Meteorological numerical mode analysis method and system based on super-resolution

Abstract

The invention discloses a meteorological numerical mode analysis method and system based on super-resolution, and belongs to the technical field of data processing. The system comprises a data acquisition module, a mode analysis module, a numerical value processing module and a visualization module. The data acquisition module is used for acquiring an algorithm library, a historical log and an original meteorological image of a specified region; the mode analysis module divides grids on the original meteorological image, calculates a difference coefficient between adjacent grids according to a historical log, adjusts the size of the grids according to the difference coefficient, and marks an overlapping region; the numerical value processing module calculates an aging coefficient of the grid area through a historical log, selects an algorithm in an algorithm library for each grid area according to the aging coefficient for matching, and reconstructs an original meteorological image by adopting a super-resolution technology to generate a target meteorological image; and the visualization module displays the reconstructed target meteorological image through a data center visualization screen, establishes a generation record according to the original meteorological image and the target meteorological image, and stores the generation record in a historical log.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for analyzing meteorological numerical patterns based on super-resolution. Background Art

[0002] With the rapid development of deep learning, super-resolution technology based on convolutional neural networks and generative adversarial networks provides new ideas and methods for reducing the size of meteorological data, which can effectively generate high-resolution meteorological fields while maintaining high accuracy. This research can not only improve the performance of meteorological forecast models, but also provide more reliable data support for the monitoring and protection of meteorological trends, thereby improving the level of refinement of meteorological services.

[0003] At present, statistical experience or dynamic equation methods are usually used for meteorological downscaling. There are certain problems in establishing linear or nonlinear connections between regional meteorological variables and local meteorological variables to output high-resolution meteorological images. On the one hand, the use of numerical models to produce low-scale data will face problems such as large computing resources required and long computing time. The downscaling calculation of traditional numerical models requires the computing center to calculate continuously for tens of minutes to several hours, which cannot meet the business needs of fast preview. On the other hand, existing algorithms usually split the image into several grid areas first, so that the entire image can be processed in a sequence when processing the image. However, this method can only use a limited spatial range when using input information, fails to utilize the complementary advantages of global statistical information and strong local fitting capabilities, and cannot pay attention to the degree of correlation between grid areas. It is impossible to give full play to its potential and reduce processing time. Therefore, a more intelligent and efficient meteorological numerical model analysis technology solution is needed at this stage to solve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide a super-resolution-based meteorological numerical model analysis method and system to solve the problems raised in the above-mentioned background technology.

[0005] In order to solve the above technical problems, the present invention provides a meteorological numerical model analysis method based on super-resolution, comprising the following steps:

[0006] S100, collects algorithm libraries, historical logs and original meteorological images of a specified area, and analyzes meteorological elements contained in the original meteorological images.

[0007] S200, dividing the original meteorological image into grids, calculating the difference coefficients between adjacent grids according to the historical logs, adjusting the size of the grids according to the difference coefficients and marking the overlapping areas.

[0008] S300, calculating the timeliness coefficient of the grid area through historical logs, selecting an algorithm in the algorithm library for matching for each grid area according to the timeliness coefficient, and using super-resolution technology to reconstruct the original meteorological image to generate a target meteorological image.

[0009] S400: Display the reconstructed target meteorological image on the visualization screen of the data center, create a generation record based on the original meteorological image and the target meteorological image, and store it in a history log.

[0010] Super-resolution is an image processing technique that aims to generate high-resolution output from low-resolution input. This technique is very effective in improving image quality, image details, and image resolution.

[0011] In S100, the algorithm library refers to a software library containing different algorithms, each algorithm uses different image processing techniques to achieve super-resolution processing of images. The historical log refers to the generation record of each meteorological image processing, each generation record includes the region, operating parameters, original meteorological image and target meteorological image.

[0012] Operation parameters refer to the parameter information of image processing after the original meteorological image is divided into grid areas, including the number of grid areas, the identification code, meteorological characteristics, area, algorithm used and processing time of each grid area. Meteorological images show the distribution and changes of various meteorological elements through various contour lines and symbols. Meteorological elements refer to various physical quantities used to characterize meteorological conditions, including temperature, air pressure, humidity, wind speed and precipitation.

[0013] These algorithms together constitute a complete numerical weather forecast system, each with its own functions and working together to achieve the whole process from observation to forecast. The choice of algorithm is the key to ensure the accuracy of numerical model forecast.

[0014] The identification code is used to distinguish different grid areas and is the unique identifier of the grid area. The identification code can not only distinguish all grid areas in an original meteorological image, but also ensure the uniqueness of the grid area in all original meteorological images in the entire historical log, and distinguish different original meteorological images and different grid areas.

[0015] In S200, the specific steps are as follows:

[0016] S201, obtain the region S corresponding to the original meteorological image OMI, and mark the generation record of the region S in the historical log. Sum the number of grids in the operating parameters in all marked generation records and calculate the average value to obtain the average number of grids W ave , evenly divide W in the original meteorological image ave grid areas of equal area.

[0017] Vi Transformer is a deep learning model based on the Transformer architecture, specifically designed for image processing tasks in the field of computer vision. It has achieved remarkable success in image classification and other visual tasks. Unlike traditional convolutional neural networks, Vi Transformer is entirely based on the self-attention mechanism in the Transformer. This enables the model to learn complex relationships between the global and local in the image, helping to process long-range dependencies and capture important features in the image.

[0018] Vi Transformer is used to divide the original meteorological image into W ave The model divides the image into blocks of equal size and converts each block into a vector through an embedding layer. This block-based approach allows the model to process the entire image in a sequential manner, which, like a text sequence, provides input for the self-attention mechanism. Similar to the traditional Transformer model, ViTransformer uses a multi-head attention mechanism that can focus on different grid areas of the original meteorological image in parallel.

[0019] S202, set the sampling number m, evenly set m sampling points in each grid area, and obtain the meteorological elements of each sampling point. Count the number of meteorological element types k, analyze the position distribution of all grid areas in the original meteorological image, and combine all two adjacent grid areas into associated grid pairs. Each grid area can be combined with b adjacent grid areas at other locations to form b associated grid pairs.

[0020] S203, analyze the position coordinates (X, Y) of the sampling points in each grid area, and calculate the Euclidean distance between the sampling points in different grid areas under the associated grid pair. According to the principle of the shorter the Euclidean distance, the better the priority, the two sampling points belonging to different grid areas under the associated grid pair are sequentially associated. Obtain the meteorological elements corresponding to each sampling point, and substitute them into the formula to calculate the difference coefficient CX of each associated grid pair:

[0021]

[0022] In the formula, ρ n is the Euclidean distance between the nth pair of associated sampling points under the associated grid pair, is the specific value of the i-th meteorological element corresponding to the n-th sampling point in one of the grid areas under the associated grid pair, It is the specific value of the i-th meteorological element corresponding to the n-th sampling point in another grid area under the associated grid pair.

[0023] S204, set the index threshold d, calculate the standard deviation of the difference coefficients of all associated grid pairs as the difference index. When the difference index is greater than d, calculate the average value CX of the difference coefficients of all associated grid pairsave , the marker difference coefficient is greater than CX ave The number of occurrences g of the grid area GD1 in all the associated grid pairs of the tag is counted, as well as the sum of the difference coefficients CX of the associated grid pairs to which the grid area GD1 belongs h , substitute into the formula g×CX h The anomaly index of grid area GD1 is calculated in .

[0024] S205. Calculate the abnormal index of each grid area respectively, and count the number of grid areas with abnormal index not equal to zero, e, and divide e by 2 and round down to get u. Sort all grid areas according to the abnormal index from large to small, and select the first u grid areas as abnormal grids according to the sorting order, and compare them with the abnormal grid GD ab Other adjacent grid areas are GD ab Set the basic area l and obtain the difference coefficient CX between the adjusted grid and the corresponding abnormal grid v , substitute into the formula to get the adjustment area of ​​each adjustment grid:

[0025]

[0026] In the formula, area is the adjustment area, CX sum It is the sum of the difference coefficients between the abnormal grid and all the adjusted grids.

[0027] S206, after controlling each adjustment grid to expand an area of ​​the same size as the corresponding adjustment area in the abnormal grid, enter step S202 again to analyze the difference coefficient and calculate the difference index. When the difference index is greater than d, continue to expand the adjustment grid until the difference index is not greater than d. When the difference index is not greater than d, the grid adjustment is completed, and the overlapping areas between different grid areas are marked.

[0028] Expanding and adjusting the size of the grid to generate overlapping areas can better aggregate cross-window information. By generating overlapping cross-regions, the interaction between features in adjacent grid areas is enhanced, thereby increasing the receptive field.

[0029] In S300, the specific steps are as follows:

[0030] S301, establish a comparison set, put all the meteorological features of each grid area in all the mark generation records into the comparison set. Extract the meteorological feature FT2 of the grid area GD2 in the original meteorological image, and calculate the similarity between each element in the comparison set and the meteorological feature FT2. Establish a coding set for the grid area GD2 and set the similarity threshold SMI thr , get the similarity greater than SMI in the comparison set thr The meteorological characteristics correspond to the identification codes of the grid areas, and are all put into the code set of the grid area GD2.

[0031] S302: Obtain identification code BM in the code set P Corresponding grid area GD P , extract the grid area GD P The meteorological features in the original meteorological image and the target meteorological image are calculated, and the similarity SMI of the meteorological features between the two is calculated. p . Get the identification code BM P Corresponding grid area MJ P and processing time P , substitute into the formula to calculate the identification code BM P The time efficiency coefficient:

[0032]

[0033] Where TZ P To identify the code BM P The time efficiency coefficient of , α is a constant.

[0034] S303, respectively calculate the timeliness coefficient of each identification code in the code set, analyze the adopted algorithm corresponding to each identification code, and classify all identification codes according to whether the adopted algorithm is the same. Calculate the average timeliness coefficient of all identification codes in each category as the timeliness index of the corresponding category, mark the category with the largest timeliness index, and use the adopted algorithm of the marked category as the grid area GD P By analogy, the selection algorithm of each grid area under the original meteorological image OMI is analyzed.

[0035] S304: Group all grid areas under the original meteorological image OMI according to the selected algorithm, process the images of all grid areas in the corresponding group with each selected algorithm, and use multi-algorithm multi-threaded parallel technology to perform super-resolution downsizing on the original meteorological image OMI, thereby generating a target meteorological image TMI. Analyze the position of each grid area in the original meteorological image OMI and map it to the target meteorological image TMI.

[0036] The super-resolution process of meteorological elements requires learning the features of elements at different spatial scales rather than simple edge features such as contour details, so it is essential to introduce generative adversarial training techniques. In the early model training process of super-resolution, the generator is trained to generate high-resolution images, while the discriminator is trained to distinguish between real high-resolution images and images generated by the generator. This adversarial training can generate more realistic high-resolution images.

[0037] The basic principle of super-resolution is to use statistical experience or dynamic equation methods to establish linear or nonlinear connections between large-scale meteorological variables and regional meteorological variables, and to convert forecast data with coarser spatial resolution and longer time intervals into meteorological data with higher resolution and finer time intervals. Meteorological elements commonly used for downscaling include precipitation, temperature, air pressure, humidity and wind speed.

[0038] In actual operation, the 3km resolution data and 1km resolution data of the numerical model only differ by 3 times in resolution at the data grid level. Although the scales of their meteorological elements are different, we can use the deep learning network to learn historical data, construct 3km and 1km model distribution pairs, and learn the mapping relationship from the 3km scale weather process to the 1km scale into the network.

[0039] S305, marking the area where the grid areas on the target meteorological image TMI overlap, establishing an algorithm set for the overlapping area OA, and placing the selected algorithm of the grid area to which the overlapping area OA belongs into the algorithm set. When analyzing the overlapping area OA using the algorithm ALG in the algorithm set, the sum of the difference coefficients between all the grid areas adjacent to the overlapping area OA is calculated as the violation index of the algorithm ALG.

[0040] S306, respectively calculating the violation index of each algorithm in the algorithm set, and adjusting the image of the overlapping area OA to the target meteorological image processed by the algorithm with the lowest violation index. Similarly, adjusting the image for each overlapping area on the target meteorological image TMI, and realizing super-resolution reconstruction of the entire image from the original meteorological image OMI to the target meteorological image TMI.

[0041] In S400, the target meteorological image after reconstruction is displayed on the visualization screen of the data center. A unique identification code is set for each grid area in the original meteorological image, and the meteorological characteristics and image processing time are analyzed to generate operation parameters. The region to which the meteorological image belongs, the operation parameters, the original meteorological image, and the target meteorological image are packaged together as a generation record and stored in the historical log.

[0042] A meteorological numerical model analysis system based on super-resolution includes a data acquisition module, a model analysis module, a numerical processing module and a visualization module.

[0043] The data acquisition module is used to collect algorithm libraries, historical logs, and original meteorological images of designated areas.

[0044] The pattern analysis module divides the original meteorological image into grids, calculates the difference coefficient between adjacent grids based on historical logs, adjusts the size of the grid according to the difference coefficient, and marks the overlapping area.

[0045] The numerical processing module calculates the timeliness coefficient of the grid area through historical logs, selects an algorithm in the algorithm library for each grid area according to the timeliness coefficient, and uses super-resolution technology to reconstruct the original meteorological image to generate the target meteorological image.

[0046] The visualization module displays the reconstructed target meteorological image through the visualization screen of the data center, creates a generation record based on the original meteorological image and the target meteorological image and stores it in the historical log.

[0047] The data acquisition module includes a meteorological image acquisition unit, a historical data acquisition unit and an algorithm library acquisition unit.

[0048] The meteorological image acquisition unit is used to collect original meteorological images of the specified area. Meteorological images show the distribution and changes of various meteorological elements through various contour lines and symbols. Meteorological elements refer to various physical quantities used to characterize meteorological conditions.

[0049] The historical data collection unit is used to collect the generation record of each meteorological image processing. Each generation record includes the region, operating parameters, original meteorological image and target meteorological image.

[0050] Operation parameters refer to the parameter information for image processing after the original meteorological image is divided into grid areas, including the number of grid areas, as well as the identification code, meteorological characteristics, area, algorithm used and processing time of each grid area.

[0051] The algorithm library acquisition unit is used to acquire software libraries containing different algorithms, and each algorithm realizes super-resolution processing of images through different image processing technologies.

[0052] The pattern analysis module includes a grid division unit and an associated expansion unit.

[0053] The grid division unit is used to divide the grid area, establish associated grid pairs and calculate the difference coefficient.

[0054] First, obtain the region S corresponding to the original meteorological image OMI, and mark the generation record of the region S in the historical log. After summing up the number of grids in the operating parameters in all marked generation records, calculate the average value to get the average number of grids W ave , evenly divide W in the original meteorological image ave grid areas of equal area.

[0055] Secondly, set the sampling number m, evenly set m sampling points in each grid area, and obtain the meteorological elements of each sampling point. Count the number of meteorological element types k, analyze the position distribution of all grid areas in the original meteorological image, and combine all two adjacent grid areas into associated grid pairs. Each grid area can be combined with b adjacent grid areas at other locations into b associated grid pairs.

[0056] Finally, the position coordinates (X, Y) of the sampling points in each grid area are analyzed, and the Euclidean distance between the sampling points in different grid areas under the associated grid pair is calculated. According to the principle of the shorter the Euclidean distance, the better the priority, the two sampling points belonging to different grid areas under the associated grid pair are sequentially associated. According to the meteorological elements corresponding to each sampling point, the difference coefficient of each associated grid pair is calculated.

[0057] The associative expansion unit is used to adjust the size of the grid according to the difference coefficient and to mark the overlapping area.

[0058] First, set the index threshold d and calculate the standard deviation of the difference coefficients of all associated grid pairs as the difference index. When the difference index is greater than d, calculate the average value CX of the difference coefficients of all associated grid pairs ave , the marker difference coefficient is greater than CX ave The associated grid pairs.

[0059] Secondly, count the number of occurrences g of the grid area GD1 in all the marked associated grid pairs, and the sum of the difference coefficients CX of the associated grid pairs to which the grid area GD1 belongs h , according to the formula g×CX h The anomaly index of grid area GD1 is calculated in . The anomaly index of each grid area is calculated separately, and the number of grid areas with non-zero anomaly indexes is counted, and e is divided by 2 and rounded down to get u.

[0060] Then, all grid areas are sorted from large to small according to the anomaly index, and the first u grid areas are selected as abnormal grids according to the sorting order. ab Other adjacent grid areas are GD ab According to the difference coefficient CX between the adjusted grid and the corresponding abnormal grid v , analyze the adjustment area of ​​each adjustment grid.

[0061] Finally, after controlling each adjustment grid to expand the area of ​​the same size as the corresponding adjustment area in the abnormal grid, calculate the difference index again. If the difference index is still greater than d, continue to expand the adjustment grid until the difference index is no greater than d. When the difference index is no greater than d, the grid adjustment is completed, and the overlapping areas between different grid areas are marked.

[0062] The numerical processing module includes a time-effect analysis unit and an image processing unit.

[0063] The time-effect analysis unit is used to calculate the time-effect coefficient of each grid area and match the algorithm.

[0064] First, a comparison set is established, and all meteorological features of each grid area in all the mark generation records are put into the comparison set. The meteorological feature FT2 of the grid area GD2 in the original meteorological image is extracted, and the similarity between each element in the comparison set and the meteorological feature FT2 is calculated.

[0065] Secondly, establish a coding set for grid area GD2 and set the similarity threshold SMI thr , get the similarity greater than SMI in the comparison set thr Meteorological features correspond to the identification codes of the grid areas, and all are put into the code set of the grid area GD2. Get the identification code BM in the code set P Corresponding grid area GD P , extract the grid area GD P The meteorological features in the original meteorological image and the target meteorological image are calculated, and the similarity SMI of the meteorological features between the two is calculated. P .

[0066] Finally, according to the identification code BM P Corresponding grid area MJ P and processing time P Calculate the timeliness coefficient, calculate the timeliness coefficient of each identification code in the code set, and classify all identification codes according to whether the adopted algorithm is the same. Calculate the average timeliness coefficient of all identification codes in the same category as the timeliness index of the corresponding class, mark the class with the largest timeliness index, and use the adopted algorithm of the marked class as the grid area GD P The selected algorithm of each grid area is analyzed in this way.

[0067] The image processing unit uses super-resolution technology to reconstruct the original meteorological image and generate the target meteorological image.

[0068] First, all grid areas under the original meteorological image OMI are grouped according to the selected algorithm. Each selected algorithm processes the images of all grid areas in the corresponding group. The original meteorological image OMI is super-resolution downsized using multi-algorithm multi-threaded parallel technology to generate the target meteorological image TMI. The position of each grid area in the original meteorological image OMI is mapped to the target meteorological image TMI.

[0069] Secondly, the area where the grid areas on the target meteorological image TMI overlap is marked, and an algorithm set is established for the overlapping area OA, and the selected algorithm of the grid area to which the overlapping area OA belongs is placed in the algorithm set. When analyzing the overlapping area OA using the algorithm ALG in the algorithm set, the sum of the difference coefficients between all grid areas adjacent to the overlapping area OA is calculated as the violation index of the algorithm ALG.

[0070] Finally, the violation index of each algorithm in the algorithm set is calculated respectively, and the image of the overlapping area OA is adjusted to the target meteorological image processed by the algorithm with the lowest violation index. Similarly, the image is adjusted for each overlapping area on the target meteorological image TMI to achieve super-resolution reconstruction of the entire image from the original meteorological image OMI to the target meteorological image TMI.

[0071] The visualization module displays the reconstructed target meteorological image through the visualization screen of the data center, and packages the region to which the meteorological image belongs, the operating parameters, the original meteorological image and the target meteorological image together as a generation record, which is stored in the historical log.

[0072] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0073] Accurate grid division and difference analysis: Through the division of grid areas and the calculation of difference coefficients, the grid areas in the meteorological image can be accurately delineated, ensuring the detailed analysis of meteorological data. This precise division enables the differences between adjacent grids to be effectively identified, thus providing a more reliable basis for subsequent data processing.

[0074] Efficient time efficiency coefficient calculation: The time efficiency coefficient of each grid area is calculated using data from historical logs to ensure that the best algorithm can be dynamically selected based on the latest data. This function enables the system to quickly adapt to meteorological changes and improves the response speed and accuracy of the model.

[0075] Multi-algorithm parallel processing: The use of super-resolution technology combined with multi-algorithm parallel processing improves the overall efficiency of meteorological image processing. During the processing, various algorithms can operate on different grid areas at the same time, significantly shortening the processing time and improving the processing quality.

[0076] Visualization and recording function: not only can the generated target meteorological image be displayed, but also all relevant operating parameters, original meteorological images and generation records can be recorded. These functions facilitate users to conduct subsequent analysis and comparison, and enhance the ease of use and practical value of the system.

[0077] Abnormal grid identification and adjustment: The system has the ability to identify and adjust abnormal grids. By analyzing the difference coefficient, it can identify and adjust the grid areas with abnormalities. This ability is particularly important in meteorological analysis and helps to detect potential problems early in complex weather conditions.

[0078] The realization of the above-mentioned beneficial effects shows that the present invention has important technical advantages in enhancing the accuracy, efficiency and practicability of meteorological numerical model analysis, and provides a more efficient method for meteorological analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0080] Figure 1 It is a flow chart of the meteorological numerical model analysis method based on super-resolution of the present invention;

[0081] Figure 2 It is a structural schematic diagram of the meteorological numerical model analysis system based on super-resolution of the present invention. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0083] See also Figure 1 The present invention provides a meteorological numerical model analysis method based on super-resolution, comprising the following steps:

[0084] S100, collects algorithm libraries, historical logs and original meteorological images of a specified area, and analyzes meteorological elements contained in the original meteorological images.

[0085] S200, dividing the original meteorological image into grids, calculating the difference coefficients between adjacent grids according to the historical logs, adjusting the size of the grids according to the difference coefficients and marking the overlapping areas.

[0086] S300, calculating the timeliness coefficient of the grid area through historical logs, selecting an algorithm in the algorithm library for matching for each grid area according to the timeliness coefficient, and using super-resolution technology to reconstruct the original meteorological image to generate a target meteorological image.

[0087] S400: Display the reconstructed target meteorological image on the visualization screen of the data center, create a generation record based on the original meteorological image and the target meteorological image, and store it in a history log.

[0088] Super-resolution is an image processing technique that aims to generate high-resolution output from low-resolution input. This technique is very effective in improving image quality, image details, and image resolution.

[0089] In S100, the algorithm library refers to a software library containing different algorithms, each algorithm uses different image processing techniques to achieve super-resolution processing of images. The historical log refers to the generation record of each meteorological image processing, each generation record includes the region, operating parameters, original meteorological image and target meteorological image.

[0090] Operation parameters refer to the parameter information of image processing after the original meteorological image is divided into grid areas, including the number of grid areas, the identification code, meteorological characteristics, area, algorithm used and processing time of each grid area. Meteorological images show the distribution and changes of various meteorological elements through various contour lines and symbols. Meteorological elements refer to various physical quantities used to characterize meteorological conditions, including temperature, air pressure, humidity, wind speed and precipitation.

[0091] These algorithms together constitute a complete numerical weather forecast system, each with its own functions and working together to achieve the whole process from observation to forecast. The choice of algorithm is the key to ensure the accuracy of numerical model forecast.

[0092] The identification code is used to distinguish different grid areas and is the unique identifier of the grid area. The identification code can not only distinguish all grid areas in an original meteorological image, but also ensure the uniqueness of the grid area in all original meteorological images in the entire historical log, and distinguish different original meteorological images and different grid areas.

[0093] In S200, the specific steps are as follows:

[0094] S201, obtain the region S corresponding to the original meteorological image OMI, and mark the generation record of the region S in the historical log. Sum the number of grids in the operating parameters in all marked generation records and calculate the average value to obtain the average number of grids W ave , evenly divide W in the original meteorological image ave grid areas of equal area.

[0095] Vi Transformer is a deep learning model based on the Transformer architecture, specifically designed for image processing tasks in the field of computer vision. It has achieved remarkable success in image classification and other visual tasks. Unlike traditional convolutional neural networks, Vi Transformer is entirely based on the self-attention mechanism in the Transformer. This enables the model to learn complex relationships between the global and local in the image, helping to process long-range dependencies and capture important features in the image.

[0096] Vi Transformer is used to divide the original meteorological image into W aveThe model divides the image into blocks of equal size and converts each block into a vector through an embedding layer. This block-based approach allows the model to process the entire image in a sequential manner, which, like a text sequence, provides input for the self-attention mechanism. Similar to the traditional Transformer model, ViTransformer uses a multi-head attention mechanism that can focus on different grid areas of the original meteorological image in parallel.

[0097] S202, set the sampling number m, evenly set m sampling points in each grid area, and obtain the meteorological elements of each sampling point. Count the number of meteorological element types k, analyze the position distribution of all grid areas in the original meteorological image, and combine all two adjacent grid areas into associated grid pairs. Each grid area can be combined with b adjacent grid areas at other locations to form b associated grid pairs.

[0098] S203, analyze the position coordinates (X, Y) of the sampling points in each grid area, and calculate the Euclidean distance between the sampling points in different grid areas under the associated grid pair. According to the principle of the shorter the Euclidean distance, the better the priority, the two sampling points belonging to different grid areas under the associated grid pair are sequentially associated. Obtain the meteorological elements corresponding to each sampling point, and substitute them into the formula to calculate the difference coefficient CX of each associated grid pair:

[0099]

[0100] In the formula, ρ n is the Euclidean distance between the nth pair of associated sampling points under the associated grid pair, is the specific value of the i-th meteorological element corresponding to the n-th sampling point in one of the grid areas under the associated grid pair, It is the specific value of the i-th meteorological element corresponding to the n-th sampling point in another grid area under the associated grid pair.

[0101] S204, set the index threshold d, calculate the standard deviation of the difference coefficients of all associated grid pairs as the difference index. When the difference index is greater than d, calculate the average value CX of the difference coefficients of all associated grid pairs ave , the marker difference coefficient is greater than CX ave The number of occurrences g of the grid area GD1 in all the associated grid pairs of the tag is counted, as well as the sum of the difference coefficients CX of the associated grid pairs to which the grid area GD1 belongs h , substitute into the formula g×CX h The anomaly index of grid area GD1 is calculated in .

[0102] S205. Calculate the abnormal index of each grid area respectively, and count the number of grid areas with abnormal index not equal to zero, e, and divide e by 2 and round down to get u. Sort all grid areas according to the abnormal index from large to small, and select the first u grid areas as abnormal grids according to the sorting order, and compare them with the abnormal grid GD ab Other adjacent grid areas are GD ab Set the basic area l and obtain the difference coefficient CX between the adjusted grid and the corresponding abnormal grid v , substitute into the formula to get the adjustment area of ​​each adjustment grid:

[0103]

[0104] In the formula, area is the adjustment area, CX sum It is the sum of the difference coefficients between the abnormal grid and all the adjusted grids.

[0105] S206, after controlling each adjustment grid to expand an area of ​​the same size as the corresponding adjustment area in the abnormal grid, enter step S202 again to analyze the difference coefficient and calculate the difference index. When the difference index is greater than d, continue to expand the adjustment grid until the difference index is not greater than d. When the difference index is not greater than d, the grid adjustment is completed, and the overlapping areas between different grid areas are marked.

[0106] Expanding and adjusting the size of the grid to generate overlapping areas can better aggregate cross-window information. By generating overlapping cross-regions, the interaction between features in adjacent grid areas is enhanced, thereby increasing the receptive field.

[0107] In S300, the specific steps are as follows:

[0108] S301, establish a comparison set, put all the meteorological features of each grid area in all the mark generation records into the comparison set. Extract the meteorological feature FT2 of the grid area GD2 in the original meteorological image, and calculate the similarity between each element in the comparison set and the meteorological feature FT2. Establish a coding set for the grid area GD2 and set the similarity threshold SMI thr , get the similarity greater than SMI in the comparison set thr The meteorological characteristics correspond to the identification codes of the grid areas, and are all put into the code set of the grid area GD2.

[0109] S302: Obtain identification code BM in the code set P Corresponding grid area GD P , extract the grid area GD P The meteorological features in the original meteorological image and the target meteorological image are calculated, and the similarity SMI of the meteorological features between the two is calculated. p . Get the identification code BM PCorresponding grid area MJ P and processing time P , substitute into the formula to calculate the identification code BM P The time efficiency coefficient:

[0110]

[0111] Where TZ P To identify the code BM P The time efficiency coefficient of , α is a constant.

[0112] S303, respectively calculate the timeliness coefficient of each identification code in the code set, analyze the adopted algorithm corresponding to each identification code, and classify all identification codes according to whether the adopted algorithm is the same. Calculate the average timeliness coefficient of all identification codes in each category as the timeliness index of the corresponding category, mark the category with the largest timeliness index, and use the adopted algorithm of the marked category as the grid area GD P By analogy, the selection algorithm of each grid area under the original meteorological image OMI is analyzed.

[0113] S304: Group all grid areas under the original meteorological image OMI according to the selected algorithm, process the images of all grid areas in the corresponding group with each selected algorithm, and use multi-algorithm multi-threaded parallel technology to perform super-resolution downsizing on the original meteorological image OMI, thereby generating a target meteorological image TMI. Analyze the position of each grid area in the original meteorological image OMI and map it to the target meteorological image TMI.

[0114] The super-resolution process of meteorological elements requires learning the features of elements at different spatial scales rather than simple edge features such as contour details, so it is essential to introduce generative adversarial training techniques. In the early model training process of super-resolution, the generator is trained to generate high-resolution images, while the discriminator is trained to distinguish between real high-resolution images and images generated by the generator. This adversarial training can generate more realistic high-resolution images.

[0115] The basic principle of super-resolution is to use statistical experience or dynamic equation methods to establish linear or nonlinear connections between large-scale meteorological variables and regional meteorological variables, and to convert forecast data with coarser spatial resolution and longer time intervals into meteorological data with higher resolution and finer time intervals. Meteorological elements commonly used for downscaling include precipitation, temperature, air pressure, humidity and wind speed.

[0116] In actual operation, the 3km resolution data and 1km resolution data of the numerical model only differ by 3 times in resolution at the data grid level. Although the scales of their meteorological elements are different, we can use the deep learning network to learn historical data, construct 3km and 1km model distribution pairs, and learn the mapping relationship from the 3km scale weather process to the 1km scale into the network.

[0117] S305, marking the area where the grid areas on the target meteorological image TMI overlap, establishing an algorithm set for the overlapping area OA, and placing the selected algorithm of the grid area to which the overlapping area OA belongs into the algorithm set. When analyzing the overlapping area OA using the algorithm ALG in the algorithm set, the sum of the difference coefficients between all the grid areas adjacent to the overlapping area OA is calculated as the violation index of the algorithm ALG.

[0118] S306, respectively calculating the violation index of each algorithm in the algorithm set, and adjusting the image of the overlapping area OA to the target meteorological image processed by the algorithm with the lowest violation index. Similarly, adjusting the image for each overlapping area on the target meteorological image TMI, and realizing super-resolution reconstruction of the entire image from the original meteorological image OMI to the target meteorological image TMI.

[0119] In S400, the target meteorological image after reconstruction is displayed on the visualization screen of the data center. A unique identification code is set for each grid area in the original meteorological image, and the meteorological characteristics and image processing time are analyzed to generate operation parameters. The region to which the meteorological image belongs, the operation parameters, the original meteorological image, and the target meteorological image are packaged together as a generation record and stored in the historical log.

[0120] See also Figure 2 The present invention provides a meteorological numerical pattern analysis system based on super-resolution, including a data acquisition module, a pattern analysis module, a numerical processing module and a visualization module.

[0121] The data acquisition module is used to collect algorithm libraries, historical logs, and original meteorological images of designated areas.

[0122] The pattern analysis module divides the original meteorological image into grids, calculates the difference coefficient between adjacent grids based on historical logs, adjusts the size of the grid according to the difference coefficient, and marks the overlapping area.

[0123] The numerical processing module calculates the timeliness coefficient of the grid area through historical logs, selects an algorithm in the algorithm library for each grid area according to the timeliness coefficient, and uses super-resolution technology to reconstruct the original meteorological image to generate the target meteorological image.

[0124] The visualization module displays the reconstructed target meteorological image through the visualization screen of the data center, creates a generation record based on the original meteorological image and the target meteorological image and stores it in the historical log.

[0125] The data acquisition module includes a meteorological image acquisition unit, a historical data acquisition unit and an algorithm library acquisition unit.

[0126] The meteorological image acquisition unit is used to collect original meteorological images of the specified area. Meteorological images show the distribution and changes of various meteorological elements through various contour lines and symbols. Meteorological elements refer to various physical quantities used to characterize meteorological conditions.

[0127] The historical data collection unit is used to collect the generation record of each meteorological image processing. Each generation record includes the region, operating parameters, original meteorological image and target meteorological image.

[0128] Operation parameters refer to the parameter information for image processing after the original meteorological image is divided into grid areas, including the number of grid areas, as well as the identification code, meteorological characteristics, area, algorithm used and processing time of each grid area.

[0129] The algorithm library acquisition unit is used to acquire software libraries containing different algorithms, and each algorithm realizes super-resolution processing of images through different image processing technologies.

[0130] The pattern analysis module includes a grid division unit and an associated expansion unit.

[0131] The grid division unit is used to divide the grid area, establish associated grid pairs and calculate the difference coefficient.

[0132] First, obtain the region S corresponding to the original meteorological image OMI, and mark the generation record of the region S in the historical log. After summing up the number of grids in the operating parameters in all marked generation records, calculate the average value to get the average number of grids W ave , evenly divide W in the original meteorological image ave grid areas of equal area.

[0133] Secondly, set the sampling number m, evenly set m sampling points in each grid area, and obtain the meteorological elements of each sampling point. Count the number of meteorological element types k, analyze the position distribution of all grid areas in the original meteorological image, and combine all two adjacent grid areas into associated grid pairs. Each grid area can be combined with b adjacent grid areas at other locations into b associated grid pairs.

[0134] Finally, the position coordinates (X, Y) of the sampling points in each grid area are analyzed, and the Euclidean distance between the sampling points in different grid areas under the associated grid pair is calculated. According to the principle of the shorter the Euclidean distance, the better the priority, the two sampling points belonging to different grid areas under the associated grid pair are sequentially associated. According to the meteorological elements corresponding to each sampling point, the difference coefficient of each associated grid pair is calculated.

[0135] The associative expansion unit is used to adjust the size of the grid according to the difference coefficient and to mark the overlapping area.

[0136] First, set the index threshold d and calculate the standard deviation of the difference coefficients of all associated grid pairs as the difference index. When the difference index is greater than d, calculate the average value CX of the difference coefficients of all associated grid pairs ave , the marker difference coefficient is greater than CX ave The associated grid pairs.

[0137] Secondly, count the number of occurrences g of the grid area GD1 in all the marked associated grid pairs, and the sum of the difference coefficients CX of the associated grid pairs to which the grid area GD1 belongs h , according to the formula g×CX h The anomaly index of grid area GD1 is calculated in . The anomaly index of each grid area is calculated separately, and the number of grid areas with non-zero anomaly indexes is counted, and e is divided by 2 and rounded down to get u.

[0138] Then, all grid areas are sorted from large to small according to the anomaly index, and the first u grid areas are selected as abnormal grids according to the sorting order. ab Other adjacent grid areas are GD ab According to the difference coefficient CX between the adjusted grid and the corresponding abnormal grid v , analyze the adjustment area of ​​each adjustment grid.

[0139] Finally, after controlling each adjustment grid to expand the area of ​​the same size as the corresponding adjustment area in the abnormal grid, calculate the difference index again. If the difference index is still greater than d, continue to expand the adjustment grid until the difference index is no greater than d. When the difference index is no greater than d, the grid adjustment is completed, and the overlapping areas between different grid areas are marked.

[0140] The numerical processing module includes a time-effect analysis unit and an image processing unit.

[0141] The time-effect analysis unit is used to calculate the time-effect coefficient of each grid area and match the algorithm.

[0142] First, a comparison set is established, and all meteorological features of each grid area in all the mark generation records are put into the comparison set. The meteorological feature FT2 of the grid area GD2 in the original meteorological image is extracted, and the similarity between each element in the comparison set and the meteorological feature FT2 is calculated.

[0143] Secondly, establish a coding set for grid area GD2 and set the similarity threshold SMI thr , get the similarity greater than SMI in the comparison set thrMeteorological features correspond to the identification codes of the grid areas, and all are put into the code set of the grid area GD2. Get the identification code BM in the code set P Corresponding grid area GD P , extract the grid area GD P The meteorological features in the original meteorological image and the target meteorological image are calculated, and the similarity SMI of the meteorological features between the two is calculated. p .

[0144] Finally, according to the identification code BM P Corresponding grid area MJ P and processing time P Calculate the timeliness coefficient, calculate the timeliness coefficient of each identification code in the code set, and classify all identification codes according to whether the adopted algorithm is the same. Calculate the average timeliness coefficient of all identification codes in the same category as the timeliness index of the corresponding class, mark the class with the largest timeliness index, and use the adopted algorithm of the marked class as the grid area GD P The selected algorithm of each grid area is analyzed in this way.

[0145] The image processing unit uses super-resolution technology to reconstruct the original meteorological image and generate the target meteorological image.

[0146] First, all grid areas under the original meteorological image OMI are grouped according to the selected algorithm. Each selected algorithm processes the images of all grid areas in the corresponding group. The original meteorological image OMI is super-resolution downsized using multi-algorithm multi-threaded parallel technology to generate the target meteorological image TMI. The position of each grid area in the original meteorological image OMI is mapped to the target meteorological image TMI.

[0147] Secondly, the area where the grid areas on the target meteorological image TMI overlap is marked, and an algorithm set is established for the overlapping area OA, and the selected algorithm of the grid area to which the overlapping area OA belongs is placed in the algorithm set. When analyzing the overlapping area OA using the algorithm ALG in the algorithm set, the sum of the difference coefficients between all grid areas adjacent to the overlapping area OA is calculated as the violation index of the algorithm ALG.

[0148] Finally, the violation index of each algorithm in the algorithm set is calculated respectively, and the image of the overlapping area OA is adjusted to the target meteorological image processed by the algorithm with the lowest violation index. Similarly, the image is adjusted for each overlapping area on the target meteorological image TMI to achieve super-resolution reconstruction of the entire image from the original meteorological image OMI to the target meteorological image TMI.

[0149] The visualization module displays the reconstructed target meteorological image through the visualization screen of the data center, and packages the region to which the meteorological image belongs, the operating parameters, the original meteorological image and the target meteorological image together as a generation record, which is stored in the historical log.

[0150] Embodiment 1:

[0151] Assume that there is an identification code BT165 in the code set, and the similarity of the meteorological characteristics of the corresponding grid area in the original meteorological image and the target meteorological image is 86%. The area of ​​the grid area corresponding to the identification code BT165 is 256×300px, and the processing time is 0.2S. When the constant α is 100000, substitute it into the formula to calculate the time efficiency coefficient of the identification code BT165:

[0152]

[0153] Then the timeliness coefficient of identification code BT165 is 3.3.

[0154] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0155] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A meteorological numerical model analysis method based on super-resolution, characterized by: The method comprises the following steps: S100, collecting algorithm libraries, historical logs and original meteorological images of a specified area, and analyzing meteorological elements contained in the original meteorological images; S200, dividing the original meteorological image into grids, calculating the difference coefficients between adjacent grids according to the historical logs, adjusting the size of the grids according to the difference coefficients and marking the overlapping areas; S300, calculating the timeliness coefficient of the grid area through historical logs, selecting an algorithm in the algorithm library for matching for each grid area according to the timeliness coefficient, and reconstructing the original meteorological image using super-resolution technology to generate a target meteorological image; S400: Display the reconstructed target meteorological image on the visualization screen of the data center, create a generation record based on the original meteorological image and the target meteorological image, and store it in a history log.

2. The super-resolution meteorological numerical model analysis method according to claim 1, characterized in that: In S100, the algorithm library refers to a software library containing different algorithms, and each algorithm uses different image processing technologies to achieve super-resolution processing of images; the historical log refers to the generation record of each meteorological image processing, and each generation record includes the region to which it belongs, operating parameters, original meteorological images and target meteorological images; operating parameters refer to parameter information for image processing after the original meteorological image is divided into grid areas, specifically including the number of grid areas, and the identification code, meteorological characteristics, area, algorithm used and processing time of each grid area; meteorological images use various contour lines and symbols to show the distribution and changes of various meteorological elements; meteorological elements refer to various physical quantities used to characterize meteorological conditions.

3. The super-resolution meteorological numerical model analysis method according to claim 2, characterized in that: In S200, the specific steps are as follows: S201, obtain the region S corresponding to the original meteorological image OMI, mark the generation record of the region S in the historical log; sum up the number of grids in the operating parameters in all marked generation records and calculate the average value to obtain the average number of grids W ave , evenly divide W in the original meteorological image ave grid areas of equal area; S202, set the sampling number m, evenly set m sampling points in each grid area, and obtain the meteorological elements of each sampling point; count the number of meteorological element types k, analyze the position distribution of all grid areas in the original meteorological image, and combine all two adjacent grid areas into associated grid pairs, and each grid area can be combined with b adjacent grid areas at other locations to form b associated grid pairs; S203, analyzing the position coordinates (X, Y) of the sampling points in each grid area, calculating the Euclidean distance between the sampling points in different grid areas under the associated grid pair; according to the principle of the shorter the Euclidean distance, the better the priority, sequentially associating two sampling points belonging to different grid areas under the associated grid pair; obtaining the meteorological elements corresponding to each sampling point, substituting them into the formula to calculate the difference coefficient CX of each associated grid pair: Where ρn is the Euclidean distance between the nth pair of associated sampling points under the associated grid pair, is the specific value of the i-th meteorological element corresponding to the n-th sampling point in one of the grid areas under the associated grid pair, is the specific value of the i-th meteorological element corresponding to the n-th sampling point in another grid area under the associated grid pair; S204, set the index threshold d, calculate the standard deviation of the difference coefficients of all associated grid pairs as the difference index; when the difference index is greater than d, calculate the average value CX of the difference coefficients of all associated grid pairs ave , the marker difference coefficient is greater than CX ave The number of occurrences g of the grid area GD1 in all the associated grid pairs of the tag is counted, as well as the sum of the difference coefficients CX of the associated grid pairs to which the grid area GD1 belongs h , substitute into the formula g×CX h The anomaly index of the grid area GD1 is calculated; S205, calculate the abnormal index of each grid area respectively, and count the number of grid areas e with abnormal index not equal to zero, divide e by 2 and round down to get u; sort all grid areas from large to small according to the abnormal index, select the first u grid areas as abnormal grids according to the sorting order, and compare them with the abnormal grid GD ab Other adjacent grid areas are GD ab Adjust the grid; set the basic area l, and obtain the difference coefficient CX between the adjusted grid and the corresponding abnormal grid v , substitute into the formula to get the adjustment area of ​​each adjustment grid: In the formula, area is the adjustment area, CX sum is the sum of the difference coefficients between the abnormal grid and all the adjusted grids; S206, after controlling each adjustment grid to expand an area of ​​the same size as the corresponding adjustment area in the abnormal grid, enter step S202 again to analyze the difference coefficient and calculate the difference index. When the difference index is greater than d, continue to expand the adjustment grid until the difference index is not greater than d; when the difference index is not greater than d, the grid adjustment is completed, and the overlapping areas between different grid areas are marked.

4. The super-resolution meteorological numerical model analysis method according to claim 3, characterized in that: In S300, the specific steps are as follows: S301, establish a comparison set, put all the meteorological features of each grid area in all the mark generation records into the comparison set; extract the meteorological feature FT2 of the grid area GD2 in the original meteorological image, calculate the similarity between each element in the comparison set and the meteorological feature FT2; establish a coding set for the grid area GD2 and set the similarity threshold SMI thr , get the similarity greater than SMI in the comparison set thr Meteorological characteristics correspond to the identification codes of the grid areas, and are all put into the code set of the grid area GD2; S302: Obtain identification code BM in the code set P Corresponding grid area GD P , extract the grid area GD P The meteorological features in the original meteorological image and the target meteorological image are calculated, and the similarity SMI of the meteorological features between the two is calculated. p ; Get identification code BM P Corresponding grid area MJ P and processing time P , substitute into the formula to calculate the identification code BM P The time efficiency coefficient: Where TZ P To identify the code BM P The time-effect coefficient, α is a constant; S303, respectively calculate the timeliness coefficient of each identification code in the code set, analyze the adopted algorithm corresponding to each identification code, and classify all identification codes according to whether the adopted algorithm is the same; respectively calculate the average timeliness coefficient of all identification codes in each category as the timeliness index of the corresponding category, mark the category with the largest timeliness index, and use the adopted algorithm of the marked category as the grid area GD P The selection algorithm of each grid area under the original meteorological image OMI is analyzed by analogy; S304, grouping all grid areas under the original meteorological image OMI according to the selected algorithm, each selected algorithm processes the images of all grid areas in the corresponding group, and adopts multi-algorithm multi-threaded parallel technology to perform super-resolution downsizing processing on the original meteorological image OMI, so as to generate a target meteorological image TMI; analyzing the position of each grid area in the original meteorological image OMI, and mapping it to the target meteorological image TMI; S305, marking the area where the grid areas on the target meteorological image TMI overlap, establishing an algorithm set for the overlapping area OA, and placing the selected algorithm of the grid area to which the overlapping area OA belongs into the algorithm set; when analyzing the overlapping area OA using the algorithm ALG in the algorithm set, calculating the sum of the difference coefficients between all the grid areas adjacent to the overlapping area OA as the violation index of the algorithm ALG; S306. Calculate the violation index of each algorithm in the algorithm set respectively, and adjust the image of the overlapping area OA to the target meteorological image processed by the algorithm with the lowest violation index; and so on, adjust the image for each overlapping area on the target meteorological image TMI to achieve super-resolution reconstruction of the entire image from the original meteorological image OMI to the target meteorological image TMI.

5. The super-resolution meteorological numerical model analysis method according to claim 4, characterized in that: In S400, the reconstructed target meteorological image is displayed on the visualization screen of the data center; a unique identification code is set for each grid area in the original meteorological image, and the meteorological characteristics and image processing time are analyzed to generate operating parameters; the region to which the meteorological image belongs, the operating parameters, the original meteorological image, and the target meteorological image are packaged together as a generation record and stored in the historical log.

6. The meteorological numerical model analysis system based on super-resolution is characterized by: The system includes a data acquisition module, a pattern analysis module, a numerical processing module and a visualization module; The data acquisition module is used to collect algorithm libraries, historical logs, and original meteorological images of designated areas; The pattern analysis module divides the original meteorological image into grids, calculates the difference coefficients between adjacent grids based on historical logs, adjusts the size of the grids based on the difference coefficients, and marks the overlapping areas; The numerical processing module calculates the timeliness coefficient of the grid area through historical logs, selects an algorithm from the algorithm library for each grid area according to the timeliness coefficient, and uses super-resolution technology to reconstruct the original meteorological image to generate the target meteorological image; The visualization module displays the reconstructed target meteorological image through the visualization screen of the data center, creates a generation record based on the original meteorological image and the target meteorological image and stores it in the historical log.

7. The meteorological numerical model analysis system based on super-resolution according to claim 6, characterized in that: The data acquisition module includes a meteorological image acquisition unit, a historical data acquisition unit and an algorithm library acquisition unit; The meteorological image acquisition unit is used to collect original meteorological images of a specified area; meteorological images display the distribution and changes of various meteorological elements through various contour lines and symbols. Meteorological elements refer to various physical quantities used to characterize meteorological conditions; The historical data collection unit is used to collect the generation record of each meteorological image processing; each generation record includes the region, operating parameters, original meteorological image and target meteorological image; Operation parameters refer to the parameter information of image processing after the original meteorological image is divided into grid areas, including the number of grid areas, as well as the identification code, meteorological characteristics, area, algorithm used and processing time of each grid area; The algorithm library acquisition unit is used to acquire software libraries containing different algorithms, and each algorithm realizes super-resolution processing of images through different image processing technologies.

8. The meteorological numerical model analysis system based on super-resolution according to claim 7, characterized in that: The mode analysis module includes a grid division unit and an associated expansion unit; The grid division unit is used to divide the grid area, establish associated grid pairs and calculate the difference coefficient; First, obtain the region S corresponding to the original meteorological image OMI, and mark the generation record of the region S in the historical log; sum up the number of grids in the operating parameters in all marked generation records, and calculate the average value to obtain the average number of grids W. ave , evenly divide W in the original meteorological image ave grid areas of equal area; Secondly, set the sampling number m, evenly set m sampling points in each grid area, and obtain the meteorological elements of each sampling point; count the number of meteorological element types k, analyze the position distribution of all grid areas in the original meteorological image, and combine all two adjacent grid areas into associated grid pairs. Each grid area can be combined with b adjacent grid areas at other locations to form b associated grid pairs; Finally, the position coordinates (X, Y) of the sampling points in each grid area are analyzed, and the Euclidean distances between the sampling points in different grid areas under the associated grid pair are calculated; according to the principle of the shorter the Euclidean distance, the better the priority, the two sampling points belonging to different grid areas under the associated grid pair are sequentially associated; according to the meteorological elements corresponding to each sampling point, the difference coefficient of each associated grid pair is calculated; The association expansion unit is used to adjust the size of the grid according to the difference coefficient and mark the overlapping area; First, set the index threshold d and calculate the standard deviation of the difference coefficients of all associated grid pairs as the difference index; when the difference index is greater than d, calculate the average value CX of the difference coefficients of all associated grid pairs ave , the marker difference coefficient is greater than CX ave The associated grid pairs of Secondly, count the number of occurrences g of the grid area GD1 in all the marked associated grid pairs, and the sum of the difference coefficients CX of the associated grid pairs to which the grid area GD1 belongs h , according to the formula g×CX h Calculate the anomaly index of the grid area GD1; Calculate the anomaly index of each grid area respectively, and count the number of grid areas e whose anomaly index is not zero, and divide e by 2 and round down to get u; Then, all grid areas are sorted from large to small according to the anomaly index, and the first u grid areas are selected as abnormal grids according to the sorting order. ab Other adjacent grid areas are GD ab The adjustment grid; according to the difference coefficient CX between the adjustment grid and the corresponding abnormal grid v , analyze the adjustment area of ​​each adjustment grid; Finally, after controlling each adjustment grid to expand the area of ​​the same size as the corresponding adjustment area in the abnormal grid, the difference index is calculated again. When the difference index is still greater than d, the adjustment grid is continued to be expanded until the difference index is no greater than d; when the difference index is no greater than d, the grid adjustment is completed, and the overlapping areas between different grid areas are marked.

9. The meteorological numerical model analysis system based on super-resolution according to claim 8, characterized in that: The numerical processing module includes a time-effect analysis unit and an image processing unit; The time-effect analysis unit is used to calculate the time-effect coefficient of each grid area and match the algorithm; First, a comparison set is established, and all the meteorological features of each grid area in all the mark generation records are put into the comparison set; the meteorological feature FT2 of the grid area GD2 in the original meteorological image is extracted, and the similarity between each element in the comparison set and the meteorological feature FT is calculated; Secondly, establish a coding set for grid area GD2 and set the similarity threshold SMI thr , get the similarity greater than SMI in the comparison set thr Meteorological characteristics correspond to the identification code of the grid area, and all are put into the code set of the grid area GD2; get the identification code BM in the code set P Corresponding grid area GD P , extract the grid area GD P The meteorological features in the original meteorological image and the target meteorological image are calculated, and the similarity SMI of the meteorological features between the two is calculated. p ; Finally, according to the identification code BM P Corresponding grid area MJ P and processing time P Calculate the timeliness coefficient, calculate the timeliness coefficient of each identification code in the code set, and classify all identification codes according to whether the adopted algorithm is the same; calculate the average timeliness coefficient of all identification codes in the same category as the timeliness index of the corresponding class, mark the class with the largest timeliness index, and use the adopted algorithm of the marked class as the grid area GD P The selected algorithm of each grid area is analyzed by analogy; The image processing unit uses super-resolution technology to reconstruct the original meteorological image and generate the target meteorological image; Firstly, all grid areas under the original meteorological image OMI are grouped according to the selected algorithm. Each selected algorithm processes the images of all grid areas in the corresponding group. The original meteorological image OMI is super-resolution downsized using multi-algorithm multi-threaded parallel technology to generate the target meteorological image TMI. Map the position of each grid area in the original meteorological image OMI to the target meteorological image TMI; Secondly, the overlapping areas of the grid areas on the target meteorological image TMI are marked, and an algorithm set is established for the overlapping area OA, and the selected algorithms of the grid areas to which the overlapping area OA belongs are placed in the algorithm set; when analyzing the overlapping area OA using the algorithm ALG in the algorithm set, the sum of the difference coefficients between all the grid areas adjacent to the overlapping area OA is calculated as the violation index of the algorithm ALG; Finally, the violation index of each algorithm in the algorithm set is calculated respectively, and the image of the overlapping area OA is adjusted to the target meteorological image processed by the algorithm with the lowest violation index; and so on, the image is adjusted for each overlapping area on the target meteorological image TMI to achieve super-resolution reconstruction of the entire image from the original meteorological image OMI to the target meteorological image TMI.

10. The meteorological numerical model analysis system based on super-resolution according to claim 9, characterized in that: The visualization module displays the reconstructed target meteorological image through the visualization screen of the data center, and packages the region to which the meteorological image belongs, the operating parameters, the original meteorological image and the target meteorological image together as a generation record, which is stored in the historical log.

Images