A method and system for inverting strong convection in transmission lines

Abstract

This invention discloses a method and system for inverting strong convection along a power transmission line. The method includes: acquiring historical meteorological radar images and DEM data along the power transmission line, and fusing the historical meteorological radar images and DEM data to obtain a topographic image; inputting the topographic image into a convolutional neural network pre-constructed based on topographic weighted features, and generating an initial predicted inversion image corresponding to the topographic image based on the initial network parameters in the convolutional neural network; constructing a target loss function for the convolutional neural network based on the topographic image and the initial predicted inversion image, and training the convolutional neural network based on the target loss function to obtain a strong convection inversion model for the power transmission line; inputting the acquired real-time topographic image into the strong convection inversion model for the power transmission line, and the strong convection inversion model predicts the inversion image in real time. This method can accurately locate the core of strong convection, improve high-resolution edge details, and enhance generalization and robustness in complex scenes.

Description

Technical Field

[0001] This invention belongs to the field of meteorological safety monitoring technology for power transmission lines, and particularly relates to a method and system for inverting strong convection in power transmission lines. Background Technology

[0002] In the field of severe convective weather monitoring along power transmission lines, high-precision meteorological field inversion is crucial for the safe operation of these lines. Traditional image prediction systems based on MS-SSIM (Multiscale Structural Similarity) have significant shortcomings, making it difficult to meet the early warning requirements when power transmission lines traverse complex terrain. First, the lack of terrain factors leads to prediction bias. Traditional methods only compare image brightness, contrast, and structural similarity, neglecting the physical constraints of terrain on the meteorological field, such as mountains blocking airflow and valleys accumulating moisture. Power transmission lines often pass through complex terrain such as mountains and valleys. Ignoring terrain features can result in significant deviations between predicted severe convective location and intensity and actual meteorological changes, failing to provide accurate early warnings for line icing and lightning strike prevention. Second, the fixed-scale weighting is not adaptable enough. The micro-topography along the transmission line has a scale-sensitive impact on the meteorological field. For example, airflow shear in valley areas requires high-resolution edge detail information for prediction, but the fixed-weighting mechanism of traditional MS-SSIM cannot dynamically adjust according to terrain features, resulting in blurred edges in predicted images of key areas and large errors in the intensity of severe convective echoes, making it difficult to meet the millimeter-level accuracy requirements for power transmission line disaster early warning.

[0003] In summary, the traditional MS-SSIM method has significant limitations in meteorological prediction scenarios involving complex terrain along power transmission lines. There is an urgent need for a new inversion system that integrates terrain features and has a dynamic weighting mechanism to improve the accuracy of predicting the impact of severe convective weather on power transmission lines. Summary of the Invention

[0004] This invention provides a method and system for inverting strong convection in transmission lines, which addresses the technical problem of blurred edges in predicted images of key regions and large errors in the intensity of strong convection echoes, making it difficult to meet the millimeter-level accuracy requirements for early warning of transmission line disasters.

[0005] In a first aspect, the present invention provides a method for inverting strong convection in transmission lines, comprising:

[0006] Historical weather radar images and DEM data along the transmission line are acquired, and the historical weather radar images and DEM data are fused to obtain a terrain image;

[0007] The terrain image is input into a convolutional neural network pre-constructed based on terrain weighted features, and an initial predicted inversion image corresponding to the terrain image is generated according to the initial network parameters in the convolutional neural network.

[0008] The target loss function of the convolutional neural network is constructed based on the terrain image and the initial predicted inversion image, and the convolutional neural network is trained based on the target loss function to obtain the transmission line strong convection inversion model.

[0009] The acquired real-time terrain image is input into the transmission line strong convection inversion model, and the transmission line strong convection inversion model outputs a real-time predicted inversion image corresponding to the real-time terrain image.

[0010] Secondly, the present invention provides a transmission line strong convection inversion system, comprising:

[0011] The fusion module is configured to acquire historical meteorological radar images and DEM data along the transmission line, and fuse the historical meteorological radar images and DEM data to obtain a terrain image;

[0012] The generation module is configured to input the terrain image into a convolutional neural network pre-constructed based on terrain weighted features, and generate an initial predicted inversion image corresponding to the terrain image according to the initial network parameters in the convolutional neural network;

[0013] The training module is configured to construct a target loss function for the convolutional neural network based on the terrain image and the initial predicted inversion image, and to train the convolutional neural network based on the target loss function to obtain a strong convection inversion model for transmission lines.

[0014] The output module is configured to input the acquired real-time terrain image into the transmission line strong convection inversion model, and the transmission line strong convection inversion model outputs a real-time predicted inversion image corresponding to the real-time terrain image.

[0015] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the transmission line strong convection inversion method according to any embodiment of the present invention.

[0016] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the transmission line strong convection inversion method according to any embodiment of the present invention.

[0017] The transmission line strong convection inversion method and system proposed in this application differs from the traditional MS-SSIM method, which only focuses on image brightness, contrast, and structural similarity in a fixed manner. It creatively integrates multi-dimensional terrain feature vectors into the core computational framework of the loss function, constructing a dynamic weighting mechanism with terrain physical constraints. Specifically:

[0018] Brightness similarity calculation incorporates the squared Euclidean distance of terrain feature vectors. As an attenuation term, it forces the model to pay close attention to the spatial coupling relationship between meteorological echoes and terrain when making predictions, which significantly reduces the systematic bias of strong convection location caused by ignoring the physical effects of terrain.

[0019] Contrast similarity calculation innovatively embeds terrain complexity adjustment factors. Through local altitude standard deviation and slope gradient norm The contrast weight of areas with drastic slope changes is dynamically amplified, which effectively suppresses the blurring effect produced by traditional methods at the edges of complex terrain and improves the accuracy of echo intensity prediction.

[0020] Structural similarity is calculated using terrain structure weighting coefficients and correction terms. It strengthens the physical constraints of terrain categories on the distribution of meteorological field structure, ensuring that the predicted airflow patterns are more in line with the laws of fluid dynamics.

[0021] The technical effects achieved by the transmission line strong convection inversion method and system of this application are: accurate positioning of the strong convection core, improvement of high-resolution edge details, and enhancement of generalization and robustness in complex scenarios. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart of a method for inverting strong convection in transmission lines provided in an embodiment of the present invention;

[0024] Figure 2 This invention provides a schematic diagram of real meteorological images and model prediction results for a specific embodiment.

[0025] Figure 3 This is a structural block diagram of a high-convection inversion system for transmission lines provided in an embodiment of the present invention;

[0026] Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figure 1 The diagram shows a flowchart of a method for inverting strong convection in transmission lines according to this application.

[0029] like Figure 1 As shown, the method for inverting strong convection in transmission lines specifically includes the following steps:

[0030] Step S101: Acquire historical meteorological radar images and DEM data along the transmission line, and fuse the historical meteorological radar images and DEM data to obtain a terrain image.

[0031] In this step, historical meteorological radar images of terrain including mountains, plains, and canyons along the transmission line are collected. Combined with the DEM data of the transmission line corridor, pixel-level terrain feature vectors are extracted. :

[0032] ,

[0033] In the formula, Represents pixels The altitude at that location This represents the slope value at that point. This represents the terrain category code value: 0 represents plains, 1 represents hillsides, 2 represents valleys, and 3 represents ridges, which are used to correspond to the distribution area of ​​power line towers.

[0034] Step S102: Input the terrain image into a convolutional neural network pre-constructed based on terrain weighted features, and generate an initial prediction inversion image corresponding to the terrain image according to the initial network parameters in the convolutional neural network.

[0035] Step S103: Construct the target loss function of the convolutional neural network based on the terrain image and the initial predicted inversion image, and train the convolutional neural network based on the target loss function to obtain the transmission line strong convection inversion model.

[0036] In this step, the terrain image is subjected to Gaussian filtering and downsampling to obtain the target terrain image;

[0037] The initial predicted inversion image is divided into at least one scale with different resolutions to obtain at least one initial predicted inversion sub-image.

[0038] Calculate the brightness similarity, contrast similarity, and structural similarity between the at least one initial predicted inversion sub-image and the target terrain image. The expression for calculating the brightness similarity between the at least one initial predicted inversion sub-image and the target terrain image is as follows:

[0039] ,

[0040] In the formula, For the first Pixel brightness similarity value For the target terrain image in pixels The average brightness within the central local window For the initial prediction inversion sub-image in pixels The average brightness within the central local window This is a constant term in the calculation of brightness similarity. These are the preset terrain feature weighting coefficients. For pixels in the target terrain image Terrain feature vector, For the pixels in the initial predicted inversion sub-image Terrain feature vector, The squared Euclidean distance between two terrain feature vectors quantifies the degree of deviation between the predicted terrain feature and the target terrain feature; the smaller the distance, the better the terrain feature matches. It is an exponential function;

[0041] The expression for calculating the contrast similarity between the at least one initial predicted inversion sub-image and the target terrain image is as follows:

[0042]

[0043] In the formula, For the first Pixel contrast similarity value This is the adjustment factor for contrast based on terrain complexity. For the target terrain image in pixels The standard deviation of brightness within a local window centered on the value. For the initial prediction inversion sub-image in pixels The standard deviation of brightness within a local window centered on the value. This is a constant term used in the contrast similarity calculation. The standard deviation of elevation in a local area of ​​the target terrain image. , These are the terrain feature weighting coefficients (empirical parameters). Altitude The spatial gradient of slope in the target terrain image. For slope gradient Norm;

[0044] The expression for calculating the structural similarity between the at least one initial predicted inversion sub-image and the target terrain image is as follows:

[0045]

[0046] In the formula, For the first Pixel structural similarity value, For the target terrain image and the initial predicted inversion sub-image at the pixel level The covariance within a local window reflects the degree of linear correlation between the brightness values ​​of two images. For the constant term in the structural similarity calculation, This represents the terrain structure weighting coefficient. This is a terrain structure similarity correction term. For target terrain image in pixels Standard deviation of brightness within a local window For the initial prediction inversion sub-image in pixels Standard deviation of brightness within a local window.

[0047] The terrain weighting value for each pixel location in each initial predicted inversion sub-image is calculated based on the brightness similarity, contrast similarity, and structural similarity. The expression for calculating the terrain weighting value for each pixel location in each initial predicted inversion sub-image is as follows:

[0048] ,

[0049] In the formula, For the first Terrain-weighted structural similarity values ​​of pixels. For the first Pixel brightness similarity value For the first Pixel contrast similarity value For the first Structural similarity value of pixels.

[0050] Based on the terrain weighted value of each pixel in an initial predicted inversion sub-image, the overall similarity between the initial predicted inversion sub-image and the target terrain image is calculated. The expression for calculating the overall similarity between the initial predicted inversion sub-image and the target terrain image is as follows:

[0051] ,

[0052] In the formula, The overall similarity between a given initial predicted inversion sub-image and the target terrain image. The initial number of rows in the predicted inversion sub-image. The initial number of columns for the predicted inversion sub-image. For the first The terrain-weighted structural similarity value of each pixel.

[0053] The various overall similarities are fused across scales to obtain the global similarity between the initial predicted inversion image and the target terrain image. The expression for calculating the global similarity between the initial predicted inversion image and the target terrain image is as follows:

[0054] ,

[0055] ,

[0056] ,

[0057] In the formula, This is the terrain-scale weighting coefficient. This represents the overall similarity between the initial predicted inversion sub-image and the target terrain image at a certain resolution scale. The total number of resolution scales participating in cross-scale fusion. This is the gradient operator for the target terrain image at a certain scale. This is the gradient operator for the initial predicted inverted sub-image at the corresponding scale. Let the norm of the gradient of the target terrain image be denoted as . Let be the norm of the initial predicted inversion sub-image gradient, and t denote the index of the resolution scale. The initial number of rows in the predicted inversion sub-image. The initial number of columns for the predicted inversion sub-image. For pixels in the target terrain image The gradient of the terrain feature vector, For pixels in the target terrain image of Norm.

[0058] The target loss function of the convolutional neural network is constructed based on the global similarity, wherein the expression of the target loss function is:

[0059] ,

[0060] In the formula, Let be the target loss function. This represents the global similarity.

[0061] Step S104: The acquired real-time terrain image is input into the transmission line strong convection inversion model, and the transmission line strong convection inversion model outputs a real-time predicted inversion image corresponding to the real-time terrain image.

[0062] In summary, the method in this application, unlike the traditional MS-SSIM which only focuses on image brightness, contrast, and structural similarity, creatively integrates multi-dimensional terrain feature vectors into the core computational framework of the loss function, constructing a dynamic weighting mechanism with terrain physical constraints. Specifically:

[0063] Brightness similarity calculation incorporates the squared Euclidean distance of terrain feature vectors. As an attenuation term, it forces the model to pay close attention to the spatial coupling relationship between meteorological echoes and terrain when making predictions, which significantly reduces the systematic bias of strong convection location caused by ignoring the physical effects of terrain.

[0064] Contrast similarity calculation innovatively embeds terrain complexity adjustment factors. Through local altitude standard deviation and slope gradient norm The contrast weight of areas with drastic slope changes is dynamically amplified, which effectively suppresses the blurring effect produced by traditional methods at the edges of complex terrain and improves the accuracy of echo intensity prediction.

[0065] Structural similarity is calculated using terrain structure weighting coefficients and correction terms. It strengthens the physical constraints of terrain categories on the distribution of meteorological field structure, ensuring that the predicted airflow patterns are more in line with the laws of fluid dynamics.

[0066] In one specific embodiment, taking the prediction of severe convective weather in a mountainous area as an example, historical meteorological radar images of the area are acquired through a meteorological radar system, with a time span of nearly 5 years, covering different seasons and periods of high incidence of severe convection; digital elevation model data of the corresponding area are collected simultaneously, with a resolution of 30 meters × 30 meters.

[0067] Based on DEM data, pixel-level terrain feature vectors are extracted using GIS spatial analysis techniques:

[0068]

[0069] in, The elevation value of each pixel was obtained directly using the DEM. The altitude of this mountainous area ranges from 800 to 2500 meters. Using the local window difference method, with a window size of 3×3 pixels, the slope distribution from 0 to 47 degrees was obtained; The terrain is divided into four categories using a terrain classification algorithm: 0 (plain), 1 (hillside), 2 (valley), and 3 (ridge), corresponding to the distribution areas of power line towers.

[0070] The terrain feature vectors are standardized, with each dimension's feature value normalized to the [0,1] interval. The meteorological radar image and terrain feature map are cropped to 256×256 pixels to construct the input dataset, where the training, validation, and test sets are divided in a 7:2:1 ratio.

[0071] Four resolution scales are set (original image, 1 / 2 resolution, 1 / 4 resolution, and 1 / 8 resolution). For each resolution scale, the terrain weighting coefficient for each pixel is calculated using the following formula:

[0072]

[0073] For each scale, brightness similarity, contrast similarity, and structural similarity are calculated. In the brightness similarity calculation, parameters are adjusted... Brightness-constrained adaptive attenuation in valley regions (encoding 2) is achieved through terrain vector difference index; contrast similarity gain coefficient. Terrain complexity function weights , On the slope Automatic contrast enhancement weighting for specific regions; structural similarity fusion coefficient. Cosine similarity is used to ensure that the structure of the leeward slope of the mountain is more consistent with the characteristics of hydrodynamics.

[0074] The global similarity is calculated using the following formula across the four resolution scales used in the multi-scale fusion setup:

[0075] ,

[0076] ,

[0077] ,

[0078] terrain The norm gradient mean of the terrain feature vectors at each scale is dynamically allocated.

[0079] The loss function is constructed using the following formula for model training:

[0080] ,

[0081] The model was trained using the Adam optimizer with a learning rate of [missing information]. After 500 iterations, on the validation set... The indicator reached 0.92, compared to the traditional An improvement of 23%. The edge sharpness of predictions for key micro-topographic areas such as valley confluences has significantly improved, and the error in the location of the rainstorm center has been reduced to within 1.5 kilometers. After prediction by the residual feedback prediction system based on topographic weighted features, the results of the actual meteorological images and model predictions are as follows: Figure 2 As shown.

[0082] Please see Figure 3 The diagram shows a structural block diagram of a power transmission line strong convection inversion system according to this application.

[0083] like Figure 3 As shown, the transmission line strong convection inversion system 200 includes a fusion module 210, a generation module 220, a training module 230, and an output module 240.

[0084] The fusion module 210 is configured to acquire historical meteorological radar images and DEM data along the transmission line, and fuse the historical meteorological radar images and DEM data to obtain a terrain image; the generation module 220 is configured to input the terrain image into a convolutional neural network pre-constructed based on terrain weighted features, and generate an initial predicted inversion image corresponding to the terrain image according to the initial network parameters in the convolutional neural network; the training module 230 is configured to construct a target loss function for the convolutional neural network based on the terrain image and the initial predicted inversion image, and train the convolutional neural network based on the target loss function to obtain a transmission line strong convection inversion model; the output module 240 is configured to input the acquired real-time terrain image into the transmission line strong convection inversion model, and the transmission line strong convection inversion model outputs a real-time predicted inversion image corresponding to the real-time terrain image.

[0085] It should be understood that Figure 3 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 3 The various modules in the document will not be described in detail here.

[0086] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the transmission line strong convection inversion method in any of the above method embodiments.

[0087] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0088] Historical weather radar images and DEM data along the transmission line are acquired, and the historical weather radar images and DEM data are fused to obtain a terrain image;

[0089] The terrain image is input into a convolutional neural network pre-constructed based on terrain weighted features, and an initial predicted inversion image corresponding to the terrain image is generated according to the initial network parameters in the convolutional neural network.

[0090] The target loss function of the convolutional neural network is constructed based on the terrain image and the initial predicted inversion image, and the convolutional neural network is trained based on the target loss function to obtain the transmission line strong convection inversion model.

[0091] The acquired real-time terrain image is input into the transmission line strong convection inversion model, and the transmission line strong convection inversion model outputs a real-time predicted inversion image corresponding to the real-time terrain image.

[0092] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the transmission line strong convection inversion system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely disposed relative to a processor, which can be connected to the transmission line strong convection inversion system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0093] Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 4 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 4Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the transmission line strong convection inversion method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the transmission line strong convection inversion system. The output device 340 may include a display screen or other display device.

[0094] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0095] In one implementation, the above-described electronic device is applied in a high-convection inversion system for transmission lines, serving as a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0096] Historical weather radar images and DEM data along the transmission line are acquired, and the historical weather radar images and DEM data are fused to obtain a terrain image;

[0097] The terrain image is input into a convolutional neural network pre-constructed based on terrain weighted features, and an initial predicted inversion image corresponding to the terrain image is generated according to the initial network parameters in the convolutional neural network.

[0098] The target loss function of the convolutional neural network is constructed based on the terrain image and the initial predicted inversion image, and the convolutional neural network is trained based on the target loss function to obtain the transmission line strong convection inversion model.

[0099] The acquired real-time terrain image is input into the transmission line strong convection inversion model, and the transmission line strong convection inversion model outputs a real-time predicted inversion image corresponding to the real-time terrain image.

[0100] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of severe convective inversion retrieval for a power line, characterized in that, The method comprises the following steps: obtain historical weather radar images along the power transmission line and DEM data, and fuse the historical weather radar images and the DEM data to obtain a terrain image; input the terrain image into a convolutional neural network constructed based on terrain weighting features in advance, generate an initial predicted inversion image corresponding to the terrain image according to initial network parameters in the convolutional neural network; construct a target loss function of the convolutional neural network according to the terrain image and the initial predicted inversion image, and train the convolutional neural network based on the target loss function to obtain a power transmission line severe convective inversion model, wherein the target loss function of the convolutional neural network is constructed according to the terrain image and the initial predicted inversion image, and the target loss function of the convolutional neural network comprises: perform Gaussian filtering and downsampling operation on the terrain image to obtain a target terrain image; divide the initial predicted inversion image into at least one scale with different resolutions to obtain at least one initial predicted inversion sub-image; calculate the brightness similarity, contrast similarity and structural similarity between the at least one initial predicted inversion sub-image and the target terrain image; calculate the terrain weighting value of each pixel position in each initial predicted inversion sub-image according to the brightness similarity, contrast similarity and structural similarity; calculate the overall similarity between a certain initial predicted inversion sub-image and the target terrain image according to the terrain weighting value of each pixel position in the certain initial predicted inversion sub-image; fuse the global similarity between the initial predicted inversion image and the target terrain image according to the overall similarity; the expression of the target loss function of the convolutional neural network is: ， In the formula, is a target loss function, is a global similarity; input the obtained real-time terrain image into the power transmission line severe convective inversion model, and the power transmission line severe convective inversion model outputs a real-time predicted inversion image corresponding to the real-time terrain image.

2. A method of severe convective inversion retrieval for a power transmission line according to claim 1, characterized in that, Wherein, the expression for calculating the brightness similarity between the at least one initial predicted inversion sub-image and the target terrain image is: ， In the formula, is the initial predicted inversion sub-image, is the brightness similarity value of the pixel point, is the brightness mean value of the local window centered at the pixel in the target terrain image, is the brightness mean value of the local window centered at the pixel in the initial predicted inversion sub-image, is the constant term of the brightness similarity calculation, is the preset terrain feature weight coefficient, is the terrain feature vector of the pixel in the target terrain image, is the terrain feature vector of the pixel in the initial predicted inversion sub-image, is the Euclidean distance square between the two terrain feature vectors, which quantifies the deviation degree of the predicted terrain feature and the target terrain feature, and the smaller the distance is, the more matched the terrain features are; is an exponential function; the expression for calculating the contrast similarity between the at least one initial predicted inversion sub-image and the target terrain image is: In the formula, For the first Pixel contrast similarity value This is the adjustment factor for contrast based on terrain complexity. For the target terrain image in pixels The standard deviation of brightness within a local window centered on the value. For the initial prediction inversion sub-image in pixels The standard deviation of brightness within a local window centered on the value. This is a constant term used in the contrast similarity calculation. The standard deviation of elevation in a local area of ​​the target terrain image. , These are the terrain feature weighting coefficients. Altitude The spatial gradient of slope in the target terrain image. For slope gradient Norm; the expression for calculating the structural similarity between the at least one initial predicted inversion sub-image and the target terrain image is: In the formula, For the first Pixel structural similarity value, For the target terrain image and the initial predicted inversion sub-image at the pixel level The covariance within a local window reflects the degree of linear correlation between the brightness values ​​of two images. For the constant term in the structural similarity calculation, This represents the terrain structure weighting coefficient. This is a terrain structure similarity correction term. For target terrain image in pixels Standard deviation of brightness within a local window For the initial prediction inversion sub-image in pixels Standard deviation of brightness within a local window.

3. A method of severe convective inversion retrieval for a power transmission line according to claim 1, characterized in that, Wherein, the expression for calculating the terrain weighting value of each pixel position in each initial predicted inversion sub-image is: ， In the formula, is the first a terrain weighted structural similarity value of the pixel point, is the first a brightness similarity value of the pixel point, is the first a contrast similarity value of the pixel point, is the first a structural similarity value of the pixel point.

4. A method of severe convective inversion retrieval for a power transmission line according to claim 1, characterized in that, Wherein, the expression for calculating the overall similarity between a certain initial predicted inversion sub-image and the target terrain image is: ， In the formula, is the overall similarity of a certain initial prediction inversion sub-image and the target terrain image, is the number of rows of the initial prediction inversion sub-image, is the number of columns of the initial prediction inversion sub-image, is the first is the terrain weighted structural similarity value of the pixel point.

5. A method of severe convective inversion detection for a power transmission line according to claim 1, wherein, Wherein, the expression for calculating the global similarity between the initial predicted inversion image and the target terrain image is: ， ， ， wherein, is a terrain scale weight coefficient, is the overall similarity between the initial predicted inversion sub-image and the target terrain image at a certain resolution scale, is the total number of resolution scales involved in cross-scale fusion, is the gradient operator of the target terrain image at a certain scale, is the gradient operator of the initial predicted inversion sub-image at the corresponding scale, is the norm of the target terrain image gradient, is the norm of the initial predicted inversion sub-image gradient, t represents the index of the resolution scale, is the number of rows of the initial predicted inversion sub-image, is the number of columns of the initial predicted inversion sub-image, is the terrain feature vector gradient of pixel in the target terrain image, is the norm of the terrain feature vector gradient of pixel in the target terrain image.

6. A power line severe convective inversion system, characterized by, The method comprises the following steps: a fusion module configured to obtain historical weather radar images along the power transmission line and DEM data, and fuse the historical weather radar images and the DEM data to obtain a terrain image; The generating module is configured to input the terrain image into a convolutional neural network constructed in advance based on terrain weighting features, and generate an initial predicted inversion image corresponding to the terrain image according to initial network parameters in the convolutional neural network. The training module is configured to construct a target loss function of the convolutional neural network according to the terrain image and the initial predicted inversion image, and train the convolutional neural network based on the target loss function to obtain the power line strong convection inversion model, wherein the target loss function of the convolutional neural network is constructed according to the terrain image and the initial predicted inversion image, and includes: performing Gaussian filtering and downsampling operations on the terrain image to obtain a target terrain image; dividing the initial predicted inversion image into at least one scale with different resolutions to obtain at least one initial predicted inversion sub-image; calculating brightness similarity, contrast similarity and structural similarity between the at least one initial predicted inversion sub-image and the target terrain image; calculating terrain weighting values of each pixel position in each initial predicted inversion sub-image according to the brightness similarity, the contrast similarity and the structural similarity; calculating overall similarity between a certain initial predicted inversion sub-image and the target terrain image according to the terrain weighting values of each pixel position in the certain initial predicted inversion sub-image; performing cross-scale fusion on each overall similarity to obtain global similarity between the initial predicted inversion image and the target terrain image; constructing the target loss function of the convolutional neural network according to the global similarity, wherein an expression of the target loss function is: ， In the formula, is a target loss function, is a global similarity; The output module is configured to input the obtained real-time terrain image into the power line strong convection inversion model, and the power line strong convection inversion model outputs a real-time predicted inversion image corresponding to the real-time terrain image.

7. An electronic device, comprising: comprise: at least one processor, and a memory connected with the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1 to 5.

Images