A task-space consistency autonomous navigation method and system for substation multi-intelligent inspection equipment

Abstract

The application discloses a kind of task-space consistency autonomous navigation method and system for substation multi-intelligent inspection equipment.Belongs to substation intelligent inspection, space intelligent modeling, visual-language-navigation, robot autonomous navigation and multi-intelligent device collaborative control technical field.The application constructs task-space coupling graph by collecting substation multi-source data;The input inspection task is parsed as a candidate structured task unit, and consistency check is carried out.After verification, the system generates executable navigation segments composed of approach section, observation section, local obstacle avoidance section and safety exit section;Type closed loop sample is generated after task, and is hierarchically written to task-space coupling graph according to exception type.The application improves the reliability, controllability and continuous optimization capability of natural language inspection task conversion into safe and executable navigation action in the complex environment of substation.

Description

Technical Field

[0001] This invention relates to the fields of intelligent substation inspection, spatial intelligent modeling, vision-language-navigation, robot autonomous navigation, and multi-intelligent device collaborative control, and particularly to a task-space consistent autonomous navigation method and system for multi-intelligent substation inspection equipment. Background Technology

[0002] Substations are densely packed with equipment and have complex spatial relationships, typically including a variety of objects such as main transformers, circuit breakers, disconnect switches, instrument transformers, surge arresters, busbars, frames, fences, trenches, inspection roads, energized compartments, maintenance areas, and temporary safety fences. With the development of intelligent inspection technology, various intelligent inspection equipment, such as wheeled inspection robots, quadrupedal inspection robots, and drones, are gradually being applied to substation inspections.

[0003] Existing intelligent substation inspection methods typically rely on preset maps, preset inspection points, and fixed inspection routes for navigation. While some methods incorporate semantic maps, path optimization, multi-source sensor fusion, voice interaction, or large-scale model analysis, their focus is primarily on inspection route generation, improved positioning accuracy, obstacle recognition, voice command invocation, or inspection result diagnosis. For complex substation inspection scenarios, the following problems still exist.

[0004] First, existing consistency checks primarily involve verification of perception layer data, such as accuracy checks between multiple sensor data, fusion checks between positioning data, deviation checks between map data and the real-time environment, or obstacle recognition result checks. Their main purpose is to improve the reliability of robot positioning, obstacle avoidance, or path execution. However, in natural language inspection task scenarios, the primary problem the system faces is not the accuracy of sensor data, but whether the semantics of the inspection task expressed by the user can be correctly and safely executed in the actual substation space. For example, when maintenance personnel input tasks such as "inspect the high-voltage side bushing of main transformer No. 1," "check the status of the 220kV bus tie bay disconnector," or "perform infrared temperature measurement on main transformer No. 2," the task instructions may contain ambiguities in equipment names, non-existent equipment locations, inconsistencies between target equipment and electrical bays, inaccessible observable surfaces, insufficient safety distances, or inadequate capabilities of the current inspection equipment. If the system does not perform consistency verification between task semantics and spatial entities before task execution, but directly generates paths and performs inspections, problems such as arriving at the wrong interval, being able to reach but not being able to observe, being able to observe but not being able to see clearly, being able to see clearly but not meeting the safe distance, or discovering that the inspection results are invalid only after the task is executed may occur.

[0005] Second, existing substation spatial modeling primarily focuses on traversable space, i.e., whether inspection equipment can reach a certain location. While some technologies adjust the attitude of cameras, pan-tilt units, or sensors after reaching the inspection point, they typically treat the observed attitude as an auxiliary action or local control parameter after path planning, failing to consider the observable pose as an independent modeling dimension alongside traversable space. The essence of substation inspection tasks goes beyond simply being able to reach the location; it also requires inspection equipment to effectively observe the target equipment or location within a safe distance, at an appropriate angle, at an appropriate distance, and with appropriate sensor attitude. Therefore, generating inspection paths solely based on traversable space is insufficient to address issues such as being able to reach the location but not being able to see it, being able to see it but not clearly, and being able to see it clearly but not safely.

[0006] Third, the existing navigation logic is usually "task input - path planning - execution inspection". This logic can automate the path in normal scenarios, but for substation safety-critical scenarios, if the task semantics, equipment entities, electrical bays, equipment locations, observable poses, safety boundaries and equipment capabilities are not pre-verified before path planning and task execution, it is easy to push erroneous tasks directly to the execution stage.

[0007] Fourth, existing closed-loop feedback typically feeds back inspection failures, manual takeovers, or replanning records as general samples to the model, without distinguishing between task semantic ambiguity, abnormal equipment entity binding, abnormal access space, abnormal observable pose, abnormal electrical safety boundary, and abnormal equipment capability mismatch. It also fails to write back different anomalies to different levels of the spatial model, resulting in unclear feedback objects and inaccurate model updates.

[0008] Fifth, when existing large-scale AI models or vision-language models are applied to inspection scenarios, they tend to directly use the model output as the basis for task execution, lacking a controllable execution mechanism suitable for substations with high safety requirements. For substation scenarios with densely packed energized equipment and serious consequences of malfunctions, the AI ​​model output should be used as a candidate result, rather than bypassing task-space consistency verification and electrical safety verification to directly drive the inspection equipment to execute.

[0009] Therefore, it is necessary to propose a task-space consistency autonomous navigation method for multi-intelligent inspection equipment in substations. This method unifies the modeling of natural language inspection tasks, substation spatial entities, observable poses, electrical safety boundaries, and inspection equipment capabilities. It adopts a "verify first, execute later" navigation paradigm, which allows the inspection task to be judged correctly, safely, and effectively before execution. Summary of the Invention

[0010] This invention addresses the shortcomings of existing technologies by providing a task-space consistency autonomous navigation method and system for multiple intelligent inspection devices in substations. It solves the problems in existing substation inspection technologies, such as the difficulty in reliably converting natural language tasks into spatial navigation actions, the lack of independent modeling of traversable paths and observable poses, the lack of semantic and spatial consistency verification before task execution, insufficient matching of multiple device capabilities, and inaccurate closed-loop feedback.

[0011] Existing consistency verification technologies primarily target the accuracy and reliability of multi-sensor data, positioning data, or environmental perception data, addressing the issue of whether the robot's perception is accurate. In contrast, the task-space consistency verification of this invention targets the consistency between the semantics of the natural language inspection task and the spatial entities, electrical bays, observable poses, safety boundaries, and equipment capabilities of the substation, addressing the issue of whether the inspection task can be executed correctly and safely.

[0012] Existing technologies typically interpret spatial models as traversable spaces and generate inspection routes based on traversable paths. Even when some existing technologies consider optimal observation attitude, they often adjust the attitude of cameras, pan-tilt units, or sensors after reaching the inspection point, without treating observable pose as an independent modeling dimension alongside traversable space. This invention further incorporates observable pose as an independent level into task-space consistency verification and executable navigation segment generation. This enables the system to not only determine whether the inspection equipment can reach the vicinity of the target, but also whether it can complete effective observation with an attitude that meets the requirements of detection actions and safety constraints. This reflects the technical characteristic of separating "arrival" and "observation" in substation inspection tasks.

[0013] The typical logic of existing technologies is "task input - path planning - execution inspection". This invention adopts the logic of "task input - task-space consistency verification - execution if successful, clarification or replacement if unsuccessful - navigation fragment generation - execution consistency comparison - typed layered write-back". This mode can detect problems such as equipment ambiguity, part errors, interval conflicts, unreachable observation, and insufficient safety distance before task execution, preventing erroneous tasks from directly entering the execution stage. It is particularly suitable for safety-critical scenarios such as substations.

[0014] Furthermore, the feedback mechanism of this invention does not perform general model training on all failed samples. Instead, it writes back to different levels of the task-space coupling graph based on the cause of the anomaly: task semantic ambiguity, equipment entity binding anomaly, passage space anomaly, observable pose anomaly, electrical safety boundary anomaly, and equipment capability mismatch anomaly. This achieves precise feedback and targeted optimization. This technique can solve the problems of unclear feedback objects and inaccurate model updates in existing closed-loop learning.

[0015] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0016] In a first aspect, the present invention provides a task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations, characterized by comprising the following steps: S1: Collect multi-source data from the substation and construct a task-space coupling diagram of the substation; S2: Receive the input inspection task instruction and parse the inspection task instruction into candidate structured task units; S3: Perform task-space consistency verification between the candidate structured task units and the substation task-space coupling diagram; if the task-space consistency verification fails, the candidate navigation segment generation stage is prohibited, and a feedback strategy is generated based on the verification failure type; if the task-space consistency verification passes, candidate observable poses are retrieved in the task-space coupling diagram based on the observable surfaces corresponding to the target equipment parts, and candidate navigation segments that match the inspection equipment capabilities are selected from the candidate observable poses; and the corresponding equipment actions, spatial node positions, sensor actions, observation attitudes, and safety boundaries are bound to the candidate navigation segments. S4: Perform a comprehensive evaluation of the candidate navigation segments to obtain the navigation segment evaluation value, and select the candidate navigation segments whose navigation segment evaluation values ​​meet the set conditions to form the target navigation task; S5: The inspection equipment performs target navigation tasks, collects the inspection equipment pose, environmental perception data, target equipment identification results and safety distance status in real time, and performs an execution consistency comparison with the equipment actions, spatial node positions, sensor actions, observation attitude and safety boundaries bound to the target navigation segment; when the execution consistency comparison result shows that there is an anomaly, it performs local replanning, task clarification, safety waiting, equipment switching or safe exit control according to the anomaly type. S6: After the inspection task is completed, collect the inspection task process data and generate categorized closed-loop samples. S7: Perform hierarchical back-writing of the substation task-space coupling diagram based on the typified closed-loop samples.

[0017] Optionally, in step S1, the substation multi-source data includes substation three-dimensional spatial data, equipment ledger data, primary wiring topology data, inspection road data, live-line safety distance data, historical inspection trajectory data, and capability data of various types of inspection equipment; the substation task-space coupling diagram includes an equipment entity layer, a passage space layer, an observable pose layer, an electrical safety constraint layer, an inspection task layer, and an equipment capability layer; wherein, the passage space layer is used to describe whether the inspection equipment can reach a certain spatial location, and the observable pose layer is set in parallel with the passage space layer to describe whether the inspection equipment can effectively observe the target equipment part at a certain spatial location and attitude; The substation task-space coupling diagram is represented as follows:

[0018] in, This diagram illustrates the task-space coupling of a substation. Represents a set of device entity nodes; Represents the set of nodes in the travel space; Represents the set of observable pose nodes; Represents the set of inspection task nodes; This represents the set of capability nodes for inspection equipment; Indicates a passable connecting edge; Represent observable relation edges; Indicates electrical safety constraint edges; This indicates a device capability matching edge.

[0019] Optionally, in step S2, the inspection task instruction is parsed into candidate structured task units through a vision-language-navigation model or a task parsing model; the candidate structured task unit includes at least the target device, device location, detection action, expected observation method, task priority, safety constraints, and candidate inspection device; The candidate structured task unit is represented as follows:

[0020] in, U This represents a candidate structured task unit obtained by parsing text, speech, or human-computer interface input. o Indicates the target device; b Indicates the location of the target equipment; a Indicates a detection action; v Indicates the desired observation method; q Indicates task priority; c Indicates safety constraints; d This indicates candidate inspection equipment.

[0021] Optionally, in step S3, the task-space consistency verification includes equipment entity consistency verification, equipment part consistency verification, electrical interval consistency verification, observable pose consistency verification, passage space consistency verification, electrical safety consistency verification, and equipment capability consistency verification. Specifically, equipment entity consistency verification is used to determine whether the target equipment name, equipment number, equipment alias, and the entity in the equipment register uniquely correspond; equipment part consistency verification is used to determine whether the target equipment part has a corresponding observable surface; electrical interval consistency verification is used to determine whether the interval to which the target equipment belongs is consistent with the interval information in the task instruction; observable pose consistency verification is used to determine whether there are candidate observable poses that meet the inspection action requirements; passage space consistency verification is used to determine whether there are candidate passage paths that meet the equipment's passage capability; electrical safety consistency verification is used to determine whether the candidate passage paths and candidate observable poses meet the live-line safety distance requirements; and equipment capability consistency verification is used to determine whether the candidate inspection equipment possesses the motion capability, sensor capability, endurance capability, and communication capability required to complete the task.

[0022] Optionally, in step S3, the task-space consistency check result is expressed as:

[0023] in, ; K This indicates the overall result of the task-space consistency check; K e This indicates the result of the device entity consistency verification. K b This indicates the results of the equipment part consistency verification. K i This indicates the result of the electrical bay consistency verification. K o This indicates the observable pose consistency verification result; K p This indicates the consistency check result of the passage space; K s This indicates the electrical safety conformity verification result; K d This indicates the result of the equipment capability consistency verification. K= 1 indicates that the inspection task meets the task-space consistency requirements, allowing it to enter the executable navigation fragment generation stage; K =0 indicates that the inspection task cannot be executed directly.

[0024] Optionally, in step S3, the feedback strategy includes task clarification information, alternative inspection plans, equipment replacement execution suggestions, safe waiting strategies, and safe exit strategies; the inspection equipment capabilities include movement mode, passage size, obstacle crossing capability, sensor type, observation altitude, endurance status, and communication status; the candidate navigation segments include an approach segment, an observation segment, a local obstacle avoidance segment, and a safe exit segment; wherein, the approach segment is used to control the inspection equipment to move from its current position to a candidate observation area near the target equipment; the observation segment is used to control the inspection equipment to maintain its observation posture and perform image acquisition, infrared temperature measurement, meter identification, sound acquisition, or partial discharge detection actions; the local obstacle avoidance segment is used to perform local detour, waiting, or posture adjustment when the approach segment or observation segment is affected by dynamic obstacles; the safe exit segment is used to control the inspection equipment to exit to a safe position when the task is completed, the risk exceeds the limit, or the task cannot continue.

[0025] Optionally, in step S4, the comprehensive evaluation includes task matching degree, feasibility, observability, electrical safety margin, dynamic obstacle risk, equipment capability adaptability, and task execution cost. The evaluation value of the navigation segment is determined as follows:

[0026] in, Indicates inspection equipment Execute candidate navigation fragments Evaluation value; Indicates candidate navigation segments; Indicates candidate inspection equipment; Indicates candidate navigation fragments Task matching degree with the current inspection task; Indicates candidate navigation fragments Observability of the target equipment parts; Indicates candidate navigation fragments Electrical safety margin; Indicates candidate navigation fragments Inspection equipment Equipment capability compatibility; Indicates candidate navigation fragments The execution cost; Indicates candidate navigation fragments The risk of dynamic obstacles; arrive Indicates the weighting coefficient; .

[0027] Optionally, in step S5, the consistency comparison result is expressed as:

[0028] in, Y This indicates the result of the consistency comparison. This indicates the positional deviation between the real-time position of the inspection equipment and the position of a spatial node in the target navigation segment; This indicates the real-time entity matching deviation between the identified device entity and the target device entity; This indicates the observation attitude deviation between the real-time observation direction and the normal to the observable surface of the target; This indicates the deviation of the safety margin between the real-time safety distance and the electrical safety boundary; This indicates that passage is obstructed.

[0029] Optionally, in step S6, the task process data includes task instructions, task parsing results, target navigation segments, actual running trajectory, execution consistency comparison results, replanning records, exception types, manual takeover records, and task completion quality. The typed closed-loop sample is represented as follows:

[0030] in, B Represents a typed closed-loop sample; U Indicates a candidate structured task unit; Indicates a target navigation segment; X Indicates the actual running trajectory; Y This indicates the result of the consistency comparison. Z Indicates the exception type; A This indicates the result of manual confirmation or the quality of task completion; Exception types Z Represented as:

[0031] in, This indicates an anomaly in the semantic ambiguity of the task. This indicates an error in the device entity binding; This indicates an abnormal passage space; This indicates an observable pose anomaly; This indicates an abnormal electrical safety boundary; This indicates an abnormality due to a mismatch between equipment capabilities.

[0032] Secondly, the present invention provides a task-space consistency autonomous navigation system for multiple intelligent inspection devices in substations, characterized in that it includes: The data acquisition module is used to collect multi-source data from the substation. The task-space coupling graph construction module is used to construct the task-space coupling graph of the substation. The task parsing module is used to receive the input inspection task instructions and parse the inspection task instructions into candidate structured task units. The task-space consistency verification module is used to perform task-space consistency verification between candidate structured task units and the substation task-space coupling diagram. If the task-space consistency verification fails, the candidate navigation segment generation stage is prohibited, and a feedback strategy is generated based on the verification failure type. If the task-space consistency verification passes, the candidate navigation segment generation stage is entered. The navigation fragment generation module is used to retrieve candidate observable poses in the task-space coupling graph based on the observable surface corresponding to the target device part after the task-space consistency verification is passed, generate candidate navigation fragments, and bind the corresponding device actions, space node positions, sensor actions, observation attitudes and safety boundaries to the candidate navigation fragments; The equipment capability matching module filters candidate navigation segments that match the capabilities of the inspected equipment from the candidate observable poses. The target navigation task generation module is used to comprehensively evaluate candidate navigation segments, obtain navigation segment evaluation values, and select candidate navigation segments whose navigation segment evaluation values ​​meet the set conditions to form a target navigation task. The consistency comparison module is used to collect the pose of the inspection equipment, environmental perception data, target equipment identification results, and safety distance status in real time when the inspection equipment is performing target navigation tasks. It then performs an execution consistency comparison with the collected results and the equipment actions, spatial node positions, sensor actions, observation attitudes, and safety boundaries bound to the target navigation segments. When the execution consistency comparison results show that there is an anomaly, it performs local replanning, task clarification, safe waiting, equipment switching, or safe exit control according to the anomaly type. The closed-loop write-back module is used to collect task process data after the inspection task is completed, generate categorized closed-loop samples, and perform layered write-back of the substation task-space coupling diagram based on the categorized closed-loop samples.

[0033] The beneficial effects of this invention are: 1. The consistency verification of this invention is not an accuracy verification between multi-sensor data, but a consistency verification between the semantics of natural language inspection tasks and the spatial entities, electrical bays, observable poses, safety boundaries, and inspection equipment capabilities of substations, thus solving the problem of whether the task semantics can be executed correctly and safely.

[0034] 2. This invention uses the observable pose layer as an independent modeling dimension alongside the access space layer. It not only determines whether the inspection equipment can reach a certain location, but also whether it can complete effective observation at that location. This solves the problems of being able to reach a location but not being able to see it, being able to see it but not clearly, and being able to see it clearly but not safely in substation inspections.

[0035] 3. This invention adopts a navigation mode of verification before execution. Before path planning and task execution, it performs verification of entities, intervals, locations, observations, passage, safety, and equipment capabilities. If the verification fails, it does not execute directly, but instead performs task clarification, generates alternative solutions, suggests equipment replacement, or performs safe waiting or safe exit, which significantly improves the navigation reliability in substation safety-critical scenarios.

[0036] 4. This invention transforms natural language tasks into executable navigation segments consisting of an approach segment, an observation segment, a local obstacle avoidance segment, and a safe exit segment, and binds each segment to equipment actions, sensor actions, observation attitude, and safety boundaries, thereby improving the controllability of the inspection task execution process.

[0037] 5. This invention classifies inspection anomalies into task semantic ambiguity, equipment entity binding anomaly, access space anomaly, observable pose anomaly, electrical safety boundary anomaly, and equipment capability mismatch anomaly, and writes them back to different levels of the task-space coupling graph to achieve accurate feedback and continuous optimization.

[0038] 6. This invention limits the output of large models or vision-language-navigation models to candidate task parsing and candidate navigation fragments. The final execution must undergo task-space consistency verification and electrical safety consistency verification, avoiding safety risks caused by AI models directly controlling equipment, and improving the controllability, interpretability and engineering applicability of AI models in substation inspection scenarios. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the overall process of the task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to the present invention; Figure 2 This is a schematic diagram of the six layers of nodes and relationships in the task-space coupling graph of an embodiment of the present invention; Figure 3 This is a schematic diagram of the candidate pose selection and proximity path generation results in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the capability matching results of different inspection devices for navigation segments in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the consistency comparison and local replanning response results in an embodiment of the present invention. Figure 6 This is a schematic diagram illustrating the comparative test results of simulated inspection tasks in an embodiment of the present invention; Figure 7 This is a schematic diagram of the continuous optimization results after typed layered write-back in an embodiment of the present invention. Detailed Implementation

[0040] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0041] Example 1 This embodiment provides a task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations, with the specific process as follows: Figure 1 As shown, it includes the following steps: S1 collects 3D spatial data of the substation, equipment ledger data, primary wiring topology data, inspection road data, live safety distance data, historical inspection trajectory data, and capability data of various types of inspection equipment to construct a substation task-space coupling diagram. The substation task-space coupling diagram includes an equipment entity layer, a passage space layer, an observable pose layer, an electrical safety constraint layer, an inspection task layer, and an equipment capability layer. Among them, the passage space layer describes whether the inspection equipment can reach a certain spatial location. The observable pose layer is set in parallel with the passage space layer and is used to describe whether the inspection equipment can effectively observe the target equipment part under a certain spatial location and attitude. The observable pose nodes in the observable pose layer include at least spatial location, observation direction, sensor attitude, target observable surface, observation distance, observation angle, occlusion status, imaging quality level, and electrical safety margin.

[0042] The substation task-space coupling diagram is represented as follows:

[0043] In the formula, This diagram illustrates the task-space coupling of a substation. Represents a set of device entity nodes; Represents the set of nodes in the travel space; Represents the set of observable pose nodes; Represents the set of inspection task nodes; This represents the set of capability nodes for inspection equipment; Indicates a passable connecting edge; Represent observable relation edges; Indicates electrical safety constraint edges; This indicates a device capability matching edge.

[0044] In this embodiment, the system collects 3D point cloud data, BIM model, substation layout plan, equipment ledger, primary wiring diagram, inspection road data, live-line safety distance table, historical inspection trajectories, and parameters of various types of inspection equipment from the substation. The system then uniformly converts the above data to the substation's internal coordinate system and constructs a task-space coupling diagram.

[0045] The task-space coupling diagram includes an equipment entity layer, a passage space layer, an observable pose layer, an electrical safety constraint layer, an inspection task layer, and an equipment capability layer. The equipment entity layer represents equipment entities such as main transformers, circuit breakers, disconnect switches, instrument transformers, surge arresters, busbars, frames, and cabinets, and records electrical bays, equipment names, equipment numbers, equipment aliases, spatial coordinates, and equipment component information. The passage space layer represents inspection roads, passable areas, steps, ditches, access control systems, fences, temporary obstruction areas, and safe exit areas. The passage space layer addresses whether the inspection equipment can reach a specific location. The observable pose layer represents the set of spatial poses that allow the inspection equipment to effectively observe target equipment or equipment parts. The observable pose layer addresses whether the inspection equipment can see, clearly see, and perform detection actions. Observable pose nodes include spatial location, observation direction, sensor attitude, target observable surface, observation distance, observation angle, occlusion status, imaging quality level, and electrical safety margin. The electrical safety constraint layer represents safe distances to energized equipment, restricted areas, maintenance work areas, temporary fenced areas, and safety boundaries corresponding to different voltage levels. The inspection task layer represents inspection task templates, equipment aliases, inspection actions, inspection priorities, task clarification rules, and alternative task rules. The equipment capability layer represents the movement methods, passage dimensions, obstacle-crossing capabilities, observation height, sensor types, endurance status, communication status, and obstacle avoidance capabilities of equipment such as wheeled robots, quadruped robots, drones, and rail-mounted robots.

[0046] Through the above modeling, the system no longer just builds a regular access map, but a task-space coupling graph that can describe the relationship between "task semantics, device entities, observable poses, safety boundaries, and device capabilities".

[0047] S2 receives inspection task instructions via text, voice, or human-machine interface input, and parses the inspection task instructions into candidate structured task units through a vision-language-navigation model or a task parsing model. Each candidate structured task unit includes at least the target device, device location, detection action, desired observation method, task priority, safety constraints, and candidate inspection device. The vision-language-navigation model or task parsing model is only used to generate candidate structured task units and does not directly control the inspection device to perform navigation actions. The candidate structured task unit is represented as follows:

[0048] In the formula, U This represents a candidate structured task unit obtained by parsing text, speech, or human-computer interface input. o Indicates the target device; b Indicates the location of the target equipment; a Indicates a detection action; vIndicates the desired observation method; q Indicates task priority; c Indicates safety constraints; d This indicates candidate inspection equipment.

[0049] In this embodiment, the user issues inspection tasks via text input, voice input, or an HMI interface. For example, the user inputs: "Inspect the high-voltage side bushing of main transformer No. 1, and collect infrared and visible light images." The system first converts speech data into text data, and then uses a vision-language-navigation model or a task parsing model to parse the task into candidate structured task units: Target equipment: Main transformer No. 1; Equipment location: High-voltage side bushing; Detection actions: Infrared image acquisition, visible light image acquisition; Desired observation method: frontal observation or lateral observation; Task priority: routine inspection or high-priority inspection; Safety constraints: Meet the safety distance requirements for the current voltage level; Candidate inspection equipment: wheeled robots, quadruped robots, or drones.

[0050] In this embodiment, the vision-language-navigation model or the large AI model only outputs candidate structured task units and does not directly control the inspection equipment to execute them. These candidate structured task units must undergo subsequent task-space consistency verification before an executable navigation segment can be generated.

[0051] In other words, the large AI model or vision-language-navigation model in this embodiment is used to parse candidate tasks from text, voice, or HMI input and generate candidate structured task units or candidate navigation fragments. The large AI model or vision-language-navigation model does not directly output final control commands, nor does it directly drive the movement of inspection equipment. Its output must undergo task-space consistency verification, electrical safety consistency verification, and equipment capability consistency verification. Therefore, this embodiment controls the AI ​​model at the candidate generation layer, using substation safety rules, spatial consistency, and electrical safety boundaries as final execution constraints, thus improving the controllability and safety of the AI ​​model in substation inspection scenarios.

[0052] S3, perform task-space consistency verification between candidate structured task units and substation task-space coupling diagram; task-space consistency verification is not an accuracy verification between multi-sensor perception data, but an execution consistency verification between natural language task semantics and substation spatial entities, electrical bays, observable poses, electrical safety boundaries and inspection equipment capabilities. S4, Task-Space Consistency Verification includes equipment entity consistency verification, equipment location consistency verification, electrical bay consistency verification, observable pose consistency verification, passage space consistency verification, electrical safety consistency verification, and equipment capability consistency verification. When all verifications pass, the executable navigation segment generation stage is allowed. When any verification fails, direct issuance of navigation execution commands is prohibited, and task clarification information, alternative inspection plans, equipment replacement execution suggestions, safety waiting strategies, or safety exit strategies are generated based on the verification failure type. Equipment entity consistency verification matches the target equipment name, equipment number, equipment alias, and electrical bay information in the candidate structured task unit with the equipment ledger data and primary wiring topology data to determine if the target equipment uniquely corresponds to the target. If multiple candidate equipment entities exist, task clarification information containing the candidate equipment name, bay to which it belongs, spatial location, and confirmation question is generated. Equipment location consistency verification matches the equipment location in the candidate structured task unit with the equipment component model in the equipment entity layer to determine if the equipment location has a corresponding observable surface. If no corresponding observable surface exists, alternative observation locations or manual confirmation suggestions are generated. Electrical bay one... Consistency check compares the electrical interval of the target equipment with the interval information explicitly or implicitly included in the task command. If they are inconsistent, navigation execution commands are prohibited from being generated, and an interval conflict warning is output. Observable pose consistency check searches for candidate observation poses in the observable pose layer based on the observable surfaces of the target equipment, inspection actions, and sensor types, determining if any candidate observation poses satisfy the observation direction, observation distance, imaging angle, and sensor operating range. Access space consistency check searches the access space layer based on the candidate observation poses and the current position of the inspection equipment. Candidate travel paths are used to determine whether they meet the requirements for the travel size, obstacle crossing ability, and movement mode of the inspection equipment. Electrical safety consistency verification is used to calculate the distance margin between the inspection equipment and live equipment, restricted areas, and temporary work areas based on the candidate travel paths and candidate observation poses, and to determine whether it meets the safety distance requirements under the corresponding voltage level. Equipment capability consistency verification is used to determine whether the current inspection equipment can complete the corresponding navigation segment and inspection action based on the movement mode, travel size, maximum observation height, sensor configuration, remaining endurance, communication status, and obstacle avoidance ability of the inspection equipment.

[0053] S5, after the task-space consistency verification passes, retrieves candidate observable poses in the observable pose layer based on the observable surface corresponding to the target equipment location. Combining the movement mode, passage size, obstacle-crossing capability, sensor type, observation height, endurance status, and communication status of different inspection equipment, it selects candidate navigation segments from the candidate observable poses that match the capabilities of the inspection equipment. Candidate navigation segments include an approach segment, an observation segment, a local obstacle avoidance segment, and a safe exit segment. The approach segment controls the inspection equipment to move from its current position to a candidate observation area near the target equipment. The observation segment controls the inspection equipment to maintain its observation posture and perform actions such as image acquisition, infrared temperature measurement, meter recognition, sound acquisition, or partial discharge detection. The local obstacle avoidance segment performs local detours, waiting, or posture adjustment when affected by dynamic obstacles in the approach or observation segment. The safe exit segment controls the inspection equipment to exit to a safe position when the task is completed, the risk exceeds limits, or the task cannot continue. Each candidate navigation segment is bound to corresponding equipment actions, spatial node positions, sensor actions, observation postures, and safety boundaries. The task-space consistency check result is represented as follows:

[0054] In the formula: K This indicates the overall result of the task-space consistency check; K e This indicates the result of the device entity consistency verification. K b This indicates the results of the equipment part consistency verification. K i This indicates the result of the electrical bay consistency verification. K o This indicates the observable pose consistency verification result; K p This indicates the consistency check result of the passage space; K s This indicates the electrical safety conformity verification result; K d This indicates the result of the equipment capability consistency verification.

[0055] in:

[0056] when hour: K= A value of 1 indicates that the inspection task meets the task-space consistency requirements, allowing entry into the executable navigation fragment generation stage; when any verification item is 0: K=0 indicates that the inspection task cannot be executed directly. The system prohibits the direct issuance of navigation execution commands and generates task clarification information, alternative inspection plans, equipment replacement execution suggestions, safety waiting strategies, or safety exit strategies based on the verification failure type.

[0057] In other words, when the overall task-space consistency check result K is 1, the system allows entry into the navigation segment generation and execution phase. When the overall task-space consistency check result K is 0, the system prohibits direct execution of navigation actions and outputs equipment entity clarification information, equipment location clarification information, electrical bay conflict information, alternative observation pose, passage obstruction warning, safety risk warning, or equipment replacement execution suggestion based on the type of check failure.

[0058] For the target equipment part b Candidate observable poses p The observable qualified results are expressed as follows:

[0059] In the formula: Indicates candidate observable pose p For the target equipment part b Observable and acceptable results; This indicates that the line of sight is unobstructed. This indicates that the observed distance is within acceptable limits. This indicates that the observed angle is within acceptable limits. This indicates that the sensor's operating range is within acceptable limits. This indicates that the electrical safety distance is within acceptable limits. ;when At that time, it indicates the candidate observable pose. p Meet the target equipment parts b Effective observation requirements; when At that time, it indicates the candidate observable pose. p If the requirements for effective observation are not met, the system will either remove the candidate observable pose or generate an alternative observation scheme.

[0060] In this embodiment, the system performs a task-space consistency check on the candidate structured task units and the task-space coupling graph. This consistency check is not an accuracy check between multi-sensor perception data, but rather an execution consistency check between the natural language task semantics and the substation spatial entities, electrical bays, observable poses, electrical safety boundaries, and equipment capabilities.

[0061] Entity consistency verification is used to determine whether the No. 1 main transformer is uniquely matched in the equipment ledger, equipment alias table, and spatial model. If multiple candidate equipment entities exist, the system generates clarification information, such as: "Please select the No. 1 main transformer body, the No. 1 main transformer 220kV side surge arrester, or the No. 1 main transformer high voltage side bushing for inspection." The part consistency check is used to determine whether the high-voltage side bushing belongs to the target equipment entity and to determine whether there is a corresponding observable surface for that part.

[0062] The bay consistency check is used to determine whether the electrical bay to which the target equipment belongs is consistent with the bay information explicitly or implicitly included in the task instruction. For example, if the user instruction specifies a 220kV bus tie bay, but the target equipment actually belongs to a 110kV bay, the system will output a bay conflict warning and prohibit execution.

[0063] Observable pose consistency verification is used to determine whether there are candidate observation poses that meet the requirements of observation direction, observation distance, imaging angle, and sensor operating range. For example, if a wheeled robot can reach the vicinity of a target device, but the target part is occluded or the height is insufficient to complete the observation, the observable pose verification for that device will fail.

[0064] The passage space consistency check is used to determine whether there is a path between the current position and the candidate observation pose that meets the device's motion mode, passage size and obstacle crossing capability.

[0065] Electrical safety consistency verification is used to determine whether candidate paths and candidate observation poses meet the requirements for live safety distances, restricted areas, and temporary work areas.

[0066] Equipment capability consistency verification is used to determine whether the current inspection equipment has the sensor capabilities, observation altitude, endurance, and communication capabilities required to complete the task.

[0067] When all consistency checks pass, the system allows the generation of an executable navigation fragment. When any check fails, the system does not directly issue navigation commands, but instead generates task clarification information, alternative inspection plans, equipment replacement execution suggestions, safety waiting policies, or safety exit policies based on the type of failure.

[0068] In this embodiment, the system models the passage space layer and the observable pose layer in parallel.

[0069] The passage space layer describes the locations and connections where inspection equipment can walk, fly, or dock. Its core evaluation criteria are whether the path is connected, whether it meets the equipment's passage dimensions, whether there are obstacles, and whether it is possible to reach the vicinity of the target.

[0070] The observable pose layer describes whether the inspection equipment can effectively observe the target equipment part in a certain spatial position and attitude. Its core evaluation is whether the observation surface is visible, whether the observation distance is appropriate, whether the observation angle meets the requirements, whether the sensor field of view covers the target part, whether the imaging quality meets the detection action, and whether the pose meets the electrical safety requirements.

[0071] For a candidate observable pose p of the target device part b, the system determines whether it meets the following conditions: First, the line of sight between p and the target equipment part b is not obstructed; Second, the distance between p and the target device part b is within the effective working distance range of the corresponding sensor; Third, the angle between the observation direction corresponding to p and the normal of the observable surface of the target equipment part b is less than the set threshold. Fourth, the posture of the inspection equipment corresponding to p can be achieved by the current inspection equipment; Fifth, p meets the safety distance requirements under the current voltage level.

[0072] In this way, the present invention can distinguish between reachable locations and observable poses. Even if a location is passable, it will only be considered as a candidate pose for an executable observation segment if the observation angle, observation distance, sensor field of view, and safety boundary are met.

[0073] In this embodiment, after task-space consistency verification, the system does not directly output the complete path. Instead, it breaks down the inspection task into an approach segment, an observation segment, a local obstacle avoidance segment, and a safe exit segment. The approach segment controls the inspection equipment to move from its current position to a candidate observation area near the target equipment. The observation segment controls the inspection equipment to maintain its observation posture and performs actions such as image acquisition, infrared thermometry, meter reading, sound acquisition, or partial discharge detection. The local obstacle avoidance segment is used to perform local detours, deceleration, posture adjustment, or reselection of the observation position when the approach or observation segment is affected by dynamic obstacles. The safe exit segment controls the inspection equipment to exit to a safe position when the task is completed, the risk exceeds limits, or the task cannot continue. Each navigation segment includes a start point, an end point, path nodes, corresponding inspection equipment, sensor actions, target observation posture, safety boundary, and abnormal exit conditions.

[0074] For example, for wheeled inspection robots, the approach segment is bound to ground road nodes and passage width constraints; the observation segment is bound to gimbal angle, infrared camera movement, and visible light camera movement; and the safe exit segment is bound to the nearest safe waiting point. For drones, the approach segment is bound to flight altitude, no-fly zones, and obstacle avoidance radius; the observation segment is bound to hovering posture, gimbal pitch angle, and safe distance; and the safe exit segment is bound to return altitude and return path.

[0075] In this embodiment, the system matches the capabilities of different inspection devices based on candidate navigation segments. For tasks where the ground is flat and the inspection target is at a normal inspection height, the system prioritizes wheeled inspection robots. For areas with steps, ditches, or temporary obstacles, the system prioritizes quadruped robots. For targets that are not observable from the ground, such as high-altitude structures, busbar connection points, or targets not observable from the ground, the system selects unmanned aerial vehicles (UAVs). For periodic inspection tasks within a fixed track coverage area, the system may select rail-mounted robots. If a single device cannot complete all navigation segments, the system can split the same inspection task into multiple segments, which are then executed by different devices. For example, a ground robot completes the approach segment and the low-altitude observation segment, while a UAV completes the high-altitude observation segment. Both share target device entity and observable surface information through a task-space coupling graph.

[0076] S6. A comprehensive evaluation is performed on the candidate navigation segments to obtain the evaluation value of the navigation segments, and the candidate navigation segments whose evaluation values ​​meet the set conditions are selected to form the target navigation task. The comprehensive evaluation includes at least the task matching degree, drivability, observability, electrical safety margin, dynamic obstacle risk, equipment capability adaptability and task execution cost. S7, during the target navigation task performed by the inspection equipment, collects the position and pose of the inspection equipment, environmental perception data, target equipment identification results and safety distance status in real time, and compares the real-time collected results with the information bound to the target navigation segment for execution consistency. S8, when the consistency comparison result shows that there is any of the following situations: target equipment entity mismatch, observation pose deviation, passage area obstruction, dynamic obstacle entry, insufficient safety distance or insufficient equipment capacity, local replanning, task clarification, safe waiting, equipment replacement, or safe exit control is performed according to the anomaly type. The evaluation value of a navigation segment is determined as follows:

[0077] In the formula: Indicates inspection equipment j Execute candidate navigation fragments Evaluation value; Indicates candidate navigation segments; j Indicates candidate inspection equipment; Indicates candidate navigation fragments Task matching degree with the current inspection task; Indicates candidate navigation fragments Observability of the target equipment parts; Indicates candidate navigation fragments Electrical safety margin Indicates candidate navigation fragments Inspection equipment jEquipment capability compatibility; Indicates candidate navigation fragments The execution cost; Indicates candidate navigation fragments The risk of dynamic obstacles; arrive Indicates the weighting coefficient; .

[0078] Among them, electrical safety margin Determine as follows:

[0079] In the formula: x Indicates candidate navigation fragments Sampling position on; Indicates sampling position x The actual distance between the equipment and the boundary of the restricted area or temporary work area; Indicates the current voltage level u Corresponding safety distance requirements; u Indicates the voltage level corresponding to the target area or target equipment; when When, it indicates a candidate navigation segment. Meets electrical safety distance requirements; when When, it indicates a candidate navigation segment. If the electrical safety distance requirement is not met, the system will either remove the candidate navigation segment or generate a safe exit strategy.

[0080] Equipment capability adaptability According to the inspection equipment j The movement mode, passage width, obstacle crossing ability, maximum observation height, sensor configuration, remaining battery life, and communication status are determined as follows:

[0081] In the formula: Indicates candidate navigation fragments Inspection equipment j Equipment capability compatibility; Indicates the movement mode matching item; Indicates the pass size matching item; Indicates sensor capability matching items; This indicates the battery life matching option; Indicates the communication status matching item; to This represents the weighting coefficient for equipment capability adaptability. Where: ;and ;when When the value is less than the set capability adaptation threshold, it indicates that the inspection equipment... jCandidate navigation fragments are not suitable for execution The system generates recommendations for switching devices.

[0082] The results of the consistency comparison are expressed as follows:

[0083] In the formula: Y This indicates the result of the consistency comparison. This indicates the positional deviation between the real-time position of the inspection equipment and the position of a spatial node in the target navigation segment; This indicates the real-time entity matching deviation between the identified device entity and the target device entity; This indicates the observation attitude deviation between the real-time observation direction and the normal to the observable surface of the target; This indicates the deviation of the safety margin between the real-time safety distance and the electrical safety boundary; This indicates a state where passage is obstructed. When: If any of the above exceeds the corresponding threshold, the system will perform local replanning, task clarification, safe waiting, device switching, or safe exit control based on the exception type.

[0084] In this embodiment, during the inspection process, the system collects real-time data on device pose, environmental perception, target device identification results, and safety distance status, and compares these with the target navigation segment for consistency. If the system detects that the real-time identified device entity is inconsistent with the target device entity, it stops the observation segment and triggers task clarification. If the observed attitude deviates from the target observable surface but the safety distance meets the requirements, it performs attitude fine-tuning or local replanning. If the passage area is blocked by temporary fences, vehicles, personnel, or tools, it determines whether there is an alternative passage edge; if so, it performs local replanning; if not, it performs safe waiting or returns to a safe area. If the real-time safety distance is less than the electrical safety boundary, it immediately performs a safe exit. If the current device cannot continue execution due to insufficient obstacle crossing ability, observation height, sensor capability, endurance, or communication capability, it pauses the current device task and generates a device replacement execution suggestion.

[0085] S9. After the task is completed, collect the task instructions, task parsing results, target navigation fragments, actual running trajectory, execution consistency comparison results, replanning records, anomaly types, manual takeover records, and task completion quality to generate typified closed-loop samples. S10, based on the typified closed-loop samples, the substation task-space coupling graph is written back in layers; among them, task semantic ambiguity anomalies are written back to the inspection task layer, equipment entity binding anomalies are written back to the equipment entity layer, passage space anomalies are written back to the passage space layer, observable pose anomalies are written back to the observable pose layer, electrical safety boundary anomalies are written back to the electrical safety constraint layer, and equipment capability mismatch anomalies are written back to the equipment capability layer. This avoids treating different failure causes as the same type of training samples for model updates.

[0086] The categorized closed-loop sample is represented as:

[0087] In the formula: B Represents a typed closed-loop sample; U Indicates a candidate structured task unit; Indicates a target navigation segment; X Indicates the actual running trajectory; Y This indicates the result of the consistency comparison. Z Indicates the exception type; A This indicates the result of manual confirmation or the quality of task completion.

[0088] Among them, abnormal types Z include:

[0089] In the formula: This indicates an anomaly in the semantic ambiguity of the task. This indicates an error in the device entity binding; This indicates an abnormal passage space; This indicates an observable pose anomaly; This indicates an abnormal electrical safety boundary; This indicates an abnormality due to a mismatch between equipment capabilities.

[0090] The relationship between exception types and hierarchical write-back is represented as follows:

[0091] In the formula: This represents the hierarchical write-back object corresponding to exception type Z; Indicates the level of inspection tasks; Represents the device entity layer; Indicates the passage space layer; Indicates the observable pose layer; Indicates the electrical safety constraint layer; This indicates the device capability layer.

[0092] By establishing a one-to-one correspondence between the above-mentioned anomaly types and hierarchical write-back objects, the system avoids treating different failure reasons as similar training samples for model updates, thereby improving the accuracy of closed-loop feedback.

[0093] In this embodiment, after the task is completed, the system generates a typified closed-loop sample. This sample includes the original task instructions, candidate structured task units, target navigation segments, actual running trajectory, execution consistency comparison results, anomaly types, manual takeover records, and task completion quality.

[0094] The system performs layered write-back based on anomaly type. The first type is task semantic ambiguity anomalies, manifested as multiple candidate device entities corresponding to the same natural language device name, or inconsistencies between the device abbreviation / alias in the user command and the device ledger. This type of anomaly is written back to the inspection task layer to update device alias relationships, task templates, and clarification question templates. The second type is device entity binding anomalies, manifested as inconsistencies between the device entity identified by the inspection equipment upon arrival at the site and the target device. This type of anomaly is written back to the device entity layer to correct the binding relationships between device entity nodes and spatial coordinates, visual features, or device numbers. The third type is passage space anomalies, manifested as planned paths being blocked by fences, vehicles, personnel, tools, or temporary construction areas during execution. This type of anomaly is written back to the passage space layer to update the status of passable edges, temporary blockage markers, and passage costs. The fourth type is observable pose anomalies, manifested as the inspection equipment being able to reach the vicinity of the target, but failing to complete effective observation due to obstruction, insufficient angle, insufficient height, insufficient sensor field of view, or the target part being invisible. The first type of anomaly is written back to the observable pose layer to update observable edges, observation quality levels, and candidate observation pose states. The fifth type is electrical safety boundary anomalies, which manifest as the distance between the candidate path or candidate observation pose and energized equipment, restricted areas, or work areas not meeting safety requirements. This type of anomaly is written back to the electrical safety constraint layer to update safety boundaries, restricted areas, and risk costs. The sixth type is equipment capability mismatch anomalies, which manifest as the current inspection equipment being unable to complete the task due to insufficient obstacle-crossing capability, passage size, maximum observation height, sensor type, endurance, or communication status. This type of anomaly is written back to the equipment capability layer to update the corresponding equipment capability parameters and task allocation rules.

[0095] By using the above-mentioned categorized and layered write-back method, the system avoids the problem of inaccurate feedback caused by inputting all failed samples into the model for training, making subsequent task parsing, path generation, observation pose selection, and device matching more targeted.

[0096] In summary, this embodiment provides a task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations. This method collects three-dimensional spatial data of the substation, equipment ledgers, primary wiring topology, inspection routes, live-line safety distances, historical trajectories, and inspection equipment capability data. It constructs a task-space coupling graph consisting of an equipment entity layer, a passage space layer, an observable pose layer, an electrical safety constraint layer, an inspection task layer, and an equipment capability layer. The observable pose layer and the passage space layer are set up side-by-side to determine whether the inspection equipment can effectively observe the target equipment location at a given spatial position and orientation. The system parses the inspection task input via text, voice, or HMI into candidate structured task units and performs consistency checks on equipment entities, equipment locations, electrical bays, observable poses, passage space, electrical safety, and equipment capabilities. This consistency check is not a multi-sensor data accuracy check, but rather an execution consistency check between the natural language task semantics and the substation spatial entities. After successful verification, the system generates an executable navigation segment consisting of an approach segment, an observation segment, a local obstacle avoidance segment, and a safe exit segment. Upon completion of the task, a categorized closed-loop sample is generated and written back to the task-space coupling graph in layers according to anomaly type. This invention improves the reliability, controllability, and continuous optimization capability of transforming natural language inspection tasks into safe and executable navigation actions in complex substation environments.

[0097] Example 2 To further illustrate the feasibility of the technical solution of this invention and its resulting technical effects, this embodiment constructs a non-limiting inspection test scenario in a substation digital twin simulation platform. This scenario corresponds to a typical equipment area of ​​a 220kV substation, containing 186 equipment entity nodes, 413 access space nodes, and 1262 candidate observable pose nodes, and is configured with four types of inspection equipment: wheeled inspection robots, quadrupedal inspection robots, drones, and rail-mounted robots. An example of the six-layer nodes and relationships in the task-space coupling graph is shown below. Figure 2 As shown.

[0098] This embodiment sets 120 inspection task instructions, including 60 normal tasks, 15 tasks with semantic ambiguity in equipment names or parts, 15 tasks with conflicts between equipment entities and electrical partitions, 15 tasks with obstructed passage or obscured observable poses, and 15 tasks with inadequate electrical safety boundaries or equipment capabilities. The observation angle threshold for candidate observable poses is set to 35°, the position deviation trigger threshold is set to 0.35m, the observation attitude deviation trigger threshold is set to 12°, and the equipment capability adaptation threshold for candidate navigation segments is set to 0.70.

[0099] For the task of "inspecting the high-voltage side bushing of main transformer No. 1 and acquiring infrared and visible light images," the system retrieved seven candidate observable poses (p1 to p7) around the target equipment. P3 was eliminated due to structural obstruction, P4 due to insufficient electrical safety margin, and P7 due to the observation angle exceeding a threshold. The remaining candidate poses underwent comprehensive evaluation, and P2 was ultimately selected as the target observation pose. A navigation segment consisting of an approach segment, an observation segment, and a safe exit segment was generated. The results of candidate pose selection and approach segment path generation are as follows: Figure 3 As shown.

[0100] exist Figure 3 In the implementation results shown, the observation distance between the selected observation pose p2 and the target device is 2.7m, the angle between the observation direction and the normal of the target observable surface is 18°, and the safety margin relative to the minimum safe distance for the corresponding voltage level is 0.62m. The generated approach path length is 13.6m, the path does not enter the restricted area, and it is able to maintain the target attitude of the infrared camera and the visible light camera in the observation section.

[0101] Furthermore, the system calculates the equipment capability adaptability of the four types of inspection equipment to the approach section, observation section, and safe exit section, respectively, and the results are as follows: Figure 4 As shown, the wheeled inspection robot has a fit of 0.92 and 0.88 for the approach and safe exit sections, respectively, but its fit for the observation section is only 0.58 due to insufficient observation height; the UAV has a fit of 0.95 for the observation section. Therefore, the system assigns the approach and safe exit sections to the wheeled inspection robot and the high-altitude observation section to the UAV, thus forming a cross-device collaborative target navigation task.

[0102] At 44 seconds into the task execution, a dynamic obstacle entered the original approach path in the simulation scenario, and the normalized values ​​of the position deviation and observation attitude deviation exceeded the anomaly trigger threshold. The consistency comparison module triggered local replanning, and the alternative passage edge switch was completed at 54 seconds. After replanning, both normalized deviations decreased to below 0.60, and the inspection equipment did not breach the electrical safety boundary throughout the entire process. The results of the consistency comparison and local replanning response are as follows: Figure 5 As shown.

[0103] To compare the technical effects, this embodiment uses the conventional "task-path-execution" method, which does not perform task-space consistency pre-verification and directly plans the path based solely on the task parsing results, as a control. Tests were conducted on 120 simulated inspection tasks. The effective task completion rate of the method of this invention increased from 75.8% to 94.2%, the invalid observation rate decreased from 16.7% to 3.3%, the safety boundary alarm rate decreased from 9.2% to 0.8%, and the manual takeover rate decreased from 22.5% to 6.7%. The results are as follows: Figure 6As shown, this demonstrates that pre-checking for task-space consistency, independent modeling of observable poses, and matching of device capabilities can reduce the likelihood of erroneous tasks directly entering the execution phase.

[0104] This embodiment also writes back task semantic ambiguity anomalies, device entity binding anomalies, passage space anomalies, observable pose anomalies, electrical safety boundary anomalies, and device capability mismatch anomalies to their respective levels. After five consecutive rounds of typified layered writeback, the task semantic unique binding rate increased from 86% to 97%, the effective observable pose selection rate increased from 82% to 96%, and the first execution success rate increased from 78% to 95%. Continuous optimization results are as follows: Figure 7 As shown.

[0105] The simulation environment, number of tasks, thresholds, and result values ​​described above are used to illustrate the implementation process and technical effects of the present invention and are non-limiting examples. In actual engineering applications, the corresponding parameters and values ​​can be adjusted or replaced according to the substation voltage level, equipment layout, inspection equipment model, and actual test data.

[0106] Example 3 This embodiment proposes a task-space consistency autonomous navigation system for multiple intelligent inspection devices in substations. It implements the task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations described in the previous embodiment. The system includes: a data acquisition module, a task-space coupling graph construction module, a task parsing module, a task-space consistency verification module, a navigation fragment generation module, a device capability matching module, a target navigation task generation module, an execution consistency comparison module, and a closed-loop write-back module. The data acquisition module is used to collect three-dimensional spatial data of the substation, equipment ledger data, primary wiring topology data, inspection road data, live-line safety distance data, historical inspection trajectory data, and capability data of various types of inspection equipment. The task-space coupling graph construction module is used to construct a task-space coupling graph consisting of an equipment entity layer, a passage space layer, an observable pose layer, an electrical safety constraint layer, an inspection task layer, and an equipment capability layer. The task parsing module is used to parse inspection task instructions input via text, voice, or human-machine interface into candidate structured task units. The task-space consistency verification module is used to verify the consistency of candidate structured task units with the task-space coupling graph in terms of equipment entity consistency, equipment location consistency, electrical interval consistency, observable pose consistency, passage space consistency, electrical safety consistency, and equipment capability consistency. If the task-space consistency verification fails, the candidate navigation segment generation stage is prohibited. The system generates feedback strategies based on the failure type of the verification. If the task-space consistency verification passes, it enters the candidate navigation segment generation stage. The navigation segment generation module generates candidate navigation segments, including approach segments, observation segments, local obstacle avoidance segments, and safe exit segments, after the task-space consistency verification passes. It also binds the corresponding equipment actions, spatial node positions, sensor actions, observation attitudes, and safety boundaries to the candidate navigation segments. The equipment capability matching module filters candidate navigation segments based on the capabilities of the inspection equipment. The target navigation task generation module comprehensively evaluates the candidate navigation segments to obtain navigation segment evaluation values ​​and selects candidate navigation segments whose evaluation values ​​meet the set conditions to form the target navigation task. The execution consistency comparison module compares the consistency between the real-time status of the inspection equipment and the target navigation segments during task execution. The closed-loop write-back module performs layered write-back of the task-space coupling graph based on typified closed-loop samples.

[0107] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0108] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A task-space consistent autonomous navigation method for multiple intelligent inspection devices in substations, comprising the following steps: S1: Collect multi-source data from the substation and construct a task-space coupling diagram of the substation; S2: Receive the input inspection task instruction and parse the inspection task instruction into candidate structured task units; S3: Perform task-space consistency verification between the candidate structured task units and the substation task-space coupling diagram; if the task-space consistency verification fails, the candidate navigation segment generation stage is prohibited, and a feedback strategy is generated based on the verification failure type; if the task-space consistency verification passes, candidate observable poses are retrieved in the task-space coupling diagram based on the observable surfaces corresponding to the target equipment parts, and candidate navigation segments that match the inspection equipment capabilities are selected from the candidate observable poses; and the corresponding equipment actions, spatial node positions, sensor actions, observation attitudes, and safety boundaries are bound to the candidate navigation segments. S4: Perform a comprehensive evaluation of the candidate navigation segments to obtain the navigation segment evaluation value, and select the candidate navigation segments whose navigation segment evaluation values ​​meet the set conditions to form the target navigation task; S5: The inspection equipment performs target navigation tasks, collects the inspection equipment pose, environmental perception data, target equipment identification results and safety distance status in real time, and performs an execution consistency comparison with the equipment actions, spatial node positions, sensor actions, observation attitude and safety boundaries bound to the target navigation segment; when the execution consistency comparison result shows that there is an anomaly, it performs local replanning, task clarification, safety waiting, equipment switching or safe exit control according to the anomaly type. S6: After the inspection task is completed, collect the inspection task process data and generate categorized closed-loop samples. S7: Perform hierarchical back-writing of the substation task-space coupling diagram based on the typified closed-loop samples.

2. The task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to claim 1, characterized in that: In step S1, the substation multi-source data includes substation three-dimensional spatial data, equipment ledger data, primary wiring topology data, inspection road data, live-line safety distance data, historical inspection trajectory data, and capability data of various types of inspection equipment; the substation task-space coupling diagram includes an equipment entity layer, a passage space layer, an observable pose layer, an electrical safety constraint layer, an inspection task layer, and an equipment capability layer; wherein, the passage space layer is used to describe whether the inspection equipment can reach a certain spatial location, and the observable pose layer is set in parallel with the passage space layer to describe whether the inspection equipment can effectively observe the target equipment part under a certain spatial location and attitude; The substation task-space coupling diagram is represented as follows: in, This diagram illustrates the task-space coupling of a substation. Represents a set of device entity nodes; Represents the set of nodes in the travel space; Represents the set of observable pose nodes; Represents the set of inspection task nodes; This represents the set of capability nodes for inspection equipment; Indicates a passable connecting edge; Represent observable relation edges; Indicates electrical safety constraint edges; This indicates a device capability matching edge.

3. The task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to claim 1, characterized in that: In step S2, the inspection task instruction is parsed into candidate structured task units through a vision-language-navigation model or a task parsing model; the candidate structured task unit includes at least the target equipment, equipment location, detection action, expected observation method, task priority, safety constraints, and candidate inspection equipment; The candidate structured task unit is represented as follows: in, U This represents a candidate structured task unit obtained by parsing text, speech, or human-computer interface input. o Indicates the target device; b Indicates the location of the target equipment; a Indicates a detection action; v Indicates the desired observation method; q Indicates task priority; c Indicates safety constraints; d This indicates candidate inspection equipment.

4. The task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to claim 1, characterized in that: In step S3, the task-space consistency verification includes equipment entity consistency verification, equipment part consistency verification, electrical interval consistency verification, observable pose consistency verification, passage space consistency verification, electrical safety consistency verification, and equipment capability consistency verification. Specifically, equipment entity consistency verification determines whether the target equipment name, equipment number, equipment alias, and the entity in the equipment register uniquely correspond; equipment part consistency verification determines whether a corresponding observable surface exists for the target equipment part; electrical interval consistency verification determines whether the interval to which the target equipment belongs is consistent with the interval information in the task instruction; observable pose consistency verification determines whether there are candidate observable poses that meet the inspection action requirements; passage space consistency verification determines whether there are candidate passage paths that meet the equipment's passage capability; electrical safety consistency verification determines whether the candidate passage paths and candidate observable poses meet the live-line safety distance requirements; and equipment capability consistency verification determines whether the candidate inspection equipment possesses the necessary motion capability, sensor capability, endurance capability, and communication capability to complete the task.

5. The task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to claim 1, characterized in that: In step S3, the task-space consistency check result is expressed as follows: in, ; K This indicates the overall result of the task-space consistency check; K e This indicates the result of the device entity consistency verification. K b This indicates the results of the equipment part consistency verification. K i This indicates the result of the electrical bay consistency verification. K o This indicates the observable pose consistency verification result; K p This indicates the consistency check result of the passage space; K s This indicates the electrical safety conformity verification result; K d This indicates the result of the equipment capability consistency verification. K= 1 indicates that the inspection task meets the task-space consistency requirements, allowing it to enter the executable navigation fragment generation stage; K =0 indicates that the inspection task cannot be executed directly.

6. The task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to claim 1, characterized in that: In step S3, the feedback strategy includes task clarification information, alternative inspection plan, equipment replacement execution suggestion, safe waiting strategy and safe exit strategy; The capabilities of the inspection equipment include movement mode, passage size, obstacle crossing capability, sensor type, observation altitude, endurance status, and communication status; the candidate navigation segments include an approach segment, an observation segment, a local obstacle avoidance segment, and a safe exit segment; wherein, the approach segment is used to control the inspection equipment to move from its current position to a candidate observation area near the target equipment; The observation section is used to control the inspection equipment to maintain its observation posture and perform actions such as image acquisition, infrared temperature measurement, meter identification, sound acquisition, or partial discharge detection; the local obstacle avoidance section is used to perform local detour, waiting, or posture adjustment when the approach or observation section is affected by dynamic obstacles; the safe exit section is used to control the inspection equipment to exit to a safe position when the task is completed, the risk exceeds the limit, or the task cannot be continued.

7. The task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to claim 1, characterized in that: In step S4, the comprehensive evaluation includes task matching degree, feasibility, observability, electrical safety margin, dynamic obstacle risk, equipment capability adaptability, and task execution cost. The evaluation value of the navigation segment is determined as follows: in, Indicates inspection equipment Execute candidate navigation fragments Evaluation value; Indicates candidate navigation segments; Indicates candidate inspection equipment; Indicates candidate navigation fragments Task matching degree with the current inspection task; Indicates candidate navigation fragments Observability of the target equipment parts; Indicates candidate navigation fragments Electrical safety margin; Indicates candidate navigation fragments Inspection equipment Equipment capability compatibility; Indicates candidate navigation fragments The execution cost; Indicates candidate navigation fragments The risk of dynamic obstacles; arrive Indicates the weighting coefficient; .

8. The task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to claim 1, characterized in that: In step S5, the consistency comparison result is expressed as follows: in, Y This indicates the result of the consistency comparison. This indicates the positional deviation between the real-time position of the inspection equipment and the position of a spatial node in the target navigation segment; This indicates the real-time entity matching deviation between the identified device entity and the target device entity; This indicates the observation attitude deviation between the real-time observation direction and the normal to the observable surface of the target; This indicates the deviation of the safety margin between the real-time safety distance and the electrical safety boundary; This indicates that passage is obstructed.

9. A task-space consistency autonomous navigation method for multiple intelligent inspection devices in substations according to claim 1, characterized in that: In step S6, the task process data includes task instructions, task parsing results, target navigation segments, actual running trajectory, execution consistency comparison results, replanning records, exception types, manual takeover records, and task completion quality. The typed closed-loop sample is represented as follows: in, B Represents a typed closed-loop sample; U Indicates a candidate structured task unit; Indicates a target navigation segment; X Indicates the actual running trajectory; Y This indicates the result of the consistency comparison. Z Indicates the exception type; A This indicates the result of manual confirmation or the quality of task completion; Exception types Z Represented as: in, This indicates an anomaly in the semantic ambiguity of the task. This indicates an error in the device entity binding; This indicates an abnormal passage space; This indicates an observable pose anomaly; This indicates an abnormal electrical safety boundary; This indicates an abnormality due to a mismatch between equipment capabilities.

10. A task-space consistent autonomous navigation system for multiple intelligent inspection devices in substations, characterized in that, include: The data acquisition module is used to collect multi-source data from the substation. The task-space coupling graph construction module is used to construct the task-space coupling graph of the substation. The task parsing module is used to receive the input inspection task instructions and parse the inspection task instructions into candidate structured task units. The task-space consistency verification module is used to perform task-space consistency verification between candidate structured task units and the substation task-space coupling diagram. If the task-space consistency verification fails, the candidate navigation segment generation stage is prohibited, and a feedback strategy is generated based on the verification failure type. If the task-space consistency verification passes, the candidate navigation segment generation stage is entered. The navigation fragment generation module is used to retrieve candidate observable poses in the task-space coupling graph based on the observable surface corresponding to the target device part after the task-space consistency verification is passed, generate candidate navigation fragments, and bind the corresponding device actions, space node positions, sensor actions, observation attitudes and safety boundaries to the candidate navigation fragments; The equipment capability matching module filters candidate navigation segments that match the capabilities of the inspected equipment from the candidate observable poses. The target navigation task generation module is used to comprehensively evaluate candidate navigation segments, obtain navigation segment evaluation values, and select candidate navigation segments whose navigation segment evaluation values ​​meet the set conditions to form a target navigation task. The consistency comparison module is used to collect the pose of the inspection equipment, environmental perception data, target equipment identification results, and safety distance status in real time when the inspection equipment is performing target navigation tasks. It then performs an execution consistency comparison with the equipment actions, spatial node positions, sensor actions, observation attitude, and safety boundaries bound to the target navigation segment. When the execution consistency comparison result indicates an anomaly, it performs local replanning, task clarification, safe waiting, equipment switching, or safe exit control according to the anomaly type. The closed-loop write-back module is used to collect task process data after the inspection task is completed, generate categorized closed-loop samples, and perform layered write-back of the substation task-space coupling diagram based on the categorized closed-loop samples.

Images