Three-dimensional lightning proximity prediction method and system

Abstract

The invention discloses a three-dimensional lightning approach prediction method and system, and the method comprises the steps: constructing a three-dimensional grid covering a lightning activity space range of a target region, carrying out the data enhancement of historical three-dimensional lightning data for representing the impact of lightning on a peripheral region, and carrying out the three-dimensional lightning approach prediction through combining with an innovative deep learning model. Compared with a lightning prediction model which only considers a two-dimensional plane or a single height in the prior art, the method can capture the dynamic change of the lightning in a three-dimensional space, and significantly improves the prediction precision and the spatial analysis capability.

Description

Technical Field

[0001] The invention belongs to the field of atmospheric science technology, and in particular relates to a three-dimensional lightning imminent prediction method and system. Background Art

[0002] Lightning is a long-distance instantaneous discharge phenomenon with huge energy generated by the collision of a large number of positive and negative electrons. It usually occurs inside clouds, between clouds, and between clouds and the ground. It has the characteristics of high frequency, strong current and high energy. Among them, cloud-to-ground lightning poses the most serious threat to human socio-economic and personal safety, often causing damage to ground buildings, electronic equipment failures, and thousands of casualties every year. Since lightning often occurs frequently in a certain area, the regularity of its occurrence in close proximity makes the prediction of lightning strike areas of great significance. At present, traditional methods for lightning prediction mainly include numerical analysis methods and machine learning methods, but these two methods have certain limitations in complex natural environments. In recent years, deep learning has made significant progress in time series prediction tasks. Converting the lightning strike area prediction problem into a time series prediction problem and using deep learning technology for prediction has become a new research direction. However, most of the existing lightning prediction research based on deep learning remains on a two-dimensional plane. Some technicians have conducted research on three-dimensional lightning, such as the prior art with publication number CN115453479A. This prior art identifies the lightning strike area, but it still stacks the two-dimensional radar product data to form three-dimensional data. The three-dimensional data formed by data stacking is still different from the actual three-dimensional data. Compared with traditional two-dimensional plane research, the three-dimensional lightning prediction method can better utilize the spatial distribution information of lightning activity, especially in complex environments such as cloud height changes and tortuous lightning paths, which helps to achieve more accurate analysis of lightning locations. However, the current research on three-dimensional lightning prediction based on three-dimensional lightning location data using deep learning models is still blank. Summary of the invention

[0003] In order to solve the problems existing in the prior art, the present invention proposes a three-dimensional lightning approach prediction method and system.

[0004] The technical solution of the present invention is as follows:

[0005] A three-dimensional lightning nowcasting method, comprising:

[0006] Construct a three-dimensional grid covering the spatial extent of lightning activity in the target area;

[0007] Mapping the historical three-dimensional lightning location data to the three-dimensional grid to obtain the historical three-dimensional lightning grid data;

[0008] Based on the historical three-dimensional lightning grid data, a three-dimensional lightning near-forecast model taking into account changes in spatial characteristics is trained;

[0009] The trained three-dimensional lightning nowcasting model is used to perform three-dimensional lightning nowcasting.

[0010] Furthermore, the specific steps of constructing a three-dimensional grid covering the spatial range of lightning activity in the target area include:

[0011] The grid range of the three-dimensional grid in the plane where the longitude and longitude are located is determined according to the longitude and latitude range of the target area, and the distribution of the three-dimensional grid in the plane where the longitude and longitude are located is delineated based on the length and width of the grid unit. The total number of grids in the longitude is recorded as N lon , the row index in longitude is recorded as col_index, and the total number of grids in latitude is recorded as N lat , the column index on the latitude is recorded as row_index;

[0012] According to the height range of the target area and the height of the grid unit, the distribution of the three-dimensional grid in the height direction is determined. The total number of height layers of the grid is recorded as N lay , the height layer index is recorded as height_index;

[0013] This forms a three-dimensional grid N lon ×N lat ×N lay .

[0014] Furthermore, the specific steps of mapping the historical three-dimensional lightning location data to the three-dimensional grid to obtain the historical three-dimensional lightning grid data include:

[0015] The historical three-dimensional lightning location data is divided into a number of temporally continuous historical three-dimensional lightning location data sequences according to a certain time step;

[0016] The height data, longitude data and latitude data in each historical three-dimensional lightning location data sequence are used to map the height layer index of the three-dimensional grid, the row index on the plane where the longitude and latitude are located, and the column index. The grid cells where lightning occurs in the three-dimensional grid are subjected to data mapping operations for indicating the occurrence of lightning and data enhancement operations for indicating the impact of lightning on the surrounding areas, thereby obtaining historical three-dimensional lightning grid data that records the height layer index, row index, column index and grid cell value of lightning occurrence.

[0017] Furthermore, the specific steps of the data mapping operation for indicating the occurrence of lightning include:

[0018] For each of the historical three-dimensional lightning location data sequences, the grid cell value of the mapping position (height_index, row_index, col_index) in the three-dimensional grid corresponding to each lightning occurrence position is increased by 1.

[0019] Furthermore, the specific steps of the data enhancement operation for representing the impact of lightning on the surrounding area include:

[0020] For each of the historical three-dimensional lightning location data series, center point processing and neighboring point processing are performed on the grid cells corresponding to the mapping position in the three-dimensional grid and the grid cells around the grid cells;

[0021] The specific method of the center point processing is: taking the grid cell where the lightning occurs as the center grid cell and increasing its value by 1 on the original basis to mark the occurrence of lightning at the location;

[0022] The specific method of the neighboring point processing is: the grid cells around the central grid cell are regarded as neighboring grids and their values ​​are increased by 1 on the original basis.

[0023] Furthermore, the three-dimensional lightning proximity prediction model taking into account the changes in spatial features adopts a three-dimensional convolutional neural network 3DAttUnet-GAN. The three-dimensional convolutional neural network 3DAttUnet-GAN is based on the Unet neural network model, expands the two-dimensional convolution operation therein into a three-dimensional convolution operation, introduces an attention mechanism, and is formed in combination with a generative adversarial network; the three-dimensional lightning proximity prediction model includes a 3DAttUnet encoding-decoding structure and a generative adversarial network, the generator of the generative adversarial network is a 3DAttUnet encoding-decoding structure, and the discriminator is a binary classification network; the generator performs feature encoding and decoding on the input data and generates a prediction result; the generative adversarial network inputs the prediction result generated by the generator into the discriminator in combination with real data, and the generator is trained by the authenticity evaluation result output by the discriminator.

[0024] Furthermore, the 3DAttUnet encoding-decoding structure includes encoder, decoder, and attention mechanism modules;

[0025] The encoder part consists of 5 layers of 3D convolution modules. Each layer uses depth-separable 3D convolution, ReLU activation function, and integrates attention mechanism module and maximum pooling layer.

[0026] The decoder uses trilinear interpolation or deconvolution for upsampling, restores the spatial and temporal resolution of the input features layer by layer, and uses depthwise separable 3D convolution combined with an attention module to further enhance the features.

[0027] The attention mechanism modules include temporal attention, channel attention and spatial attention. Temporal attention is applied to time series to capture dynamic changes in time series; channel attention is applied to altitude layers to dynamically adjust the importance of altitude layers; spatial attention is applied to the plane space of each altitude layer to adaptively adjust the spatial position weight of each altitude layer.

[0028] Furthermore, based on the historical three-dimensional lightning grid data, the specific steps of training the three-dimensional lightning near-fall prediction model taking into account the changes in spatial characteristics include:

[0029] The historical three-dimensional lightning grid data are divided into a training set and a test set at a ratio of 8:2, and the training set is further divided into a training subset and a validation subset at a ratio of 8:2;

[0030] During the model training process, a timestamp file is set to form a continuous sliding time array with a length of 20 time steps for the three-dimensional lightning grid data, that is, the data of the first 10 time steps are used to predict the data of the next 10 time steps. The learning rate, data batch processing scale and loss function are optimized to improve the model's attention to lightning concentration areas, and the validation set is used for verification.

[0031] The test set is used for inference prediction, and multiple evaluation indicators are used to evaluate the prediction accuracy of the trained model.

[0032] Furthermore, it also includes visualizing the three-dimensional lightning imminent prediction results, and correspondingly generating a three-dimensional stereogram with lightning frequency marks and spatial feature marks, and a three-view drawing consisting of a top view, a front view, and a side view of the three-dimensional lightning imminent prediction results.

[0033] A three-dimensional lightning approach prediction system, comprising: a three-dimensional grid construction module, a three-dimensional lightning grid data construction module, a model training module and a prediction module;

[0034] The three-dimensional grid construction module is used to construct a three-dimensional grid covering the spatial range of lightning activity in the target area;

[0035] The three-dimensional lightning data construction module is used to map the historical three-dimensional lightning location data into the three-dimensional grid to obtain the historical three-dimensional lightning grid data;

[0036] The model training module is used to train a three-dimensional lightning near-fall prediction model taking into account changes in spatial characteristics based on the historical three-dimensional lightning grid data;

[0037] The prediction module is used to perform three-dimensional lightning proximity prediction using the trained three-dimensional lightning proximity prediction model.

[0038] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor is used to call and run the computer program stored in the memory to execute any of the methods described above.

[0039] A computer-readable storage medium stores a computer program, wherein the computer program implements the steps of any of the above methods when executed by a processor.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention proposes a three-dimensional lightning proximity prediction method, which constructs a three-dimensional grid covering the spatial range of lightning activity in the target area. Compared with the lightning prediction model in the prior art that only considers a two-dimensional plane or a single height, this method can capture the dynamic changes of lightning in three-dimensional space, and significantly improves the prediction accuracy and spatial resolution capability.

[0042] The method of the present invention enhances the data diversity of the data set and the feature capture capability of the model by performing data enhancement on the historical three-dimensional lightning data to represent the impact of lightning on the surrounding area. At the same time, the three-dimensional lightning proximity prediction model that takes into account the changes in spatial features in the method of the present invention combines three-dimensional convolution, attention mechanism and generative adversarial network, which significantly improves the prediction accuracy of the model.

[0043] The method of the present invention can more comprehensively capture the distribution and diffusion effects of lightning in three-dimensional space by processing historical three-dimensional lightning location data and adding data enhancement operations. It is not limited to a single plane or local area, but reflects the impact of lightning activity from three dimensions of longitude, latitude and altitude. This processing significantly improves the diversity of the data set and enables the model to fully understand the dynamic changes of lightning in three-dimensional space. At the same time, the three-dimensional lightning proximity prediction model used in the present invention combines three-dimensional convolution, an attention mechanism integrating time series, channels and space, and a generative adversarial network, which can comprehensively capture the changing laws of three-dimensional spatial features. Compared with two-dimensional prediction, this three-dimensional analysis method significantly improves the accuracy and spatial resolution of the model. The triple attention mechanism integrating time series, channels and space further enhances the model's ability to capture the spatial and temporal characteristics of lightning in dynamic changes, and significantly improves the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Flow chart of the three-dimensional lightning approach prediction method in the embodiment;

[0045] Figure 2 A schematic diagram of constructing a three-dimensional grid in an embodiment;

[0046] Figure 3 Schematic diagram of data mapping operation and data enhancement operation in the embodiment;

[0047] Figure 4 Schematic diagram of the structure of the three-dimensional lightning approach prediction model in the embodiment;

[0048] Figure 5 It is a schematic diagram of the prediction results in the embodiment. DETAILED DESCRIPTION

[0049] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent modifications of the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0050] Embodiment 1:

[0051] A three-dimensional lightning approach prediction method of the present invention, as Figure 1 As shown, the specific steps are as follows:

[0052] S1. Construct a three-dimensional grid covering the spatial range of lightning activity in the target area;

[0053] S2, mapping the historical three-dimensional lightning location data to the three-dimensional grid to obtain historical three-dimensional lightning grid data;

[0054] S3. Based on the historical three-dimensional lightning grid data, training a three-dimensional lightning near-fall prediction model taking into account changes in spatial characteristics;

[0055] S4. Perform three-dimensional lightning nowcasting prediction using the trained three-dimensional lightning nowcasting prediction model.

[0056] Embodiment 2:

[0057] This embodiment is further designed on the basis of the first embodiment in that the specific steps of constructing a three-dimensional grid covering the lightning activity spatial range of the target area in this embodiment include:

[0058] The grid range of the three-dimensional grid in the plane where the longitude and longitude are located is determined according to the longitude and latitude range of the target area, and the distribution of the three-dimensional grid in the plane where the longitude and longitude are located is delineated based on the length and width of the grid unit. The total number of grids in the longitude is recorded as N lon , the row index in longitude is recorded as col_index, and the total number of grids in latitude is recorded as N lat , the column index on the latitude is recorded as row_index;

[0059] According to the height range of the target area and the height of the grid unit, the distribution of the three-dimensional grid in the height direction is determined. The total number of height layers of the grid is recorded as N lay , the height layer index is recorded as height_index;

[0060] This forms a three-dimensional grid N lon ×N lat ×N lay ,like Figure 2 shown.

[0061] Embodiment three:

[0062] This embodiment is further designed on the basis of the second embodiment in that in this embodiment, the historical three-dimensional lightning location data is mapped to the three-dimensional grid, and the specific steps of obtaining the historical three-dimensional lightning grid data include:

[0063] The historical three-dimensional lightning location data is divided into a number of temporally continuous historical three-dimensional lightning location data sequences according to a certain time step;

[0064] The height data, longitude data and latitude data in each historical three-dimensional lightning location data sequence are used to map the height layer index of the three-dimensional grid, the row index on the plane where the longitude and latitude are located, and the column index, such as Figure 3 As shown, grid cells where lightning occurs in the three-dimensional grid are subjected to data mapping operations for representing the occurrence of lightning and data enhancement operations for representing the impact of lightning on the surrounding areas, thereby obtaining historical three-dimensional lightning grid data that records the lightning occurrence height layer index, row index, column index and grid cell value.

[0065] Embodiment 4:

[0066] This embodiment is further designed on the basis of the third embodiment in that: Figure 3 As shown in the figure, the specific steps of the data mapping operation used to represent the occurrence of lightning in this example include:

[0067] For each of the historical three-dimensional lightning location data sequences, the grid cell value of the mapping position (height_index, row_index, col_index) in the three-dimensional grid corresponding to each lightning occurrence position is increased by 1. After the increase in value, the value of the grid cell in the three-dimensional grid represents the number of lightning occurrences at the position.

[0068] Embodiment five:

[0069] This embodiment is further designed on the basis of the third embodiment in that: Figure 3 As shown in FIG. 1 , the specific steps of the data enhancement operation used in this example to represent the impact of lightning on the surrounding area include:

[0070] For each historical three-dimensional lightning location data sequence, center point processing and neighbor point processing are performed on the grid cells corresponding to the mapping position in the three-dimensional grid and the grid cells around the grid cells according to the number of lightning strikes.

[0071] The specific method of the center point processing is: taking the grid cell where the lightning occurs as the center grid cell and increasing its value by 1 on the original basis to mark the occurrence of lightning at the location;

[0072] The specific method of the neighboring point processing is: the grid cells around the central grid cell are regarded as neighboring grids and their values ​​are increased by 1 on the original basis to reflect the potential impact of lightning on the area.

[0073] The neighboring grids are selected according to the size of the altitude layer. If the size of the altitude layer is greater than or equal to the set threshold, the neighboring grids are the 8 neighboring grid cells at the same altitude layer as the central grid cell; otherwise, the neighboring grids are the 26 grid cells spatially adjacent to the central grid cell.

[0074] Embodiment six:

[0075] This embodiment is further designed on the basis of the third embodiment in that the three-dimensional lightning approach prediction model taking into account the spatial feature changes in this embodiment adopts a three-dimensional convolutional neural network 3DAttUnet-GAN. The three-dimensional convolutional neural network 3DAttUnet-GAN is based on the Unet neural network model, expands the two-dimensional convolution operation therein into a three-dimensional convolution operation, introduces an attention mechanism, and is formed in combination with a generative adversarial network (GAN);

[0076] The three-dimensional lightning near-forecast model includes a 3DAttUnet encoding-decoding structure and a generative adversarial network (GAN).

[0077] The generator of the generative adversarial network (GAN) is a 3DAttUnet encoder-decoder structure, and the discriminator is a binary classification network;

[0078] The generator performs feature encoding and decoding on the input data and generates a prediction result;

[0079] The generative adversarial network inputs the prediction results generated by the generator into the discriminator in combination with real data (real images), and trains the generator based on the authenticity evaluation results output by the discriminator;

[0080] Embodiment seven:

[0081] This embodiment is further designed on the basis of the sixth embodiment in that: Figure 4As shown in the figure, the 3DAttUnet encoding-decoding structure in this example mainly includes an encoder, a decoder, and an attention mechanism module. The encoder part consists of 5 layers of 3D convolution modules, each layer uses depth-separable 3D convolution, Relu activation function, and integrates attention mechanism module and maximum pooling layer; specifically, it includes 1 layer of 3D convolution layer with 64 input channels, using 64 3×3×3 convolution kernels, followed by Relu activation function and attention mechanism module, and finally using the maximum pooling layer for downsampling, 1 layer of 3D convolution layer with 128 input channels, using 128 3×3×3 convolution kernels, followed by Relu activation function and attention mechanism module, and finally using the maximum pooling layer for downsampling, 1 layer of 3D convolution layer with 128 input channels, using 128 3×3×3 convolution kernels, followed by Relu activation function and attention mechanism module, and finally using the maximum pooling layer for downsampling, The channel is 256, using 256 3×3×3 convolution kernels, followed by Relu activation function and attention mechanism module, and finally using the maximum pooling layer for downsampling of the three-dimensional convolution layer, 1 layer of input channels is 512, using 512 3×3×3 convolution kernels, followed by Relu activation function and attention mechanism module, and finally using the maximum pooling layer for downsampling of the three-dimensional convolution layer, 1 layer of input channels is 1024, using 1024 3×3×3 convolution kernels, followed by Relu activation function and attention mechanism module, and finally using the maximum pooling layer for downsampling of the three-dimensional convolution layer;

[0082] The decoder uses trilinear interpolation or deconvolution for upsampling operations, restores the spatial and temporal resolution of the input features layer by layer, and uses depthwise separable 3D convolution combined with an attention module to further enhance the features. Specifically, it includes 4 layers of 3D upsampling convolution layers corresponding to the encoder and an output layer. The output layer converts the number of channels of the feature map into the number of target categories through a 1×1×1 3D convolution, and limits the output result to between 0 and 1 through a Sigmoid activation function to meet the specific task requirements.

[0083] The attention mechanism modules include temporal attention, channel attention and spatial attention. Temporal attention is applied to time series to capture dynamic changes in time series. Channel attention is applied to altitude layers to dynamically adjust the importance of altitude layers, helping the model enhance the modeling ability of spatial-altitude relationships between altitude layers. Spatial attention is applied to the plane space of each altitude layer to adaptively adjust the spatial position weight of each altitude layer, allowing the model to focus more on key areas in the plane space.

[0084] The 3DAttUnet-GAN three-dimensional convolutional neural network refers to the core structure of the generator of the 3DAttUnet encoding-decoding structure as the generative adversarial network. The discriminator is mainly divided into feature extraction and classification. The feature extraction part consists of 4 layers of three-dimensional convolutional layers (kernel size is 3×3×3) and Relu activation function to extract features from the predicted image or label data. The classification part consists of 3 fully connected layers, which are used to map the extracted features to binary classification outputs. It accepts the output of the feature extraction part as input, and the final output is a scalar value indicating the authenticity or category of the input data.

[0085] Overall architecture process:

[0086] Generation phase: The input data is encoded and decoded through the 3DAttUnet network to generate prediction results.

[0087] Adversarial stage: The generated predictions are fed into the discriminator together with the real image, and the discriminator outputs its authenticity assessment (real / fake).

[0088] Optimization process: Through adversarial learning, the generator and the discriminator compete with each other. Under the guidance of the discriminator's feedback, the generator continuously improves the generation effect and finally generates high-precision results.

[0089] Upper part: 3DAttUnet encoding-decoding structure

[0090] Input (cube on the left): The model receives input 3D volume data, in this case 3D lightning grid data.

[0091] Encoding path (yellow and blue cubes): The encoding part of the network extracts high-level features layer by layer through multiple layers of 3D convolution and downsampling. In this process, the spatial dimension of the feature map gradually decreases, but the number of channels increases, forming more expressive abstract features.

[0092] Skip connections (horizontal arrows): The model uses skip connections to pass shallow features in the encoding path directly to the decoding path. These skip connections are the core of the U-Net structure, which can fuse multi-scale features in the decoding stage, ensuring that high-resolution details are preserved during the reconstruction process.

[0093] Decoding path (yellow and blue cubes): The decoding part uses upsampling operations to gradually restore the spatial resolution of the feature map, while using jump connections to integrate the features of the encoding stage. The purpose of this process is to generate an output that matches the dimension of the input data, gradually restore fine-grained information, and achieve accurate segmentation or reconstruction.

[0094] Attention mechanism module: The model introduces a specific attention mechanism module, which includes temporal attention, channel attention and spatial attention, which are respectively applied to time series, height layers and plane space at each height. Through adaptive weighting, the model's attention to key areas is enhanced, the influence of irrelevant features is suppressed, and thus the prediction accuracy is improved.

[0095] Output (cube on the right): The model finally generates a prediction result, and the output three-dimensional feature map will be consistent with the original input in spatial dimension.

[0096] Lower part: Generative Adversarial Network (GAN)

[0097] Generator (3DAttUnet-GAN): The 3DAttUnet structure acts as a generator in the GAN framework, responsible for generating predicted 3D images from input data. The generator encodes and decodes the input data layer by layer and outputs a prediction result that conforms to the target distribution.

[0098] Discriminator: The discriminator in the GAN framework is a binary classification network. Its input includes the prediction results generated by the generator and the real image data (the "Truth" part). The discriminator is responsible for determining whether the input image is real or fake. The feedback of the discriminator is used to guide the optimization of the generator so that the generator can generate high-quality output that is closer to the real data.

[0099] Embodiment eight:

[0100] This embodiment is further designed on the basis of the first embodiment in that, in this embodiment, based on the historical three-dimensional lightning grid data, the specific steps of training the three-dimensional lightning approach prediction model taking into account the spatial feature changes include:

[0101] The historical three-dimensional lightning grid data are divided into a training set and a test set at a ratio of 8:2, and the training set is further divided into a training subset and a validation subset at a ratio of 8:2;

[0102] During the model training process, a timestamp file is set to form a continuous sliding time array with a length of 20 time steps for the three-dimensional lightning grid data, that is, the data of the first 10 time steps are used to predict the data of the next 10 time steps. The learning rate, data batch processing scale and loss function are optimized to improve the model's attention to lightning concentration areas, and the validation set is used for verification.

[0103] The test set is used for inference prediction, and multiple evaluation indicators (hit rate (POD), false alarm rate (FAR), threat score (TS) and Hansson score (HSS)) are used to evaluate the prediction accuracy of the trained model.

[0104] Embodiment nine:

[0105] This embodiment is further designed on the basis of the first embodiment in that it also includes visual processing of the three-dimensional lightning approach prediction results, and correspondingly generates a three-dimensional stereogram with lightning frequency marks and spatial feature marks and three views consisting of a top view, a front view and a side view of the three-dimensional lightning approach prediction results. Among them, the three-dimensional stereogram: the three-dimensional stereogram is used to show the three-dimensional distribution of lightning as a whole. Through the three-dimensional stereogram, the distribution and frequency of lightning at different height levels and geographical locations can be intuitively seen. The figure can represent the number of lightning occurrences at different locations through changes in color and transparency, so that users can clearly identify the high-frequency area and spatial distribution characteristics of lightning. Top view: Looking down on the entire area from above, showing the distribution of lightning on the geographical plane. This view can clearly show the distribution density of lightning at different longitudes and latitudes. Front view: Showing the distribution of lightning in the vertical direction from a frontal perspective, reflecting the frequency changes of lightning at different height levels. This view helps to understand the occurrence law of lightning at different height layers. Side view: Showing the distribution of lightning on the cross section from a side perspective, providing another perspective to observe the height distribution and change trend of lightning.

[0106] Embodiment ten:

[0107] A three-dimensional lightning approach prediction system of the present invention comprises: a three-dimensional grid construction module, a three-dimensional lightning grid data construction module, a model training module and a prediction module;

[0108] The three-dimensional grid construction module is used to construct a three-dimensional grid covering the spatial range of lightning activity in the target area;

[0109] The three-dimensional lightning data construction module is used to map the historical three-dimensional lightning location data into the three-dimensional grid to obtain the historical three-dimensional lightning grid data;

[0110] The model training module is used to train a three-dimensional lightning near-fall prediction model taking into account changes in spatial characteristics based on the historical three-dimensional lightning grid data;

[0111] The prediction module is used to perform three-dimensional lightning proximity prediction using the trained three-dimensional lightning proximity prediction model.

[0112] Embodiment eleven:

[0113] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor is used to call and run the computer program stored in the memory to execute any of the methods described above.

[0114] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0115] Application examples:

[0116] This example simulates the implementation of the three-dimensional lightning proximity prediction method of the present invention. The simulated research area is Ningbo City, Zhejiang Province, and the historical three-dimensional lightning location data used are real lightning that occurred in the geographical scope of Ningbo City, Zhejiang Province from February to September 2024.

[0117] When constructing a three-dimensional grid, this example uses the lower left corner of the study area as the reference point of the three-dimensional grid. Setting the reference point can ensure that the lightning location can be correctly mapped to the corresponding unit of the three-dimensional grid. In this example, the size of the grid unit on the latitude and longitude plane is set to 0.04°, the altitude layer is set to 2 kilometers, and 5 altitude layer intervals are set, 0-2 kilometers, 2-4 kilometers, 4-6 kilometers, 6-8 kilometers, 8 kilometers and above.

[0118] An N can be formed by the following formula lon ×N lat The horizontal grid:

[0119]

[0120] In the process of mapping the three-dimensional grid, the height level where the lightning occurred is determined by comparing the height where the lightning occurred with the preset height interval. The level is represented by an index value and is used to map to the height dimension of the three-dimensional grid.

[0121]

[0122] Among them, height_index is the height index, and Δh is the interval unit of the set height layer, 2KM.

[0123] Calculation of horizontal grid position: The row and column position where lightning occurs in the grid network is calculated using the following formula:

[0124]

[0125] Among them, row_index is the row index, latitude is the latitude of lightning occurrence, base_lat is the latitude of the base point, and grid_size is the grid cell size of 0.04°.

[0126]

[0127] Where col_index is the column index, longitude is the longitude where the lightning occurred, base_lon is the longitude of the base point, and grid_size is the grid cell size of 0.04°.

[0128] During the model training process, the model parameters are continuously optimized and adjusted, including the learning rate, data batch size, and loss function. The present invention adopts a weighted comprehensive loss function, which consists of the mean square error (MSE) and the mean absolute error (MAE). MSE reflects the square error between the predicted value and the true value, and is used to check the short-term prediction performance; MAE measures the absolute error and focuses on the long-term consistency of the prediction trend. Combining the two can evaluate the model performance from multiple angles, and give higher weights to areas where lightning frequently occurs, so that the model pays more attention to the characteristic changes in these areas. The training process continues until a high-precision three-dimensional lightning proximity prediction model is constructed and saved.

[0129] The trained 3D lightning approach prediction model is used to perform 3D lightning approach prediction. The prediction results are visualized as follows: Figure 5 shown. Figure 5 It includes a 3D view and three projection views in different directions. The 3D view shows the 3D lightning distribution predicted by the model. The lightning prediction points show obvious aggregation in the 3D space and are concentrated in certain areas, indicating that the model has successfully captured the main locations where lightning occurs. The XY projection view (top view) clearly shows the distribution of lightning in the XY plane (i.e., the surface plane). It can be seen that lightning is mainly concentrated in certain specific geographical locations, while it is relatively sparse in other areas. This is consistent with the fact that actual lightning usually presents a local concentrated distribution. The XZ projection view (front view) shows the distribution of lightning on the X-axis (horizontal distance) and Z-axis (altitude). It can be observed that most of the lightning occurs in a lower altitude range (the Z-axis is close to the ground), which is consistent with the altitude where lightning frequently occurs in the atmospheric troposphere. The YZ projection view (side view) further verifies the distribution trend of lightning on the Y-axis (latitude) and Z-axis (altitude). The figure also shows that lightning is mainly concentrated in areas with lower altitudes, indicating that the model has successfully captured the range of lightning occurrence in the altitude dimension.

[0130] The above description is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a technician familiar with the technical field within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A three-dimensional lightning approach prediction method, characterized in that: include: Construct a three-dimensional grid covering the spatial extent of lightning activity in the target area; Mapping the historical three-dimensional lightning location data to the three-dimensional grid to obtain the historical three-dimensional lightning grid data; Based on the historical three-dimensional lightning grid data, a three-dimensional lightning near-forecast model taking into account changes in spatial characteristics is trained; The trained three-dimensional lightning nowcasting model is used to perform three-dimensional lightning nowcasting.

2. The three-dimensional lightning imminent prediction method according to claim 1, characterized in that: The specific steps of constructing a three-dimensional grid covering the spatial range of lightning activity in the target area include: The grid range of the three-dimensional grid in the plane where the longitude and longitude are located is determined according to the longitude and latitude range of the target area, and the distribution of the three-dimensional grid in the plane where the longitude and longitude are located is delineated based on the length and width of the grid unit. The total number of grids in the longitude is recorded as N lon , the row index in longitude is recorded as col_index, and the total number of grids in latitude is recorded as N lat , the column index on the latitude is recorded as row_index; According to the height range of the target area and the height of the grid unit, the distribution of the three-dimensional grid in the height direction is determined. The total number of height layers of the grid is recorded as N lay , the height layer index is recorded as height_index; This forms a three-dimensional grid N lon ×N lat ×N lay .

3. The three-dimensional lightning imminent prediction method according to claim 2, characterized in that: The specific steps of mapping the historical three-dimensional lightning location data to the three-dimensional grid to obtain the historical three-dimensional lightning grid data include: The historical three-dimensional lightning location data is divided into a number of temporally continuous historical three-dimensional lightning location data sequences according to a certain time step; The height data, longitude data and latitude data in each historical three-dimensional lightning location data sequence are used to map the height layer index of the three-dimensional grid, the row index on the plane where the longitude and latitude are located, and the column index. The grid cells where lightning occurs in the three-dimensional grid are subjected to data mapping operations for indicating the occurrence of lightning and data enhancement operations for indicating the impact of lightning on the surrounding areas, thereby obtaining historical three-dimensional lightning grid data that records the height layer index, row index, column index and grid cell value of lightning occurrence.

4. The three-dimensional lightning imminent prediction method according to claim 3, characterized in that: The specific steps of the data mapping operation for indicating the occurrence of lightning include: For each of the historical three-dimensional lightning location data sequences, the grid cell value of the mapping position (height_index, row_index, col_index) in the three-dimensional grid corresponding to each lightning occurrence position is increased by 1.

5. The three-dimensional lightning imminent prediction method according to claim 3, characterized in that: The specific steps of the data enhancement operation for representing the impact of lightning on the surrounding area include: For each of the historical three-dimensional lightning location data series, center point processing and neighboring point processing are performed on the grid cells corresponding to the mapping position in the three-dimensional grid and the grid cells around the grid cells; The specific method of the center point processing is: taking the grid cell where the lightning occurs as the center grid cell and increasing its value by 1 on the original basis to mark the lightning occurrence at the location; The specific method of the neighboring point processing is: the grid cells around the central grid cell are regarded as neighboring grids and their values ​​are increased by 1 on the original basis.

6. The three-dimensional lightning imminent prediction method according to claim 1, characterized in that: The three-dimensional lightning proximity prediction model taking into account the changes in spatial features adopts a three-dimensional convolutional neural network 3DAttUnet-GAN. The three-dimensional convolutional neural network 3DAttUnet-GAN is based on the Unet neural network model, expands the two-dimensional convolution operation therein into a three-dimensional convolution operation, introduces an attention mechanism, and is formed in combination with a generative adversarial network; the three-dimensional lightning proximity prediction model includes a 3DAttUnet encoding-decoding structure and a generative adversarial network, the generator of the generative adversarial network is a 3DAttUnet encoding-decoding structure, and the discriminator is a binary classification network; the generator performs feature encoding and decoding on the input data and generates a prediction result; the generative adversarial network inputs the prediction result generated by the generator into the discriminator in combination with real data, and the generator is trained by the authenticity evaluation result output by the discriminator.

7. The three-dimensional lightning imminent prediction method according to claim 6, characterized in that: The 3DAttUnet encoding-decoding structure includes an encoder, a decoder and an attention mechanism module; the encoder part consists of 5 layers of three-dimensional convolution modules, each layer uses a depth-separable three-dimensional convolution, a Relu activation function, and integrates an attention mechanism module and a maximum pooling layer; The decoder uses trilinear interpolation or deconvolution for upsampling, restores the spatial and temporal resolution of the input features layer by layer, and uses depthwise separable 3D convolution combined with an attention module to further enhance the features. The attention mechanism modules include temporal attention, channel attention and spatial attention. Temporal attention is applied to time series to capture dynamic changes in time series; channel attention is applied to altitude layers to dynamically adjust the importance of altitude layers; spatial attention is applied to the plane space of each altitude layer to adaptively adjust the spatial position weight of each altitude layer.

8. The three-dimensional lightning imminent prediction method according to claim 1, characterized in that: The specific steps of training the three-dimensional lightning near-fall prediction model taking into account the changes in spatial characteristics based on the historical three-dimensional lightning grid data include: The historical three-dimensional lightning grid data are divided into a training set and a test set at a ratio of 8:2, and the training set is further divided into a training subset and a validation subset at a ratio of 8:2; During the model training process, a timestamp file is set to form a continuous sliding time array with a length of 20 time steps for the three-dimensional lightning grid data, that is, the data of the first 10 time steps are used to predict the data of the next 10 time steps. The learning rate, data batch processing scale and loss function are optimized to improve the model's attention to lightning concentration areas, and the validation set is used for verification. The test set is used for inference prediction, and multiple evaluation indicators are used to evaluate the prediction accuracy of the trained model.

9. The three-dimensional lightning imminent prediction method according to claim 1, characterized in that: It also includes visualization processing of three-dimensional lightning imminent prediction results, correspondingly generating a three-dimensional stereogram with lightning frequency marks and spatial feature marks and three-view drawings consisting of a top view, a front view and a side view of the three-dimensional lightning imminent prediction results.

10. A three-dimensional lightning approach prediction system, characterized in that: include: Three-dimensional grid construction module, three-dimensional lightning grid data construction module, model training module and prediction module; The three-dimensional grid construction module is used to construct a three-dimensional grid covering the spatial range of lightning activity in the target area; The three-dimensional lightning data construction module is used to map the historical three-dimensional lightning location data into the three-dimensional grid to obtain the historical three-dimensional lightning grid data; The model training module is used to train a three-dimensional lightning near-fall prediction model taking into account changes in spatial characteristics based on the historical three-dimensional lightning grid data; The prediction module is used to perform three-dimensional lightning proximity prediction using the trained three-dimensional lightning proximity prediction model.

Images