Wind power hoisting risk prediction and sudden stop control system based on edge AI

Abstract

The invention relates to the technical field of wind power hoisting, and discloses a wind power hoisting risk prediction and sudden stop control system based on edge AI, and the system comprises a multi-source fusion sensing module which is used for obtaining the environment of a hoisting site, and the multi-dimensional physical state data of hoisting equipment and fan parts in real time; the edge AI calculation and fusion early warning module is used for receiving and fusing the multi-dimensional physical state data in real time and outputting a comprehensive risk level; and the digital twinning and intelligent control module is used for executing multi-stage visual early warning or issuing an emergency stop instruction to a hoisting equipment control system according to a preset strategy. According to the invention, a high-precision integrated three-dimensional digital base map is generated through a terrain and forest region integrated scanning unit, and a forest growth model is coupled to carry out future clearance conflict prediction; digitization and predictable management of a complex construction environment are achieved, path interference and ecological damage caused by unclear environment are avoided from the source, and operation and maintenance cost and ecological risks are directly reduced.

Description

Technical Field

[0001] This invention relates to the field of wind power installation technology, specifically a wind power installation risk prediction and emergency stop control system based on edge AI. Background Technology

[0002] As a crucial component of clean energy, wind power generation is expanding into hilly, mountainous, and forested areas with complex terrain. The installation of large wind turbine components (such as towers, nacelles, and rotors) in these regions presents a combination of challenges: complex terrain, forest disturbances, variable weather, and extremely high safety risks. Traditional installation methods, relying primarily on manual experience, static design drawings, and conventional weather forecasts, have the following significant drawbacks: 1. Lagging and inaccurate environmental perception capabilities: Pre-construction topographic surveys are mostly static and discrete ground measurements, making it difficult to obtain continuous, high-precision three-dimensional topographic data for the entire site. In particular, it is impossible to accurately quantify the height, density, and dynamic growth of the forest canopy. This leads to blind spots in on-site road planning, equipment location selection, and hoisting path design. Interference between hoisting equipment or components and growing tree canopies often occurs, forcing frequent road rerouting or tree pruning, which is costly and ecologically damaging.

[0003] 2. The monitoring of the hoisting process is opaque and fragmented: Existing technologies have limited and isolated means of monitoring the key states of the crane itself (such as real-time overturning moment, outrigger ground pressure, and structural stress), often limited to basic parameters such as lifting weight and radius; for the real-time spatial attitude of the hoisted wind turbine components, there is a lack of effective and accurate monitoring methods. The safety judgment of the entire hoisting process relies heavily on the personal experience and visual observation of operators and commanders, lacking quantitative and integrated data support.

[0004] 3. Passive and delayed risk warnings: Existing safety monitoring systems mostly rely on threshold alarms, meaning they are triggered only after parameters (such as wind speed) exceed fixed limits, making them "post-event" alarms. Complex, multi-factor coupled risks (such as the superposition of sudden gusts of wind and large-angle crane rotation) cannot be predicted in advance. Furthermore, data processing typically involves uploading to remote servers, resulting in decision-making delays of up to seconds, which cannot meet the stringent requirements of millisecond-level emergency stop responses during hoisting operations.

[0005] 4. The accident analysis methods are rudimentary, making it difficult to accumulate experience: Once a dangerous situation or accident occurs, investigations can only rely on blurry video recordings and scattered records, making it difficult to accurately trace the state and interaction processes of various systems at the moment of the accident, thus hindering root cause analysis. Valuable lessons cannot be effectively transformed into "knowledge" for optimizing operating procedures, training content, and system algorithms, leading to recurring similar problems. Summary of the Invention

[0006] The purpose of this invention is to provide a wind power installation risk prediction and emergency stop control system based on edge AI to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The wind turbine installation risk prediction and emergency stop control system based on edge AI includes: The multi-source fusion sensing module is used to acquire multi-dimensional physical state data of the environment, hoisting equipment, and wind turbine components at the hoisting site in real time; The edge AI computing and fusion early warning module is used to receive and fuse multi-dimensional physical state data in real time, and perform low-latency calculations through the built-in risk prediction model to output a comprehensive risk level. The digital twin and intelligent control module is used to construct a digital twin scenario synchronized with the physical hoisting process, receive the comprehensive risk level, and execute multi-level visual early warnings or issue emergency stop commands to the hoisting equipment control system according to preset strategies.

[0008] As a further embodiment of the present invention: the multi-source fusion sensing module includes: The integrated topographic and forest area scanning unit generates a three-dimensional digital base map containing topography and canopy height based on UAV lidar point cloud, multispectral imagery, real-time dynamic measurement, and global navigation satellite system. The hoisting equipment status monitoring matrix is ​​used to monitor the operating status of hoisting equipment in real time; it includes force sensors, inclinometers, pressure gauges, and anemometers deployed at key parts of the hoist. The wind turbine component attitude detection matrix includes: an inertial measurement unit for real-time monitoring of the motion status of key wind turbine components, and a laser rangefinder for measuring the gap between mating surfaces.

[0009] As a further aspect of the present invention: the edge AI computing and fusion early warning module includes: Heterogeneous data protocol gateway, used to uniformly access sensor data from different sources; Edge AI inference servers are used to run lightweight risk prediction models and calculate dynamic risk levels based on preset security constraints; The hoisting window scheduling unit optimizes the time window for future hoisting operations based on genetic algorithms and micro-weather forecasts.

[0010] As a further aspect of the present invention: the digital twin and intelligent control module includes: The real-time twin engine uses a 3D digital base map generated by an integrated terrain and forest area scanning unit, as well as real-time monitoring data from the hoisting equipment status monitoring matrix and the wind turbine component attitude detection matrix to drive a virtual hoisting scene. The warning and emergency stop logic unit provides warnings in different ways based on different risk levels; The accident simulation and backtracking unit is used to reproduce the hoisting process based on historical data and to conduct accident analysis.

[0011] As a further aspect of the present invention, the specific method for generating a three-dimensional digital base map in the integrated topographic and forest area scanning unit is as follows: S11. Plan the flight path of the UAV based on the mission area and deploy a real-time dynamic measurement and global navigation satellite system ground reference network; then use the UAV system equipped with lidar, multispectral camera, airborne inertial measurement unit and global navigation satellite system to collect data synchronously along the set route and obtain laser point cloud data and multispectral image data with precise spatiotemporal labels. S12. Process and classify the laser point cloud data, separate the ground point cloud and the vegetation point cloud, generate a digital terrain model based on the ground point cloud, and generate a digital surface model based on all point clouds including the vegetation point cloud. S13. Calculate the elevation difference between the digital surface model and the digital terrain model to obtain the canopy height layer information, so as to form an integrated three-dimensional digital base map that combines the terrain and canopy height. S14. Assign a time-dimensional growth prediction model to the canopy height information in the integrated three-dimensional digital base map; and simulate the canopy height distribution after a specified time period in the future to generate a predictive digital surface model. S15. Perform clearance calculation and conflict detection on the predictive digital surface model and the preset overhead line design elevation or the passage envelope space of hoisting equipment to identify current and future spatial conflict risk points. S16. Based on the integrated 3D digital base map and conflict detection results, optimize the site selection and foundation treatment scheme for crawler cranes, and plan the entry and operation paths of lifting equipment.

[0012] As a further aspect of the present invention: the method for calculating dynamic risk levels in an edge AI inference server is as follows: S21. Acquire and synchronize multi-source sensing data in real time through the protocol gateway, and construct a standardized feature vector for the current moment; S22. Input the feature vector into the lightweight risk prediction model to obtain a multidimensional risk assessment vector; S23. Compare the probability values ​​of each item in the multidimensional risk assessment vector with the preset dynamic safety constraints to determine the current dynamic comprehensive risk level; S24. Based on the dynamic comprehensive risk level, output the corresponding early warning signal or control command.

[0013] As a further aspect of the present invention: the method for scheduling hoisting operation time windows in the hoisting window scheduling unit is as follows: S31. Obtain micro-meteorological forecast data, site environmental status data, and a list of hoisting tasks to be executed within the future predetermined period; S32. Based on predetermined dynamic safety constraints, match and model the micro-weather forecast data, site environmental status data and hoisting task list to construct an optimization model with the goal of maximizing the safe operation window. S33. Use a genetic algorithm to solve the optimization model and generate a scheduling scheme for future hoisting operation time windows that meets safety constraints; S34. Output the hoisting operation time window scheduling plan, and trigger corresponding early warning information based on the proximity of the current time to the operation window in the plan.

[0014] As a further aspect of the present invention: in the accident simulation and retrospective unit, the method for reproducing the hoisting process and analyzing the accident is as follows: S41. Synchronously record and store multi-source time-series data throughout the entire hoisting operation process; S42. Upon receiving the backtracking instruction, based on the selected time range, the stored multi-source time-series data is retrieved to drive the digital twin model to perform high-fidelity scene reconstruction and playback. S43. Based on the reconstructed scenario, use analysis tools to conduct interactive analysis of the events in order to locate the root cause; S44. Based on the analysis results, automatically generate structured analysis reports and improvement suggestions.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention generates a high-precision integrated three-dimensional digital base map through an integrated topographic and forest area scanning unit, and couples it with a forest growth model to predict future airspace conflicts. It solves the drawbacks of traditional static surveying, realizes digital and predictable management of complex construction environments, avoids path interference and ecological damage caused by unclear environment from the source, and transforms passive periodic pruning and relocation into precise preventive maintenance, directly reducing operation and maintenance costs and ecological risks. By deploying edge AI servers at the hoisting site and running lightweight risk prediction models, multi-source data can be fused, analyzed, and output dynamic risk levels in a short time. This enables the prediction of complex risks such as overturning and collisions in the next few seconds and the execution of tiered control from prompts and warnings to automatic emergency stops. This significantly shifts the safety defense line from post-event alarms to pre-event prediction and instantaneous blocking during the event, providing core technical support for achieving the goal of zero major accidents. By integrating micro-weather forecasting with genetic algorithm-based scheduling optimization, the system can recommend globally optimal safe operating time windows for hoisting tasks in the coming days. This significantly improves the utilization efficiency of equipment and personnel, reduces blind waiting and risky operations, and directly contributes to the economic goals of shortening the single-machine hoisting cycle and reducing the overall project duration. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a wind power installation risk prediction and emergency stop control system based on edge AI. Figure 2 This is a flowchart illustrating a wind turbine installation risk prediction and emergency stop control method based on edge AI. Detailed Implementation

[0017] Please see Figure 1 In this embodiment of the invention, the wind turbine installation risk prediction and emergency stop control system based on edge AI includes: The multi-source fusion sensing module is used to acquire multi-dimensional physical state data of the hoisting site environment, hoisting equipment, and wind turbine components in real time; specifically, it includes: The integrated topographic and forest area scanning unit generates a 3D digital base map containing topography and canopy height based on UAV LiDAR point clouds, multispectral imagery, and real-time dynamic measurement and Global Navigation Satellite System (RTK-GNSS). The specific method for generating the 3D digital base map is as follows: S11. Plan the UAV flight path based on the mission area and deploy a real-time dynamic measurement and Global Navigation Satellite System (RTK-GNSS) ground reference network; then use the UAV system equipped with LiDAR, multispectral camera, and airborne inertial measurement unit and Global Navigation Satellite System (GNSS / IMU) to synchronously collect data along the set path, and obtain laser point cloud data and multispectral image data with precise spatiotemporal labels; for example, plan the UAV flight path for the roads and all camera positions in the field to ensure that the LiDAR point cloud density is ≥200 points / square meter and the ground resolution of the multispectral image is ≤10cm.

[0018] At least three RTK-GNSS ground reference stations are evenly deployed within the survey area and connected to the national coordinate system to provide a unified centimeter-level absolute coordinate reference for all data.

[0019] For the eucalyptus forests on both sides of the road, the flight design must ensure that LiDAR can effectively penetrate the canopy gaps to obtain understory topographic data.

[0020] S12. Process and classify the laser point cloud data, separating the ground point cloud and the vegetation point cloud. Vegetation points include eucalyptus tree canopies, and ground points include cranes and towers. Then, generate a digital terrain model (DTM) based on the ground point cloud, and generate a digital surface model (DSM) based on all point clouds, including the vegetation point cloud.

[0021] S13. Calculate the elevation difference between the digital surface model (DSM) and the digital terrain model (DTM) to obtain the canopy height layer information, so as to form an integrated three-dimensional digital base map that combines the terrain and canopy height.

[0022] S14. Assign a time-dimensional growth prediction model to the canopy height information in the integrated 3D digital base map; and simulate the canopy height distribution after a specified time period in the future to generate a predictive digital surface model (p-DSM).

[0023] S15. Perform clearance calculations and conflict detection between the predictive digital surface model (p-DSM) and the preset design elevation of overhead lines or the passage envelope of hoisted equipment, identifying current and future spatial conflict risk points; for example: Clearance calculation: Assuming that the clearance between the tree canopy and the 110kV transmission line needs to be greater than 7 meters, and the tree height at a certain location is currently 15m, the simulation shows that it will be 20m after 1 year and 40m after 5 years; the clearance with the line is calculated again; thus accurately predicting when and where insufficient clearance will occur. Conflict detection: For planned hoisting paths (especially the 150m boom slewing range of the main hoist), during simulated hoisting, the expected travel envelope of the wind turbine components (such as the 117t impeller assembly) is spatially calculated on the 3D base map and collided with the predicted DSM at a future point in time to identify hoisting interference points that may be caused by canopy growth in advance.

[0024] For example, identify current and future spatial conflict risk points as follows: urgent conflict points that must be pruned now; preventative locations that need pruning in the next N years; and road sections where existing on-site roads or hoisting paths may need to be adjusted due to canopy growth. This will transform the blind pruning or passive rerouting every 3-4 years into proactive management that is "on demand, precise, and preventative," directly avoiding annual ineffective pruning costs and secondary risks.

[0025] S16. Based on the integrated 3D digital base map and conflict detection results, optimize the site selection and foundation treatment scheme for crawler cranes, and plan the entry and operation paths of lifting equipment.

[0026] The hoisting equipment status monitoring matrix is ​​used to monitor the operating status of hoisting equipment in real time; it includes force sensors, inclinometers, pressure gauges and anemometers deployed in key parts of the hoist (such as hooks, main beams, booms, hydraulic systems and pneumatic circuit systems); For example, a pin-type force sensor can be used to measure the load on the hook and the force at key structural force transmission points. For example, the spatial attitude angles of the boom (such as pitch angle and slewing angle) can be measured by a fiber optic inclinometer. For example, hydraulic system pressure gauges can be used to monitor outrigger reaction force, mechanism load, and system pressure status. For example, high-precision anemometers can be used to monitor instantaneous wind speed, gust wind speed, and wind direction data. The wind turbine component attitude detection matrix includes: an inertial measurement unit for real-time monitoring of the motion status of key wind turbine components (such as tower, blades, and nacelle), such as monitoring the torsion angle and flapping angle of the blades during hoisting.

[0027] And laser rangefinders used to measure the gap between mating surfaces; for example, to measure in real time the gap and parallelism between the two mating flanges between towers and between the tower and the nacelle.

[0028] The edge AI computing and fusion early warning module, deployed at the hoisting site, receives and fuses multi-dimensional physical state data in real time, performs low-latency calculations using a built-in risk prediction model, and outputs a comprehensive risk level. Specifically, it includes: Heterogeneous data protocol gateway, used to uniformly access sensor data from different sources; for example, it supports OPC-UA, MQTT, and ROS2 protocols.

[0029] An edge AI inference server is used to run a lightweight risk prediction model and calculate a dynamic risk level based on preset security constraints; the method for calculating the dynamic risk level is as follows: S21. Acquire and synchronize multi-source sensing data in real time through the protocol gateway, such as crane data from OPC-UA, sensor data from MQTT, and component attitude data from ROS2. The multi-source sensing data includes: real-time wind speed and overturning moment of the hoisting equipment, real-time attitude angles of the wind turbine components, and the ground settlement rate of the hoisting area; and construct a standardized feature vector for the current moment; such as standardized feature vector, in, , These are instantaneous and gust wind speeds, respectively. For wind direction, , These are the lifting moment and the calculated wind load moment (based on the actual lifting weight). Pitch angle Gust wind speed (real-time calculation) The roll angle of the blade. , These are the maximum gap value and maximum deviation value of the laser rangefinder, respectively. This represents the instantaneous settlement rate of the foundation settlement monitoring instrument.

[0030] S22. Input the feature vector into the lightweight risk prediction model to obtain a multi-dimensional risk assessment vector; wherein, the multi-dimensional risk assessment vector includes: overturning probability, collision probability, docking failure probability, and comprehensive risk level; for example, the multi-dimensional risk assessment vector ;in, The predicted probability of overturning within a future time period (e.g., 10 seconds) is 0 to 1. , is the predicted probability of a collision (between parts or with the tree canopy) occurring within a future time period (e.g., 10 seconds); This is a predicted probability of docking failure (based on attitude and gap trends). The risk level is determined by a comprehensive risk rating (e.g., Level 0 - Green, Level 1 - Yellow, Level 2 - Orange, Level 3 - Red). The risk prediction model is a pruned and quantized deep learning model or gradient boosting tree model, and its inference process is completed on the edge AI accelerator card.

[0031] S23. Compare the probability values ​​of each item in the multidimensional risk assessment vector with the preset dynamic safety constraints to determine the current dynamic comprehensive risk level; wherein, the dynamic safety constraints include: wind speed threshold, gust threshold, overturning moment ratio threshold, component attitude angle threshold, flange gap deviation threshold, and foundation settlement rate threshold. The dynamic comprehensive risk level classification is as follows: Level 0 - Green, Level 1 - Yellow, Level 2 - Orange, Level 3 - Red; among them, Level 3 executes an emergency stop decision, Level 2 executes an early warning decision, Level 1 executes a notification decision, and Level 0 executes no decision; the specific execution logic is as follows: Red Emergency Stop Decision Logic: If the collision probability is greater than the first probability threshold, or the flange gap deviation is greater than the first deviation threshold (e.g., 2mm) and the wind turbine components are in a preset docking proximity space, or the foundation settlement rate is greater than the first rate threshold (e.g., 3mm / h), then the risk level is determined to be the highest level, and an emergency stop command is triggered; that is, The component is in the final meter of docking. ; Orange alert decision-making logic:

[0032] Yellow alert decision-making logic: ; Green normal decision-making logic: No action.

[0033] S24. Based on the dynamic comprehensive risk level, output the corresponding early warning signal or control command; Assuming the risk level is the highest, an emergency stop command will be sent to the crane control system within a delay of less than 200 milliseconds; for example, (Red): Immediately (<50ms) send an “emergency stop” digital signal to the crane via a hardwired or high-priority real-time network to interrupt the current operation.

[0034] Assuming the risk level is secondary high, an early warning message containing the specific reasons for the risk will be sent to the digital twin platform and the on-site command terminal; for example, (Orange): The command twin platform highlights the alarm and pushes detailed warning information to the commander's APP ("Gusts exceed limits, torque rate 92%").

[0035] The hoisting window scheduling unit optimizes future hoisting operation time windows based on genetic algorithms and micro-weather forecasts; the method for scheduling hoisting operation time windows is as follows: S31. Obtain micro-meteorological forecast data, site environmental status data, and a list of hoisting tasks to be executed within the predetermined future period; among which: Micro-weather forecast data includes: wind speed, gusts, wind direction, temperature, probability of precipitation, and visibility; Site environmental status data includes: canopy height data at specific future time points based on historical scanning data and forest growth models, used to determine the clearance conditions of the hoisting path; The hoisting task list includes: task type (tower, impeller, etc.), required hoisting time (e.g., hoisting the impeller requires 4 hours of continuous operation), priority, and dependencies (e.g., the nacelle can only be hoisted after the third tower section is hoisted).

[0036] S32. Based on predetermined dynamic safety constraints (e.g., wind speed ≤ 8m / s, gusts ≤ 10m / s, no thunderstorms, visibility > 1km), match and model the micro-meteorological forecast data, site environmental status data, and hoisting task list to construct an optimization model with the goal of maximizing the safe operation window; for example, the main objective of the optimization model is to maximize the "total safe operation time window" of all hoisting tasks in the future planning period (e.g., the next 5 days).

[0037] Secondary objective (multi-objective optimization): (1) Minimize the total project duration: Prioritize tasks on the critical path; (2) Minimize equipment turnover: continuously schedule tasks at the same machine position; (3) Minimize ecological disturbance: High-risk operations should be carried out during periods when the forest growth simulation shows sufficient airspace to avoid triggering pruning.

[0038] S33. Use a genetic algorithm to solve the optimization model and generate a future hoisting operation time window scheduling scheme that meets the safety constraints. Assume that a hilly wind farm needs to complete the hoisting of two turbine positions, F12 and F13, within the next 3 days (each turbine position includes: 3 sections of tower, nacelle, and rotor, for a total of 5 tasks).

[0039] The genetic algorithm solves the optimization model to obtain a high-scoring solution: By concentrating all manpower and equipment, all five hoisting tasks at position F12 were completed consecutively within the ideal 15-hour window on the second day (a total of 20 hours, but some tasks could overlap or be closely linked). Although the tasks at position F12 slightly exceeded the 15-hour window (meaning the last small task might be completed with a slight risk of exceeding the limit), it ensured that the tasks at position F13 could be safely scheduled for the next weather cycle, maximizing overall safety and efficiency.

[0040] S34. Output the hoisting operation time window scheduling plan, and trigger corresponding early warning information based on the proximity of the current time to the operation window in the plan; For example, the command center's large screen displays the scheduling plan and marks it as follows: "Day 2, 10:00 AM to 2:00 AM the following day, focus on tackling the F12 turret position. Warning: The final section of the tower hoisting may face gusts approaching the threshold; please prepare for emergencies."

[0041] The digital twin and intelligent control module is used to construct a digital twin scenario synchronized with the physical hoisting process, receive comprehensive risk levels, and execute multi-level visual early warnings or issue emergency stop commands to the hoisting equipment control system according to preset strategies. Specifically, it includes: The real-time twin engine uses a 3D digital base map generated by an integrated terrain and forest area scanning unit, as well as real-time monitoring data from the hoisting equipment status monitoring matrix and the wind turbine component attitude detection matrix to drive a virtual hoisting scene. Assuming that during the smooth lifting of the 117t impeller, the main hook pulley block pin, as monitored by the lifting equipment status monitoring matrix, experiences a force of 1147 kN (117000 kg × 9.8 m / s²); the main tilt angle of the boom is 78°; the pressure on the four outriggers [P1,P2,P3,P4] = [29,28,31,29] MPa; and the gust of wind is 13 m / s.

[0042] Suppose that when hoisting a 55-meter-long blade, the data monitored by the inertial measurement unit shows that it has twisted by 1.2° around its own axis; the system will determine whether this twist is within the allowable range (e.g., <1.5°). If it exceeds the limit, it will prompt that it needs to be corrected by adjusting the slings or waiting for the wind to stabilize, in order to prevent collision with the installed components or tower.

[0043] The warning and emergency stop logic unit issues warnings in different ways according to different risk levels; for example, risk levels include: level 0 - green, level 1 - yellow, level 2 - orange, and level 3 - red. When the risk level is green, there is no risk and no action is required. When the risk level is yellow, an alert will be displayed on the virtual machine interface; When the risk level is orange, send an audible and visual alarm to the on-site terminal; When the risk level is red, an emergency stop signal is automatically sent to the crane controller within 200ms. Assume that 4 to 8 laser rangefinders are evenly distributed around the circumference of the flange at the top of the lower tower section; during the hoisting of the upper tower section, the laser rangefinders continuously emit laser beams upward to measure the distance to the bottom surface of the flange of the upper tower section.

[0044] If the measured readings are: [Position 1: 12.3mm, Position 2: 11.8mm, Position 3: 14.5mm, Position 4: 10.1mm].

[0045] Therefore, the calculation yields: Average gap: (12.3+11.8+14.5+10.1) / 4 = 12.175mm, which meets the macroscopic gap requirements for bolt insertion (e.g., 10-20mm). Parallelism: The difference between the maximum and minimum values ​​is 14.5 - 10.1 = 4.4 mm; this value far exceeds the user-set alarm threshold of 2 mm. The system determines that although the average clearance is appropriate, there is a serious non-parallelism or center misalignment; if the component is forcibly lowered under these conditions, it will cause localized flange collision, misalignment of bolt holes, and thread seizure. The system immediately issued an audible and visual alarm: "Flange is seriously non-parallel, deviation 4.4mm! Stop descent and perform horizontal correction!" and highlighted the direction and magnitude of the offset on the digital twin interface.

[0046] The accident simulation and backtracking unit is used to reproduce the hoisting process based on historical data and perform accident analysis. The methods for reproducing the hoisting process and analyzing accidents are as follows: S41. Synchronously record and store multi-source time-series data throughout the entire hoisting operation process, including raw sensor data, intermediate risk prediction data, control command data, and environmental scenario data.

[0047] S42. Upon receiving the backtracking instruction, based on the selected time range, the stored multi-source time-series data is retrieved to drive the digital twin model to perform high-fidelity scene reconstruction and playback.

[0048] S43. Based on the reconstructed scenario, use analysis tools to conduct interactive analysis of the events to locate the root cause; the analysis tools include: Mechanical causation analysis tools are used to extract and compare time-series curves of key mechanical parameters related to the event's mechanism; For example, a comparison reveals that the wind load moment is [value missing] 2 seconds before the emergency stop is triggered. A sudden gust of wind increased the torque by 70%, contributing 70% of the total torque increase, while the lifting torque... The changes were gradual; this directly points to the sudden wind conditions as the main cause, rather than improper operation. A time-series causal chain analysis tool used to align and visualize the causal evolution of multiple key parameters during an event process; For example, gusts of wind Blade roll angle Collision probability Four curves are generated from the four key parameters of risk level, and they are aligned vertically to analyze the causal evolution relationship. Assume a simulation analysis tool is used to modify variables in historical data within a digital twin model and to simulate and extrapolate the results of the modified scenario. The variables in the modified historical data include: crane boom elevation angle, load weight, wind speed and direction, and the 3D model of temporarily added auxiliary tooling. For example, in the case of an emergency stop caused by excessive wind speed, investigators could ask: "What would have happened if the crane boom angle had been lowered by 5 degrees?" Simulation results show that the torque rate will decrease by 8%, and the risk level may only rise to an orange warning rather than a red emergency stop.

[0049] S44. Based on the analysis results, automatically generate a structured analysis report and improvement suggestions, which include: an event overview, a data-based timeline, root cause conclusions, an assessment of each responsible link, and specific improvement suggestions for equipment, operation, or management processes.

[0050] Please see Figure 2 In this embodiment of the invention, the wind power installation risk prediction and emergency stop control method based on edge AI includes the following steps: S1. Construct a unified coordinate system three-dimensional digital base map of the hoisting site and deploy a multi-source sensor network; S2. Sensor data, meteorological data, and base map data are aggregated and fused in real time through edge AI servers; S3. Use a risk prediction model to perform real-time calculations and assess the current and future short-term hoisting comprehensive risk level; S4. Synchronously map physical state and risk level in the digital twin platform; S5. Implement corresponding early warning strategies based on risk levels, and automatically trigger emergency stop control of hoisting equipment when the risk level reaches the highest level.

[0051] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A wind turbine installation risk prediction and emergency stop control system based on edge AI, characterized in that, include: The multi-source fusion sensing module is used to acquire multi-dimensional physical state data of the environment, hoisting equipment, and wind turbine components at the hoisting site in real time; The edge AI computing and fusion early warning module is used to receive and fuse multi-dimensional physical state data in real time, and perform low-latency calculations through the built-in risk prediction model to output a comprehensive risk level. The digital twin and intelligent control module is used to construct a digital twin scenario synchronized with the physical hoisting process, receive the comprehensive risk level, and execute multi-level visual early warnings or issue emergency stop commands to the hoisting equipment control system according to preset strategies.

2. The wind power installation risk prediction and emergency stop control system based on edge AI according to claim 1, characterized in that, The multi-source fusion sensing module includes: The integrated topographic and forest area scanning unit generates a three-dimensional digital base map containing topography and canopy height based on UAV lidar point cloud, multispectral imagery, real-time dynamic measurement, and global navigation satellite system. The hoisting equipment status monitoring matrix is ​​used to monitor the operating status of hoisting equipment in real time; it includes force sensors, inclinometers, pressure gauges, and anemometers deployed at key parts of the hoist. The wind turbine component attitude detection matrix includes: an inertial measurement unit for real-time monitoring of the motion status of key wind turbine components, and a laser rangefinder for measuring the gap between mating surfaces.

3. The wind power installation risk prediction and emergency stop control system based on edge AI according to claim 1, characterized in that, The edge AI computing and fusion early warning module includes: Heterogeneous data protocol gateway, used to uniformly access sensor data from different sources; Edge AI inference servers are used to run lightweight risk prediction models and calculate dynamic risk levels based on preset security constraints; The hoisting window scheduling unit optimizes the time window for future hoisting operations based on genetic algorithms and micro-weather forecasts.

4. The wind power installation risk prediction and emergency stop control system based on edge AI according to claim 1, characterized in that, The digital twin and intelligent control module includes: The real-time twin engine uses a 3D digital base map generated by an integrated terrain and forest area scanning unit, as well as real-time monitoring data from the hoisting equipment status monitoring matrix and the wind turbine component attitude detection matrix to drive a virtual hoisting scene. The warning and emergency stop logic unit provides warnings in different ways based on different risk levels; The accident simulation and backtracking unit is used to reproduce the hoisting process based on historical data and to conduct accident analysis.

5. The wind turbine hoisting risk prediction and emergency stop control system based on edge AI according to claim 2, characterized in that, The specific method for generating a 3D digital base map in the integrated topographic and forest area scanning unit is as follows: S11. Plan the flight path of the UAV based on the mission area and deploy a real-time dynamic measurement and global navigation satellite system ground reference network; then use the UAV system equipped with lidar, multispectral camera, airborne inertial measurement unit and global navigation satellite system to collect data synchronously along the set route and obtain laser point cloud data and multispectral image data with precise spatiotemporal labels. S12. Process and classify the laser point cloud data, separate the ground point cloud and the vegetation point cloud, generate a digital terrain model based on the ground point cloud, and generate a digital surface model based on all point clouds including the vegetation point cloud. S13. Calculate the elevation difference between the digital surface model and the digital terrain model to obtain the canopy height layer information, so as to form an integrated three-dimensional digital base map that combines the terrain and canopy height. S14. Assign a time-dimensional growth prediction model to the canopy height information in the integrated three-dimensional digital base map; and simulate the canopy height distribution after a specified time period in the future to generate a predictive digital surface model. S15. Perform clearance calculation and conflict detection on the predictive digital surface model and the preset overhead line design elevation or the passage envelope space of hoisting equipment to identify current and future spatial conflict risk points. S16. Based on the integrated 3D digital base map and conflict detection results, optimize the site selection and foundation treatment scheme for crawler cranes, and plan the entry and operation paths of lifting equipment.

6. The wind turbine hoisting risk prediction and emergency stop control system based on edge AI according to claim 3, characterized in that, The method for calculating dynamic risk levels in edge AI inference servers is as follows: S21. Acquire and synchronize multi-source sensing data in real time through the protocol gateway, and construct a standardized feature vector for the current moment; S22. Input the feature vector into the lightweight risk prediction model to obtain a multidimensional risk assessment vector; S23. Compare the probability values ​​of each item in the multidimensional risk assessment vector with the preset dynamic safety constraints to determine the current dynamic comprehensive risk level; S24. Based on the dynamic comprehensive risk level, output the corresponding early warning signal or control command.

7. The wind turbine hoisting risk prediction and emergency stop control system based on edge AI according to claim 3, characterized in that, The method for scheduling hoisting operation time windows in the hoisting window scheduling unit is as follows: S31. Obtain micro-meteorological forecast data, site environmental status data, and a list of hoisting tasks to be executed within the future predetermined period; S32. Based on predetermined dynamic safety constraints, match and model the micro-weather forecast data, site environmental status data and hoisting task list to construct an optimization model with the goal of maximizing the safe operation window. S33. Use a genetic algorithm to solve the optimization model and generate a scheduling scheme for future hoisting operation time windows that meets safety constraints; S34. Output the hoisting operation time window scheduling plan, and trigger corresponding early warning information based on the proximity of the current time to the operation window in the plan.

8. The wind power installation risk prediction and emergency stop control system based on edge AI according to claim 1, characterized in that, In the accident simulation and retrospective unit, the methods for reproducing the hoisting process and analyzing the accident are as follows: S41. Synchronously record and store multi-source time-series data throughout the entire hoisting operation process; S42. Upon receiving the backtracking instruction, based on the selected time range, the stored multi-source time-series data is retrieved to drive the digital twin model to perform high-fidelity scene reconstruction and playback. S43. Based on the reconstructed scenario, use analysis tools to conduct interactive analysis of the events in order to locate the root cause; S44. Based on the analysis results, automatically generate structured analysis reports and improvement suggestions.

Images