Real-time Early Warning System and Method for Substation Anomalies Based on Humanoid Inspection Robot

The substation anomaly real-time early warning system using a humanoid inspection robot solves the problem of limited walking speed and battery life of inspection robots in substations, enabling real-time early warning and continuous inspection under special circumstances, and ensuring the timely detection of substation equipment anomalies and the timely transmission of alarm information.

CN123122022APending Publication Date: 2026-11-17STATE GRID LIAONING ELECTRIC POWER CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611456286.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing humanoid inspection robots have limitations in walking speed, battery life, and continuous operation capability during substation inspections, resulting in task turnaround time delays. They are unable to balance the efficiency of full station coverage with the requirements of real-time early warning, and are prone to interruption under special circumstances, affecting the continuity of inspections and the timeliness of alarms.

Method used

A real-time early warning system for substation anomalies based on a humanoid inspection robot is adopted. It includes a communication module, an early warning processing module, an alarm output module, a background early warning platform, and a collaborative support module. Through incremental transmission, minimum control surface priority strategy, dual-mode cycle switching, and link anomaly scoring, it ensures that the reachability of key alarm information and the rapid access to anomaly locations can still be maintained when communication fluctuates or the link is interrupted.

Benefits of technology

It can maintain the accessibility of critical alarm information even when communication fluctuates or the link is interrupted, shorten the time delay of anomaly detection, improve the efficiency of anomaly confirmation, reduce the probability of fault stagnation, and improve the continuity and real-time performance of inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122022A_ABST
    Figure CN123122022A_ABST
Patent Text Reader

Abstract

A real-time early warning system and method for substation anomalies based on a humanoid inspection robot. The system includes a humanoid inspection robot, a communication module, an early warning processing module, an alarm output module, a backend early warning platform, and a collaborative support module. The humanoid inspection robot collects inspection data to obtain fused mapping information, and uses an incremental transmission mechanism on the uplink side to achieve switching between the minimum control plane and the data plane. Based on the fused mapping information, anomaly early warning information is generated, converted into alarm information, and pushed to the operation and maintenance terminal. The anomaly early warning information is further classified and released; a dual-mode cycle switching is set up for regular inspection cycle and alarm priority cycle; and a weighted fusion of multiple anomaly indicators is implemented, and the link anomaly level is obtained based on the fused data, generating a link anomaly event and triggering a self-rescue handling process, determining the target handling action, and issuing control commands. This invention improves the continuity of the alarm link and the timeliness of anomaly detection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of electrical technology, specifically relating to a real-time early warning system and method for substation anomalies based on a humanoid inspection robot. Background Technology

[0002] As a crucial hub in the power system, substations operate with their primary and secondary equipment constantly energized. Issues such as equipment defects, poor contact at joints leading to localized overheating, abnormal noises, and environmental anomalies are often sudden and difficult to detect. To improve inspection efficiency and reduce the time and risks associated with manual inspections, a solution is proposed: employing humanoid inspection robots to perform substation inspection tasks.

[0003] However, when existing humanoid inspection robots complete multi-view data acquisition and data encapsulation at target equipment, their walking speed, battery life, and continuous operation capabilities are limited, which can easily lead to task turnaround delays when it is necessary to cover a large number of equipment points or to quickly reach abnormal points. It is difficult to balance the efficiency of full-site coverage and the requirements of real-time early warning.

[0004] Furthermore, during the inspection process, special circumstances such as malfunctions, channel blockages, communication fluctuations, or insufficient energy can easily cause the inspection task to be interrupted. When inspecting in high-temperature weather or near high-temperature equipment, the robot's body temperature rise may cause sensor drift, control unit frequency reduction, or protection shutdown, making the detection of high-temperature equipment more prone to interruption, affecting the continuity of inspection and the timeliness of alarms. This, in turn, affects the continuous output and real-time delivery of abnormal warning and alarm information, and can also easily lead to the need for frequent manual intervention, affecting automation and intelligence. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a real-time early warning system and method for substation anomalies based on a humanoid inspection robot.

[0006] The first aspect of the present invention discloses a real-time early warning system for substation anomalies based on a humanoid inspection robot. The system includes: a humanoid inspection robot, a communication module, an early warning processing module, an alarm output module, a background early warning platform, and a collaborative support module. The humanoid inspection robot collects inspection data to construct real-scene frames, and performs feature matching with a preset 3D model to obtain fused mapping information. The communication module uses an incremental transmission mechanism on the uplink side based on inspection data and fusion mapping information. When the communication quality deteriorates, the minimum control plane priority strategy is triggered. When the communication quality recovers, the data plane transmission is restored and the buffer retransmission is triggered. The early warning processing module generates abnormal early warning information based on inspection data and fused mapping information; The alarm output module converts abnormal warning information into alarm information and pushes it to the operation and maintenance terminal; The aforementioned back-end early warning platform manages inspection tasks and classifies and releases abnormal early warning information. The collaborative support module includes two types of control logic: normal cycle support and special situation support. The normal cycle support switches between two modes: regular inspection cycle mode and alarm priority cycle mode. The special situation support performs weighted fusion of multiple abnormal indicators to obtain a link abnormality score and outputs the link abnormality level based on the link abnormality score. It generates a link abnormality event and triggers a self-rescue process, determines the target handling action, and issues control commands.

[0007] Further, the fusion mapping information is obtained, including: By acquiring multiple frames of images, after distortion removal, time synchronization and keyframe filtering, a three-dimensional Gaussian splash representation is constructed based on the keyframe depth information or point cloud information. The rendering perspective is determined according to the pose of the humanoid inspection robot and synthesized into a real scene frame. From the real scene frame, feature matching is performed on the equipment outline corner points, wiring end points and the 1:1 3D model in the station to obtain the translational deviation and rotational deviation between the real scene frame and the model, and the alignment error is calculated. The 3D Gaussian splash representation is transformed to the coordinate system of the 1:1 3D model within the station, and the fusion mapping information is output.

[0008] Furthermore, the fusion mapping information includes device identifier, point identifier, real-scene frame index, robot pose index, position of abnormal area in real-scene frame, corresponding position of abnormal area in 1:1 3D model in the station, and alignment error identifier. When the alignment error is less than or equal to the preset error threshold, output the fusion mapping information; When the alignment error exceeds a preset error threshold, an alignment correction process is triggered, which includes repositioning, reselecting keyframes, or switching adjacent point references.

[0009] Furthermore, for the uplink side, an incremental transmission mechanism is adopted. The humanoid inspection robot uses the most recently successfully uploaded device reference packet as the baseline on the edge side. After performing quality filtering and anomaly screening on the currently collected data, it only sends the data determined to be abnormal and its associated real-scene frame index to the uplink, as well as the key normal data summary extracted according to the device importance weight, point coverage interval and energy budget, and the unconfirmed incremental packets cached during the link interruption; normal non-key frames do not enter the uplink data plane.

[0010] Further, the aforementioned anomaly screening is performed, specifically as follows: First, determine whether the image or real-scene frame meets the requirements for sharpness, occlusion ratio, jitter, and lighting based on the quality label field. If the conditions are not met, the data is marked as low quality. If the conditions are met, the equipment status parameters are compared with the historical reference range of the corresponding equipment, and the presence of anomalies is determined by combining the results of appearance defect identification, temperature change, gas concentration and acoustic anomaly identification. When the equipment status parameters exceed the historical benchmark range, the appearance defect identification result reaches the appearance abnormality condition, the temperature change result reaches the preset temperature rise condition, the gas concentration result reaches the preset concentration condition, or the acoustic abnormality result reaches the acoustic abnormality condition, it is judged that there is an abnormality, and the corresponding data is marked as abnormal data. Otherwise, the humanoid inspection robot will select data from the normal state of the equipment according to the importance of the equipment, the time interval for data collection, and the coverage requirements of the data points.

[0011] Furthermore, the communication module is configured as follows: When heartbeats are continuously lost, end-to-end transmission delay continuously exceeds the first preset delay threshold, packet loss rate continuously exceeds the first preset packet loss threshold, uplink bandwidth is lower than data plane transmission requirements, or buffered data fails to be retransmitted continuously, communication quality is determined to be degraded. When the communication quality deterioration continues for a first preset number of times or a first preset duration, and there are currently incomplete inspection tasks, incomplete retest tasks, link abnormal events, or pending control commands, the minimum control plane priority strategy is triggered; after the communication quality is restored, data plane transmission is performed and buffer retransmission is triggered.

[0012] Furthermore, the dual-mode beat switching includes: When an alarm priority cycle is triggered, the abnormal point is inserted at the beginning of the point residence acquisition sequence, and the residence time of low-risk points is shortened or the channel coverage frequency is reduced, provided that the time window constraint is met; when the arrival time window constraint is not met, the correction scheduling is triggered. The correction scheduling includes adjusting the coverage order, changing the travel path, or delaying the start time of the next point's dwell time. In the dual-mode cycle switching, the conditions for triggering the alarm priority cycle include: the warning level reaches the threshold, the time limit for the abnormal point retest is less than the preset threshold, or the link abnormality level reaches the medium level but the retest can still be performed; the conditions for restoring the normal cycle include: the abnormal point retest is completed and the alarm status is updated to confirmed or deactivated, and the link abnormality level drops back to the low level and continues for the preset duration.

[0013] Further, the link anomaly level is obtained, including: The link anomaly score is obtained by weighting the communication anomaly index, location anomaly index, energy anomaly index, task anomaly index, fault anomaly index, and channel anomaly index. The link anomaly score is compared with a preset set of grading thresholds to obtain the link anomaly level and generate a corresponding link anomaly event. The communication anomaly indicators are determined based on the number of heartbeat loss, the number of times the end-to-end latency exceeds the threshold, and the number of times the packet loss rate continuously exceeds the threshold. The location anomaly index is determined based on the number of times the location reliability is below the threshold and the number of times the location drift continuously exceeds the threshold. The energy anomaly index is determined based on the remaining energy being below a threshold and the expected insufficient range. The task anomaly indicators are determined based on task timeouts and execution phase stalls. The fault abnormality indicators are determined based on the results of drive, braking, joint, sensor, or self-test. The channel anomaly indicators are determined based on channel blockage, insufficient passage width, and the number of failed escape attempts.

[0014] Further, the link anomaly event is generated, including: The preset set of hierarchical thresholds includes a first threshold, a second threshold, and a third threshold, wherein the first threshold < the second threshold < the third threshold; If the link anomaly score is less than the first threshold, the link anomaly level is normal. If the link anomaly score is greater than or equal to the first threshold and less than the second threshold, the link anomaly level is mild anomaly level. If the link anomaly score is greater than or equal to the second threshold and less than the third threshold, then the link anomaly level is moderate. If the link anomaly score is greater than or equal to the third threshold, the link anomaly level is classified as severe anomaly. Based on the link anomaly level, combined with the anomaly type, occurrence time, humanoid inspection robot pose, current task stage and associated alarm identifier corresponding to the detected multiple anomaly indicators, the link anomaly event is generated and the self-rescue process is triggered; the link anomaly event is stored in association with the fault evidence package, retest record and alarm status after the link is restored. The self-rescue procedure includes getting out of trouble and backing up, safely stopping, retesting and retrying, and requesting manual intervention.

[0015] Furthermore, the generation of the fault evidence package specifically involves: When the link anomaly level reaches the preset level or a fault anomaly is detected, the humanoid inspection robot generates a fault evidence package. The fault evidence package includes event identifier, associated alarm identifier, device identifier, location identifier, timestamp, robot pose, anomaly type set, summary of the most recent key inspection data, most recent preset duration state sequence, current task stage, list of completed collection items and list of incomplete collection items; When the link is available, the communication module prioritizes reporting the fault evidence packet. When the link is unavailable, it writes the fault evidence packet and the corresponding real-scene frame into the local cache and retransmits them after the link is restored.

[0016] Further, determining the target action and issuing control commands includes: The comprehensive cost is calculated by weighting the time delay cost, risk cost, and energy cost required to execute the candidate disposal action; The candidate action corresponding to the minimum comprehensive cost is selected from the set of candidate actions and taken as the target action. The latency cost is calculated based on the ratio of the estimated total time required to execute the candidate action to the alarm recovery time. The estimated total time includes the movement time, the action execution time, and the control command transmission and status feedback time. The upper limit of the latency cost is 1. The risk cost is calculated by weighting the robot's own risk value, the device's neighborhood risk value, and the channel risk value. The energy cost is calculated based on the ratio of the estimated total energy consumption required to perform the candidate disposal action to the robot's current remaining energy. The estimated total energy consumption includes movement energy consumption, execution energy consumption, and communication and computing energy consumption, and the upper limit of the energy cost is 1. The control commands issued include stopping, reversing, and retesting.

[0017] Furthermore, based on the inspection data, anomaly warning information is generated, including: The early warning processing module, based on inspection data, compares real-time scene frames with historical baseline images, and combines the coordinate alignment relationships corresponding to equipment identification, point identification, robot pose, and spatial point cloud to identify defects, difference areas, and confidence levels. It also combines multimodal sensor data, preset criteria, time-series algorithms, and historical degradation curves to determine anomalies and output anomaly determination results. The abnormal warning information is classified into four levels: normal, mild, moderate and severe.

[0018] The second aspect of this invention discloses a real-time early warning method for substation anomalies based on a humanoid inspection robot, which implements the technical solution of the real-time early warning system for substation anomalies as described in the first aspect of this invention, including: S1: The humanoid robot collects inspection data to construct real-world scene frames, and performs feature matching with the preset 3D model to obtain fused mapping information; S2: Based on inspection data and fusion mapping information, an incremental transmission mechanism is adopted for the uplink side. When the communication quality deteriorates, the minimum control plane priority strategy is triggered. When the communication quality recovers, the data plane transmission is restored and the buffer retransmission is triggered. S3: Generates anomaly warning information based on inspection data and fused mapping information; S4: Convert the generated abnormal warning information into alarm information and push it to the operation and maintenance terminal; S5: Manage inspection tasks and classify and issue abnormal early warning information; S6: Normal cycle time ensures dual-mode cycle time switching between regular inspection cycle time mode and alarm priority cycle time mode; the special situation ensures weighted fusion of multiple abnormal indicators to obtain a link abnormality score and outputs the link abnormality level based on the link abnormality score, generates a link abnormality event and triggers the self-rescue handling process, determines the target handling action and issues control commands.

[0019] The beneficial effects of this invention are that, compared with the prior art, 1. This invention establishes a dual-mode switching mechanism between a regular inspection cycle mode and an alarm priority cycle mode in the inspection scheduling, so that abnormal points are prioritized and trigger a retest closed loop in the alarm priority cycle mode; at the same time, combined with incremental transmission, minimum control plane keep-alive and buffer retransmission mechanism under link interruption, the key alarm information is still available and the evidence data can be supplemented when communication fluctuates or the link is interrupted, thereby reducing the probability of alarm link interruption and shortening the abnormal arrival delay.

[0020] 2. This invention generates a point-based collection sequence and a path coverage inspection route, and imposes constraints on coverage and energy budget. At the same time, it judges the execution deviation based on the arrival time window and triggers the correction scheduling, so that the robot can still suppress the task turnaround time when covering a large number of equipment points. When an anomaly occurs, it prioritizes the completion of key evidence collection through a phased evidence supplementation method of retesting and retrying, and then completes the full collection when conditions permit, thereby improving the efficiency of anomaly confirmation without sacrificing full station coverage.

[0021] 3. This invention integrates fault anomalies, channel blockages, location anomalies, communication anomalies, and energy anomalies into an event-based handling process through a link anomaly scoring and handling strategy library. It also combines fault evidence packages with handling result association records to achieve an executable closed loop of rollback, entry into a safe docking point, retesting and retrying, and remote takeover. Furthermore, it sets up a detection session suspension and breakpoint recovery mechanism for high-temperature environments or high-temperature equipment neighborhood inspections. When the machine body temperature exceeds the limit, the detection of high-temperature equipment is suspended and switched to the low-temperature equipment location. After the machine body temperature drops, the unfinished data collection items are restored, thereby reducing the probability of fault stagnation caused by high temperature and improving continuous operation capability. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the system flow of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a logical diagram of the collaborative protection module of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this invention are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0024] As one embodiment of the present invention, such as Figure 1 As shown, a real-time early warning system for substation anomalies based on a humanoid inspection robot is disclosed. The system includes: a humanoid inspection robot, a communication module, an early warning processing module, an alarm output module, a backend early warning platform, and a collaborative support module. The humanoid inspection robot is used to collect inspection data and upload the collected inspection data to the backend early warning platform through the communication module. The communication module is used for inspection data interaction and link protection between the humanoid inspection robot and the backend early warning platform. The early warning processing module is used to generate anomaly early warning information based on the inspection data and output it to the alarm module and the collaborative support module. The backend early warning platform is used to manage substation inspection tasks and classify and publish anomaly early warning information. The collaborative support module is used to execute dual-mode cycle scheduling and special situation protection decisions based on the anomaly early warning information. The backend early warning platform sends scheduling instructions to the humanoid inspection robot through the communication module. The alarm output module is used to convert the anomaly early warning information into alarm information and push it to the operation and maintenance terminal, and respond to retest trigger actions.

[0025] The humanoid inspection robot is used to collect inspection data associated with target equipment identification and time information at the substation site and report its own status information.

[0026] In one embodiment, the humanoid inspection robot includes a mobility mechanism, a computing control unit, a multi-source inspection sensor assembly, and a real-scene image generation unit; the multi-source inspection sensor assembly includes two or more of a binocular camera, an infrared thermal imager, a gas sensor, and an acoustic sensor.

[0027] In this embodiment, the real-scene image generation unit is used to generate real-scene image frames corresponding to the three-dimensional model in the station and establish a binding relationship with the inspection data. The real-scene image generation unit specifically includes: a binocular imaging unit to acquire multiple frames of images of the target equipment location; a calculation and control unit to perform distortion correction, time synchronization and key frame filtering on the images; and a Gaussian splash representation to be constructed based on the depth information or point cloud information of the key frames, representing the scene as multiple three-dimensional Gaussian elements.

[0028] The computational control unit combines the robot pose corresponding to the keyframes to unify the spatial 3D points obtained from different keyframes into the coordinate system of the station's 3D model, and performs outlier removal and duplicate point merging on the unified spatial 3D points to obtain the spatial point cloud of the target equipment location.

[0029] The computational control unit constructs a 3D Gaussian splash representation of the target device location based on keyframe depth information or point cloud information; determines the rendering viewpoint according to the robot pose and binocular installation pose, and projects and synthesizes the 3D Gaussian elements into a real-world frame under the current viewpoint; uses this real-world frame as a registration medium to extract the device contour corner points, wiring end points, etc., and match them with the corresponding features in the 1:1 3D model in the station to obtain the translational and rotational deviations between the real-world frame and the model, and calculates the alignment error; and accordingly, transforms the 3D Gaussian splash representation to the coordinate system of the 1:1 3D model in the station and outputs the fusion mapping information.

[0030] When the alignment error is less than or equal to the preset error threshold, the calculation control unit outputs the fusion mapping information; when the alignment error is greater than the preset error threshold, the calculation control unit triggers the alignment correction process, which includes one or more of repositioning, reselecting keyframes, or switching adjacent point references; the fusion mapping information includes device identifier, point identifier, real-scene frame index, robot pose index, the position of the abnormal area in the real-scene frame, the corresponding position of the abnormal area in the 1:1 3D model in the station, and the alignment error identifier, which are used to associate the alarm information with the abnormal area in the real-scene frame and the local position of the device in the 1:1 3D model in the station.

[0031] The binding of the inspection data and the real-scene frame is achieved through structured encapsulation. The encapsulated fields include at least: device identifier, location identifier, timestamp, parameter field set, real-scene frame index, and quality identifier field. The parameter field set includes device status parameter field, image acquisition parameter field, environmental parameter field, robot status parameter field, and link status parameter field. The quality identifier field is used to mark factors that affect availability, such as occlusion ratio, jitter, and insufficient lighting.

[0032] The communication module is used to exchange task information and inspection data between the background early warning platform and the humanoid inspection robot.

[0033] In one implementation, the background early warning platform and the humanoid inspection robot establish a two-way link through a dedicated mobile communication network; the communication module organizes the transmission by message type: uplink messages are used to carry inspection data and execution status information, and downlink messages are used to carry inspection tasks, retest instructions and collaborative control instructions.

[0034] The uplink side adopts an incremental transmission mechanism: the humanoid inspection robot compresses the images and real-world frames at the edge and performs initial screening for anomalies, uploading only abnormal data and key normal data.

[0035] The incremental transmission mechanism involves the robot at the edge using the most recently successfully uploaded device baseline packet as a baseline. After performing quality filtering and initial anomaly screening on the currently collected data, it only sends data determined to be abnormal and its associated real-world frame index to the uplink, along with key normal data summaries extracted according to device importance weight, point coverage interval, and energy budget, as well as unconfirmed incremental packets cached during link interruptions. Normal non-critical frames do not enter the uplink data plane. The synchronization period is preset to be triggered upon completion of resident data collection or a 3–10s heartbeat, configured by the backend.

[0036] The initial anomaly screening specifically involves a humanoid inspection robot first determining whether an image or real-world frame meets requirements for clarity, occlusion ratio, jitter, and lighting based on quality identification fields. Data that does not meet these requirements is marked as low-quality data. For data that meets the quality requirements, the humanoid inspection robot further compares the device status parameters with the historical baseline range of the corresponding device, and combines this with the results of appearance defect identification, temperature change, gas concentration, and acoustic anomaly to determine if any anomalies exist. If the device status parameters exceed the historical baseline range, the appearance defect identification results meet the appearance anomaly conditions, the temperature change results meet the preset temperature rise conditions, the gas concentration results meet the preset concentration conditions, or the acoustic anomaly results meet the acoustic anomaly conditions, then an anomaly is identified and the corresponding data is marked as abnormal data. Otherwise, no anomalies are identified. The humanoid inspection robot selects data from the normal state of the device according to the importance of the device, the collection time interval, and the point coverage requirements. The synchronization cycle is executed according to a preset cycle. When the link is interrupted or the communication quality deteriorates, the robot switches to local caching mode to cache inspection data, real-world frame indexes, fusion mapping information, and status sequences. The state sequence is a conventional time-series log in which the inspection robot records its own operating status in chronological order. This invention uses a set of fields including timestamp, robot pose, current task stage, target device location, remaining battery power, walking status, sensor working status, communication latency, packet loss rate, cache status, and fault identifier. This state sequence is cached together with fusion mapping information and real-scene frame index for use in restoring the operation process before and after the link is restored and for assisting in the calculation of link anomaly scores.

[0037] To ensure control commands are reachable under special circumstances, the communication module is configured with a minimum control plane. The minimum control plane carries heartbeat, emergency stop, rollback, retest and retry, request for manual intervention triggers, and status summary fields. The status summary field is a simplified set of fields used to characterize the robot's current controllable state, task state, and link state when data plane transmission is restricted or large-capacity data uploads are paused. The status summary field includes robot identifier, timestamp, robot pose, current task stage, target device location, remaining battery power, communication quality level, cached data volume, most recent alarm identifier, most recent link anomaly event identifier, fault identifier, and pending control command identifier.

[0038] The communication module continuously acquires heartbeat return results, end-to-end transmission latency, packet loss rate, uplink bandwidth, and the number of consecutive retransmission failures, and compares these indicators with corresponding thresholds. When one or more of the following situations occur: continuous heartbeat loss, end-to-end transmission latency continuously exceeding the latency threshold, packet loss rate continuously exceeding the packet loss rate threshold, uplink bandwidth lower than the data plane transmission requirement, or continuous retransmission failures of cached data, communication quality is determined to be degraded. When the communication quality degraded state continues for a preset number of times or for a preset duration, and there are one or more situations such as incomplete inspection tasks, incomplete retest tasks, link abnormal events, or pending control commands, when the switching conditions are met, the communication module enters the minimum control plane priority strategy, suspends or reduces the data plane transmission priority of image, real-scene frame, and inspection data fragment, and prioritizes the transmission of heartbeat, emergency docking, backoff, retest retry, request for manual handling trigger, and status summary fields. When the heartbeat recovers stably, the end-to-end transmission latency is lower than the recovery threshold, the packet loss rate is lower than the recovery threshold, and the preset recovery duration is reached, the communication module determines that the recovery conditions are met, resumes data plane transmission, and triggers cached retransmission.

[0039] The early warning processing module is used to generate anomaly judgment results based on inspection data, and generate anomaly early warning information accordingly.

[0040] In one implementation, the early warning processing module calls the historical database to perform comparative analysis; the historical database stores equipment operation data, fault case data, and Gaussian splashing real-world historical archives associated with equipment locations over the past three years.

[0041] The early warning processing module includes an appearance defect recognition submodule and a trend early warning submodule. The appearance defect recognition submodule uses a lightweight convolutional neural network model to compare real-time scene frames with historical benchmark images. Specifically, the background early warning platform retrieves the corresponding historical benchmark image based on the device identifier, location identifier, and scene frame index, and performs region alignment between the real-time scene frames and the historical benchmark image based on fusion mapping information to form a real-time comparison area and a historical benchmark area. The appearance defect recognition submodule inputs the real-time comparison area and the historical benchmark area into the lightweight convolutional neural network model to extract the graphs from both. The system identifies and distinguishes features, outputting defect identification results, difference region information, and confidence information. The defect identification results include the presence and type of appearance defects; the difference region information identifies the location and range of appearance defects within real-time scene frames; and the confidence information indicates the reliability of the defect identification results. The lightweight convolutional neural network model has no more than one million parameters. The trend warning submodule uses a time-series algorithm to model continuous inspection indicators such as temperature and vibration, and predicts abnormal trends by referencing historical degradation curves of similar equipment. Trend prediction can provide degradation warnings two to five days in advance.

[0042] The trend warning module is triggered based on a weekly window temperature increment threshold, satisfying the following formula: ; ; in, For equipment In the window Temperature increment within; For a moment Temperature value; The temperature value at the start of the window; This is the temperature trend warning threshold. 1.5℃ is acceptable; If the value is 1, a trend warning will be triggered; otherwise, it will not be triggered.

[0043] The early warning processing module outputs anomaly determination results and generates anomaly early warning information. The anomaly early warning information includes anomaly type, associated device identifier, occurrence time, anomaly location identifier, early warning level, and confidence level. The early warning level characterizes at least one of the severity, urgency, or processing priority of the anomaly determination result. The early warning processing module can determine the early warning level based on at least one of the following: anomaly type, importance of associated devices, anomaly location, confidence level, historical inspection or retest results, and environmental interference.

[0044] The abnormal location marker is associated with the real-view frame index and is used to locate the local abnormal area of ​​the device on the display interface.

[0045] The alarm output module is used to output abnormal warning information as alarm information and push it to the operation and maintenance terminal and / or the on-site alarm device.

[0046] The alarm output module is used to output abnormal warning information as alarm information and push it to the operation and maintenance terminal and / or the on-site alarm device; the alarm information includes alarm identifier, associated device identifier, abnormality type, warning level, occurrence time, abnormal location identifier and evidence index information; the evidence index information is used to point to the inspection data segment or key real-scene frame associated with the alarm.

[0047] The background early warning platform is used to manage substation inspection tasks and to classify and release abnormal early warning information.

[0048] The background early warning platform overlays abnormal location markers on the display interface to form a heat map of abnormal locations; the abnormal location markers can be generated in combination with robot location information and can use ultra-wideband positioning to achieve centimeter-level positioning accuracy to support accurate re-measurement of abnormal points.

[0049] When maintenance personnel trigger a retest on the alarm display interface, the alarm output module associates the retest trigger action with the corresponding alarm identifier and records it. The background early warning platform sends a retest instruction to the humanoid inspection robot, and the robot returns to the corresponding location to collect and transmit the retest data again. After the retest result is returned, the alarm output module updates the alarm status to confirmed or deactivated, and archives the alarm confirmation result and handling record.

[0050] The collaborative protection module is used to protect and control risks that affect the continuity and real-time performance of the alarm link during the inspection and scheduling process of the background early warning platform.

[0051] The collaborative support module includes two types of control logic: normal cycle time support and special situation support. The normal cycle time support is used to generate a point-based data collection sequence and path coverage for the humanoid inspection robot based on the inspection task. During the execution of the humanoid inspection robot, it switches between a normal inspection cycle time mode and an alarm-priority cycle time mode. Figure 3 As shown.

[0052] Conditions for triggering alarm priority beats: The warning level reaches the threshold, or the time limit for retesting the abnormal point is less than the preset threshold, or the link abnormality level reaches the medium level but retesting can still be performed.

[0053] Conditions for restoring normal rhythm: The abnormal point is retested and the alarm status is updated to confirmed or cleared, and the link abnormality level drops back to a low level and continues for the preset duration.

[0054] In the regular inspection cycle mode, the collaborative support module applies coverage constraints to the path coverage inspection route within a preset time window, and applies energy budget constraints to the energy consumption of the point-based collection sequence and the path coverage inspection route. In the alarm priority cycle mode, the collaborative support module increases the retest priority of abnormal points and adjusts the update frequency of the path coverage inspection route according to the warning level of the abnormal warning information. It inserts abnormal points at the beginning of the point-based collection sequence, and shortens the residence time of low-risk points or reduces the coverage frequency of some channels while meeting the time window constraints.

[0055] The collaborative support module generates an arrival time window based on the predicted dwell time and collection duration, and determines whether the arrival time window constraint is met based on the expected or actual arrival time of the humanoid inspection robot. When the arrival time window constraint is not met, a correction scheduling is triggered. The correction scheduling includes one or more of the following: adjusting the coverage order, changing the travel path, or delaying the start time of the next point's dwell time.

[0056] The special circumstances protection is used to obtain a link anomaly score by weighted fusion of multiple anomaly indicators when a fault anomaly, mission anomaly, positioning anomaly, communication anomaly, energy anomaly, or channel anomaly is detected. Based on the link anomaly score, the link anomaly level is output, a link anomaly event is generated, and a self-rescue handling process is triggered. The self-rescue handling process includes reversing from a set of candidate handling actions such as getting out of trouble, entering a safe docking point, retesting and retrying, and requesting manual handling. The target handling action or a combination of target handling actions is determined from the set of candidate handling actions, and corresponding control commands are issued.

[0057] The link anomaly event is a business-level anomaly signal generated by the collaborative support module after determining the link anomaly level. It is used to characterize that the overall operational support status of the robot has reached the corresponding anomaly level. The link anomaly event can serve as the basis for triggering self-rescue procedures, switching the linkage communication module to the minimum control plane priority strategy, associating and archiving it with fault evidence packages and alarm records, and retransmitting it after the link is restored.

[0058] Link anomaly scoring The calculation formula is: ; in, Scoring of communication anomaly indicators; Scoring of abnormal indicators; Scoring of energy anomaly indicators; Scoring of task anomaly indicators; Scoring of fault and abnormal indicators; Scoring of channel anomaly indicators; , , , , , These are the corresponding weight coefficients, and + + + + + =1, such as =0.1, =0.3, =0.1, =0.2, =0.2, =0.1; such as It is of a high level. It is of medium level. It is a low-level item.

[0059] The number of heartbeats lost, the number of times the end-to-end latency exceeded the threshold, and the number of times the packet loss rate continuously exceeded the threshold were determined. The determination is based on the number of times the location reliability is below the threshold and the number of times the location drift continuously exceeds the threshold; Determined based on remaining energy being below a threshold and insufficient expected range; Determined based on task timeout and execution phase stagnation; Determined based on the self-test results of the drive, braking, joint, sensor, or computing control unit, and includes abnormal posture and fall events; The link anomaly level is determined based on channel congestion, insufficient passage width, and the number of failed escape attempts; the link anomaly level is determined by... The abnormal events in the link are obtained by mapping with the set of hierarchical thresholds and generating the abnormal events in the link. The abnormal events in the link include event identifier, event type, occurrence time, robot position, current task stage and associated alarm identifier.

[0060] The hierarchical threshold set includes a first threshold. Second threshold and the third threshold If the link is abnormal, a score will be given. If the link anomaly level is normal, then the link anomaly level is normal. If the link anomaly level is low, the link anomaly level is classified as mild. If the link anomaly level is moderate, then the link anomaly level is moderate. If the link anomaly level is not met, the link anomaly level is classified as severe. For example, the first threshold is 0.3, the second threshold is 0.6, and the third threshold is 0.85.

[0061] After triggering the self-rescue procedure, the collaborative support module determines the target action from a set of candidate actions, including escape and retreat, entering a safe stopping point, retesting and retrying, and requesting manual intervention, based on the link anomaly level, current task status, robot status, and environmental communication conditions. The comprehensive cost of the candidate actions satisfies the following formula: ; in, For candidate disposal actions; As a comprehensive cost; "Delay cost" refers to the expected alarm recovery delay or alarm continuity impact delay after executing candidate handling actions. Risk cost is used to characterize the safety risks to equipment and humanoid inspection robots; Energy cost, used to characterize the consumption of range; , , These are the corresponding weight coefficients, and ,like =0.4, =0.3, =0.3. The collaborative support module updates the weighting coefficients based on the on-site risk level, alarm continuity requirements, and remaining energy status, and selects the candidate action with the lowest overall cost as the target action.

[0062] The delay cost The time required to execute the candidate action is determined based on the estimated time required and the allowed recovery time for the alarm, satisfying the following formula: ; in, The movement time corresponding to the candidate action is determined based on the path length from the robot's current position to the target position and the speed limit. The duration of the action execution is read from the action timetable according to the type of action. The number of retreat segments and the duration of a single retreat segment correspond to the escape and retreat. The duration of the docking action corresponds to the entry into the safe docking point. The number of retest data collection items and the duration of a single data collection item correspond to the retest and retry. The duration of the remote takeover establishment corresponds to the request for manual handling. To control the duration of command transmission and status feedback, it is determined based on the round-trip delay and retransmission count of the current communication link; Allowed recovery time for alarm links; This indicates taking the smaller value.

[0063] The action schedule is pre-stored or dynamically maintained as a reference execution time for different action types. For example, the retreat mechanism can be determined based on the number of retreat segments and the duration of each segment; an example of a single retreat segment duration could be 3 seconds or determined based on the robot's movement speed and safe retreat distance. The retest and retry mechanism can be determined based on the number of retest data acquisition items and the duration of each item; an example of a single item duration could be 5 seconds or determined based on the time consumed by image / infrared / acoustic data acquisition. Entering a safe docking point can be determined based on the docking action duration, which may include the time consumed by deceleration, alignment, stabilization, and attitude adjustment. Requesting manual intervention can correspond to the remote takeover establishment time, which can be determined based on the time consumed by establishing the human-machine interaction link, authentication, and transfer of control.

[0064] The risk cost Based on robot safety risks, equipment neighborhood risks, and passageway risks, the following formula is satisfied: ; in, The risk value of the robot body is determined based on the robot's self-inspection results, posture status, body temperature, and remaining braking capacity. The risk value for the vicinity of the equipment is determined based on the safe distance from the energized area of ​​the target equipment, the temperature of the vicinity of high-temperature equipment, and the alarm level of the equipment. The channel risk value is determined based on the channel blockage status, channel width, location confidence, and number of escape failures. , and These are the corresponding risk weight coefficients, and ,like =0.2, =0.5, =0.3; Each risk value can be assigned 0.2, 0.6 and 1 for low risk, medium risk and high risk respectively, and determined by the background early warning platform based on the sensor detection results and the station safety constraint table.

[0065] The energy cost The energy consumed in performing the candidate action is determined based on the robot's remaining energy, satisfying the following formula: ; in, The energy consumption for performing the candidate disposal action is determined based on the path length, robot movement power, and movement time. The energy consumption for action execution is determined based on the execution power and execution duration corresponding to the action type. The energy consumption for communication and computing is determined based on the amount of data uploaded, the amount of data retransmitted, the duration of computing tasks, and the power of the communication module. This represents the robot's current remaining energy.

[0066] The collaborative assurance module calculates the comprehensive cost of each candidate action in the candidate action set. The system selects the candidate action with the lowest overall cost as the target action. When the difference in overall cost between two candidate actions is less than a preset threshold, the selection is made in the order of safety priority, alarm continuity priority, and energy priority. The selection result is written into the link anomaly event and archived in conjunction with the fault evidence package, the handling result, and the alarm status.

[0067] The above self-rescue and response procedures further include the following measures: When the link anomaly level reaches the preset level or a fault anomaly is detected, the humanoid inspection robot generates a fault evidence package. The fault evidence package includes: event identifier, associated alarm identifier, device identifier, location identifier, timestamp, robot pose, anomaly type set, summary of the most recent key inspection data, the most recent preset duration state sequence, current task stage, list of completed collection items, and list of incomplete collection items. The communication module prioritizes reporting the fault evidence package when the link is available, and writes the fault evidence package and the corresponding real-scene frame to the local cache when the link is unavailable, and retransmits it after the link is restored. The background early warning platform associates and stores the fault evidence package with link anomaly events, alarm records, retest records, and handling records.

[0068] The communication module divides the link content into a data plane and a minimum control plane; the data plane is used to transmit inspection data, real-time image frames, and analysis results; the minimum control plane is used to transmit heartbeat, emergency stop, backoff, retest and retry, manual intervention request triggers, and status summary fields; when When the preset threshold is reached and the preset number of times is reached, the communication module switches to the minimum control plane priority strategy to ensure that control commands such as docking, backoff, and retesting remain accessible; when the communication quality is restored and the restoration conditions are met, the communication module resumes data plane transmission and triggers buffer retransmission.

[0069] The background early warning platform predefines a set of safe docking points in the 3D model of the station and configures docking constraints for each safe docking point. The configuration of docking constraints includes docking point location, allowed docking area, minimum channel width, safety distance, communication coverage status, and neighborhood risk level. When the link anomaly level reaches a high level or an energy anomaly is triggered, the collaborative protection module first performs accessibility screening on each safe docking point, eliminating safe docking points that have no feasible path, the minimum channel width of the path is less than the robot's passage width, the path has continuous obstacles blocking it, or the expected energy consumption to reach exceeds the remaining available energy. The remaining safe docking points are used as candidate docking points, and the target docking point is determined by combining channel accessibility, docking point risk cost, and expected alarm recovery delay. When assessing the accessibility and safety of candidate paths to safe docking points, at least the following should be considered: whether the path width meets the robot's passage requirements, whether there are persistent or unavoidable obstacles in the path, and whether the localization reliability meets the passage requirements. If any key condition is not met, the corresponding path or safe docking point will be excluded from the candidate range.

[0070] The risk cost of a stop is determined based on the distance between the safe stop and the live equipment, the temperature of the stop area, the ground condition, and the communication coverage. The higher the risk, the greater the risk cost of the stop. For candidate stops that pass the accessibility screening, the collaborative support module compares them in the following order: priority of route accessibility score, priority of lower stop risk cost, priority of shorter expected arrival delay, and priority of lower expected arrival energy consumption, and determines the target stop. For example, the set of safe docking points includes a first safe docking point, a second safe docking point, and a third safe docking point. The third safe docking point has a path width less than the robot's required width, resulting in a path width score of 0. Therefore, the third safe docking point is eliminated during the accessibility screening stage. The first safe docking point has a path accessibility score of 1, a medium-risk docking point risk cost, and an estimated arrival delay of 40 seconds. The second safe docking point has a path accessibility score of 1, a low-risk docking point risk cost, and an estimated arrival delay of 55 seconds. Since both the first and second safe docking points are passable, and the second safe docking point has a lower risk cost and its estimated arrival delay is still within the allowable range, the collaborative support module identifies the second safe docking point as the target docking point and issues a control command to enter the second safe docking point.

[0071] The collaborative protection module issues an instruction to enter the safe docking point and switches the robot's state to docking preservation state after the robot arrives; in the docking preservation state, heartbeat reporting and cached data transmission are still maintained to ensure the continuity of the evidence chain.

[0072] When an abnormality or local obstacle is detected that prevents further progress, the collaborative support module triggers a retreat action. The retreat adopts a segmented retreat method with a preset retreat step size. After each retreat segment, the passage's passability is reassessed. If passability cannot be restored after retreating to the preset number of segments, the collaborative support module switches to entering a safe stopping point or requesting manual intervention. The background early warning platform writes the reason for the retreat trigger, the number of retreat segments, the last reachable location, and the list of incomplete points into the task breakpoint record for subsequent risk avoidance and rearrangement during recovery inspections.

[0073] When an alarm has been generated and requires retesting for confirmation, but the retesting fails or the retesting data is missing, the collaborative assurance module triggers a retesting retry. The retesting retry includes two stages: rapid evidence completion retesting and complete item retesting. Rapid evidence completion retesting is used to prioritize the collection of key collection items directly related to alarm confirmation after the arrival of the anomaly point. Key collection items include one or more of the following: real-world image frames of the anomaly location, infrared temperature measurement data, abnormal sound clips, gas detection data, and robot pose information. After the key collection items are collected and the corresponding evidence index is generated, the background early warning platform determines whether the current link status, remaining energy, retesting completion time limit, and robot self-check status meet the conditions for complete item retesting. If the current link can transmit complete data or the local cache space meets the cache requirements and the remaining energy... If the energy threshold for retesting is exceeded, the retesting completion time limit still allows continued stay, and the robot is not in a fault or high-temperature protection state, the rapid evidence completion retest will switch to the complete item retest. The complete item retest is used to complete one or more of the following according to the retesting collection item list: multi-view image acquisition of the target equipment location, complete infrared scanning, acoustic acquisition, gas detection, and historical benchmark position re-photographing. During the complete item retest, if link quality deterioration, remaining energy below the threshold, insufficient retesting completion time limit, robot self-testing abnormality, or new high-priority alarm events occur, the complete item retest will be stopped, and the evidence index formed by the rapid evidence completion retest will be retained. The background early warning platform will write the incomplete collection items into the retesting breakpoint record, and the incomplete collection items will continue to be executed after the link, energy, or robot status recovers.

[0074] When a fault or location anomaly persists and automatic handling fails a preset number of times, the collaborative support module triggers a request for manual handling. The background early warning platform pushes the link anomaly event to the operation and maintenance terminal and opens the remote takeover entry. In the remote takeover state, the background early warning platform sets domain and speed limits on the operation commands that can be issued, allowing only operations such as moving, turning, stopping, and reversing to be performed within the safe area and safe speed range. The background early warning platform writes the takeover start time, takeover end time, takeover operation sequence, and handling results into the handling record.

[0075] When the remaining energy is below the energy threshold or the expected range is insufficient to complete the inspection of the next target location, the collaborative support module triggers the action of entering a safe docking point or returning to the supply station; the background early warning platform obtains the charging progress of the supply station via Bluetooth and forms the estimated recovery time; the background early warning platform breaks down the inspection task into completed segments and incomplete segments, alarms are given priority for reordering the incomplete segments, and the resumption inspection task is issued after charging is completed, while triggering cache retransmission and retesting to complete the task.

[0076] The aforementioned special circumstance protection measures further include the logic for handling intermittent high-temperature detection and suspension / recovery: The calculation and control unit acquires the machine body temperature and the temperature of the surrounding area of ​​the equipment; the machine body temperature is obtained from the machine body temperature sensor; the temperature of the surrounding area of ​​the equipment is obtained from the infrared thermal imaging unit or the ambient temperature sensor; when the machine body temperature reaches a preset pause threshold or the temperature of the surrounding area of ​​the equipment reaches a preset ambient temperature threshold, a high-temperature risk event is generated and reported or cached and re-transmitted; the high-temperature risk event is recorded. And used to update the risk cost of candidate disposal actions. .

[0077] When a high-temperature risk event is triggered and the current task includes the detection of high-temperature equipment locations, the background early warning platform sets the detection session corresponding to the high-temperature equipment location to a suspended state and generates a suspension record. The suspension record includes: equipment identifier, location identifier, session identifier, list of completed collection items, list of incomplete collection items, set of recent real-scene frame indexes, summary of recent inspection parameter fields, machine body temperature, temperature of the equipment's neighboring area, suspension timestamp, current task stage and breakpoint position. The suspension record is written to the task breakpoint record and associated with the cache retransmission mechanism.

[0078] After suspension, the collaborative support module temporarily suspends the high-temperature equipment points from the point-based collection sequence, rearranges the point-based collection sequence according to the low-temperature equipment point set and task priority, and controls the humanoid inspection robot to perform low-temperature equipment detection; during the interruption, it continuously reports heartbeat and status summary, and transmits the collected data, suspension records and fault evidence packages according to the incremental transmission and cache retransmission mechanism.

[0079] When the robot's body temperature drops to the preset recovery threshold and remains stable for a preset duration, the background early warning platform reads the suspension record and issues a breakpoint recovery command. The preset recovery threshold is determined based on the allowable operating temperature of the robot's internal key components, the measurement error of the temperature sensor, the historical temperature range of normal inspections, and the safety margin. The preset recovery threshold is lower than the preset pause threshold to avoid frequent suspension and recovery within the critical temperature range. The preset stabilization duration is determined based on the robot's heat dissipation inertia and the temperature sampling cycle. It can be set to ensure that the robot's body temperature is below the preset recovery threshold for multiple consecutive sampling cycles, or to ensure that the temperature remains below the preset recovery threshold for two to five minutes. The breakpoint recovery command carries a session identifier, a target point identifier, a list of incomplete data collection items, and a recovery completion time limit. The robot returns to the corresponding high-temperature equipment point to continue executing the incomplete data collection items.

[0080] Before resuming execution, the computational control unit performs consistency verification on the sensors related to high temperature detection, and performs one or more of the following when the verification fails: re-acquisition, recalibration, or switching to alternative acquisition items; the recovered segment data sets segment identifiers and establishes an associated index with the suspended segment data, so that the alarm evidence index and traceability query remain segmentable and associative.

[0081] The collaborative protection module associates and records abnormal events in the link with the results of handling actions, and outputs them to the alarm output module, so that the alarm display interface presents the abnormal link alarms and handling status.

[0082] As an embodiment of the present invention, a real-time early warning method for substation anomalies based on a humanoid inspection robot is disclosed, employing the specific implementation method described in the above-mentioned real-time early warning system for substation anomalies, with reference to... Figure 2 .

[0083] S1: The humanoid robot collects inspection data to construct real-world scene frames, and performs feature matching with the preset 3D model to obtain fused mapping information; S2: Based on inspection data and fusion mapping information, an incremental transmission mechanism is adopted for the uplink side. When the communication quality deteriorates, the minimum control plane priority strategy is triggered. When the communication quality recovers, the data plane transmission is restored and the buffer retransmission is triggered. S3: Generates anomaly warning information based on inspection data and fused mapping information; S4: Convert the generated abnormal warning information into alarm information and push it to the operation and maintenance terminal; S5: Manage inspection tasks and classify and issue abnormal early warning information; S6: Normal cycle time ensures dual-mode cycle time switching between regular inspection cycle time mode and alarm priority cycle time mode; the special situation ensures weighted fusion of multiple abnormal indicators to obtain a link abnormality score and outputs the link abnormality level based on the link abnormality score, generates a link abnormality event and triggers the self-rescue handling process, determines the target handling action and issues control commands.

[0084] As an embodiment of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, it employs the specific implementation method described above for the real-time early warning method for substation anomalies.

[0085] As an embodiment of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, adopts the specific implementation method described above for the real-time early warning method for substation anomalies.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A real-time early warning system for substation anomalies based on a humanoid inspection robot, characterized in that, The system includes: a humanoid inspection robot, a communication module, an early warning processing module, an alarm output module, a back-end early warning platform, and a collaborative support module; The humanoid inspection robot collects inspection data to construct real-scene frames, and performs feature matching with a preset 3D model to obtain fused mapping information. The communication module uses an incremental transmission mechanism on the uplink side based on inspection data and fusion mapping information. When the communication quality deteriorates, the minimum control plane priority strategy is triggered. When the communication quality recovers, the data plane transmission is restored and the buffer retransmission is triggered. The early warning processing module generates abnormal early warning information based on inspection data and fused mapping information; The alarm output module converts abnormal warning information into alarm information and pushes it to the operation and maintenance terminal; The aforementioned back-end early warning platform manages inspection tasks and classifies and releases abnormal early warning information. The collaborative support module includes two types of control logic: normal cycle support and special situation support. The normal cycle support switches between two modes: regular inspection cycle mode and alarm priority cycle mode. The special situation support performs weighted fusion of multiple abnormal indicators to obtain a link abnormality score and outputs the link abnormality level based on the link abnormality score. It generates a link abnormality event and triggers a self-rescue process, determines the target handling action, and issues control commands.

2. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 1, characterized in that, Obtaining the fusion mapping information includes: By acquiring multiple frames of images, after distortion removal, time synchronization and keyframe filtering, a three-dimensional Gaussian splash representation is constructed based on the keyframe depth information or point cloud information. The rendering perspective is determined according to the pose of the humanoid inspection robot and synthesized into a real scene frame. From the real scene frame, feature matching is performed on the equipment outline corner points, wiring end points and the 1:1 3D model in the station to obtain the translational deviation and rotational deviation between the real scene frame and the model, and the alignment error is calculated. The 3D Gaussian splash representation is transformed to the coordinate system of the 1:1 3D model within the station, and the fusion mapping information is output.

3. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 2, characterized in that, The fusion mapping information includes device identifier, point identifier, real-scene frame index, robot pose index, the position of the abnormal area in the real-scene frame, the corresponding position of the abnormal area in the 1:1 3D model in the station, and alignment error identifier. When the alignment error is less than or equal to the preset error threshold, output the fusion mapping information; When the alignment error exceeds a preset error threshold, an alignment correction process is triggered, which includes repositioning, reselecting keyframes, or switching adjacent point references.

4. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 1, characterized in that, The incremental transmission mechanism adopted on the uplink side is a humanoid inspection robot that uses the most recently successfully uploaded device reference packet as the baseline on the edge side. After performing quality filtering and anomaly screening on the currently collected data, it only sends the data that is judged to be abnormal and its associated real-scene frame index to the uplink, as well as the key normal data summary extracted according to the device importance weight, point coverage interval and energy budget, and the unconfirmed incremental packets cached during the link interruption. Normal non-critical frames do not enter the uplink data plane.

5. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 4, characterized in that, The aforementioned initial anomaly screening is performed as follows: First, determine whether the image or real-scene frame meets the requirements for sharpness, occlusion ratio, jitter, and lighting based on the quality label field. If the conditions are not met, the data is marked as low quality. If the conditions are met, the equipment status parameters are compared with the historical reference range of the corresponding equipment, and the presence of anomalies is determined by combining the results of appearance defect identification, temperature change, gas concentration and acoustic anomaly identification. When the equipment status parameters exceed the historical benchmark range, the appearance defect identification result reaches the appearance abnormality condition, the temperature change result reaches the preset temperature rise condition, the gas concentration result reaches the preset concentration condition, or the acoustic abnormality result reaches the acoustic abnormality condition, it is judged that there is an abnormality, and the corresponding data is marked as abnormal data. Otherwise, the humanoid inspection robot will select data from the normal state of the equipment according to the importance of the equipment, the time interval for data collection, and the coverage requirements of the data points.

6. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 1, characterized in that, The communication module is configured as follows: When heartbeats are continuously lost, end-to-end transmission delay continuously exceeds the first preset delay threshold, packet loss rate continuously exceeds the first preset packet loss threshold, uplink bandwidth is lower than data plane transmission requirements, or buffered data fails to be retransmitted continuously, communication quality is determined to be degraded. When the communication quality deterioration continues for a first preset number of times or a first preset duration, and there are currently incomplete inspection tasks, incomplete retest tasks, link abnormal events, or pending control commands, the minimum control plane priority strategy is triggered. Once communication quality is restored, perform data plane transmission and trigger buffer retransmission.

7. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 1, characterized in that, The dual-mode beat switching includes: When an alarm priority cycle is triggered, the abnormal point is inserted at the beginning of the point residence acquisition sequence, and the residence time of low-risk points is shortened or the channel coverage frequency is reduced, provided that the time window constraint is met; when the arrival time window constraint is not met, the correction scheduling is triggered. The correction scheduling includes adjusting the coverage order, changing the travel path, or delaying the start time of the next point's dwell time. In the dual-mode cycle switching, the conditions for triggering the alarm priority cycle include: the warning level reaches the threshold, the time limit for the retest of the abnormal point is less than the preset threshold, or the link abnormality level reaches the medium level but the retest can still be performed; the conditions for restoring the normal cycle include: the retest of the abnormal point is completed and the alarm status is updated to confirmed or deactivated, and the link abnormality level drops back to the low level and continues for the preset duration.

8. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 1, characterized in that, The link anomaly level is obtained, including: The link anomaly score is obtained by weighting the communication anomaly index, location anomaly index, energy anomaly index, task anomaly index, fault anomaly index, and channel anomaly index. The link anomaly score is compared with a preset set of grading thresholds to obtain the link anomaly level and generate a corresponding link anomaly event. The communication anomaly indicators are determined based on the number of heartbeat loss, the number of times the end-to-end latency exceeds the threshold, and the number of times the packet loss rate continuously exceeds the threshold. The location anomaly index is determined based on the number of times the location reliability is below the threshold and the number of times the location drift continuously exceeds the threshold. The energy anomaly index is determined based on the remaining energy being below a threshold and the expected insufficient range. The task anomaly indicators are determined based on task timeouts and execution phase stalls. The fault abnormality indicators are determined based on the results of drive, braking, joint, sensor, or self-test. The channel anomaly indicators are determined based on channel blockage, insufficient passage width, and the number of failed escape attempts.

9. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 8, characterized in that, The generation of the link anomaly event includes: The preset set of hierarchical thresholds includes a first threshold, a second threshold, and a third threshold, wherein the first threshold < the second threshold < the third threshold; If the link anomaly score is less than the first threshold, the link anomaly level is normal. If the link anomaly score is greater than or equal to the first threshold and less than the second threshold, the link anomaly level is mild anomaly level. If the link anomaly score is greater than or equal to the second threshold and less than the third threshold, then the link anomaly level is moderate. If the link anomaly score is greater than or equal to the third threshold, the link anomaly level is classified as severe anomaly. Based on the link anomaly level, combined with the anomaly type, occurrence time, humanoid inspection robot pose, current task stage and associated alarm identifier corresponding to the detected multiple anomaly indicators, the link anomaly event is generated and the self-rescue process is triggered; the link anomaly event is stored in association with the fault evidence package, retest record and alarm status after the link is restored. The self-rescue procedure includes getting out of trouble and backing up, safely stopping, retesting and retrying, and requesting manual intervention.

10. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 9, characterized in that, The generation of the fault evidence package is specifically as follows: When the link anomaly level reaches the preset level or a fault anomaly is detected, the humanoid inspection robot generates a fault evidence package. The fault evidence package includes event identifier, associated alarm identifier, device identifier, location identifier, timestamp, robot pose, anomaly type set, summary of the most recent key inspection data, most recent preset duration state sequence, current task stage, list of completed collection items and list of incomplete collection items; When the link is available, the communication module prioritizes reporting the fault evidence packet. When the link is unavailable, it writes the fault evidence packet and the corresponding real-scene frame into the local cache and retransmits them after the link is restored.

11. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 1, characterized in that, Determining the target action and issuing control commands includes: The comprehensive cost is calculated by weighting the time delay cost, risk cost, and energy cost required to execute the candidate disposal action; The candidate action corresponding to the minimum comprehensive cost is selected from the set of candidate actions and taken as the target action. The latency cost is calculated based on the ratio of the estimated total time required to execute the candidate action to the alarm recovery time. The estimated total time includes the movement time, the action execution time, and the control command transmission and status feedback time. The upper limit of the latency cost is 1. The risk cost is calculated by weighting the robot's own risk value, the device's neighborhood risk value, and the channel risk value. The energy cost is calculated based on the ratio of the estimated total energy consumption required to perform the candidate disposal action to the robot's current remaining energy. The estimated total energy consumption includes movement energy consumption, execution energy consumption, and communication and computing energy consumption, and the upper limit of the energy cost is 1. The control commands issued include stopping, reversing, and retesting.

12. The substation anomaly real-time early warning system based on a humanoid inspection robot according to claim 1, characterized in that, Anomaly warning information is generated based on the inspection data, including: The early warning processing module, based on inspection data, compares real-time scene frames with historical baseline images, and combines the coordinate alignment relationships corresponding to equipment identification, point identification, robot pose, and spatial point cloud to identify defects, difference areas, and confidence levels. It also combines multimodal sensor data, preset criteria, time-series algorithms, and historical degradation curves to determine anomalies and output anomaly determination results. The abnormal warning information is classified into four levels: normal, mild, moderate and severe.

13. A method for real-time early warning of substation anomalies based on a humanoid inspection robot, comprising executing the real-time early warning system for substation anomalies based on a humanoid inspection robot as described in any one of claims 1-12, characterized in that, include: S1: The humanoid robot collects inspection data to construct real-world scene frames, and performs feature matching with the preset 3D model to obtain fusion mapping information; S2: Based on inspection data and fusion mapping information, an incremental transmission mechanism is adopted for the uplink side. When the communication quality deteriorates, the minimum control plane priority strategy is triggered. When the communication quality recovers, the data plane transmission is restored and the buffer retransmission is triggered. S3: Generates anomaly warning information based on inspection data and fused mapping information; S4: Convert the generated abnormal warning information into alarm information and push it to the operation and maintenance terminal; S5: Manage inspection tasks and classify and issue abnormal early warning information; S6: Normal rhythm protection switches between two rhythm modes: regular inspection rhythm mode and alarm priority rhythm mode; Special situation protection performs weighted fusion of multiple abnormal indicators to obtain a link abnormality score and outputs the link abnormality level based on the link abnormality score, generates a link abnormality event and triggers the self-rescue handling process, determines the target handling action and issues control commands.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements the substation anomaly real-time early warning system according to any one of claims 1-12.

15. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the substation anomaly real-time early warning system according to any one of claims 1-12.