Hybrid prediction method and prediction system for lightning strike probability of offshore wind turbines

Abstract

This invention discloses a hybrid prediction method for the lightning strike probability of offshore wind turbines: 1. Constructing a parameterized static potential and electric field background library; 2. Setting up an n-level virtual envelope surface, with a sparse electrode grid on the outermost layer to solve the electrostatic field and construct the background library, obtaining the virtual attachment points of the next-level envelope surface as new electrodes, and repeating the solution and path prediction step by step; 3. Starting from the nth-level electrode grid point, using a path search with decreasing potential constraints and end-point field strength rewards, obtaining the attachment points of the next-level virtual surface in the region formed by the virtual envelope surface, and recording the path points between the start and end points as the leading development path; 4. Weighted fusion of multiple simulation results from each starting point of the multi-electrode grid to form a lightning strike probability distribution cloud map. This invention also discloses a hybrid prediction system for the lightning strike probability of offshore wind turbines. The hybrid prediction method for the lightning strike probability of offshore wind turbines in this invention solves the problems of excessive computation and long time consumption of fractal methods, achieving a balance between high accuracy and high efficiency.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine lightning strike probability prediction technology. Specifically, this invention relates to a hybrid prediction method for the probability of lightning strikes on offshore wind turbines, and also to a hybrid prediction system for implementing the above-mentioned hybrid prediction method for the probability of lightning strikes on offshore wind turbines. Background Technology

[0002] Wind farms are typically located in open plains, hills, or near-shore areas. The towering wind turbines are highly susceptible to damage from lightning strikes. According to recent failure statistics, lightning-induced blade damage accounts for over 50% of all failures, posing a significant threat to the safe operation of wind turbines. Offshore wind turbines operate in the high-salt-spray, high-humidity marine atmosphere. Salt spray alters the conductivity of the blade surface, significantly increasing the probability of lightning strikes and making lightning risk assessment more complex.

[0003] Currently, methods for calculating lightning strike probability mainly fall into two categories: one is the electrogeometric model (EGM) based on empirical formulas, which struggles to reflect the actual physical process of discharge; the other is the discharge simulation method based on fractal theory, which, while capable of simulating the randomness and bifurcation characteristics of leader development, suffers from massive computational demands and long processing times. Particularly for wind turbine lightning simulation models, which include millions of finite element elements, obtaining a statistically significant lightning strike probability distribution requires running thousands or even tens of thousands of fractal simulations, each involving complex electric field recalculations, making it difficult to meet the needs of rapid engineering evaluation.

[0004] The above calculation methods also waste computing resources. Even with a fractal strategy using variable step size optimization, thousands of calculations are still required. Especially when the lightning leader is far from the wind turbine, the calculation results of the overall electric field distribution of most wind turbines are highly similar, which is a huge waste of computing resources. Summary of the Invention

[0005] The first objective of this invention is to provide a hybrid prediction method for the probability of lightning strikes on offshore wind turbines, which solves the problems of excessive computation and long time consumption of traditional fractal methods, and achieves a balance between high accuracy and high efficiency.

[0006] The second objective of this invention is to provide a hybrid prediction system for implementing the above-described hybrid prediction method for the probability of lightning strikes on offshore wind turbines.

[0007] The first technical solution adopted in this invention is a hybrid prediction method for the probability of lightning strikes on offshore wind turbines, the specific method of which is as follows:

[0008] S1. Construct a parameterized static potential and electric field background library;

[0009] S2. Multi-level iterative construction of electrode mesh and background library: n-level virtual envelope surfaces are set with the wind turbine as the center; sparse electrode meshes are laid out on the outermost layer, and the electrostatic field is solved to construct the background library; the path is predicted in the influence area to obtain the virtual attachment point of the next level envelope surface and use it as a new electrode. The computational domain is gradually reduced, and the solution and path prediction are repeated until the surface of the wind turbine is reached.

[0010] S3. Pilot development simulation based on hybrid path prediction model: The hybrid path prediction model is used to search along the path of potential decrease constraint and terminal field strength reward to find the next level virtual surface attachment point until the wind turbine surface attachment point, and obtain the pilot development path; and the electrode-multi-attachment point model is used to obtain the probability distribution characteristics.

[0011] S4. Generation of lightning probability distribution: Based on the weighted fusion of simulation results at each starting point of the multi-electrode grid, a cloud map of the lightning probability distribution on the surface of the wind turbine is formed.

[0012] The invention is further characterized by:

[0013] The specific method for S1 is as follows:

[0014] S1.1 Establish a three-dimensional geometric model of the offshore wind turbine, including blades, lightning arresters, nacelle, and tower; for floating wind turbines, further establish the geometric configurations corresponding to different tilt angles and blade azimuth angles.

[0015] S1.2 Set material properties and environmental parameters. Specifically, seawater is set as an ideal conductor boundary, and equivalent surface conductivity is set on the blade surface to simulate different salt spray pollution levels, which are divided into clean, lightly polluted and heavily polluted.

[0016] S1.3 Set a set of typical lightning current waveform parameters, including amplitude, wavefront time and half-peak time;

[0017] S1.4. For different parameter combinations in steps S1.1 to S1.3, use the electrostatic field solver to calculate the spatial potential distribution of the entire solution domain. and electric field distribution This forms a parameterized library of static potential and electric field backgrounds.

[0018] The specific method for S2 is as follows:

[0019] S2.1 Define a multi-level virtual envelope surface:

[0020] Centered on the wind turbine, multiple virtual envelope surfaces are set up, which divide the space into n levels, where n≥3, and each level corresponds to a stage of pioneering development;

[0021] S2.2, Construct the first-level mesh and background library:

[0022] A sparse starting point grid is generated on the first-level envelope surface, with each grid node representing a downward leader starting point; for all possible operating conditions, the electric field distribution of the entire space from the envelope surface to the fan surface is calculated using an electrostatic field solver to form the first background library;

[0023] S2.3, First-level path prediction and virtual attachment point generation:

[0024] Starting from each starting point, a hybrid path prediction model is used to simulate the process of the leader evolving from the first-level envelope to the second-level envelope. The endpoint of the path prediction is not the wind turbine surface, but a point on the second-level envelope. The endpoints of all paths on the second-level envelope constitute a set of virtual attachment points, which connect all discrete grid nodes on the path to generate the first-level leader path set.

[0025] S2.4, Second-level mesh and background library:

[0026] Based on the virtual attachment points generated in the first level, the mesh is refined on the second-level envelope surface to generate new starting points. At this time, the computational domain is reduced to the space from the second-level envelope surface to the wind turbine surface. Only the region from the second-level envelope surface to the wind turbine surface needs to be meshed with high resolution to recalculate the electric field distribution and form the second background library.

[0027] S2.5, Second-level path prediction and generation of the next-level virtual attachment point:

[0028] Starting from each new starting point on the second-level virtual envelope, path prediction is performed again to simulate the process of the leader evolving to the third-level virtual envelope, and new virtual attachment points are generated on the third-level virtual envelope.

[0029] S2.6, nth level mesh and background library:

[0030] Based on the virtual attachment points generated at level n, the starting point mesh is generated by densifying the virtual envelope surface at level n, and the local electric field from the virtual envelope surface at level n to the wind turbine surface is calculated to form the background library at level n.

[0031] S2.7 Final Path Prediction:

[0032] The final path prediction is performed from the starting point on the nth level envelope surface, directly calculating the flash point on the wind turbine surface;

[0033] S2.8, Iteration Termination Condition:

[0034] When the change in probability distribution between two adjacent iterations is less than a preset threshold, the result is considered to have converged and the iteration is stopped; otherwise, S2.4 to S2.7 can be repeated on the latest envelope surface to further refine the mesh.

[0035] The specific method for S3 is as follows:

[0036] S3.1 Data Extraction and Preprocessing:

[0037] Extract the three-dimensional electric potential and electric field mesh data between the starting point and the target envelope from the background library at the current level, and construct the electric potential array. and electric field strength array Among them, the potential array The potential of the envelope surface from the starting point to the target surface is monotonically decreasing;

[0038] S3.2 Determination of the starting point and candidate endpoint:

[0039] Starting point: The coordinates of the starting point of the current level, denoted as... ,in The set of all starting points;

[0040] Candidate endpoints: On the target envelope, candidate points are selected based on the local maxima of the electric field intensity. , recorded as , The set of all candidate endpoints;

[0041] S3.3, Shortest Path Search with Constraints:

[0042] Treating the 3D mesh as a graph, with nodes as mesh points and edges connecting 26 neighboring nodes, for a given candidate endpoint... The search starts from the point of origin, under the condition of satisfying the potential decrease constraint. To the candidate endpoint Find the optimal path and calculate the path from the starting point. sm To the finish line tn Comprehensive indicators Select to make The minimum endpoint and its path are used as the prediction result;

[0043] S3.4, Competition among multiple candidate endpoints and pseudo-random statistical processing:

[0044] For each candidate destination, a path search is performed once to calculate the optimal path cost to that destination. Then calculate the comprehensive index. ,choose The smallest endpoint and its path; where, The weighting coefficient is determined through scaled-down experiments or historical data, and its value ranges from 0.1 to 0.5. End point t The electric field strength;

[0045] To obtain the probability distribution characteristics, Monte Carlo sampling of the background electric field library can be combined with repeated execution of S3.3 to obtain multiple possible attachment points corresponding to the same starting point, thus achieving pseudo-random processing similar to fractal calculation.

[0046] The specific method for S3.3 is as follows:

[0047] Movement constraint: Movement is only allowed in the direction of decreasing potential, i.e.:

[0048] (1);

[0049] in, The current node potential, The potential of the surrounding 26 neighboring nodes;

[0050] The edge cost definition for shortest path search:

[0051] (2);

[0052] in For nodes electric field strength, For nodes electric field strength, It is a constant;

[0053] From the starting point a certain point in the middle Candidate set to the endpoint a certain point in the middle cumulative path cost between Cost of edges between all adjacent nodes on the path sum:

[0054] (3);

[0055] in, Accumulate cost for the path; P To start from the beginning To the finish line The path passes through i 10 nodes form a sequence ;

[0056] Introducing the finish line strength reward, the path is obtained. Comprehensive indicators :

[0057] (4);

[0058] in, End point t electric field strength, The weighting coefficient is determined through scaled-down experiments or historical data, and its value ranges from 0.1 to 0.5.

[0059] Use A Or Dijkstra's algorithm calculates from all starting sets to candidate endpoint set The optimal path is obtained, yielding the minimum Compare its comprehensive indicators ;

[0060] Select all The smallest endpoint and its path are used as the starting point set. to candidate endpoint set The prediction results are as follows: if the target of the current level is a virtual envelope surface, the end point of the path is recorded as the virtual attachment point of the next level; if the target is the wind turbine surface, the flash point is recorded.

[0061] In formula (2), ε is taken as 1 / 1000 or 10 of the minimum non-zero electric field intensity in the background library. -6 The minimum non-zero field strength is times the field strength.

[0062] The specific method for S4 is as follows:

[0063] S4.1 Input data:

[0064] Starting point set Each starting point Having spatial coordinates and the volume element it represents Volume element It can be calculated using Voronoi partitioning or grid rules;

[0065] Simulation results: For each starting point , conducted Sub-independent random Monte Carlo simulation. The simulation results for each simulation are recorded as follows:

[0066] : Whether the wind turbine was hit; where 1 indicates hit and 0 indicates no hit; if hit, record the coordinates of the lightning strike point on the wind turbine surface. ;

[0067] Among them, subscript Starting point Serial number marking, Number of simulations performed for this starting point mark, ;

[0068] S4.2, Surface element division of the fan:

[0069] The surface of the fan is divided into Individual Each element has geometric information, and a spatial index is established to quickly locate the element to which the flash point belongs.

[0070] S4.3 Weight Calculation:

[0071] The total volume of the sampling space is Each starting point The weights are:

[0072] (5);

[0073] Under the assumption of uniform distribution, weights Proportional to the starting point Having spatial coordinates and the volume element it represents The actual spatial proportion of the corresponding area;

[0074] S4.4 Weighted Statistics:

[0075] Iterate through all starting points and all simulations, accumulating the weighted hit count on each face element, and simultaneously accumulating the total weighted simulation count. :

[0076] (6);

[0077] in, As weight, The number of random Monte Carlo simulations corresponding to the weights;

[0078] S4.5 Calculate the probability of a surface element being struck by lightning:

[0079] A random Monte Carlo simulation pilot ultimately hits a surface element. Lightning intercept probability estimate for:

[0080] (7);

[0081] This value satisfies ,margin The total probability that the pilot misses the wind turbine; where For face element The total number of times the target was hit in the simulation; The total weighted number of simulations for the leading hit in random Monte Carlo simulations;

[0082] If the probability density of surface element lightning is required Further calculations can be performed:

[0083] (8);

[0084] in, For face element Area; subscript The number of facets to divide the surface area of ​​the fan;

[0085] S4.6, Generate a cloud map:

[0086] The probability of each facet receiving lightning. or probability density By mapping the color to the surface element and rendering each surface element on the 3D model of the wind turbine, a cloud map of the lightning probability distribution can be obtained.

[0087] The second technical solution adopted in this invention is: a hybrid prediction system for the lightning strike probability of offshore wind turbines using the above prediction method, including a static electric field background library construction module, a multi-level iterative grid generation module, a hybrid path prediction module, and a lightning strike probability generation module;

[0088] The Static Electric Field Background Library Construction Module is used to build a parameterized static potential and electric field background library;

[0089] The multi-level iterative mesh generation module is used to construct electrode meshes and background libraries through multi-level iterations;

[0090] The hybrid path prediction module is used to perform pilot development simulations of prediction models based on hybrid paths;

[0091] The lightning strike probability generation module is used to generate the lightning strike probability distribution.

[0092] The beneficial effects of this invention are:

[0093] (1) The present invention proposes a hybrid prediction method for the probability of lightning strikes on offshore wind turbines. It proposes a three-stage hybrid prediction method of "static electric field pre-calculation, multi-electrode grid sampling, and hybrid path prediction model" to efficiently evaluate the probability of lightning strikes on offshore wind turbines. Its core is to decouple the high-cost calculation link in the traditional fully stochastic fractal simulation. Through the organic combination of static electric field pre-calculation, multi-electrode grid sampling, and hybrid path prediction model (rapid screening by steepest descent method, accurate optimization by cost distance algorithm, etc.), a balance between accuracy and efficiency is achieved.

[0094] (2) The hybrid prediction method for the probability of lightning strikes on offshore wind turbines in this invention adopts a multi-level iterative strategy from coarse to fine, gradually refining the calculation area at different distance scales to avoid the waste of resources caused by high-precision calculation in the whole space. Attached Figure Description

[0095] Figure 1 This is a flowchart of the hybrid prediction method for the probability of lightning strikes on offshore wind turbines according to the present invention;

[0096] Figure 2This is a schematic diagram of the process of progressively dividing the envelope and finding the path in the hybrid prediction method for the probability of lightning strikes on offshore wind turbines in this invention.

[0097] Figure 3 This is a schematic diagram of the electrode positions on the first-level virtual envelope surface, which is set 500 meters above the wind turbine in Embodiment 1 of the present invention;

[0098] Figure 4 This is a schematic diagram of the electrode positions on the second-level virtual envelope surface, which is set 50 meters above the wind turbine in Embodiment 1 of the present invention;

[0099] Figure 5 This is a schematic diagram of the electrode positions on the third-level virtual envelope surface, which is set 5 meters above the wind turbine in Embodiment 1 of the present invention;

[0100] Figure 6 This is a schematic diagram of the lightning path calculation results in Embodiment 1 of the present invention. Detailed Implementation

[0101] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0102] This invention provides a hybrid prediction method for the probability of lightning strikes on offshore wind turbines, such as... Figure 1 As shown, the specific method is as follows:

[0103] S1. Construct a parameterized static potential and electric field background library; the specific method is as follows:

[0104] S1.1 Establish a three-dimensional geometric model of the offshore wind turbine, including blades, lightning arresters, nacelle, and tower; for floating wind turbines, further establish the geometric configurations corresponding to different tilt angles and blade azimuth angles.

[0105] S1.2 Set material properties and environmental parameters. Specifically, seawater is set as an ideal conductor boundary, and equivalent surface conductivity is set on the blade surface to simulate different salt spray pollution levels, which are divided into clean, lightly polluted and heavily polluted.

[0106] S1.3 Set a set of typical lightning current waveform parameters, including amplitude, wavefront time and half-peak time;

[0107] S1.4. For different parameter combinations in steps S1.1 to S1.3, use the electrostatic field solver to calculate the spatial potential distribution of the entire solution domain. and electric field distribution This forms a parameterized library of static potential and electric field backgrounds.

[0108] Among them, the electric potential ensures a downward trend towards wind turbines, while the electric field ensures the optimal path.

[0109] S2. Multi-level iterative construction of electrode mesh and background library: An n-level virtual envelope surface is set up with the wind turbine as the center; a sparse electrode mesh is laid out on the outermost layer, and the electrostatic field is solved to construct the background library; within the influence region, the path is predicted to obtain the virtual attachment points of the next-level envelope surface and used as new electrodes. The computational domain is gradually reduced, and the solution and path prediction are repeated until the wind turbine surface is reached; the specific method is as follows:

[0110] S2.1 Define a multi-level virtual envelope surface:

[0111] Centered on the wind turbine, multiple virtual envelope surfaces are set up, which divide the space into n levels, where n≥3, and each level corresponds to a stage of pioneering development;

[0112] The multi-level virtual envelope surface actually set in this invention may specifically include a far-field envelope surface, a mid-field envelope surface, and a near-field envelope surface;

[0113] For example: the far-field envelope is 500 meters away from the wind turbine surface, the mid-field envelope is 50 meters away from the wind turbine surface, and the near-field envelope is 5 meters away from the wind turbine surface.

[0114] S2.2, Construct the first-level mesh and background library:

[0115] A sparse starting point grid (e.g., grid spacing of 80 meters) is generated on the first-level envelope surface, with each grid node representing a downlink leader starting point; for all possible operating conditions (wind turbine attitude, salt spray pollution, lightning current waveform), the electric field distribution of the entire space from the envelope surface to the wind turbine surface is calculated using an electrostatic field solver to form the first background library;

[0116] Since the electric field changes gradually when the electrodes are far from the wind turbine, a coarser grid can be used for this stage of calculation.

[0117] S2.3, First-level path prediction and virtual attachment point generation:

[0118] Starting from each starting point, a hybrid path prediction model is used to simulate the process of the leader evolving from the first-level envelope to the second-level envelope. The endpoint of the path prediction is not the wind turbine surface, but a point on the second-level envelope. The endpoints of all paths on the second-level envelope constitute a set of virtual attachment points, which represent intermediate positions that the leader may reach. Connecting all discrete grid nodes on the path generates the first-level leader path set.

[0119] S2.4, Second-level mesh and background library:

[0120] Based on the virtual attachment points generated in the first stage, the mesh is refined on the second-stage envelope surface (e.g., mesh spacing of 20 meters) to generate new starting points. At this time, the computational domain is reduced to the space from the second-stage envelope surface to the wind turbine surface. Only the region from the second-stage envelope surface to the wind turbine surface needs to be divided into high-resolution meshes and the electric field distribution needs to be recalculated to form the second background library.

[0121] Because the electrode is relatively closer to the wind turbine than the first-stage path, the area is reduced, and a medium-sized grid is used for this stage of calculation.

[0122] S2.5, Second-level path prediction and generation of the next-level virtual attachment point:

[0123] Starting from each new starting point on the second-level virtual envelope, path prediction is performed again to simulate the process of the leader evolving to the third-level virtual envelope, and new virtual attachment points are generated on the third-level virtual envelope.

[0124] S2.6, nth level mesh and background library:

[0125] Based on the virtual attachment points generated at level n, the starting point mesh is generated by densifying the virtual envelope surface at level n, and the local electric field from the virtual envelope surface at level n to the wind turbine surface is calculated to form the background library at level n.

[0126] Since the electrode is relatively close to the wind turbine, the area is further reduced, and special local structures of the wind turbine (such as the blade tip, the blade containing the lightning rod, etc.) can be selected for calculation. This level of calculation uses a finer grid.

[0127] S2.7 Final Path Prediction:

[0128] The final path prediction is performed from the starting point on the nth level envelope surface, directly calculating the flash point on the wind turbine surface;

[0129] S2.8, Iteration Termination Condition:

[0130] When the change in probability distribution between two adjacent iterations is less than a preset threshold, the result is considered to have converged and the iteration is stopped; otherwise, S2.4 to S2.7 can be repeated on the latest envelope surface to further refine the mesh.

[0131] In the prediction method of this invention, the virtual envelope may not necessarily be level 3, but it is at least level 3, based on the actual situation.

[0132] S3. Pilot development simulation based on hybrid path prediction model: The hybrid path prediction model is used to search along the path of potential decrease constraint and terminal field strength reward to find the next level virtual surface attachment point until the wind turbine surface attachment point, and obtain the pilot development path; and the electrode-multi-attachment point model is used to obtain the probability distribution characteristics.

[0133] A hybrid path prediction model is used to calculate the attachment point positions from the starting electrode to the next level virtual envelope surface to obtain a leader downlink path; an electrode-multi-attachment point model is used to calculate the multiple attachment point positions corresponding to one electrode to obtain probability distribution characteristics.

[0134] This invention uses a hybrid path prediction model to calculate one attachment point at a time. The attachment points obtained in each calculation are not consistent. Therefore, in order to reflect the probability, Monte Carlo sampling is required. After Monte Carlo sampling, one electrode and multiple attachment points can be obtained. Then, the electrode-multiple attachment point model is used to calculate the probability distribution characteristics.

[0135] The specific method is as follows:

[0136] S3.1 Data Extraction and Preprocessing:

[0137] Extract the three-dimensional electric potential and electric field mesh data between the starting point and the target envelope from the background library at the current level, and construct the electric potential array. and electric field strength array Among them, the potential array The potential of the envelope surface from the starting point to the target surface is monotonically decreasing;

[0138] S3.2 Determination of the starting point and candidate endpoint:

[0139] Starting point: The coordinates of the starting point of the current level, denoted as... ,in The set of all starting points;

[0140] Candidate endpoints: On the target envelope, candidate points are selected based on the local maxima of the electric field intensity (e.g., greater than 8 neighborhoods). , recorded as , The set of all candidate endpoints;

[0141] S3.3, Shortest Path Search with Constraints:

[0142] Treating the 3D mesh as a graph, with nodes as mesh points and edges connecting 26 neighboring nodes, for a given candidate endpoint... The search starts from the point of origin, under the condition of satisfying the potential decrease constraint. To the candidate endpoint Find the optimal path and calculate the path from the starting point. sm To the finish line tn Comprehensive indicators Select to make The minimum endpoint and its path are used as the prediction result;

[0143] The specific method is as follows:

[0144] Movement constraint: Movement is only allowed in the direction of decreasing potential, i.e.:

[0145] (1);

[0146] in, The current node potential, The potential of the surrounding 26 neighboring nodes;

[0147] The edge cost definition for shortest path search:

[0148] (2);

[0149] in For nodes electric field strength, For nodes electric field strength, The value should be a constant to avoid division by zero;

[0150] in The value is taken as 1 / 1000 or 10 of the minimum non-zero electric field strength in the background library. -6 The minimum non-zero field strength is times the field strength.

[0151] From the starting point a certain point in the middle Candidate set to the endpoint a certain point in the middle cumulative path cost between Cost of edges between all adjacent nodes on the path sum:

[0152] (3);

[0153] in, Accumulate cost for the path; P To start from the beginning To the finish line The path passes through i 10 nodes form a sequence ;

[0154] Introducing the finish line strength reward, the path is obtained. Comprehensive indicators :

[0155] (4);

[0156] in, End point t electric field strength, The weighting coefficient is determined through scaled-down experiments or historical data, and its value ranges from 0.1 to 0.5.

[0157] Use A Or Dijkstra's algorithm calculates from all starting sets to candidate endpoint set The optimal path is obtained, yielding the minimum Compare its comprehensive indicators ;

[0158] Select all The smallest endpoint and its path are used as the starting point set. to candidate endpoint set The prediction results are as follows: if the target of the current level is a virtual envelope surface, the end point of the path is recorded as the virtual attachment point of the next level; if the target is the wind turbine surface, the flash point is recorded.

[0159] In this context, "graph" in "viewed as a graph" refers to a connected data structure. In graph theory, a graph is defined as an ordered pair of... To support Dijkstra's algorithm and A For the algorithm to run, the graph usually needs to have the attribute of "weight".

[0160] Vertex set It represents a non-empty set, whose elements are called vertices (or nodes). In path planning, they represent points, intersections, or states on the map.

[0161] Edge set Represents a set containing either unordered pairs (undirected graph) or ordered pairs (directed graph). If two vertices... and If there is a direct connection, then .

[0162] In this invention, the weight refers to the cost defined by equation (2). .

[0163] S3.4, Competition among multiple candidate endpoints and pseudo-random statistical processing:

[0164] Considering that an electric arc originating from a starting point may eventually attach to multiple endpoints, and given the limited number of potential candidate endpoints (usually a few), a path search is performed separately for each candidate endpoint to calculate the optimal path cost to that endpoint. Then calculate the comprehensive index. ,choose The smallest endpoint and its path; where, The weighting coefficient is determined through scaled-down experiments or historical data, and its value ranges from 0.1 to 0.5. End point t The electric field strength;

[0165] The above process is a single deterministic path prediction. To obtain the probability distribution characteristics, Monte Carlo sampling of the background electric field library can be combined (e.g., randomly selecting different lightning current parameters, salt spray levels, or adding electric field perturbations), and S3.3 can be executed multiple times to obtain multiple possible attachment points corresponding to the same starting point, thus achieving pseudo-random processing similar to fractal calculation.

[0166] This invention proposes a hybrid path prediction model based on a hybrid path search method using decreasing potential constraints and electric field strength rewards. The three-dimensional space is discretized into a mesh graph. Monotonically decreasing potential is used as the movement constraint, the reciprocal of the electric field strength of adjacent nodes is used as the edge cost, and the field strength at the destination is introduced as a reward. The preliminary path prediction is transformed into a constrained multi-objective shortest path problem. The process of progressively dividing the envelope and finding the path is as follows: Figure 2 As shown.

[0167] This invention uses the Monte Carlo method to realistically simulate actual physical processes, making the problem as consistent with reality as possible, and thus obtaining very satisfactory results.

[0168] S4. Generation of lightning strike probability distribution: Multiple simulation results (including non-hit events) for each starting point in the multi-electrode grid are weighted and summarized to finally generate a lightning strike probability distribution cloud map on the wind turbine surface. Since the grid is densified in key areas and the density of starting points is uneven, starting point weights are introduced to correct sampling bias and ensure that the probability distribution reflects the lightning strike risk under a uniform distribution in real space.

[0169] The specific method is as follows:

[0170] S4.1 Input data:

[0171] Starting point set Each starting point Having spatial coordinates and the volume element it represents Volume element It can be calculated using Voronoi partitioning or grid rules;

[0172] Simulation results: For each starting point , conducted Sub-independent random Monte Carlo simulation. The simulation results for each simulation are recorded as follows:

[0173] : Whether the wind turbine was hit; where 1 indicates hit and 0 indicates no hit; if hit, record the coordinates of the lightning strike point on the wind turbine surface. ;

[0174] Among them, subscript Starting point Serial number marking, Number of simulations performed for this starting point mark, ;

[0175] S4.2, Surface element division of the fan:

[0176] The surface of the wind turbine (blades, nacelle, tower) is divided into... Individual Each facet element has geometric information (area). (such as center point coordinates), and establish a spatial index to quickly locate the surface element to which the flash point belongs;

[0177] S4.3 Weight Calculation:

[0178] The total volume of the sampling space is Each starting point The weights are:

[0179] (5);

[0180] Under the assumption of uniform distribution, weights Proportional to the starting point Having spatial coordinates and the volume element it represents The actual spatial proportion of the corresponding area;

[0181] S4.4 Weighted Statistics:

[0182] Iterate through all starting points and all simulations, accumulating the weighted hit count on each face element, and simultaneously accumulating the total weighted simulation count. :

[0183] (6);

[0184] in, As weight, The number of random Monte Carlo simulations corresponding to the weights;

[0185] S4.5 Calculate the probability of a surface element being struck by lightning:

[0186] A random Monte Carlo simulation (starting points are randomly and uniformly distributed in the sampling space) shows the leader eventually hitting a surface element. Lightning intercept probability estimate for:

[0187] (7);

[0188] This value satisfies ,margin The total probability corresponding to the pilot not hitting the wind turbine; among which is the surface element The total number of times hit in the simulation; is the total weighted simulation times of the pilot hitting in the random Monte Carlo simulation;

[0189] If the lightning strike probability density of the surface element (probability per unit area) is required, it can be further calculated as:

[0190] (8);

[0191] Among which, is the area of the surface element ; the subscript is the number of surface elements divided in the surface area of the wind turbine;

[0192] S4.6. Generate a cloud map:

[0193] Map the lightning strike probability or probability density of each surface element to colors. For example, the Jet / Rainbow color table can be used. By representing the change of probability through color change and rendering each surface element on the three-dimensional model of the wind turbine, the lightning strike probability distribution cloud map can be obtained. <​​​​​​​​​​​​​​​​​​​In this invention, the static electric field is pre-calculated: before the lightning path simulation begins, the spatial electric field distribution of the entire solution domain is calculated in one go with high precision for all possible operating conditions (wind turbine geometry, salt spray pollution level, lightning current waveform parameters), forming a "static electric field background library". All subsequent pilot development simulations can quickly obtain the electric field value at any location by looking up tables or interpolation, without needing to repeatedly solve for the electromagnetic field.

[0201] In this invention, multi-electrode grid sampling is used: a discrete "virtual electrode" grid is systematically arranged in the key spatial area above the wind turbine and on the windward side. Each grid node represents the starting point of the downlink leader of a simulation. By replacing pure random sampling with grid sampling, key coverage of high-risk areas is ensured.

[0202] Through the above strategy, this invention reduces the computational load to one-tenth of that of traditional fully random fractal methods while ensuring physical rationality, providing an efficient evaluation tool for the lightning protection design of offshore wind turbines.

[0203] Example 1

[0204] The specific method for assessing the probability of lightning strike on a 5 MW offshore wind turbine is as follows:

[0205] S1. Establish a 3D model of the wind turbine including blades, nacelle, and tower. The blade length is 70 meters. Set the equivalent surface conductivity on the blade surface: 10 for clean conditions. -12 S / m, light pollution is set to 10 -8 S / m, heavy pollution is set to 10. -6 S / m. The first return stroke waveform (30kA, 1 / 200μs) and subsequent return stroke waveform (10kA, 0.25 / 100μs) recommended by IEC 61400-24 standard were selected as typical lightning current parameters.

[0206] S2. Establish a first-level virtual envelope surface 500 meters above the wind turbine, generating a sparse electrode grid with a spacing of 80 meters (approximately 300 starting points). Use COMSOL to calculate the full-space electric field from 500 meters to the wind turbine surface, forming the first background library, such as... Figure 3 As shown.

[0207] S3. Perform first-level path prediction for each starting point, simulating the process of the leader evolving from a 500-meter envelope to a 50-meter envelope, and obtain a virtual attachment point set on the 50-meter envelope. Based on these attachment points, generate a second-level electrode grid with a spacing of 20 meters on the 50-meter envelope, such as... Figure 4 As shown.

[0208] S4. Narrow the computational area to 50 meters above the wind turbine surface and recalculate the local electric field to form a second background pool. Perform path prediction from the starting point of the second-level grid to obtain virtual attachment points on the 5-meter envelope surface, such as... Figure 5 As shown.

[0209] S5. Similarly, a third-level electrode grid (5-meter spacing) is generated on the 5-meter envelope surface. The local electric field (third background library) from 5 meters to the wind turbine surface is calculated. The final path prediction is then performed to obtain the actual flashover point, such as... Figure 6 As shown.

[0210] S6. Statistically analyze the final flash points and generate a probability distribution cloud map.

[0211] Based on the characteristics of the fan and the electrodes, when calculating the background library, only the points of interest (such as the flash pillar, copper mesh, protruding edges, etc.) need to be calculated, without having to calculate the entire machine.

[0212] Example 2

[0213] This embodiment uses a hybrid prediction method for the probability of lightning strikes on offshore wind turbines. The specific method is as follows:

[0214] S1. Construct a parameterized static potential and electric field background library;

[0215] S2. Multi-level iterative construction of electrode mesh and background library: n-level virtual envelope surfaces are set with the wind turbine as the center; sparse electrode meshes are laid out on the outermost layer, and the electrostatic field is solved to construct the background library; the path is predicted in the influence area to obtain the virtual attachment point of the next level envelope surface and use it as a new electrode. The computational domain is gradually reduced, and the solution and path prediction are repeated until the surface of the wind turbine is reached.

[0216] S3. Pilot development simulation based on hybrid path prediction model: The hybrid path prediction model is used to search along the path of potential decrease constraint and terminal field strength reward to find the next level virtual surface attachment point until the wind turbine surface attachment point, and obtain the pilot development path; and the electrode-multi-attachment point model is used to obtain the probability distribution characteristics.

[0217] S4. Generation of lightning probability distribution: Based on the weighted fusion of simulation results at each starting point of the multi-electrode grid, a cloud map of the lightning probability distribution on the surface of the wind turbine is formed.

[0218] Example 3

[0219] The hybrid prediction method for the probability of lightning strikes on offshore wind turbines in this embodiment, based on Embodiment 2, has the following specific method in S1:

[0220] S1.1 Establish a three-dimensional geometric model of the offshore wind turbine, including blades, lightning arresters, nacelle, and tower; for floating wind turbines, further establish the geometric configurations corresponding to different tilt angles and blade azimuth angles.

[0221] S1.2 Set material properties and environmental parameters. Specifically, seawater is set as an ideal conductor boundary, and equivalent surface conductivity is set on the blade surface to simulate different salt spray pollution levels, which are divided into clean, lightly polluted and heavily polluted.

[0222] S1.3 Set a set of typical lightning current waveform parameters, including amplitude, wavefront time and half-peak time;

[0223] S1.4. For different parameter combinations in steps S1.1 to S1.3, use the electrostatic field solver to calculate the spatial potential distribution of the entire solution domain. and electric field distribution This forms a parameterized library of static potential and electric field backgrounds.

[0224] Example 4

[0225] The hybrid prediction method for the probability of lightning strikes on offshore wind turbines in this embodiment, based on Embodiment 2, has the following specific method in S2:

[0226] S2.1 Define a multi-level virtual envelope surface:

[0227] Centered on the wind turbine, and according to the general distance scale of pilot development, multiple levels of virtual envelope surfaces are set up. These virtual envelope surfaces divide the space into n levels, where n≥3, and each level corresponds to a stage of pilot development.

[0228] The multi-level virtual envelope surface set in this embodiment specifically includes a far-field envelope surface, a mid-field envelope surface, and a near-field envelope surface;

[0229] For example: the far-field envelope is 500 meters away from the wind turbine surface, the mid-field envelope is 50 meters away from the wind turbine surface, and the near-field envelope is 5 meters away from the wind turbine surface.

[0230] S2.2 Construct the first-level mesh and background library (e.g., 500-meter level):

[0231] A sparse starting point grid (e.g., grid spacing of 80 meters) is generated on a 500-meter envelope surface, with each grid node representing a downlink leader starting point; for all possible operating conditions (wind turbine attitude, salt spray pollution, lightning current waveform), the electric field distribution of the entire space from the envelope surface to the wind turbine surface is calculated using an electrostatic field solver to form the first background library;

[0232] Since the electric field changes gradually when the electrode is 500 meters away from the wind turbine, a coarser grid can be used for this stage of calculation.

[0233] S2.3, First-level path prediction and virtual attachment point generation:

[0234] Starting from each starting point, a hybrid path prediction model is used to simulate the process of the leader evolving from a 500-meter envelope to a 50-meter envelope. The endpoint of the path prediction is not the wind turbine surface, but a point on the 50-meter envelope. The endpoints of all paths on the 50-meter envelope constitute a set of virtual attachment points, which represent the intermediate positions that the leader may reach. Connecting all discrete grid nodes on the path generates the first-level leader path set.

[0235] S2.4, Second-level grid and background library (50-meter level):

[0236] Based on the virtual attachment points generated in the first stage, the mesh is refined on the 50-meter envelope surface (e.g., the mesh spacing is 20 meters) to generate new starting points. At this time, the calculation area is reduced to the space from the 50-meter envelope surface to the wind turbine surface. Only the area from the 50-meter envelope surface to the wind turbine surface needs to be divided into high-resolution meshes and the electric field distribution needs to be recalculated to form the second background library.

[0237] Because the area is reduced when the electrode is 50m away from the wind turbine, a medium grid is used for this stage of calculation.

[0238] S2.5, Second-level path prediction and generation of the next-level virtual attachment point:

[0239] Starting from each new starting point on the 50-meter virtual envelope, path prediction is performed again to simulate the process of the leader evolving to the 5-meter virtual envelope, and new virtual attachment points are generated on the 5-meter virtual envelope.

[0240] S2.6, nth level mesh and background library (5-meter level):

[0241] Based on the virtual attachment points generated in the second stage, the starting point mesh is generated by densifying the mesh on the 5-meter virtual envelope surface. The local electric field from the 5-meter virtual envelope surface to the wind turbine surface is calculated to form the third background library.

[0242] Since the area is further reduced when the electrode is 5m away from the wind turbine, it is also possible to select the local structures of the wind turbine that are of particular interest (such as the blade tip, the blade containing the lightning rod, etc.) for calculation. This level of calculation uses a finer grid.

[0243] S2.7 Final Path Prediction:

[0244] The final path prediction is performed from the starting point on the 5-meter envelope surface, directly calculating the flashover point on the wind turbine surface;

[0245] S2.8, Iteration Termination Condition:

[0246] When the change in probability distribution between two adjacent iterations is less than a preset threshold, the result is considered to have converged and the iteration is stopped; otherwise, S2.4 to S2.7 can be repeated on the latest envelope surface to further refine the mesh.

[0247] Example 5

[0248] The hybrid prediction method for the probability of lightning strikes on offshore wind turbines in this embodiment, based on Embodiment 2, has the following specific method in S3:

[0249] S3.1 Data Extraction and Preprocessing:

[0250] Extract the three-dimensional electric potential and electric field mesh data between the starting point and the target envelope from the background library at the current level, and construct the electric potential array. and electric field strength array Among them, the potential array The potential of the envelope surface from the starting point to the target surface is monotonically decreasing;

[0251] S3.2 Determination of the starting point and candidate endpoint:

[0252] Starting point: The coordinates of the starting point of the current level, denoted as... ,in The set of all starting points;

[0253] Candidate endpoints: On the target envelope, candidate points are selected based on the local maxima of the electric field intensity (e.g., greater than 8 neighborhoods). , recorded as , The set of all candidate endpoints;

[0254] S3.3, Shortest Path Search with Constraints:

[0255] Treating the 3D mesh as a graph, with nodes as mesh points and edges connecting 26 neighboring nodes, for a given candidate endpoint... The search starts from the point of origin, under the condition of satisfying the potential decrease constraint. To the candidate endpoint Find the optimal path and calculate the path from the starting point. sm To the finish line tn Comprehensive indicators Select to make The minimum endpoint and its path are used as the prediction result;

[0256] The specific method is as follows:

[0257] Movement constraint: Movement is only allowed in the direction of decreasing potential, i.e.:

[0258] (1);

[0259] in, The current node potential, The potential of the surrounding 26 neighboring nodes;

[0260] The edge cost definition for shortest path search:

[0261] (2);

[0262] in For nodes electric field strength, For nodes electric field strength, The value should be a constant to avoid division by zero;

[0263] in The value is taken as 1 / 1000 or 10 of the minimum non-zero electric field strength in the background library. -6 The minimum non-zero field strength is times the field strength.

[0264] From the starting point a certain point in the middle Candidate set to the endpoint a certain point in the middle cumulative path cost between Cost of edges between all adjacent nodes on the path sum:

[0265] (3);

[0266] in, Accumulate cost for the path; P To start from the beginning To the finish line The path passes through i 10 nodes form a sequence ;

[0267] Introducing the finish line strength reward, the path is obtained. Comprehensive indicators :

[0268] (4);

[0269] in, End point t electric field strength, The weighting coefficient is determined through scaled-down experiments or historical data, and its value ranges from 0.1 to 0.5.

[0270] Use A Or Dijkstra's algorithm calculates from all starting sets to candidate endpoint set The optimal path is obtained, yielding the minimum Compare its comprehensive indicators ;

[0271] Select all The smallest endpoint and its path are used as the starting point set. to candidate endpoint set The prediction results are as follows: if the target of the current level is a virtual envelope surface, the end point of the path is recorded as the virtual attachment point of the next level; if the target is the wind turbine surface, the flash point is recorded.

[0272] S3.4, Competition among multiple candidate endpoints and pseudo-random statistical processing:

[0273] Considering that an electric arc originating from a starting point may eventually attach to multiple endpoints, and given the limited number of potential candidate endpoints (usually a few), a path search is performed separately for each candidate endpoint to calculate the optimal path cost to that endpoint. Then calculate the comprehensive index. ,choose The smallest endpoint and its path; where, The weighting coefficient is determined through scaled-down experiments or historical data, and its value ranges from 0.1 to 0.5. End point t The electric field strength;

[0274] The above process is a single deterministic path prediction. To obtain the probability distribution characteristics, Monte Carlo sampling of the background electric field library can be combined (e.g., randomly selecting different lightning current parameters, salt spray levels, or adding electric field perturbations), and S3.3 can be executed multiple times to obtain multiple possible attachment points corresponding to the same starting point, thus achieving pseudo-random processing similar to fractal calculation.

[0275] Example 6

[0276] The hybrid prediction method for the probability of lightning strikes on offshore wind turbines in this embodiment, based on embodiment 2, has the following specific method in S4:

[0277] S4.1 Input data:

[0278] Starting point set Each starting point Having spatial coordinates and the volume element it represents Volume element It can be calculated using Voronoi partitioning or grid rules;

[0279] Simulation results: For each starting point , conducted Sub-independent random Monte Carlo simulation. The simulation results for each simulation are recorded as follows:

[0280] : Whether the wind turbine was hit; where 1 indicates hit and 0 indicates no hit; if hit, record the coordinates of the lightning strike point on the wind turbine surface. ;

[0281] Among them, subscript Starting point Serial number marking, Number of simulations performed for this starting point mark, ;

[0282] S4.2, Surface element division of the fan:

[0283] The surface of the wind turbine (blades, nacelle, tower) is divided into... Individual Each facet element has geometric information (area). (such as center point coordinates), and establish a spatial index to quickly locate the surface element to which the flash point belongs;

[0284] S4.3 Weight Calculation:

[0285] The total volume of the sampling space is Each starting point The weights are:

[0286] (5);

[0287] Under the assumption of uniform distribution, weights Proportional to the starting point Having spatial coordinates and the volume element it represents The actual spatial proportion of the corresponding area;

[0288] S4.4 Weighted Statistics:

[0289] Iterate through all starting points and all simulations, accumulating the weighted hit count on each face element, and simultaneously accumulating the total weighted simulation count. :

[0290] (6);

[0291] in, As weight, The number of random Monte Carlo simulations corresponding to the weights;

[0292] S4.5 Calculate the probability of a surface element being struck by lightning:

[0293] A random Monte Carlo simulation (starting points are randomly and uniformly distributed in the sampling space) shows the leader eventually hitting a surface element. Lightning intercept probability estimate for:

[0294] (7);

[0295] This value satisfies ,margin The total probability that the pilot misses the wind turbine; where For face element The total number of times the target was hit in the simulation; The total weighted number of simulations for the leading hit in random Monte Carlo simulations;

[0296] If the probability density of surface element lightning is required The probability per unit area can be further calculated:

[0297] (8);

[0298] in, For face element Area; subscript The number of facets to divide the surface area of ​​the fan;

[0299] S4.6, Generate a cloud map:

[0300] The probability of each facet receiving lightning. or probability density Mapping to colors, for example using the Jet / Rainbow color table, allows color variations to represent changes in probability. By rendering each facet on the 3D model of the wind turbine, a cloud map of the lightning probability distribution can be obtained.

Claims

1. A hybrid prediction method for the probability of lightning strikes on offshore wind turbines, characterized in that, The specific method is as follows: S1. Construct a parameterized static potential and electric field background library; S2. Multi-level iterative construction of electrode mesh and background library: n-level virtual envelope surfaces are set with the wind turbine as the center; sparse electrode meshes are laid out on the outermost layer, and the electrostatic field is solved to construct the background library; the path is predicted in the influence area to obtain the virtual attachment point of the next level envelope surface and use it as a new electrode. The computational domain is gradually reduced, and the solution and path prediction are repeated until the surface of the wind turbine is reached. S3. Pilot development simulation based on hybrid path prediction model: The hybrid path prediction model is used to search along the path of potential decrease constraint and terminal field strength reward to find the next level virtual surface attachment point until the wind turbine surface attachment point, and obtain the pilot development path; and the electrode-multi-attachment point model is used to obtain the probability distribution characteristics. S4. Generation of lightning probability distribution: Based on the weighted fusion of simulation results at each starting point of the multi-electrode grid, a cloud map of the lightning probability distribution on the surface of the wind turbine is formed. The specific method of S3 is as follows: S3.1 Data Extraction and Preprocessing: Extract the three-dimensional electric potential and electric field mesh data between the starting point and the target envelope from the background library at the current level, and construct the electric potential array. and electric field strength array Among them, the potential array The potential of the envelope surface from the starting point to the target surface is monotonically decreasing; S3.2 Determination of the starting point and candidate endpoint: Starting point: The coordinates of the starting point of the current level, denoted as... ,in The set of all starting points; Candidate endpoints: On the target envelope, candidate points are selected based on the local maxima of the electric field intensity. , recorded as , The set of all candidate endpoints; S3.3, Shortest Path Search with Constraints: Treating the 3D mesh as a graph, with nodes as mesh points and edges connecting 26 neighboring nodes, for a given candidate endpoint... The search starts from the point of origin, under the condition of satisfying the potential decrease constraint. To the candidate endpoint Find the optimal path and calculate the path from the starting point. sm To the finish line tn Comprehensive indicators Select to make The minimum endpoint and its path are used as the prediction result; S3.4, Competition among multiple candidate endpoints and pseudo-random statistical processing: For each candidate destination, a path search is performed once to calculate the optimal path cost to that destination. Then calculate the comprehensive index. ,choose The smallest endpoint and its path; where, The weighting coefficient is determined through scaled-down experiments or historical data, and its value ranges from 0.1 to 0.

5. End point t The electric field strength; To obtain the probability distribution characteristics, Monte Carlo sampling of the background electric field library is combined with repeated execution of S3.3 to obtain multiple possible attachment points corresponding to the same starting point, thus achieving pseudo-random processing similar to fractal calculation.

2. The hybrid prediction method for the probability of lightning strikes on offshore wind turbines according to claim 1, characterized in that, The specific method of S1 is as follows: S1.1 Establish a three-dimensional geometric model of the offshore wind turbine, including blades, lightning arresters, nacelle, and tower; for floating wind turbines, further establish the geometric configurations corresponding to different tilt angles and blade azimuth angles. S1.2 Set material properties and environmental parameters. Specifically, seawater is set as an ideal conductor boundary, and equivalent surface conductivity is set on the blade surface to simulate different salt spray pollution levels, which are divided into clean, lightly polluted and heavily polluted. S1.3 Set a set of typical lightning current waveform parameters, including amplitude, wavefront time and half-peak time; S1.

4. For different parameter combinations in steps S1.1 to S1.3, use the electrostatic field solver to calculate the spatial potential distribution of the entire solution domain. and electric field distribution This forms a parameterized library of static potential and electric field backgrounds.

3. The hybrid prediction method for the lightning strike probability of offshore wind turbines according to claim 1, characterized in that, The specific method of S2 is as follows: S2.1 Define a multi-level virtual envelope surface: Centered on the wind turbine, multiple virtual envelope surfaces are set up, which divide the space into n levels, where n≥3, and each level corresponds to a stage of pioneering development; S2.2, Construct the first-level mesh and background library: A sparse starting point grid is generated on the first-level envelope surface, with each grid node representing a downward leader starting point; for all possible operating conditions, the electric field distribution of the entire space from the envelope surface to the fan surface is calculated using an electrostatic field solver to form the first background library; S2.3, First-level path prediction and virtual attachment point generation: Starting from each starting point, a hybrid path prediction model is used to simulate the process of the leader evolving from the first-level envelope to the second-level envelope. The endpoint of the path prediction is not the wind turbine surface, but a point on the second-level envelope. The endpoints of all paths on the second-level envelope constitute a set of virtual attachment points, which connect all discrete grid nodes on the path to generate the first-level leader path set. S2.4, Second-level mesh and background library: Based on the virtual attachment points generated in the first level, the mesh is refined on the second-level envelope surface to generate new starting points. At this time, the computational domain is reduced to the space from the second-level envelope surface to the wind turbine surface. Only the region from the second-level envelope surface to the wind turbine surface needs to be meshed with high resolution to recalculate the electric field distribution and form the second background library. S2.5, Second-level path prediction and generation of the next-level virtual attachment point: Starting from each new starting point on the second-level virtual envelope, path prediction is performed again to simulate the process of the leader evolving to the third-level virtual envelope, and new virtual attachment points are generated on the third-level virtual envelope. S2.6, nth level mesh and background library: Based on the virtual attachment points generated at level n, the starting point mesh is generated by densifying the virtual envelope surface at level n, and the local electric field from the virtual envelope surface at level n to the wind turbine surface is calculated to form the background library at level n. S2.7 Final Path Prediction: The final path prediction is performed from the starting point on the nth level envelope surface, directly calculating the flash point on the wind turbine surface; S2.8, Iteration Termination Condition: When the change in probability distribution between two adjacent iterations is less than a preset threshold, the result is considered to have converged and the iteration is stopped; otherwise, S2.4~S2.7 are repeated on the latest envelope surface to further refine the mesh.

4. The hybrid prediction method for the probability of lightning strikes on offshore wind turbines according to claim 3, characterized in that, The specific method of S3.3 is as follows: Movement constraint: Movement is only allowed in the direction of decreasing potential, i.e.: （1）； in, The current node potential, The potential of the surrounding 26 neighboring nodes; The edge cost definition for shortest path search: （2）； in For nodes electric field strength, For nodes electric field strength, It is a constant; From the starting point a certain point in the middle Candidate set to the endpoint a certain point in the middle cumulative path cost between Cost of edges between all adjacent nodes on the path sum: （3）； in, Accumulate cost for the path; P To start from the beginning To the finish line The path passes through i 10 nodes form a sequence ; Introducing the finish line strength reward, the path is obtained. Comprehensive indicators : （4）； in, End point t electric field strength, The weighting coefficient is determined through scaled-down experiments or historical data, and its value ranges from 0.1 to 0.

5. Compute using A* or Dijkstra's algorithm from all starting sets. to candidate endpoint set The optimal path is obtained, which minimizes... Compare its comprehensive indicators ; Select all The smallest endpoint and its path are used as the starting point set. to candidate endpoint set The prediction results are as follows: if the target of the current level is a virtual envelope surface, the end point of the path is recorded as the virtual attachment point of the next level; if the target is the wind turbine surface, the flash point is recorded.

5. The hybrid prediction method for the lightning strike probability of offshore wind turbines according to claim 4, characterized in that, In formula (2), ε is taken as 1 / 1000 or 10 of the minimum non-zero electric field intensity in the background library. -6 The minimum non-zero field strength is times the field strength.

6. The hybrid prediction method for the lightning strike probability of offshore wind turbines according to claim 5, characterized in that, The specific method of S4 is as follows: S4.1 Input data: Starting point set Each starting point Having spatial coordinates and the volume element it represents Volume element Calculated using Voronoi partitioning or grid rules; Simulation results: For each starting point , conducted Sub-independent random Monte Carlo simulation. The simulation results for each simulation are recorded as follows: : Whether the wind turbine was hit; where 1 indicates hit and 0 indicates no hit; if hit, record the coordinates of the lightning strike point on the wind turbine surface. ; Among them, subscript Starting point Serial number marking, Number of simulations performed for this starting point mark, ; S4.2, Surface element division of the fan: The surface of the fan is divided into Individual Each element has geometric information, and a spatial index is established to quickly locate the element to which the flash point belongs. S4.3 Weight Calculation: The total volume of the sampling space is Each starting point The weights are: （5）； Under the assumption of uniform distribution, weights Proportional to the starting point Having spatial coordinates and the volume element it represents The actual spatial proportion of the corresponding area; S4.4 Weighted Statistics: Iterate through all starting points and all simulations, accumulating the weighted hit count on each face element, and simultaneously accumulating the total weighted simulation count. : （6）； in, As weight, The number of random Monte Carlo simulations corresponding to the weights; S4.5 Calculate the probability of a surface element being struck by lightning: A random Monte Carlo simulation pilot ultimately hits a surface element. Lightning intercept probability estimate for: （7）； This value satisfies ,margin The total probability that the pilot misses the wind turbine; where For face element The total number of times the target was hit in the simulation; The total weighted number of simulations for the leading hit in random Monte Carlo simulations; If the probability density of surface element lightning is required Further calculations: （8）； in, For face element Area; subscript The number of face cells to divide the surface area of ​​the fan; S4.6, Generate a cloud map: The probability of each facet receiving lightning. or probability density The color is mapped and each surface element is rendered on the 3D model of the wind turbine to obtain the lightning probability distribution cloud map.

7. A hybrid prediction system for the lightning strike probability of offshore wind turbines using the prediction method described in any one of claims 1 to 6, characterized in that, It includes a static electric field background library construction module, a multi-level iterative grid generation module, a hybrid path prediction module, and a lightning strike probability generation module; The static electric field background library construction module is used to construct a parameterized static potential and electric field background library; The multi-level iterative mesh generation module is used to construct electrode meshes and background libraries through multi-level iterations. The hybrid path prediction module is used to perform pilot development simulations of a prediction model based on hybrid paths; The lightning strike probability generation module is used to generate the lightning strike probability distribution.

Images