AI auxiliary attitude correction algorithm for main transformer replacement of large hydropower station

Abstract

The invention discloses a large hydropower station main transformer replacement AI auxiliary attitude correction algorithm. The algorithm comprises the steps that main transformer replacement auxiliary positioning equipment is deployed, and a holder of a transverse visual image deviation measuring device and a tripod of a longitudinal laser distance measuring device are adjusted; in the early stage of traction, data are continuously collected through the measuring devices, data fusion and track prediction are carried out through an AI auxiliary analysis prediction algorithm in the main control equipment of the positioning device, and a real-time deviation correction operation prompt is provided for an operator through the mobile display panel system until the tail end butt joint deviation is smaller than a set value; based on the method of S3, deviation rectification and traction are carried out at the same time, when traction is carried out within the last 2-meter distance, the tail end image positioning process is started, and tail end positioning of the high-voltage bushing is completed; the problems that in the prior art, the measurement precision is not stable, the operation efficiency is low, and workers operate in dangerous areas are effectively solved, and the safety, precision and efficiency of main transformer replacement operation are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of transformer replacement traction technology, and in particular relates to an AI-assisted attitude correction algorithm for the replacement of main transformers in large hydropower stations. Background Technology

[0002] In the operation and maintenance of large hydropower stations, the main transformer, as the core power equipment of the entire power generation system, requires regular maintenance and replacement, which is a crucial and highly technically complex operation. According to the industry standard "DL / T 573-2010 Guidelines for the Maintenance of Power Transformers," when the main transformer needs to be removed from the powerhouse for maintenance, it must be smoothly pulled to the maintenance position using a pre-designed dedicated track transport system and traction equipment. After maintenance is completed, it is precisely pulled back to its installation position along the original track, ultimately achieving precise connection between the A / B / C phase bushings on its high-voltage side and the corresponding conductors in the GIS pipeline. The precision requirements for this final connection are extremely stringent, typically requiring the spatial deviation of the center axis in both the lateral and longitudinal directions to be strictly controlled within ±10 mm. Any deviation beyond this tolerance may lead to abnormal stress at the electrical connection, seal failure, or partial discharge, thus creating serious operational safety hazards. Therefore, how to achieve high-precision positioning and high-efficiency end-point docking of the main transformer, a heavy-duty equipment weighing hundreds of tons, during long-distance traction in the specific industrial environment of a hydropower plant with complex structure and limited space has always been a core technical problem that has long existed in this field and urgently needs to be solved.

[0003] Currently, the closest and most widely adopted existing technology in the industry is a mechanical measurement and positioning system that relies entirely on manual operation. The specific implementation of this traditional technical solution is as follows: First, operators use a frame level to calibrate the track plane to establish a horizontal benchmark. Then, multiple sets of nylon plumb lines are suspended on-site to establish a spatial measurement benchmark—one set of plumb lines is suspended directly above the track wheels of the main transformer transfer trolley for visual monitoring of lateral misalignment between the wheel and rail, i.e., to prevent "rail biting"; another set of plumb lines is precisely fixed above the center point of the GIS conductor and the center of the main transformer's high-voltage bushing, and through repeated adjustments, they are brought to the same horizontal plane after being lowered. Throughout the traction operation, 2 to 3 operators are required to work together. The main transformer must be stopped every 0.5 to 1 meter it moves. The operators need to bend down into the narrow space at the bottom of the main transformer, which is usually less than 500 mm high, or climb to the top of the high-voltage equipment. They use a steel ruler or steel tape measure to manually read and record the projection distance of each set of plumb lines on the rail surface or marking plate. The offset is calculated manually, and then the operators use their experience to guide the adjustment of the length of the left and right traction ropes or the output of the traction machine to correct the main transformer's traveling posture and final stopping position.

[0004] Although this traditional method largely supported the operation and maintenance of hydropower stations during a specific historical period, its inherent technical limitations have become glaringly apparent in the context of modern industry's pursuit of "high precision, high efficiency, and high safety," and it suffers from the following numerous insurmountable defects: 1. Low measurement accuracy and poor stability: The core measurement benchmark—the plumb line—inevitably experiences a 2-5 mm sway due to the suspension height, equipment vibration, and airflow disturbances, causing the benchmark itself to drift unpredictably. Simultaneously, in space-constrained measurement locations, it's difficult for personnel to maintain a perpendicular line of sight to the steel ruler's scale, inevitably leading to parallax and reading errors. The combination of manual readings and benchmark drift makes it difficult to consistently guarantee overall measurement accuracy, with actual alignment deviations often exceeding the ±15 mm safety tolerance.

[0005] 2. The operational process is extremely inefficient: To achieve the coplanar requirement for the final docking, debugging just three sets of plumb lines takes 15 to 24 minutes. The entire traction process requires frequent starting and stopping of the traction equipment, and a 5-10 meter transfer path often takes more than 3 hours, severely impacting critical maintenance windows. If the final docking fails and rework is required, simply redeploying the traction rigging in the reverse direction can take an additional 4 hours or more, significantly delaying the entire unit's power restoration schedule.

[0006] 3. Poor spatial adaptability and operational safety: The operating space at the bottom of the main transformer is extremely narrow, and the ground is often paved with uneven pebbles, making personnel movement and tool operation extremely difficult. The high-voltage riser pipes at the top of the main transformer are densely distributed, with a spacing of only 150-300 mm, severely restricting the position and movement of workers, making operations such as plumb line hanging and marking extremely inconvenient and easily obstructed. More importantly, personnel must perform high-risk operations in a complex environment near high-voltage equipment, in a wet and slippery environment with residual insulating oil, posing a high risk of tool falls and personnel slipping.

[0007] 4. Complete lack of dynamic monitoring and real-time correction capabilities: This technology relies on discrete point sampling measurement, resulting in an extremely low data update frequency. It cannot capture minute instantaneous deviations caused by track joints, vibrations, etc., during traction intervals. This data blind spot causes significant delays in adjustment operations, making preventative and precise correction impossible. It can only rely on the operator's experience for "post-event remediation," which is prone to secondary deviations due to over-adjustment.

[0008] 5. Outdated technology and poor economic efficiency: This solution relies entirely on manual experience and mechanical measuring tools. Measurement results are significantly affected by the operator's skill level and sense of responsibility. Data cannot be automatically recorded and traced, which is seriously out of step with the trend of automation and digitalization in modern industry. At the same time, frequent downtime results in the traction equipment's idle waiting time accounting for more than 35%, leading to low equipment utilization. Moreover, each operation requires 2-3 man-days, resulting in high labor and time costs.

[0009] In summary, existing technologies, due to their inherent limitations, rely entirely on manual experience and rudimentary tools. Their technical means are significantly lagging behind the development level of modern industrial measurement and intelligent control, and cannot meet the urgent needs for millimeter-level accuracy, continuous processes, and high-level safety assurance in the replacement of main transformers in large hydropower stations. Therefore, it is necessary to propose an AI-assisted attitude correction algorithm for the replacement of main transformers in large hydropower stations to solve the above problems. Summary of the Invention

[0010] The technical problem to be solved by this invention is to provide an AI-assisted attitude correction algorithm for the replacement of main transformers in large hydropower stations, in order to solve the problems of existing technologies. The technical solution adopted by this invention is as follows: The replacement of main transformers in large hydropower stations uses AI-assisted attitude correction algorithms, including: The process of replacing the AI-assisted attitude correction algorithm on the main transformer of a large hydropower station includes the following steps: S1, Deploy auxiliary positioning equipment for main transformer replacement: Deploy the longitudinal laser ranging device, the lateral visual image deviation measurement device, the end image positioning and monitoring device, the operation scene observation device, the image processing host and the main control equipment of the positioning device to the designated location and power them on; S2, System Calibration: Adjust the gimbal of the horizontal visual image offset measuring device to ensure that the track wheel and track are within the frame; Adjust the tripod of the longitudinal laser rangefinder to ensure that the laser point hits the rangefinder target set on the upper plane of the main transformer; S3, AI-assisted traction positioning: During the initial stage of traction, data is continuously collected by various measuring devices. The AI-assisted analysis and prediction algorithm in the main control device of the positioning device performs data fusion and trajectory prediction, and provides real-time correction operation prompts to the operator through the mobile display tablet system until the end docking deviation is less than the set value. S4, End Image Localization: The S3-based method corrects deviations while pulling, and when the pulling reaches the last 2 meters, the end image positioning process is initiated to complete the end positioning of the high-voltage bushing.

[0011] Preferably, when deploying the auxiliary positioning device for main transformer replacement in step S1, the auxiliary positioning device for main transformer replacement and its deployment location are as follows: The longitudinal laser ranging device is installed at the starting end of the main transformer traction track, perpendicular to the front surface of the main transformer, and is used to monitor the movement distance of the front surface of the transformer; the longitudinal laser ranging device includes a dual-beam laser intelligent ranging detector and a tripod for supporting and adjusting the detector. A lateral visual image deviation measurement device is installed at the four wheels on the bottom of the front surface of the transformer to monitor the deviation of the track wheels from the track. The lateral visual image deviation measurement device includes an image visual deviation measurement detector, a lateral deviation measurement mounting bracket for fixing it to the transformer skirt, and a quick-clamping device for quickly fixing the detector to the mounting bracket. The end-point image positioning and monitoring device is deployed on the top of the high-voltage bushing for positioning and aligning the high-voltage bushing and GIS conductor. A work scene observation device, wherein the work scene observation device is a scene image monitoring camera with a magnetic base or air suction cup; The image processing host is communicatively connected to the lateral visual image deviation measurement device, the end image positioning monitoring device, and the work scene observation device. The positioning device's main control unit is communicatively connected to the longitudinal laser ranging device and the image processing host, and integrates an AI-assisted analysis and prediction algorithm; The mobile display tablet system is communicatively connected to the main control device of the positioning device. The equipment power supply system provides power to the image processing host and the main control equipment of the positioning device.

[0012] Preferably, in step S2, the lateral deviation measurement mounting bracket is fixed by a handwheel fixing mechanism. The handwheel fixing mechanism includes a U-shaped bracket that engages with the skirt of the main transformer. A locking wheel is threaded to the top of the bracket, and pads are connected to the bottom of the locking wheel and the inner wall of the bracket for clamping and locking. The bottom of the handwheel fixing mechanism is slidably connected to an adjusting bracket with a cross slide structure to achieve position adjustment in the longitudinal and lateral directions. A pan-tilt unit is connected to the bottom of the adjusting bracket via a universal joint, and the image visual lateral deviation measurement detector is mounted on the fast-locking device via the pan-tilt unit.

[0013] Preferably, in step S3, the AI-assisted analysis and prediction algorithm performs real-time fusion calculation of longitudinal displacement and lateral offset, and predicts the direction of the main transformer's movement trajectory trend line.

[0014] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes real-time calculation of the main transformer attitude parameters, as follows: The gap values ​​between the four track wheels and the track were collected in real time by four sets of lateral visual image offset measurement devices, respectively. , , and ; The feed rates of the left and right traction ropes are collected using a longitudinal laser rangefinder. and Let the distance between the front and rear wheels be L, and the distance between the traction points be... Real-time calculation of the three core attitude parameters of the main transformer: Lateral offset of centroid : ; When Dx > 0, the principal variable's mass center is biased to the right; otherwise, it is biased to the left. Center of mass deflection angle : ; ; ; when When the value is greater than 0, the main transformer shifts to the right; otherwise, it shifts to the left. Center of mass feed : ; When Dy > 0, it indicates that the main transformer is being pulled in.

[0015] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the calculation of correction angular velocity based on PD control, as follows: The lateral offset of the current centroid of the main transformer obtained from the above control cycle calculation and current centroid deflection angle As input, the proportional-derivative controller calculates the correction angular velocity required to bring the main transformer back to the centerline. The calculation formula is as follows: ; in, For proportional control operators, For differential control operators, This represents the current forward velocity of the main variable's center of mass.

[0016] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the decomposition of left and right traction velocities, as follows: The calculated correction angular velocity Combined with the current forward speed of the main variable's center of mass and the distance between traction points Decomposed into the speed of the left and right traction ropes perpendicular to the surface of the main transformer box. and : ; in, The current forward velocity of the main variable's center of mass. This refers to the distance between traction points.

[0017] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the speed conversion of the traction rope along the track direction and command output, as follows: The traction speed perpendicular to the box body and According to the current deflection angle of the principal variable's center of mass Converted to actual traction speed along the track direction and The calculation formula is as follows: ; ; The calculated and As a quantitative correction command, it is output to the traction control system via the mobile display tablet system to adjust the feed speed of the left and right traction ropes.

[0018] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes state updates and iterations, as follows: At a control time step Then, update the motion state of the main transformer, including: New deflection angle ; New speed of progress ; New lateral offset ; Will , and This serves as the initial state for the next control cycle, and the process returns to step S301 for iterative calculation until the centroid offset of the main transformer is reached. and the angle of deflection of the center of mass All values ​​are reduced to zero, thus completing the correction.

[0019] Preferably, the end-image localization process in step S4 includes: S401, The end image positioning and monitoring detector in the end image positioning and monitoring device is deployed at the top of the high voltage bushing to collect images of the high voltage bushing and GIS conductor; The end-image positioning monitoring detector works in conjunction with the mounting frame. The mounting frame includes a cross structure composed of four telescopic rods. The monitoring detector is positioned at the center of the four telescopic rods. The telescopic rods are used to adjust the position of the end-image positioning monitoring detector and are locked by a handle to prevent directional deviation. The mounting frame also includes a tapered pin rotatably connected to the end of the telescopic rod for fixing it to the mounting hole or extension of the high-pressure bushing riser pipe. The telescopic rod has a threaded sleeve structure and is marked with graduations. It is locked by a locking bolt. The target's mounting plate is connected to the threaded holes on the GIS conductor via the four corner holes, and a weight is suspended at the center of the mounting plate and hangs down naturally. S402, the end image positioning and monitoring device is fixed to the top of the high-voltage bushing of the main transformer using its dedicated mounting bracket; by adjusting the telescopic rod and direction adjustment handle of the mounting bracket, the line of sight of the detector is basically coincident with the axis of the high-voltage bushing, and its field of view can cover the GIS conductor area in front. The target is precisely fixed at the center of the GIS conductor, and its weight structure ensures that its geometric center is aligned with the axis of the GIS conductor in a natural hanging state. The image processing host calibrates the focal length, principal point, distortion coefficient, and position and attitude relative to the orbital coordinate system of the deployed end image positioning and monitoring device detector, and establishes the transformation relationship from image pixel coordinates to actual physical space coordinates. S403, the end-point image positioning monitoring detector measures the deviation between the high-voltage bushing axis and the GIS conductor axis through image processing algorithms, as detailed below: When the main transformer enters the final connection stage, the system executes the following cyclic calculation process: Image capture: The end-point image positioning and monitoring device acquires images in real time, including the front-end features of the high-voltage bushing and the target on the GIS conductor, at a frequency of not less than 10Hz; Feature extraction and recognition: The image processing host receives the image and runs visual algorithms. Target recognition and center localization: Using the specific geometric pattern of the target, the target is accurately identified in the image by combining Hough transform or contour extraction with ellipse fitting algorithm, and its image center coordinates are calculated with sub-pixel accuracy. Bushing reference feature positioning: At the same time, the edge of the high-pressure bushing flange is identified as the reference feature, and its image center coordinates are calculated; Spatial bias calculation: Calculate the pixel deviation between the target center and the cannula reference center in the image coordinate system; Using pre-calibrated camera parameters and known actual target dimensions, pixel deviation is calculated as the actual physical deviation between the high-voltage bushing axis and the GIS conductor axis at the end mating surface perpendicular to the track plane through perspective projection model or direct linear transformation, including lateral deviation and longitudinal deviation. S404, AI-assisted decision-making and correction instruction generation: The main control equipment of the positioning device receives the actual physical deviation from the end positioning system and outputs instructions. The generated instructions are displayed through the mobile display tablet system and simultaneously sent to the traction control system for traction control.

[0020] Furthermore, the entire traction process is divided into two stages. In the early traction stage, that is, from the starting position to the last 2 meters, the AI-assisted traction positioning method in step S3 is used to perform real-time fusion calculation of longitudinal displacement and lateral offset for traction correction. The second stage is the end positioning stage within the last 2 meters. At this time, the transformer traction is almost completed, and there is no need to consider the rail biting problem. The traction correction at this time is mainly achieved based on the end positioning in step S4, without the need for the intervention of the longitudinal laser ranging device and the lateral visual image offset measurement device.

[0021] The beneficial effects of this invention are as follows: 1. This invention fundamentally solves the inherent defects of low accuracy and poor stability of traditional manual measurement methods by constructing a non-contact multi-dimensional intelligent measurement system. By deploying a longitudinal laser ranging device, a lateral visual image deviation measurement device, and an end-point image positioning monitoring device, a complete non-contact measurement system is constructed. This system uses laser and image as measurement benchmarks, completely eliminating the fundamental errors caused by physical contact and benchmark oscillation. At the same time, advanced algorithms such as sub-pixel edge detection integrated into the image processing host achieve ultra-fine recognition of edge features, upgrading the measurement process from relying on human eye interpretation to machine intelligent recognition. This makes the measurement results no longer affected by on-site environmental vibrations, operator skill level, and viewing angle, solving the long-standing pain points of unreliable measurement benchmarks and data accuracy varying with the operator in this field.

[0022] 2. This invention effectively overcomes the core shortcomings of existing technologies, such as low operational efficiency and lagging dynamic response, by establishing an AI-based real-time data fusion and dynamic correction mechanism. By enabling each system to collect high-frequency data of no less than 10Hz throughout the traction process, and using the AI-assisted analysis and prediction algorithm in the main control equipment of the positioning device for real-time fusion calculation, continuous and seamless monitoring of the main transformer's spatial attitude is achieved. This system can instantly capture minute deviation trends caused by track joints and instantaneous fluctuations in traction force, and generate quantified correction commands in advance, enabling traction operations to be carried out almost continuously. This saves the huge time cost of frequent start-stop operations, significantly improves operational efficiency and first-time success rate, and solves the historical problem of process interruption and adjustment lag caused by discrete measurement.

[0023] 3. This invention, through its modular, universally adaptable, and remote monitoring design, successfully addresses the severe challenges of poor adaptability and insufficient operational safety in existing technologies when applied across power stations. The hardware adopts a "universal module + adaptable component" architecture, while the software layer establishes a universal algorithm model decoupled from individual transformer parameters through track reference normalization and scenario parameter self-learning. This allows the equipment to be quickly deployed and reused in different power stations across the entire river basin after initial modeling. Furthermore, remote monitoring and command reception via a mobile display tablet system eliminate the need for operators to enter the high-voltage areas at the bottom or top of the transformer; they can maintain overall control from a safe area, improving the equipment's versatility, ease of use, and inherent safety level. Attached Figure Description

[0024] Figure 1 This is a plan view showing the arrangement of the device structure according to the present invention; Figure 2 This is a three-dimensional schematic diagram showing the arrangement of the device structure according to the present invention; Figure 3 This is a schematic diagram of the device structure of the present invention; Figure 4 This is a schematic diagram of the end-image positioning and monitoring device of the present invention; Figure 5 This is a schematic diagram of the transverse visual image deviation measurement device of the present invention; Figure 6 This is a schematic diagram of the target structure of the end-image positioning and monitoring device of the present invention; Figure 7 This is a schematic diagram of AI-assisted traction positioning in an embodiment of the present invention; Figure 8 The figure shows the simulation results of the algorithm verification in the embodiment of the present invention; Figure 9 This is a diagram of the pullback positioning software interface designed according to the algorithm in an embodiment of the present invention; Figure 10 This is a diagram showing the software interface of the software system in an embodiment of the present invention. Detailed Implementation

[0025] Example 1: like Figures 1-6 As shown, the replacement of the AI-assisted attitude correction algorithm for the main transformer of a large hydropower station includes the following steps: S1, Deploy auxiliary positioning equipment for main transformer replacement: Deploy the longitudinal laser ranging device, the lateral visual image deviation measurement device, the end image positioning and monitoring device, the operation scene observation device, the image processing host and the main control equipment of the positioning device to the designated location and power them on; S2, System Calibration: Adjust the gimbal of the horizontal visual image offset measuring device to ensure that the track wheel and track are within the frame; Adjust the tripod of the longitudinal laser rangefinder to ensure that the laser point hits the rangefinder target set on the upper plane of the main transformer; S3, AI-assisted traction positioning: During the initial stage of traction, data is continuously collected by various measuring devices. The AI-assisted analysis and prediction algorithm in the main control device of the positioning device performs data fusion and trajectory prediction, and provides real-time correction operation prompts to the operator through the mobile display tablet system until the end docking deviation is less than the set value. S4, End Image Localization: The S3-based method corrects deviations while pulling, and when the pulling reaches the last 2 meters, the end image positioning process is initiated to complete the end positioning of the high-voltage bushing.

[0026] Preferably, when deploying the auxiliary positioning device for main transformer replacement in step S1, the auxiliary positioning device for main transformer replacement and its deployment location are as follows: The longitudinal laser ranging device is installed at the starting end of the main transformer traction track, perpendicular to the front surface of the main transformer, and is used to monitor the movement distance of the front surface of the transformer; the longitudinal laser ranging device includes a dual-beam laser intelligent ranging detector and a tripod for supporting and adjusting the detector. A lateral visual image deviation measurement device is installed at the four wheels on the bottom of the front surface of the transformer to monitor the deviation of the track wheels from the track. The lateral visual image deviation measurement device includes an image visual deviation measurement detector, a lateral deviation measurement mounting bracket for fixing it to the transformer skirt, and a quick-clamping device for quickly fixing the detector to the mounting bracket. The end-point image positioning and monitoring device is deployed on the top of the high-voltage bushing for positioning and aligning the high-voltage bushing and GIS conductor. A work scene observation device, wherein the work scene observation device is a scene image monitoring camera with a magnetic base or air suction cup; The image processing host is communicatively connected to the lateral visual image deviation measurement device, the end image positioning monitoring device, and the work scene observation device. The positioning device's main control unit is communicatively connected to the longitudinal laser ranging device and the image processing host, and integrates an AI-assisted analysis and prediction algorithm; The mobile display tablet system is communicatively connected to the main control device of the positioning device. The equipment power supply system provides power to the image processing host and the main control equipment of the positioning device.

[0027] Preferably, in step S2, the lateral deviation measurement mounting bracket is fixed by a handwheel fixing mechanism. The handwheel fixing mechanism includes a U-shaped bracket that engages with the skirt of the main transformer. A locking wheel is threaded to the top of the bracket, and pads are connected to the bottom of the locking wheel and the inner wall of the bracket for clamping and locking. The bottom of the handwheel fixing mechanism is slidably connected to an adjusting bracket with a cross slide structure to achieve position adjustment in the longitudinal and lateral directions. A pan-tilt unit is connected to the bottom of the adjusting bracket via a universal joint, and the image visual lateral deviation measurement detector is mounted on the fast-locking device via the pan-tilt unit.

[0028] Preferably, in step S3, the AI-assisted analysis and prediction algorithm performs real-time fusion calculation of longitudinal displacement and lateral offset, and predicts the direction of the main transformer's movement trajectory trend line.

[0029] like Figure 7 As shown, preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes real-time calculation of the main transformer attitude parameters, as detailed below: The gap values ​​between the four track wheels and the track were collected in real time by four sets of lateral visual image offset measurement devices, respectively. , , and ; The feed rates of the left and right traction ropes are collected using a longitudinal laser rangefinder. and Let the distance between the front and rear wheels be L, and the distance between the traction points be... Real-time calculation of the three core attitude parameters of the main transformer: Lateral offset of centroid : ; When Dx > 0, the principal variable's mass center is biased to the right; otherwise, it is biased to the left. Center of mass deflection angle : ; ; ; when When the value is greater than 0, the main transformer shifts to the right; otherwise, it shifts to the left. Center of mass feed : ; When Dy > 0, it indicates that the main transformer is being pulled in.

[0030] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the calculation of correction angular velocity based on PD control, as follows: The lateral offset of the current centroid of the main transformer obtained from the above control cycle calculation and current centroid deflection angle As input, the proportional-derivative controller calculates the correction angular velocity required to bring the main transformer back to the centerline. The calculation formula is as follows: ; in, For proportional control operators, For differential control operators, This represents the current forward velocity of the main variable's center of mass.

[0031] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the decomposition of left and right traction velocities, as follows: The calculated correction angular velocity Combined with the current forward speed of the main variable's center of mass and the distance between traction points Decomposed into the speed of the left and right traction ropes perpendicular to the surface of the main transformer box. and : ; in, The current forward velocity of the main variable's center of mass. This refers to the distance between traction points.

[0032] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the speed conversion of the traction rope along the track direction and command output, as follows: The traction speed perpendicular to the box body and According to the current deflection angle of the principal variable's center of mass Converted to actual traction speed along the track direction and The calculation formula is as follows: ; ; The calculated and As a quantitative correction command, it is output to the traction control system via the mobile display tablet system to adjust the feed speed of the left and right traction ropes.

[0033] Preferably, in the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes state updates and iterations, as follows: At a control time step Then, update the motion state of the main transformer, including: New deflection angle ; New speed of progress ; New lateral offset ; Will , and This serves as the initial state for the next control cycle, and the process returns to step S301 for iterative calculation until the centroid offset of the main transformer is reached. and the angle of deflection of the center of mass All values ​​are reduced to zero, thus completing the correction.

[0034] Preferably, the end-image localization process in step S4 includes: S401, The end image positioning and monitoring detector in the end image positioning and monitoring device is deployed at the top of the high voltage bushing to collect images of the high voltage bushing and GIS conductor; The end-image positioning monitoring detector works in conjunction with the mounting frame. The mounting frame includes a cross structure composed of four telescopic rods. The monitoring detector is positioned at the center of the four telescopic rods. The telescopic rods are used to adjust the position of the end-image positioning monitoring detector and are locked by a handle to prevent directional deviation. The mounting frame also includes a tapered pin rotatably connected to the end of the telescopic rod for fixing it to the mounting hole or extension of the high-pressure bushing riser pipe. The telescopic rod has a threaded sleeve structure and is marked with graduations. It is locked by a locking bolt. The target's mounting plate is connected to the threaded holes on the GIS conductor via the four corner holes, and a weight is suspended at the center of the mounting plate and hangs down naturally. S402, the end image positioning and monitoring device is fixed to the top of the high-voltage bushing of the main transformer using its dedicated mounting bracket; by adjusting the telescopic rod and direction adjustment handle of the mounting bracket, the line of sight of the detector is basically coincident with the axis of the high-voltage bushing, and its field of view can cover the GIS conductor area in front. The target is precisely fixed at the center of the GIS conductor, and its weight structure ensures that its geometric center is aligned with the axis of the GIS conductor in a natural hanging state. The image processing host calibrates the focal length, principal point, distortion coefficient, and position and attitude relative to the orbital coordinate system of the deployed end image positioning and monitoring device detector, and establishes the transformation relationship from image pixel coordinates to actual physical space coordinates. S403, the end-point image positioning monitoring detector measures the deviation between the high-voltage bushing axis and the GIS conductor axis through image processing algorithms, as detailed below: When the main transformer enters the final connection stage, the system executes the following cyclic calculation process: Image capture: The end-point image positioning and monitoring device acquires images in real time, including the front-end features of the high-voltage bushing and the target on the GIS conductor, at a frequency of not less than 10Hz; Feature extraction and recognition: The image processing host receives the image and runs visual algorithms. Target recognition and center localization: Using the specific geometric pattern of the target, the target is accurately identified in the image by combining Hough transform or contour extraction with ellipse fitting algorithm, and its image center coordinates are calculated with sub-pixel accuracy. Bushing reference feature positioning: At the same time, the edge of the high-pressure bushing flange is identified as the reference feature, and its image center coordinates are calculated; Spatial bias calculation: Calculate the pixel deviation between the target center and the cannula reference center in the image coordinate system; Using pre-calibrated camera parameters and known actual target dimensions, pixel deviation is calculated as the actual physical deviation between the high-voltage bushing axis and the GIS conductor axis at the end mating surface perpendicular to the track plane through perspective projection model or direct linear transformation, including lateral deviation and longitudinal deviation. S404, AI-assisted decision-making and correction instruction generation: The main control equipment of the positioning device receives the actual physical deviation from the end positioning system and outputs instructions. The generated instructions are displayed through the mobile display tablet system and simultaneously sent to the traction control system for traction control.

[0035] Furthermore, the entire traction process is divided into two stages. In the early traction stage, that is, from the starting position to the last 2 meters, the AI-assisted traction positioning method in step S3 is used to perform real-time fusion calculation of longitudinal displacement and lateral offset for traction correction. The second stage is the end positioning stage within the last 2 meters. At this time, the transformer traction is almost completed, and there is no need to consider the rail biting problem. The traction correction at this time is mainly achieved based on the end positioning in step S4, without the need for the intervention of the longitudinal laser ranging device and the lateral visual image offset measurement device.

[0036] Example 2: The algorithm is constrained by taking into account the actual situation and the feasibility of operation for actual operators. The maximum actual driving speed is limited to 5 mm / s and the maximum deviation angle is not more than 0.5 degrees. Finally, the feed amount on the left and right sides of the main transformer is fed back. The operator can control the left and right traction force of the main transformer in real time by using the current actual position of the main transformer and the predicted left and right feed amount at the next moment, so as to finally correct the position of the main transformer.

[0037] like Figure 8 As shown, in and Under these conditions, adjusting the speed of the ropes on both sides allows the object to eventually return to the center position. However, the azimuth response curve shows that the absolute value of the maximum deflection angle exceeds 2 degrees in this situation. In reality, there are limitations on the maximum adjustable deflection angle. Therefore, it is necessary to enhance the control system's response to angles and slightly weaken the system's response to lateral deflection, i.e., increase... weakening the effect Effects and benefits. (From) Figure 10 It can be seen that the modification and Subsequently, the lateral deflection response was weakened while the yaw angle response was strengthened. During this adjustment process, the maximum absolute value of the heading angle was less than 0.5 degrees. It is evident that strengthening the lateral deflection angle resulted in a decrease in the maximum heading angle response.

[0038] like Figure 9 As shown, when acquiring traction section data, the traction section interface of the software is first accessed, and the marking measurement method is used. Specifically, a low-distortion industrial camera is used to precisely mark measurement lines on the track around the main transformer and at the track wheel stops. Then, four sets of clearance values ​​are carefully measured and recorded. The measured data is combined with the main transformer information data generated based on AI model training—that is, the main transformer motion attitude data—to generate and display the main transformer attitude data.

[0039] like Figure 10 As shown, when entering transformer information, it is necessary to strictly follow the system regulations, carefully compare with the transformer design drawings, and meticulously verify all parameters. Simultaneously, the physical transformer should be compared with the drawings to ensure mutual verification, thereby obtaining comprehensive and accurate transformer information and ensuring that the constructed model accurately matches the transformer's characteristics. After completing the above information collection and verification, the transformer information should be accurately entered into the software AI model.

Claims

1. An AI-assisted attitude correction algorithm for replacing the main transformer in a large hydropower station, characterized in that... Includes the following steps: S1, Deploy auxiliary positioning equipment for main transformer replacement: Deploy the longitudinal laser ranging device, the lateral visual image deviation measurement device, the end image positioning and monitoring device, the operation scene observation device, the image processing host and the main control equipment of the positioning device to the designated location and power them on; S2, System Calibration: Adjust the gimbal of the horizontal visual image offset measuring device to ensure that the track wheel and track are within the frame; Adjust the tripod of the longitudinal laser rangefinder to ensure that the laser point hits the rangefinder target set on the upper plane of the main transformer; S3, AI-assisted traction positioning: During the initial stage of traction, data is continuously collected by various measuring devices. The AI-assisted analysis and prediction algorithm in the main control device of the positioning device performs data fusion and trajectory prediction, and provides real-time correction operation prompts to the operator through the mobile display tablet system until the end docking deviation is less than the set value. S4, End Image Localization: The S3-based method corrects deviations while pulling, and when the pulling reaches the last 2 meters, the end image positioning process is initiated to complete the end positioning of the high-voltage bushing.

2. The AI-assisted attitude correction algorithm for replacing the main transformer in a large hydropower station according to claim 1, characterized in that, When deploying the auxiliary positioning equipment for main transformer replacement in step S1, the auxiliary positioning equipment for main transformer replacement and its deployment location are as follows: The longitudinal laser ranging device is installed at the starting end of the main transformer traction track, perpendicular to the front surface of the main transformer, and is used to monitor the movement distance of the front surface of the transformer; the longitudinal laser ranging device includes a dual-beam laser intelligent ranging detector and a tripod for supporting and adjusting the detector. A lateral visual image deviation measurement device is installed at the four wheels on the bottom of the front surface of the transformer to monitor the deviation of the track wheels from the track. The lateral visual image deviation measurement device includes an image visual deviation measurement detector, a lateral deviation measurement mounting bracket for fixing it to the transformer skirt, and a quick-clamping device for quickly fixing the detector to the mounting bracket. The end-point image positioning and monitoring device is deployed on the top of the high-voltage bushing for positioning and aligning the high-voltage bushing and GIS conductor. A work scene observation device, wherein the work scene observation device is a scene image monitoring camera with a magnetic base or air suction cup; The image processing host is communicatively connected to the lateral visual image deviation measurement device, the end image positioning monitoring device, and the work scene observation device. The positioning device's main control unit is communicatively connected to the longitudinal laser ranging device and the image processing host, and integrates an AI-assisted analysis and prediction algorithm; The mobile display tablet system is communicatively connected to the main control device of the positioning device. The equipment power supply system provides power to the image processing host and the main control equipment of the positioning device.

3. The AI-assisted attitude correction algorithm for replacing the main transformer in a large hydropower station according to claim 2, characterized in that, In step S2, the lateral deviation measurement mounting bracket is fixed by a handwheel fixing mechanism. The handwheel fixing mechanism includes a U-shaped bracket that engages with the skirt of the main transformer. A locking wheel is threaded to the top of the bracket, and pads are connected to the bottom of the locking wheel and the inner wall of the bracket for clamping and locking. The bottom of the handwheel fixing mechanism is slidably connected to an adjusting bracket with a cross slide structure to achieve position adjustment in the longitudinal and lateral directions. A pan-tilt unit is connected to the bottom of the adjusting bracket via a universal joint, and the image visual lateral deviation measurement detector is mounted on the fast-clamping device via the pan-tilt unit.

4. The positioning method for replacing the main transformer of a large hydropower station using an AI-assisted attitude correction algorithm according to claim 1, characterized in that, In step S3, the AI-assisted analysis and prediction algorithm performs real-time fusion calculation of longitudinal displacement and lateral offset, and predicts the direction of the trend line of the main transformer's movement trajectory.

5. The positioning method for replacing the main transformer of a large hydropower station using an AI-assisted attitude correction algorithm according to claim 1, characterized in that, In the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes real-time calculation of the main transformer attitude parameters, as detailed below: The gap values ​​between the four track wheels and the track were collected in real time by four sets of lateral visual image offset measurement devices, respectively. , , and ; The feed rates of the left and right traction ropes are collected using a longitudinal laser rangefinder. and Let the distance between the front and rear wheels be L, and the distance between the traction points be... Real-time calculation of the three core attitude parameters of the main transformer: Lateral offset of centroid : ； When Dx > 0, the principal variable's mass center is biased to the right; otherwise, it is biased to the left. Center of mass deflection angle : ； ； ； when When the value is greater than 0, the main transformer shifts to the right; otherwise, it shifts to the left. Center of mass feed : ； When Dy > 0, it indicates that the main transformer is being pulled in.

6. The positioning method for replacing the main transformer of a large hydropower station using an AI-assisted attitude correction algorithm according to claim 5, characterized in that, In the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the calculation of correction angular velocity based on PD control, as follows: The lateral offset of the current centroid of the main transformer obtained from the above control cycle calculation and current centroid deflection angle As input, the proportional-derivative controller calculates the correction angular velocity required to bring the main transformer back to the centerline. The calculation formula is as follows: ； in, For proportional control operators, For differential control operators, This represents the current forward velocity of the main variable's center of mass.

7. The positioning method for replacing the main transformer of a large hydropower station using an AI-assisted attitude correction algorithm as described in claim 6, characterized in that, In the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the decomposition of left and right traction velocities, as detailed below: The calculated correction angular velocity Combined with the current forward speed of the main variable's center of mass and the distance between traction points Decomposed into the speed of the left and right traction ropes perpendicular to the surface of the main transformer box. and : ； in, The current forward velocity of the main variable's center of mass. This refers to the distance between traction points.

8. The positioning method for replacing the main transformer of a large hydropower station using an AI-assisted attitude correction algorithm according to claim 6, characterized in that, In the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes the velocity conversion of the traction rope along the track direction and command output, as detailed below: The traction speed perpendicular to the box body and According to the current deflection angle of the principal variable's center of mass Converted to actual traction speed along the track direction and The calculation formula is as follows: ； ； The calculated and As a quantitative correction command, it is output to the traction control system via the mobile display tablet system to adjust the feed speed of the left and right traction ropes.

9. The positioning method for replacing the main transformer of a large hydropower station using an AI-assisted attitude correction algorithm according to claim 6, characterized in that, In the AI-assisted traction positioning step, the calculation of longitudinal displacement and lateral offset includes state updates and iterations, as detailed below: At a control time step Then, update the motion state of the main transformer, including: New deflection angle ; New speed of progress ; New lateral offset ; Will , and This serves as the initial state for the next control cycle, and the process returns to step S301 for iterative calculation until the centroid offset of the main transformer is reached. and the angle of deflection of the center of mass All values ​​are reduced to zero, thus completing the correction.

10. The AI-assisted attitude correction algorithm for replacing the main transformer in a large hydropower station according to claim 1, characterized in that, The end-image localization process in step S4 includes: S401, The end image positioning and monitoring detector in the end image positioning and monitoring device is deployed at the top of the high voltage bushing to collect images of the high voltage bushing and GIS conductor; The end-image positioning monitoring detector works in conjunction with the mounting frame. The mounting frame includes a cross structure composed of four telescopic rods. The monitoring detector is positioned at the center of the four telescopic rods. The telescopic rods are used to adjust the position of the end-image positioning monitoring detector and are locked by a handle to prevent directional deviation. The mounting frame also includes a tapered pin rotatably connected to the end of the telescopic rod for fixing it to the mounting hole or extension of the high-pressure bushing riser pipe. The telescopic rod has a threaded sleeve structure and is marked with graduations. It is locked by a locking bolt. The target's mounting plate is connected to the threaded holes on the GIS conductor via the four corner holes, and a weight is suspended at the center of the mounting plate and hangs down naturally. S402, the end image positioning and monitoring device is fixed to the top of the high-voltage bushing of the main transformer using its dedicated mounting bracket; by adjusting the telescopic rod and direction adjustment handle of the mounting bracket, the line of sight of the detector is basically coincident with the axis of the high-voltage bushing, and its field of view can cover the GIS conductor area in front. The target is precisely fixed at the center of the GIS conductor, and its weight structure ensures that its geometric center is aligned with the axis of the GIS conductor in a natural hanging state. The image processing host calibrates the focal length, principal point, distortion coefficient, and position and attitude relative to the orbital coordinate system of the deployed end image positioning and monitoring device detector, and establishes the transformation relationship from image pixel coordinates to actual physical space coordinates. S403, the end-point image positioning monitoring detector measures the deviation between the high-voltage bushing axis and the GIS conductor axis through image processing algorithms, as detailed below: When the main transformer enters the final connection stage, the system executes the following cyclic calculation process: Image capture: The end-point image positioning and monitoring device acquires images in real time, including the front-end features of the high-voltage bushing and the target on the GIS conductor, at a frequency of not less than 10Hz; Feature extraction and recognition: The image processing host receives the image and runs visual algorithms. Target recognition and center localization: Using the specific geometric pattern of the target, the target is accurately identified in the image by combining Hough transform or contour extraction with ellipse fitting algorithm, and its image center coordinates are calculated with sub-pixel accuracy. Bushing reference feature positioning: At the same time, the edge of the high-pressure bushing flange is identified as the reference feature, and its image center coordinates are calculated; Spatial bias calculation: Calculate the pixel deviation between the target center and the cannula reference center in the image coordinate system; Using pre-calibrated camera parameters and known actual target dimensions, pixel deviation is calculated as the actual physical deviation between the high-voltage bushing axis and the GIS conductor axis at the end mating surface perpendicular to the track plane through perspective projection model or direct linear transformation, including lateral deviation and longitudinal deviation. S404, AI-assisted decision-making and correction instruction generation: The main control equipment of the positioning device receives the actual physical deviation from the end positioning system and outputs instructions. The generated instructions are displayed through the mobile display tablet system and simultaneously sent to the traction control system for traction control.

Images