each power operation and maintenance sub-task, querying a matching intelligent agent from multiple intelligent agents based on the task attribute information corresponding to the power operation and maintenance sub-task, executing the power operation and maintenance sub-task through the queried intelligent agent, and obtaining a corresponding maintenance sub-result; and determining the operation and maintenance result for the power operation and maintenance scenario based on the maintenance sub-result of each power operation and maintenance sub-task.

[0159] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: acquiring a scene image sent by an image acquisition device; extracting features from the region image of the key inspection area in the scene image according to an image feature extraction model to obtain a scene vector of the scene image; acquiring maintenance text sent by an inspection task client, performing word segmentation on the maintenance text to obtain a word segmentation result; determining the text vector of the maintenance text based on the word segmentation result through a text feature extraction model; acquiring equipment status data reported by equipment in the power maintenance scenario, the equipment status data including the status values ​​of multiple status dimensions; and extracting features from the equipment status data through a status feature extraction model to obtain a status vector of the equipment status data.

[0160] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the concatenation order of the scene vector, text vector, and state vector respectively; concatenating the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data according to the concatenation order of the scene vector, text vector, and state vector to obtain the scene state vector of the power operation and maintenance scene.

[0161] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining the structured template vector of each task template in the task template library; constructing a task template matching function based on the scenario state vector and the structured template vector of each task template in the task template library; and querying the matching task template from the task template library corresponding to the power operation and maintenance scenario with the goal of maximizing the task template matching function.

[0162] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: concatenating the scene state vector and the structured template vector of the queried task template to obtain a concatenated vector; and performing a linear transformation on the concatenated vector to obtain the operation and maintenance intent vector corresponding to the queried task template.

[0163] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining the template number of the queried task template; and based on the template number, querying a matching task graph from a preset task graph library, wherein the matching task graph indicates at least one power operation and maintenance sub-task.

[0164] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0165] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0166] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0167] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A power operation and maintenance method, characterized in that, The method includes: Acquire scene images, maintenance text, and device status data of equipment in the power operation and maintenance scenario; By fusing the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data, the scene state vector of the power operation and maintenance scenario is obtained. Based on the scenario state vector, a matching task template is queried from the task template library corresponding to the power operation and maintenance scenario, and the operation and maintenance intent vector corresponding to the queried task template is determined. Based on the task template, at least one power operation and maintenance sub-task is determined, and based on the operation and maintenance intent vector, the task attribute information corresponding to at least one power operation and maintenance sub-task is determined. For each power operation and maintenance sub-task, based on the task attribute information corresponding to the power operation and maintenance sub-task, a matching intelligent agent is queried from multiple intelligent agents, and the power operation and maintenance sub-task is executed by the queried intelligent agent to obtain the corresponding operation and maintenance sub-result; Based on the operation and maintenance sub-results of each power operation and maintenance sub-task, the operation and maintenance results for the power operation and maintenance scenario are determined.

2. The method according to claim 1, characterized in that, The scene vector determination step for the scene image includes: Acquire scene images sent by the image acquisition device; Based on the image feature extraction model, feature extraction is performed on the region image of the key inspection area in the scene image to obtain the scene vector of the scene image; The steps for determining the text vector of the operation and maintenance text include: Obtain the maintenance text sent by the inspection task client, perform word segmentation on the maintenance text, and obtain the word segmentation result; Based on the word segmentation results, the text vector of the operation and maintenance text is determined through a text feature extraction model; The steps for determining the state vector of the device status data include: Acquire device status data reported by devices in power operation and maintenance scenarios, wherein the device status data includes the status values ​​of multiple status dimensions; The device state data is extracted using a state feature extraction model to obtain a state vector.

3. The method according to claim 1, characterized in that, The scene state vector of the power operation and maintenance scenario is obtained by fusing the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data, including: Determine the concatenation order of the scene vector, text vector, and state vector respectively; The scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data are concatenated in the order of scene vector, text vector, and state vector to obtain the scene state vector of the power operation and maintenance scenario.

4. The method according to claim 1, characterized in that, The step of retrieving a matching task template from the task template library corresponding to the power operation and maintenance scenario based on the scenario state vector includes: Obtain the structured template vector for each task template in the task template library; Based on the scene state vector and the structured template vector of each task template in the task template library, a task template matching function is constructed; With the maximization of the task template matching function as the query target, the matching task template is retrieved from the task template library corresponding to the power operation and maintenance scenario.

5. The method according to claim 1, characterized in that, The determined operation and maintenance intent vector corresponding to the queried task template includes: The concatenated vector is obtained by concatenating the scene state vector and the structured template vector of the queried task template. Perform a linear transformation on the concatenated vector to obtain the operation and maintenance intent vector corresponding to the queried task template.

6. The method according to claim 1, characterized in that, The step of determining at least one power operation and maintenance sub-task based on the task template includes: Retrieve the template number of the retrieved task template; Based on the template number, a matching task map is retrieved from a preset task map library, and the matching task map indicates at least one power operation and maintenance sub-task.

7. A power operation and maintenance device, characterized in that, The device includes: The data acquisition module is used to acquire scene images, maintenance text, and equipment status data of devices in the power operation and maintenance scenario. The vector fusion module is used to fuse the scene vector of the scene image, the text vector of the operation and maintenance text, and the state vector of the device status data to obtain the scene state vector of the power operation and maintenance scenario. The task determination module is used to query a matching task template from the task template library corresponding to the power operation and maintenance scenario based on the scenario state vector, determine the operation and maintenance intent vector corresponding to the queried task template, determine at least one power operation and maintenance sub-task based on the task template, and determine the task attribute information corresponding to at least one power operation and maintenance sub-task based on the operation and maintenance intent vector. The task execution module is used to query a matching intelligent agent from multiple intelligent agents for each power operation and maintenance sub-task based on the task attribute information corresponding to the power operation and maintenance sub-task, so as to execute the power operation and maintenance sub-task through the queried intelligent agent and obtain the corresponding operation and maintenance sub-result; The operation and maintenance result determination module is used to determine the operation and maintenance result for the power operation and maintenance scenario based on the operation and maintenance sub-results of each power operation and maintenance sub-task.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images