Power switching operation robot and mistake prevention control method thereof

Abstract

The application discloses a power switching operation robot and a mistaken operation prevention control method thereof. The robot comprises a robot body, a special operation arm, a special unlocking and operation tool, a mistaken operation prevention intelligent body and an intelligent mistaken operation prevention terminal. The mistaken operation prevention intelligent body is integrated in the robot body, cooperates with a station end mistaken operation prevention system, a two-ticket safety management and control system and a power monitoring system, and realizes operation ticket analysis, task scheduling, multi-source state checking, AI video analysis and embodied intelligent decision-making. The robot completes the whole process management and control of switching operation through interval double confirmation, mistaken operation prevention logic checking, voice singing ticket, accurate positioning and operation, operation video collection and other steps, in combination with a double mistaken operation prevention mechanism of system logic checking and actual field state, and effectively prevents electrical mistaken operation accidents, and improves the safety and intelligent level of switching operation.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance and anti-misoperation of power equipment, and particularly relates to a power switching operation robot and its anti-misoperation control method. Background Technology

[0002] In the process of building a new power system, smart grids and smart power plants, as the core carriers of power production, directly affect the safety and reliability of the entire power system due to their level of intelligent operation and maintenance. Switching operations are a key and core link in the operation and maintenance of power equipment in power plants and substations. Through this operation, high-voltage electrical equipment can be switched between operating, hot standby, cold standby, and maintenance states. The standardization and accuracy of this operation are important prerequisites for ensuring the safe operation of power equipment and preventing electrical misoperation accidents.

[0003] Currently, switching operations of high-voltage electrical equipment in the power industry are still primarily performed manually on-site. This model heavily relies on the experience and judgment of maintenance personnel, whose skill levels vary widely, leading to a persistent risk of human error. Statistics show that approximately 90% of electrical misoperation accidents in the past decade occurred during manual operations by maintenance personnel. Furthermore, during manual inspections and operations, maintenance personnel can only perceive equipment status through visual observation and simple tool measurements, making it difficult to achieve comprehensive, real-time monitoring of multi-dimensional parameters such as equipment temperature and partial discharge. This hinders effective prediction of long-term equipment operating trends and early warning of latent faults. In addition, large-scale switching operations are cumbersome and time-consuming. For operations such as grounding connections of ultra-high voltage and extra-high voltage equipment, a single person often cannot complete the task independently, requiring auxiliary tools or multiple personnel, resulting in high labor intensity and fatigue for workers, further increasing the probability of operational errors.

[0004] On the other hand, most existing intelligent error prevention technologies focus on system-level logic verification during the operation ticket issuance stage, such as checking the compliance of the operation ticket sequence through the "five-prevention" rules. However, these technologies lack deep integration with the on-site operation execution end, making it difficult to achieve closed-loop error prevention and control throughout the entire process from issuing operation instructions, on-site operation execution to operation status confirmation. In particular, during operation execution, real-time and accurate determination of equipment status still relies on manual on-site confirmation, failing to form an immediate linkage with the error prevention system. This creates obvious "breakpoints" in the entire switching operation process, not only reducing operational efficiency but also preventing the error prevention mechanism from being integrated throughout the entire operation, thus failing to meet the higher requirements of smart grids and smart power plants for operation and maintenance safety, intelligence, and efficiency. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a power switching operation robot and its error prevention control method. Specifically, the technical solution provided by this invention is as follows: A power switching operation robot, comprising: The humanoid robot body is equipped with a bipedal walking mechanism, a multi-level anti-collision system and a fault self-diagnosis function. It is used to autonomously navigate within the station, move to the target operating interval and perform switching operations. The dedicated operating arm is a multi-degree-of-freedom serial articulated robotic arm, mounted on the shoulder rotation platform of the robot body, and equipped with a tool quick-change interface at the end; Specialized unlocking and operating tools, including handcart operating tools, emergency tripping tools, positioning cameras, grounding switch operating devices, and partial discharge detection probes, can be automatically changed through the tool quick-change interface; The anti-misoperation intelligent agent is integrated into the robot's main unit and communicates with the station-end anti-misoperation system, the two-ticket safety management system, and the power monitoring system. It is used for operation ticket parsing, task scheduling, multi-source status verification, video analysis, knowledge storage, and embodied intelligent decision-making. The anti-misoperation intelligent agent has a built-in power industry-specific anti-misoperation model. The intelligent anti-misoperation terminal is deployed on the side of high-voltage electrical equipment and has functions such as equipment identity verification, status acquisition, remote interlocking control, and status indication.

[0006] Furthermore, the humanoid robot body adopts a free movement mode, and achieves in-situ rotation around its own vertical axis through differential movement of the two legs in conjunction with the coordinated movement of the hip and ankle joints; the navigation and positioning integrates visual sensors, radar systems, inertial measurement units and tactile sensors, and is optionally equipped with an outdoor navigation module; the multi-level anti-collision system includes non-contact detection and protection composed of lidar and ultrasonic sensors, as well as a contact-type buffer strip system.

[0007] Furthermore, the fault self-check covers three stages: before operation, during operation, and after operation, including power status self-check, communication status self-check, motion mechanism self-check, operating arm and end tool self-check, and sensor self-check. During the pre-operation self-check, if a logical conflict is detected between the current state of the equipment and the operation command to be executed, execution is refused and an instruction error alert is issued. During the post-operation self-check, if the actual state of the equipment is not consistent with the operation expectation, a status error alert is issued and manual intervention is required.

[0008] Furthermore, the anti-error intelligent agent includes an operation ticket parsing and task scheduling module, a multi-source status verification and video analysis module, an external system interaction interface module, a knowledge storage and log management module, and an embodied intelligence and continuous learning module, wherein: The operation ticket parsing and task scheduling module is used to receive and parse operation tickets, generate a sequence of instructions that can be executed by the robot, and coordinate multiple robots and multiple tasks; the operation ticket parsing uses a large power industry model for semantic understanding and structured information extraction. The multi-source status verification and video analysis module is used to collect robot vision information, intelligent anti-misoperation terminal status and electrical quantity data of power monitoring system, and to identify equipment number, indicator light status, mechanical position and operating tools through deep learning model; The external system interaction interface module is used for standardized data interaction with the power professional big data model, the two-ticket safety management system, the station-end error prevention system, the PMS / ERP system, and the power monitoring system. The knowledge storage and log management module is used to store robot operation procedures, error prevention logic rule base, equipment status judgment model and task execution logs; The embodied intelligence and continuous learning module is used to achieve multimodal perception fusion, autonomous planning and real-time adjustment, natural language interaction, remote expert guidance, and online model reinforcement and retraining.

[0009] Furthermore, the intelligent anti-misoperation terminal has a unique ID, and its surface is equipped with a QR code or barcode for device identity verification. It integrates an RFID tag internally for redundancy backup. The terminal has built-in miniature limit switches, reed switches, or mechanical contacts to collect the unlocking / locking status, door position, and temporary grounding wire connection status. The terminal has an internal locking mechanism driven by a miniature electromagnet, which performs remote unlocking after receiving the unlocking command from the anti-misoperation intelligent agent, and automatically locks or receives a locking command after the operation is completed. The terminal panel has LED indicator lights for on-site status indication.

[0010] A method for preventing errors in the control of the above-mentioned power switching operation robot includes the following steps: Operation ticket generation and preprocessing: Operation tickets are generated on the station-side error prevention system or imported from the production management system. The error prevention system performs structured parsing and item-by-item error prevention logic verification. After verification and approval, the operation ticket is transmitted to the error prevention intelligent agent. Operation ticket parsing and task issuance: The anti-misoperation intelligent agent parses the operation ticket to obtain the target interval position, device list and operation task sequence, and generates a navigation path; Double confirmation of operation interval: The robot moves to the front of the target interval and visually identifies the target to complete the double confirmation; Error prevention logic verification: For the current operation step, the error prevention agent sends an error prevention logic verification request to the station-side error prevention system. It can only continue after receiving the permission instruction, thus forming the first layer of error prevention verification. Robot on-site operation and self-check: The robot sequentially performs voice voting, locates the target, confirms the target, executes the operation, and collects video after the operation, and performs a status self-check before and after the operation. Video analysis and status feedback: The anti-misoperation intelligent agent performs AI video analysis on the images after the operation, and feeds the analysis results back to the two-ticket safety control system. It is then fused and verified with the electrical quantity status of the power monitoring system to form a second layer of anti-misoperation verification. Task Loop and End: After receiving the confirmation signal from the previous step, the anti-misoperation agent determines whether it is the last step. If so, the task ends; otherwise, it repeats the steps of anti-misoperation logic verification up to video analysis and status feedback.

[0011] Furthermore, in the dual confirmation of the operation interval, visual number recognition uses an OCR model to identify the numerical and textual numbers on the interval number plate and compares them with the operation ticket. Map number association matches the robot's current positioning coordinates with the interval coordinate range pre-stored in the map database. After both confirmations are passed, the robot sends a signal to the anti-misoperation intelligent agent indicating that the interval confirmation is complete.

[0012] Furthermore, in the error prevention logic verification, the error prevention system verifies the equipment status based on the error prevention logic rule base and the real-time data collected by the power monitoring system. If the verification is successful, an operation permission instruction containing a timestamp and digital signature is issued. If the verification fails, the error prevention agent terminates the task and prompts that the operation conditions are not met.

[0013] Furthermore, during the robot's on-site operation and self-check: the voice announcement announces the current operation step via the robot's speaker and supports two-way voice communication in the background; the positioning of the operation object uses a multi-servo algorithm combining vision and touch to bring the tool's center point close to the operation point until the deviation is less than a set threshold; when confirming the operation object, the device identifier next to the operation point is read again and compared with the operation ticket for a second time; when performing the operation, the end tool is automatically changed according to the operation type, and torque monitoring is used to determine whether mechanical jamming has occurred; after the operation, video capture includes shooting the indicator light area, mechanical indicator, or position indication in the observation window.

[0014] Furthermore, it also includes an anomaly handling mechanism: when the torque abnormally increases beyond the threshold during operation and the position feedback does not change, it is determined to be a mechanical jamming anomaly, and the robot immediately stops operation and reports it; when the communication of the robot's anti-misoperation intelligent agent is interrupted for more than a set time, the robot automatically stops all movement and enters a safe standby state, and continues or re-verifies after the communication is restored; when a state conflict is found during the pre-operation self-check or post-operation self-check, the robot stops subsequent operations and reports a state verification anomaly, waiting for manual confirmation.

[0015] Compared with the prior art, the present invention has at least the following beneficial effects: First, this invention deeply integrates the humanoid robot body with the anti-misoperation intelligent agent, constructing a closed-loop anti-misoperation control system covering the entire process from operation ticket generation, instruction issuance, on-site execution to status confirmation. Unlike existing technologies that limit anti-misoperation verification to the operation ticket system level, this invention, through the collaborative work of the anti-misoperation intelligent agent with the two-ticket safety control system, the station-end anti-misoperation system, and the intelligent anti-misoperation terminal, combines system logic verification with robot on-site video analysis, status self-check, and equipment-side lock control feedback, forming a dual-verification anti-misoperation mechanism. This eliminates the breakpoint between the anti-misoperation system and the execution end in traditional operations, effectively preventing electrical misoperation accidents from a technical perspective.

[0016] Secondly, this invention introduces embodied intelligence, endowing the robot with multimodal perception fusion, autonomous planning and real-time adjustment, natural language interaction, and continuous learning capabilities. This transforms the robot from a mere execution of preset programs into an intelligent agent with environmental understanding and autonomous decision-making abilities. It can adjust its operating strategies in real time based on dynamic environmental factors such as changes in ambient light, equipment surface conditions, and temporary obstacles. Furthermore, it can receive timely assistance from remote experts when encountering anomalies, significantly improving its operational adaptability and success rate under complex working conditions.

[0017] Furthermore, this invention integrates robot visual recognition information, status reported by intelligent anti-misoperation terminals, and electrical quantity data from the power monitoring system through a multi-source status verification module and an AI video analysis module, achieving multi-dimensional and highly redundant confirmation of the status of high-voltage electrical equipment. For different equipment types such as circuit breakers, disconnect switches, grounding switches, handcarts, and storage doors, various dual-confirmation criteria are employed, including indicator light recognition, mechanical position recognition, and electrical contact verification, greatly improving the accuracy and reliability of equipment status determination and eliminating the subjectivity and uncertainty of manual judgment.

[0018] Furthermore, this invention integrates a fault self-checking function covering the pre-operation, during-operation, and post-operation stages, as well as a multi-level anti-collision safety protection system into the robot body. Through pre-operation and post-operation self-checking mechanisms, it promptly detects command conflicts or abnormal states, effectively reducing the risks to equipment and personnel safety posed by robot malfunctions or operational errors, and ensuring the reliability and safety of the robot's long-term autonomous operation. Simultaneously, the knowledge storage and log management module in the anti-error intelligent agent fully records the entire process data of each operation. Combined with a continuous learning mechanism, it continuously optimizes the deep learning model, enabling the robot to self-evolve from historical operations and abnormal cases. By sharing anonymized learning data across sites, it continuously improves its generalization ability and operational accuracy for different equipment models and operating environments.

[0019] In summary, this invention significantly outperforms existing technologies in terms of safety, intelligence, environmental adaptability, reliability, and maintainability during switching operations, providing practical and feasible technical support for the efficient and safe operation and maintenance of smart grids and smart power plants. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0021] Figure 1 This is a schematic diagram of the anti-misoperation switching robot provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the anti-misoperation control method provided in an embodiment of the present invention.

[0022] The reference numerals in the attached figures represent: 1-Robot body, 2-Dedicated operating arm, 3-End tool, 4-Anti-misoperation intelligent agent, 5-Intelligent anti-misoperation terminal. Detailed Implementation

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

[0024] Example 1 This embodiment provides a power switching operation robot, which is a highly integrated intelligent terminal used to replace or assist manual labor in completing switching operation tasks of high-voltage electrical equipment.

[0025] like Figure 1 As shown, the power switching operation robot described in this embodiment mainly consists of a humanoid robot body 1, a dedicated operating arm 2, an end effector 3 (a dedicated unlocking and operating tool), an anti-misoperation intelligent agent 4, and an intelligent anti-misoperation terminal 5. The humanoid robot body 1 serves as the physical motion platform, the dedicated operating arm 2 and the end effector 3 serve as the operating execution mechanisms, the anti-misoperation intelligent agent 4 serves as the core decision-making and control center and provides the robot with advanced cognitive and autonomous learning capabilities, and the intelligent anti-misoperation terminal 5 serves as the identity and status perception node for the field equipment.

[0026] 1. Humanoid robot body The robot features a humanoid design, combining biomimetic movement capabilities with a compact form to adapt to the operational needs of narrow passages (minimum passage width approximately 700mm) and limited operating space in front of switchgear in high-voltage power distribution rooms. The robot is approximately 1.6m tall, with its torso and limbs constructed from a high-strength aluminum alloy and engineering plastic composite structure. This ensures structural strength while keeping the overall weight under 80kg, facilitating easy transfer between floors via elevators or ramps.

[0027] In terms of movement and navigation capabilities, the humanoid robot's lower body is equipped with a bipedal walking mechanism, enabling basic movement functions such as forward and backward straight movement, turning, and braking. To adapt to flexible maneuvering in the narrow passages of a power distribution room, the robot adopts a free-movement mode: through differential movement of its two legs combined with coordinated hip and ankle joint movements, the robot can rotate in place around its vertical axis to a target angle without stopping or backing up for adjustments. It can smoothly pass through narrow spaces, and the angular velocity of its in-place turning is adjustable, with a maximum angular velocity of up to 90 degrees. ° / s. The navigation and positioning system integrates multiple sensors, with a lidar mounted on the top of the torso (scanning radius 30m, angular resolution 0.1). ° The robot is used to construct a two-dimensional grid map of the environment and perform real-time Monte Carlo positioning. Simultaneously, an inertial measurement unit (IMU) and foot tactile sensors are installed inside the torso to provide auxiliary positioning information in areas where LiDAR features are not obvious (such as long corridors). For outdoor substation scenarios, an RTK-GPS module can be optionally added to achieve centimeter-level absolute positioning. The robot can receive control commands from the local monitoring system and flexibly switch between two working modes according to task settings and control requirements: In remote control mode, maintenance personnel can remotely send motion commands (forward, backward, turn, stop) and operation commands through the control handle or graphical interface of the backend system. The robot responds in real time and transmits video images, suitable for manual fine-tuning or system debugging in complex environments; In automatic mode, the robot autonomously completes global path planning (using the A* algorithm), local path dynamic planning (using the dynamic window method), navigation and positioning, and operation execution from the starting point to the target interval based on the preset task and navigation map, and periodically reports information such as position, speed, and task progress to the backend.

[0028] The robot is equipped with a multi-level collision avoidance and safety protection system. The first level of protection uses non-contact detection: a 16-line LiDAR (near-field version, detection range 0.1~5m) is installed on the robot's chest, back, and both sides of its waist to scan the surrounding 360° area in real time. °The robot is equipped with 12 ultrasonic sensors (detection range 0.02~4m) deployed on its limbs and lower torso to detect obstacles that are difficult for lidar to detect, such as transparent glass and black light-absorbing objects. When an obstacle is detected in any direction at a distance less than the safety threshold (0.5m in the forward direction and 0.3m in other directions), the robot immediately decelerates to 0.2m / s; if the distance further decreases to 0.2m, it brakes to a stop. Secondary protection uses contact-type buffers: mechanical collision buffer strips are installed at the robot's forearms, lower back, and knee joints, each containing a microswitch. When the robot accidentally collides with an obstacle, the buffer strip is pressed, triggering the microswitch and immediately cutting off the joint motor drive power, achieving a hardware-level emergency stop. In fully autonomous operation mode, after stopping due to an obstacle, the robot re-detects the obstacle every 2 seconds; if the obstacle is removed and three consecutive detections show no obstacle, the robot automatically resumes walking and continues executing unfinished tasks; if the obstacle is not removed after 30 seconds, the robot sends a path obstruction alarm to the backend and awaits manual intervention. In addition, the back-end system or handheld remote control is equipped with an independent emergency stop button. When pressed, the robot immediately cuts off all power output and remains in a braking state until manually reset.

[0029] The robot possesses comprehensive fault self-diagnosis and status reporting functions, covering three stages: pre-operation, during operation, and post-operation. Before each task start, the robot first performs a comprehensive self-check: Regarding power status, it reads the total voltage, individual cell voltage, current, remaining capacity (SOC), temperature, and charge / discharge cycles reported by the battery management system. The SOC must be no less than 30%; otherwise, it automatically navigates to a charging station. Individual cell voltage deviation must not exceed 50mV, and battery temperature must be within the range of -10℃ to 50℃. Regarding communication status, it sends heartbeat packets to the anti-misoperation terminal and waits for a response, measuring round-trip time (RTT) and packet loss rate. The RTT must not exceed 200ms, and the packet loss rate must not exceed 1%; otherwise, it attempts to switch communication links (e.g., from 5G to wired Ethernet or backup Wi-Fi). (Fi); Regarding the motion mechanism, the robot sequentially performs forward 0.5m, backward 0.5m, left turn 90°, right turn 90°, and braking actions. The actual displacement or angle is compared with the command value through encoder feedback, and the error must not exceed 5%. Regarding the manipulator and end effector, each joint of the robotic arm is moved sequentially to its limit position. The encoder and torque sensor feedback are checked for normal operation, and a tool quick change operation is performed to check the on / off signal of the tool in-situ sensor. Regarding sensors, the camera's autofocus and auto white balance functions are checked, the LiDAR point cloud data is checked for any abnormal gaps, and the ultrasonic sensor is checked for normal detection of standard obstacles. If any abnormality is found during the self-test, the robot immediately issues an audible and visual alarm locally via a buzzer and a red indicator light. At the same time, the abnormality information (including the fault module code, occurrence time, and detailed fault description) is uploaded to the anti-misoperation terminal and power monitoring system through a security protocol to prompt maintenance personnel to intervene in a timely manner. For non-fatal faults (such as the failure of an ultrasonic sensor), the robot can still continue to perform the task, but a warning must be marked in the task report. For fatal faults (such as drive motor failure or communication interruption), the robot is prohibited from performing the task and its movement is locked.

[0030] During the pre-operation self-check phase, after receiving an operation command for a specific device (e.g., circuit breaker 103), the robot uses a high-definition camera mounted on its head to capture images of the indicator lights on the target device before executing the operation. It then uses a built-in lightweight image recognition model (e.g., MobileNetV3) to identify the device's current status in real time. If a logical conflict is detected between the current device status and the upcoming operation command (e.g., the command is to open the circuit breaker but the indicator light is currently red and closed, or the command is to close the circuit breaker but the indicator light is currently green and open), the robot immediately stops the operation, sends an error message to the anti-misoperation terminal and the power monitoring system, including the device identifier, the current identification status, and the required operation, and refuses to execute any subsequent actions. During the post-operation self-check phase, after completing a switching operation (e.g., opening the circuit breaker), the robot again captures images of the indicator lights on the target device and compares them with the expected operation status. If the robot detects that the actual status of the equipment does not match the expected operation (for example, the instruction is to open the circuit breaker but the indicator light is still red and in the closed state, or the mechanical indicator still shows ON), the robot immediately sends a status error alert to the anti-misoperation terminal and the power monitoring system, triggering an audible and visual alarm and recording a detailed log. At this time, the robot will not automatically repeat the operation, but will wait for manual confirmation or remote instructions.

[0031] 2. Specialized operating arm and end effector The dedicated manipulator arm in this embodiment is a six-degree-of-freedom tandem articulated robotic arm, mounted on the shoulder rotation platform of the robot body to simulate the motion capabilities of a human arm. Each joint of this robotic arm is equipped with a high-precision absolute encoder (resolution 0.001). ° The robotic arm is equipped with a torque sensor (range 0~100 N·m, accuracy ±0.5%FS), a reach of 600 mm, an end-effector load capacity of 3 kg, and a repeatability of ±0.05 mm. The control of the robotic arm employs a feedforward plus PD feedback control algorithm based on a dynamic model, enabling smooth force / position hybrid control. During operation, the arm can adjust the applied force in real time based on feedback from the end-effector torque sensor, preventing damage to the operating mechanisms of high-voltage electrical equipment due to excessive operating force. A rotary joint is also provided between the robotic arm and the shoulder platform of the robot body, allowing the robotic arm to rotate ±180° in the horizontal plane, thereby expanding the operating coverage area. Simultaneously, the humanoid robot can coordinate with its gait to enable the robotic arm to reach operating points at any height on the switchgear, from a maximum of 1.8 m to a minimum of 0.3 m.

[0032] The robot's end effector is equipped with a quick-change tool interface that conforms to ISO 9409-1 standards and integrates a pneumatic locking mechanism, 24V DC power supply contacts, and RS485 communication contacts. The robot can automatically change different end effector tools according to task instructions. The quick-change device features a Hall effect-based tool presence sensor; when the tool is correctly installed and locked, the sensor outputs a high-level signal; if the tool is not properly installed, the robot refuses to perform subsequent operations. After each tool change, the robot performs a tool self-check: for rotary tools, it performs one full rotation without load to check if the encoder and torque signals are normal; for pressing tools, it performs one no-load pressing stroke to check if the pressure sensor returns to zero. Operation can only begin after the self-check passes.

[0033] This embodiment equips the robot with a variety of dedicated unlocking and operation tools, such as: (1) Handcart operating tool: The head of this tool is a hexagonal or square sleeve, the size of which matches the operating hole of the mainstream central cabinet handcart (e.g., 10mm hexagon). An angle encoder and torque sensor are integrated inside the tool. When it is necessary to switch the handcart from the working position to the test position (or vice versa), the robot inserts the handcart operating tool into the operating hole and rotates it in the specified direction (usually counterclockwise is pulling out, and clockwise is pushing in) and number of rotations (e.g., 35 rotations). During the rotation, the robot monitors the torque value in real time. The normal torque range is 5~20 N·m. If the torque suddenly increases to more than 30 N·m and lasts for 0.5 seconds, it is judged as mechanical jamming, and the rotation is stopped immediately and the abnormality is reported.

[0034] (2) Emergency tripping tool: The tool has an insulated push rod at the head, with a diameter of 6mm and a length of 100mm. It is used to press the emergency tripping button on the switchgear remotely in an emergency. The tool is equipped with a pressure sensor, which can control the pressing force between 10 and 15N to avoid damaging the button.

[0035] (3) Positioning Camera: This tool is actually a miniature vision module, containing an 8-megapixel global shutter camera and two high-brightness LED fill lights. The positioning camera is fixed at the end of the robotic arm near the tool, with its optical axis parallel to the tool's axis. When the robot needs to precisely position an operating hole or button, it uses the positioning camera to capture an image of the target area, calculates the deviation between the tool's current pose and the target pose using a visual servo algorithm, and adjusts the position of the robotic arm's end in real time until the deviation is less than 0.5mm. The positioning camera can also be used to confirm the identification tags (such as equipment numbers) of the operated object, providing a second layer of identity verification.

[0036] (4) Grounding switch operating device: This device consists of a replaceable socket wrench head and an electric rotary drive module, adaptable to different models of grounding switch operating shafts (commonly hexagonal head, square head, or internal spline). The maximum output torque is 80 N·m, and the rotation speed is 30 rpm. During operation, after the device engages the operating shaft, it is rotated clockwise to the closed position (usually the rotation angle is 90° or 180°), or counterclockwise to the open position. The robot uses an angle encoder to determine whether the rotation is in place and a torque sensor to determine whether there is jamming.

[0037] (5) Partial Discharge Detection Probe: This tool is a capacitively coupled ultrasonic sensor with a frequency range of 20~200kHz. Before or after operation, the robot can attach the probe to the designated detection point of the switchgear (such as the surface of the cable compartment or circuit breaker compartment) to collect the ultrasonic signal generated by partial discharge. After amplification and filtering, the signal is uploaded to the background for analysis to assist in judging the insulation status of the equipment. The probe is kept in contact with the cabinet by magnetic attraction or vacuum suction cup.

[0038] 3. Anti-misoperation intelligent agent The anti-error intelligent agent integrates a large-scale power industry model and serves as the core control system of the robot. Deployed within an embedded industrial control computer inside the robot, it collaborates with the station-side backend management system and the central control room cloud platform. This large-scale model is pre-trained and fine-tuned based on massive amounts of text, image, and operational rule data from the power industry, possessing capabilities such as semantic parsing of operation tickets, logical reasoning of equipment status, human-machine natural language interaction, and intelligent diagnosis of abnormal situations. It achieves a deep integration of AI intelligent analysis algorithms and operational control logic, acting as the robot's "AI brain." The anti-error intelligent agent mainly consists of modules such as operation ticket parsing and task scheduling, multi-source status verification and video analysis, external system interaction, knowledge storage and log management, embodied intelligence, and continuous learning.

[0039] (1) Operation ticket parsing and task scheduling The anti-misoperation intelligent agent first receives operation ticket data from the station-side anti-misoperation system or the two-ticket safety management system through the operation ticket parsing module, supporting XML, JSON, or the power industry standard CIM / E format. This module performs syntax and semantic validation on the operation ticket to ensure it conforms to a predefined pattern. Then, using a named entity recognition engine based on a large power industry model, it extracts the target interval (region, interval number, dual-number text), the equipment list (all high-voltage equipment and their unique identifiers), and the sequentially arranged operation sequence from the free text description (each operation step includes the operation object ID, operation type, expected state before operation, and expected state after operation). After parsing, a standardized internal task object is generated and passed to the task scheduling module.

[0040] The task scheduling module decomposes the operation sequence into a sequence of atomic instructions that the robot can execute. For each operation step, it sequentially generates navigation instructions (target interval coordinates and attitude angles), visual positioning instructions (capture parameters), error prevention verification requests (sending a logic verification request to the error prevention system and waiting for a response), operation instructions (calling the end-effector type, action type, and motion parameters), and confirmation instructions (acquiring images and calling video analysis after the operation). This module is also responsible for coordinating multiple robots and multiple tasks, preventing simultaneous entry into the same interval through a locking mechanism and planning staggered motion paths, maintaining the task queue, and supporting priority adjustment, pause, resumption, and cancellation.

[0041] (2) Multi-source state verification and video analysis The status verification module gathers equipment status information from multiple sources: visual recognition information transmitted back by the robot itself (equipment indicator lights, mechanical positions, and number plates); status reported by the intelligent anti-misoperation terminal (lock / unlock status, door opening / closing, and temporary grounding wire connection status); and electrical quantity status collected by the Supervisory Control and Data Acquisition (SCADA) system through hardwiring (circuit breaker auxiliary contacts, disconnector auxiliary contacts, grounding switch auxiliary contacts, and passive contacts of the live indicator, etc.). This module timestamps and fuses the multi-source information (using DS evidence theory), comparing it item by item with the expected status in the operation ticket. If all match, it outputs "Status verification passed"; otherwise, it outputs the inconsistent items.

[0042] The video analysis module is the core AI component of the anti-misoperation intelligent agent. After the real-time video stream (H.264 encoded, 1920×1080 resolution, 25fps) or keyframe images transmitted by the robot enter the buffer queue, the module calls a set of pre-trained deep learning models for inference: the device number OCR model is based on a deep learning text recognition algorithm (such as CRNN+CTC, SVTR or Transformer-based OCR model), supports the recognition of numbers, Chinese characters and special symbols, and has a character accuracy of ≥99.5%; the indicator light status classification model uses ResNet-50, outputting categories such as red / green solid, flashing, and off, with an accuracy of ≥98%; the mechanical position status recognition model is based on Mask R-CNN, which segments components such as isolating switch arms, grounding switch rods, and handcart position contacts and determines the opening and closing status based on geometric relationships, and is also used to identify the opening and closing of the compartment door; the operating tool detection model is based on YOLOv8, which confirms the end tool type after tool replacement. The video analysis module is GPU accelerated (the robot itself can integrate a lightweight GPU module, and the backend is also configured with NVIDIA Tesla T4 / L4 inference GPUs, etc.). The processing time per frame is ≤50ms. The analysis results are output in JSON format, including device ID, status type, recognition result, confidence level and key frame storage path.

[0043] (3) External system interaction interface The anti-misoperation intelligent agent interacts with external systems through standardized data exchange via interface modules: With the two-ticket safety management system, it uses WebSocket + JSON-RPC for bidirectional real-time communication of operation ticket issuance and operation step confirmation signals; with the substation / power plant anti-misoperation system (five-prevention system), it uses the IEC 61850 standard and the MMS or GOOSE protocol for exchanging anti-misoperation logic verification requests and results; with the PMS / ERP system, it supports operation ticket import and operation result write-back via RESTful API; with the power monitoring system (SCADA), it reads the electrical status of equipment in real time via Modbus TCP or IEC 104 protocol; and with other parts of the robot body (such as motion control and arm control), it communicates via an internal high-speed bus. The interface modules have built-in mechanisms for disconnection reconnection, message confirmation, and timeout retransmission to ensure communication stability in weak network environments. The wireless communication link between the robot body and external systems primarily uses a 5G private network or encrypted WiFi 6, while also retaining a wired Ethernet interface for debugging and backup. All transmitted data is encrypted with AES-256 and digitally signed to prevent tampering.

[0044] (4) Knowledge storage and log management The anti-misoperation intelligent agent integrates operation methods and storage media functions. The robot body has a built-in embedded industrial control computer (Intel Core i7-1185G7, 16GB DDR4, 512GB NVMe solid-state drive), while the back-end server is equipped with a large-capacity storage array. The storage media contains: Robot operation program: including navigation algorithm libraries such as Gmapping mapping, AMCL localization, A* and DWA path planning; ROS2-based motion control program (motor PID control, robotic arm inverse kinematics solving, trajectory planning); visual recognition inference engine based on OpenCV and PyTorch; state machine-driven task execution engine.

[0045] The error prevention logic rule base uses a decision table to store over 200 "five-prevention" rules, covering equipment such as circuit breakers, disconnect switches, grounding switches, handcarts, and temporary grounding wires. Each rule includes prerequisites, prohibited operations, and permitted operations (e.g., preventing disconnect switches from being opened or closed under load, and preventing grounding wires from being connected while energized). The rule base can be updated online.

[0046] Equipment status determination model: Stores deep learning weight files (.pth or .onnx format) of the aforementioned OCR model, indicator light classification model, mechanical position segmentation model, tool detection model, etc. The model version is associated with the software version and is compressed by INT8 quantization to reduce storage space and improve inference speed.

[0047] Task execution log: After each operation task is completed, a structured log (SQLite or Parquet) is automatically generated, containing task metadata, detailed records of each operation step (instructions, verification, execution status, keyframe path, video analysis results, double verification results), exception records (self-check failure, communication interruption, status conflict, mechanical jamming, etc.), and the robot's own status log (position, speed, battery level, joint angles, etc.). The log files are regularly backed up to external storage (such as NAS) and stored for no less than 3 years for post-event traceability, fault analysis, and auditing.

[0048] (5) Embodied intelligence and continuous learning Embodied intelligent agents operate as advanced plug-ins of error-preventing intelligent agents, endowing robots with environmental perception, autonomous decision-making, continuous learning, and human-machine collaboration capabilities, so that robots are no longer simply tools that execute preset programs.

[0049] Multimodal perception fusion: Embodied intelligent agents integrate data from multiple sensors, including vision (visible light and infrared thermal imaging), hearing (microphone array directional acquisition of equipment operating sounds), force (joint torque, end-effector contact force), and touch (tool and equipment contact status). After time and spatial alignment, this forms a comprehensive understanding of the operating environment and equipment status. For example, when performing a circuit breaker tripping operation, the agent not only visually confirms the indicator lights but also collects the impact sound of the mechanism through microphones and senses the smoothness of the operation through force to comprehensively determine whether the operation was successful.

[0050] Autonomous planning and real-time adjustment: Within the framework of the five-prevention logic and operation sequence specified in the operation ticket, the embodied intelligent agent can autonomously plan the optimal operation path and strategy. When reaching the target interval, it automatically selects the best robotic arm operation posture; if dust on the equipment surface makes QR code recognition difficult, it can autonomously control the end of the arm to approach and blow away the dust before re-recognizing; when changes in lighting cause a decrease in the confidence of video analysis, it automatically adjusts the camera exposure parameters or takes multiple images from different angles for comprehensive judgment.

[0051] Natural Language Interaction: The robot is equipped with a microphone array and speakers. Its embodied intelligent agent integrates Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) engines, leveraging a large-scale power industry model to achieve more accurate understanding of industry terminology and dialogue generation, supporting natural language dialogue with maintenance personnel. Maintenance personnel can verbally inquire about operation progress or issue commands such as "pause task, return to charging." Voice interaction supports mixed Chinese and English, and all commands undergo security verification via voiceprint recognition.

[0052] Remote expert guidance: When the robot encounters anomalies that it cannot handle autonomously (such as mechanical jamming or inconsistent status verification), the embodied intelligent agent establishes a real-time audio and video connection with a remote expert via a 5G network. The expert can view the robot's first-person perspective video through a VR / AR terminal and guide the robot to handle the anomaly through voice and virtual annotations (arrows, circles, etc. superimposed on the video screen). All interaction data is recorded for subsequent model training.

[0053] Embodied agents can integrate vision-language-action (VLA) models to achieve end-to-end mapping from visual observations and natural language commands to robotic arm movements, improving their generalization ability to unseen scenarios; and they can use imitation learning methods such as diffusion policies to learn complex operational skills from expert demonstrations.

[0054] Continuous Learning and Model Optimization: The embodied agent possesses online learning and incremental training capabilities. After each task is completed, the collected images, operation results (success / failure), and exception cases are automatically labeled (partially automatically, partially manually verified) and added to the training dataset. The deep learning model is fine-tuned using new datasets periodically (e.g., weekly). After the updated model is tested on a validation set (accuracy no lower than the old model), it is deployed to the robot and backend system via OTA (Over-The-Air), enabling the robot's self-evolution. Furthermore, it can learn from anonymized data shared by robots at other sites, improving its generalization ability to different equipment models and operating conditions.

[0055] Through the overall design of the above-mentioned anti-misoperation intelligent agent, the core control system of this robot realizes a full-chain function including operation ticket parsing, task scheduling, multi-source status verification, AI video analysis, external system collaboration, knowledge storage and continuous learning, and embodied intelligent decision-making, providing a safe and reliable intelligent "brain" for the power switching operation robot.

[0056] 4. Intelligent anti-misoperation terminal The intelligent anti-misoperation terminal is an intelligent locking and control device deployed on the side of high-voltage electrical equipment. Each terminal has a unique ID (32-bit UUID) and is physically bound to its associated equipment. For switchgear, the terminal is embedded next to the control panel of the cabinet door; for temporary grounding wires, the terminal is fixed to the grounding stake; for compartment doors, the terminal is installed on the door frame. This terminal integrates equipment identification verification, status acquisition, remote locking / unlocking control, and status indication, and is a key node for information interaction between the robot and the equipment side during on-site operation.

[0057] For device identification verification, the terminal surface is equipped with a QR code or barcode with a reflective coating, which encodes information such as device type, dual serial number, and terminal ID. After the robot reaches the target interval, it first scans the QR code with a high-definition camera, decodes it, and compares it with the device information in the operation ticket. As a redundancy backup, the terminal also integrates RFID tags conforming to the ISO 15693 standard (frequency 13.56MHz), and an RFID reader / writer is integrated at the end of the robot arm, which can read tag information at close range (≤5cm) to further ensure the correctness of the operated object.

[0058] The terminal incorporates multiple sensors for real-time data collection of the status of the bound devices. Locking / unlocking status is determined by detecting the bolt position using a miniature limit switch: an extended bolt indicates locking, and a retracted bolt indicates unlocking. For terminals mounted on the door, a reed switch detects door opening and closing; a magnet is fixed to the door frame, and the reed switch closes when the door is closed. For grounding terminal terminals, mechanical contacts detect whether the grounding clamp is correctly engaged with the grounding head, thus reporting whether it is connected or not. All status data can be uploaded via LoRa wireless communication (470MHz frequency, 1kbps rate) or via Bluetooth BLE 5.0 near-field communication when the robot approaches. The terminal's sampling period is 1 second, and it actively reports status changes to ensure the anti-misoperation agent can promptly obtain information from the device side.

[0059] The terminal has an internal locking mechanism driven by a miniature electromagnet, enabling remote locking and unlocking control. After the anti-maloperation agent completes logical verification, it sends an unlock command to the terminal (the command includes the device ID, timestamp, and verification code). Upon receiving the command, the terminal first verifies the Message Authentication Code (MAC) using a pre-shared key. If verification is successful, the electromagnet is energized, the latch retracts, and the device unlocks, allowing operation by a robot or human. After the operation is complete, the anti-maloperation agent can send a locking command, or the terminal can automatically lock after a set time (e.g., 30 seconds). The terminal's unlocking action is powered by a backup lithium battery, ensuring normal operation even when the main power supply is interrupted.

[0060] In addition, the terminal panel is equipped with three LED indicators for on-site status indication: a solid red LED indicates locking, a solid green LED indicates unlocking, and a flashing yellow LED indicates a communication failure. Maintenance personnel can intuitively understand the current locking status of the equipment without the need for other tools, facilitating on-site inspections and troubleshooting. Through this design, the intelligent anti-misoperation terminal, in collaboration with the robot and the anti-misoperation intelligent agent, forms a complete closed loop from identity verification and status collection to operation authorization, providing reliable equipment-side anti-misoperation protection for switching operations.

[0061] Example 2 This embodiment provides a method for preventing errors in the control of the aforementioned power switching operation robot. This method achieves closed-loop error prevention and control throughout the entire process, from operation ticket generation, verification, execution to status confirmation, through deep collaboration between the error prevention intelligent agent, the two-ticket safety management system, the station-end error prevention system (five-prevention system), and the intelligent error prevention terminal.

[0062] like Figure 2 As shown, the method specifically includes the following steps: 1. Operation ticket generation and preprocessing Maintenance personnel can directly issue operation tickets at the workstation of the anti-misoperation system in the central control room or at the station. The operation ticket should include at least the operation task name, operation start time, dual numbering of the target equipment interval (numerical number and text number), and the specific objects and operation types of each operation step arranged in sequence (e.g., opening 101 circuit breaker, closing 101-2 disconnecting switch, closing 101-7 grounding switch, etc.). In addition, users can also import standard operation tickets that already exist in power industry production management systems such as PMS and ERP into the anti-misoperation system in structured or semi-structured formats (such as XML, JSON, or CIM / E).

[0063] Upon receiving an operation ticket, the anti-misoperation system first performs structured parsing, converting the operation steps described in natural language into a unified internal sequence of operation instructions. Then, based on its built-in five-prevention logic rule library (including core rules for preventing accidental circuit breaker opening / closing, preventing disconnection / closing under load, preventing live grounding wire connection, preventing circuit breaker or disconnection / closing with grounding wire connected, and preventing accidental entry into energized areas) and real-time equipment status obtained from the power monitoring system (SCADA) (such as circuit breaker opening / closing position, trolley working / testing position, disconnection switch position, grounding switch position, and line energization status), the system performs item-by-item anti-misoperation logic verification on the parsed operation ticket. Verification includes: whether the operation sequence conforms to the five-prevention logic, whether the current equipment status meets the operational prerequisites, whether the operation object matches the actual equipment on site, and whether there are any duplicate or omitted operations.

[0064] If all anti-misoperation checks pass, the operation ticket automatically enters the approval process, where it is confirmed by the shift supervisor or operations specialist with the appropriate permissions in the electronic workflow. After approval, the anti-misoperation system transmits the structured operation ticket to the anti-misoperation intelligent agent integrated into the robot body via a secure communication protocol (e.g., a TLS encrypted channel based on national cryptographic algorithms). If the checks fail, the system returns a clear error message to the user (e.g., "Step 2: Closing the 101-2 isolating switch does not meet the conditions because the 101 circuit breaker is still in the closed state"), requiring the user to modify the operation ticket and resubmit it.

[0065] 2. Operation ticket parsing and task assignment After receiving the operation ticket, the anti-misoperation intelligent agent further parses the following key information: the location description of the target bay (e.g., "10kV distribution room, 3rd row, 3rd cabinet, bay number 103"), all relevant equipment in the bay and their dual numbers, the operation task sequence—this task sequence is an ordered list, each element containing the operation step number, the operation object ID (associated with the UUID of the intelligent anti-misoperation terminal), the operation type code (e.g., 0x01 represents opening, 0x02 represents closing, 0x03 represents the handcart being pulled out of the test position, 0x04 represents the handcart being pushed into the working position, etc.), and the expected state after the operation.

[0066] The anti-collision agent queries the robot's built-in semantic map and navigation network based on the target interval location, and uses the A* algorithm to generate the optimal collision-free path from the robot's current position (or a specified starting point) to the preset target interval station. The path consists of a series of navigation key points, each containing two-dimensional coordinates (x, y) and the robot's orientation angle θ. Subsequently, the anti-collision agent sends navigation commands to the robot's motion control module, carrying the path point sequence, the precise coordinates of the target station, and the robot's intended attitude angle. Simultaneously, the anti-collision agent initializes the task execution context and starts the state machine for the entire operation process.

[0067] 3. Double confirmation of operation interval Following navigation instructions, the robot autonomously moves to a pre-set operating position in front of the target interval (typically 0.6–0.8 meters from the cabinet door, directly facing the control panel). Upon arrival, the robot adjusts the pitch and yaw angles of its high-definition camera using its head-mounted gimbal to capture an overall image of the operating interval. This image is transmitted in real-time to the video analysis module of the anti-misoperation intelligent agent for processing, performing double confirmation of the interval.

[0068] The first level of verification involves visual identification of serial numbers: The video analysis module uses deep learning-based text recognition algorithms (such as CRNN+CTC, SVTR, or Transformer-based OCR models) to identify the numerical and textual serial numbers (such as "103" and "Feeder Line Three") on the numbered plaques in the image. The identification results are then compared with the dual serial numbers recorded in the operation ticket. Only when both are completely identical can the first level of verification pass.

[0069] The second confirmation involves map number association: the robot matches its current positioning coordinates (obtained by fusing LiDAR SLAM and wheel odometry) with pre-defined coordinate ranges in the map database. Each range is stored in the map database as a polygonal region and its associated number. The robot confirms that its coordinates fall within the polygonal region corresponding to the operation ticket's specified number, and that the orientation angle deviation does not exceed ±10°. ° .

[0070] After both confirmations are successful, the robot sends a confirmation signal to the anti-misoperation agent via the MQTT channel and awaits the next instruction. If either confirmation fails (e.g., the number recognized by OCR does not match the operation ticket, or the positioning coordinates fall within the adjacent interval range), the robot immediately stops, issues a voice alarm through the speaker, sends an interval matching error alarm to the backend, and simultaneously packages and uploads the currently captured image and positioning data for remote analysis and manual intervention by maintenance personnel.

[0071] 4. Error-proofing logic verification (first layer of verification) Before executing each operation step, the anti-misoperation agent first sends an anti-misoperation logic verification request to the station-side anti-misoperation system. Taking the first step, "opening circuit breaker 103," as an example, the request message includes the operation object ID (circuit breaker 103), the operation type (opening), and the current status of the equipment obtained by the anti-misoperation agent through real-time image recognition from the robot (e.g., the indicator light recognition model outputs "red closing light on," confirming that the circuit breaker is in the closed position). After receiving the request, the anti-misoperation system performs a comprehensive logic verification by combining the five-prevention logic rule base and the electrical quantity status collected in real-time by the SCADA system (e.g., the auxiliary contact status of circuit breaker 103 is "closed," the line current is greater than zero, etc.). The verification points include: confirming that the circuit breaker is currently allowed to open (no protection interlock, no anti-misoperation interlock), whether the operation sequence complies with regulations (in this example, the first step, which is allowed), and whether the operator's permissions are valid, etc.

[0072] If the verification passes, the anti-malfunction system issues a control command allowing operation to the anti-malfunction agent, which includes a timestamp and digital signature. If the verification fails (e.g., the circuit breaker is already in the open state, or load current is detected on the line), the anti-malfunction system returns a failure reason code and a readable description (e.g., "Circuit breaker 103 is currently in the open state; repeated opening operation violates the five-prevention rule"). Upon receiving the failure response, the anti-malfunction agent immediately terminates the current task, sends an "Operation conditions not met" alarm to the backend, displays the failure reason on the monitoring interface, and simultaneously issues a voice prompt, "Operation conditions not met, task terminated," awaiting manual intervention.

[0073] 5. Robot on-site operation and self-inspection Upon receiving permission to operate, the robot initiates the on-site operation execution subroutine.

[0074] First, there's the voice announcement. The robot uses its onboard speaker to clearly and clearly announce the current operation step, for example: "Now performing the first step: circuit breaker 103, please confirm." Simultaneously, back-end monitoring personnel can communicate with the robot in full-duplex via a two-way voice intercom system to verify the announced results or issue supplementary instructions. The volume of the voice announcement automatically adjusts according to ambient noise to ensure clear hearing for on-site personnel.

[0075] The second step is locating the target object. The robot uses a positioning camera (wide-angle, with auxiliary lighting) integrated into its end effector to search for the target operation point. For circuit breaker tripping, the operation point is the trip button; for trolley operation, the operation point is the trolley operating hole; for grounding switch operation, the operation point is the grounding switch operating axis. The positioning camera acquires images in real time and uses a visual servo algorithm to adjust the end effector position in real time based on pixel deviation. The pixel deviation is converted into actual spatial deviation using hand-eye calibration parameters. When the actual spatial deviation is less than 0.5mm, the positioning is considered complete.

[0076] Thirdly, the target device is confirmed. When the robotic arm's end effector approaches the operation point to within approximately 2 cm, the robot again uses a positioning camera or a miniature end effector to capture a magnified image of the area near the operation point. OCR or QR code recognition technology is then used to read the equipment identification markings affixed next to the operation point (such as the "103" sign next to the "open" button), and a secondary comparison is made with the equipment number in the operation ticket. This step effectively prevents accidental operation of adjacent equipment due to visual positioning errors.

[0077] Fourthly, the robot performs the operation. Depending on the operation type, the robot automatically selects and switches to the appropriate end-effector via a tool quick-change device. For button operations, the robot controls the end-effector to press the button vertically with a preset force (e.g., 10N) and stroke (6mm), holding it for 0.3 seconds before retracting. During this time, a force sensor monitors the contact force in real time. If the force value abnormally increases while the stroke does not reach the set value, it is determined that the button is stuck. For handcart in / out operations, the robot inserts the handcart operating tool into the operating hole and rotates it at a constant speed according to the specified rotation direction (pull out counterclockwise, push in clockwise) and number of rotations (e.g., 35 rotations), while monitoring the torque value. The normal torque range is 5~20 N·m. If the torque exceeds 30 N·m and lasts for 0.5 seconds, it is determined that there is mechanical jamming, and the rotation is immediately stopped. For grounding knife operations, the robot uses the grounding knife operating device to engage the operating shaft and rotates it to the limit position in the set direction (usually a rotation angle of 90 degrees). ° Or 180 ° The angle encoder determines that the object is in position.

[0078] Fifth is post-operation video capture. After the operation is completed, the robot retracts its robotic arm, adjusts the gimbal and camera, and captures the status indication images of the currently operated object. For circuit breakers, the area of ​​the open / close indicator light (expected to have the green open light on) and the mechanical indicator (expected to display "OFF") need to be captured; for handcarts, the handcart position indication inside the observation window needs to be captured (expected to display "Test Position"); for grounding switches, the grounding switch status indicator light and the mechanical position indicator need to be captured. The captured video stream or keyframe images are pushed to the anti-misoperation intelligent agent in real time.

[0079] 6. Video analysis and status feedback (secondary verification) Upon receiving the image after the operation, the anti-misoperation intelligent agent immediately activates the AI ​​video analysis module for processing. For different types of devices, the corresponding recognition model is invoked: for switch states, the indicator light classification model outputs red / green light status, and the mechanical position recognition model outputs "ON / OFF"; for isolating switches, both electrical indicator lights and mechanical linkage positions are analyzed simultaneously to form a "electrical + mechanical" dual confirmation criterion; for grounding switches, the indicator light color and mechanical indicator position are identified; for door openings, the opening / closing status is determined by detecting door gaps or latch positions using a segmentation model. The video analysis results are output in structured data format, including the device ID, the identified status, confidence level, and keyframe storage path.

[0080] Subsequently, the anti-malfunction intelligent agent feeds back the video analysis results (e.g., "Circuit breaker 103 successfully tripped, green indicator light is on, mechanical indicator shows OFF, confidence level 0.98") to the two-ticket safety control system. Upon receiving the video confirmation signal, the two-ticket safety control system automatically retrieves the electrical quantity status collected simultaneously by the power monitoring system (SCADA) (e.g., the auxiliary contact of circuit breaker 103 has changed to "open," and the line current is zero). The system uses "video confirmation" and "electrical quantity confirmation" as two independent criteria for fusion verification: if both indicate that the equipment has reached the expected state, the second verification passes; if they are inconsistent (e.g., the video shows the green light is on but the auxiliary contact is still closed), or either criterion fails to return within the timeout period, the verification fails.

[0081] After successful verification, the two-ticket safety control system sends a confirmation signal to the anti-misoperation intelligent agent, stating "Operation step confirmed complete, proceed to the next step." This signal includes a timestamp and electronic signature. If verification fails, the system automatically switches to manual confirmation mode: a prompt window pops up on the monitoring interface, displaying the robot's transmitted video, SCADA electrical data, and any discrepancies between the two. The on-duty maintenance personnel then make a comprehensive judgment and manually click "Confirm Continue" or "Terminate Task." If manual confirmation continues, the system forces the verification to pass and records the reason for manual confirmation; if the task is terminated, the anti-misoperation intelligent agent stops all operations, and the robot enters a safe standby state.

[0082] 7. Task Loop and End After receiving the confirmation signal for the previous operation, the anti-malfunction agent checks whether the sequence number of the currently completed operation step equals the total number of steps in the operation ticket. If it is not the last step, the anti-malfunction agent increments the operation step counter by 1, reads the information for the next operation step, and then repeats steps 4-6 above to execute the subsequent operations in the operation ticket sequentially. If it is the last step, the entire switching operation task ends.

[0083] After the task is completed, the anti-error intelligent agent sends a task completion notification to the station-side anti-error system and the two-ticket security control system, along with a complete summary of the task (operation ticket number, start and end times, execution results of each step, exception records, etc.). Simultaneously, the robot automatically performs a quick self-check and uploads the self-check data. Then, according to a preset strategy, it returns to the charging point to standby or enters a low-power sleep state. The backend system generates a detailed task report for maintenance personnel to archive and audit.

[0084] If any of the following abnormal situations occur during the entire operation process described above, the robot and the anti-misoperation intelligent agent will initiate the corresponding abnormal handling procedures.

[0085] (1) Mechanical jamming anomaly: During operation, if the robot detects a torque value exceeding the set safety threshold (e.g., 30 N·m) for more than 0.5 seconds via the torque sensor at the end of the arm, and the position encoder shows no change in angle or stroke, it is determined that the equipment is jammed due to a mechanical fault. The robot immediately stops the current operation, maintains the current posture of the robotic arm (to avoid injury from sudden release), and sends an "Operational anomaly - mechanical jamming" command to the anti-misoperation agent, along with the torque curve and position data. A high-priority alarm pops up in the background, prompting maintenance personnel to go to the site for handling. The robot will not automatically attempt to repeat the operation before manual intervention.

[0086] (2) Communication Interruption Anomaly: If the wireless communication link (5G / WiFi) between the robot and the anti-misoperation agent is interrupted for more than 5 seconds, the robot will immediately stop all movements (including walking and robotic arm movements) after detecting a timeout of the heartbeat packet locally, enter a safe standby state, and attempt to reconnect. After communication is restored, the robot will report its status (position, joint angles, etc.) during the interruption to the anti-misoperation agent, which will then decide whether to continue the task or re-verify. The background system records the communication interruption event and its duration for network quality analysis.

[0087] (3) Status Conflict Anomaly: During the pre-operation self-check, if the current status of the equipment identified by the robot seriously conflicts with the preconditions required by the operation ticket (e.g., the instruction is "open circuit breaker" but the circuit breaker is already in the open state), or if the actual status identified during the post-operation self-check does not match the expected status (e.g., the indicator light is still red after the "open" operation), the robot immediately stops subsequent operations, reports "status verification anomaly," and attaches detailed information and image evidence of the conflict. The anti-misoperation agent sets the task status to "abnormal pause," awaiting manual confirmation. The robot will not automatically perform any operations before receiving explicit manual instructions.

[0088] Through the close coordination of the above steps, the anti-misoperation control method provided in this embodiment realizes closed-loop management of the entire process of switching operations, from operation ticket generation, double verification, on-site execution to status confirmation, effectively preventing the occurrence of electrical misoperation accidents and significantly improving the safety and intelligence level of power system operation and maintenance.

[0089] 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 them; under the concept of the present invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the present invention as described above, which are not provided in detail for the sake of brevity; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A power switching operation robot, characterized in that, include: The humanoid robot body is equipped with a bipedal walking mechanism, a multi-level anti-collision system and a fault self-diagnosis function. It is used to autonomously navigate within the station, move to the target operating interval and perform switching operations. The dedicated operating arm is a multi-degree-of-freedom serial articulated robotic arm, mounted on the shoulder rotation platform of the robot body, and equipped with a tool quick-change interface at the end; Specialized unlocking and operating tools, including handcart operating tools, emergency tripping tools, positioning cameras, grounding switch operating devices, and partial discharge detection probes, can be automatically changed through the tool quick-change interface; The anti-misoperation intelligent agent is integrated into the robot's main unit and communicates with the station-end anti-misoperation system, the two-ticket safety management system, and the power monitoring system. It is used for operation ticket parsing, task scheduling, multi-source status verification, video analysis, knowledge storage, and embodied intelligent decision-making. The anti-misoperation intelligent agent has a built-in power industry-specific anti-misoperation model. The intelligent anti-misoperation terminal is deployed on the side of high-voltage electrical equipment and has functions such as equipment identity verification, status acquisition, remote interlocking control, and status indication.

2. The power switching operation robot as described in claim 1, characterized in that, The humanoid robot body adopts a free movement mode, and achieves in-situ rotation around its own vertical axis through differential movement of the two legs and coordinated movement of the hip and ankle joints; the navigation and positioning integrates visual sensors, radar system, inertial measurement unit and tactile sensor, and is equipped with an optional outdoor navigation module; the multi-level anti-collision system includes non-contact detection and protection composed of lidar and ultrasonic sensors and contact-type buffer strip system.

3. The power switching operation robot as described in claim 1, characterized in that, The fault self-check covers three stages: before operation, during operation, and after operation. It includes power status self-check, communication status self-check, motion mechanism self-check, operating arm and end tool self-check, and sensor self-check. During the pre-operation self-check, if a logical conflict is detected between the current state of the equipment and the operation command to be executed, the operation will be refused and an error warning will be issued. During the post-operation self-check, if the actual state of the equipment is found to be inconsistent with the operation expectation, an error warning will be issued and manual intervention will be required.

4. The power switching operation robot as described in claim 1, characterized in that, The error prevention intelligent agent includes an operation ticket parsing and task scheduling module, a multi-source status verification and video analysis module, an external system interaction interface module, a knowledge storage and log management module, and an embodied intelligence and continuous learning module, wherein: The operation ticket parsing and task scheduling module is used to receive and parse operation tickets, generate a sequence of instructions that can be executed by the robot, and coordinate multiple robots and multiple tasks; the operation ticket parsing uses a large power industry model for semantic understanding and structured information extraction. The multi-source status verification and video analysis module is used to collect robot vision information, intelligent anti-misoperation terminal status and electrical quantity data of power monitoring system, and to identify equipment number, indicator light status, mechanical position and operating tools through deep learning model; The external system interaction interface module is used for standardized data interaction with the power professional big data model, the two-ticket safety management system, the station-end error prevention system, the PMS / ERP system, and the power monitoring system. The knowledge storage and log management module is used to store robot operation procedures, error prevention logic rule base, equipment status judgment model and task execution logs; The embodied intelligence and continuous learning module is used to achieve multimodal perception fusion, autonomous planning and real-time adjustment, natural language interaction, remote expert guidance, and online model reinforcement and retraining.

5. The power switching operation robot as described in claim 1, characterized in that, The intelligent anti-misoperation terminal has a unique ID and a QR code or barcode on its surface for device identification verification. It integrates an RFID tag for redundancy backup. The terminal has built-in miniature limit switches, reed switches, or mechanical contacts to collect the unlocking / locking status, door position, and temporary grounding wire connection status. The terminal has a locking mechanism driven by a miniature electromagnet inside. After receiving the unlocking command from the anti-misoperation intelligent agent, it performs remote unlocking and automatically locks after the operation is completed or receives a locking command. The terminal panel has LED indicator lights for on-site status indication.

6. A method for preventing misoperation of the power switching operation robot according to any one of claims 1 to 5, characterized in that, Including the following steps: Operation ticket generation and preprocessing: Operation tickets are generated on the station-side error prevention system or imported from the production management system. The error prevention system performs structured parsing and item-by-item error prevention logic verification. After verification and approval, the operation ticket is transmitted to the error prevention intelligent agent. Operation ticket parsing and task issuance: The anti-misoperation intelligent agent parses the operation ticket to obtain the target interval position, device list and operation task sequence, and generates a navigation path; Double confirmation of operation interval: The robot moves to the front of the target interval and visually identifies the target to complete the double confirmation; Error prevention logic verification: For the current operation step, the error prevention agent sends an error prevention logic verification request to the station-side error prevention system. It can only continue after receiving the permission instruction, thus forming the first layer of error prevention verification. Robot on-site operation and self-check: The robot sequentially performs voice voting, locates the target, confirms the target, executes the operation, and collects video after the operation, and performs a status self-check before and after the operation. Video analysis and status feedback: The anti-misoperation intelligent agent performs AI video analysis on the images after the operation, and feeds the analysis results back to the two-ticket safety control system. It is then fused and verified with the electrical quantity status of the power monitoring system to form a second layer of anti-misoperation verification. Task Loop and End: After receiving the confirmation signal from the previous step, the anti-misoperation agent determines whether it is the last step. If so, the task ends; otherwise, it repeats the steps of anti-misoperation logic verification up to video analysis and status feedback.

7. The anti-misoperation control method as described in claim 6, characterized in that, In the dual confirmation of the operation interval, visual number recognition uses an OCR model to identify the numerical and textual numbers on the interval number plate and compares them with the operation ticket. Map number association matches the robot's current positioning coordinates with the interval coordinate range pre-stored in the map database. After both confirmations are passed, the robot sends a signal to the anti-misoperation intelligent agent to indicate that the interval confirmation is complete.

8. The anti-misoperation control method as described in claim 6, characterized in that, In the aforementioned error prevention logic verification, the error prevention system verifies the equipment status based on the error prevention logic rule base and the real-time data collected by the power monitoring system. After the verification is passed, an operation permission instruction containing a timestamp and digital signature is issued. If the verification fails, the anti-error agent terminates the task and prompts that the operation conditions are not met.

9. The anti-misoperation control method as described in claim 6, characterized in that, During the robot's on-site operation and self-check: the voice announcement announces the current operation step through the robot's speaker and supports two-way voice communication in the background; the positioning of the operation object uses a multi-servo algorithm of vision and tactile sense to make the tool's center point approach the operation point until the deviation is less than a set threshold; when confirming the operation object, the device identifier next to the operation point is read again and compared with the operation ticket for a second time; when performing the operation, the end tool is automatically changed according to the operation type, and torque monitoring is used to determine whether mechanical jamming has occurred; after the operation, video acquisition includes shooting the indicator light area, mechanical indicator, or position indication in the observation window.

10. The anti-misoperation control method as described in claim 6, characterized in that, It also includes anomaly handling mechanisms: when the torque abnormally increases beyond the threshold during operation and the position feedback does not change, it is determined to be a mechanical jamming anomaly, and the robot immediately stops operation and reports it; when the communication of the robot's anti-misoperation intelligent agent is interrupted for more than the set time, the robot automatically stops all movement and enters a safe standby state, and continues or re-verifies after the communication is restored; when a state conflict is found during the pre-operation self-check or post-operation self-check, the robot stops subsequent operations and reports a state verification anomaly, waiting for manual confirmation.

Images