A method and system for managing lightning protection detection points

Abstract

This invention provides a method and system for managing lightning protection detection points. The method divides the area to be detected into grids and generates a regional lightning protection risk index for each grid using a spatiotemporal graph convolutional network. It acquires its own state data and external environment data, inputs them into a pre-trained attention-based MLP neural network model, and outputs a health score for each lightning protection detection point. The external environment data includes the regional lightning protection risk index. The method then determines whether the health score is greater than a threshold. If so, it uses an LSTM model supplemented by an XGBoost model to determine a continuous predicted health score based on the health score data, the number of lightning strikes, and average temperature and humidity. Finally, it determines whether the predicted health score is greater than a threshold. If so, the corresponding lightning protection detection point is removed. If the health score of the lightning protection detection point is not greater than the threshold, the corresponding lightning protection detection point is retained, effectively achieving dynamic management based on real-time risk.

Description

Technical Field

[0001] This invention belongs to the field of lightning protection testing point management technology, and specifically relates to a lightning protection testing point management method and system. Background Technology

[0002] Lightning protection testing is the inspection and evaluation of the lightning protection capabilities of buildings, facilities, or equipment. Its main purpose is to ensure that these objects can operate safely and stably during thunderstorms through a series of testing methods, thereby avoiding property damage and personal injury caused by lightning strikes.

[0003] Currently, lightning protection testing points are set up on buildings, and all lightning protection testing points are tested by manual inspections on a regular basis. However, since each lightning protection testing point faces different risks, not all lightning protection testing points need to be tested every time, which undoubtedly consumes a lot of manpower and resources. Summary of the Invention

[0004] Based on this, the present invention provides a method and system for managing lightning protection detection points, which aims to change the rigid mode of traditional periodic inspections and realize dynamic management based on real-time risks.

[0005] A first aspect of this invention provides a method for managing lightning protection detection points, the method comprising:

[0006] The area to be detected is divided into grids, and a dynamic regional lightning protection risk index is generated for each grid based on a spatiotemporal graph convolutional network.

[0007] Acquire the self-state data and external environment data of the corresponding lightning protection detection points within the grid, input the self-state data and the external environment data into a pre-trained MLP neural network model based on the attention mechanism, and output the health score of the lightning protection detection points. The external environment data includes the regional lightning protection risk index.

[0008] Determine whether the health score of the lightning protection testing point is greater than the threshold.

[0009] If the health score of the lightning protection testing point is determined to be greater than the threshold, then based on the health score data of the corresponding lightning protection testing point, the number of lightning strikes, and the average temperature and humidity, the LSTM model is used, supplemented by the XGBoost model, to determine the continuous predicted health score.

[0010] Determine whether the predicted health score is greater than a threshold;

[0011] If the predicted health score is greater than the threshold, the corresponding lightning protection detection point will be removed.

[0012] If the health score of a lightning protection testing point is determined to be no greater than the threshold, the corresponding lightning protection testing point will be retained.

[0013] Furthermore, the step of dividing the area to be detected into a grid and generating a dynamic regional lightning protection risk index for each grid based on a spatiotemporal graph convolutional network includes:

[0014] Acquire lightning activity data, geological and soil data, surrounding building distribution data, and meteorological data; perform gridding processing on the lightning activity data, geological and soil data, surrounding building distribution data, and meteorological data; and extract feature vectors for each grid.

[0015] After preprocessing the extracted feature vectors, the target feature vectors are determined by filtering using Pearson correlation coefficient and random forest feature importance.

[0016] Based on the target feature vector and time, a two-dimensional graph structure is constructed, and a neural network model is trained based on the two-dimensional graph structure.

[0017] The feature vectors and time dimension features of the nodes in the two-dimensional graph structure are input into the trained neural network model, and the output is the regional lightning protection risk index, where each grid is a node.

[0018] Furthermore, the neural network model consists of an input layer, a spatiotemporal convolutional layer, a pooling layer, and a fully connected layer arranged sequentially. The spatiotemporal convolutional layer includes two ST-GCN modules, each containing spatial convolution and temporal convolution, with ReLU as the activation function. The spatial convolution uses Chebyshev multinomial filtering. The temporal convolution has a 1D convolution kernel size of 3. The pooling layer uses pooling with a time step of 2.

[0019] Furthermore, the self-status data includes grounding performance data, SPD status data, physical integrity data, historical performance data, and environmental data. The grounding performance data includes the current grounding resistance value, grounding resistance change rate, and duration of grounding resistance exceeding the standard. The SPD status data includes leakage current value and the rate of increase in the number of actions. The physical integrity data includes corrosion level and structural defect score. The historical performance data includes the number of faults within a preset time and the time since the last maintenance. The environmental data includes the comprehensive temperature and humidity index of the location, the predicted corrosion rate, and the frequency of lightning strikes. The external environmental data also includes regional risk trends, the proportion of surrounding high-risk grids, microenvironmental risk coefficient, lightning warning correlation, and historical lightning strike density.

[0020] Furthermore, the attention-based MLP neural network model includes an input layer, a feature encoding layer, an attention layer, and an output layer. The feature encoding layer includes a three-layer fully connected network with BatchNorm and LeakyReLU activation functions. The attention layer calculates attention weights for features using a single-layer MLP.

[0021] Furthermore, in the step of determining continuous predicted health scores based on the health score data, number of lightning strikes, and average temperature and humidity of the corresponding lightning protection detection points, the mean absolute error of the LSTM model and the XGBoost model on the validation set is calculated respectively, and the corresponding weights are assigned using the inverse error method. Finally, the results output by the LSTM model and the XGBoost model are weighted and summed to obtain the predicted health score.

[0022] Furthermore, the step of retaining the corresponding lightning protection detection point if the health score of the detection point is not greater than the threshold includes:

[0023] The continuous predicted health score of a single lightning protection detection point is correlated with the scores of other points within a preset range to identify risk transmission paths, and regional collaborative maintenance is carried out based on the risk transmission paths.

[0024] A second aspect of this invention provides a lightning protection detection point management system for implementing the lightning protection detection point management method described in the first aspect, the lightning protection detection point management system comprising:

[0025] The segmentation module is used to divide the area to be detected into grids and generate a dynamic regional lightning protection risk index for each grid based on the spatiotemporal graph convolutional network.

[0026] The input module is used to acquire the self-state data and external environment data of the corresponding lightning protection detection points within the grid, input the self-state data and the external environment data into a pre-trained MLP neural network model based on the attention mechanism, and output the health score of the lightning protection detection points. The external environment data includes the regional lightning protection risk index.

[0027] The first judgment module is used to determine whether the health score of the lightning protection detection point is greater than the threshold.

[0028] The determination module is used to determine the continuous predicted health score if the health score of the lightning protection detection point is greater than the threshold. This is done by using an LSTM model supplemented by an XGBoost model, based on the health score data of the corresponding lightning protection detection point, the number of lightning strikes, and the average temperature and humidity.

[0029] The second judgment module is used to determine whether the predicted health score is greater than a threshold.

[0030] The elimination module is used to eliminate the corresponding lightning protection detection point if the predicted health score is greater than the threshold.

[0031] The retention module is used to retain the corresponding lightning protection detection point if the health score of the lightning protection detection point is not greater than the threshold.

[0032] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the lightning protection detection point management method provided in the first aspect.

[0033] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the lightning protection detection point management method provided in the first aspect.

[0034] This invention provides a method and system for managing lightning protection detection points. The method divides the area to be detected into grids and generates a dynamic regional lightning protection risk index for each grid using a spatiotemporal graph convolutional network. It acquires the self-state data and external environment data of the corresponding lightning protection detection points within the grid, inputs these data into a pre-trained attention-based MLP neural network model, and outputs a health score for each lightning protection detection point. The external environment data includes the regional lightning protection risk index. The method then determines whether the health score of the lightning protection detection point is greater than a threshold. If the health score is greater than a threshold, the method proceeds accordingly. If the health score of a lightning protection detection point is greater than the threshold, then based on the health score data, number of lightning strikes, and average temperature and humidity of the corresponding lightning protection detection point, an LSTM model supplemented by an XGBoost model is used to determine a continuous predicted health score; it is then determined whether the predicted health score is greater than the threshold; if the predicted health score is greater than the threshold, the corresponding lightning protection detection point is removed; if the health score of the lightning protection detection point is not greater than the threshold, the corresponding lightning protection detection point is retained. Specifically, the above method changes the rigid mode of traditional periodic inspections and realizes dynamic management based on real-time risks. Attached Figure Description

[0035] Figure 1 The flowchart illustrates the implementation of a lightning protection detection point management method according to Embodiment 1 of the present invention.

[0036] Figure 2 This is a structural block diagram of a lightning protection detection point management system provided in Embodiment 2 of the present invention;

[0037] Figure 3This is a structural block diagram of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0038] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0039] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0041] Example 1

[0042] According to an embodiment of the present invention, a method for managing lightning protection detection points is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0043] This first embodiment provides a method for managing lightning protection testing points, which can be used in electronic devices, such as computers. It should be noted that the application scenario involves first generating testing points based on the building's lightning protection testing targets, assigning each testing point a unique identification code NFC chip, and then affixing the NFC chip to the lightning protection testing target. During the initial testing, the testing personnel scan the NFC QR code to activate the NFC chip and simultaneously add the corresponding testing point. After scanning the NFC QR code, installing and activating the NFC chip, the building testing points are finally uploaded in the form of a testing report.

[0044] Please see Figure 1 , Figure 1The flowchart of a lightning protection detection point management method provided in Embodiment 1 of the present invention is shown, specifically including steps S01 to S07.

[0045] Step S01: Divide the area to be detected into a grid and generate a dynamic regional lightning protection risk index for each grid based on the spatiotemporal graph convolutional network.

[0046] Specifically, lightning activity data, geological and soil data, surrounding building distribution data, and meteorological data are acquired. The lightning activity data, geological and soil data, surrounding building distribution data, and meteorological data are then processed into a grid. Feature vectors are extracted for each grid. In this embodiment of the invention, the feature vector extracted for each grid includes the annual average number of thunderstorm days, the number of lightning strikes in the past 30 days, and the average peak value of lightning current from the lightning activity data; the weighted soil resistivity and soil type risk coefficient from the geological and soil data; the average building height, building coverage, and terrain undulation from the surrounding building distribution data; and the average humidity in the past 7 days, the average atmospheric electric field intensity, and the lightning warning level score from the meteorological data.

[0047] After preprocessing the extracted feature vectors, the target feature vectors are determined by filtering using Pearson correlation coefficient and random forest feature importance. In other words, after the extracted feature vectors are processed by missing value imputation and normalization, the target feature vectors are determined by filtering using Pearson correlation coefficient and random forest feature importance based on Python.

[0048] Based on the target feature vector and time, a two-dimensional graph structure is constructed, and a neural network model is trained based on the two-dimensional graph structure. In this embodiment of the invention, the nodes of the two-dimensional graph structure are defined as follows: each 100m×100m grid is a node, and the node feature is the target feature vector; the edges of the two-dimensional graph structure are defined as follows: the "k-nearest neighbor" rule (k=8) is adopted, that is, each node is connected to the 8 nearest grids, and the edge weight = 1 / distance (distance unit: km), the closer the distance, the greater the weight; in order to divide the data into time slices by hours (24 time slices per day), a "space-time" two-dimensional graph structure is constructed, and the time series length of the input model is set to 12 hours.

[0049] The feature vectors and temporal features of nodes in a two-dimensional graph structure are input into a trained neural network model, which outputs a regional lightning protection risk index. The neural network model consists of an input layer, a spatiotemporal convolutional layer, a pooling layer, and a fully connected layer arranged sequentially. The spatiotemporal convolutional layer includes two ST-GCN modules, each containing spatial convolution and temporal convolution, with ReLU activation function. The spatial convolution uses Chebyshev multinomial filtering. The temporal convolution has a 1D convolution kernel size of 3. The pooling layer uses pooling with a time step of 2. The fully connected layer consists of two layers (128→64→1), and the output is a risk index (0-100) for each grid.

[0050] Step S02: Obtain the self-state data and external environment data of the corresponding lightning protection detection point within the grid, input the self-state data and the external environment data into a pre-trained MLP neural network model based on the attention mechanism, and output the health score of the lightning protection detection point. The external environment data includes the regional lightning protection risk index.

[0051] It should be noted that the self-status data includes grounding performance data, SPD status data, physical integrity data, historical performance data, and environmental data. The grounding performance data includes the current grounding resistance value, grounding resistance change rate, and duration of grounding resistance exceeding the standard. The SPD status data includes leakage current value and the rate of increase in the number of actions. The physical integrity data includes corrosion level and structural defect score. The historical performance data includes the number of faults within a preset time and the time since the last maintenance. The environmental data includes the comprehensive temperature and humidity index of the location, the predicted corrosion rate, and the frequency of lightning strikes. The external environmental data also includes regional risk trends, the proportion of high-risk grids in the surrounding area, the micro-environment risk coefficient, the correlation of lightning warnings, and the historical lightning strike density.

[0052] In this embodiment of the invention, the current grounding resistance value is the most recent measurement value (Ω), the grounding resistance change rate is (current value - value 3 months ago) / value 3 months ago × 100%, the grounding resistance exceeding the standard duration is the number of hours continuously exceeding the threshold (e.g., 10Ω), the leakage current value is the most recent monitoring value (μA), the action frequency growth rate is (number of actions in the last 3 months - the previous 3 months) / the previous 3 months × 100%, the corrosion level is the image recognition result (level 1-5, level 5 is the most severe), the structural defect score is the weighted sum of various defects (fracture × 0.4 + deformation × 0.3 + loosening × 0.3), the number of faults within the preset time is the cumulative value of the past year (0 times / 1-2 times / ≥3 times), the time since the last maintenance is the number of days (taking the natural logarithm to reduce the impact of extreme values), and the point temperature... The comprehensive humidity index is (temperature / 35 + humidity / 100) / 2 (normalized to 0-1), the corrosion rate prediction value is the LSTM prediction result based on temperature, humidity and historical data (mm / year), the lightning strike frequency is the number of times the lightning current counter has been recorded in the past 3 months, the regional risk trend is (current grid lightning protection risk index - lightning protection risk index 7 days ago) / lightning protection risk index 7 days ago × 100%, the proportion of high-risk grids in the surrounding area is the number of grids with a score ≥80 in the surrounding 9 grids / 9, the microenvironment risk coefficient is humid area (1.2) / waterlogged area (1.5) / dry area (0.8), the lightning warning correlation is the warning level of the area × 0.2 + atmospheric electric field strength × 0.8, and the historical lightning strike density is the average annual number of lightning strikes within 100m of this point in the past 5 years.

[0053] Specifically, the attention-based MLP neural network model includes an input layer, a feature encoding layer, an attention layer, and an output layer. The input layer is the aforementioned 18-dimensional feature vector. The feature encoding layer includes a 3-layer fully connected network (128→64→32), with BatchNorm and LeakyReLU activation functions added. The attention layer calculates attention weights for the 32-dimensional features through a single-layer MLP, enabling the model to automatically focus on key features (such as grounding resistance). The output layer includes one neuron (sigmoid activation), outputting a health score of 0-100.

[0054] Step S03: Determine whether the health score of the lightning protection detection point is greater than the threshold. If yes, proceed to step S04; otherwise, proceed to step S07.

[0055] In step S04, based on the health score data, number of lightning strikes, and average temperature and humidity of the corresponding lightning protection detection points, an LSTM model is used, supplemented by an XGBoost model, to determine the continuous predicted health score.

[0056] Specifically, the LSTM model includes an input layer, hidden layers, fully connected layers, and an output layer. The hidden layer uses two LSTM units: the first layer has 64 hidden nodes (capturing low-level spatiotemporal features, such as the short-term correlation between monthly lightning strikes and health), and the second layer has 32 hidden nodes (further abstracting the features of the first layer to capture long-term dependencies, such as the cumulative impact of annual environmental changes on health). Each layer adds a Dropout layer (dropout rate = 0.2) to prevent overfitting. The fully connected layer maps the 32-dimensional features output from the second LSTM layer to 16 dimensions (introducing non-linearity through the ReLU activation function), and then maps them to a 1-dimensional output (corresponding to the predicted health score for one month). The output layer uses a linear activation function (since the health score is a continuous value, probability normalization is not required), and the output range is constrained to between 0 and 100.

[0057] The XGBoost model chosen is the regression-based XGBoost (because the output is a continuous health score), employing a tree ensemble strategy (multiple decision trees collaboratively predicting). In this embodiment of the invention, the number of trees (n_estimators) is set to 200 to ensure sufficient model fitting ability while avoiding overfitting due to too many trees; the maximum tree depth (max_depth) is set to 6 to control the complexity of a single tree (excessive depth can easily lead to overfitting to training set noise); the learning rate (learning_rate) is set to 0.03 to improve prediction accuracy by iteratively optimizing model parameters; the subsample ratio (subsample) is set to 0.85, randomly selecting 85% of the samples during training for each tree to enhance the model's generalization ability; and the column sample ratio (colsample_bytree) is set to 0.8, randomly selecting 80% of the features during training for each tree to reduce the redundant influence between features. In addition, the XGBoost model does not directly support the time step input of time series data. It needs to be expanded into static features, and each feature is labeled with a time dimension (such as "number of lightning strikes in the first month" and "comprehensive temperature and humidity index in the 12th month") to ensure that the model can identify the differences in the impact of features on health at different time points.

[0058] It should be noted that the mean absolute error (MAE) of the LSTM model and the XGBoost model on the validation set is calculated separately, and the corresponding weights are assigned using the inverse error ratio method (weight = 1 / MAE / (1 / MAE_LSTM+1 / MAE_XGBoost). Finally, the outputs of the LSTM model and the XGBoost model are weighted and summed to obtain the predicted health score. For example, the predicted health score = LSTM predicted value × 0.7 + XGBoost predicted value × 0.3.

[0059] Step S05: Determine whether the predicted health score is greater than the threshold. If so, proceed to step S06.

[0060] In step S06, the corresponding lightning protection detection points are removed.

[0061] This is understandable, indicating that the lightning protection testing point has been in a stable and healthy state for a long time. Therefore, the lightning protection testing point can be removed to reduce the workload.

[0062] In step S07, the corresponding lightning protection detection points are retained.

[0063] Specifically, the retained lightning protection testing points need to be manually tested. Furthermore, the continuous predicted health scores of a single lightning protection testing point are correlated with the scores of other points within a preset range to identify risk transmission paths. Based on these risk transmission paths, regional collaborative maintenance is carried out. It should be noted that first, the correlation coefficient (Pearson correlation coefficient) between the predicted health scores of the target point and surrounding points (within a 300m range) is calculated. An absolute value of the correlation coefficient ≥ 0.7 is considered "strong correlation," 0.5-0.7 is "medium correlation," and < 0.5 is "weak correlation." For example, the correlation coefficient between lightning rod A (score 72→68→62) and surrounding grounding device B (score 70→66→60) in a certain community is 0.82 (strong correlation), and the correlation coefficient between grounding device B and SPD (Surge Protective Device) C (score 75→74→73) is 0.35 (weak correlation), indicating that risk transmission exists between A and B.

[0064] Furthermore, through cross-correlation analysis, the direction of risk transmission can be determined (e.g., if point A decays for one month and point B begins to decay, it indicates that A is the source of transmission and B is the affected point), and the lag time can be calculated (e.g., if the health of B begins to decline 28 days after A decays), providing a time window for "source intervention" (e.g., after maintaining point A, the decay of point B can be prevented within 28 days).

[0065] It should be noted that by unifying the time granularity of health data between the target location and related locations within 300m (preferably monthly, and weekly when data is sufficient), and after preprocessing including linear interpolation to fill missing values, first-order difference detrending, and Z-score standardization, the correlation coefficients of the two locations are calculated under different lags by sliding the time series data. The lag time with the strongest correlation is found. If location B shows the strongest response after a change in location A within a specific lag step (e.g., 1 month or 4 weeks), and the absolute value of the correlation coefficient is ≥0.7 and the lag step is within 1-3 time units, then A can be identified as the transmission source and B as the affected point. The lag step can then be converted into a specific time (e.g., 4 weeks = 28 days), providing a time window for source intervention (e.g., checking B within 21 days after maintaining A). It can also identify the "transmission chain" (e.g., A→B→C) within the region and verify the intervention effect.

[0066] In summary, the lightning protection detection point management method in the above embodiments of the present invention divides the area to be detected into a grid and generates a dynamic regional lightning protection risk index for each grid based on a spatiotemporal graph convolutional network; acquires the self-state data and external environment data of the corresponding lightning protection detection points within the grid, inputs the self-state data and external environment data into a pre-trained attention-based MLP neural network model, and outputs a health score for the lightning protection detection points, wherein the external environment data includes the regional lightning protection risk index; determines whether the health score of the lightning protection detection points is greater than a threshold; if so, If the health score of a lightning protection testing point is greater than the threshold, then based on the health score data, number of lightning strikes, and average temperature and humidity of the corresponding lightning protection testing point, an LSTM model supplemented by an XGBoost model is used to determine a continuous predicted health score; it is then determined whether the predicted health score is greater than the threshold; if the predicted health score is greater than the threshold, the corresponding lightning protection testing point is removed; if the health score of the lightning protection testing point is not greater than the threshold, the corresponding lightning protection testing point is retained. Specifically, the above method changes the rigid mode of traditional periodic inspections and realizes dynamic management based on real-time risks.

[0067] Example 2

[0068] Please see Figure 2 , Figure 2 This is a structural block diagram of a lightning protection detection point management system provided in Embodiment 2 of the present invention. This lightning protection detection point management system 200 is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0069] Specifically, the lightning protection detection point management system 200 includes: a division module 21, an input module 22, a first judgment module 23, a determination module 24, a second judgment module 25, a rejection module 26, and a retention module 27, wherein:

[0070] The segmentation module 21 is used to divide the area to be detected into grids and generate a dynamic regional lightning protection risk index for each grid based on the spatiotemporal graph convolutional network.

[0071] Input module 22 is used to acquire the self-state data and external environment data of the corresponding lightning protection detection points within the grid, input the self-state data and the external environment data into a pre-trained MLP neural network model based on an attention mechanism, and output the health score of the lightning protection detection points. The external environment data includes the regional lightning protection risk index, and the self-state data includes grounding performance data, SPD status data, physical integrity data, historical performance data, and environmental data. The grounding performance data includes the current grounding resistance value, grounding resistance change rate, and duration of grounding resistance exceeding the standard. The SPD status data includes leakage current value and the rate of increase in the number of actions. The physical integrity data... The historical performance data includes the number of failures within a preset time period and the time since the last maintenance, based on corrosion level and structural defect score. The environmental data includes the comprehensive index of temperature and humidity at the location, the predicted corrosion rate, and the frequency of lightning strikes. The external environmental data also includes regional risk trends, the proportion of high-risk grids in the surrounding area, the micro-environment risk coefficient, the correlation of lightning warnings, and the historical lightning strike density. The MLP neural network model based on the attention mechanism includes an input layer, a feature encoding layer, an attention layer, and an output layer. The feature encoding layer includes a 3-layer fully connected network with BatchNorm and LeakyReLU activation functions. The attention layer calculates attention weights for features through a single-layer MLP.

[0072] The first judgment module 23 is used to determine whether the health score of the lightning protection detection point is greater than the threshold.

[0073] The determination module 24 is used to determine the continuous predicted health score if the health score of the lightning protection detection point is greater than the threshold. Based on the health score data, number of lightning strikes and average temperature and humidity of the corresponding lightning protection detection point, the LSTM model is used and supplemented by the XGBoost model to determine the health score. The mean absolute error of the LSTM model and the XGBoost model on the validation set is calculated respectively, and the corresponding weights are assigned using the inverse error method. Finally, the outputs of the LSTM model and the XGBoost model are weighted and summed to obtain the predicted health score.

[0074] The second judgment module 25 is used to determine whether the predicted health score is greater than a threshold.

[0075] The elimination module 26 is used to eliminate the corresponding lightning protection detection point if the predicted health score is greater than the threshold.

[0076] The retention module 27 is used to retain the corresponding lightning protection detection point if the health score of the lightning protection detection point is not greater than the threshold.

[0077] Furthermore, in some optional embodiments of the present invention, the dividing module 21 includes:

[0078] The acquisition unit is used to acquire lightning activity data, geological and soil data, surrounding building distribution data, and meteorological data, and to perform gridding processing on the lightning activity data, geological and soil data, surrounding building distribution data, and meteorological data, and to extract feature vectors for each grid.

[0079] The preprocessing unit is used to preprocess the extracted feature vectors and then determine the target feature vectors by filtering them using Pearson correlation coefficient and random forest feature importance.

[0080] The construction unit is used to construct a two-dimensional graph structure based on the target feature vector and time, and to train a neural network model based on the two-dimensional graph structure. The neural network model consists of an input layer, a spatiotemporal convolutional layer, a pooling layer, and a fully connected layer arranged sequentially. The spatiotemporal convolutional layer includes two ST-GCN modules, each containing spatial convolution and temporal convolution, with ReLU as the activation function. The spatial convolution uses Chebyshev multinomial filtering. The temporal convolution has a 1D convolution kernel size of 3. The pooling layer uses pooling with a time step of 2.

[0081] The input unit is used to input the feature vectors and time dimension features of the nodes in the two-dimensional graph structure into the trained neural network model, and outputs the regional lightning protection risk index, where each grid is a node.

[0082] Furthermore, in some optional embodiments of the present invention, the lightning protection detection point management system 200 further includes:

[0083] The correlation analysis module is used to perform correlation analysis between the continuous predicted health scores of a single lightning protection detection point and the scores of other points within a preset range, identify the risk transmission path, and perform regional collaborative maintenance based on the risk transmission path.

[0084] Example 3

[0085] In another aspect, the present invention also proposes an electronic device, please refer to [link to relevant documentation]. Figure 3The electronic device shown is an embodiment of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, it implements the lightning protection detection point management method described above.

[0086] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0087] The memory 20 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a plug-in hard disk, SmartMediaCard (SMC), SecureDigital (SD) card, FlashCard, etc., equipped on the electronic device. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or will be output.

[0088] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0089] This invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the lightning protection detection point management method described above.

[0090] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0091] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0092] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0093] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0094] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for managing lightning protection testing points, characterized in that, The method includes: The area to be detected is divided into grids, and a dynamic regional lightning protection risk index is generated for each grid based on a spatiotemporal graph convolutional network. Acquire the self-state data and external environment data of the corresponding lightning protection detection points within the grid, input the self-state data and the external environment data into a pre-trained MLP neural network model based on the attention mechanism, and output the health score of the lightning protection detection points. The external environment data includes the regional lightning protection risk index. Determine whether the health score of the lightning protection testing point is greater than the threshold. If the health score of the lightning protection testing point is determined to be greater than the threshold, then based on the health score data of the corresponding lightning protection testing point, the number of lightning strikes, and the average temperature and humidity, the LSTM model is used, supplemented by the XGBoost model, to determine the continuous predicted health score. Determine whether the predicted health score is greater than a threshold; If the predicted health score is greater than the threshold, the corresponding lightning protection detection point will be removed. If the health score of a lightning protection testing point is determined to be no greater than the threshold, the corresponding lightning protection testing point is retained. The neural network model consists of an input layer, a spatiotemporal convolutional layer, a pooling layer, and a fully connected layer arranged sequentially. The spatiotemporal convolutional layer includes two ST-GCN modules, each containing spatial convolution and temporal convolution, with ReLU as the activation function. The spatial convolution uses Chebyshev multinomial filtering. The temporal convolution has a 1D convolution kernel size of 3. The pooling layer uses pooling with a time step of 2. In the step of determining the continuous predicted health score based on the health score data, number of lightning strikes and average temperature and humidity of the corresponding lightning protection detection point, the mean absolute error of the LSTM model and the XGBoost model on the validation set are calculated respectively, and the corresponding weights are assigned by the inverse error method. Finally, the outputs of the LSTM model and the XGBoost model are weighted and summed to obtain the predicted health score. The step of retaining the corresponding lightning protection detection point if the health score of the detection point is not greater than the threshold includes: The continuous predicted health score of a single lightning protection detection point is correlated with the scores of other points within a preset range to identify risk transmission paths, and regional collaborative maintenance is carried out based on the risk transmission paths.

2. The lightning protection detection point management method according to claim 1, characterized in that, The steps of dividing the area to be detected into a grid and generating a dynamic regional lightning protection risk index for each grid based on a spatiotemporal graph convolutional network include: Acquire lightning activity data, geological and soil data, surrounding building distribution data, and meteorological data; perform gridding processing on the lightning activity data, geological and soil data, surrounding building distribution data, and meteorological data; and extract feature vectors for each grid. After preprocessing the extracted feature vectors, the target feature vectors are determined by filtering using Pearson correlation coefficient and random forest feature importance. Based on the target feature vector and time, a two-dimensional graph structure is constructed, and a neural network model is trained based on the two-dimensional graph structure. The feature vectors and time dimension features of the nodes in the two-dimensional graph structure are input into the trained neural network model, and the output is the regional lightning protection risk index, where each grid is a node.

3. The lightning protection detection point management method according to claim 2, characterized in that, The self-status data includes grounding performance data, SPD status data, physical integrity data, historical performance data, and environmental data. The grounding performance data includes the current grounding resistance value, grounding resistance change rate, and duration of grounding resistance exceeding the standard. The SPD status data includes leakage current value and the rate of increase in the number of actions. The physical integrity data includes corrosion level and structural defect score. The historical performance data includes the number of faults within a preset time and the time since the last maintenance. The environmental data includes the comprehensive temperature and humidity index of the location, the predicted corrosion rate, and the frequency of lightning strikes. The external environmental data also includes regional risk trends, the proportion of high-risk grids in the surrounding area, micro-environmental risk coefficient, lightning warning correlation, and historical lightning strike density.

4. The lightning protection detection point management method according to claim 3, characterized in that, The attention-based MLP neural network model includes an input layer, a feature encoding layer, an attention layer, and an output layer. The feature encoding layer consists of three fully connected layers with BatchNorm and LeakyReLU activation functions. The attention layer calculates attention weights for features using a single-layer MLP.

5. A lightning protection detection point management system, characterized in that, For implementing the lightning protection detection point management method as described in any one of claims 1-4, the system comprises: The segmentation module is used to divide the area to be detected into grids and generate a dynamic regional lightning protection risk index for each grid based on the spatiotemporal graph convolutional network. The input module is used to acquire the self-state data and external environment data of the corresponding lightning protection detection points within the grid, input the self-state data and the external environment data into a pre-trained MLP neural network model based on the attention mechanism, and output the health score of the lightning protection detection points. The external environment data includes the regional lightning protection risk index. The first judgment module is used to determine whether the health score of the lightning protection detection point is greater than the threshold. The determination module is used to determine the continuous predicted health score if the health score of the lightning protection detection point is greater than the threshold. This is done by using an LSTM model supplemented by an XGBoost model, based on the health score data of the corresponding lightning protection detection point, the number of lightning strikes, and the average temperature and humidity. The second judgment module is used to determine whether the predicted health score is greater than a threshold. The elimination module is used to eliminate the corresponding lightning protection detection point if the predicted health score is greater than the threshold. The retention module is used to retain the corresponding lightning protection detection point if the health score of the lightning protection detection point is not greater than the threshold.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the lightning protection detection point management method as described in any one of claims 1-4.

7. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the lightning protection detection point management method as described in any one of claims 1-4.

Images