A fan blade lightning strike risk early warning method and device based on artificial intelligence

Abstract

The application discloses a fan blade lightning stroke risk early warning method and device based on artificial intelligence, relates to the technical field of fan lightning protection early warning, and comprises the following steps: collecting real-time meteorological data, lightning monitoring data and blade operation state data of a wind power plant area, and constructing a multi-source feature data matrix; based on preset blade geometric parameters, a blade surface electric field distribution model is established, and the lightning stroke sensitivity coefficient of each blade is calculated; the multi-source feature data matrix and the lightning stroke sensitivity coefficient are input into the constructed deep neural network model, and the lightning stroke risk grade and lightning stroke probability prediction value of each fan are output; when the lightning stroke probability prediction value is greater than a preset risk threshold value, an early warning signal is generated and sent to a fan control system, and according to the lightning stroke risk grade, a preset coping strategy library is matched, a target pitch angle adjustment amount and a target rotating speed adjustment amount are acquired, and the fan is controlled to execute a blade lightning protection action. The application improves the initiative and safety of fan lightning protection.

Description

Technical Field

[0001] This invention relates to the field of wind turbine lightning protection early warning technology, and in particular to a method and device for early warning of lightning strike risk on wind turbine blades based on artificial intelligence. Background Technology

[0002] With the continuous growth of wind power installed capacity and the improvement of single-unit power ratings, lightning protection safety of wind turbines operating in complex meteorological environments has gradually become one of the key factors affecting the stable operation of wind farms. Especially in high-altitude, high-humidity, and strong convection areas, the increased length, composite materials, and complex internal structure of wind turbine blades significantly enhance their response characteristics to electromagnetic fields, leading to a substantial increase in the probability of lightning strikes. Traditional lightning protection designs typically achieve passive protection through methods such as setting metal receptors, conductors, or grounding grids. However, under variable meteorological conditions and dynamic operating states, the electric field distribution on the blade surface changes rapidly with variations in pitch angle, rotational speed, and azimuth angle, making accurate prediction and active adjustment difficult. Furthermore, existing lightning strike monitoring methods are mostly based on lightning current signal detection or acoustic identification of blade damage, lacking a pre-emptive intelligent assessment and response decision-making mechanism for lightning strike risks, resulting in insufficient early warning timeliness, delayed response, and high damage risk. With the development of artificial intelligence and multi-source data fusion analysis technology, how to correlate and model meteorological, lightning and blade operation data to achieve real-time prediction and intelligent response to lightning strike risks has become an important research direction in the field of wind power safety monitoring.

[0003] CN107729680B discloses a method for assessing the probability of lightning strikes on wind turbine blades, belonging to the field of wind power lightning protection technology. This method addresses the issue of traditional analyses neglecting upward lightning strikes by proposing the use of the finite element method to calculate the electric field distribution of the wind turbine under the influence of thunderclouds and downward leaders. It also determines whether the sampling points on the wind turbine blade surface form an initial upward leader during the development of the downward leader, thereby establishing a blade lightning strike probability assessment model to calculate the probability of lightning strikes at any location. This technology can accurately assess the lightning strike probability distribution on the blade surface under static conditions, providing a theoretical basis for lightning protection system design. However, this scheme mainly focuses on the static electric field distribution relationship between the thundercloud and wind turbine system, without incorporating real-time wind turbine operating parameters (such as pitch angle and speed changes) and dynamic meteorological monitoring data, nor combining deep learning or multi-source feature fusion mechanisms for adaptive prediction. Therefore, it is difficult to meet the needs of rapid risk assessment and early warning in complex operating environments.

[0004] CN111306010B discloses a method and system for detecting lightning damage to wind turbine blades. This method collects sound signals from the wind turbine to determine the presence of lightning and further extracts the sweeping sound signals from the blades for energy analysis, thereby identifying whether the blades have suffered lightning damage. This technology can, to some extent, achieve real-time detection of blade damage after a lightning strike, avoiding safety hazards caused by operating with damage. However, this solution falls under the category of post-event detection, only identifying damage after a lightning strike occurs, and cannot provide pre-event lightning risk warnings or adjust operational strategies. Furthermore, its monitoring features are limited to a single dimension, failing to involve the fusion modeling of multi-source meteorological data, lightning activity parameters, and blade electric field characteristics, lacking in-depth analysis and prediction capabilities regarding the risk formation mechanism.

[0005] In summary, existing wind turbine blade lightning strike assessment and detection technologies mostly focus on static probability modeling or post-incident damage identification, lacking dynamic prediction and proactive defense mechanisms based on real-time operating status, meteorological changes, and lightning monitoring data. This makes it difficult to achieve accurate quantification and intelligent response to lightning strike risks. This invention proposes an artificial intelligence-based wind turbine blade lightning strike risk early warning method, which effectively improves the proactive prevention capabilities and operational safety of wind turbine lightning protection systems. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0007] Given that existing wind turbine blade lightning protection methods mostly rely on static structural design and post-event testing, making it difficult to achieve real-time prediction and proactive response to lightning strike risks under complex weather and operating conditions, this invention is proposed.

[0008] Therefore, the problem to be solved by this invention is how to achieve intelligent prediction and early warning control of lightning strike risk on wind turbine blades based on multi-source meteorological and operational data and combined with artificial intelligence models, thereby improving the initiative and safety of wind turbine lightning protection.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a method for early warning of lightning strike risk on wind turbine blades based on artificial intelligence, comprising, Collect real-time meteorological data, lightning monitoring data, and blade operating status data of the wind farm area to construct a multi-source feature data matrix; Based on the blade operating status data and preset blade geometric parameters, a blade surface electric field distribution model is established, and the lightning susceptibility coefficient of each blade is calculated. The multi-source feature data matrix and the lightning sensitivity coefficient are input into the constructed deep neural network model, which outputs the lightning risk level and lightning probability prediction value of each wind turbine. When the predicted probability of a lightning strike exceeds the preset risk threshold, an early warning signal is generated and sent to the wind turbine control system. At the same time, the system matches the preset response strategy library according to the lightning strike risk level, obtains the target pitch angle adjustment and the target speed adjustment, and controls the wind turbine to perform blade lightning protection actions.

[0010] As a preferred embodiment of the artificial intelligence-based wind turbine blade lightning strike risk early warning method of the present invention, the method for establishing the electric field distribution model on the blade surface is as follows: Obtain the preset blade geometric parameters and establish a three-dimensional geometric model of the blade; Based on the spatial azimuth angle of the blades, the spatial attitude of each blade at the current moment is determined, and the three-dimensional geometric model of the blades is spatially transformed according to the spatial attitude to obtain the real-time spatial position model of each blade. Extract the topographic elevation data and surrounding building height data of the wind farm area, construct a three-dimensional terrain model of the wind farm environment, and embed the real-time spatial location model into the three-dimensional terrain model of the wind farm environment to obtain the blade-environment coupling model; By setting parameters for thundercloud height and thundercloud charge density, and using the finite element method to calculate the spatial electric field distribution based on the blade-environment coupling model, the spatial electric field distribution includes the electric field intensity value and electric field direction vector of each grid node on the blade surface.

[0011] As a preferred embodiment of the artificial intelligence-based wind turbine blade lightning strike risk early warning method of the present invention, the deep neural network model includes a temporal feature extraction branch and a static feature extraction branch; the temporal feature extraction branch adopts a long short-term memory network structure, including an input layer, a first LSTM hidden layer, a second LSTM hidden layer, and a temporal feature output layer; the static feature extraction branch adopts a fully connected neural network structure, including an input layer, a first fully connected hidden layer, a second fully connected hidden layer, and a static feature output layer.

[0012] As a preferred embodiment of the artificial intelligence-based wind turbine blade lightning strike risk early warning method of the present invention, the method includes: acquiring the target pitch angle adjustment and the target rotational speed adjustment, and controlling the wind turbine to perform blade lightning protection actions, including: Traverse the list of high-risk wind turbines, extract the turbine number, maximum lightning strike probability prediction value and final lightning strike risk level of each wind turbine, determine the early warning triggering conditions and generate graded early warning signals; The graded early warning signals are encapsulated into early warning data packets and sent to the wind turbine control system via a communication protocol. Based on the lightning strike risk level, the system queries the preset response strategy library to find the matching basic response strategy and extracts the pitch angle increase coefficient and speed reduction coefficient. Based on the lightning strike sensitivity coefficient, identify the blade with the highest lightning strike sensitivity and calculate the target lightning protection position angle for this blade to reach the lightning protection safety position; The target pitch angle adjustment is calculated based on the pitch angle amplification factor and the current pitch angle; simultaneously, the target speed adjustment is calculated based on the speed reduction factor and the current impeller speed. Generate a sequence of control commands for blade lightning protection actions based on the target lightning protection position angle, the target pitch angle adjustment, and the target rotational speed adjustment. The wind turbine control system executes each control command sequentially according to the order of the blade lightning protection action control command sequence, guiding the blade with the highest lightning strike sensitivity to rotate to the target lightning protection position angle, thus completing the blade lightning protection action.

[0013] As a preferred embodiment of the artificial intelligence-based wind turbine blade lightning strike risk early warning method of the present invention, wherein: the method for obtaining the lightning strike sensitivity coefficient is as follows: The blade surface is divided into K discrete segments along the radial direction of the three-dimensional geometric model of the blade. The electric field intensity values ​​of the corresponding grid nodes of each discrete segment are extracted, and the mean electric field intensity of each discrete segment is calculated. Based on the mean electric field strength, the lightning attachment probability of each discrete segment is calculated using a preset lightning strike leader development model. For each discrete section, based on the lightning strike attachment probability and the distance of the discrete section from the blade tip, the local lightning strike risk index is calculated, and the local lightning strike risk index of each discrete section is weighted and summed to obtain the lightning strike sensitivity coefficient.

[0014] As a preferred embodiment of the artificial intelligence-based wind turbine blade lightning strike risk early warning method of the present invention, wherein: the method for obtaining the lightning strike risk level and the predicted lightning strike probability value is as follows: Temporally relevant features are extracted from a multi-source feature data matrix to form a temporal feature input tensor. The input tensor is then used to input the temporal feature extraction branch, and a hidden state vector is output. Simultaneously, static features at the current moment are extracted from the multi-source feature data matrix to form a static feature input vector, which is then input into the static feature extraction branch to output a feature vector. The attention weights for temporal and static features are calculated through an attention fusion layer. The hidden state vector and the feature vector are then normalized and weighted and fused to obtain a fused feature vector. The fused feature vector is input into a dual-output task layer, wherein the dual-output task layer includes a probability prediction output head and a risk level classification output head; The probability prediction output head is used to calculate the predicted value of lightning strike probability, and the risk level output head is used to classify the lightning strike risk level.

[0015] As a preferred embodiment of the artificial intelligence-based wind turbine blade lightning strike risk early warning method of the present invention, the blade lightning protection action control command sequence includes three stages of control commands: the first stage is a speed reduction command, wherein the speed reduction command adjusts the impeller speed according to the target speed adjustment amount within the response time parameter; the second stage is a pitch angle increase command, wherein the pitch angle increase command adjusts the pitch angle according to the target pitch angle adjustment amount within the response time parameter; the third stage is a blade position adjustment command, wherein the blade position adjustment command guides the blade with the highest lightning strike sensitivity to rotate to the target lightning protection position angle by fine-tuning the impeller speed.

[0016] Secondly, embodiments of the present invention provide an artificial intelligence-based wind turbine blade lightning strike risk early warning device, which includes: The multi-source data acquisition module is used to collect real-time meteorological data, lightning monitoring data and blade operating status data in the wind farm area, and to construct a multi-source feature data matrix. The blade electric field module establishes a blade surface electric field distribution model based on the blade operating status data and preset blade geometric parameters, and calculates the lightning sensitivity coefficient of each blade. The lightning strike risk prediction module is used to input the multi-source feature data matrix and the lightning strike sensitivity coefficient into the constructed deep neural network model, and output the lightning strike risk level and lightning strike probability prediction value of each wind turbine. The early warning and control execution module generates an early warning signal and sends it to the wind turbine control system when the predicted lightning strike probability value is greater than the preset risk threshold. At the same time, it matches the preset response strategy library according to the lightning strike risk level, obtains the target pitch angle adjustment amount and the target speed adjustment amount, and controls the wind turbine to perform blade lightning protection actions.

[0017] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any step of the above-described artificial intelligence-based wind turbine blade lightning risk early warning method.

[0018] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the above-described artificial intelligence-based wind turbine blade lightning strike risk warning method.

[0019] Compared with existing technologies, the advantages of this invention are as follows: By collecting real-time meteorological data, lightning monitoring data, and blade operating status data from wind farms, a multi-source feature data matrix is ​​constructed, achieving comprehensive capture of factors influencing lightning strike risk; a surface electric field distribution model is established based on blade operating status and geometric parameters, and a lightning strike sensitivity coefficient is calculated. This combines electromagnetic field theory with the real-time status of the wind turbine, quantifying the electric field concentration under different attitudes using the finite element method and converting it into lightning strike attachment probability, establishing a hybrid intelligent mechanism that integrates physical and data-driven approaches, thus shortening the early warning response time; and integrating multi-source data and sensitive... The degree coefficient input adopts a deep neural network with a dual-branch architecture and attention fusion mechanism. It captures temporal evolution features through LSTM, processes static features through a fully connected network, and realizes multi-task joint learning of probability prediction and hierarchical classification. When the lightning strike probability exceeds the threshold, it matches the response strategy according to the risk level and identifies the blades most vulnerable to lightning strikes. Through a three-stage control sequence of reducing the rotational speed, increasing the pitch angle, and precisely guiding the highly sensitive blades to a safe position, it realizes closed-loop control from passive monitoring to active protection, thereby reducing the probability of lightning strike attachment, reducing the blade damage accident rate, and reducing the economic loss of a single lightning strike. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart of an artificial intelligence-based method for early warning of lightning strike risks on wind turbine blades. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0022] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] As mentioned in the background section, existing wind turbine blade lightning strike assessment and detection technologies mostly focus on static probability modeling or post-incident damage identification, lacking dynamic prediction and proactive defense mechanisms based on real-time operating status, meteorological changes, and lightning monitoring data. This makes it difficult to achieve accurate quantification and intelligent response to lightning strike risks. To address these issues, this invention provides an artificial intelligence-based method for early warning of lightning strike risks in wind turbine blades.

[0025] Reference Figure 1 , Figure 1 This is a flowchart illustrating an artificial intelligence-based early warning method for lightning strike risks on wind turbine blades, according to an embodiment of the present invention. Figure 1 As shown, an artificial intelligence-based method for early warning of lightning strike risks on wind turbine blades includes: S1: Collect real-time meteorological data, lightning monitoring data and blade operation status data of the wind farm area to construct a multi-source feature data matrix; Specifically, the wind farm central monitoring system establishes communication connections with the meteorological monitoring station, lightning location system, and wind turbine SCADA system, respectively, with a data acquisition cycle of 1 second. The meteorological monitoring station transmits meteorological monitoring information through the Modbus-TCP protocol; the lightning location system transmits lightning location information through the Ethernet interface; and the wind turbine SCADA system transmits wind turbine operation information through the Profibus-DP protocol.

[0026] S1.1: Collect real-time meteorological data, lightning monitoring data, and blade operating status data from meteorological monitoring stations, lightning location systems, and wind turbine SCADA systems respectively, and obtain a set of meteorological characteristic parameters, a lightning activity characteristic vector, and a set of blade spatial position parameters; Specifically, real-time meteorological data includes ambient temperature, relative humidity, atmospheric pressure, wind speed, wind direction angle, precipitation intensity, and visibility in the wind farm area; atmospheric dew point temperature is calculated based on ambient temperature and relative humidity; atmospheric density is calculated based on atmospheric pressure and ambient temperature, and real-time meteorological data, atmospheric dew point temperature, and atmospheric density are used as a set of meteorological characteristic parameters.

[0027] Furthermore, the lightning monitoring data includes the time of lightning occurrence, lightning location coordinates, lightning intensity, lightning type identifier, and lightning polarity identifier. A plane coordinate system is established with the center of the wind farm as the origin. The distance between each lightning location coordinate and the center of the wind farm is calculated. Lightning events with a distance less than the preset monitoring radius are filtered out. The number of lightning strikes, average lightning intensity, and maximum lightning intensity within the most recent time window are counted to construct a lightning activity feature vector.

[0028] Furthermore, the blade operating status data includes the unit number ID of each wind turbine, impeller speed, pitch angle, yaw angle, blade position angle, blade surface temperature, and blade vibration acceleration; based on the blade position angle, the spatial coordinates of the blade tip are calculated to form a set of blade spatial position parameters.

[0029] Preferably, a collection timestamp is added to each data point in the meteorological characteristic parameter set, lightning activity feature vector, and leaf spatial position parameter set, and data alignment is performed at 1-second intervals. If data from any data source is missing at a certain moment, the data value from the previous moment is used to fill the gap. A reasonable range of values ​​for each parameter is set. When a parameter in the meteorological characteristic parameter set exceeds the reasonable range, it is determined to be an outlier. The outlier is corrected by linear interpolation of data from adjacent moments, forming a time-aligned sequence of meteorological characteristic parameters, lightning activity feature vector, and leaf spatial position parameter sequence B.

[0030] S1.2: For each blade of the wind turbine, calculate the spatial distance between the tip spatial coordinates and the coordinates of the nearest lightning location, obtain the blade-lightning spatial correlation coefficient of each blade, and form a spatial correlation coefficient vector.

[0031] S1.3: Based on the blade operating status data, calculate the blade tip linear velocity, and statistically analyze the root mean square value and peak value within the most recent time window to obtain the blade operating status feature set; Specifically, when the number of lightning strikes in the lightning activity feature vector is greater than 0, the change in blade surface temperature and the abrupt change in blade vibration acceleration within the time window before and after the lightning event are extracted to form a lightning response feature vector; the blade tip linear velocity, root mean square value of vibration acceleration, peak value of vibration acceleration and lightning response feature vector are combined to form a blade operating state feature set.

[0032] S1.4: The meteorological characteristic parameter sequence, lightning activity characteristic vector sequence, leaf spatial position parameter sequence, spatial correlation coefficient vector, and leaf operation status characteristic set are concatenated column by column to form a multi-source characteristic data matrix with a feature dimension of 30. It should be noted that the number of rows in the multi-source feature data matrix corresponds to the length of the time series, and the number of columns corresponds to the total number of feature parameters. The multi-source feature data matrix is ​​normalized by using the minimum-maximum normalization method to scale the feature values ​​of each column to the interval [0, 1], thus obtaining the normalized multi-source feature data matrix.

[0033] For example, the meteorological monitoring station transmits data every second, including ambient temperature at 15°C, relative humidity at 65%, atmospheric pressure at 1013 hPa, wind speed at 12 m / s, wind direction angle at 120°, precipitation intensity at 2 mm / h, and visibility at 10 km, and calculates atmospheric dew point temperature at 8.5°C and atmospheric pressure at 1.225 kg / m³.3 Atmospheric density forms a set of meteorological characteristic parameters; simultaneously, the lightning location system provides lightning event information, such as a lightning strike occurring 3 kilometers from the center of the wind farm, with an intensity of -25kA, a cloud-to-ground lightning type, and a negative polarity; lightning strikes within a 5-kilometer monitoring radius are filtered, and the number of strikes (e.g., 3), average intensity (-20kA), and maximum intensity (-30kA) within the last 10 minutes are statistically analyzed to construct a lightning activity characteristic vector; the wind turbine SCADA system transmits the operating data of each wind turbine, such as the rotor speed of wind turbine No. 1 being 12 rpm, the pitch angle being 5°, the yaw angle being 30°, the blade position angle being 150°, the surface temperature being 10°C, and the vibration acceleration being 0.5 m / s². 2 The spatial coordinates of the blade tip (e.g., x=50m, y=80m, z=120m) are calculated based on the blade position angle, forming a set of blade spatial position parameters. All data are aligned at 1-second intervals, missing values ​​are filled with data from the previous moment, and outliers are corrected through linear interpolation, ultimately forming a time-aligned multi-source sequence. The spatial distance between the blade tip of wind turbine No. 1 and the nearest lightning location (e.g., 2.5 km) is calculated, resulting in a blade-lightning spatial correlation coefficient of 0.8, forming a vector. Based on the blade operating state, the blade tip linear velocity (e.g., 75 m / s) is calculated, and the root mean square value of the vibration acceleration (0.6 m / s²) is statistically analyzed. 2 ) and peak value (1.2m / s 2 When the number of lightning strikes is greater than 0, extract the change in blade surface temperature (e.g., -2°C) and the abrupt change in vibration acceleration (0.3 m / s²) before and after the lightning strike. 2 The lightning response feature vector is formed by combining all features (including meteorological sequence, lightning vector, location sequence, correlation coefficient and operational status features) into a 30-dimensional matrix and normalizing them to scale each feature value to the [0,1] interval.

[0034] S2: Based on the blade spatial azimuth angle and preset blade geometric parameters, establish a blade surface electric field distribution model and calculate the lightning susceptibility coefficient of each blade; S2.1: Obtain the preset blade geometric parameters and establish a three-dimensional geometric model of the blade; In an optional embodiment, the three-dimensional geometric model of the blade is established as follows: Preset blade geometric parameters include the total blade length, radially distributed chord length data, torsion angle distribution data, and the dielectric constant of the blade material; establish multiple cross-sectional profiles along the blade spanwise, each profile described by a standard airfoil curve equation, determine the cross-sectional dimensions based on the chord length parameters at that location, and perform rotational transformations on the cross-sections according to the torsion angle distribution data; connect the cross-sectional profiles using spline interpolation to generate a smooth blade surface; define a cylindrical segment structure for the connecting flange at the blade root, and form a tapered tip geometry at the blade tip; discretize the blade surface using a mesh, employing triangular or quadrilateral mesh elements to cover the entire blade surface, ensuring the mesh density meets the accuracy requirements of subsequent finite element electric field calculations; establish a local coordinate system for the blade in three-dimensional space, with the blade root center as the origin and the blade axis as the principal axis; assign the dielectric constant of the blade material to the model, completing the construction of the three-dimensional geometric model of the blade. S2.2: Based on the spatial azimuth angle of the blades, determine the spatial attitude of each blade at the current moment, and transform the three-dimensional geometric model of the blades according to the spatial attitude to obtain the real-time spatial position model of each blade. It should be noted that the spatial attitude includes the blade root coordinates, the blade axial direction vector, and the blade cross-section normal vector.

[0035] S2.3: Extract the topographic elevation data and surrounding building height data of the wind farm area, construct a three-dimensional topographic model of the wind farm environment, and embed the real-time spatial location model into the three-dimensional topographic model of the wind farm environment to obtain the blade-environment coupling model; S2.4: Set the height and charge density parameters of the thundercloud layer. Based on the blade-environment coupling model, use the finite element method to calculate the spatial electric field distribution, which includes the electric field intensity value and electric field direction vector of each grid node on the blade surface. The preferred formula for the spatial electric field distribution is as follows: ; in, The electric field strength in space, For spatial coordinates, Let's say the coordinates of the charge source. For charge density, The vacuum permittivity, Atmospheric attenuation coefficient, As an environmental correction factor, The background electric field intensity, For the Laplace operator, The volume of charge distribution in a thundercloud. Let be the integral volume element, representing a triple integral over the region of charge distribution.

[0036] S2.5: Divide the blade surface into K discrete segments along the radial direction of the three-dimensional geometric model of the blade, extract the electric field intensity value of the corresponding grid node of each discrete segment, and calculate the average electric field intensity of each discrete segment. Specifically, the formula for the mean electric field strength is as follows: ; in, Let be the mean electric field intensity of the k-th discrete segment. Let K be the surface area of ​​the k-th discrete segment. For the surface integration region of the k-th discrete segment, This is the weighting coefficient for the influence of the electric field gradient. Let be the gradient vector of the electric field intensity.

[0037] S2.6: Based on the mean electric field strength, the lightning attachment probability of each discrete segment is calculated using a preset lightning strike leader development model; In an optional embodiment, a preset lightning leader development model is constructed based on leader development theory and electric field breakdown criteria; the initial height and initial charge of the downlink leader in the thundercloud are set to simulate the process of the downlink leader developing towards the ground in a step-like manner; at each stage of leader development, the potential difference and electric field intensity distribution between the leader head and each discrete segment on the blade surface are calculated; when the local electric field intensity of a certain segment exceeds the critical breakdown field strength threshold, it is determined that the segment has the capability to develop an uplink leader; the principle of electric field vector superposition is adopted to comprehensively consider the thundercloud charge, downlink... The contributions of leader channel charge and induced charge on the blade surface to the space electric field are investigated. A leader development speed model is introduced, and the probability of the upward leader developing upward from each segment of the blade surface is calculated based on the empirical relationship between electric field strength and leader speed. When the gap between the upward and downward leaders is less than the final jump distance, a lightning attachment event is determined to have occurred. Multiple random simulations are performed using the Monte Carlo method to count the frequency of successful upward leader development and lightning attachment in each discrete segment. This frequency is normalized and used as the lightning attachment probability for that segment.

[0038] It should be noted that the lightning attachment probability characterizes the likelihood that a lightning leader will preferentially attach to this discrete segment.

[0039] S2.7: For each discrete section, calculate the local lightning strike risk index based on the lightning strike attachment probability and the distance of the discrete section from the blade tip, and then sum the local lightning strike risk indices of each discrete section by weight to obtain the lightning strike sensitivity coefficient. Furthermore, the specific formula for the localized lightning strike risk index is as follows: ; in, Let be the local lightning strike risk index for the i-th discrete segment. Let be the lightning strike attachment probability of the i-th segment. This is the distance attenuation coefficient. Let be the distance from the i-th segment to the blade tip. The total length of the blade. The Gaussian error function is... Let be the average electric field strength of the i-th segment. The national standard for the electric field strength that triggers a lightning strike. The standard deviation of the electric field strength; Furthermore, the formula for calculating the lightning strike sensitivity coefficient is as follows: ; in, The lightning sensitivity coefficient of the blade. Let be the area weight coefficient of the i-th segment. Let i be the local lightning strike risk index for the i-th segment. Let be the azimuth angle of the i-th segment. The optimal lightning protection azimuth angle is... The standard deviation of the azimuth angle. Here, K represents the weighting system for the risk change rate, and K is the total number of discrete segments dividing the blade surface. It is the hyperbolic tangent function; S2.8: For the three blades of the wind turbine, calculate the corresponding lightning sensitivity coefficients respectively, and mark the blade with the maximum lightning sensitivity coefficient as the high-risk blade at the current moment.

[0040] For example, preset blade geometry parameters include a length of 45 meters, chord length distribution data, torsion angle data, and dielectric constant (e.g., 3.5). A smooth 3D blade model is generated through spline interpolation. Based on the current blade spatial azimuth angle (e.g., yaw angle 30°, pitch angle 5°), the model is spatially transformed to obtain a real-time position model. An environmental 3D model is constructed by combining wind farm terrain elevation and building data, and then embedded into the blade model. The thundercloud height is set to 2000 meters, and the charge density to 0.1 C / m³. 3 The spatial electric field distribution was calculated using the finite element method, with the formula considering charge density, atmospheric attenuation, and environmental correction. The blade surface was divided into 10 discrete segments, and the mean electric field strength of each segment was calculated (e.g., 15 kV / m for segment k). Based on a lightning leader development model (simulating downlink leader development and assessing uplink leader probability), the lightning strike attachment probability of each segment was calculated (e.g., 0.05 for segment i). Combining this with a distance of 10 meters from the blade tip, a local lightning strike risk index was calculated, and then a weighted sum was obtained to obtain a lightning strike sensitivity coefficient of 0.75. The high-risk blade was marked as blade number 1.

[0041] S3: Input the multi-source feature data matrix and lightning sensitivity coefficient into the deep neural network model to output the lightning risk level and lightning probability prediction value of each wind turbine; In an optional embodiment, the deep neural network model includes a temporal feature extraction branch and a static feature extraction branch. The temporal feature extraction branch adopts a long short-term memory network structure, including an input layer, a first LSTM hidden layer, a second LSTM hidden layer, and a temporal feature output layer. The first LSTM hidden layer has 128 neurons, and the second LSTM hidden layer has 64 neurons. The static feature extraction branch adopts a fully connected neural network structure, including an input layer, a first fully connected hidden layer, a second fully connected hidden layer, and a static feature output layer. The first fully connected hidden layer has 64 neurons, and the second fully connected hidden layer has 32 neurons. Both the first LSTM hidden layer and the first fully connected hidden layer use the ReLU activation function.

[0042] S3.1: Extract time-related features from the multi-source feature data matrix to form a time-series feature input tensor, input the time-series feature extraction branch, and output the hidden state vector; It should be noted that the time-series related features include wind speed, wind direction angle, precipitation intensity, number of lightning strikes, average lightning intensity, impeller speed, blade position angle, and lightning sensitivity coefficient; the dimension of the time-series feature input tensor is 8; the dimension of the hidden state vector is 64; the time-series feature input tensor is calculated by forward propagation through the first LSTM hidden layer and the second LSTM hidden layer in sequence.

[0043] S3.2: Simultaneously, extract the static features at the current time from the multi-source feature data matrix to form a static feature input vector, input to the static feature extraction branch, and output the feature vector; It should be noted that the static features include static feature parameters such as ambient temperature, relative humidity, atmospheric pressure, atmospheric dew point temperature, atmospheric density, maximum lightning intensity, distance to lightning location, blade pitch angle, yaw angle, tip spatial coordinates, spatial correlation coefficient, blade tip linear velocity, root mean square value of vibration acceleration, peak value of vibration acceleration, change in blade surface temperature in the lightning response feature vector, and abrupt change in vibration acceleration. The static feature parameters are combined into a static feature input vector with a dimension of 16; the feature vector has a dimension of 32. The static feature input vector is input into the static feature extraction branch, and the static feature input vector is forward propagated through the first fully connected hidden layer and the second fully connected hidden layer in sequence. The feature vector output by the second fully connected hidden layer is used as the output of the static feature output layer.

[0044] S3.3: Calculate the attention weights of temporal features and static features through the attention fusion layer, and perform normalized weighted fusion of the hidden state vector and feature vector to obtain the fused feature vector; Preferably, an attention fusion layer is established, which is used to perform weighted fusion of the hidden state vector and the feature vector.

[0045] S3.5: Construct a dual-output task layer by fusing feature vector inputs, wherein the dual-output task layer includes a probability prediction output head and a risk level classification output head; It should be noted that the probability prediction output head uses a single-neuron fully connected layer structure; the risk level classification output head uses a 4-neuron fully connected layer plus a Softmax activation function structure.

[0046] S3.6: Calculate the predicted probability value of lightning strikes through the probability prediction output head, and classify the lightning strike risk level through the risk level classification output head; Specifically, the fused feature vector is mapped to a predicted lightning strike probability value, using the following formula: ; in, This is the predicted probability of lightning strike on the blade. Let be the weight coefficient of the j-th feature. To fuse the j-th component of the feature vector, where m is the dimension of the feature vector, Let S be the weight of the lightning strike sensitivity coefficient. To predict the length of the time window, is the time decay constant.

[0047] It should be noted that the predicted value of the lightning strike probability ranges from [0, 1], which represents the probability that blade 1 will be struck by lightning within 30 minutes in the future prediction time window.

[0048] Preferably, when the number of lightning strikes is greater than 0, the predicted lightning strike probability remains unchanged; when the number of lightning strikes in the lightning activity feature vector is equal to 0, the predicted lightning strike probability is attenuated and corrected.

[0049] Furthermore, the predicted maximum and average lightning strike probabilities for this wind turbine blade are calculated using the following formulas: ; in, The maximum probability of being struck by lightning. Let j be the probability of the j-th leaf. Geometric mean weighting Let v be the average probability, v be the variance weight, and Var be the variance function.

[0050] Furthermore, based on the predicted maximum and average lightning strike probabilities, the overall lightning strike risk index of the wind turbine is calculated using the following formula: ; in, The overall lightning strike risk index for wind turbines. The weighting coefficients are those with the highest probability. This is the predicted value for the maximum probability of a lightning strike. The weighting coefficients are the average probability coefficients. This is the predicted average probability of a lightning strike. The weighting coefficients for entropy values. The function is used to calculate the entropy of the probability distribution of the three blades. It is a vector consisting of the probabilities of the three leaf blades.

[0051] It should be noted that the maximum lightning strike probability prediction value represents the risk level of the blades of the wind turbine most susceptible to lightning strikes; the average lightning strike probability prediction value represents the overall lightning exposure level of the wind turbine; and the overall lightning strike risk index of the wind turbine ranges from [0, 1].

[0052] Specifically, the fused feature vector is mapped to a probability distribution vector of four risk levels, as shown in the following formula: ; in, Let be the probability distribution vector for the i-th risk level. Let be the logit value for the i-th risk level. The noise intensity coefficient is the Gumbel distribution. Let Gumbel be a random variable with a standard distribution. It is an exponential function.

[0053] In an optional embodiment, the category with the highest probability value in the probability distribution vector is selected as the first-level lightning risk. To ensure the consistency between the risk level and the overall lightning risk index of the wind turbine, risk level thresholds are defined: when the overall lightning risk index is less than the first threshold, it corresponds to the first-level risk; when the first threshold is greater than or equal to the overall lightning risk index and less than the second threshold, it corresponds to the second-level risk; when the second threshold is greater than or equal to the overall lightning risk index and less than the third threshold, it corresponds to the third-level risk; and when the overall lightning risk index is greater than or equal to the third threshold, it corresponds to the fourth-level risk.

[0054] Furthermore, the first-level lightning risk level is compared with the threshold classification result based on the overall lightning risk index of the wind turbine. If the two are inconsistent, the threshold classification result based on the overall lightning risk index of the wind turbine is adopted as the final lightning risk level to ensure the interpretability of the risk level judgment. The final lightning risk level of the wind turbine, the overall lightning risk index of the wind turbine, the maximum lightning probability prediction value, and the individual lightning probability prediction value of each blade are output.

[0055] Furthermore, the sub-step of step S3 is repeated for all wind turbines in the wind farm to obtain the final lightning strike risk level, overall lightning strike risk index, and maximum lightning strike probability prediction value for each turbine. The lightning strike risk assessment results of all wind turbines are sorted by unit number ID to form a wind farm lightning strike risk assessment matrix. Wind turbines in the wind farm lightning strike risk assessment matrix whose maximum lightning strike probability prediction value is greater than a preset risk threshold are screened out. For the screened high-risk wind turbines, their unit number ID, final lightning strike risk level L_final, overall lightning strike risk index, and blade lightning strike probability prediction value are extracted and packaged into a high-risk wind turbine list. The high-risk wind turbine list and the wind farm lightning strike risk assessment matrix are passed to step S4 to trigger the generation of early warning signals and the matching of response strategies.

[0056] It should be noted that each row of the wind farm lightning strike risk assessment matrix corresponds to one wind turbine; the four risk levels—Level 1, Level 2, Level 3, and Level 4—correspond to low, medium, high, and extremely high risk, respectively; the preset risk thresholds include a first threshold, a second threshold, and a third threshold; the first threshold is determined based on historical lightning strike statistics and the probability analysis of minor blade damage; by collecting lightning strike records from wind farms over many years, statistically analyzing the frequency of lightning strikes under different meteorological conditions and blade operating states, and combining this with the probability of repairable damage such as minor ablation and scratches on the blade surface, the first threshold is determined using the percentile method; the second threshold is... The first threshold is determined based on the analysis of the recipient's operating characteristic curve and economic loss assessment. Using the prediction results of the deep neural network model on the validation set, true positive rate and false positive rate curves are plotted under different thresholds. The threshold point that maximizes the Youden index is selected as the initial value of the second threshold. The third threshold is determined based on the critical conditions for severe blade damage and safety margin requirements. Referring to blade lightning strike test data and electric field simulation results, the critical lightning current intensity and electric field intensity combination that leads to severe consequences such as blade structural damage and penetrating damage are identified. These critical conditions are mapped to the overall lightning strike risk index of the wind turbine, and a safety factor is introduced for conservative correction to obtain the third threshold.

[0057] For example, the deep neural network model processes multi-source features and sensitivity coefficients; the temporal feature extraction branch takes into input 8-dimensional tensors such as wind speed and lightning strike count, and outputs a 64-dimensional hidden state vector through an LSTM layer; the static feature extraction branch takes into input 16-dimensional vectors such as temperature and humidity, and outputs a 32-dimensional feature vector through a fully connected layer. An attention fusion layer weights both to obtain a fused feature vector; the dual-output task layer calculates the predicted lightning strike probability (e.g., p=0.85, representing the probability of a lightning strike within the next 30 minutes) and the risk level probability distribution (e.g., level four risk has the highest probability); based on the maximum probability and the average probability, it calculates the overall lightning strike risk index of the wind turbine (e.g., ...). =0.8), and based on the threshold, the final risk level is divided (e.g., level four risk), high-risk wind turbines are screened, and a list is generated and passed to S4.

[0058] S4: When the predicted probability of lightning strike is greater than the preset risk threshold, an early warning signal is generated and sent to the wind turbine control system. At the same time, the preset response strategy library is matched according to the lightning strike risk level to obtain the target pitch angle adjustment and target speed adjustment, and the wind turbine is controlled to perform blade lightning protection actions.

[0059] S4.1: Traverse the wind turbines in the high-risk wind turbine list, extract the unit number, maximum lightning strike probability prediction value and final lightning strike risk level of the wind turbine, determine the early warning triggering conditions and generate graded early warning signals; Specifically, the warning trigger condition is determined by whether the predicted maximum lightning strike probability is greater than a preset risk threshold. When the predicted maximum lightning strike probability is greater than the preset risk threshold, the warning process is triggered, and a corresponding graded warning signal is generated based on the final lightning strike risk level. If the final lightning strike risk level is level two, a yellow warning signal is generated; if the final lightning strike risk level is level three, an orange warning signal is generated; and if the final lightning strike risk level is level four, a red warning signal is generated. Warning priorities are assigned to each graded warning signal, with the yellow warning signal having a priority of 2, the orange warning signal having a priority of 3, and the red warning signal having a priority of 4. The expected warning duration is calculated based on the number of lightning strikes and the average lightning intensity in the lightning activity feature vector sequence output from step S1.

[0060] S4.2: Encapsulate the graded early warning signals into early warning data packets and send them to the wind turbine control system via a communication protocol; It should be noted that the wind farm central monitoring system uses the Profibus-DP communication protocol to send early warning data packets to the corresponding wind turbine control system; after receiving the early warning data packets, the wind turbine control system displays the early warning information on the human-machine interface and stores it in the early warning log database.

[0061] S4.3: Based on the lightning strike risk level, query the preset response strategy library for matching basic response strategies, and extract the pitch angle increase coefficient and speed reduction coefficient; Furthermore, a pre-defined response strategy library is established, which contains multi-level response strategies for different final lightning strike risk levels. Within this library, basic strategies are defined for medium-risk (Level 2 risk), high-risk (Level 3 risk), and extremely high-risk (Level 4 risk). Each basic response strategy includes corresponding pitch angle amplification coefficient, rotational speed reduction coefficient, and response time parameters. Based on the final lightning strike risk level in the warning data packet, a matching basic response strategy is retrieved from the pre-defined response strategy library. The pitch angle amplification coefficient, rotational speed reduction coefficient, and response time parameters are then extracted from the basic response strategy.

[0062] S4.4: Based on the lightning strike sensitivity coefficient, identify the blade with the highest lightning strike sensitivity and calculate the target lightning protection position angle for this blade to reach the lightning protection safety position; Specifically, the lightning sensitivity coefficients corresponding to the blades are obtained from the early warning data packets. The values ​​of the lightning sensitivity coefficients are compared, and the blade with the maximum value of the lightning sensitivity coefficient is marked as the blade with the highest lightning sensitivity. The current blade position angle of each blade is obtained from the blade operation status data in step S1. The blade lightning protection safe position area is defined as within ±30° in the vertical downward direction, and the corresponding blade position angle range is 240° to 300°. For the blade with the highest lightning sensitivity, if the current blade position angle of the blade is not within the range of 240° to 300°, the minimum rotation angle for the blade to reach the blade lightning protection safe position area is calculated, and the target lightning protection position angle of the blade with the highest lightning sensitivity is set to 270°. The target lightning protection position angle of the remaining blades is calculated based on the constraint condition of 120° between the blades.

[0063] S4.5: Calculate the target pitch angle adjustment based on the pitch angle amplification factor and the current pitch angle; simultaneously calculate the target speed adjustment based on the speed reduction factor and the current impeller speed. Furthermore, the current pitch angle and impeller speed are obtained from the blade operating status data in step S1; the current pitch angle is multiplied by the pitch angle amplification factor to obtain the target pitch angle; the difference between the target pitch angle and the current pitch angle is calculated to obtain the initial pitch angle adjustment amount; if the initial pitch angle adjustment amount is greater than 30°, the initial pitch angle adjustment amount is corrected to 30°; the current impeller speed is multiplied by the speed reduction factor to obtain the target speed; the difference between the target speed and the current impeller speed is calculated to obtain the initial speed adjustment amount; a fine-tuning coefficient is calculated based on the overall wind turbine lightning risk index in step S3; the initial pitch angle adjustment amount is multiplied by the fine-tuning coefficient to obtain the target pitch angle adjustment amount; the initial speed adjustment amount is multiplied by the fine-tuning coefficient to obtain the target speed adjustment amount.

[0064] S4.6: Generate a sequence of blade lightning protection action control commands based on the target lightning protection position angle, the target pitch angle adjustment, and the target speed adjustment; In an optional embodiment, the blade lightning protection action control command sequence includes three stages of control commands. The first stage is a speed reduction command, which adjusts the impeller speed according to the target speed adjustment amount within the response time parameter. The second stage is a pitch angle increase command, which adjusts the pitch angle according to the target pitch angle adjustment amount within the response time parameter. The third stage is a blade position adjustment command, which guides the blade with the highest lightning strike sensitivity to rotate to the target lightning protection position angle by fine-tuning the impeller speed. A timestamp and priority are added to each control command. The blade lightning protection action control command sequence is encapsulated into a control command data packet according to the execution order.

[0065] S4.7: The wind turbine control system executes each control command in sequence according to the order of the blade lightning protection action control command sequence, guiding the blade with the highest lightning strike sensitivity to rotate to the target lightning protection position angle, and completing the blade lightning protection action; Specifically, the wind turbine control system executes blade lightning protection actions and monitors the execution status. After receiving the control command data packet, the wind turbine control system executes each control command in sequence according to the order of the blade lightning protection action control command sequence. When executing the speed reduction command, the generator speed is reduced by adjusting the inverter output frequency, while monitoring the speed deviation between the actual speed and the target speed. When executing the pitch angle increase command, the blade pitch angle is increased by driving the pitch angle control motor, while monitoring the pitch angle deviation between the actual pitch angle and the target pitch angle.

[0066] In an optional embodiment, when the absolute value of the rotational speed deviation is less than the rotational speed control accuracy threshold and the absolute value of the pitch angle deviation is less than the pitch angle control accuracy threshold, the first and second stage control commands are determined to be completed. When executing the blade position adjustment command, the blade with the highest lightning strike sensitivity is rotated to the target lightning protection position angle by fine-tuning the impeller speed. During the execution of the blade lightning protection action control command sequence, the impeller speed, pitch angle, blade position angle, blade vibration acceleration, and generator output power are collected in real time to form execution status monitoring data. When all control commands are completed, an execution completion confirmation signal is generated, and the execution completion confirmation signal and execution status monitoring data are sent back to the wind farm central monitoring system.

[0067] Furthermore, after receiving the confirmation signal that the execution is completed, the wind farm central monitoring system re-collects the blade operating status data of the wind turbine; based on the blade operating status data after the lightning protection action is executed and the current meteorological characteristic parameter sequence and lightning activity characteristic vector sequence, the calculation process of step S2 is re-executed to obtain the lightning sensitivity coefficient after the lightning protection action is executed; and the change in the lightning sensitivity coefficient before and after the lightning protection action is executed is calculated.

[0068] In an optional embodiment, when the change in the lightning sensitivity coefficient is greater than 0, it indicates that the lightning protection action reduces the risk of lightning strikes to the blades. The lightning protection action effect evaluation index is calculated based on the change in the lightning sensitivity coefficient. The final lightning risk level, basic response strategy, target pitch angle adjustment, target speed adjustment, and lightning protection action effect evaluation index of this early warning response are recorded as strategy effect records and stored in the strategy effect database. If the number of strategy effect records in the strategy effect database reaches a preset number, the lightning protection action effect evaluation index is used as the optimization objective function, and the pitch angle increase coefficient and speed reduction coefficient in the preset response strategy library are updated using the gradient descent method.

[0069] For example, when the predicted probability of a lightning strike is greater than 0.7, a red alert is generated; a matching strategy is queried from the response strategy library to obtain the pitch angle increase coefficient (e.g., 1.5) and speed reduction coefficient (e.g., 0.7), the blade with the highest lightning strike sensitivity (e.g., blade No. 1) is identified, and its target lightning protection position angle (270°) is calculated; based on the current pitch angle (5°) and impeller speed (12 rpm), the target pitch angle adjustment amount (e.g., 2.5° adjustment if increased to 7.5°) and the target speed adjustment amount (e.g., -3.6 rpm adjustment if decreased to 8.4 rpm) are calculated, and a control command sequence is generated to sequentially execute speed reduction, pitch angle increase, and blade position adjustment to guide the high-risk blade to a safe position; after execution, the status data is monitored, the lightning strike sensitivity coefficient is recalculated, the lightning protection effect is evaluated (e.g., coefficient decreases by 0.2), and the strategy library parameters are updated to optimize future responses.

[0070] In summary, this invention constructs a multi-source feature data matrix by collecting real-time meteorological data, lightning monitoring data, and blade operating status data from wind farms, achieving comprehensive capture of factors influencing lightning strike risk. Based on blade operating status and geometric parameters, a surface electric field distribution model is established and a lightning sensitivity coefficient is calculated. Electromagnetic field theory is combined with the real-time status of the wind turbine. The electric field concentration under different attitudes is quantified using the finite element method and converted into lightning strike probability, establishing a hybrid intelligent mechanism that integrates physical and data-driven approaches, shortening the early warning response time. Multi-source data and sensitivity coefficients are input into a deep neural network employing a dual-branch architecture and attention fusion mechanism. LSTM captures temporal evolution features, a fully connected network processes static features, and multi-task joint learning of probability prediction and hierarchical classification is achieved. When the lightning strike probability exceeds a threshold, a response strategy is matched according to the risk level, and the blade most susceptible to lightning strikes is identified. A three-stage control sequence—reducing rotational speed, increasing pitch angle, and precisely guiding highly sensitive blades to a safe position—achieves closed-loop control from passive monitoring to active protection, reducing the lightning strike probability, lowering the blade damage accident rate, and reducing the economic loss from a single lightning strike.

[0071] Based on the teachings of the above embodiments, other aspects of the present invention also disclose an artificial intelligence-based wind turbine blade lightning strike risk early warning device, comprising: The multi-source data acquisition module is used to collect real-time meteorological data, lightning monitoring data and blade operating status data in the wind farm area, and to construct a multi-source feature data matrix. The blade electric field module establishes a blade surface electric field distribution model based on blade operating status data and preset blade geometric parameters, and calculates the lightning susceptibility coefficient of each blade. The lightning strike risk prediction module is used to input the multi-source feature data matrix and lightning strike sensitivity coefficient into the constructed deep neural network model, and output the lightning strike risk level and lightning strike probability prediction value for each wind turbine. The early warning and control execution module generates an early warning signal and sends it to the wind turbine control system when the predicted lightning strike probability value is greater than the preset risk threshold. At the same time, it matches the preset response strategy library according to the lightning strike risk level, obtains the target pitch angle adjustment amount and the target speed adjustment amount, and controls the wind turbine to perform blade lightning protection actions.

[0072] This embodiment also provides a computer device applicable to the case of a wind turbine blade lightning strike risk early warning method based on artificial intelligence, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the wind turbine blade lightning strike risk early warning method based on artificial intelligence as proposed in the above embodiment.

[0073] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0074] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for early warning of lightning strike risks to wind turbine blades based on artificial intelligence as proposed in the above embodiments.

[0075] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0076] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of lightning strike risk in wind turbine blades based on artificial intelligence, characterized in that: include, Collect real-time meteorological data, lightning monitoring data, and blade operating status data of the wind farm area to construct a multi-source feature data matrix; Based on the blade operating status data and preset blade geometric parameters, a blade surface electric field distribution model is established, and the lightning susceptibility coefficient of each blade is calculated. The multi-source feature data matrix and the lightning sensitivity coefficient are input into the constructed deep neural network model, which outputs the lightning risk level and lightning probability prediction value of each wind turbine. When the predicted probability of a lightning strike exceeds the preset risk threshold, an early warning signal is generated and sent to the wind turbine control system. At the same time, the system matches the preset response strategy library according to the lightning strike risk level, obtains the target pitch angle adjustment and the target speed adjustment, and controls the wind turbine to perform blade lightning protection actions.

2. The method for early warning of lightning strike risk on wind turbine blades based on artificial intelligence as described in claim 1, characterized in that: The method for establishing the electric field distribution model on the blade surface is as follows: Obtain the preset blade geometric parameters and establish a three-dimensional geometric model of the blade; Based on the spatial azimuth angle of the blades, the spatial attitude of each blade at the current moment is determined, and the three-dimensional geometric model of the blades is spatially transformed according to the spatial attitude to obtain the real-time spatial position model of each blade. Extract the topographic elevation data and surrounding building height data of the wind farm area, construct a three-dimensional terrain model of the wind farm environment, and embed the real-time spatial location model into the three-dimensional terrain model of the wind farm environment to obtain the blade-environment coupling model; By setting parameters for thundercloud height and thundercloud charge density, and using the finite element method to calculate the spatial electric field distribution based on the blade-environment coupling model, the spatial electric field distribution includes the electric field intensity value and electric field direction vector of each grid node on the blade surface.

3. The method for early warning of lightning strike risk on wind turbine blades based on artificial intelligence as described in claim 1, characterized in that: The deep neural network model includes a temporal feature extraction branch and a static feature extraction branch. The temporal feature extraction branch adopts a long short-term memory network structure, including an input layer, a first LSTM hidden layer, a second LSTM hidden layer, and a temporal feature output layer. The static feature extraction branch adopts a fully connected neural network structure, including an input layer, a first fully connected hidden layer, a second fully connected hidden layer, and a static feature output layer.

4. The method for early warning of lightning strike risk on wind turbine blades based on artificial intelligence as described in claim 1, characterized in that: Obtain the target pitch angle adjustment and target speed adjustment, and control the wind turbine to perform blade lightning protection actions, including: Traverse the list of high-risk wind turbines, extract the turbine number, maximum lightning strike probability prediction value and final lightning strike risk level of each wind turbine, determine the early warning triggering conditions and generate graded early warning signals; The graded early warning signals are encapsulated into early warning data packets and sent to the wind turbine control system via a communication protocol. Based on the lightning strike risk level, the system queries the preset response strategy library to find the matching basic response strategy and extracts the pitch angle increase coefficient and speed reduction coefficient. Based on the lightning strike sensitivity coefficient, identify the blade with the highest lightning strike sensitivity and calculate the target lightning protection position angle for this blade to reach the lightning protection safety position; The target pitch angle adjustment is calculated based on the pitch angle amplification factor and the current pitch angle; simultaneously, the target speed adjustment is calculated based on the speed reduction factor and the current impeller speed. Generate a sequence of control commands for blade lightning protection actions based on the target lightning protection position angle, the target pitch angle adjustment, and the target rotational speed adjustment. The wind turbine control system executes each control command sequentially according to the order of the blade lightning protection action control command sequence, guiding the blade with the highest lightning strike sensitivity to rotate to the target lightning protection position angle, thus completing the blade lightning protection action.

5. The method for early warning of lightning strike risk on wind turbine blades based on artificial intelligence as described in claim 4, characterized in that: The method for obtaining the lightning strike sensitivity coefficient is as follows: The blade surface is divided into K discrete segments along the radial direction of the three-dimensional geometric model of the blade. The electric field intensity values ​​of the corresponding grid nodes of each discrete segment are extracted, and the mean electric field intensity of each discrete segment is calculated. Based on the mean electric field strength, the lightning attachment probability of each discrete segment is calculated using a preset lightning strike leader development model. For each discrete section, based on the lightning strike attachment probability and the distance of the discrete section from the blade tip, the local lightning strike risk index is calculated, and the local lightning strike risk index of each discrete section is weighted and summed to obtain the lightning strike sensitivity coefficient.

6. The method for early warning of lightning strike risk on wind turbine blades based on artificial intelligence as described in claim 4, characterized in that: The method for obtaining the lightning strike risk level and the predicted lightning strike probability is as follows: Temporally relevant features are extracted from a multi-source feature data matrix to form a temporal feature input tensor. The input tensor is then used to input the temporal feature extraction branch, and a hidden state vector is output. Simultaneously, static features at the current moment are extracted from the multi-source feature data matrix to form a static feature input vector, which is then input into the static feature extraction branch to output a feature vector. The attention weights for temporal and static features are calculated through an attention fusion layer. The hidden state vector and the feature vector are then normalized and weighted and fused to obtain a fused feature vector. The fused feature vector is input into a dual-output task layer, wherein the dual-output task layer includes a probability prediction output head and a risk level classification output head; The probability prediction output head is used to calculate the predicted value of lightning strike probability, and the risk level output head is used to classify the lightning strike risk level.

7. The method for early warning of lightning strike risk on wind turbine blades based on artificial intelligence as described in claim 4, characterized in that: The blade lightning protection action control command sequence includes three stages of control commands. The first stage is a speed reduction command, wherein the content of the speed reduction command is to adjust the impeller speed according to the target speed adjustment amount within the response time parameter. The second stage is the pitch angle increase command, which means adjusting the pitch angle according to the target pitch angle adjustment amount within the response time parameter; the third stage is the blade position adjustment command, which means guiding the blade with the highest lightning strike sensitivity to rotate to the target lightning protection position angle by fine-tuning the impeller speed.

8. An artificial intelligence-based wind turbine blade lightning strike risk early warning device, based on the artificial intelligence-based wind turbine blade lightning strike risk early warning method according to any one of claims 1 to 7, characterized in that: include, The multi-source data acquisition module is used to collect real-time meteorological data, lightning monitoring data and blade operating status data in the wind farm area, and to construct a multi-source feature data matrix. The blade electric field module establishes a blade surface electric field distribution model based on the blade operating status data and preset blade geometric parameters, and calculates the lightning sensitivity coefficient of each blade. The lightning strike risk prediction module is used to input the multi-source feature data matrix and the lightning strike sensitivity coefficient into the constructed deep neural network model, and output the lightning strike risk level and lightning strike probability prediction value of each wind turbine. The early warning and control execution module is used to generate an early warning signal and send it to the wind turbine control system when the predicted probability of lightning strikes is greater than the preset risk threshold. At the same time, it matches the preset response strategy library according to the lightning strike risk level, obtains the target pitch angle adjustment amount and the target speed adjustment amount, and controls the wind turbine to perform blade lightning protection actions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the artificial intelligence-based wind turbine blade lightning risk early warning method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based wind turbine blade lightning risk early warning method as described in any one of claims 1 to 7.

Images