CN117934482B - Lightning probability prediction method, device and equipment for wind turbine and storage medium - Google Patents

Abstract

The application provides a lightning probability prediction method, device and equipment of a wind turbine and a storage medium, and relates to the field of electric digital data processing. According to the method, the characteristic that lightning is tip corona discharge is utilized, the wind turbine is divided into a plurality of small areas from the wind turbine, the corona times and the lightning hit times of each small area are identified and counted, the lightning strike probability is obtained by dividing the lightning hit times by the corona times, a prediction model capable of predicting future lightning strike probability is obtained by training a plurality of previous lightning strike probabilities through a machine learning algorithm, and the future lightning strike probability of the wind turbine can be predicted through the prediction model. According to the method, environmental factors of the wind motor are not considered, calculation and modeling are directly performed from two results of corona generation and lightning stroke reception, so that the calculated result and the modeled model both have the data characteristics, the prediction method has good universality, the calculation force burden is small, and complex calculation steps and models are avoided.

Description

Lightning probability prediction method, device and equipment for wind turbine and storage medium

Technical Field

The application relates to the field of electric digital data processing, in particular to a lightning stroke probability prediction method, device and equipment of a wind turbine and a storage medium.

Background

The wind power generation is to convert the kinetic energy of wind into electric energy, the wind energy is a clean and pollution-free renewable energy source, the wind power generation drives the windmill blades to rotate through wind power, and the rotating speed is improved through a speed increaser to promote the generator to generate electricity, and the wind power generation does not need to use fuel, does not generate radiation or air pollution, and is a renewable new energy source. Wind power generation mainly converts wind energy into mechanical work through a wind motor (wind driven generator), and the mechanical work drives a rotor to rotate, and finally, alternating current is output.

In order to ensure good operation efficiency of the wind turbine, the wind turbine is usually required to be built in a relatively flat area so as to avoid influence on wind speed caused by buildings, mountains, forests and the like. Because the length of the sleeve and the blade of the wind motor is longer, the blade of the wind motor is positioned at a high position, and is easy to be struck by lightning in thunderstorm weather.

At present, the lightning stroke models of wind motors are in an initial research stage, the lightning stroke models proposed by various students are different, so that the references, the restrictions and the realizabilities of the lightning stroke models are different, the lightning stroke models are not uniform, the universality is poor, the restriction conditions are many, the lightning stroke probabilities obtained by the staff according to different lightning stroke models are also different, most of the lightning stroke models have no prediction function, and the staff can not easily obtain the future hit probability of the wind motors by utilizing the existing lightning stroke models, so that countermeasures cannot be prepared in advance.

Disclosure of Invention

The application mainly aims to provide a lightning probability prediction method, device and equipment for a wind motor and a storage medium, so as to solve the problems that in the prior art, the lightning probabilities obtained by different lightning models are same, most of the lightning models have no prediction function, and therefore workers cannot easily obtain the future lightning probability of the wind motor by using the existing lightning models, and countermeasures cannot be prepared in advance.

In order to achieve the above object, the present application provides the following technical solutions:

the wind turbine comprises a tower drum with a fixed frame arranged in a preset area and three blades rotatably arranged on the tower drum, wherein the lightning strike probability prediction method comprises the following steps:

dividing the area of each blade respectively, wherein each blade forms at least two blade subareas;

Dividing the tower barrel into areas to form at least two tower barrel areas;

Respectively acquiring a plurality of historical corona times of each blade subarea and each tower bobbin area based on a preset time interval;

Respectively acquiring a plurality of historical hit times of each blade subarea and each tower cone area based on the preset time interval;

acquiring the ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning stroke probability;

Respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model;

And respectively predicting future lightning stroke probability of each blade subarea and each tower cone area through the prediction model based on a preset prediction step number.

As a further improvement of the present application, the step of respectively acquiring a plurality of historical corona times of each blade subregion and each tower tube region based on a preset time interval includes:

Acquiring a corona generation visual image of the wind motor through ultraviolet imaging;

acquiring a minimum bounding box of a corona region of the corona occurrence visual image through a target detection algorithm;

judging whether the minimum bounding box exists in each blade subarea or each tower cone area or not respectively;

If the minimum bounding box exists, respectively increasing the number of times of corona once in a blade subarea or a tower cone area in which the minimum bounding box exists;

And respectively acquiring all historical corona times of each blade subarea and each tower cone area based on the preset time interval.

As a further improvement of the present application, acquiring a minimum bounding box of a corona region of the corona-generating visual image by a target detection algorithm includes:

Dividing the corona generation visual image into a plurality of grids through grids with a first preset density;

predicting a preset number of first-order bounding boxes based on each grid respectively, wherein each first-order bounding box comprises at least one grid;

Defining the highest confidence coefficient of the corona region, and respectively acquiring the confidence coefficient of each first-order boundary box;

Acquiring a first-order boundary box with highest confidence coefficient from all confidence coefficients as a second-order boundary box;

Calculating the intersection ratio of the second-order boundary frame and the first-order boundary frame respectively;

Reserving a second-order boundary box with the cross ratio being greater than or equal to a preset ratio as a third-order boundary box;

The third-order boundary frame with the highest confidence coefficient is reserved and used as a fourth-order boundary frame;

And acquiring a union set of all fourth-order bounding boxes, wherein the union set is the minimum bounding box.

As a further improvement of the present application, the step of respectively obtaining a plurality of historical hit times of each blade subregion and each tower tube region based on the preset time interval includes:

Obtaining a lightning strike hitting visual image of the wind turbine through continuous recording;

Acquiring a lightning stroke termination point in the lightning stroke hitting visual image through a sliding window algorithm;

Judging whether the lightning stroke termination point exists in each blade subarea or each tower cylinder area or not respectively;

If the lightning stroke exists, respectively increasing the number of times of hitting the blade subareas or the tower cylinder areas with the lightning stroke end points for one time;

And respectively acquiring all historical hit times of each blade subarea and each tower cone area based on the preset time interval.

As a further improvement of the present application, acquiring a lightning stroke termination point in the visual image of the lightning stroke hit by a sliding window algorithm includes:

Binarizing the lightning strike hit visual image to obtain a black-white visual image;

dividing the black-and-white visual image through a grid with a second preset density;

Respectively obtaining the color of each grid, assigning 0 to the grid with black color and 1 to the grid with white color;

Defining each grid and eight adjacent grids as a3 x3 window;

traversing all grids through the 3 x 3 window in one step on all grids;

Deleting grids with all 1 values and grids with all 0 values in the 3 multiplied by 3 window based on all traversal results;

Acquiring reserved grids and defining the reserved grids as hit boundaries;

and acquiring a closed end of the hit boundary, wherein the closed end is the lightning stroke termination point.

As a further improvement of the present application, a prediction model is obtained by learning and training the historical lightning probability of each blade subregion and each tower barrel region respectively through a machine learning algorithm, comprising:

Generating a plurality of acquisition time points based on a preset time interval;

Acquiring historical lightning stroke probability of each blade subarea and each tower cone area based on all acquisition time points, and generating a data set based on the historical accumulated probability of each blade subarea and the historical accumulated probability of each tower cone area;

integrating all the data sets to form a total data set;

normalizing the total data set to obtain a normalized data set;

dividing the normalized data set into a training set and a verification set according to a preset proportion;

defining a topological relation of a neural network model, wherein the topological relation comprises an input layer, an implicit layer and an output layer which are sequentially connected through signals;

Inputting the training set into the input layer, carrying out iteration for preset times through the neural network model, and respectively acquiring root mean square errors of the training set and the current training result based on each iteration;

And obtaining the minimum value in all root mean square errors, and obtaining a training result corresponding to the minimum value as the prediction model.

As a further improvement of the present application, the neural network model is characterized by the formula (1):

（1）；

Wherein, The neural network model is preset; /(I)Is the input layer and/>，/>For the input layer/>Each input node corresponds to a training set,/>Is the first of the input layersThe/>, of the input nodes to the hidden layerThe weights of the input nodes; /(I)For connecting to the hidden layerBias of the individual input nodes; /(I)Is a transfer function, and/>The numbers in brackets of the subscripts are the number of layers.

In order to achieve the above purpose, the present application further provides the following technical solutions:

The lightning probability prediction device of the wind turbine is applied to the lightning probability prediction method, and comprises the following steps:

The blade area dividing module is used for dividing areas of each blade respectively, and each blade forms at least two blade subareas;

The tower barrel region dividing module is used for dividing the tower barrel region to form at least two tower barrel regions;

The historical corona frequency acquisition module is used for respectively acquiring a plurality of historical corona frequencies of each blade subarea and each tower bobbin area based on a preset time interval;

the historical hit frequency acquisition module is used for respectively acquiring a plurality of historical hit frequencies of each blade subarea and each tower bobbin area based on the preset time interval;

The historical lightning probability definition module is used for acquiring the ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning probability;

the prediction model acquisition module is used for respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model;

And the future lightning strike probability acquisition module is used for respectively predicting the future lightning strike probability of each blade subarea and each tower bobbin area through the prediction model based on the preset prediction step number.

In order to achieve the above purpose, the present application further provides the following technical solutions:

An electronic device comprising a processor, a memory coupled to the processor, the memory storing program instructions executable by the processor; and the processor executes the program instructions stored in the memory to realize the method for predicting the lightning stroke probability of the wind motor.

In order to achieve the above purpose, the present application further provides the following technical solutions:

A storage medium having stored therein program instructions which, when executed by a processor, implement a method for predicting a probability of a lightning strike by a wind turbine as described above.

According to the application, each blade is divided into areas, and each blade forms at least two blade subareas; dividing the tower barrel into areas to form at least two tower barrel areas; respectively acquiring a plurality of historical corona times of each blade subarea and each tower bobbin area based on a preset time interval; respectively acquiring a plurality of historical hit times of each blade subarea and each tower bobbin area based on a preset time interval; acquiring a ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning stroke probability; respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model; and respectively predicting future lightning stroke probability of each blade subarea and each tower barrel area through a prediction model based on the preset prediction step number. According to the method, the characteristics that lightning is tip corona discharge are utilized, the wind turbine is divided into a plurality of small areas from the wind turbine, the corona times and the lightning hit times of each small area are identified and counted, the lightning strike probability is obtained by dividing the lightning hit times by the corona times, a prediction model capable of predicting future lightning strike probability is obtained by training a plurality of previous lightning strike probabilities through a machine learning algorithm, and the future lightning strike probability of the wind turbine can be predicted through the prediction model. According to the method, environmental factors (such as weather law, wind speed distribution, altitude, wind motor height and model number) of the wind motor are not considered, calculation and modeling are directly carried out from two results of generating corona and receiving lightning, so that the calculated result and the modeled type both have the data characteristics (such as different data results due to different regional characteristics, weather characteristics, geographical characteristics and the like), the prediction method has good universality, the preorder factors of the wind motor are not considered, calculation and prediction are directly carried out from the result with the characteristics, and the calculation load is small and complex calculation steps and models are avoided.

Drawings

FIG. 1 is a schematic diagram of steps in a method for predicting probability of lightning strike of a wind turbine according to an embodiment of the present application;

FIG. 2 is a schematic diagram of functional blocks of an embodiment of a wind turbine lightning probability prediction apparatus according to the present application;

FIG. 3 is a schematic diagram of an embodiment of an electronic device of the present application;

FIG. 4 is a schematic diagram illustrating the structure of an embodiment of a storage medium according to the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

The terms "first," "second," "third," and the like in this disclosure are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "first," "second," and "third" may explicitly or implicitly include at least one such feature. In the description of the present application, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise. All directional indicators (such as up, down, left, right, front, and back … …) in the embodiments of the present application are merely used to explain the relative positional relationship, movement, etc. between the components in a particular gesture (as shown in the drawings), and if the particular gesture changes, the directional indicator changes accordingly. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those listed steps or elements but may include other steps or elements not listed or inherent to such process, method, article, or apparatus.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.

As shown in fig. 1, the present embodiment provides an embodiment of a lightning probability prediction method of a wind turbine, in which the wind turbine includes a tower having a fixed frame disposed in a predetermined area, and three blades rotatably mounted on the tower.

Preferably, the rotor of the wind motor is generally a 120 ° annular array of three blades, each blade typically having a length of 20 to 44 meters and a chord length of 2 to 3.7 meters, the thickness of the area swept during rotation being about 1.4 to 2.6 meters. The wind power machine has uniform specification, and the specific structure, style and model of the wind power machine are not repeated in the implementation.

Specifically, the lightning probability prediction method of the present embodiment specifically includes the following steps:

And S1, respectively dividing the areas of each blade, wherein each blade forms at least two blade subareas.

Preferably, the length of the blade is 20 meters to 44 meters, and a dividing line can be generated every 10cm based on the length direction of the blade, namely the length of the blade subregion in the length direction of the blade is 10cm, and the width of the blade subregion is equal to the width of the blade.

And S2, dividing the tower barrel into areas to form at least two tower barrel areas.

Preferably, the tower is a cylinder with a higher height, and the cylinder may be divided every 10cm in the height direction of the cylinder.

Preferably, since lightning has a tip discharge characteristic, cylinders not located at the ends may not be divided.

And step S3, respectively acquiring a plurality of historical corona times of each blade subarea and each tower bobbin area based on a preset time interval.

And S4, respectively acquiring a plurality of historical hit times of each blade subarea and each tower bobbin area based on a preset time interval.

Preferably, the number of corona generated and hit times of the wind power machine per hour in each thunderstorm weather can be counted.

And S5, acquiring the ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning stroke probability.

Preferably, the corona is a sufficiently unnecessary condition for lightning, i.e. the generation of the corona does not necessarily generate lightning, which must first generate the corona; if corona exists but no thunder exists, the condition that the thunderstorm weather is not thunder and lightning exists is indicated, and the wind motor is leaked to generate corona; if thunderstorm weather is met and corona exists, but lightning is not hit, the electric field strength is insufficient, or the downlink pilot strength is insufficient, or the uplink pilot strength is insufficient.

And S6, respectively learning and training the historical lightning stroke probability of each blade subarea and each tower bobbin area through a machine learning algorithm to obtain a prediction model.

Preferably, the machine learning algorithm mainly comprises a genetic algorithm, a Q learning algorithm, a support vector machine algorithm and a neural network algorithm, and the neural network algorithm is preferred.

And S7, respectively predicting future lightning stroke probability of each blade subarea and each tower barrel area through a prediction model based on the preset prediction step number.

Further, step S3, obtaining a plurality of historical corona times of each blade sub-region and each tower bobbin region based on a preset time interval, specifically includes the following steps:

step S31, obtaining a visual image of corona generation of the wind motor through ultraviolet imaging.

Preferably, the corona phenomenon has remarkable characteristics in the ultraviolet frequency band, and the ultraviolet imaging means are also adopted in the prior art to obtain the visual image of the corona phenomenon, and step S31 of the embodiment is a conventional application.

Step S32, acquiring a minimum bounding box of a corona region of the corona generating visual image through a target detection algorithm.

Preferably, the target detection algorithm of the present embodiment mainly includes a conventional target detection algorithm VJ, HOG, DPMDetector; a deep learning Two-stage target detection algorithm RCNN, SPPNet, fastRCNN, fasterRCNN; an algorithm FPN, cascadeRCNN for target detection Trick; deep learning one-stage target detection algorithm Yolo, X, SSD, retinaNet; a deep learning Anchor-free target detection algorithm CornerNet, centerNet, FCOS; based on the transducer target detection algorithm DETR.

Preferably, this embodiment prefers Yolo algorithm.

Step S33, judging whether a minimum bounding box exists in each blade subarea or each tower cylinder area respectively, and if the minimum bounding box exists, executing step S34.

Step S34, the vane subarea or the tower cylinder area with the minimum bounding box is respectively increased by one corona times.

And step S35, respectively acquiring all historical corona times of each blade subarea and each tower bobbin area based on a preset time interval.

Further, in step S32, a minimum bounding box of a corona region of the corona generating visual image is acquired through a target detection algorithm, and the steps specifically include the following steps:

In step S321, the corona generating visual image is divided into a plurality of grids by grids of a first preset density.

In step S322, a preset number of first-order bounding boxes are respectively predicted based on each grid, where each first-order bounding box includes at least one grid.

Step S323, defining that the corona area has the highest confidence coefficient, and respectively acquiring the confidence coefficient of each first-order bounding box.

In step S324, the first-order bounding box with the highest confidence is obtained from all the confidence coefficients, and is used as the second-order bounding box.

In step S325, the intersection ratios of the second-order bounding boxes and the first-order bounding boxes are calculated.

Preferably, the intersection ratio is the intersection of the second order bounding box with each first order bounding box, respectively, divided by the union of the second order bounding box with each first order bounding box, respectively, to obtain the ratio (which may be an area ratio).

In step S326, a second-order bounding box with the cross ratio greater than or equal to the preset ratio is reserved as a third-order bounding box.

In step S327, the third-order bounding box with the highest confidence is reserved as the fourth-order bounding box.

In step S328, the union of all the fourth-order bounding boxes is obtained, and the union is the minimum bounding box.

Preferably, each grid is used for predictionThe coordinates and width and height of the three-order bounding boxes, and the confidence of each three-order bounding box, i.e. the prediction/>, is required for each gridA value.

It will be appreciated that each grid requires predictionPersonal/>; Wherein/>Is the offset of the center of the third order bounding box relative to the grid,/>Is the proportion of the third-order boundary box relative to the picture after the size adjustment,/>The confidence of the grid is 1 or 0.

Preferably, confidence may be understood as the accuracy of whether there is a target within the current grid and a third order bounding box.

Illustrating: setting a target in the picture after the size adjustment, and setting the width and the height of the picture after the size adjustment asThen:

Dividing the picture into 7×7 (s×s) grids on average, if there is one grid located at the center of the target, the coordinates of the grid are Let the coordinates of the center of the target be/>Then according to/>The offset is calculated.

Preferably, in the actual detection, if the predicted third-order bounding box and the actual bounding box overlap perfectly, the value of the overlap ratio is 1. In the practical application process, the value of the preset ratio can be set to 0.5 to determine whether the predicted bounding box is correct, and the more accurate the bounding box is, the more positively correlated the cross ratio is.

Preferably, yolo algorithm also requires training of third order bounding boxes to improve accuracy of target detection.

And next, training the training model through a preset target training set, and iteratively adjusting the weight and bias of the training model by a certain number of times through a back propagation algorithm so as to reduce the value of a loss function of the training model.

Preferably, the loss function is。

Wherein,For/>/>, Of the gridWhether the three-order bounding boxes are responsible for the indication function of the target or not is judged to be 1 or 0; /(I)、/>、/>、/>、/>Corresponds to the/>, respectivelyPersonal/>Predicted values.

It is understood that the loss function includes a coordinate value deviation of the third-order bounding box, a deviation of the confidence, a deviation of the prediction probability (or a class deviation).

Wherein,For the mid-point loss of the third-order bounding box in coordinate value deviation,/>Is the third-order boundary box width and height loss in coordinate value deviation,/>For the deviation of confidence,/>To predict deviations in probability (or class deviations).

Wherein,For the localization error penalty, it is generally the case/>；/>The S multiplied by S grids are the grids; the number of the third-order bounding boxes; /(I) And/>For/>Estimated values of the midpoint abscissa and the ordinate of the three-order bounding boxes; /(I)And/>For/>Estimating the width and height of each third-order boundary box; /(I)For/>Confidence of the three-order bounding boxes; /(I)For/>An estimate of the confidence level of the three-order bounding box; /(I)For confidence prediction loss, it is generally the case/>；For/>Class probabilities of the three-order bounding boxes; /(I)For/>Estimating the class probability of each third-order bounding box; And/> In/>Correspondence/>。

It should be noted that, since each grid does not necessarily contain an object, if there is no object in the grid, this will result inThe value of (2) is 0, so that the gradient span in the subsequent back propagation algorithm is overlarge, so that the introduction/>To control the loss of the predicted position of the third-order bounding box and the introduction/>There is no loss of targets within the control single grid.

Note that the symbol meaning of the yolo algorithm is not mutually consistent with the symbol meaning of other positions in this embodiment.

Further, step S4, based on a preset time interval, respectively obtaining a plurality of historical hit times of each blade sub-region and each tower bobbin region, where the step specifically includes the following steps:

in step S41, a visual image of the lightning strike hit of the wind motor is obtained through continuous recording.

And S42, acquiring a lightning stroke termination point in the visual image hit by the lightning stroke through a sliding window algorithm.

And step S43, judging whether each blade subarea or each tower cylinder area has a lightning stroke termination point or not, and if so, executing step S44.

And step S44, respectively increasing the number of times of hitting on the blade subareas or the tower cylinder areas with lightning stroke termination points.

Step S45, all historical hit times of each blade subarea and each tower drum area are respectively obtained based on a preset time interval.

Further, in step S42, a lightning stroke termination point in the visual image of the lightning stroke hit is obtained through a sliding window algorithm, and the step specifically includes the following steps:

Step S421, binarizing the lightning strike hit visual image to obtain a black-and-white visual image.

Step S422, dividing the black-and-white visual image by the grid of the second preset density.

Step S423, the color of each grid is obtained, the grid with black color is assigned to 0, and the grid with white color is assigned to 1.

In step S424, each grid and eight adjacent grids are defined as a 3×3 window.

Step S425, traversing all grids through a 3×3 window in one step on all grids.

Step S426, deleting grids with all 1 'S and all 0' S in the 3×3 window based on all the traversal results.

In step S427, the reserved mesh is acquired and defined as a hit boundary.

In step S428, a closed end of the hit boundary is obtained, and the closed end is the lightning stroke termination point.

Preferably, one end of the hit boundary is contacted with the wind turbine, and the obtained detection result has a closing characteristic; the other end of the hit boundary faces the cloud end, is two approximately parallel straight lines or curves, and the other end of the hit boundary is not intersected, so that the hit boundary is not closed.

Further, in step S6, the historical lightning probability of each blade subregion and each tower barrel region is respectively learned and trained by a machine learning algorithm to obtain a prediction model, and the steps specifically include the following steps:

step S61, generating a plurality of acquisition time points based on a preset time interval.

Step S62, historical lightning stroke probability of each blade subarea and each tower cone area is respectively acquired based on all acquisition time points, and a data set is generated based on the historical accumulated probability of each blade subarea and the historical accumulated probability of each tower cone area.

In step S63, all the data sets are integrated to form a total data set.

And step S64, carrying out normalization processing on the total data set to obtain a normalized data set.

Preferably, the normalization method of zero-mean normalization (Z-score normalization) is preferred in this embodiment, and this method gives the mean (mean) and standard deviation (standard deviation) of the raw data to normalize the data, and the processed data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1. For the normalization method, in this embodiment, batch normalization (Batch Normalization) may be used, compared with simple normalization when training is performed on the previous neural network, only normalization is performed on the input layer data, but no normalization is performed on the middle layer, although normalization is performed on the data set of the input node, the data distribution of the input data after matrix multiplication is more likely to be changed greatly, and as the number of network layers of the hidden layer is deepened continuously, the change of the data distribution will be larger and larger, so that the normalization is performed on the middle layer of the neural network by batch normalization, and the training effect is better.

Step S65, dividing the normalized data set into a training set and a verification set according to a preset proportion.

Preferably, the preset ratio is typically 80%: the 20% ratio divides the image data into a training set and a verification set, namely 80% of the data is the training set and 20% of the data is the verification set.

Step S66, defining the topological relation of the neural network model, wherein the topological relation comprises an input layer, an implicit layer and an output layer which are connected in sequence in a signal mode.

Step S67, the training set is input to the input layer, iteration is carried out for preset times through the neural network model, and root mean square errors of the verification set and the current training result are respectively obtained based on each iteration.

Step S68, the minimum value in all root mean square errors is obtained, and a training result corresponding to the minimum value is obtained as a prediction model.

Preferably, the root mean square error is. Wherein/>Is root mean square error,/>For the number of training sets,/>For/>True value of the training set,/>For/>Training results after the training of the training sets are completed.

It should be noted that the root mean square error formula is only used for principle illustration, and the letter symbols and meanings of the root mean square error formula are not mutually communicated with other formulas in the embodiment.

Further, the neural network model is characterized by formula (1):

（1）。

Wherein, The method comprises the steps of presetting a neural network model; /(I)Is an input layer and/>，/>Is the input layer of the first/>Each input node corresponds to a training set,/>Is the input layer of the first/>Input node to hidden layer/>The weights of the input nodes; /(I)For connection to the hidden layer >Bias of the individual input nodes; Is a transfer function, and/> The numbers in brackets of the subscripts are the number of layers.

Preferably, the number in brackets of the symbol corner mark in formula (1) of this embodiment is the number of layers, e.gThe superscript (2) in (a) is the second layer, i.e. the hidden layer,/>The upper corner marks (1, 2) of (a) are from the first layer to the second layer, namely from the input layer to the hidden layer.

The sign meaning of the transfer function is not communicated with other places.

Preferably, training a model to train a neural network typically requires providing a large amount of data, i.e., a data set; the data sets are generally divided into three categories, namely training set (TRAINING SET), validation set (test set) and test set (test set) described above.

Wherein, one epoch (increase in number) is a process equal to one training with all samples in the training set, which means one forward propagation (forward pass) and one backward propagation (back pass); when the number of samples (i.e., training sets) of one epoch is too large, excessive time may be consumed for performing one training, and it is not necessary to use all data of the training set for each training, the whole training set needs to be divided into a plurality of small blocks, that is, a plurality of batches for performing the training; one epoch is made up of one or more latches, which are part of a training set, with only a portion of the data being used for each training process, i.e., one latch, and one iteration being used for training one latch.

Preferably, the neural network training specifically comprises a Perceptron (Perceptron) composed of two layers of neurons, an input layer receiving external input signals and transmitting the external input signals to an output layer, wherein the output layer is M-P neurons, and the output layer is provided withAs a step function, and given a training dataset, weight/>(/>=1, 2,..N) and training threshold/>Can be obtained through learning.

It should be noted that the step function is not interconnected with the symbolic meaning of the other formulas in the embodiment, and the step function is only schematically illustrated and does not participate in the calculation of the other formulas.

Preferably, the number of times of training the neural network in this embodiment may be set to 500 times.

Preferably, the learning rate of 1 st to 250 th epochs may be set to 0.01, the learning rate of 251 st to 325 th epochs may be set to 0.001, and the learning rate of 326 th to 1000 th epochs may be set to 0.0001.

It can be understood that the neural network training of this embodiment mainly includes the following ideas:

① Initializing weight and bias items in a network, and initializing parameter values (the weight of an output unit, the bias items, the weight of a hidden unit and the bias items are all parameters of a model) to obtain the output value of each layer of elements for activating forward propagation, thereby obtaining the value of a loss function.

② And activating forward propagation to obtain the output value of each layer and the expected value of the loss function of each layer.

③ According to the loss function, calculating an error term of the output unit and an error term of the hidden unit, calculating various errors, calculating a gradient of a parameter with respect to the loss function or calculating a partial derivative according to a calculus chain law. Solving partial derivatives for vectors or matrixes in the composite function, wherein the editing derivatives of the internal functions of the composite function are always multiplied left; for scalar bias derivative in the composite function, the derivative of the internal function of the composite function can be multiplied left or right.

④ The weights and bias terms in the neural network are updated.

⑤ And repeating ②~④ until the loss function is smaller than a preset threshold value or the iteration times are used up, and outputting the parameter at the moment to be the current optimal parameter.

In the embodiment, each blade is divided into areas, and each blade forms at least two blade subareas; dividing the tower barrel into areas to form at least two tower barrel areas; respectively acquiring a plurality of historical corona times of each blade subarea and each tower bobbin area based on a preset time interval; respectively acquiring a plurality of historical hit times of each blade subarea and each tower bobbin area based on a preset time interval; acquiring a ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning stroke probability; respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model; and respectively predicting future lightning stroke probability of each blade subarea and each tower barrel area through a prediction model based on the preset prediction step number. According to the embodiment, the characteristic that lightning is tip corona discharge is utilized, the wind turbine is divided into a plurality of small areas from the wind turbine, the corona times and the lightning hit times of each small area are identified and counted, the lightning strike probability is obtained by dividing the lightning hit times by the corona times, the previous multiple lightning strike probabilities are trained through a machine learning algorithm, a prediction model capable of predicting future lightning strike probability is obtained, and the future lightning strike probability of the wind turbine can be predicted through the prediction model. According to the method, environmental factors (such as weather law, wind speed distribution, altitude, wind motor height and model) of the wind motor are not considered, calculation and modeling are directly performed from two results of generating corona and receiving lightning, so that the calculated result and the modeled model both have the data characteristics (such as different data results due to different regional characteristics, weather characteristics, geographical characteristics and the like), the prediction method of the embodiment has good universality, the preorder factors of the wind motor are not considered, calculation and prediction are directly performed from the result with the characteristics, and the calculation load of the embodiment is small and no complex calculation steps and model are caused.

As shown in fig. 2, the present embodiment provides an embodiment of a lightning probability prediction apparatus of a wind turbine, and in the present embodiment, the lightning probability prediction apparatus is applied to the lightning probability prediction method in the above embodiment, and the apparatus includes a blade area dividing module 1, a tower area dividing module 2, a historical corona number obtaining module 3, a historical hit number obtaining module 4, a historical lightning probability defining module 5, a prediction model obtaining module 6, and a future lightning probability obtaining module 7 that are electrically connected in order.

The blade area dividing module 1 is used for dividing areas of each blade respectively, and each blade forms at least two blade subareas; the tower barrel region dividing module 2 is used for dividing the tower barrel region to form at least two tower barrel regions; the historical corona frequency acquisition module 3 is used for respectively acquiring a plurality of historical corona frequencies of each blade subarea and each tower bobbin area based on a preset time interval; the historical hit frequency acquisition module 4 is used for respectively acquiring a plurality of historical hit frequencies of each blade subarea and each tower bobbin area based on a preset time interval; the historical lightning probability definition module 5 is used for acquiring the ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning probability; the prediction model obtaining module 6 is used for respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model; the future lightning strike probability acquisition module 7 is used for respectively predicting the future lightning strike probability of each blade subarea and each tower barrel area through a prediction model based on the preset prediction steps.

Further, the historical corona frequency acquisition module 3 specifically comprises a first historical corona frequency acquisition sub-module, a second historical corona frequency acquisition sub-module, a third historical corona frequency acquisition sub-module, a fourth historical corona frequency acquisition sub-module and a fifth historical corona frequency acquisition sub-module which are electrically connected in sequence; the first historical corona frequency acquisition sub-module is electrically connected with the tower barrel area dividing module, and the fifth historical corona frequency acquisition sub-module is electrically connected with the historical hit frequency acquisition module.

The first historical corona frequency acquisition sub-module is used for acquiring a corona generation visual image of the wind motor through ultraviolet imaging; the second historical corona frequency acquisition submodule is used for acquiring a minimum bounding box of a corona region of the corona generation visual image through a target detection algorithm; the third historical corona frequency acquisition submodule is used for judging whether a minimum bounding box exists in each blade subarea or each tower bobbin area or not respectively; the fourth historical corona frequency acquisition submodule is used for respectively increasing the corona frequency once in a blade subarea or a tower bobbin area with the minimum bounding box if the minimum bounding box exists; the fifth historical corona frequency acquisition submodule is used for respectively acquiring all the historical corona frequency of each blade subarea and each tower bobbin area based on a preset time interval.

Further, the second historical corona frequency acquisition submodule specifically comprises a first historical corona frequency acquisition unit, a second historical corona frequency acquisition unit, a third historical corona frequency acquisition unit, a fourth historical corona frequency acquisition unit, a fifth historical corona frequency acquisition unit, a sixth historical corona frequency acquisition unit, a seventh historical corona frequency acquisition unit and an eighth historical corona frequency acquisition unit which are electrically connected in sequence; the first historical corona frequency acquisition unit is electrically connected with the first historical corona frequency acquisition submodule, and the eighth historical corona frequency acquisition unit is electrically connected with the third historical corona frequency acquisition submodule.

The first historical corona frequency acquisition unit is used for dividing the corona occurrence visual image into a plurality of grids through grids with a first preset density; the second historical corona frequency acquisition unit is used for respectively predicting a preset number of first-order boundary boxes based on each grid, and each first-order boundary box comprises at least one grid; the third historical corona frequency acquisition unit is used for defining that the corona region has the highest confidence coefficient and respectively acquiring the confidence coefficient of each first-order bounding box; the fourth historical corona frequency acquisition unit is used for acquiring a first-order boundary box with highest confidence coefficient from all confidence coefficients as a second-order boundary box; the fifth historical corona frequency acquisition unit is used for calculating the intersection ratio of the second-order boundary frame and the first-order boundary frame respectively; the sixth historical corona frequency acquisition unit is used for reserving a second-order boundary box with the cross ratio being greater than or equal to a preset ratio as a third-order boundary box; the seventh historical corona frequency acquisition unit is used for reserving a third-order boundary box with highest confidence coefficient as a fourth-order boundary box; the eighth historical corona frequency acquisition unit is used for acquiring the union of all the fourth-order bounding boxes, and the union is the minimum bounding box.

Further, the history hit frequency acquisition module specifically comprises a first history hit frequency acquisition sub-module, a second history hit frequency acquisition sub-module, a third history hit frequency acquisition sub-module, a fourth history hit frequency acquisition sub-module and a fifth history hit frequency acquisition sub-module which are electrically connected in sequence; the first historical hit frequency acquisition sub-module is electrically connected with the fifth historical corona frequency acquisition sub-module, and the fifth historical hit frequency acquisition sub-module is electrically connected with the historical lightning probability definition module.

The first historical hit frequency acquisition sub-module is used for continuously recording and acquiring lightning hit visual images of the wind motor; the second historical hit frequency acquisition sub-module is used for acquiring lightning stroke termination points in the lightning stroke hit visual image through a sliding window algorithm; the third historical hit frequency acquisition submodule is used for judging whether lightning stroke termination points exist in each blade subarea or each tower bobbin area respectively; the fourth historical hit frequency acquisition submodule is used for respectively increasing hit frequency of a blade subarea or a tower bobbin area with a lightning stroke termination point if the lightning stroke termination point exists; the fifth historical hit frequency acquisition submodule is used for respectively acquiring all historical hit frequencies of each blade subarea and each tower bobbin area based on a preset time interval.

Further, the second history hit frequency acquisition submodule specifically includes a first history hit frequency acquisition unit, a second history hit frequency acquisition unit, a third history hit frequency acquisition unit, a fourth history hit frequency acquisition unit, a fifth history hit frequency acquisition unit, a sixth history hit frequency acquisition unit, a seventh history hit frequency acquisition unit, and an eighth history hit frequency acquisition unit, which are electrically connected in sequence; the first history hit frequency acquisition unit is electrically connected with the first history hit frequency acquisition submodule, and the eighth history hit frequency acquisition unit is electrically connected with the third history hit frequency acquisition submodule.

The first historical hit frequency acquisition unit is used for performing binarization processing on the lightning strike hit visual image to obtain a black-white visual image; the second history hit frequency acquisition unit is used for dividing black-and-white visual images through grids with a second preset density; the third historical hit frequency acquisition unit is used for respectively acquiring the color of each grid, assigning 0 to the grid with black color and 1 to the grid with white color; a fourth history hit number acquisition unit for defining each grid and eight adjacent grids as a 3×3 window; the fifth historical hit number acquisition unit is used for traversing all grids through a 3×3 window by taking one grid as one step on all grids; the sixth historical hit frequency acquisition unit is used for deleting grids with all the values of 1 and grids with all the values of 0 in the 3 multiplied by 3 window based on all the traversal results; a seventh history hit number acquisition unit for acquiring the reserved grid and defining the reserved grid as a hit boundary; the eighth historical hit frequency acquisition unit is used for acquiring a closed end of the hit boundary, wherein the closed end is a lightning stroke termination point.

Further, the prediction model acquisition module specifically comprises a first prediction model acquisition sub-module, a second prediction model acquisition sub-module, a third prediction model acquisition sub-module, a fourth prediction model acquisition sub-module, a fifth prediction model acquisition sub-module, a sixth prediction model acquisition sub-module, a seventh prediction model acquisition sub-module and an eighth prediction model acquisition sub-module which are electrically connected in sequence; the first prediction model acquisition sub-module is electrically connected with the fifth historical hit frequency acquisition sub-module, and the eighth prediction model acquisition sub-module is electrically connected with the future lightning stroke probability acquisition module.

The first prediction model acquisition submodule is used for generating a plurality of acquisition time points based on a preset time interval; the second prediction model acquisition submodule is used for respectively acquiring historical lightning stroke probability of each blade subarea and each tower cone area based on all acquisition time points, and generating a data set based on the historical cumulative probability of each blade subarea and the historical cumulative probability of each tower cone area; the third prediction model acquisition sub-module is used for integrating all data sets to form a total data set; the fourth prediction model acquisition sub-module is used for carrying out normalization processing on the total data set to obtain a normalized data set; the fifth prediction model acquisition submodule is used for dividing the normalized data set into a training set and a verification set according to a preset proportion; the sixth prediction model acquisition submodule is used for defining a topological relation of the neural network model, wherein the topological relation comprises an input layer, an implicit layer and an output layer which are connected in sequence in a signal manner; the seventh prediction model obtaining submodule is used for inputting the training set into the input layer, carrying out iteration for preset times through the neural network model, and respectively obtaining root mean square errors of the verification set and the current training result based on each iteration; the eighth prediction model obtaining sub-module is used for obtaining the minimum value in all root mean square errors and obtaining a training result corresponding to the minimum value as a prediction model.

Further, the neural network model in the sixth predictive model acquisition sub-module is characterized by the formula (1):

（1）。

Wherein, The method comprises the steps of presetting a neural network model; /(I)Is an input layer and/>，/>Is the input layer of the first/>Each input node corresponds to a training set,/>Is the input layer of the first/>Input node to hidden layer/>The weights of the input nodes; /(I)For connection to the hidden layer >Bias of the individual input nodes; Is a transfer function, and/> The numbers in brackets of the subscripts are the number of layers.

It should be noted that, the present embodiment is a functional module embodiment based on the foregoing method embodiment, and expansion, preference, illustration, notice and the like of the present embodiment may be referred to the foregoing method embodiment, which is not repeated herein.

In the embodiment, each blade is divided into areas, and each blade forms at least two blade subareas; dividing the tower barrel into areas to form at least two tower barrel areas; respectively acquiring a plurality of historical corona times of each blade subarea and each tower bobbin area based on a preset time interval; respectively acquiring a plurality of historical hit times of each blade subarea and each tower bobbin area based on a preset time interval; acquiring a ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning stroke probability; respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model; and respectively predicting future lightning stroke probability of each blade subarea and each tower barrel area through a prediction model based on the preset prediction step number. According to the embodiment, the characteristic that lightning is tip corona discharge is utilized, the wind turbine is divided into a plurality of small areas from the wind turbine, the corona times and the lightning hit times of each small area are identified and counted, the lightning strike probability is obtained by dividing the lightning hit times by the corona times, the previous multiple lightning strike probabilities are trained through a machine learning algorithm, a prediction model capable of predicting future lightning strike probability is obtained, and the future lightning strike probability of the wind turbine can be predicted through the prediction model. According to the method, environmental factors (such as weather law, wind speed distribution, altitude, wind motor height and model) of the wind motor are not considered, calculation and modeling are directly performed from two results of generating corona and receiving lightning, so that the calculated result and the modeled model both have the data characteristics (such as different data results due to different regional characteristics, weather characteristics, geographical characteristics and the like), the prediction method of the embodiment has good universality, the preorder factors of the wind motor are not considered, calculation and prediction are directly performed from the result with the characteristics, and the calculation load of the embodiment is small and no complex calculation steps and model are caused.

Fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present application. As shown in fig. 3, the electronic device 8 includes a processor 81 and a memory 82 coupled to the processor 81.

The memory 82 stores program instructions for implementing the wind turbine lightning strike probability prediction method of any of the embodiments described above.

The processor 81 is configured to execute program instructions stored in the memory 82 for performing a lightning probability prediction of the wind turbine.

The processor 81 may also be referred to as a CPU (Central Processing Unit ). The processor 81 may be an integrated circuit chip with signal processing capabilities. Processor 81 may also be a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

Further, fig. 4 is a schematic structural diagram of a storage medium according to an embodiment of the present application, and referring to fig. 4, the storage medium 9 according to an embodiment of the present application stores a program instruction 91 capable of implementing all the methods described above, where the program instruction 91 may be stored in the storage medium in the form of a software product, and includes several instructions for making a computer device (which may be a personal computer, a server, or a network device) or a processor (processor) execute all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, an optical disk, or other various media capable of storing program codes, or a terminal device such as a computer, a server, a mobile phone, a tablet, or the like.

In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of elements is merely a logical functional division, and there may be additional divisions of actual implementation, e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other forms.

In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units. The foregoing is only the embodiments of the present application, and the patent scope of the application is not limited thereto, but is also covered by the patent protection scope of the application, as long as the equivalent structure or equivalent flow changes made by the description and the drawings of the application or the direct or indirect application in other related technical fields are adopted.

The embodiments of the present application have been described in detail above, but they are merely examples, and the present application is not limited to the above-described embodiments. It will be apparent to those skilled in the art that any equivalent modifications or substitutions to the present application are within the scope of the present application, and therefore, equivalent changes and modifications, improvements, etc., which do not depart from the spirit and scope of the present application, are intended to be covered by the present application.

Claims (4)

1. The wind turbine comprises a tower drum with a fixed frame arranged in a preset area and three blades rotatably arranged on the tower drum, and the wind turbine lightning probability prediction method is characterized by comprising the following steps of:

dividing the area of each blade respectively, wherein each blade forms at least two blade subareas;

Dividing the tower barrel into areas to form at least two tower barrel areas;

Respectively acquiring a plurality of historical corona times of each blade subarea and each tower bobbin area based on a preset time interval;

Respectively acquiring a plurality of historical hit times of each blade subarea and each tower cone area based on the preset time interval;

acquiring the ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning stroke probability;

Respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model;

predicting future lightning stroke probability of each blade subarea and each tower cone area through the prediction model based on a preset prediction step number;

based on a preset time interval, respectively obtaining a plurality of historical corona times of each blade subarea and each tower bobbin area, including:

Acquiring a corona generation visual image of the wind motor through ultraviolet imaging;

acquiring a minimum bounding box of a corona region of the corona occurrence visual image through a target detection algorithm;

judging whether the minimum bounding box exists in each blade subarea or each tower cone area or not respectively;

If the minimum bounding box exists, respectively increasing the number of times of corona once in a blade subarea or a tower cone area in which the minimum bounding box exists;

Acquiring all historical corona times of each blade subarea and each tower cone area based on the preset time interval;

acquiring a minimum bounding box of a corona region of the corona occurrence visual image through a target detection algorithm, wherein the minimum bounding box comprises:

Dividing the corona generation visual image into a plurality of grids through grids with a first preset density;

predicting a preset number of first-order bounding boxes based on each grid respectively, wherein each first-order bounding box comprises at least one grid;

Defining the highest confidence coefficient of the corona region, and respectively acquiring the confidence coefficient of each first-order boundary box;

Acquiring a first-order boundary box with highest confidence coefficient from all confidence coefficients as a second-order boundary box;

Calculating the intersection ratio of the second-order boundary frame and the first-order boundary frame respectively;

Reserving a second-order boundary box with the cross ratio being greater than or equal to a preset ratio as a third-order boundary box;

The third-order boundary frame with the highest confidence coefficient is reserved and used as a fourth-order boundary frame;

acquiring a union set of all fourth-order bounding boxes, wherein the union set is the minimum bounding box;

Based on the preset time interval, respectively obtaining a plurality of historical hit times of each blade subarea and each tower cone area, wherein the method comprises the following steps:

Obtaining a lightning strike hitting visual image of the wind turbine through continuous recording;

Acquiring a lightning stroke termination point in the lightning stroke hitting visual image through a sliding window algorithm;

Judging whether the lightning stroke termination point exists in each blade subarea or each tower cylinder area or not respectively;

If the lightning stroke exists, respectively increasing the number of times of hitting the blade subareas or the tower cylinder areas with the lightning stroke end points for one time;

Acquiring all historical hit times of each blade subarea and each tower cone area based on the preset time interval;

Acquiring a lightning stroke termination point in the lightning stroke hitting visual image through a sliding window algorithm, wherein the method comprises the following steps:

Binarizing the lightning strike hit visual image to obtain a black-white visual image;

dividing the black-and-white visual image through a grid with a second preset density;

Respectively obtaining the color of each grid, assigning 0 to the grid with black color and 1 to the grid with white color;

Defining each grid and eight adjacent grids as a3 x3 window;

traversing all grids through the 3 x 3 window in one step on all grids;

Deleting grids with all 1 values and grids with all 0 values in the 3 multiplied by 3 window based on all traversal results;

Acquiring reserved grids and defining the reserved grids as hit boundaries;

acquiring a closed end of the hit boundary, wherein the closed end is the lightning stroke termination point;

Respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model, wherein the method comprises the following steps of:

Generating a plurality of acquisition time points based on a preset time interval;

Acquiring historical lightning stroke probability of each blade subarea and each tower cone area based on all acquisition time points, and generating a data set based on the historical accumulated probability of each blade subarea and the historical accumulated probability of each tower cone area;

integrating all the data sets to form a total data set;

normalizing the total data set to obtain a normalized data set;

dividing the normalized data set into a training set and a verification set according to a preset proportion;

defining a topological relation of a neural network model, wherein the topological relation comprises an input layer, an implicit layer and an output layer which are sequentially connected through signals;

Inputting the training set into the input layer, carrying out iteration for preset times through the neural network model, and respectively acquiring root mean square errors of the training set and the current training result based on each iteration;

Obtaining the minimum value in all root mean square errors, and obtaining a training result corresponding to the minimum value as the prediction model;

the neural network model is characterized by formula (1):

（1）；

Wherein, Modeling the neural network; /(I)Is the input layer and/>，/>For the input layer/>Each input node corresponds to a training set,/>For the input layer/>The/>, of the input nodes to the hidden layerThe weights of the input nodes; /(I)For connecting to the hidden layerBias of the individual input nodes; /(I)Is a transfer function, and/>The numbers in brackets of the subscripts are the number of layers.

2. A lightning probability prediction apparatus of a wind turbine, the lightning probability prediction apparatus of a wind turbine being applied to the lightning probability prediction method according to claim 1, characterized in that the lightning probability prediction apparatus comprises:

The blade area dividing module is used for dividing areas of each blade respectively, and each blade forms at least two blade subareas;

The tower barrel region dividing module is used for dividing the tower barrel region to form at least two tower barrel regions;

The historical corona frequency acquisition module is used for respectively acquiring a plurality of historical corona frequencies of each blade subarea and each tower bobbin area based on a preset time interval;

the historical hit frequency acquisition module is used for respectively acquiring a plurality of historical hit frequencies of each blade subarea and each tower bobbin area based on the preset time interval;

The historical lightning probability definition module is used for acquiring the ratio of the historical corona times to the historical hit times based on the same blade subarea or the same tower bobbin area, and defining the ratio as the historical lightning probability;

the prediction model acquisition module is used for respectively learning and training the historical lightning stroke probability of each blade subarea and each tower cone area through a machine learning algorithm to obtain a prediction model;

And the future lightning strike probability acquisition module is used for respectively predicting the future lightning strike probability of each blade subarea and each tower bobbin area through the prediction model based on the preset prediction step number.

3. An electronic device comprising a processor, and a memory coupled to the processor, the memory storing program instructions executable by the processor; the processor, when executing the program instructions stored in the memory, implements the lightning probability prediction method of claim 1.

4. A storage medium having stored therein program instructions which, when executed by a processor, implement a method for predicting probability of lightning strike according to claim 1.