Autonomous inspection system of offshore wind plant operation and maintenance unmanned aerial vehicle

Abstract

The invention discloses an autonomous inspection system of an offshore wind plant operation and maintenance unmanned aerial vehicle, and relates to the technical field of inspection operation and maintenance. An autonomous inspection system of an offshore wind plant operation and maintenance unmanned aerial vehicle comprises an autonomous inspection preparation module and an autonomous inspection adjustment module. According to the method, historical operation and maintenance data and meteorological risk factors are fused, the routing inspection priority of each wind turbine generator set is scientifically evaluated, optimal path planning is realized through a digital twinning and simulated annealing algorithm, and the path rationality and routing inspection efficiency are improved; by means of a knowledge graph and a hypergraph neural network, a path risk avoiding strategy can be optimized in real time, and safe execution of an inspection task under severe sea conditions is guaranteed; task allocation is dynamically adjusted according to the unmanned aerial vehicle cluster state, task merging and path sharing are achieved, loads are effectively balanced, energy consumption is reduced, repeated flight is avoided, and the overall cooperation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of patrol inspection and operation and maintenance technology, and in particular to an autonomous patrol inspection system for offshore wind farm operation and maintenance drones. Background Art

[0002] As the scale of offshore wind farms continues to expand and the number of wind turbines increases dramatically, traditional inspection methods that rely on manual or semi-autonomous drones can no longer meet the operation and maintenance needs in high-frequency, high-risk offshore environments. Existing technologies generally have problems such as low path planning efficiency, uneven task coverage, insufficient environmental perception, and lack of dynamic adjustment capabilities. As a result, it is difficult for operation and maintenance drones to achieve efficient, comprehensive, and safe autonomous inspections under complex weather conditions.

[0003] Therefore, it is necessary to combine multi-source heterogeneous data fusion analysis to improve the perception and adaptability of operation and maintenance drones, and ensure that all wind turbines in offshore wind farms are inspected safely, efficiently and intelligently in one mission. Summary of the Invention

[0004] The present invention aims to provide an autonomous inspection system for an offshore wind farm operation and maintenance drone, thereby improving the perception and adaptability of the operation and maintenance drone.

[0005] An autonomous inspection system for offshore wind farm operation and maintenance drones, including: The autonomous inspection preparation module includes a risk investigation unit and an inspection preparation unit; the risk investigation unit is used to obtain meteorological data of offshore wind farms and radar point cloud data of offshore wind farms; risk prediction is performed based on the meteorological data of offshore wind farms and radar point cloud data of offshore wind farms to obtain the offshore wind farm inspection risk feature set and the corresponding offshore wind farm environmental situation map; it is assumed that the offshore wind farm contains M wind turbines D m , m=1, 2, ..., M; wind turbine D is marked in the offshore wind farm environmental situation map m location; The inspection preparation unit is used to set a total of N operation and maintenance drones J n At the same time, inspection work is carried out, n=1, 2, ..., N; operation and maintenance drone J n The corresponding power state is E n ; Get wind turbine D m Historical operation and maintenance records of P m ; Based on historical operation and maintenance records P m and offshore wind farm environmental situation map to determine wind turbine D m Inspection and maintenance priority X m ; Based on inspection and maintenance priority X m and offshore wind farm environmental situation map for autonomous inspection planning, providing all operation and maintenance drones nDistribute the initial autonomous inspection path L n ; The autonomous inspection adjustment module comprises an autonomous inspection unit; the autonomous inspection unit is configured to obtain dynamic variable parameters in real time during autonomous inspection, and update the autonomous inspection path L n in real time according to the dynamic variable parameters n to obtain a new autonomous inspection path L n until the autonomous inspection is completed.

[0006] As a preferred technical solution of the present application, the specific steps of risk prediction based on offshore wind farm meteorological data and offshore wind farm radar point cloud data include: The offshore wind farm meteorological data and the offshore wind farm radar point cloud data are enhanced by using a knowledge graph to obtain a fusion offshore wind farm gas phase feature tensor; an offshore wind farm meteorological risk correlation model is constructed based on the fusion offshore wind farm gas phase feature tensor; a pre-trained hypergraph neural network is used to analyze the offshore wind farm meteorological risk correlation model to obtain a dynamic meteorological risk factor graph and a meteorological risk propagation probability matrix; The offshore wind farm radar point cloud data is updated based on the dynamic meteorological risk factor graph and the meteorological risk propagation probability matrix to obtain an offshore wind farm environment situation map; the positions of the wind turbine D m are marked in the offshore wind farm environment situation map, and the corresponding meteorological risk prediction factor C m of the wind turbine D m is output. The dynamic meteorological risk factor graph and the meteorological risk propagation probability matrix are used for feature output to form an offshore wind farm inspection risk feature set.

[0007] As a preferred technical solution of the present application, the specific steps of judging the inspection and operation priority X m of the wind turbine D m based on the historical operation record P m and the offshore wind farm environment situation map include: The historical operation record P m contains the historical failure times, the latest failure maintenance type, and the latest failure downtime of the wind turbine D m ; The meteorological risk prediction factor C m corresponding to the wind turbine D m in the offshore wind farm environment situation map is extracted, and a wind turbine multi-dimensional evaluation matrix Z m is constructed based on the historical operation record P m and the meteorological risk prediction factor C m . A risk index decay function is established to the historical operation record P mMatching risk attenuation factor F m ; Based on the multi-dimensional evaluation matrix Z of wind turbine m Performing risk semantic analysis, outputting the multi-dimensional feature vector Z of wind turbine operation and maintenance m ’; Based on the multi-dimensional feature vector Z of wind turbine operation and maintenance m ’、risk attenuation factor F m and wind turbine operation and maintenance dynamic game model, priority matching is performed, and the inspection and maintenance priority X is output m .

[0008] As a preferred technical solution of the present application, based on the inspection and maintenance priority X m and the offshore wind farm environment situation map, the specific steps of autonomous inspection planning include: Based on the digital twin analysis method and the simulated annealing algorithm, a plurality of initial path planning individuals are constructed for the inspection and maintenance priority X m and the offshore wind farm environment situation map; an initial path planning population is constructed based on the plurality of initial path planning individuals; The target function is to maximize the total inspection priority coverage value, minimize the total path risk cost, and maximize the autonomous inspection efficiency, and the optimal initial path planning individual is selected from the plurality of initial path planning individuals; Based on the optimal initial path planning individual, an initial autonomous inspection path L n is assigned to all operation and maintenance unmanned aerial vehicles J n .

[0009] As a preferred technical solution of the present application, in the autonomous inspection process, the inspection dynamic variable parameters are obtained in real time; wherein the inspection dynamic variable parameters of the operation and maintenance unmanned aerial vehicle J n include: self-state parameters, external meteorological parameters, inspected wind turbine state parameters, and cluster cooperation state parameters; for the operation and maintenance unmanned aerial vehicle J n , the operation and maintenance unmanned aerial vehicle adjacent to the operation and maintenance unmanned aerial vehicle J n is set as L nk .

[0010] As a preferred technical solution of the present application, the specific steps of updating the autonomous inspection path L n of the operation and maintenance unmanned aerial vehicle J n in real time include: Every Δt judges whether the inspection dynamic variable parameters of the operation and maintenance unmanned aerial vehicle J n meet the path change standard; if the operation and maintenance unmanned aerial vehicle J n meets the path change standard, a candidate path set is generated based on the operation and maintenance unmanned aerial vehicle J n ; The operation and maintenance unmanned aerial vehicle Lnk Can the operation and maintenance drone be merged? n If the result is yes, the operation and maintenance drone L is selected based on the candidate path set. nk Update autonomous inspection paths in real time, and update operation and maintenance drones in real time n Autonomous inspection path L n ; If the result is no, then the operation and maintenance drone J is selected based on the candidate path set. n Real-time update of autonomous inspection path L n .

[0011] As a preferred technical solution of the present invention, one autonomous inspection means that all wind turbines in the offshore wind farm are inspected at least once, and each wind turbine is inspected by at least one operation and maintenance drone.

[0012] The present invention has the following advantages: 1. The present invention ensures that each wind turbine in a wind farm is inspected by at least one drone through autonomous inspection, significantly improving the operation and maintenance integrity and risk monitoring accuracy of offshore wind power equipment; integrating historical operation and maintenance data with meteorological risk factors, scientifically assessing the inspection priority of each wind turbine, and realizing optimal path planning through digital twin and simulated annealing algorithms, thereby improving path rationality and inspection efficiency; with the help of knowledge graphs and hypergraph neural networks, path risk avoidance strategies can be optimized in real time to ensure the safe execution of inspection tasks in severe sea conditions; dynamically adjusting task allocation according to the status of the drone cluster to realize task merging and path sharing, effectively balancing loads, reducing energy consumption, avoiding repeated flights, and improving overall collaborative efficiency.

[0013] 2. The present invention uses a feature-learning-based path change model and a local simulated annealing algorithm to determine and update the path in real time according to the drone's status and environmental changes, thereby enhancing the ability to respond to emergencies. At the same time, through game modeling and risk attenuation mechanisms, it can maintain mission continuity and scheduling stability in the face of random interference and historical uncertainties, and has strong engineering practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a structural diagram of an autonomous inspection system for an offshore wind farm operation and maintenance drone used in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0016] An autonomous inspection system for offshore wind farm operation and maintenance drones, see Figure 1 Shown, including: The autonomous inspection preparation module includes a risk investigation unit and an inspection preparation unit; the risk investigation unit is used to obtain offshore wind farm meteorological data and offshore wind farm radar point cloud data; risk prediction is performed based on the offshore wind farm meteorological data and the offshore wind farm radar point cloud data, to obtain an offshore wind farm inspection risk feature set and a corresponding offshore wind farm environment situation map; it is set that the offshore wind farm includes M wind turbine generators D m , m = 1, 2, …, M; the positions of the wind turbine generators D m are marked in the offshore wind farm environment situation map; The inspection preparation unit is used to set that N maintenance unmanned aerial vehicles J n perform inspection work at the same time, n = 1, 2, …, N; the corresponding power supply states of the maintenance unmanned aerial vehicles J n are E n ; historical maintenance records P m of the wind turbine generators D m are obtained; based on the historical maintenance records P m and the offshore wind farm environment situation map, an inspection and maintenance priority X m of the wind turbine generators D m is determined; based on the inspection and maintenance priority X m and the offshore wind farm environment situation map, autonomous inspection planning is performed, and an initial autonomous inspection path L n is allocated to all the maintenance unmanned aerial vehicles J n ; The autonomous inspection adjustment module includes an autonomous inspection unit; the autonomous inspection unit is used to obtain inspection dynamic variable parameters in real time during autonomous inspection, and update the autonomous inspection path L n in real time for the maintenance unmanned aerial vehicles J n , to obtain a new autonomous inspection path L n , until the autonomous inspection is completed.

[0017] The specific steps of performing risk prediction based on the offshore wind farm meteorological data and the offshore wind farm radar point cloud data include: The offshore wind farm meteorological data and the offshore wind farm radar point cloud data are subjected to feature enhancement using a knowledge graph, to obtain a fusion offshore wind farm gas phase feature tensor; an offshore wind farm meteorological risk correlation model is constructed based on the fusion offshore wind farm gas phase feature tensor; a pre-trained hypergraph neural network is used to perform risk analysis on the offshore wind farm meteorological risk correlation model, to obtain a dynamic meteorological risk factor graph and a meteorological risk propagation probability matrix; The offshore wind farm meteorological data includes wind speed, wind direction, air temperature, humidity, precipitation, air pressure, cloud map and other meteorological data, which is obtained from meteorological monitoring stations, offshore meteorological buoys, satellite remote sensing or other third-party meteorological platforms, etc.; Offshore wind farm radar point cloud data is constructed from 3D meteorological structure information above the wind farm, covering air turbulence, cloud shape, wave height, wind shear structure, and other information. It is acquired through high-precision marine radar systems and lidar sensors and is used to restore the three-dimensional spatial information of the wind farm. A meteorological knowledge graph for offshore wind farms is constructed, and entities and relationships are established for variables such as wind speed, wind direction, temperature, humidity, precipitation, thunderstorms, and wave height. In the feature enhancement step, sparse point completion and scene segmentation are performed on the radar point cloud data to extract the characteristics of the wind turbine infrastructure and the surrounding environment. A graph convolutional network (GCN) is used to propagate features of meteorological variables on the knowledge graph to construct a fused offshore wind farm gas phase feature tensor. Based on the fused offshore wind farm gas phase feature tensor, the offshore wind farm meteorological risk association model is used to analyze the causal relationship and synergy between different variables. The offshore wind farm meteorological risk association model is a graph structure model used to characterize the complex relationship between multiple meteorological factors and equipment risks in wind farms. Its essence is a high-dimensional graph model that integrates multi-source data, such as a heterogeneous graph, hypergraph, or dynamic causal graph, with the purpose of revealing the potential risk impact mechanism of meteorological changes on the safe operation of wind turbines. Among them, the vertex set in the offshore wind farm meteorological risk association model represents observed variables such as environmental parameters / structural responses; the edge set in the offshore wind farm meteorological risk association model represents the multi-parameter coupling relationship. Pre-trained hypergraph neural networks are used for joint modeling and reasoning of multi-source heterogeneous data. Dynamic meteorological risk factor maps are used to visualize the potential impact of regional meteorological variables on wind power equipment. Meteorological risk propagation probability matrices are used to predict the spatiotemporal evolution of risks within the wind farm space. The training process of the hypergraph neural network mainly includes the following steps: First, construct training samples containing multi-source heterogeneous data (such as wind speed, wave height, humidity, radar reflection intensity, etc.) and wind power equipment status. Each sample consists of a meteorological feature tensor of a time segment and a corresponding risk event label (such as equipment failure / warning); then, establish a hypergraph structure, associate multiple related meteorological variables with wind power equipment nodes through hyperedges, and encode each node feature as a vector input into the HGNN; the training method adopts supervised learning, using cross entropy or MSE loss function to optimize the node risk prediction value and actual risk label (such as 0 for normal and 1 for high risk); the goal is to minimize the prediction error so that the model can accurately output the risk score of each node and the propagation weight of the hyperedge; the sign of the end of training is when the loss function on the validation set stops decreasing for multiple consecutive rounds or reaches the set minimum error threshold, which means that the model is considered to have converged and has good generalization ability; Based on the dynamic meteorological risk factor map and the meteorological risk propagation probability matrix, the offshore wind farm radar point cloud data is updated to obtain the offshore wind farm environmental situation map; in the offshore wind farm environmental situation map, the wind turbine D is marked.m The position of the wind turbine generator set D m Corresponding meteorological risk prediction factor C m ; The original offshore wind farm radar point cloud data is modeled in three dimensions. The spatial influence range of each factor in the dynamic meteorological risk factor map is introduced and superimposed on the point cloud spatial coordinate system. The risk propagation probability matrix is ​​used to predict the meteorological risk changes in each region within several future time windows. A multi-channel spatial attention mechanism can be used to fuse the spatial coordinates in the point cloud with the risk score in the map, enhancing the semantic weight of high-risk areas in the three-dimensional model, and obtaining an offshore wind farm environmental situation map, which includes: the geographical location of wind turbines identified by structured point cloud, the meteorological risk heat of different regions formed by the map mapping and propagation matrix prediction results, and the risk evolution trajectory or trend arrow based on the dynamic reasoning of the propagation probability matrix. The meteorological risk prediction factor represents the location of the wind turbine. It searches for all meteorological factor nodes that act on the spatial range, extracts their risk scores and propagation coefficients, and calculates the weighted impact value of the equipment's exposure to meteorological risks during the forecast period based on the propagation probability matrix. Based on the dynamic meteorological risk factor map and the meteorological risk propagation probability matrix, feature output is performed to form a risk feature set for offshore wind farm inspections. The meteorological risk prediction factors of all wind turbines are integrated, and a feature fusion algorithm is used to form a multidimensional risk representation vector.

[0018] Based on historical operation and maintenance records m and offshore wind farm environmental situation map to determine wind turbine D m Inspection and maintenance priority X m The specific steps include: Historical operation and maintenance records m The wind turbine D m The number of historical faults, the type of recent fault repair, and the duration of recent fault downtime; Extract wind turbine D from offshore wind farm environmental situation map m Corresponding meteorological risk prediction factor C m , based on historical operation and maintenance records P m and meteorological risk prediction factor C m Constructing a multi-dimensional evaluation matrix Z for wind turbines m ; From the historical operation and maintenance records m Feature extraction is performed in combination with meteorological risk prediction factor C m Construct a one-dimensional wind turbine multidimensional evaluation matrix Z m ; Establish a risk exponential decay function for historical operation and maintenance records P m Matching risk attenuation factor Fm ; the risk index decay function is specifically F m = e -γ（tnow-tm） , where e is a mathematical constant, tnow represents the current time, tm represents the historical operation and maintenance record P m , where tm represents the time when the last failure occurred; and γ is an adjustment parameter, which is set by a human being; based on the multi-dimensional evaluation matrix Z m of the wind turbine, risk semantic analysis is performed, and a multi-dimensional feature vector Z m of the wind turbine operation and maintenance is output; is input into a semantic extraction network such as an MLP (Multi-Layer Perceptron) or an attention mechanism network; meteorological risk factors and historical characteristics of equipment are weighted and fused, and a risk semantic representation vector is output: each dimension in the vector represents a certain risk preference, and a multi-dimensional feature vector Z m ’ of the wind turbine operation and maintenance is output; based on the multi-dimensional feature vector Z m ’ of the wind turbine operation and maintenance, the risk decay factor F m , and the dynamic game model of the wind turbine operation and maintenance, priority matching is performed, and an inspection and maintenance priority X m is output; The dynamic game model of the wind turbine operation and maintenance is specifically: ; where i is a continuous time variable, which is counted from the start of the priority task; f (·) represents a priority dynamic evolution function, which is set by a professional technician; and σ represents an environmental noise intensity coefficient, which is set by a human being; represents a standard Wiener process, and a Brownian motion differential is used to represent unpredictable random disturbances; where represents the dynamic change rate of the inspection and maintenance priority X m over time i, which reflects how the inspection priority of the wind turbine D m changes over time and environmental conditions; the role of the priority dynamic evolution function is to calculate the trend change of the priority according to the current state and historical data; represents a random disturbance term, which simulates unpredictable environmental noise; through a dynamic differential equation form, it can respond to the state of the wind turbine (such as sudden failure, weather change) and the demand for cooperation of the UAV cluster in real time; and are introduced to ensure that the model not only considers the long-term historical failure rules (such as higher priority for frequently malfunctioning units), but also weakens the influence of obsolete data through the risk decay factor F m ; Before the priority dynamic evolution function is calculated, the wind turbine operation and maintenance multi-dimensional feature vector is converted into a vector scalar, and then the calculation is performed. The setting means of the priority dynamic evolution function needs to comprehensively consider the real-time state of the wind turbine, historical operation and maintenance data, and environmental risk factors, so as to ensure that the adjustment of the inspection priority is scientific and efficient; first of all, professional and technical personnel need to define the basic form of the function according to actual needs, and usually adopt a nonlinear function to capture the complex relationship between the priority and the multi-dimensional features; for example, a function based on weighted summation or neural network mapping can be constructed, taking the wind turbine operation and maintenance multi-dimensional feature vector, the current priority and the risk decay factor as input, and outputting the trend of the priority change through parameterized modeling. The key of this step is to ensure that the function can reflect the dynamic influence of different factors on the priority, for example, the high-frequency fault unit should obtain a higher priority promotion rate; after determining the function form, parameter calibration is needed through historical data; using the past operation records and meteorological data of the wind farm, combined with the actual inspection results, the weight coefficients or neural network parameters in the function are trained; the training process can adopt gradient descent or other optimization algorithms, and the goal is to minimize the error between the priority prediction value and the actual demand; for example, if a unit fails due to not timely inspection, backtracking analysis should adjust the function parameters, so that the priority can rise faster under similar conditions, this step relies on high-quality data labeling and iterative optimization to ensure that the prediction ability of the function meets the needs of the actual scene.

[0019] In order to enhance the adaptability of the function, a dynamic adjustment mechanism also needs to be introduced. Due to the complex and changeable environment of offshore wind farms, the function needs to be able to fine-tune according to real-time feedback. For example, when the hypergraph neural network detects a new weather risk pattern, the internal parameters of the function can be updated through online learning techniques. In addition, the time sensitivity of the risk decay factor also needs to be dynamically adjusted, such as increasing the weight coefficient of recent faults during typhoon season. This process usually requires embedded logic or external control interfaces to allow operation and maintenance personnel to intervene in the adjustment range of key parameters based on experience. Finally, the output of the function needs to work with other modules, and the priority dynamic change needs to be converted into specific drone task instructions, so the output range of the function must match the input specifications of the task allocation module; for example, if the task allocation module uses a priority score of 0 to 100, the output of the function needs to be mapped to this interval through scaling or truncation; at the same time, the calling frequency of the function needs to be synchronized with the state update period of the drone, usually set to calculate once every fixed time step, to ensure the timeliness of priority adjustment, this integration process relies on system-level timing control and data pipeline design to ensure seamless connection between the dynamic game model and modules such as path planning and risk prediction.

[0020] Through the above steps, the historical data, current risk and unmanned aerial vehicle state can be comprehensively judged to avoid static planning failure; an exponential decay weight is introduced to reduce old data interference and improve real-time performance; the actual scene of multiple machines, multiple tasks and multiple disturbances can be simulated, which is suitable for complex environment dynamic changes, and the comprehensive health index of the wind turbine is no longer dependent on a single indicator, and has behavior pattern recognition capability.

[0021] Based on the inspection operation priority X m The specific steps of autonomous inspection planning based on the offshore wind farm environment situation map include: Based on the digital twin analysis method and the simulated annealing algorithm, the inspection operation priority X m and the offshore wind farm environment situation map to construct a plurality of initial path planning individuals; based on the plurality of initial path planning individuals to construct an initial path planning population; Based on the offshore wind farm GIS data, radar point cloud and environment situation map, a three-dimensional space simulation model is constructed for simulation analysis; each initial path planning individual is represented as the arrangement of the inspection sequence planned by all operation unmanned aerial vehicles; Taking the maximum total inspection priority coverage value, the minimum total path risk cost and the maximum autonomous inspection efficiency as the objective function to perform population iteration, and selecting the optimal initial path planning individual from the plurality of initial path planning individuals; Based on the optimal initial path planning individual, the initial autonomous inspection path L n is allocated to all operation unmanned aerial vehicles J n . In the autonomous inspection process, the inspection dynamic variable parameters are obtained in real time; wherein, the inspection dynamic variable parameters of the operation unmanned aerial vehicle J n include: self-state parameters, external meteorological parameters, inspected wind turbine state parameters, and cluster cooperation state parameters; for the operation unmanned aerial vehicle J n , the operation unmanned aerial vehicle adjacent to the operation unmanned aerial vehicle J n is L nk . Wherein, the self-state parameters are power, motor temperature, remaining range, flight stability and the like; the external meteorological parameters are current wind speed, humidity, wave height and the like; the inspected wind turbine state parameters are whether the inspected wind turbine has found a fault and whether it needs secondary inspection; the cluster cooperation state parameters represent the current state, idle degree and path overlap of other unmanned aerial vehicles; The specific steps of real-time updating the autonomous inspection path L n of the operation unmanned aerial vehicle J n include: Every Δt judges whether the inspection dynamic variable parameters of the operation unmanned aerial vehicle J n meet the path change standard; if the operation unmanned aerial vehicle Jn If the path change criteria are met, then the operation and maintenance drone J n Generate a set of candidate paths; the unit of Δt is seconds and can be set to 30 seconds or other values, which is manually set; the path change standard is determined by combining the dynamic inspection variable parameters with the trained machine learning model; feature learning is performed on the dynamic inspection variable parameters to obtain the dynamic inspection variable parameter features; the dynamic inspection variable parameter features are input into the trained machine learning model to determine whether the path change standard is met; The machine learning model can use the MLP model as the basic model. The training set is obtained by recording the drone status and environmental parameters from a large number of real or simulated inspection tasks. The label "whether the path change was executed at that time" is manually or systematically marked. The label value is 1 if the path has actually changed, and 0 if it has not changed. The model is trained based on the training set. When the model output judgment accuracy reaches the set standard, the training is stopped to obtain the trained machine learning model. The set standard is set manually. Use the local simulated annealing algorithm to generate path perturbations and obtain a set of candidate paths; First, the inspection path of the current maintenance drone is used as the initial solution, and the local perturbation range is set (such as exchanging, inserting, or removing some wind turbine nodes). Then, the objective function is defined, with maximizing task priority coverage, minimizing risk cost, and rationalizing flight distance as the multi-objective trade-off criteria. In each iteration, a path perturbation operation is randomly selected to generate a new solution. The objective function value of the new path is calculated, and the decision on whether to accept the solution is made based on the current temperature parameter and the Metropolis criterion. The probability of accepting a poor solution is controlled by gradually lowering the temperature, making it possible to escape the local optimum. Finally, the iteration ends at the set number of iterations or temperature threshold conditions, and multiple high-quality, locally optimal candidate paths are output to form a candidate path set for subsequent dynamic path update selection. Determine the operation and maintenance drone L nk Can the operation and maintenance drone be merged? n If the result is yes, the operation and maintenance drone L is selected based on the candidate path set. nk Update autonomous inspection paths in real time, and update operation and maintenance drones in real time n Autonomous inspection path L n ; If the result is no, then the operation and maintenance drone J is selected based on the candidate path set n Real-time update of autonomous inspection path L n ; The specific steps are: If the operation and maintenance drone L nk There are few inspection tasks left or most of them have been completed, that is, their task load is low and they have the ability to accept additional tasks; nk With the operation and maintenance drone Jn The spatial trajectory of the current path, if the two paths are basically the same or there is a high degree of spatial overlap, it means that the task transfer will not lead to a significant increase in the path cost; confirm the operation and maintenance drone L nk The flight status of the UAV is good, such as sufficient battery power, stable operation of the equipment, and not in the return or fault state. If all the above conditions are met, the mission is considered to be mergeable. At this time, the optimal path will be selected from the generated candidate paths and the operation and maintenance UAV L will be replanned. nk Inspection paths, integrated from the operation and maintenance drone J n The updated path is recorded as L nk ';At the same time, the original allocation to the operation and maintenance drone J n The task of the drone will be stripped off and the operation and maintenance of the drone will be arranged n Perform one of the following two actions: Return: if the battery is insufficient or no backup task is assigned; or Switch to the backup task: if the system has a low priority or inspection supplement task reserved for it, it will be assigned to continue the operation; On the contrary, if any of the conditions are not met (such as path direction conflict, insufficient battery, too heavy a task, etc.), the task will not be merged; at this time, the optimal path in the candidate path set generated by the local simulated annealing algorithm will be directly assigned to the operation and maintenance drone J. n Update paths individually to reflect current state changes.

[0022] An autonomous inspection means that all wind turbines in the offshore wind farm are inspected at least once, and each wind turbine is inspected by at least one operation and maintenance drone; Ensure that each wind turbine is inspected by a drone at least once, eliminate the omission of key equipment, significantly reduce the risk of wind turbine fault warning blind spots, and improve overall operational safety and management controllability; through initial task allocation and multi-objective optimization path planning based on inspection priority, combined with simulated annealing algorithm and digital twin environment modeling, maximize the flight efficiency and task completion of drones, reduce repeated flights and energy waste; introduce cluster state perception and dynamic task collaboration mechanism, support real-time task redistribution and path merging between drones, achieve dynamic load balancing, and effectively deal with drone performance differences, power fluctuations and sudden failures Obstacles; weather risk prediction, propagation matrix and real-time situation map update are introduced during the inspection process. The drone path can be automatically adjusted according to environmental changes, avoiding flight in high-risk areas and improving the safe completion rate of inspection tasks; the path change adopts an adaptive machine learning model judgment mechanism, combined with the local simulated annealing algorithm and task merging strategy, which has self-repair and self-scheduling capabilities in emergency situations to ensure that the overall inspection task is not interrupted or missed; it integrates priority dynamic game modeling, risk factor fusion analysis, feature learning and reasoning judgment, so that inspection decision-making moves from static rules to intelligent learning, effectively improving system autonomy and task adaptability.

[0023] It is to be understood that all of the above modifications and alterations can be made to the above-described arrangements without departing from the spirit or scope of the present application, which is defined in the appended claims. Parts of this description which have not been described in detail are part of the prior art known to those skilled in the art.

Claims

1. An autonomous inspection system for offshore wind farm operation and maintenance drones, characterized by: include: Autonomous inspection preparation module, including risk investigation unit and inspection preparation unit; The risk investigation unit is used to obtain meteorological data and radar point cloud data of offshore wind farms; Based on the meteorological data and radar point cloud data of offshore wind farms, risk prediction is performed to obtain the offshore wind farm inspection risk feature set and the corresponding offshore wind farm environmental situation map; assuming that the offshore wind farm contains M wind turbines D m , m=1, 2, ..., M; wind turbine D is marked in the offshore wind farm environmental situation map m location; The inspection preparation unit is used to set a total of N operation and maintenance drones J n At the same time, inspection work is carried out, n=1, 2, ..., N; operation and maintenance drone J n The corresponding power state is E n ; Get wind turbine D m Historical operation and maintenance records of P m ; Based on historical operation and maintenance records P m and offshore wind farm environmental situation map to determine wind turbine D m Inspection and maintenance priority X m ; Based on inspection and maintenance priority X m and offshore wind farm environmental situation map for autonomous inspection planning, providing all operation and maintenance drones n Assign the initial autonomous inspection path L n ; An autonomous inspection and adjustment module includes an autonomous inspection unit; The autonomous inspection unit is used to obtain dynamic and variable inspection parameters in real time during the autonomous inspection process. n Real-time update of autonomous inspection path L n , get the new autonomous inspection path L n , until the end of an independent inspection.

2. The autonomous inspection system for offshore wind farm operation and maintenance drones according to claim 1 is characterized in that: The specific steps for risk prediction based on offshore wind farm meteorological data and offshore wind farm radar point cloud data include: The knowledge graph is used to enhance the features of offshore wind farm meteorological data and offshore wind farm radar point cloud data, and a fused offshore wind farm gas phase feature tensor is obtained. A meteorological risk association model for offshore wind farms is constructed based on the fused offshore wind farm gas phase feature tensor. A pre-trained hypergraph neural network is used to perform risk analysis on the offshore wind farm meteorological risk association model, and a dynamic meteorological risk factor map and meteorological risk propagation probability matrix are obtained. Based on the dynamic meteorological risk factor map and the meteorological risk propagation probability matrix, the offshore wind farm radar point cloud data is updated to obtain the offshore wind farm environmental situation map; in the offshore wind farm environmental situation map, the wind turbine D is marked. m The position of the wind turbine generator set D m Corresponding meteorological risk prediction factor C m ; Based on the dynamic meteorological risk factor map and the meteorological risk propagation probability matrix, feature output is performed to form the offshore wind farm inspection risk feature set.

3. The autonomous inspection system for offshore wind farm operation and maintenance drones according to claim 2 is characterized in that: Based on historical operation and maintenance records m and offshore wind farm environmental situation map to determine wind turbine D m Inspection and maintenance priority X m The specific steps include: Historical operation and maintenance records m The wind turbine D m The number of historical faults, the type of recent fault repair, and the duration of recent fault downtime; Extract wind turbine D from offshore wind farm environmental situation map m Corresponding meteorological risk prediction factor C m , based on historical operation and maintenance records P m and meteorological risk prediction factor C m Constructing a multi-dimensional evaluation matrix Z for wind turbines m ; Establish a risk exponential decay function for historical operation and maintenance records P m Matching risk attenuation factor F m ; Based on the multi-dimensional evaluation matrix Z of wind turbines m Perform risk semantic analysis and output the wind turbine operation and maintenance multi-dimensional feature vector Z m '; Based on the multidimensional characteristic vector Z of wind turbine operation and maintenance m ', risk attenuation factor F m Priority matching is performed with the wind turbine operation and maintenance dynamic game model to output the inspection and maintenance priority X m .

4. The autonomous inspection system for offshore wind farm operation and maintenance drones according to claim 3 is characterized in that: Based on inspection and maintenance priority X m The specific steps for autonomous inspection planning based on the offshore wind farm environmental situation map include: Based on digital twin analysis method and simulated annealing algorithm, the inspection and maintenance priority X m and an offshore wind farm environmental situation map to construct a number of initial path planning individuals; and construct an initial path planning population based on the number of initial path planning individuals; The objective function is to maximize the total inspection priority coverage value, minimize the total risk cost of the path, and maximize the autonomous inspection efficiency. The optimal initial path planning individual is selected from several initial path planning individuals through population iteration. Based on the optimal initial path planning individual, for all operation and maintenance drones J n Assign the initial autonomous inspection path L n .

5. The autonomous inspection system for offshore wind farm operation and maintenance drones according to claim 4 is characterized in that: During the autonomous inspection process, dynamic and variable inspection parameters are obtained in real time; among them, the operation and maintenance drone J n Dynamic variable parameters of inspection include: self-state parameters, external meteorological parameters, state parameters of inspected wind turbines, cluster collaboration state parameters; n , Setting up and operating drones J n The adjacent operation and maintenance drone is L nk .

6. The autonomous inspection system for offshore wind farm operation and maintenance drones according to claim 5, characterized in that: For operation and maintenance drone J n Real-time update of autonomous inspection path L n The specific steps include: Every Δt, the maintenance drone J is judged n Whether the dynamic variable parameters of the inspection meet the path change standards; if the operation and maintenance drone J n If the path change criteria are met, then the operation and maintenance drone J n Generate a set of candidate paths; Determine the operation and maintenance drone L nk Can the operation and maintenance drone be merged? n If the result is yes, the operation and maintenance drone L is selected based on the candidate path set. nk Update autonomous inspection paths in real time, and update operation and maintenance drones in real time n Autonomous inspection path L n ; If the result is no, then the operation and maintenance drone J is selected based on the candidate path set. n Real-time update of autonomous inspection path L n .

7. The autonomous inspection system for offshore wind farm operation and maintenance drones according to claim 6, characterized in that: An autonomous inspection means that all wind turbines in the offshore wind farm are inspected at least once, and each wind turbine is inspected by at least one operation and maintenance drone.

Images