Operation and maintenance guidance method and device based on fused graph neural network

Abstract

This application discloses an operation and maintenance guidance method and apparatus based on a fused graph neural network, relating to the field of intelligent operation and maintenance technology. The disclosed operation and maintenance guidance method based on a fused graph neural network includes: constructing a digital twin model of a target building based on operation and maintenance work order information and determining extreme environment scenario data based on the operation and maintenance work order information and the target building digital twin model; constructing an initial fused graph neural network based on the target building digital twin model and the extreme environment scenario data; inputting the current location of maintenance personnel, environmental data, and historical operation and maintenance records into the initial fused graph neural network to obtain a target operation and maintenance path; and generating navigation instructions based on the target operation and maintenance path to complete the operation and maintenance guidance based on the fused graph neural network. This achieves accurate planning of operation and maintenance paths and generation of targeted navigation instructions in extreme environments, effectively ensuring the safety of maintenance personnel, improving the efficiency of operation and maintenance decision-making, and enhancing the rationality of path planning.

Description

Technical Field

[0001] This application relates to the field of intelligent operation and maintenance technology, and in particular to an operation and maintenance guidance method and device based on fused graph neural networks. Background Technology

[0002] Traditional intelligent operation and maintenance (O&M) solutions are typically based on static models. These models, or pre-built 3D models or 2D drawings, are invoked after a work order is triggered to provide maintenance personnel with basic equipment location and path information. While this approach can guide personnel in general environments, it fails to reflect environmental changes in real time when sudden events such as high-temperature leaks, dust blockages, or electromagnetic interference occur. This forces O&M personnel to rely on personal experience for judgment, reducing efficiency and significantly increasing safety risks. The main drawback of traditional solutions is the lack of real-time perception and decision support capabilities for dynamic and extreme environments, making it difficult to provide safe and efficient navigation and data assistance to O&M personnel under drastically changing conditions.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide an operation and maintenance guidance method and device based on fused graph neural networks, which aims to solve the technical problem of how to ensure that operation and maintenance personnel can safely and efficiently complete operation navigation and decision-making in extreme environments when the path is dynamically blocked and the environmental information is incomplete.

[0005] To achieve the above objectives, this application proposes an operation and maintenance guidance method based on a fused graph neural network, the method comprising:

[0006] A digital twin model of the target building is constructed based on the maintenance work order information, and extreme environment scenario data is determined based on the maintenance work order information and the digital twin model of the target building.

[0007] Based on the target building digital twin model and the extreme environment scenario data, an initial fusion graph neural network is constructed;

[0008] The current location of maintenance personnel, environmental data, and historical maintenance records are input into the initial fusion graph neural network to obtain the target maintenance path;

[0009] Navigation instructions are generated based on the target operation and maintenance path to complete operation and maintenance guidance based on a fused graph neural network.

[0010] In addition, to achieve the above objectives, this application also proposes an operation and maintenance guidance device based on a fused graph neural network. The operation and maintenance guidance device based on a fused graph neural network includes: a data determination module, used to construct a digital twin model of the target building based on operation and maintenance work order information and determine extreme environment scenario data based on the operation and maintenance work order information and the digital twin model of the target building.

[0011] The network construction module is used to construct an initial fusion graph neural network based on the target building digital twin model and the extreme environment scenario data;

[0012] The path generation module is used to input the current location of maintenance personnel, environmental data, and historical operation and maintenance records into the initial fusion graph neural network to obtain the target operation and maintenance path;

[0013] The operation and maintenance guidance module is used to generate navigation instructions based on the target operation and maintenance path to complete the operation and maintenance guidance based on the fused graph neural network.

[0014] Furthermore, to achieve the above objectives, this application also proposes an operation and maintenance guidance device based on a fused graph neural network. The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the operation and maintenance guidance method based on the fused graph neural network described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the operation and maintenance guidance method based on the fused graph neural network described above.

[0016] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the operation and maintenance guidance method based on a fused graph neural network as described above.

[0017] One or more technical solutions proposed in this application have at least the following technical effects:

[0018] By employing a technical approach that involves constructing a digital twin model of the target building based on maintenance work order information and determining extreme environment scenario data, combining this model with scenario data to build an initial fusion graph neural network, and then inputting the current location of maintenance personnel, environmental data, and historical maintenance records into the network to obtain the target maintenance path, and finally generating navigation instructions based on the path, this approach solves the problem of existing intelligent maintenance systems lacking real-time safety assurance and accurate data support for maintenance personnel in extreme environments, leading to difficulties in navigation, slow decision-making, and high safety risks for maintenance personnel. Compared with existing technologies that rely solely on static building information model geometric data or two-dimensional drawings and lack dynamic environment adaptation and intelligent path planning, this approach achieves accurate mapping of extreme environment scenarios through digital twin models, intelligent path planning by integrating multi-dimensional dynamic data through fusion graph neural networks, and precise implementation of maintenance guidance through navigation instructions. Ultimately, this approach achieves real-time assurance of maintenance personnel's operational safety in extreme environments, dynamic adaptation of maintenance path planning, and efficient and accurate maintenance guidance, reducing reliance on the personal experience of maintenance personnel. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating an embodiment of the operation and maintenance guidance method based on a fused graph neural network provided in this application.

[0022] Figure 2 This is a flowchart illustrating Embodiment 2 of the operation and maintenance guidance method based on fused graph neural networks in this application.

[0023] Figure 3 This is a simplified flowchart illustrating the operation and maintenance guidance method based on a fused graph neural network provided in Embodiment 2 of this application.

[0024] Figure 4 This is a schematic diagram of the module structure of the operation and maintenance guidance device based on the fused graph neural network according to an embodiment of this application;

[0025] Figure 5 This is a schematic diagram of the device structure of the hardware operating environment involved in the operation and maintenance guidance method based on fused graph neural network in the embodiments of this application.

[0026] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0027] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0028] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0029] The main solution of this application embodiment is as follows: construct a digital twin model of the target building based on the maintenance work order information, and determine extreme environment scenario data based on the maintenance work order information and the target building digital twin model; construct an initial fusion graph neural network based on the target building digital twin model and the extreme environment scenario data; input the current maintenance personnel location, environmental data, and historical maintenance records into the initial fusion graph neural network to obtain the target maintenance path; generate navigation instructions based on the target maintenance path to complete maintenance guidance based on the fusion graph neural network.

[0030] In this embodiment, for ease of description, the following description will focus on identifying the operation and maintenance guidance device based on the fused graph neural network.

[0031] To address the challenges of ensuring safe and efficient navigation and decision-making for maintenance personnel in extreme environments, particularly when paths are dynamically obstructed or environmental information is incomplete, this application provides a solution. This solution utilizes a technical approach: constructing a digital twin model of the target building based on maintenance work order information and determining extreme environment scenario data. This model is then combined with the scenario data to construct an initial fusion graph neural network. The current location of maintenance personnel, environmental data, and historical maintenance records are input into the network to obtain the target maintenance path. Finally, navigation instructions are generated based on this path. This approach solves the problem of existing intelligent maintenance systems lacking real-time safety assurance and accurate data support for maintenance personnel in extreme environments, leading to difficulties in navigation, slow decision-making, and high safety risks. Compared to existing technologies that rely solely on static building information model geometric data or two-dimensional drawings and lack dynamic environment adaptation and intelligent path planning, this solution achieves accurate mapping of extreme environment scenarios through a digital twin model, intelligent path planning by integrating multi-dimensional dynamic data through a fusion graph neural network, and precise implementation of maintenance guidance through navigation instructions. Ultimately, this achieves real-time assurance of maintenance personnel's operational safety in extreme environments, dynamic adaptation of maintenance path planning, and efficient and accurate maintenance guidance, reducing reliance on the personal experience of maintenance personnel.

[0032] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as an operation and maintenance guidance device based on a fused graph neural network. The following description uses an operation and maintenance guidance device based on a fused graph neural network as an example to illustrate this embodiment and the subsequent embodiments.

[0033] Based on this, embodiments of this application provide an operation and maintenance guidance method based on a fused graph neural network, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the operation and maintenance guidance method based on fused graph neural networks in this application.

[0034] In this embodiment, the operation and maintenance guidance method based on fused graph neural network includes steps S10~S40:

[0035] Step S10: Construct a digital twin model of the target building based on the maintenance work order information and determine extreme environment scenario data based on the maintenance work order information and the digital twin model of the target building;

[0036] It should be noted that maintenance work order information is a data set that records the key content of maintenance tasks. It usually includes information such as faulty device identification, fault type, and maintenance area. This information is the basis for starting the maintenance process and determining maintenance objectives.

[0037] In addition, the target building digital twin model is a digital model built based on the actual structure and equipment information of the building. It can map the physical spatial structure of the building, the three-dimensional coordinates of the equipment, the equipment relationship, etc., providing visualization and data support for subsequent operation and maintenance analysis.

[0038] Furthermore, extreme environment scenario data describes the characteristics of extreme environments within the operation and maintenance area, including the deployment locations of environmental sensors, the distribution of risk areas, and environmental parameter thresholds. This data is used to reflect the specific conditions of extreme environments such as high temperature and high dust.

[0039] Understandably, the faulty equipment identifier, fault type, and maintenance area range are extracted from the maintenance work order information. Based on this extracted information, the sub-model corresponding to the maintenance area range in the building's overall digital twin model is retrieved to supplement the detailed 3D model data of the faulty equipment and the associated equipment model data, thereby constructing the target building's digital twin model. Based on the target building's digital twin model, the specific location area of ​​the faulty equipment in the model is located, i.e., the model location area.

[0040] Furthermore, by querying historical and real-time environmental data of the model's location area, the extreme environmental types of the area are determined, such as high temperature, high dust, strong electromagnetic fields, and low temperature. Combining the spatial structure data of the target building's digital twin model with the determined extreme environmental types, the distribution patterns of environmental parameters and the locations of risk points are analyzed. Historical and real-time data of environmental sensors in the area are retrieved to determine the safety and risk thresholds of environmental parameters. Finally, extreme environmental scenario data containing sensor locations, risk area coordinates, and environmental parameter thresholds are obtained.

[0041] In one feasible implementation, step S10 may include steps S11 to S15:

[0042] Step S11: Extract the faulty device identifier, fault type, and maintenance area from the maintenance work order information;

[0043] It should be noted that the faulty equipment identifier is a code or identification information used to uniquely identify the faulty equipment. This identifier can accurately locate the specific equipment that needs to be repaired, avoiding confusion between maintenance tasks for different equipment.

[0044] Furthermore, the fault type is a classification description of the problem that occurs in the faulty equipment, such as electrical faults, mechanical faults, etc. Different fault types correspond to different maintenance strategies and data requirements, providing direction for subsequent operation and maintenance preparation.

[0045] Understandably, the maintenance area is the spatial range around the faulty equipment where maintenance work needs to be carried out. It is usually a certain radius area centered on the faulty equipment to clarify the spatial boundaries of the maintenance work. The three key pieces of information, namely the faulty equipment identifier, the fault type, and the maintenance area, are extracted one by one from the maintenance work order information to provide core input data for the subsequent construction of the target building digital twin model.

[0046] Step S12: Construct a digital twin model of the target building based on the faulty equipment identifier, the fault type, and the maintenance area.

[0047] It should be noted that the target building digital twin model contains digital information such as the three-dimensional coordinates of the faulty equipment, its spatial relationship, and the building structure. This information is the basis for locating the faulty equipment model's location area.

[0048] Understandably, the operation and maintenance area is used to select sub-models related to operation and maintenance tasks from the building-wide digital twin model, avoiding loading redundant data across the entire domain and improving model building efficiency; based on the extracted faulty equipment identifier, fault type, and operation and maintenance area, the sub-model corresponding to the operation and maintenance area in the building-wide digital twin model is first retrieved.

[0049] Furthermore, based on the faulty equipment identifier, the basic model of the faulty equipment is located in the sub-model, and then detailed three-dimensional model data of the faulty equipment is supplemented according to the fault type, such as internal component structure, key parameter annotation, etc. At the same time, the model data of related equipment associated with the faulty equipment is supplemented, and finally the digital twin model of the target building is completed.

[0050] By combining core operation and maintenance information to construct a digital twin model of the target building, we can focus on the specific areas and equipment required for operation and maintenance. This ensures that the model contains complete spatial and equipment association information without redundant data interference, providing an accurate digital carrier for subsequent location of faulty equipment and analysis of extreme environments.

[0051] Step S13: Locate the model location area of ​​the faulty equipment in the maintenance work order information based on the target building digital twin model;

[0052] Understandably, in the target building's digital twin model, the corresponding 3D model of the faulty equipment is found based on the faulty equipment identifier; with this 3D model as the core, and combined with the spatial range required for maintenance operations in the operation and maintenance scenario, an area containing the faulty equipment and its surrounding related space is delineated, and this area is the model location area of ​​the faulty equipment.

[0053] By accurately locating the model location of faulty equipment using a digital twin model of the target building, the scope of operation and maintenance analysis can be focused on the space directly related to the faulty equipment, avoiding invalid analysis of irrelevant areas, and providing a clear spatial range for subsequent accurate determination of extreme environment types and analysis of environmental parameters.

[0054] Step S14: Determine the extreme environment type of the model location region;

[0055] It should be noted that extreme environment types indicate the types of environmental parameters that need to be analyzed in detail. For example, high-temperature environments require analysis of temperature parameters, and high-dust environments require analysis of dust concentration parameters, ensuring that the analysis focuses on key risk factors.

[0056] Understandably, extreme environment scenario data is a dataset describing the extreme environmental characteristics of the model's location area, including sensor locations, risk area coordinates, environmental parameter thresholds, etc., used to support the subsequent construction of the initial fusion graph neural network and the planning of operation and maintenance paths; combined with the spatial structure data of the target building's digital twin model, the distribution patterns of environmental parameters corresponding to extreme environment types in the model's location area are analyzed, such as the trend of temperature spreading outward along the equipment's heat dissipation vents in high-temperature environments, and the corner locations where dust easily accumulates in high-dust environments.

[0057] Step S15: Analyze environmental parameters based on the target building digital twin model and the extreme environment type to obtain extreme environment scenario data.

[0058] Understandably, extreme environment scenario data is a dataset describing the extreme environmental characteristics of the model's location area, including sensor locations, risk area coordinates, environmental parameter thresholds, etc., used to support the subsequent construction of the initial fusion graph neural network and the planning of operation and maintenance paths; combined with the spatial structure data of the target building's digital twin model, the distribution patterns of environmental parameters corresponding to extreme environment types in the model's location area are analyzed, such as the trend of temperature spreading outward along the equipment's heat dissipation vents in high-temperature environments, and the corner locations where dust easily accumulates in high-dust environments.

[0059] Furthermore, based on the distribution patterns, the locations of risk points within the area are determined, such as high-temperature risk points and dust accumulation risk points. At the same time, historical and real-time data from environmental sensors within the model's location area are retrieved, and combined with operational safety standards, the safety and risk thresholds for each environmental parameter are determined. By integrating information such as sensor locations, risk area coordinates, and environmental parameter thresholds, extreme environmental scenario data is finally obtained.

[0060] Based on the analysis of environmental parameters using the target building's digital twin model and extreme environment types, we can accurately obtain extreme environmental characteristic data of the model's location area. This provides comprehensive and realistic environmental data support for defining nodes and edges when constructing the initial fusion graph neural network and assessing risk costs when planning the target operation and maintenance path.

[0061] Step S20: Construct an initial fusion graph neural network based on the target building digital twin model and the extreme environment scenario data;

[0062] It should be noted that the initial fusion graph neural network is the initial model of fusion graph neural network technology. It integrates multi-dimensional data such as buildings, environment, and equipment in the form of a graph structure, and has the ability to extract features and make decisions from multi-source data. It is the core model for generating target operation and maintenance paths.

[0063] Additionally, the geometric data of the building information model is the data describing the spatial form of the building and equipment in the digital twin model of the target building. It includes the three-dimensional coordinates of the equipment, the dimensions of the building passages, the location of the walls, etc., and is used to build the spatial foundation for the initial fusion graph neural network.

[0064] Furthermore, non-geometric data in Building Information Modeling (BIM) refers to the data describing the non-spatial attributes of buildings and equipment in the target building's digital twin model. This includes information such as equipment materials, service life, and maintenance cycles, and is used to supplement the attribute information of the initial fusion graph neural network.

[0065] Understandably, device association data describes the connection relationships between devices, including power supply relationships, data flow transmission relationships, etc., and is used to reflect the logical relationships between device nodes in the initial fusion graph neural network; environmental sensor deployment location is information recorded in the extreme environment scenario data about the installation location of environmental sensors, usually presented in three-dimensional coordinates, and is used to determine the location of sensor nodes in the initial fusion graph neural network.

[0066] Furthermore, the risk area distribution data is information describing the range of high-risk areas in extreme environment scenario data, including the boundary coordinates and risk levels of areas such as high temperature areas and high dust areas, which is used to construct virtual risk nodes in the initial fusion graph neural network. The initial fusion graph node is the basic building block of the initial fusion graph neural network, including maintenance personnel starting candidate nodes, target equipment nodes, sensor nodes, and virtual risk nodes. Each node carries the feature information of the corresponding entity or region.

[0067] It is understandable that the initial fusion graph edges are the elements connecting the nodes of the initial fusion graph, including physical access edges and device logical association edges. Physical access edges are used to represent the passable physical paths, and device logical association edges are used to mark the direction of energy flow or data flow between devices. Geometric data of building information model, non-geometric data of building information model and device association data are extracted from the target building digital twin model. Environmental sensor deployment location and risk area distribution data are extracted from extreme environment scenario data.

[0068] Furthermore, based on the extracted geometric data, non-geometric data, equipment association data, environmental sensor deployment locations, and risk area distribution data of the building information model, initial fusion graph nodes and edges are defined; static and dynamic attributes are set for physical access edges, with static attributes including path length and basic access cost, and dynamic attributes used to associate environmental data calculation results; and the direction of energy flow or data flow between equipment is marked for logically associated edges; based on the defined initial fusion graph nodes and edges, the initial fusion graph neural network is constructed.

[0069] In one feasible implementation, step S20 may include steps S21 to S24:

[0070] Step S21: Based on the target building digital twin model, determine the geometric data of the building information model, the non-geometric data of the building information model, and the equipment association data;

[0071] It should be noted that the geometric data of the building information model describes the spatial form of the building and equipment, including the three-dimensional coordinates of the equipment, the dimensions of the building passages, the location of the walls, etc. This data is used to define the spatial location of the nodes in the initial fusion map and the physical length of the edges in the initial fusion map.

[0072] Furthermore, non-geometric data in Building Information Modeling (BIM) describes the non-spatial attributes of buildings and equipment, including equipment materials, service life, maintenance cycles, and other information. This data is used to supplement the attribute information of nodes in the initial fusion graph, thereby improving the completeness of node features.

[0073] It is understandable that device association data describes the connection relationships between devices, including power supply relationships, data flow transmission relationships, etc. This data is used to define the logical association edges of devices in the initial fusion graph. Geometric data, non-geometric data, and device association data of the building information model are selected and extracted from the target building digital twin model to provide data support for defining the nodes and edges of the initial fusion graph.

[0074] Step S22: Obtain the deployment location and risk area distribution data of environmental sensors based on the extreme environment scenario data;

[0075] It should be noted that the environmental sensor deployment location records the information of where the environmental sensor is installed in the physical space, usually presented in the form of three-dimensional coordinates. This data is used to define the position of the sensor node in the initial fusion map to ensure that the sensor node is consistent with the actual sensor spatial position.

[0076] Furthermore, the risk area distribution data describes the range of high-risk areas, including the boundary coordinates and risk levels of areas such as high-temperature zones and high-dust zones. This data is used to define the location and risk attributes of virtual risk nodes in the initial fusion map, reflecting the risk distribution of extreme environments.

[0077] It is understandable that data on the deployment locations of environmental sensors and the distribution of risk areas are filtered and extracted from extreme environment scenario data, and redundant data that is irrelevant to the initial fusion graph node and edge definitions is excluded, so as to ensure that the extracted data can directly support the subsequent construction of the initial fusion graph structure.

[0078] Focusing on extracting key environmental data from extreme environment scenario data ensures that the initial fusion graph neural network constructed subsequently contains risk and perception information of extreme environments, enabling the model to adapt to extreme environments and providing environmental data support for the subsequent generation of safe target operation and maintenance paths.

[0079] Step S23: Based on the geometric data of the building information model, the non-geometric data of the building information model, the equipment association data, the deployment location of environmental sensors, and the risk area distribution data, define the initial fusion graph nodes and the initial fusion graph edges. The initial fusion graph nodes include maintenance personnel starting candidate nodes, target equipment nodes, sensor nodes, and virtual risk nodes. The initial fusion graph edges include physical access edges and equipment logical association edges.

[0080] It should be noted that the initial fusion graph node is the basic building block of the initial fusion graph neural network, the maintenance personnel initial candidate node is the node corresponding to the possible initial position of the maintenance personnel, the target equipment node is the node corresponding to the faulty equipment, the sensor node is the node corresponding to the environmental sensor, and the virtual risk node is the node corresponding to the high-risk area.

[0081] Furthermore, the initial fusion graph edges are elements that connect the nodes of the initial fusion graph. Physical access edges are used to represent passable physical paths and connect nodes related to access, such as the initial candidate node for maintenance personnel and the target device node. Device logical association edges are used to mark the direction of energy flow or data flow between devices and connect device nodes that have logical associations.

[0082] Understandably, the spatial coordinates and path lengths of physical access edges of each node are determined based on the geometric data of the Building Information Model (BIM). Attributes such as material and service life are added to the nodes by combining non-geometric data from the BIM. Logical association edges and their logical directions are defined based on the equipment association data. The coordinates of sensor nodes are defined based on the deployment locations of environmental sensors. The coordinates and risk levels of virtual risk nodes are defined based on the risk area distribution data. Finally, the initial fusion graph nodes and edges are defined.

[0083] Step S24: Construct an initial fusion graph neural network based on the initial fusion graph nodes and the initial fusion graph edges.

[0084] Understandably, the attribute information of the initial fusion graph nodes, such as spatial coordinates, material, and risk level, is transformed into node feature vectors to form a node feature matrix; the attribute information of the initial fusion graph edges, such as path length and logical direction, is transformed into edge feature vectors to form an edge feature matrix; the node feature matrix and the edge feature matrix are used as initial inputs to build a network architecture that includes a graph attention module and a decision module, the parameters of each module are configured, and the initial fusion graph neural network is completed.

[0085] Step S30: Input the current location of maintenance personnel, environmental data, and historical operation and maintenance records into the initial fusion graph neural network to obtain the target operation and maintenance path;

[0086] It should be noted that the current location of the maintenance personnel is the real-time spatial location information of the maintenance personnel at the operation and maintenance site. It is usually obtained through Ultra Wide Band (UWB), Bluetooth Low Energy (BLE), or 5th Generation Mobile Communication Technology (5G) and is used to determine the current node in the initial fusion graph neural network.

[0087] Additionally, environmental data reflects the real-time environmental conditions at the operation and maintenance site, including parameters such as temperature, dust concentration, and harmful gas concentration. This data is collected by environmental sensors and used to assess the environmental risks of the route.

[0088] Furthermore, historical operation and maintenance records are data that record information related to past operation and maintenance tasks, including the frequency of path congestion, the time taken to handle faults, etc., which are used to provide historical experience reference for current path planning.

[0089] Understandably, the target maintenance path is the optimal path from the current location of the maintenance personnel to the target equipment node, output by the initial fusion graph neural network. This path comprehensively considers time cost and environmental risk cost, ensuring that the maintenance personnel can reach the target equipment safely and efficiently. The current location of the maintenance personnel is obtained through positioning technology and mapped to the current node in the initial fusion graph neural network. Environmental data collected by environmental sensors are collected, and data such as temperature, dust concentration, and harmful gas concentration are associated with the corresponding sensor nodes in the initial fusion graph neural network.

[0090] Furthermore, historical operation and maintenance records are retrieved, and data such as path congestion frequency and fault handling time are associated with the corresponding physical access edges. The current node information, associated environmental data, and associated historical operation and maintenance records are input into the initial fusion graph neural network. The initial fusion graph neural network performs feature extraction and calculation on the input data to generate multiple candidate paths from the current node to the target device node.

[0091] Understandably, the comprehensive cost of each candidate path is calculated, which includes the time cost based on path length and the risk cost based on environmental data. The candidate path with the lowest comprehensive cost is selected as the target operation and maintenance path.

[0092] Step S40: Generate navigation instructions based on the target operation and maintenance path to complete operation and maintenance guidance based on fused graph neural network.

[0093] It should be noted that navigation instructions are instructions used to guide maintenance personnel to carry out maintenance work according to the target maintenance path. They include first-person perspective path guidance instructions and data pop-up display instructions. These instructions are presented to maintenance personnel through the digital twin mobile terminal.

[0094] Additionally, digital twin mobile devices are terminal devices used by maintenance personnel at the maintenance site, such as industrial tablets or augmented reality (AR) glasses, to receive target maintenance routes, display navigation instructions, and provide feedback on maintenance site information.

[0095] Furthermore, the operation and maintenance guidance based on the fusion graph neural network relies on the target operation and maintenance path and navigation instructions generated by the fusion graph neural network to guide maintenance personnel throughout the operation and maintenance process, aiming to ensure the safety and efficiency of the operation and maintenance process.

[0096] Understandably, the process involves obtaining the target maintenance path from the initial fusion graph neural network output; generating navigation instructions adapted to the digital twin mobile terminal based on the target maintenance path; the first-person perspective path guidance instructions in the navigation instructions are used to prompt maintenance personnel to turn, go straight, or avoid obstacles; and the data pop-up display instructions are used to control the digital twin mobile terminal to display the maintenance data in layers.

[0097] Furthermore, the generated navigation instructions are sent to the digital twin mobile terminal. Maintenance personnel receive and view the navigation instructions through the digital twin mobile terminal, and follow the instructions to go to the target equipment to carry out maintenance work, thereby completing the maintenance guidance based on the fused graph neural network.

[0098] Navigation instructions generated based on the target maintenance path can intuitively provide maintenance personnel with path guidance and data support, avoiding the need for maintenance personnel to rely on personal experience to explore and move forward in extreme environments, reducing safety risks, improving the efficiency and accuracy of maintenance work, and ensuring the practicality and reliability of maintenance guidance.

[0099] In one feasible implementation, step S40 may include steps S41 to S43:

[0100] Step S41: Generate an operation and maintenance perspective based on the target operation and maintenance path and send the operation and maintenance perspective to the digital twin mobile terminal, so that the digital twin mobile terminal can provide feedback on the operation and maintenance site information based on the target operation and maintenance path;

[0101] It should be noted that the operation and maintenance perspective is a visualization angle generated based on the target operation and maintenance path. It is usually a first-person perspective and can simulate the visual effect when maintenance personnel walk along the path, allowing maintenance personnel to intuitively see the path and the surrounding environment.

[0102] Furthermore, digital twin mobile devices are terminal devices used by maintenance personnel at the maintenance site, such as industrial tablets or augmented reality glasses, which have the function of receiving perspective data, collecting on-site information, and providing feedback.

[0103] Understandably, the maintenance site information is data collected and fed back by the digital twin mobile terminal based on the target maintenance path, including the positional deviation data between the maintenance personnel and the target maintenance path, and real-time data collection of on-site environmental parameters. The perspective data generated based on the target maintenance path is sent to the digital twin mobile terminal. After receiving it, the digital twin mobile terminal uses its own positioning module and environmental acquisition module to obtain maintenance site information such as the deviation between the current position and the path of the maintenance personnel and the real-time temperature on site, and feeds this information back to the backend.

[0104] Step S42: Determine the target maintenance data package based on the maintenance site information;

[0105] It should be noted that the target maintenance data package is a targeted data set compiled based on on-site maintenance information and maintenance needs. It includes basic equipment information, fault-related data, risk response tips, etc., and can provide maintenance personnel with accurate on-site decision support.

[0106] Furthermore, determining the target maintenance data package requires considering the specific circumstances of the maintenance site. If the location deviation exceeds the preset value, path correction data is added; if the environmental parameters exceed the threshold, risk warning data is added. At the same time, basic data such as device ID and core parameters are retained.

[0107] Understandably, when receiving maintenance information from the digital twin mobile terminal, the system first analyzes whether the deviation between the maintenance personnel's location and the target maintenance path is within a reasonable range. If the deviation exceeds 1 meter, path correction prompts are added. Then, it analyzes whether the on-site environmental parameters exceed safety thresholds. If the temperature is too high, high-temperature protection and operational precautions are added. Finally, it integrates the device ID, current core parameters, and historical maintenance records related to the current fault to form the target maintenance data package.

[0108] Determining the target maintenance data package based on real-time on-site maintenance information ensures that the data package not only meets the actual needs of the site but also avoids redundant data interference. This allows maintenance personnel to quickly obtain key support information, avoids decision-making delays caused by complex data, and improves maintenance efficiency.

[0109] Step S43: Generate navigation instructions based on the target maintenance data package to complete maintenance guidance based on fused graph neural network.

[0110] Understandably, based on the path correction data and target maintenance path in the target maintenance data package, first-person perspective path guidance instructions are generated, such as "Turn left 5 meters ahead to correct the path" and "Pass through the high-temperature area quickly." Based on the device ID, core parameters, and historical maintenance records in the data package, layered pop-up display instructions are generated, controlling the digital twin mobile terminal to display the device ID and core parameters at the first layer, historical maintenance records at the second layer, and risk warning data at the third layer. The two types of instructions are integrated into a complete navigation instruction and sent to the digital twin mobile terminal. Maintenance personnel follow the instructions to move and operate, ultimately completing the maintenance guidance based on the fused graph neural network.

[0111] Furthermore, the operation and maintenance guidance based on the fusion graph neural network relies on the target operation and maintenance path and navigation instructions generated by the initial fusion graph neural network to guide maintenance personnel throughout the entire operation and maintenance process from start to finish, aiming to ensure safety and efficiency.

[0112] Based on the target maintenance data package, targeted navigation instructions are generated, which can provide maintenance personnel with integrated guidance on path guidance and data support. This avoids maintenance personnel relying on personal experience to figure things out in extreme environments, reduces safety risks, and at the same time ensures that maintenance operations are accurately matched with fault requirements, thereby improving the quality of maintenance task completion.

[0113] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The operation and maintenance guidance method based on fused graph neural networks includes steps S31 to S34 in step S30:

[0114] Step S31: Input the current location of maintenance personnel, environmental data, and historical operation and maintenance records into the initial fusion graph neural network to determine the current node and the physical access edge of the current node;

[0115] It should be noted that the current location of the maintenance personnel is the real-time spatial location information of the maintenance personnel at the operation and maintenance site. It is usually obtained through Ultra Wide Band (UWB), Bluetooth Low Energy (BLE), or 5th Generation Mobile Communication Technology (5G) and is used to locate the corresponding node in the initial fusion graph neural network.

[0116] Additionally, environmental data reflects the real-time environmental conditions at the operation and maintenance site, including parameters such as temperature, dust concentration, and harmful gas concentration. This data is collected by environmental sensors and used to associate sensor nodes and physical access edges in the initial fusion graph neural network.

[0117] Furthermore, historical operation and maintenance records are data that record information related to past operation and maintenance tasks, including path congestion frequency, fault handling time, etc., which are used to supplement the physical access edges in the initial fusion graph neural network with historical attribute information.

[0118] Furthermore, physical access edges are the accessible path edges connecting the current node and surrounding nodes in the initial fusion graph neural network. Their average length and distribution density determine the rationality of the preset hop count range, ensuring that the effective path within the safe field of vision of maintenance personnel can be covered within the hop count range.

[0119] Understandably, the current node is the node in the initial fusion graph neural network that matches the current maintenance personnel's location, and the physical access edge is the edge in the initial fusion graph neural network that connects the current node with surrounding nodes and represents a passable physical path. The current maintenance personnel's location, environmental data, and historical maintenance records are input into the initial fusion graph neural network. The network first maps the current maintenance personnel's location to the corresponding current node, then associates the environmental data with the sensor nodes around the current node and the connected physical access edges, and associates the historical maintenance records with the corresponding physical access edges, finally determining the current node and the physical access edge of the current node.

[0120] In one feasible implementation, step S31 may include steps S311 to S314:

[0121] Step S311: Determine the current temperature, dust concentration, and harmful gas concentration based on environmental data;

[0122] It should be noted that the current temperature is the real-time air or equipment ambient temperature at the maintenance site, used to determine whether the environment is in an extreme high or low temperature state, which directly affects the work safety of maintenance personnel and the operating status of equipment.

[0123] Furthermore, dust concentration is the amount of suspended dust in the air at the maintenance site. High dust concentration can affect visibility and personnel health, and is an important indicator for assessing the risk of path travel. Harmful gas concentration is the amount of gases in the air at the maintenance site that are harmful to the human body or equipment. Exceeding the safety threshold will directly threaten personnel safety and requires close monitoring.

[0124] Understandably, the process involves filtering and extracting three key parameters from the collected environmental data: current temperature, dust concentration, and harmful gas concentration. This removes redundant information irrelevant to risk assessment, providing accurate environmental risk data for determining the current node and physical access edges.

[0125] Step S312: Input the current location of the maintenance personnel, the current temperature, the dust concentration, and the harmful gas concentration into the initial fusion graph neural network to determine the current node;

[0126] It should be noted that the current location of the maintenance personnel is the real-time spatial coordinate of the maintenance personnel at the operation and maintenance site, which is used to find the corresponding spatial location anchor point in the initial fusion graph neural network.

[0127] In addition, the current temperature, dust concentration, and harmful gas concentration are parameters that reflect the real-time risks at the maintenance site. These parameters can help the initial fusion graph neural network to determine the node risk attributes corresponding to the surrounding environment of maintenance personnel, ensuring that the current node matches the actual environmental state.

[0128] Furthermore, the current node is the node in the initial fusion graph neural network that matches the real-time location of the maintenance personnel and the risks of the surrounding environment. This node not only contains spatial coordinate information but also relates to the risk characteristics of the surrounding environment, and is the starting point for subsequent path planning.

[0129] Understandably, the spatial coordinates of the current maintenance personnel's location, along with risk parameters such as the current temperature, dust concentration, and harmful gas concentration, are input into the initial fusion graph neural network. The network initially locates the possible node range through spatial coordinate matching, and then compares the risk parameters with the preset environmental risk attributes of each node to finally determine the current node that perfectly matches the actual location of the maintenance personnel and the surrounding environment.

[0130] Step S313: Determine the path congestion frequency and fault handling time data based on the historical operation and maintenance records;

[0131] It should be noted that the path congestion frequency is a statistical count of the number of times a physical path has been slowed down due to obstruction by personnel or equipment during historical operation and maintenance. It reflects the historical traffic efficiency of the path, and the higher the frequency, the more likely the path is to experience traffic delays.

[0132] Furthermore, fault handling time data is a record of the time spent handling faults on devices around or related to a certain physical path during historical operation and maintenance. It can indirectly reflect the fault occurrence rate and maintenance complexity of devices around the path, and provide experience reference for current path selection.

[0133] Understandably, data related to physical paths is filtered from historical operation and maintenance records, categorized and statistically analyzed according to path identifiers, and the path congestion frequency corresponding to each path is extracted. At the same time, the processing time data of equipment failures around each path is compiled to form historical attribute data for subsequent determination of physical access edges.

[0134] Step S314: Input the path congestion frequency and fault handling time data into the initial fusion graph neural network to determine the physical access edge of the current node.

[0135] It should be noted that the physical access edges of the current node are the edges in the initial fusion graph neural network that connect the current node with other surrounding nodes and represent accessible physical paths. The attributes of these edges not only include real-time environmental risks, but also need to be improved with historical operation and maintenance data to improve efficiency and fault-related attributes.

[0136] Furthermore, the initial fusion graph neural network has data association capabilities, which can match the input historical data with the existing physical access edges in the network, supplement each edge with corresponding historical attributes, thereby clarifying which edges around the current node are the preferred path edges with high historical access efficiency and low fault correlation.

[0137] Understandably, the path congestion frequency and fault handling time data are input into the initial fusion graph neural network. The network matches the path identifiers in the data with the physical access edges around the current node one by one. Attribute labels of path congestion frequency and fault handling time are added to each successfully matched physical access edge. Combining these historical attributes with real-time environmental attributes, the physical access edge of the current node is finally determined.

[0138] Step S32: Determine a preset hop count range based on the physical access edge, and extract features from the subgraph within the preset hop count range of the current node to obtain the target subgraph features;

[0139] It should be noted that the preset hop count range is a range of node hop counts defined around the current node for filtering subgraphs. In this embodiment, it is usually set to 5 hops. This range can cover the area within 50 meters around the current node, which can include enough path selections and avoid the subgraph being too large, thus reducing the computational efficiency.

[0140] Furthermore, a subgraph is a local graph structure within a preset hop count range in the initial fusion graph neural network, containing nodes and edges within that range. The target subgraph features are the extraction and integration results of node attributes, edge attributes, and relationships in the subgraph, which are used to generate candidate paths in the future.

[0141] Furthermore, the target subgraph features include the attributes and association information of nodes and edges within the subgraph. This information is the core basis for the initial fusion graph neural network to generate candidate paths, and can support the feasibility and rationality of the network analysis path.

[0142] Understandably, based on the average length of the physical access edge of the current node and the safety requirements in extreme environments, a preset hop count range is determined; with the current node as the center, all nodes and edges within the preset hop count range are selected to form a subgraph; the feature extraction module of the initial fusion graph neural network is called to extract attributes such as the spatial coordinates and risk level of the nodes in the subgraph, and attributes such as the path length and dynamic risk weight of the edges are extracted. These attributes are then aggregated to finally obtain the target subgraph features.

[0143] In one feasible implementation, step S32 may include steps S321 to S325:

[0144] Step S321: Obtain a preset sensing range, and determine a preset hop count range based on the physical access edge and the preset sensing range;

[0145] It should be noted that the preset perception range is a physical space range set by combining the safety vision and operational needs of maintenance personnel in extreme environments. It is usually a 50-meter area around the current node. This range can cover the effective environment that personnel can perceive, and can also avoid the computational redundancy caused by the subsequent subgraph being too large.

[0146] Furthermore, the preset hop count range is the number of node jumps defined around the current node. It is calculated by dividing the preset perception range by the average length of the physical access edge and is used to clarify the node selection boundary of the subsequent subgraph.

[0147] Understandably, the process involves first obtaining a pre-defined sensing range, then calculating the average length of the physical access edges around the current node, dividing the pre-defined sensing range by this average length to obtain the number of node jumps. This number is the pre-defined jump range, ensuring that the subgraph within the pre-defined jump range can cover the spatial area corresponding to the pre-defined sensing range.

[0148] Step S322: Expand from the current node according to the preset hop count range to obtain a subgraph node set and determine the edge set of the subgraph node set;

[0149] It should be noted that the subgraph node set is a collection of all initial fusion graph nodes within a preset hop count range, including the current node, surrounding path nodes, sensor nodes, and virtual risk nodes, which together constitute the node basis of the subgraph.

[0150] Furthermore, the edge set is the set of initial fused graph edges that connect each node in the subgraph node set, including physical access edges and device logical association edges. The edge set and the subgraph node set together constitute the complete structure of the subgraph.

[0151] It is understandable that, starting from the current node, the system jumps outwards sequentially according to a preset number of jumps, incorporating the corresponding node with each jump, until all nodes within the preset number of jumps are covered, forming a subgraph node set; then, all edges connecting any two nodes in the subgraph node set are selected to form an edge set, thus obtaining the basic structure of the subgraph containing nodes and edges.

[0152] Step S323: Extract the static attributes of the subgraph node set, and determine the initial features based on the static attributes, wherein the static attributes include spatial coordinates and component material;

[0153] It should be noted that the subgraph node set contains all nodes within a preset number of hops. Each node carries preset static attributes, which are inherent features of the node that do not change with the environment and form the basis for extracting initial features.

[0154] In addition, the spatial coordinates in the static attributes are the three-dimensional position data of the nodes in the initial fusion graph neural network, which are used to reflect the spatial distribution relationship of the nodes and are the spatial basis of path planning; the component material is the material property of the entity corresponding to the node (such as equipment, passage wall). Different materials have different tolerances to extreme environments, which affect the risk association attributes of the node.

[0155] Furthermore, the initial features are digital representations of the static attributes of the subgraph node set, transforming attributes such as spatial coordinates and component materials into vector forms that can be recognized by the neural network, providing a basic feature carrier for subsequent integration with dynamic risk factors.

[0156] Understandably, the spatial coordinates and component material information of each node in the subgraph node set are extracted one by one, and redundant attributes unrelated to the initial features are removed from the nodes; the spatial coordinates are converted into numerical vectors, and the component materials are converted into numerical labels according to preset rules (such as assigning a value of 1 to high-temperature resistant materials and a value of 0 to ordinary materials); the spatial coordinate vectors and material labels of each node are integrated to form the initial features of each node, and finally the initial feature set of the subgraph node set is obtained.

[0157] Step S324: Calculate the dynamic risk factor of the initial feature, and allocate the dynamic risk factor to the edge set of the subgraph node set according to the preset decay rule to obtain the time series feature;

[0158] It should be noted that the dynamic risk factor is a dynamic value calculated based on real-time environmental data, reflecting the environmental risks around the node. This value is updated as the environment changes and is used to correct the risk attributes of the initial characteristics.

[0159] In addition, the preset attenuation rule is used to describe the law of risk change with distance. In this embodiment, the Gaussian attenuation rule is adopted, that is, the risk value decreases as the distance between the node and the risk source increases, which is consistent with the actual situation of risk diffusion in the physical world.

[0160] Furthermore, the time-series feature is the edge feature containing dynamic risk attributes obtained after allocating dynamic risk factors to the edge set according to a preset decay rule. This feature can reflect the real-time risk status of the edge and provide dynamic data support for subsequent subgraph updates.

[0161] Understandably, based on real-time environmental data such as current temperature and dust concentration, combined with the spatial coordinates of each node, the dynamic risk factors around each node are calculated; according to the preset attenuation rule, the contribution value of the dynamic risk factors of the two ends of each edge to the edge is calculated, and the contribution values ​​are superimposed as the dynamic risk attribute of the edge; the dynamic risk attribute of the edge is integrated with the original static attribute of the edge to form the temporal characteristics of each edge, and finally the temporal characteristic set of the edge set is obtained.

[0162] Step S325: Update the subgraph node set according to the temporal features to obtain the target subgraph features.

[0163] It should be noted that the target subgraph features are the updated overall features of the subgraph, which include the features of the subgraph node set and the temporal features of the edge set. The two are related through the connection relationship between nodes and edges to form structured subgraph feature data, which is used to generate candidate paths in the future.

[0164] Furthermore, updating the subgraph node set involves inversely associating the temporal features of the edge set with the corresponding nodes, allowing the node features to contain dynamic risk information of the surrounding edges, and ensuring that the node features can reflect the risk status of its associated paths.

[0165] Understandably, based on the temporal characteristics of each edge in the edge set, the two nodes connected by that edge are found; the dynamic risk attributes of the edge are assigned to the corresponding two nodes according to a preset weight, and the initial features of the nodes are updated; the updated node features and the temporal characteristics of the edges are integrated to form a complete feature set containing node-edge relationships, static attributes and dynamic risk attributes, and this set is the target subgraph feature.

[0166] Step S33: Generate a set of candidate paths from the current node to the target device node based on the target subgraph features;

[0167] It should be noted that the candidate path set contains multiple potential paths from the current node to the target device node. The comprehensive cost of each candidate path is the core indicator for evaluating the quality of the path. By comprehensively considering the time consumption and risk level of the path, we can ensure that the selected path is both efficient and safe.

[0168] Furthermore, the target device node is the device to be maintained or the destination node.

[0169] Understandably, the target subgraph features are input into the path search module of the initial fusion graph neural network. The module takes the current node as the starting point and the target device node as the ending point, and combines the passability attributes of the edges in the subgraph with the risk attributes of the nodes to analyze the connected paths between the nodes. Path filtering conditions are set, such as excluding edges with dynamic risk weights exceeding a threshold and limiting the total path length to no more than 1.5 times the straight-line distance from the current node to the target device node. Based on the filtering conditions, 3-5 differentiated paths are selected from all connected paths to form a candidate path set.

[0170] In one feasible implementation, step S33 may include steps S331 to S333:

[0171] Step S331: Calculate the expected return for movement based on the features of the target subgraph;

[0172] It should be noted that the target subgraph features are structured data that includes the updated features of the subgraph node set and the time-series features of the edge set. The node features include spatial coordinates, component materials, and dynamic risk information of associated edges, while the edge features include path length, dynamic risk weights, etc. These features are the core basis for calculating the expected return of movement.

[0173] Additionally, the expected return on movement is an estimated value of the overall benefits that can be obtained in the future after moving from the current node to the surrounding neighboring nodes. This value takes into account the path efficiency (such as time cost) and security risks (such as dynamic risk weights) after the movement. The higher the value, the better the overall benefits of the movement direction.

[0174] Furthermore, calculating the expected return on movement relies on the decision module of the initial fusion graph neural network. This module performs calculations on the node and edge features in the target subgraph, converting the feature data into quantified return values, providing a basis for decision-making in subsequent path search.

[0175] Understandably, the features of the current node, the features of its surrounding neighboring nodes, and the temporal features of the physical access edges between the current node and its neighboring nodes are extracted from the features of the target subgraph. These features are then input into the decision-making module, which uses a preset algorithm to comprehensively evaluate the time cost savings and risk reductions of moving to each neighboring node, and outputs the expected return for moving to each neighboring node.

[0176] Step S332: Set path search constraints based on the dynamic risk weight threshold of the physical access edge and the total path length threshold;

[0177] It should be noted that the dynamic risk weight threshold of the physical passage edge is a pre-set critical value used to determine whether the edge has the safety of passage. This threshold is determined based on the safety standard of extreme environment. If the dynamic risk weight of the edge is greater than this threshold, it means that the path risk corresponding to the edge is too high and is not suitable for passage.

[0178] In addition, the total path length threshold is a pre-set critical value used to limit the maximum length of candidate paths. It is usually set to 1.5 times the straight-line distance from the current node to the target device node. This threshold can prevent excessively long paths from causing low operation and maintenance efficiency, and ensure that candidate paths have both security and efficiency.

[0179] Furthermore, the path search constraints are a selection rule composed of a dynamic risk weight threshold and a total path length threshold. Paths generated during the subsequent path search process must meet both of these conditions to be included in the candidate path set, ensuring the basic safety and efficiency of the candidate paths.

[0180] Understandably, the dynamic risk weight threshold of physical access edges is determined based on the type of extreme environment, such as setting the threshold to 0.8 in high-temperature environments and 0.7 in high-dust environments; the straight-line distance from the current node to the target device node is calculated, and the total path length threshold is determined at a ratio of 1.5; the dynamic risk weight of physical access edges ≤ dynamic risk weight threshold and the total path length ≤ total path length threshold are taken as two core rules to form the path search constraints.

[0181] Step S333: Starting from the current node and ending at the target device node, perform path search based on the greedy strategy, the expected return of the movement, and the constraints to generate a candidate path set.

[0182] Understandably, starting from the current node, we first filter out the neighboring nodes around the current node that meet the constraints; from these neighboring nodes, we select the node with the highest expected return for movement as the next moving node according to a greedy strategy; using this node as the new starting point, we repeat the above filtering and selection process until we reach the target device node and form a complete path; we repeat the above search process multiple times to generate 3-5 different complete paths, forming a candidate path set.

[0183] In the t-th iteration, when expanding the current path to neighboring nodes to generate a new path, its attributes are calculated as follows:

[0184]

[0185] In the formula, The cumulative risk of the new path; This represents the cumulative risk along the original path. The path dynamic risk function, i.e., the connection and Real-time risk weights of edges (from risk quantification in step one). This is the end node of the original path; for The neighboring nodes.

[0186]

[0187] In the formula, The cumulative time for the new path; This represents the cumulative time along the original path. The path travel time function, i.e., the connection and The static passage time of the edge; This is the end node of the original path; for The neighboring nodes.

[0188]

[0189] In the formula, The cumulative expected return for the new path; This represents the cumulative expected return along the original path. The expected return function for the mobile activity (from the IRL / Q network); The cumulative risk of the new path; This represents the maximum acceptable cumulative risk constraint for the path. The dynamic attenuation coefficient ( (This is used to control the penalty effect of accumulated risk on returns). It is a natural exponential function.

[0190] Furthermore, it is necessary to filter the set of valid paths. The formula for calculating the set of valid paths is as follows:

[0191]

[0192] In the formula, For the first The set of all legal and feasible paths expanded by round-by-round iteration; This represents the maximum acceptable cumulative risk constraint for the path. This is the maximum acceptable cumulative time constraint for the path.

[0193]

[0194] In the formula, For the first The current bundle (best path set) in each round of iteration; To select a function; The cumulative expected return of the path; Beam Width indicates the number of data points used to determine the beam width. The optimal number of paths to retain in descending order. Finally, by integrating, we obtain the final set of paths that contain all paths that reach the goal and satisfy the constraints.

[0195] Step S34: Calculate the comprehensive cost of the candidate paths in the candidate path set, and determine the target operation and maintenance path based on the comprehensive cost.

[0196] It should be noted that the overall cost consists of time cost and risk cost. Time cost is calculated based on path length and travel speed, reflecting the time taken for the path. Risk cost is calculated based on the dynamic risk weights of the edges, reflecting the degree of environmental risk of the path.

[0197] Furthermore, the target maintenance path is the path with the lowest overall cost selected from the candidate path set. This path can guide maintenance personnel to the target equipment node with the least time and risk cost in extreme environments.

[0198] Understandably, the time cost of each candidate path is calculated by summing the lengths of all physical access edges contained in the path to obtain the total length, and then dividing the total length by the average travel speed under extreme conditions to obtain the time cost. The risk cost of each candidate path is calculated by summing the dynamic risk weights of all physical access edges contained in the path to obtain the risk cost. Weighting coefficients for time cost and risk cost are set according to the level of extreme environment, with higher weights for risk cost in high-risk environments. The comprehensive cost of each candidate path is calculated through weighted summation. Finally, the comprehensive cost of all candidate paths is compared, and the path with the lowest comprehensive cost is selected as the target operation and maintenance path.

[0199] In one feasible implementation, step S34 may include steps S341 to S345:

[0200] Step S341: Define a comprehensive cost index based on the path-level characteristics of the candidate path. The comprehensive cost index includes time cost, security risk, and historical reliability.

[0201] It should be noted that time cost is an indicator for evaluating the efficiency of path passage, which is related to the total path length and passage speed; security risk is an indicator for evaluating the environmental risk of the path, which is related to the dynamic risk weight of the physical passage edges; historical reliability is an indicator for evaluating the past performance of the path, which is related to the frequency of path congestion and the time taken to handle faults in the historical operation and maintenance records.

[0202] Understandably, based on the path-level characteristics of each candidate path, the core evaluation dimensions of the path in terms of efficiency, safety, and historical performance are identified; time cost, safety risk, and historical reliability are defined as the three components of the comprehensive cost index, with each component corresponding to a type of key information in the path-level characteristics, forming a complete multi-dimensional evaluation index system.

[0203] Defining multi-dimensional comprehensive cost indicators based on path-level characteristics can avoid the one-sidedness of evaluating paths with a single indicator, and ensure that subsequent scoring calculations can comprehensively consider the efficiency, safety and historical reliability of the path, providing a comprehensive evaluation basis for selecting the optimal path.

[0204] Step S342: Calculate the comprehensive cost by scoring the candidate paths in the candidate path set according to the comprehensive cost index;

[0205] It should be noted that the time cost score is calculated by the total path length and the average travel speed under extreme conditions. The shorter the total path length and the more suitable the travel speed, the lower the time cost score, and the lower the score represents better efficiency. The safety risk score is calculated by summing the dynamic risk weights of all physical travel edges in the path. The lower the total weight, the lower the safety risk score, and the lower the score represents better safety.

[0206] Furthermore, the historical reliability score is calculated using data on route congestion frequency and fault handling time. The lower the congestion frequency and the shorter the fault handling time, the lower the historical reliability score, and the lower the score represents better historical performance. The overall cost is the result of a weighted sum of the three index scores according to preset weights. The lower the overall cost, the better the overall route.

[0207] Understandably, the time cost score for each candidate path is calculated by dividing the total path length by the average travel speed to obtain the time value, and then converting it into a score according to a preset classification rule; the safety risk score is calculated by summing the dynamic risk weights of all physical access edges in the path, and directly using this as the safety risk score; the historical reliability score is calculated by summing the path congestion frequency and fault handling time according to a preset ratio, and using this as the historical reliability score; the weights of the three scores are set according to the extreme environment level, and the weighted sum is used to obtain the comprehensive cost of each candidate path.

[0208] By converting comprehensive cost indicators into calculable comprehensive costs through quantitative scoring, the overall merits and demerits of each candidate path can be objectively and accurately measured, avoiding biases caused by subjective judgments and providing a quantitative basis for subsequent path selection.

[0209] Step S343: Select the initial operation and maintenance path based on the comprehensive cost under preset security constraints;

[0210] It should be noted that the preset security constraints are pre-defined rules used to ensure the basic security of the initial operation and maintenance path. These rules are based on extreme environment security standards.

[0211] In addition, the initial operation and maintenance path is a path selected from the candidate path set that meets the preset security constraints and has a relatively low overall cost. This path is the basis for subsequent path value assessment. It is necessary to ensure that it meets the basic security requirements before further assessing its value.

[0212] Furthermore, the initial operation and maintenance path selection requires first verifying the security constraints of each path in the candidate path set, eliminating paths that do not meet the security requirements, and then selecting the path with the lowest overall cost from the remaining paths as the initial operation and maintenance path to ensure that the path has both security and initial optimality.

[0213] Understandably, each path in the candidate path set is checked one by one to see if it meets the preset security constraints. If the security risk score of a path exceeds the security risk threshold, or if there is a physical access edge with a dynamic risk weight greater than the threshold, it is excluded. From the remaining paths that meet the security constraints, the overall cost is compared, and the path with the lowest overall cost is selected as the initial operation and maintenance path.

[0214] Step S344: Evaluate the path value based on the initial operation and maintenance path to obtain the path value;

[0215] It should be noted that path value is a quantitative value that measures the added value of the initial operation and maintenance path. The higher the value, the more additional support the path can provide for operation and maintenance work in addition to basic security and efficiency, such as the presence of emergency equipment storage points near the path or the path being able to quickly reach the location of related equipment.

[0216] Furthermore, path value assessment needs to be combined with the needs of operation and maintenance scenarios, and scores should be given from dimensions such as the distance between the path and emergency resources, the correlation between the path and related equipment, and the sufficiency of the path's operational space. The path value is then obtained by weighted summation of the scores of each dimension.

[0217] Understandably, the process involves querying the location of emergency resources around the initial maintenance path, calculating the shortest distance between the path and the emergency resources (the closer the distance, the higher the score); analyzing whether the initial maintenance path can quickly reach the associated devices of the faulty equipment (the stronger the association, the higher the score); assessing whether the operating space along the path meets the needs of equipment maintenance (the more space, the higher the score); and finally, weighting and summing the scores of each dimension according to preset weights to obtain the path value of the initial maintenance path.

[0218] Step S345: Determine the target operation and maintenance path based on the path value.

[0219] Furthermore, determining the target maintenance path requires first setting a path value threshold. If the path value of the initial maintenance path is greater than or equal to the threshold, it means that its added value meets the standard and can be directly used as the target maintenance path. If it is less than the threshold, the path with the second lowest overall cost needs to be re-selected from the candidate paths that meet the security constraints for value evaluation until a path whose value meets the standard is found.

[0220] Understandably, the path value of the initial maintenance path is compared with a preset path value threshold. If the path value is greater than or equal to the threshold, the initial maintenance path is determined as the target maintenance path. If the path value is less than the threshold, the path with the second lowest overall cost among the candidate paths that meet the security constraints is selected as the new initial maintenance path. The value assessment steps are repeated until a path whose path value meets the standard is found and it is determined as the target maintenance path.

[0221] Determining the target operation and maintenance path by combining path value can ensure that the final target path is not only safe and efficient, but also has high added value, which can provide more support for operation and maintenance work. It can avoid problems such as insufficient emergency resources and insufficient operating space during operation and maintenance due to neglecting added value, and improve the overall success rate and security of operation and maintenance tasks.

[0222] For example, to help understand the implementation process of the operation and maintenance guidance method based on fused graph neural networks obtained by combining this embodiment with the above embodiment one, please refer to... Figure 3 , Figure 3A simplified flowchart of an operation and maintenance guidance method based on a fused graph neural network is provided, specifically:

[0223] Starting with dynamic sensing, abrupt changes in sensor readings are required to capture real-time changes in the environment, such as temperature and dust concentration. Next, a risk diffusion function is calculated using sensor readings and the maintenance worker's location to generate a path subgraph. Environmental risks are assessed using sensor data, and a dynamic subgraph reflecting the current environmental conditions is constructed accordingly. Then, meta-path attention and node fusion representations are generated from the path subgraph. Graph neural network technology is used to analyze the paths and nodes in the subgraph, generating a comprehensive representation that integrates environmental risk and path information to facilitate further decision-making. A comprehensive cost calculation is performed, evaluating the costs of all possible paths, including time and risk costs, to determine the optimal path. By comparing the costs of different paths, the optimal next action is selected, ensuring that maintenance personnel can safely and efficiently reach the target location. The calculated optimal path and related data are pushed to the maintenance personnel's digital twin terminal, such as AR glasses or industrial tablets, to provide real-time navigation and data support.

[0224] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the operation and maintenance guidance method based on the fused graph neural network of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0225] This application also provides an operation and maintenance guidance device based on a fused graph neural network. Please refer to [reference needed]. Figure 4 The operation and maintenance guidance device based on the fused graph neural network includes:

[0226] The data determination module 10 is used to construct a digital twin model of the target building based on the maintenance work order information and to determine extreme environment scenario data based on the maintenance work order information and the digital twin model of the target building.

[0227] Network construction module 20 is used to construct an initial fusion graph neural network based on the target building digital twin model and the extreme environment scenario data;

[0228] The path generation module 30 is used to input the current location of maintenance personnel, environmental data, and historical operation and maintenance records into the initial fusion graph neural network to obtain the target operation and maintenance path;

[0229] The operation and maintenance guidance module 40 is used to generate navigation instructions based on the target operation and maintenance path to complete the operation and maintenance guidance based on the fused graph neural network.

[0230] The operation and maintenance guidance device based on fused graph neural networks provided in this application, employing the operation and maintenance guidance method based on fused graph neural networks in the above embodiments, can solve the technical problem of how to ensure that operation and maintenance personnel can safely and efficiently complete operation navigation and decision-making in extreme environments when path dynamics are obstructed and environmental information is incomplete. Compared with the prior art, the beneficial effects of the operation and maintenance guidance device based on fused graph neural networks provided in this application are the same as the beneficial effects of the operation and maintenance guidance method based on fused graph neural networks provided in the above embodiments, and other technical features in the operation and maintenance guidance device based on fused graph neural networks are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0231] This application provides an operation and maintenance guidance device based on a fused graph neural network. The operation and maintenance guidance device based on the fused graph neural network includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the operation and maintenance guidance method based on the fused graph neural network in the above embodiment 1.

[0232] The following is for reference. Figure 5 This document illustrates a structural schematic diagram of an operation and maintenance guidance device based on a fused graph neural network, suitable for implementing embodiments of this application. The operation and maintenance guidance device based on a fused graph neural network in these embodiments may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 5 The operation and maintenance guidance device based on fused graph neural network shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0233] like Figure 5As shown, the maintenance guidance device based on a fused graph neural network may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the maintenance guidance device based on the fused graph neural network. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the maintenance guidance equipment based on fused graph neural networks to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows maintenance guidance equipment based on fused graph neural networks with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0234] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0235] The operation and maintenance guidance device based on fused graph neural networks provided in this application, employing the operation and maintenance guidance method based on fused graph neural networks in the above embodiments, can solve the technical problem of how to ensure that operation and maintenance personnel can safely and efficiently complete operation navigation and decision-making in extreme environments when path dynamics are obstructed and environmental information is incomplete. Compared with the prior art, the beneficial effects of the operation and maintenance guidance device based on fused graph neural networks provided in this application are the same as the beneficial effects of the operation and maintenance guidance method based on fused graph neural networks provided in the above embodiments, and other technical features in this operation and maintenance guidance device based on fused graph neural networks are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0236] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0237] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0238] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the operation and maintenance guidance method based on the fused graph neural network in the above embodiments.

[0239] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), Erasable Programmable Read Only Memory (EPROM), optical fiber, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0240] The aforementioned computer-readable storage medium may be included in the operation and maintenance guidance device based on the fused graph neural network; or it may exist independently and not assembled into the operation and maintenance guidance device based on the fused graph neural network.

[0241] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the fusion graph neural network-based operation and maintenance guidance device, cause the fusion graph neural network-based operation and maintenance guidance device to: ddd

[0242] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0243] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0244] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0245] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-described operation and maintenance guidance method based on fused graph neural networks. This solves the technical problem of ensuring that operation and maintenance personnel can safely and efficiently complete job navigation and decision-making even in extreme environments where path dynamics are obstructed and environmental information is incomplete. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the operation and maintenance guidance method based on fused graph neural networks provided in the above embodiments, and will not be repeated here.

[0246] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the operation and maintenance guidance method based on the fused graph neural network described above.

[0247] The computer program product provided in this application can solve the technical problem of how to ensure that maintenance personnel can safely and efficiently complete operation navigation and decision-making in extreme environments, even when path dynamics are blocked and environmental information is incomplete. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the maintenance guidance method based on fused graph neural networks provided in the above embodiments, and will not be repeated here.

[0248] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for operation and maintenance guidance based on a fused graph neural network, characterized in that, The method includes: A digital twin model of the target building is constructed based on the maintenance work order information, and extreme environment scenario data is determined based on the maintenance work order information and the digital twin model of the target building. Based on the target building digital twin model and the extreme environment scenario data, an initial fusion graph neural network is constructed; The current location of maintenance personnel, environmental data, and historical maintenance records are input into the initial fusion graph neural network to obtain the target maintenance path; Navigation instructions are generated based on the target operation and maintenance path to complete operation and maintenance guidance based on fused graph neural network; The step of constructing an initial fusion graph neural network based on the target building digital twin model and the extreme environment scenario data includes: Based on the target building digital twin model, determine the geometric data, non-geometric data, and equipment association data of the building information model; Based on the extreme environment scenario data, obtain data on the deployment locations of environmental sensors and the distribution of risk areas; Based on the geometric data of the building information model, the non-geometric data of the building information model, the equipment association data, the deployment location of the environmental sensors, and the risk area distribution data, initial fusion graph nodes and initial fusion graph edges are defined. The initial fusion graph nodes include maintenance personnel starting candidate nodes, target equipment nodes, sensor nodes, and virtual risk nodes. The initial fusion graph edges include physical access edges and equipment logical association edges. An initial fusion graph neural network is constructed based on the nodes and edges of the initial fusion graph.

2. The method as described in claim 1, characterized in that, The step of inputting the current maintenance personnel location, environmental data, and historical maintenance records into the initial fusion graph neural network to obtain the target maintenance path includes: The current location of maintenance personnel, environmental data, and historical operation and maintenance records are input into the initial fusion graph neural network to determine the current node and the physical access edge of the current node; A preset hop count range is determined based on the physical access edges, and feature extraction is performed on the subgraph within the preset hop count range of the current node to obtain the target subgraph features; Generate a set of candidate paths from the current node to the target device node based on the target subgraph features; Calculate the comprehensive cost of the candidate paths in the candidate path set, and determine the target operation and maintenance path based on the comprehensive cost.

3. The method as described in claim 2, characterized in that, The step of inputting the current maintenance personnel location, environmental data, and historical maintenance records into the initial fusion graph neural network to determine the current node and the physical access edge of the current node includes: Determine the current temperature, dust concentration, and harmful gas concentration based on environmental data; The current location of the maintenance personnel, the current temperature, the dust concentration, and the concentration of harmful gases are input into the initial fusion graph neural network to determine the current node; Determine the frequency of path congestion and the time taken to handle faults based on historical operation and maintenance records; The path congestion frequency and fault handling time data are input into the initial fusion graph neural network to determine the physical access edge of the current node.

4. The method as described in claim 2, characterized in that, The steps of determining a preset hop count range based on the physical access edges and extracting features from the subgraph within the preset hop count range of the current node to obtain the target subgraph features include: Obtain a preset sensing range, and determine a preset hop count range based on the physical access edge and the preset sensing range; Expand from the current node according to the preset hop count range to obtain a subgraph node set and determine the edge set of the subgraph node set; Extract the static attributes of the subgraph node set, and determine the initial features based on the static attributes, wherein the static attributes include spatial coordinates and component material; Calculate the dynamic risk factor of the initial feature, and allocate the dynamic risk factor to the edge set of the subgraph node set according to the preset decay rule to obtain the time series feature; The subgraph node set is updated based on the temporal features to obtain the target subgraph features.

5. The method as described in claim 2, characterized in that, The step of generating a set of candidate paths from the current node to the target device node based on the target subgraph features includes: Calculate the expected return on movement based on the characteristics of the target subgraph; Set path search constraints based on the dynamic risk weight threshold of the physical access edge and the total path length threshold; Starting from the current node and ending at the target device node, a path search is performed based on a greedy strategy, the expected reward of the movement, and the constraints to generate a set of candidate paths.

6. The method as described in claim 2, characterized in that, The step of calculating the comprehensive cost of the candidate paths in the candidate path set and determining the target operation and maintenance path based on the comprehensive cost includes: A comprehensive cost index is defined based on the path-level characteristics of the candidate paths. The comprehensive cost index includes time cost, security risk, and historical reliability. The candidate paths in the candidate path set are scored and calculated based on the comprehensive cost index to obtain the comprehensive cost; Under preset security constraints, an initial operation and maintenance path is selected based on the overall cost. The path value is obtained by evaluating the path value based on the initial operation and maintenance path. The target operation and maintenance path is determined based on the path value.

7. The method as described in claim 1, characterized in that, The step of generating navigation instructions based on the target operation and maintenance path to complete the operation and maintenance guidance based on the fusion graph neural network includes: An operation and maintenance perspective is generated based on the target operation and maintenance path, and the operation and maintenance perspective is sent to the digital twin mobile terminal, so that the digital twin mobile terminal can provide feedback on the operation and maintenance site information based on the target operation and maintenance path; The target maintenance data package is determined based on the aforementioned maintenance site information; Navigation instructions are generated based on the target maintenance data package to complete maintenance guidance based on a fused graph neural network.

8. The method as described in claim 1, characterized in that, The steps of constructing a digital twin model of the target building based on the maintenance work order information and determining extreme environment scenario data based on the maintenance work order information and the digital twin model of the target building include: Extract the faulty device identifier, fault type, and maintenance area from the maintenance work order information; Based on the faulty device identifier, the fault type, and the maintenance area, construct a digital twin model of the target building; Based on the target building digital twin model, locate the model location area of ​​the faulty equipment in the maintenance work order information; Determine the extreme environment type of the model's location region; Based on the digital twin model of the target building and the extreme environment type, environmental parameters are analyzed to obtain extreme environment scenario data.

9. An operation and maintenance guidance device based on a fused graph neural network, characterized in that, The operation and maintenance guidance device based on a fused graph neural network executes the operation and maintenance guidance method based on a fused graph neural network as described in any one of claims 1 to 8, wherein the device comprises: The data determination module is used to construct a digital twin model of the target building based on the maintenance work order information and to determine extreme environment scenario data based on the maintenance work order information and the target building digital twin model. The network construction module is used to construct an initial fusion graph neural network based on the target building digital twin model and the extreme environment scenario data; The path generation module is used to input the current location of maintenance personnel, environmental data, and historical operation and maintenance records into the initial fusion graph neural network to obtain the target operation and maintenance path; The operation and maintenance guidance module is used to generate navigation instructions based on the target operation and maintenance path to complete the operation and maintenance guidance based on the fused graph neural network.

Images