Intelligent correction method for heavy rainfall falling area based on mode set information

Abstract

The invention relates to the field of heavy rainfall forecasting, and provides an intelligent correction method for a heavy rainfall falling area based on pattern set information, and the method comprises the steps: carrying out the separation of an intensity ranking sequence and a relative percentile space matrix of heavy rainfall data through a preset ranking method, and obtaining the heavy rainfall intensity information and the heavy rainfall falling area; through the correction model, calculation is carried out based on the heavy rainfall falling area, and a corrected falling area result is obtained; according to the correction falling area result, the heavy rainfall intensity information and the correction falling area result are combined and restored, and a heavy rainfall forecast correction result is obtained; the rainfall intensity and the falling area are separated by adopting a preset ranking method, deviation correction is carried out on a forecast rainfall falling area correction problem which is neglected in the past through a deep learning method, and then reversible reduction is carried out on the corrected rainfall falling area information and an ensemble forecast rainfall intensity sequence. The rainfall intensity forecasting advantage of ensemble forecasting is reserved, the rainfall falling area deviation is reduced, and therefore more accurate rainfall forecasting is obtained, and the rainfall forecasting capacity can be enhanced.

Description

Technical Field

[0001] This invention relates to the field of heavy precipitation forecasting, and in particular to an intelligent correction method for heavy precipitation location based on model ensemble information. Background Technology

[0002] Heavy rainfall not only causes secondary disasters such as flash floods, mudslides, and landslides, but also results in agricultural losses. Therefore, accurately forecasting heavy rainfall has always been a crucial task in weather forecasting; however, current weather forecasting models still fall short of meeting societal needs. Although ensemble forecasting can provide precipitation ensemble and probabilistic forecast information by constructing forecasts from multiple ensemble members, extracting effective information from the massive amounts of precipitation ensemble forecasts remains the biggest challenge.

[0003] In recent years, deep learning models, through multi-layer nonlinear transformation structures, have been able to autonomously mine abstract features and potential patterns in data and build end-to-end complex mapping systems, and have been gradually applied to post-processing of weather model forecasts.

[0004] However, current weather forecasting post-processing techniques based on deep learning lack post-processing correction techniques for precipitation location based on precipitation ensemble forecasts, resulting in significant biases in precipitation location. Precipitation intensity and precipitation location are both crucial aspects of precipitation forecasting. While there are numerous correction methods for precipitation intensity, the lack of methods for correcting precipitation location limits the current post-correction forecasting capabilities of precipitation forecasts. Summary of the Invention

[0005] This invention provides an intelligent correction method for heavy precipitation areas based on pattern ensemble information, which addresses the shortcomings of existing precipitation area correction methods by integrating deep learning into the precipitation area forecast correction of ensemble forecasts, thereby improving the accuracy of heavy precipitation forecasts.

[0006] This invention provides a method for intelligent correction of heavy precipitation areas based on pattern set information, comprising the following steps.

[0007] By using a pre-defined ranking method, the intensity ranking sequence and relative percentile spatial matrix of heavy precipitation data are separated to obtain heavy precipitation intensity information and heavy precipitation area, respectively. Calculations were performed based on the areas of heavy precipitation to obtain corrected precipitation results; Based on the corrected precipitation location results, the heavy precipitation intensity information and the corrected precipitation location results are merged and restored to obtain the heavy precipitation forecast correction results.

[0008] This invention provides a method for intelligent correction of heavy precipitation areas based on pattern set information. By using a pre-defined ranking method, the intensity ranking sequence and relative percentile spatial matrix of heavy precipitation data are separated to obtain heavy precipitation intensity information and heavy precipitation area information, respectively. This also includes: By using a pre-defined statistical method, the intensity deviation of the original heavy precipitation forecast is corrected to obtain heavy precipitation data.

[0009] According to the present invention, a method for intelligent correction of heavy precipitation areas based on pattern set information is provided. This method separates the intensity ranking sequence and relative percentile space matrix of heavy precipitation data using a preset ranking method to obtain heavy precipitation intensity information and heavy precipitation areas, respectively. By using a pre-defined ranking method, the intensity and location of heavy precipitation data are ranked separately, resulting in an intensity ranking sequence and a relative percentile spatial matrix. The intensity ranking sequence and the relative percentile spatial matrix are separated to obtain the heavy precipitation intensity information and the heavy precipitation area, respectively.

[0010] According to the present invention, a method for intelligent correction of heavy precipitation area based on pattern set information is provided. This method ranks the intensity and area of ​​heavy precipitation data using a preset ranking method, obtaining an intensity ranking sequence and a relative percentile space matrix, including: By using a pre-defined ranking method, the intensity of heavy precipitation data is ranked to obtain an intensity ranking sequence; The areas affected by heavy rainfall are ranked using a pre-defined ranking method, resulting in sorted area data. If there are rankings with the same value in the sorted data, the ranking of the same value is the average of all rankings of that same value as the final ranking; wherein, the rankings of the same value are the same percentile. Based on the sorted data and the ranking of the same value, a relative percentile space matrix is ​​obtained.

[0011] According to the present invention, a method for intelligent correction of heavy precipitation areas based on pattern set information is provided. The heavy precipitation data includes: actual precipitation data and forecast precipitation data; the heavy precipitation intensity information includes: actual precipitation intensity sequence and forecast precipitation intensity sequence; the heavy precipitation area includes: actual precipitation area and forecast precipitation area; the method further includes separating the intensity ranking sequence and relative percentile space matrix of the heavy precipitation data using a preset ranking method to obtain the heavy precipitation intensity information and the heavy precipitation area, including: Separate the intensity ranking sequence and the relative percentile spatial matrix of the actual precipitation data to obtain the actual precipitation intensity sequence and the actual precipitation field. Based on the spatial relative size information of the actual precipitation field, the actual precipitation area is determined; Separate the intensity ranking sequence and relative percentile spatial matrix of the forecast precipitation data to obtain the forecast precipitation intensity sequence and forecast precipitation field; Based on the spatial relative size of the forecast precipitation field, the area of ​​forecast precipitation is determined.

[0012] According to the present invention, a method for intelligent correction of heavy precipitation areas based on model ensemble information is provided. The magnitude information of the ensemble member forecast fields includes: the maximum precipitation intensity of all ensemble members, the minimum precipitation intensity of all ensemble members, and the normalization coefficient of the maximum precipitation intensity of the current ensemble member. The step of calculating the corrected precipitation area based on the heavy precipitation area using a correction model includes: Based on heavy precipitation data, determine the maximum precipitation intensity and the minimum precipitation intensity of all ensemble members; In the calculation of each ensemble member, based on the maximum precipitation intensity of the current ensemble member, the normalization coefficient of the maximum precipitation intensity of the current ensemble member is obtained by normalizing the calculation using the minimum and maximum precipitation intensities of all ensemble members. The magnitude information of the forecast fields of the ensemble members is processed through a fully connected layer to obtain the first intermediate value; By using a splicing layer, the area of ​​heavy rainfall and the first intermediate value are spliced ​​together to obtain the second intermediate value; By using a self-attention mechanism and convolutional layers, local features are extracted and convolutional processing is performed on the second intermediate value to obtain the corrected landing area result.

[0013] The present invention also provides an intelligent correction device for heavy precipitation areas based on pattern set information, comprising the following modules: The separation module is used to separate the intensity ranking sequence and relative percentile spatial matrix of heavy precipitation data using a preset ranking method, so as to obtain the heavy precipitation intensity information and the heavy precipitation area respectively. The correction module is used to calculate the corrected precipitation area based on the heavy precipitation area using the correction model; The merging module is used to merge and restore the heavy precipitation intensity information and the corrected precipitation area results based on the corrected precipitation area results, so as to obtain the corrected heavy precipitation forecast results.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent correction method for heavy precipitation areas based on pattern set information as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent correction method for heavy precipitation areas based on pattern set information as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the intelligent correction method for heavy precipitation areas based on pattern set information as described above.

[0017] This invention provides an intelligent correction method for heavy precipitation areas based on ensemble pattern information. It separates the intensity ranking sequence and relative percentile spatial matrix of heavy precipitation data using a pre-defined ranking method, obtaining heavy precipitation intensity information and heavy precipitation areas respectively. A correction model is then used to calculate the corrected precipitation areas based on these areas. Finally, the heavy precipitation intensity information and the corrected precipitation areas are merged and restored to obtain the corrected heavy precipitation forecast. Compared to current weather forecasting-deep learning post-processing techniques, which do not include post-processing correction techniques for precipitation areas based on ensemble forecasts, this invention uses a pre-defined ranking method to separate precipitation intensity and area, correcting the previously neglected issue of forecast precipitation area correction through deep learning. The corrected precipitation area information is then reversibly restored from the ensemble forecast precipitation intensity sequence. This not only retains the advantages of ensemble forecast precipitation intensity prediction but also further reduces precipitation area bias, resulting in more accurate precipitation forecasts and enhancing precipitation forecasting capabilities. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the intelligent correction method for heavy precipitation areas based on pattern set information provided by the present invention.

[0020] Figure 2 This is an example provided by the present invention, showing the corrected and normalized result of the precipitation area for the next 24 hours, reported from 00:00 UTC on July 19, 2021.

[0021] Figure 3 This is an example provided by the present invention, showing the CRPS score of the landing area correction model for 24-72 hours in advance forecasts from June to August 2021.

[0022] Figure 4 This is an example provided by the present invention, showing the weather BS score 24-72 hours in advance from June to August 2021 and the weather BS score from 12:00 on July 19 to 00:00 on July 20.

[0023] Figure 5This is an example provided by the present invention, showing the correction of the landing area 24 hours in advance, the spatial distribution of the 75th percentile predicted by the original ensemble, and its actual distribution.

[0024] Figure 6 This is a schematic diagram of the intelligent correction device for heavy precipitation areas based on pattern set information provided by the present invention.

[0025] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The following is combined with Figures 1-7 This invention is described.

[0028] Figure 1 This is a flowchart illustrating the intelligent correction method for heavy precipitation areas based on pattern set information provided by the present invention. Figure 1 As shown, the method includes the following: Step 101: Using a preset ranking method, separate the intensity ranking sequence and relative percentile space matrix of heavy precipitation data to obtain heavy precipitation intensity information and heavy precipitation area.

[0029] In step 101 above, the intensity ranking sequence and relative percentile space matrix of heavy precipitation data are separated from large to small using a preset ranking method. The intensity ranking sequence is one-dimensional, while the relative percentile space matrix is ​​two-dimensional.

[0030] Integrating deep learning into the precipitation area forecast correction of ensemble forecasts creates a precipitation area correction method that combines ensemble forecasts and deep learning. This is beneficial for further improving the accuracy of heavy precipitation forecasts and enhancing the ability to predict and warn of disasters in advance.

[0031] In this invention, the area of ​​heavy precipitation corresponds to the percentile relationship of the original forecast field, and the corrected area result represents the corrected percentile relationship of the forecast field. By leveraging the nonlinear feature extraction capabilities of deep learning, this invention constructs an intelligent correction model for heavy precipitation area correction, effectively reducing the area forecast bias in precipitation ensemble forecasts and thus improving the accuracy of heavy precipitation forecasts. This technology integrates the advantages of deep learning methods and precipitation ensemble forecasts, improving the area bias problem in heavy precipitation weather events.

[0032] This invention first performs data preprocessing. Data preprocessing includes: acquiring forecast data: ensemble forecast systems from various operational centers, multi-source gridded precipitation analysis data, and station observation data to construct forecast and observation precipitation datasets; forecast datasets: precipitation data from model forecasts issued by various operational centers such as the European Centre for Medium-Range Weather Forecasts (ECMWF), the China Meteorological Administration (CMA), and the National Centers for Environmental Prediction (NCEP); observation datasets: data from multiple sources, such as observation data from actual stations and three-source fusion gridded precipitation data issued by the National Meteorological Information Center.

[0033] Optionally, step 101 includes steps B1 to B2: Step B1: Rank the intensity and location of heavy precipitation data using a preset ranking method to obtain the intensity ranking sequence and the relative percentile space matrix.

[0034] Step B2: Separate the intensity ranking sequence and the relative percentile space matrix to obtain the heavy precipitation intensity information and the heavy precipitation area, respectively.

[0035] In steps B1 to B2 above, in order to train the precipitation area correction model, it is first necessary to effectively separate the intensity information and precipitation area information of the actual precipitation and the precipitation ensemble forecast. By sorting algorithms, including sorting from smallest to largest or from largest to smallest, the actual precipitation intensity sequence and the precipitation ensemble forecast intensity sequence are obtained.

[0036] Regarding separation, for example, for a 3x3 matrix 2, 4, 5 / 3, 4, 6 / 6, 0, 2; The intensity information, ordered from smallest to largest, is 0, 2, 2, 3, 4, 4, 5, 6, 6. Based on the relative intensity information, it can be mapped to 1 / 9, 2 / 9, 3 / 9, 4 / 9, 5 / 9, 6 / 9, 7 / 9, 8 / 9, 9 / 9. The landing area information can then be transformed into a standardized matrix, represented by 9 quantiles. This standardized matrix, also known as the landing area matrix, can be considered as the percentiles of the original forecast field.

[0037] Optionally, step B1 includes steps B11 to B14: Step B11: Rank the intensity of heavy precipitation data using a preset ranking method to obtain an intensity ranking sequence.

[0038] Step B12: Rank the areas of heavy precipitation in the data using a preset ranking method; obtain the sorted area data.

[0039] Step B13: If there are rankings with the same value in the sorted data, the ranking of the same value is the average of all rankings of the same value as the final ranking; wherein, the rankings of the same value are the same percentile.

[0040] Step B14: Based on the sorted data and the ranking of the same value, obtain the relative percentile space matrix.

[0041] In steps B11 to B14 above, the sorted data is the ranking of heavy precipitation data, which is processed into a one-dimensional sequence to obtain heavy precipitation intensity information; at the same time, the relative percentiles in the form of a two-dimensional matrix are retained as the heavy precipitation areas; the preset ranking method for the heavy precipitation areas is a method for calculating the relative percentile relationship.

[0042] When calculating the area of ​​heavy rainfall, if there are rankings with the same value, the ranking of the same value is taken as the average of all rankings of the same value as the final ranking; where the rankings of the same value are the same percentile; the area of ​​heavy rainfall is obtained based on the sorted data and the rankings of the same value.

[0043] Optionally, the heavy precipitation data includes: actual precipitation data and forecast precipitation data; the heavy precipitation intensity information includes: actual precipitation intensity sequence and forecast precipitation intensity sequence; the heavy precipitation area includes: actual precipitation area and forecast precipitation area; Step 101 includes steps C1 to C4: Step C1: Separate the intensity ranking sequence and relative percentile spatial matrix of the actual precipitation data to obtain the actual precipitation intensity sequence and the actual precipitation field.

[0044] Step C2: Determine the precipitation area based on the spatial relative size information of the actual precipitation field.

[0045] Step C3: Separate the intensity ranking sequence and relative percentile spatial matrix of the forecast precipitation data to obtain the forecast precipitation intensity sequence and forecast precipitation field.

[0046] Step C4: Determine the predicted precipitation area based on the spatial relative size information of the predicted precipitation field.

[0047] Optionally, before step 101, step A1 is also included: Step A1: Correct the intensity deviation of the original heavy precipitation forecast using a preset statistical method to obtain heavy precipitation data.

[0048] In step A1 above, conventional precipitation intensity correction methods are used to correct the intensity of the precipitation ensemble forecast. This primarily involves correcting systematic biases in precipitation intensity using statistical methods, such as Frequency Matching Method (FMM) and Probability Matching Method (PMM). However, this invention mainly focuses on area correction; intensity correction is merely a preliminary step. The intensity correction is simply performed member-by-member. After member-by-member intensity correction, the probabilistic distribution characteristics of the ensemble forecast, which originally had probabilistic significance, are not practically affected in application.

[0049] Step 102: Using the correction model, calculate based on the heavy precipitation area to obtain the corrected precipitation area result.

[0050] In step 102 above, to construct an algorithm for adjusting precipitation areas using ensemble forecast areas as input, correction, and real-world precipitation areas as output during the training phase, an intensity sequence is calculated for each individual member, resulting in five independent sequences for five members. The representative precipitation ensemble forecast intensity serves as the intensity parameter for each ensemble member, and together with the precipitation forecast area distribution, a deep learning correction framework is constructed. This framework allows for the inference of corrected precipitation areas. A correction framework is built using a trainable neural network, incorporating a deep learning model with self-attention mechanisms and convolutional layers. After training, an end-to-end correction process is established between the original and corrected precipitation areas.

[0051] Optionally, the magnitude information of the ensemble member forecast fields includes: the maximum precipitation intensity of all ensemble members, the minimum precipitation intensity of all ensemble members, and the normalization coefficient of the maximum precipitation intensity of the current ensemble forecast member; Step 102 includes steps B1 to B5: Step B1: Based on the heavy precipitation data, determine the maximum precipitation intensity and the minimum precipitation intensity of all ensemble members.

[0052] Step B2: In the calculation of each ensemble member, based on the maximum precipitation intensity of the current ensemble member, the normalization coefficient of the maximum precipitation intensity of the current ensemble member is obtained by normalizing the calculation using the minimum and maximum precipitation intensities of all ensemble members.

[0053] Step B3: Process the magnitude information of the forecast field of the ensemble members through a fully connected layer to obtain the first intermediate value.

[0054] Step B4: By using a splicing layer, the area of ​​heavy rainfall and the first intermediate value are spliced ​​together to obtain the second intermediate value.

[0055] Step B5: Using a self-attention mechanism and convolutional layers, local features are acquired and convolutional processing is performed on the second intermediate value to obtain the corrected precipitation area result. Step 103: Based on the corrected precipitation area result, the heavy precipitation intensity information and the corrected precipitation area result are merged and restored to obtain the heavy precipitation forecast correction result.

[0056] In step 103 above, precipitation correction and reconstruction based on forecast intensity and corrected precipitation area: by using the separated precipitation intensity sequence and the corrected precipitation area result, inverse reconstruction is performed according to the percentile relationship of the original forecast field, and the precipitation intensity sequence is placed into the corresponding position according to the ranking of the precipitation area to restore the forecast of each set member.

[0057] The corrected precipitation distribution results here refer to the end-to-end output of the neural network. The resulting distribution matrix after separation has a spatial relative order. Similarly, the end-to-end output of the neural network in this application is also a 3x3 matrix similar to the example above, containing spatial relative size information. The inverse restoration process refers to reversing the principle of the separation process. The separated intensity sequence and the corrected distribution from the end-to-end output of the neural network are sequentially placed into the corrected distribution matrix, forming a 3x3 precipitation matrix.

[0058] To further explain the present invention, the following specific examples are provided.

[0059] Example 1. This example includes four steps: data preprocessing, precipitation intensity area separation, construction of precipitation area correction model, and precipitation correction reconstruction based on forecast intensity and correction area.

[0060] Data preprocessing: 24-hour cumulative precipitation forecasts from the European Centre for Medium-Range Weather Forecasts' Global Ensemble Prediction System (ECMWF-GEPS) for June-August of four years from 2019 to 2022 were obtained; three-source fusion precipitation analysis data were obtained from the China Meteorological Administration's Multi-Source Precipitation Analysis System (CMPAS); June-August 2021 was selected as validation data, and the other data were used as training datasets, totaling 540 samples.

[0061] Precipitation intensity area separation: The area information of actual precipitation and precipitation ensemble forecast is used to calculate the area data by ranking in ascending order. The calculation process of the area is as follows: Given an array X Sort the data in ascending order to obtain the landing area information matrix X. R in, The element in row i and column j Ranking; Let be the precipitation at the location (i, j) with longitude i and latitude j; At position (i, j) The relative percentile; n is the matrix dimension product, i.e., q × s for a given example, used to convert sort size to relative percentile; [c1] When a given array X contains multiple identical values ​​(e.g.) (using the average of multiple rankings) As a ranking, when multiple intensity values ​​are the same, they should be consistent, so grid points with the same intensity should also have a consistent rank.

[0062] in, Let m be the smallest sequence position among multiple identical values ​​(let's assume it's p) in the same sorting; m is the next sequence position in the sequence that does not satisfy (i.e., is greater than or less than p), meaning there are m-1 identical numbers in array X that are equal to p. [2][c3] Construction of the precipitation area correction model: The model extracts the forecast intensity information of different members from different start dates in the training dataset to characterize the ensemble forecast precipitation intensity and the current member's precipitation forecast intensity parameters. This, along with the precipitation area distribution, is used to construct a deep learning correction framework. The corrected precipitation area results can then be inferred through this framework. The input data for the forecast intensity information (magnitude) includes, for example, the overall ensemble forecast intensity (e.g., the maximum value of each member) and the maximum value of the current input member, thereby introducing a characteristic description of the ensemble forecast.

[0063] By separating precipitation intensity and precipitation location order, the ensemble forecast precipitation information is separated. Separating the forecast precipitation locations effectively increases the forecasting capability of precipitation relative to its location. Figure 2 As shown, Figure 2 This is the corrected and normalized result of the precipitation forecast for the next 24 hours, starting from 00:00 UTC on July 19, 2021. Here, ac and df represent the corrected and original forecast control forecast members and the first two ensemble members, respectively, and g is the observed normalized percentile for the precipitation area. Ensemble member 1 ( Figure 2 b and Figure 2 Compared to e), the percentile for the northwestern part of the selected region decreased after settlement correction, compared to Figure 2 The percentile distribution shown by the g-percentile of actual precipitation is more consistent.

[0064] Example 2: In this embodiment, the precipitation correction method provided by the present invention was used to correct the precipitation forecast for June-August 2021. Figure 3The RPS scores for various forecast (correction) methods are presented for forecasts 24-72 hours in advance. (By...) Figure 3 It is evident that the CRPS scores decreased in the 24-72 hour lead time, with an average decrease of approximately 0.2, indicating that the ensemble forecast precipitation accuracy can be improved after the location correction. Figure 4 This displays the overall weather forecast (BS) scores for June-August 2021, before and after the weather area revision. For example... Figure 4 As shown in a, after the location correction, the BS score for sunny / rainy weather decreased by as much as 53%, improving the weather forecast accuracy. Figure 4 The 'b' in the figure represents the BS rating for weather conditions from 12:00 on July 19th to 00:00 on July 20th. Figure 5 This shows the 75th percentile precipitation distribution of each forecast (corrected) 24 hours in advance. It includes the 75th percentile spatial distribution of the 24-hour advance precipitation corrections (b and e) and the actual distribution (a and d) of the original ensemble forecasts (c and f). Ac and df are from the 12:00 UTC forecast on July 21, 2021 and 00:00 UTC forecast on July 20, 2021, respectively. Looking at the 24-hour precipitation forecast from 12:00 UTC on July 21 (…), Figure 5 After the localization correction, the area of ​​heavy precipitation shifted southward, which is more consistent with the actual situation and increases the forecasting capability for heavy precipitation. In the 24-hour precipitation forecast starting at 00:00 UTC on July 20, 2021, the overall precipitation center shifted southeastward after the correction, and the corrected localization shift is more consistent with the actual situation. This indicates that the localization correction can have a certain indicative significance for the overall shift of precipitation and can effectively correct the bias in the localization of ensemble forecast precipitation.

[0065] This invention provides an intelligent correction method for heavy precipitation areas based on ensemble pattern information. It separates the intensity ranking sequence and relative percentile spatial matrix of heavy precipitation data using a pre-defined ranking method, obtaining heavy precipitation intensity information and heavy precipitation areas respectively. A correction model is then used to calculate the corrected precipitation areas based on these areas. Finally, the heavy precipitation intensity information and the corrected precipitation areas are merged and restored to obtain the corrected heavy precipitation forecast. Compared to current weather forecasting-deep learning post-processing techniques, which do not include post-processing correction techniques for precipitation areas based on ensemble forecasts, this invention uses a pre-defined ranking method to separate precipitation intensity and area, correcting the previously overlooked issue of forecast precipitation area correction through deep learning. The corrected precipitation area information is then reversibly restored from the ensemble forecast precipitation intensity sequence. This not only retains the advantages of ensemble forecast precipitation intensity prediction but also further reduces precipitation area bias, resulting in more accurate precipitation forecasts and enhancing precipitation forecasting capabilities.

[0066] The following describes the intelligent correction device for heavy precipitation areas based on pattern set information provided by the present invention. The intelligent correction device for heavy precipitation areas based on pattern set information described below can be referred to in correspondence with the intelligent correction method for heavy precipitation areas based on pattern set information described above.

[0067] Figure 6 This is a flowchart illustrating the intelligent correction device for heavy precipitation areas based on pattern set information provided by the present invention. Figure 6 As shown, the device includes the following: The separation module 601 is used to separate the intensity ranking sequence and the relative percentile space matrix of heavy precipitation data by using a preset ranking method, so as to obtain the heavy precipitation intensity information and the heavy precipitation area respectively.

[0068] The correction module 602 is used to calculate the corrected precipitation area based on the heavy precipitation area using the correction model.

[0069] The merging module 603 is used to merge and restore the heavy precipitation intensity information and the corrected precipitation area results based on the corrected precipitation area results, so as to obtain the heavy precipitation forecast correction results.

[0070] This invention provides an intelligent correction device for heavy precipitation areas based on ensemble pattern information. It separates the intensity ranking sequence and relative percentile spatial matrix of heavy precipitation data using a preset ranking method, obtaining heavy precipitation intensity information and heavy precipitation areas respectively. A correction model is then used to calculate the corrected precipitation areas based on these areas. Finally, the heavy precipitation intensity information and the corrected precipitation areas are merged and restored to obtain the corrected heavy precipitation forecast. Compared to current weather forecasting-deep learning post-processing techniques, which do not include post-processing correction techniques for precipitation areas based on ensemble forecasts, this invention uses a preset ranking method to separate precipitation intensity and area, correcting the previously neglected issue of forecast precipitation area correction through deep learning. The corrected precipitation area information is then reversibly restored from the ensemble forecast precipitation intensity sequence. This not only retains the advantages of ensemble forecast precipitation intensity prediction but also further reduces precipitation area bias, resulting in more accurate precipitation forecasts and enhancing precipitation forecasting capabilities.

[0071] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840. The processor 810, communications interface 820, and memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions from the memory 830 to execute a smart correction method for heavy precipitation areas based on pattern set information.

[0072] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0073] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the intelligent correction method for heavy precipitation areas based on pattern set information provided by the above methods.

[0074] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the intelligent correction method for heavy precipitation areas based on pattern set information provided by the above methods.

[0075] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0076] 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., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0077] 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 for intelligent correction of heavy precipitation area based on mode set information, characterized in that, The method comprises the following steps: The intensity ranking sequence and the relative percentage space matrix of the heavy rainfall data are separated by a preset ranking method, and the heavy rainfall intensity information and the heavy rainfall area are obtained respectively; The revised area result is obtained by calculating based on the heavy rainfall area through a revision model; The heavy rainfall intensity information and the revised area result are combined and restored according to the revised area result, and the heavy rainfall forecast revision result is obtained. 2.The method of claim 1, wherein, Before the intensity ranking sequence and the relative percentage space matrix of the heavy rainfall data are separated by a preset ranking method, the heavy rainfall data is obtained by intensity bias revision of the original heavy rainfall forecast through a preset statistical method. The intensity ranking sequence and the relative percentage space matrix of the heavy rainfall data are separated by a preset ranking method, and the heavy rainfall intensity information and the heavy rainfall area are obtained respectively, which comprises the following steps: 3.The method of claim 1, wherein, The intensity and the area in the heavy rainfall data are ranked respectively by a preset ranking method, and the intensity ranking sequence and the relative percentage space matrix are obtained; The intensity ranking sequence and the relative percentage space matrix are separated, and the heavy rainfall intensity information and the heavy rainfall area are obtained respectively. The intensity and the area in the heavy rainfall data are ranked respectively by a preset ranking method, and the intensity ranking sequence and the relative percentage space matrix are obtained, which comprises the following steps:

4. The method of claim 3, wherein the method is characterized by, The intensity in the heavy rainfall data is ranked by a preset ranking method, and the intensity ranking sequence is obtained; The area in the heavy rainfall data is ranked by a preset ranking method, and the ranked area data is obtained; In the case that there is the same value ranking in the ranked area data, the average value of all rankings of the same value is taken as the final ranking; wherein the same value ranking is the same percentage; The relative percentage space matrix is obtained according to the ranked area data and the ranking of the same value. The heavy rainfall data comprises the real-time rainfall data and the forecast rainfall data; the heavy rainfall intensity information comprises the real-time rainfall intensity sequence and the forecast rainfall intensity sequence; the heavy rainfall area comprises the real-time rainfall area and the forecast rainfall area; the intensity ranking sequence and the relative percentage space matrix of the heavy rainfall data are separated by a preset ranking method, and the heavy rainfall intensity information and the heavy rainfall area are obtained, which comprises the following steps:

5. The method of claim 1, wherein the method is characterized by, The intensity ranking sequence and the relative percentage space matrix of the real-time rainfall data are separated, and the real-time rainfall intensity sequence and the real-time rainfall field are obtained; The real-time rainfall area is determined according to the spatial relative size information of the real-time rainfall field; The intensity ranking sequence and the relative percentage space matrix of the forecast rainfall data are separated, and the forecast rainfall intensity sequence and the forecast rainfall field are obtained; The forecast rainfall area is determined according to the spatial relative size information of the forecast rainfall field. The magnitude information of the ensemble member forecast field comprises the maximum rainfall intensity of all ensemble members, the minimum rainfall intensity of all ensemble members and the normalized coefficient of the maximum rainfall intensity of the current ensemble forecast member; the revised area result is obtained by calculating based on the heavy rainfall area through a revision model, which comprises the following steps:

6. The method of claim 1, wherein the method is characterized by, ​ According to the heavy precipitation data, determine the maximum precipitation intensity of all ensemble members, the minimum precipitation intensity of all ensemble members; In the calculation of each ensemble member, based on the maximum precipitation intensity of the current ensemble member, the minimum precipitation intensity and the maximum precipitation intensity of all ensemble members are normalized to obtain the normalized coefficient of the maximum precipitation intensity of the current ensemble prediction member; Through the full connection layer, the magnitude information of the ensemble member prediction field is processed to obtain a first intermediate value; Through the splicing layer, the heavy precipitation area and the first intermediate value are spliced to obtain a second intermediate value; Through the self-attention mechanism and the convolution layer, the second intermediate value is locally characterized and convoluted to obtain the revised area result.

7. A device for intelligent correction of heavy precipitation area based on mode set information, characterized in that, It comprises: a separation module for separating the intensity ranking sequence and the relative percentage space matrix of the heavy precipitation data by a preset ranking method to obtain the heavy precipitation intensity information and the heavy precipitation area respectively; a revision module for calculating the revised area result based on the heavy precipitation area through a revision model; a merging module for merging the heavy precipitation intensity information and the revised area result according to the revised area result to obtain the heavy precipitation forecast revision result.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor executes the computer program to realize the heavy precipitation area intelligent revision method based on the mode set information according to any one of claims 1 to 6. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the heavy precipitation area intelligent revision method based on the mode set information according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to realize the heavy precipitation area intelligent revision method based on the mode set information according to any one of claims 1 to 6.

Images