Design method of lightning arrester layout position and wind power blade lightning protection system

Abstract

The invention belongs to the technical field of wind power blade lightning protection, and discloses a lightning arrester layout position design method which comprises a data storage module, a data analysis module, a position layout prediction module and an upper computer. The data storage module is used for collecting and storing artificial historical experience data and electric field intensity simulation data around the lightning arrester, then transmitting the data to the data analysis module for conversion, performing normalization processing on the obtained original data, eliminating abnormal values, and performing data enhancement on the original data through an overlapping sampling method; the data analysis module transmits the processed data to a position layout prediction module, and the position layout prediction module transmits an enhanced data set to a multi-channel weighted fusion convolutional network model for position layout prediction; and finally, transmitting the data to an upper computer for display, summarizing a layout rule, and selecting optimal position layout data of the lightning arrester. And the lightning receiving probability and the safety performance of fan operation are improved.

Description

Technical Field

[0001] The present invention relates to a design method for the layout position of a lightning arrester and a lightning protection system for a wind turbine blade, belonging to the technical field of lightning protection for wind turbine blades. Background Art

[0002] With the continuous development of the wind power industry, the blades of wind turbines are getting longer and longer, and the damage events of lightning to components such as wind turbine blades are also increasing continuously. Since the cost of wind turbine blades is high and the maintenance cost is large, once damaged by lightning strikes, it will cause huge economic losses. Therefore, reasonable lightning protection design for the blades is required.

[0003] The layout design of the lightning arrester for a wind turbine blade is a key link in the lightning protection of a wind turbine. The main function of the lightning arrester is to attract lightning and introduce it into the ground to protect the wind turbine from lightning damage. The layout of the lightning arrester needs to consider the influence of the dynamic changes of the blade on the space electric potential and the induced electric field intensity at different positions of the blade under the lightning environment.

[0004] The layout design of the lightning arrester for a wind turbine blade usually includes the following key points. Design of the tip lightning arrester: The tip is the part most easily struck by lightning, so the design of the tip lightning arrester is crucial. Usually, a lightning arrester made of metal material is fixed at the tip through the blade fixing flange, and the bottom connecting flange is connected to the blade root through the down-lead laid on the truss. Design of the lightning arrester on the blade surface: In addition to the tip lightning arrester, multiple groups of lightning arresters are also arranged on the blade surface. These lightning arresters are symmetrically arranged along the axial direction of the blade on the outer surface of the blade and are connected to the tip lightning arrester through a conductive shunt strip to enhance the drainage ability. Installation device and method: The installation of the lightning arrester requires precise positioning and firm connection. The installation device includes an upper lightning arrester installation component and a lower lightning arrester installation component, and the accurate installation of the lightning arrester is ensured through a positioning plug and a guide shaft.

[0005] In the prior art, the reasonable layout of the lightning arrester mainly relies on the historical experience of engineers. As the length of the blade increases, the reliability decreases. If the layout of the lightning arrester fails, it will cause the failure of the lightning protection system of the blade, and then lead to the damage of the wind turbine by lightning strikes.

[0006] After retrieval, CN202110322278.0 discloses a wind turbine blade, including a blade body, a first main beam, a lightning protection layer, a paint layer, a tip lightning arrester and a second main beam. The blade body has a tip, a blade body and a root arranged in sequence; a carbon fiber main beam is arranged in the blade body and the root, and the second main beam is arranged at least in the blade body and is made of a non-conductive fiber material and does not participate in receiving the electricity of lightning strikes. When the wind turbine blade rotates, it can withstand a greater torque and the tip area will not be broken; no carbon fiber main beam is arranged in the tip area, and no lightning protection layer and paint layer are arranged on the outer wall surface of the tip area.

[0007] CN202011186545.8 discloses a wind turbine blade, comprising a main beam cap, a leading edge bonding component and a trailing edge bonding component. The front end of the main beam cap is provided with a leading edge bonding component, and the rear end of the main beam cap is provided with a trailing edge bonding component. The front end of the leading edge bonding component is pre-embedded with evenly distributed pre-embedded holes for installing a blade root flange. The blade root flange is connected to a blade root baffle through resin putty and a baffle reinforcement agent. The inner wall of the main beam cap is also installed with a leading edge web and a trailing edge web for supporting the main beam cap.

[0008] The above technical solutions are mainly designed for the structure of the lightning rod. The location layout of the lightning rod is generally based on a preliminary range obtained by numerical calculation, and then the specific location is determined by relying on the historical experience of engineers, which is often unreliable.

[0009] Therefore, it is very necessary to invent a method that can summarize the layout rules of lightning arresters in wind power lightning protection systems and ultimately select the optimal new blade-shaped lightning arrester position layout data to improve the lightning arrestment probability and the safety performance of wind turbine operation. Summary of the invention

[0010] In view of the above-mentioned prior art, the reasonable arrangement of lightning rods mainly relies on the historical experience of engineers. As the length of the blades increases and the structure changes, the reliability decreases accordingly. The arrangement of lightning rods based on experience alone cannot meet the design requirements of the blades, which will cause the lightning protection system of the blades to fail, and further lead to the risk of the wind turbine being damaged by lightning. The present invention discloses a method for designing the layout position of lightning rods, which accumulates data samples of successful design cases in the past, adopts a deep learning method, combines the simulation design sample data of the new blade type, and intelligently summarizes the rules through fusion training, and finally selects the optimal lightning rod position layout data for the new blade type.

[0011] In order to achieve the above object, the technical solution adopted by the present invention is:

[0012] Disclosed is a method for designing the layout position of a lightning receptor, comprising a data storage module, a data analysis module, a position layout prediction module and a host computer; the data storage module is used for collecting and storing artificial historical experience data and electric field strength simulation data around the lightning receptor, and then transmitting the data to the data analysis module for conversion, normalizing the obtained raw data, eliminating abnormal values, and enhancing the raw data by an overlapping sampling method; the data analysis module transmits the processed data to the position layout prediction module, and the position layout prediction module transmits the enhanced data set to a multi-channel weighted fusion convolutional network model to perform position layout prediction; finally, the data is transmitted to the host computer for display, the layout rules are summarized, and the optimal lightning receptor position layout data is selected.

[0013] Further, the data analysis module is connected to the data acquisition module; the data analysis module includes a data preprocessing algorithm, which preprocesses the artificial historical experience data and the simulated data of the electric field intensity around the lightning arrester, and detects and removes abnormal mutation signals.

[0014] Further, the data preprocessing algorithm includes the following steps;

[0015] S1: Normalize the data and map the data to the range of 0 - 1 for processing;

[0016] S2: Detect and remove abnormal mutation values from the normalized data, reduce the computational complexity, and reduce the influence of abnormal mutation values on the false judgment of fault warning and diagnosis;

[0017] S3: Perform overlapping sampling and data enhancement on the normalized data to highlight data features.

[0018] Further, the position layout prediction module includes a deep learning algorithm, and the network model adopted by the deep learning algorithm is a multi-channel weighted fusion convolutional network; different receptive fields are formed by units with different convolutional kernel sizes in each channel for feature extraction, and the size of the receptive field changes with the change of the convolutional kernel. The relationship between them is as follows: where n represents the number of convolutional pooling layers; rf n-1 represents the receptive field size of the n - 1 layer; rf n represents the receptive field size of the n layer; s n-1 represents the stride of the n - 1 layer; k n-1 represents the convolution size of the n - 1 layer; rf last represents the size of the receptive field of the last layer, which is equal to the size of the convolutional kernel k last of this layer.

[0019] Further, after feature extraction through different convolutional kernels, the outputs obtained from each channel are: where is the output of the l - th convolutional layer of channel j; is the i - th feature input of the (l - 1)-th convolution of channel j, and there are k feature inputs in total; is the convolutional kernel of the (l - 1)-th layer of channel j.

[0020] Further, after feature extraction based on the dynamic receptive field, the kurtosis value is used as the weight to measure the difference in importance. Assume that k j represents the kurtosis value of the vibration signal on channel j, and m represents the number of channels. Then the weight of this channel is: After obtaining the weights of each channel, it is necessary to perform weighted fusion on them: where X lis the output of the l-th convolutional layer; is the feature output of the (l-1)-th convolutional layer in channel j; is the bias of the (l-1)-th convolutional layer in channel j; is the weight of the (l-1)-th convolutional layer in channel j; f l-1 (x) is the activation function of the (l-1)-th convolutional layer; m is the number of channels.

[0021] Furthermore, it also includes a reference data test platform for establishing a reference database, transmitting the established reference database to the position layout prediction module to train the multi-channel weighted fusion convolutional network multiple times, and saving the optimal training model parameters.

[0022] Another object of the present invention is a lightning protection system for wind turbine blades manufactured using the above design method.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] The method for designing the layout position of the lightning arrester of the lightning protection system of the present invention is used in the field of wind power generation. Sufficient artificial historical experience data and simulation data of the electric field intensity around the lightning arrester are collected and stored, and then the data is transmitted to the data analysis module and the position layout prediction module. The data analysis module is equipped with a data preprocessing algorithm, and the position layout prediction module is equipped with a deep learning algorithm. First, the signal is normalized and mapped to the range of 0-1, and then the preprocessed data is transmitted to the deep learning network model to predict the layout position of the lightning arrester of the lightning protection system, summarize the rules, and the network model can adopt a convolutional algorithm with fewer layers.

[0025] The method for designing the layout position of the lightning arrester of the lightning protection system of the present invention predicts the layout position of the lightning arrester of the lightning protection system in real time through a deep learning fault diagnosis algorithm; finally, the upper computer responds to the real-time state of the lightning protection system. Compared with the existing technology, the present invention can summarize the rules for the layout of the lightning arrester of the wind power lightning protection system, and finally select the optimal layout data of the lightning arrester for the new blade type, improving the lightning receiving probability and the safety performance of the fan operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a schematic diagram of the method for designing the layout position of the lightning arrester of the lightning protection system described in the present invention.

[0027] Figure 2 is a schematic diagram of the multi-channel weighted fusion convolutional network of the method for designing the layout position of the lightning arrester of the lightning protection system described in the present invention.

[0028] Figure 3 is a schematic flow diagram of the method for designing the layout position of the lightning arrester of the lightning protection system described in the present invention.

[0029] Figure 4This is a schematic diagram of the specific structure of the lightning protection system lightning arrester layout position design method of the present invention.

[0030] Among them, 1 - data storage module, 2 - data analysis module, 3 - position layout prediction module, 4 - host computer, 5 - reference data test platform. Specific implementation manner

[0031] For the convenience of elaborating and understanding the present invention, the following combines the attached Figures 1-4 A detailed description of the embodiments of the present invention will be given.

[0032] The implementation scheme of the design method for the layout position of the lightning arrester of the present invention includes the following steps:

[0033] The first step: Collect and store sufficient artificial historical experience data and simulation data of the electric field intensity around the lightning arrester.

[0034] The second step: Perform data normalization and enhancement preprocessing through the data analysis module.

[0035] The third step: Transmit the processed data to the position layout prediction module, and use the reference database to train the deep learning network model multiple times to summarize the position layout law of the lightning arrester of the wind turbine blade lightning protection system.

[0036] The fourth step: Reflect the real-time prediction result on the host computer. The specific implementation is described as follows.

[0037] Embodiment 1

[0038] As Figure 1 and Figure 4 shown, the lightning arrester layout position design method of this embodiment includes a data storage module 1, a data analysis module 2, a position layout prediction module 3, a host computer 4, and a data test platform 5. The data storage module 1 is connected to the data analysis module 2. The data storage module 1 is used to collect and store artificial historical experience data of the lightning arrester layout positions of different types of blades and simulation data of the electric field intensity around the lightning arrester. The data analysis module 1 is connected to the data acquisition module for data processing; the data analysis module 1 includes a data preprocessing algorithm, and the data preprocessing algorithm preprocesses the artificial historical experience data and the simulation data of the electric field intensity around the lightning arrester, detects and removes abnormal mutation signals. The data analysis module 2 transmits the processed data to the position layout prediction module 3, and the position layout prediction module 3 conveys the enhanced data set to the trained multi-channel weighted fusion deep learning algorithm and convolutional network model for position layout prediction, and obtains a weight error ≤ threshold. The host computer 4 is connected to the position layout prediction module 3. The host computer is used to respond to the real-time state and is also used to analyze the real-time state information. The flow of the lightning protection system lightning arrester layout position design method is as Figure 3As shown. Finally, it is transmitted to the host computer 4 for display, the layout rules are summarized, and the optimal lightning arrester position layout data (reasonable lightning arrester position parameters) are selected. The reference data test platform 5 is used to establish a reference database, and the established reference database is transmitted to the position layout prediction module 3 to train the multi-channel weighted fusion convolutional network multiple times, and the optimal training model parameters are saved. The reference database can collect the normal service data of each wind farm throughout the year.

[0039] The preprocessing algorithm includes: S1. Normalize the data and map the data to the range of 0-1; S2. Detect and remove abnormal mutation values from the normalized data to reduce the computational complexity and the influence of abnormal mutation values on the false judgment of fault warning and diagnosis; S3. Perform overlapping sampling and data enhancement on the normalized data to highlight the data features. The deep learning algorithm uses a network model that is a multi-channel weighted fusion convolutional network, and the network model structure is as Figure 2 shown.

[0040] The specific process is first data division, and then immediately multi-channel weighted convolution can extract the prominent features in the signal. The multi-channel convolution uses units with different kernel sizes in each channel to form different receptive fields for feature extraction. The size of the receptive field changes with the change of the convolution kernel, and the following relationship exists between the two: In the expression: n represents the number of convolutional pooling layers; rf n-1 represents the receptive field size of the n-1 layer; rf n represents the receptive field size of the n layer; s n-1 represents the stride of the n-1 layer; k n-1 represents the convolution size of the n-1 layer; rf last represents the size of the receptive field of the last layer, which is equal to the size of the convolution kernel k last of this layer.

[0041] After feature extraction through different convolution kernels, the output obtained for each channel is: In the expression: is the output of the l-th convolutional layer of channel j; is the i-th feature input of the (l-1)-th convolution of channel j, and there are k feature inputs in total; is the convolution kernel of the (l-1)-th layer of channel j.

[0042] After feature extraction based on the dynamic receptive field, the kurtosis value is used as the weight to measure the difference in importance. Assuming that k j represents the kurtosis value of the vibration signal on channel j and m represents the number of channels, then the weight of this channel is: After obtaining the weights of each channel, it is necessary to perform weighted fusion on them: In the expression: X lis the output of the l-th convolutional layer; is the feature output of the (l - 1)-th convolutional layer in channel j; is the bias of the (l - 1)-th convolutional layer in channel j; is the weight of the (l - 1)-th convolutional layer in channel j; f l-1 (x) is the activation function of the (l - 1)-th convolutional layer; m is the number of channels.

[0043] A fully connected layer is connected after the last pooling layer. The extracted features are flattened into a row, and then operations are performed using fully connected neurons.

[0044] In summary, the original data is first processed into multiple samples of the same length. The processed samples are divided into a training set and a test set according to a ratio of 8:2. First, the training set is fed into the network model for model training, and then the test set is used for verification. The model parameters with the best effect are determined. Subsequently, the collected artificial historical experience data and the simulated data of the electric field intensity around the lightning arrester are fed into the network model to summarize the rules for the layout of the lightning arrester in the wind power lightning protection system, and finally, the optimal layout data of the new blade type of the lightning arrester is selected.

[0045] The method for designing the layout position of the lightning arrester in the lightning protection system in the solution of the present invention is applied to the field of wind power generation. Sufficient artificial historical experience data and simulated data of the electric field intensity around the lightning arrester are collected and stored, and then the data is transmitted to the data analysis module and the position layout prediction module. The data analysis module has a data preprocessing algorithm, and the position layout prediction module has a deep learning algorithm. First, the signal is normalized and mapped to the range of 0 - 1, and then the preprocessed data is transmitted to the deep learning network model to predict the position layout of the lightning arrester in the lightning protection system and summarize the rules. The network model can adopt a convolutional algorithm with fewer layers.

[0046] The above are only embodiments of the present invention. The invention is not limited to the fields involved in this embodiment. Common general knowledge such as specific structures and characteristics in the solution is not described in detail here. It should be noted that for those skilled in the art, without departing from the content of the present invention, several deformations and improvements can still be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be based on the content of its claims, and the specific implementation manners described in the specification can be used to interpret the content of the claims.

Claims

1. A method for designing the layout position of a lightning arrester, characterized in that: It includes a data storage module, a data analysis module, a location layout prediction module and a host computer; The data storage module is used to collect and store artificial historical experience data and electric field strength simulation data around the lightning arrester, and then transmit them to the data analysis module for conversion, normalize the obtained raw data, eliminate outliers, and enhance the raw data by overlapping sampling method; The data analysis module transmits the processed data to the location layout prediction module, which transmits the enhanced data set to the multi-channel weighted fusion convolutional network model for location layout prediction; finally, it is transmitted to the host computer for display, summarizes the layout rules, and selects the optimal lightning arrester location layout data.

2. The method for designing the layout position of the lightning arrester according to claim 1, characterized in that: The data analysis module is connected to the data acquisition module; the data analysis module includes a data preprocessing algorithm, which preprocesses artificial historical experience data and electric field strength simulation data around the lightning arrester to detect and remove abnormal mutation signals.

3. The method for designing the layout position of the lightning arrester according to claim 2, characterized in that: The data preprocessing algorithm comprises the following steps: S1: Normalize the data and map it to the range of 0-1; S2: Detect and remove abnormal mutation values ​​from the normalized data to reduce the computational complexity and the impact of abnormal mutation values ​​on fault warning and diagnosis errors; S3: Overlap sampling is performed on the data processed in step S2 to enhance the data and highlight the data features.

4. The method for designing the layout position of the lightning receptor according to claim 3, characterized in that: The location layout prediction module includes a deep learning algorithm, which uses a multi-channel weighted fusion convolutional network; units with different convolution kernel sizes are used in each channel to form different receptive fields for feature extraction. The size of the receptive field changes with the change of the convolution kernel, and the two have the following relationship: Where n represents the number of convolutional pooling layers; rf n-1 Represents the receptive field size of the n-1th layer; rf n Represents the size of the receptive field of layer n; s n-1 represents the step length of n-1 layers; k n-1 Represents the size of n-1 convolution layers; rf last Represents the size of the receptive field band of the last layer, which is equal to the convolution kernel k of this layer last size.

5. The method for designing the layout position of the lightning receptor according to claim 4, characterized in that: After feature extraction through different convolution kernels, the output of each channel is: in is the output of the lth convolutional layer of channel j; The i-th feature input of the l-1th convolution layer of channel j, with a total of k feature inputs; is the convolution kernel of the l-1th layer of channel j.

6. The method for designing the layout position of the lightning receptor according to claim 5, characterized in that: After feature extraction based on the dynamic receptive field, the kurtosis value is used as the weight to measure the difference in importance. Assuming k j represents the kurtosis value of the vibration signal on channel j, m represents the number of channels, and the weight of the channel is: After obtaining the weights of each channel, they need to be weighted fused: Where X l is the output of the lth convolutional layer; is the feature output of the l-1th convolutional layer of channel j; is the bias of the l-1th layer of channel j; is the weight of the l-1th layer of channel j; f l-1 (x) is the activation function of the l-1th convolutional layer; m is the number of channels.

7. The method for designing the layout position of a lightning receptor according to any one of claims 1 to 6, characterized in that: It also includes a reference data testing platform for establishing a reference database, transmitting the established reference database to the position layout prediction module to perform multiple training on the multi-channel weighted fusion convolutional network and save the optimal training model parameters.

8. A wind turbine blade lightning protection system manufactured using the design method described in any one of claims 1 to 7.

Images