Multi-model integrated precipitation forecast method and system based on generalized triangular hat theory

Abstract

The present invention discloses a multi-mode integrated precipitation forecasting method and system based on the generalized triangular hat theory, which relates to the field of meteorological forecasting technology. The method obtains multi-source numerical model data in real time, eliminates systematic deviations through spatial standardization and real-time frequency matching preprocessing, and uses the generalized triangular hat theory to dynamically analyze the covariance matrix of the difference sequence between models to quantify uncertainty, and dynamically allocates weight coefficients based on uncertainty, and finally weighted integration outputs standardized precipitation forecast products. Without relying on historical or actual observation data, the TS score in rainstorm and heavy rainstorm forecasts can be improved by 49%-124% compared with international advanced models (such as ECMWF), and it shows high accuracy and strong adaptability in areas with scarce meteorological stations and transitional weather scenes such as typhoons and fronts. The integrated results are directly connected to the business platform of the National Meteorological Center to support real-time disaster warning decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological forecasting, and in particular to a multi-mode integrated precipitation forecasting method and system based on generalized tricorne theory. Background Art

[0002] As the core means of modern meteorological forecasting, numerical weather forecasting has been widely used in the prediction of various weather processes. At present, meteorological agencies at home and abroad generally adopt multi-model integrated forecasting methods to overcome the limitations of single model forecasting by integrating the forecast results of different numerical models (such as ECMWF and NCEP at the global scale, and GRAPES_3KM at the mesoscale). In the existing technology, the mainstream multi-model integration method usually performs weight allocation based on historical error evaluation, that is, by analyzing the forecast performance of each model in the past period of time (such as TS score, root mean square error, etc.), a higher weight is given to the model with smaller error, so as to optimize the integration result. However, the above method has the following significant problems in practical application: Insufficient dynamic adaptability: In the process of transitional precipitation weather (such as typhoon landing and frontal transit), the error characteristics of the model may mutate, and the static weight allocation based on historical data cannot timely reflect the particularity of the current weather process, resulting in a decrease in the accuracy of integrated forecasting. Data dependency limitation: In the early stage of precipitation, due to insufficient accumulation of historical data, it is difficult to accurately assess the model error, and the reliability of weight allocation is low; the weight coefficient is calibrated based on actual observation data, but in areas with sparse meteorological stations (such as plateaus and oceans), the lack of observation data causes the method to fail. Insufficient error separation: Existing methods do not effectively distinguish between systematic errors (such as inherent model bias) and random errors (such as differences in weather process sensitivity) in model forecasts, resulting in weight allocation that fails to accurately reflect the model's true forecasting capabilities.

[0003] For example, the integration method currently used by the China National Meteorological Center has achieved certain results in the integration of global models (ECMWF, NCEP) and regional models (such as GRAPES_3KM). However, during extreme weather events such as the heavy rain in North China in 2023, the TS score of the integrated forecast of heavy rain was only about 10% higher than that of the best single model due to the failure to dynamically adjust the weights, which failed to fully tap the potential of multi-model integration. In addition, existing studies have attempted to optimize weights through Bayesian model averaging (BMA) or machine learning, but they still rely on a large amount of historical observation data and have high computational complexity, making it difficult to meet the real-time requirements of operational forecasts. Summary of the invention

[0004] In view of this, the present invention proposes a multi-mode integrated precipitation forecasting method and system based on the generalized tricorne theory, which can dynamically quantify the uncertainty of multi-mode precipitation forecasting based on the generalized tricorne theory, realize real-time weight allocation without real-time observation, and improve the accuracy of short-term precipitation integrated forecasting. The present invention provides the following technical solutions:

[0005] A multi-model integrated precipitation forecasting method based on generalized tricorn hat theory, the method comprising:

[0006] Real-time access to short-term precipitation forecast data from multiple numerical weather forecast models;

[0007] Preprocessing the short-term precipitation forecast data to generate standardized precipitation data;

[0008] Based on the generalized tricornut theory, uncertainty characteristics analysis is performed on standardized precipitation data to obtain uncertainty analysis results for each numerical weather forecast model.

[0009] Assign weight coefficients to the corresponding numerical weather forecast models based on the uncertainty analysis results;

[0010] The standardized precipitation data is weighted and integrated based on the weight coefficients, and a multi-model integrated precipitation forecast product is output.

[0011] Optionally, the method of performing preprocessing on the short-term precipitation forecast data to generate standardized precipitation data includes:

[0012] Performing spatial consistency processing on the precipitation forecast data, specifically, using a bilinear interpolation method to uniformly process the plurality of precipitation forecast data into grid precipitation data of equal longitude and latitude covering the same area and spatial resolution;

[0013] The grid point precipitation data is subjected to a real-time frequency matching algorithm to eliminate systematic deviations, so as to output standardized precipitation data with random errors retained.

[0014] Optionally, the method of performing uncertainty characteristic analysis on standardized precipitation data based on the generalized tricorne theory to obtain uncertainty analysis results of each numerical weather forecast model includes:

[0015] Expanding the spatial fields of a plurality of said numerical weather prediction models into a one-dimensional sequence;

[0016] A numerical weather forecast model is randomly selected as the reference field;

[0017] Calculate the precipitation difference series between the remaining models and the reference field in the same spatial and temporal range to generate a difference matrix;

[0018] Calculating a covariance matrix of the precipitation difference sequence based on the difference matrix;

[0019] The random error terms corresponding to each mode are analyzed based on the covariance matrix to construct the uncertainty analysis results of each mode.

[0020] Optionally, the method of analyzing the random error terms corresponding to each mode based on the covariance matrix to construct the uncertainty analysis results of each mode includes:

[0021] A symmetric noise covariance matrix is ​​introduced to represent the random error characteristics of precipitation data in each model;

[0022] A system of equations is constructed through the mathematical relationship between the covariance matrix and the noise covariance matrix, and the errors of the precipitation data of each model are assumed to be independent of each other, so as to solve the noise covariance matrix by minimizing the free parameters method;

[0023] The noise covariance matrix is ​​analyzed to obtain the random error term corresponding to each mode as the uncertainty analysis result of each mode.

[0024] Optionally, the one-dimensional sequence is expressed as: , any one-dimensional sequence can be decomposed into: ,in, Represents the actual field of precipitation data at each time. represents the error term in any mode;

[0025] The calculation of the precipitation difference sequence is expressed as: , where i=1,2,…,N-1.

[0026] Optionally, the precipitation difference sequence is converted into a matrix representation:

[0027] , where M is the number of grid points of grid precipitation data of each model;

[0028] The covariance matrix of the precipitation difference sequence is:

[0029] , where cov() is the covariance operator;

[0030] The noise covariance matrix R is A symmetric matrix of the format:

[0031] , where the relationship between the covariance matrix R and the covariance matrix S of the precipitation difference sequence is: S=J·R·J T ,in:

[0032] ;

[0033] The system of equations is: , where i and j are the ordinal numbers of the numerical weather prediction models.

[0034] Optionally, the method for allocating weight coefficients to corresponding numerical weather forecast modes based on uncertainty analysis results includes:

[0035] Based on the principle that the larger the uncertainty, the smaller the weight coefficient, the distribution weight coefficient of each numerical weather forecast model is solved. The formula is:

[0036] , where r ii is the diagonal element in the noise covariance matrix R, specifically the random error covariance of the i-th numerical weather forecast model.

[0037] Optionally, the method of performing weighted integration on the standardized precipitation data based on the weight coefficient and outputting a multi-mode integrated precipitation forecast product includes:

[0038] Based on the weight coefficients, the multi-model integrated precipitation forecast product is output, and the integrated formula is:

[0039] .

[0040] The present invention further discloses a multi-mode integrated precipitation forecasting system based on the generalized triangular hat theory, comprising:

[0041] The data processing module is used to obtain short-term precipitation forecast data of multiple numerical weather forecast models in real time; it is also used to,

[0042] Preprocessing the short-term precipitation forecast data to generate standardized precipitation data;

[0043] The feature analysis module is used to perform uncertainty feature analysis on the standardized precipitation data based on the generalized tricornut theory to obtain the uncertainty analysis results of each numerical weather forecast model;

[0044] A weight calculation module, used to assign weight coefficients to corresponding numerical weather forecast modes based on uncertainty analysis results;

[0045] The forecast output module is used to perform weighted integration on the standardized precipitation data based on the weight coefficients and output a multi-mode integrated precipitation forecast product.

[0046] The present invention further discloses a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0047] According to the technical solution of the present invention, by acquiring multi-source numerical model data in real time, after preprocessing such as spatial standardization and real-time frequency matching to eliminate systematic deviations, the covariance matrix of the difference sequence between models is dynamically analyzed using the generalized triangular hat theory to quantify uncertainty, and weight coefficients are dynamically allocated based on uncertainty, and finally weighted integration is used to output standardized precipitation forecast products. Its technical effect is reflected in the following aspects: this method does not need to rely on historical or actual observation data when allocating weight coefficients, and can achieve the optimal integration of the results of multiple numerical forecast models, and achieves a 49%-124% improvement in TS scores compared with international advanced models (such as ECMWF) in rainstorm and heavy rainstorm forecasts, and shows high accuracy and strong adaptability in areas with scarce meteorological stations and in transitional weather scenes such as typhoons and fronts. The integrated results are directly connected to the business platform of the National Meteorological Center to support real-time disaster warning decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] For the purpose of illustration and not limitation, the present invention is now described in conjunction with the embodiments of the present invention and the accompanying drawings, in which:

[0049] Figure 1 is a flow chart of a multi-model integrated precipitation forecasting method based on the generalized triangular hat theory in an embodiment of the present invention;

[0050] Figure 2 is a schematic structural diagram of a multi-mode integrated precipitation forecast system based on the generalized triangular hat theory in an embodiment of the present invention;

[0051] Figure 3 It is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the implementation mode of the present application will be clearly and completely described below in conjunction with the drawings in the implementation mode of the present application. Obviously, the described implementation mode is only a part of the implementation mode of the present application, not all the implementation modes. Based on the implementation mode in the present application, all other implementation modes obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0053] It should be noted that, in the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0054] refer to Figure 1 This embodiment discloses a multi-mode integrated precipitation forecasting method based on the generalized triangular hat theory, the method comprising:

[0055] S100: Real-time acquisition of short-term precipitation forecast data of multiple numerical weather forecast models. In this embodiment, the latest model short-term precipitation forecast data is retrieved in real time from the big data cloud platform of the Meteorological Bureau. Exemplarily, the short term can be 0-84 hours. Among them, when searching, the search members include 3 global scales such as ECMWF, GERMAN, and NCEP, as well as 4 mesoscales such as GRAPES_3KM, BEIJING_MR, GUANGZHOU_MR, and SHANGHAI_MR, accumulating precipitation forecast products of 7 numerical weather forecast models. The multi-model integrated forecasting method of this embodiment is mainly used when the model members are greater than 3 or more.

[0056] This implementation gives an example of obtaining short-term precipitation forecasts of multiple numerical weather forecast modes: the weather event is that Typhoon No. 9 of xx landed in xx on xx, and precipitation forecasts for a specific area are required. Select the following multiple numerical weather forecast modes:

[0057] Global models: ECMWF (0.1°×0.1°), NCEP-GFS (0.25°×0.25°), GERMAN-ICON (0.125°×0.125°).

[0058] Mesoscale model: GRAPES_3KM (3km), BEIJING_MR (North China region, 3km), GUANGZHOU_MR (South China region, 3km).

[0059] Obtain precipitation forecast data for the next 24 hours from the above multiple models.

[0060] S200: After obtaining multiple short-term precipitation forecasts, preprocess the short-term precipitation forecast data to generate standardized precipitation data. Since multiple weather forecast data are obtained, there are significant differences in the spatial resolution and product projection methods of different numerical forecast model products. In order to minimize the impact of these differences, the bilinear interpolation method is used to uniformly process the multiple precipitation forecast data into grid precipitation data of equal longitude and latitude covering the same area and spatial resolution. For example, the obtained numerical model forecast products are uniformly processed into grid precipitation forecast products of equal longitude and latitude covering the same area and with a spatial resolution of 0.05°×0.05°. That is, the bilinear interpolation algorithm is used to unify heterogeneous data into 0.05°×0.05° grid data of equal longitude and latitude to achieve spatial consistency. The input data is the original precipitation forecast field of each model (for example: 0.1° grid of ECMWF, 3 km grid of GRAPES_3KM). Further grid mapping is performed on this, and the original data is projected onto the target equal longitude and latitude grid (such as 100°-130° east longitude, 15°-45° north latitude). The four original grid points around the target grid point are distance-weighted. For areas outside the model coverage, the data are filled in using the neighboring extrapolation method, and a grid field with a uniform resolution is output (for example: 0.05°×0.05°, covering the entire China).

[0061] After completing the spatial consistency processing, it is necessary to further eliminate the systematic deviations existing in each numerical model while retaining its random error characteristics. This technical solution adopts a real-time frequency matching algorithm to achieve deviation correction by dynamically comparing the statistical distribution of model forecasts and high-precision reference data. Specifically, the inherent error examples of each model are manifested as the overall offset of precipitation, or the difference in the sensitivity of the model to weather processes (such as convective triggering and terrain effects). Based on the spatially standardized grid precipitation field (0.05°×0.05°), the cumulative frequencies of model forecasts and reference data at different precipitation levels (such as 0.1mm, 10mm, and 50mm) are statistically analyzed. For example, the cumulative frequency of precipitation ≥50mm in the model in 24 hours is 15%, while the reference data is 10%, indicating that the model systematically overestimates the probability of heavy rain. For each precipitation magnitude interval, the frequency ratio of the model and the reference data is calculated, and a nonlinear correction curve is generated to dynamically adjust the model forecast value. For example, if the frequency of the model in the 50-100mm interval is 1.2 times that of the reference data, the forecast value of this interval is scaled to 1 / 1.2≈0.83 times the original value. Then, according to the current weather process, the preset correction parameters are selected. For example, typhoon precipitation is mainly convective, and the correction focuses on the deviation correction of short-term heavy precipitation (>20mm / h); the plum rain front precipitation focuses on the magnitude adjustment of continuous precipitation. Then, the precipitation of each grid point in the model is dynamically adjusted according to the correction function to achieve only the correction of the systematic deviation part (such as the overall offset) and retain the random error of the model to the weather (such as the difference in weather sensitivity). By performing preprocessing on the short-term precipitation forecast data, the current forecast is directly compared with the real-time reference data to avoid the lag of historical statistics. While eliminating the systematic deviation of the model, the random error reflecting its true forecasting ability is retained. It significantly improved the comparability of multi-modal data and laid a solid foundation for subsequent uncertainty analysis based on the generalized tricorne hat theory.

[0062] S300: Based on the generalized tricornut theory, uncertainty feature analysis is performed on the standardized precipitation data to obtain uncertainty analysis results for each numerical weather forecast model. Specifically, the spatial fields of the multiple numerical weather forecast models are expanded into a one-dimensional sequence. In this embodiment, the two-dimensional grid field of each model is expanded into a one-dimensional sequence in row priority order, which is expressed as , and any one-dimensional sequence can be decomposed into: ,in, Represents the actual field of precipitation data at each time. Represents the error term under any mode. Based on this, the calculation of the precipitation difference series is expressed as: , where i=1,2,...,N-1. Further, the precipitation difference sequence is converted into a matrix representation: , where M is the number of grid points of the grid precipitation data of each model. A numerical weather forecast model is randomly selected as the reference field, and the precipitation difference sequences of the remaining models and the reference field in the same time and space range are calculated to generate a difference matrix. The covariance matrix of the precipitation difference sequence is calculated based on the difference matrix. The covariance matrix of the precipitation difference sequence is: , where cov() is the covariance operator. Based on the covariance matrix, the random error term corresponding to each mode is parsed to construct the uncertainty analysis result of each mode.

[0063] Furthermore, a symmetrical noise covariance matrix is ​​introduced to represent the random error characteristics of precipitation data of each mode, and a set of equations is constructed through the mathematical relationship between the covariance matrix and the noise covariance matrix, and it is assumed that the errors of precipitation data of each mode are independent of each other, so as to solve the noise covariance matrix using the method of minimizing free parameters. The noise covariance matrix is ​​analyzed to obtain the random error terms corresponding to each mode as the uncertainty analysis results of each mode. Specifically, the noise covariance matrix R is A symmetric matrix of the format:

[0064] , where S = J·R·J T , which represents the relationship between R and S. J is a matrix representation: .

[0065] The above equations constructed by the mathematical relationship between the covariance matrix and the noise covariance matrix are: , where i and j are the ordinal numbers of the numerical weather prediction models.

[0066] S400: Based on the uncertainty analysis result, a weight coefficient is allocated to the corresponding numerical weather forecast mode. That is, after obtaining the equation group, the weight coefficient allocated to each numerical weather forecast mode is solved according to the principle that the larger the uncertainty, the smaller the weight coefficient. The formula is:

[0067] , where r ii is the diagonal element in the noise covariance matrix R, specifically the random error covariance of the ith numerical weather forecast model, so r ii= This means that the larger the weight, the smaller the random error of the model, the higher the forecast reliability, and the ii The smaller it is, the smaller the random deviation of the model from the actual precipitation field and the more stable the forecast.

[0068] This embodiment provides an exemplary calculation process:

[0069] Assuming that three modes are integrated, the diagonal elements of the noise covariance matrix R are: , , An example of how the weights are calculated is:

[0070] .

[0071] S500: Based on the weight coefficient, the standardized precipitation data is weighted and integrated, and a multi-mode integrated precipitation forecast product is output. Specifically, based on the weight coefficient, a multi-mode integrated precipitation forecast product is output, and the integration formula is: .

[0072] This embodiment provides an exemplary integrated forecast output process:

[0073] For each grid point (x, y) and forecast time t, calculate the weighted precipitation value: ,in, It is represented by the grid precipitation forecast value after integration, N represents the total number of models involved in the integration, represents the weight coefficient of the i-th mode, represents the standardized precipitation data of the ith mode.

[0074] The overall example is as follows:

[0075] Integrate forecast data from three models: ECMWF, NCEP, and GRAPES_3KM;

[0076] The obtained forecast data are standardized, as shown in Table 1:

[0077] Table 1

[0078]

[0079] Perform a weighted calculation:

[0080] Grid point 1 (116.0°E, 39.9°N):

[0081] Result=0.45×85+0.35×92+0.20×80=38.25+32.2+16.0=86.45mm,

[0082] Grid point 2 (116.1°E, 40.0°N):

[0083] Result=0.45×78+0.35×85+0.20×70=35.1+29.75+14.0=78.85mm.

[0084] In summary, this technical solution proposes a dynamic multi-model integrated precipitation forecasting method based on the generalized tricornut theory. By acquiring precipitation data from global models such as ECMWF and NCEP and regional models such as GRAPES_3KM in real time, bilinear interpolation is used to unify the data to a 0.05°×0.05° grid, and a real-time frequency matching algorithm is used to eliminate systematic deviations to generate standardized data. Subsequently, the covariance matrix of the inter-model difference sequence is analyzed based on the generalized tricornut theory to quantify the random error variance of each model. The weight coefficients are dynamically allocated according to the principle of “the smaller the error, the higher the weight”, and finally the weighted integration is used to output the forecast product in the MICAPS standard format. Without the support of historical real-time data, this method can improve the TS scores of 24-hour forecasts of heavy rain and torrential rain by 49% and 124% respectively compared with the ECMWF model. It is particularly suitable for areas with scarce meteorological stations and transitional weather processes such as typhoons and fronts. In actual application, its high precision and business practicality have been verified in major disastrous weather such as the heavy rain in North China in 2023. The integrated results are directly connected to the business platform of the National Meteorological Center to support real-time disaster warning and flood control decision-making.

[0085] refer to Figure 2 This embodiment further discloses a multi-mode integrated precipitation forecasting system based on the generalized tricorn hat theory, including:

[0086] The data processing module 21 is used to obtain short-term precipitation forecast data of multiple numerical weather forecast models in real time; and is also used to:

[0087] Preprocessing the short-term precipitation forecast data to generate standardized precipitation data includes: performing spatial consistency processing on the precipitation forecast data, specifically, using a bilinear interpolation method to uniformly process a plurality of the precipitation forecast data into grid precipitation data of equal longitude and latitude covering the same area and spatial resolution; using a real-time frequency matching algorithm to eliminate systematic deviations from the grid precipitation data to output standardized precipitation data with random errors retained;

[0088] The feature analysis module 22 is used to perform uncertainty feature analysis on the standardized precipitation data based on the generalized tricorn hat theory to obtain the uncertainty analysis results of each numerical weather forecast model, including: expanding the spatial fields of multiple numerical weather forecast models into a one-dimensional sequence; randomly selecting a numerical weather forecast model as a reference field; calculating the precipitation difference sequence of the remaining models and the reference field in the same time and space range to generate a difference matrix; calculating the covariance matrix of the precipitation difference sequence based on the difference matrix; parsing the random error terms corresponding to each model based on the covariance matrix to construct the uncertainty analysis results of each model, including: introducing a symmetrical noise covariance matrix to represent the random error characteristics of the precipitation data of each model; constructing a set of equations through the mathematical relationship between the covariance matrix and the noise covariance matrix, and assuming that the errors of the precipitation data of each model are independent of each other, so as to solve the noise covariance matrix using the method of minimizing free parameters; parsing the noise covariance matrix to obtain the random error terms corresponding to each model as the uncertainty analysis results of each model; the one-dimensional sequence is expressed as: , any one-dimensional sequence can be decomposed into: ,in, Represents the actual field of precipitation data at each time. represents the error term under any mode; the calculation of the precipitation difference series is expressed as: , where i=1,2,...,N-1; the precipitation difference sequence is converted into a matrix representation:

[0089] , where M is the number of grid points of grid precipitation data of each model;

[0090] The covariance matrix of the precipitation difference sequence is:

[0091] , where cov() is the covariance operator;

[0092] The noise covariance matrix R is A symmetric matrix of the format:

[0093] , where the relationship between the covariance matrix R and the covariance matrix S of the precipitation difference sequence is: S=J·R·J T ,in:

[0094] ;

[0095] The system of equations is: , where i and j are the ordinal numbers of the numerical weather prediction models;

[0096] The weight calculation module 23 is used to allocate weight coefficients to the corresponding numerical weather forecast modes based on the uncertainty analysis results, including: using the principle that the larger the uncertainty, the smaller the weight coefficient, to solve the allocation weight coefficient of each numerical weather forecast mode, the formula is:

[0097] , where r i j is the element in the noise covariance matrix R, specifically the random error covariance between the i-th numerical weather forecast model and the j-th numerical weather forecast model;

[0098] The forecast output module 24 is used to perform weighted integration on the standardized precipitation data based on the weight coefficients and output a multi-mode integrated precipitation forecast product.

[0099] Figure 3 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as Figure 3 As shown, the electronic device 50 includes: a processor 501 (processor), a memory 502 (memory) and a bus 503;

[0100] The processor 501 and the memory 502 communicate with each other via the bus 503 ; the processor 501 is used to call the program instructions in the memory 502 to execute the methods provided by the above-mentioned method implementation methods.

[0101] This embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions. The computer instructions enable a computer to execute the methods provided by the above-mentioned method embodiments.

[0102] A person skilled in the art can understand that all or part of the steps for implementing the above-mentioned method implementation method can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium, which, when executed, executes the steps of the above-mentioned method implementation method; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various storage media that can store program codes.

[0103] The device implementation described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present implementation scheme. Those of ordinary skill in the art may understand and implement it without creative effort.

[0104] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each implementation mode or some parts of the implementation mode.

[0105] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-model integrated precipitation forecasting method based on the generalized tricorn hat theory, characterized in that: The method comprises: Real-time access to short-term precipitation forecast data from multiple numerical weather forecast models; Preprocessing the short-term precipitation forecast data to generate standardized precipitation data, the preprocessing comprising: using a bilinear interpolation algorithm to unify heterogeneous data into 0.05°×0.05° latitude and longitude grid data to achieve spatial consistency, and output a grid field with uniform resolution; based on the spatially standardized grid field, respectively counting the cumulative frequencies of model forecasts and reference data at different precipitation levels, and generating a nonlinear correction curve to eliminate systematic deviations and retain random errors; Based on the generalized tricorne theory, the uncertainty characteristics of standardized precipitation data are analyzed to obtain the uncertainty analysis results of each numerical weather forecast model, including: expanding the spatial fields of multiple numerical weather forecast models into a one-dimensional sequence; randomly selecting a numerical weather forecast model as a reference field; calculating the precipitation difference sequence of the remaining models and the reference field in the same time and space range to generate a difference matrix; based on the difference matrix, the covariance matrix of the precipitation difference sequence is calculated, and the formula is: S=J·R·J T , where S is the covariance matrix of the difference matrix, J is the structure matrix, and J T represents the transpose of J, R is the noise covariance matrix; based on the covariance matrix, the random error term corresponding to each mode is parsed to obtain the uncertainty analysis result of each mode; Based on the uncertainty analysis results, a weight coefficient is assigned to the corresponding numerical weather forecast model, and the formula is: ,in, is the weight coefficient of the i-th mode, r ii is the i-th diagonal element in the noise covariance matrix R; Based on the weight coefficient, the standardized precipitation data is weighted integrated and a multi-model integrated precipitation forecast product is output. The integration formula is: ,in is the standardized precipitation data under the i-th mode.

2. The multi-mode integrated precipitation forecasting method according to claim 1, characterized in that: The method of performing preprocessing on short-term precipitation forecast data to generate standardized precipitation data includes: Performing spatial consistency processing on the precipitation forecast data, specifically, using a bilinear interpolation method to uniformly process the plurality of precipitation forecast data into grid precipitation data of equal longitude and latitude covering the same area and spatial resolution; The grid point precipitation data is subjected to a real-time frequency matching algorithm to eliminate systematic deviations, so as to output standardized precipitation data with random errors retained.

3. The multi-mode integrated precipitation forecasting method according to claim 1, characterized in that: The method of analyzing the random error terms corresponding to each mode based on the covariance matrix to construct the uncertainty analysis results of each mode includes: A symmetric noise covariance matrix is ​​introduced to represent the random error characteristics of precipitation data in each model; A system of equations is constructed through the mathematical relationship between the covariance matrix and the noise covariance matrix, and the errors of the precipitation data of each model are assumed to be independent of each other, so as to solve the noise covariance matrix by minimizing the free parameters method; The noise covariance matrix is ​​analyzed to obtain the random error term corresponding to each mode as the uncertainty analysis result of each mode.

4. The multi-mode integrated precipitation forecasting method according to claim 3, characterized in that: The one-dimensional sequence is expressed as: , any one-dimensional sequence can be decomposed into: ,in, Represents the actual field of precipitation data at each time point, represents the error term in any mode; The calculation of the precipitation difference sequence is expressed as: , where i=1,2,…,N-1.

5. The multi-mode integrated precipitation forecasting method according to claim 4, characterized in that: Convert the precipitation difference sequence into a matrix representation: , where M is the number of grid points of grid precipitation data of each model; The covariance matrix of the precipitation difference sequence is: , where cov() is the covariance operator; The noise covariance matrix R is a symmetric matrix in N*N format: , where the relationship between the covariance matrix R and the covariance matrix S of the precipitation difference sequence is: S=J·R·J T ,in: ； The system of equations is: , where i and j are the ordinal numbers of the numerical weather prediction models.

6. A multi-model integrated precipitation forecast system based on the generalized tricorn hat theory, characterized in that: include: A data processing module is used to obtain short-term precipitation forecast data of multiple numerical weather forecast models in real time; Also used for Preprocessing the short-term precipitation forecast data to generate standardized precipitation data, the preprocessing comprising: using a bilinear interpolation algorithm to unify heterogeneous data into 0.05°×0.05° latitude and longitude grid data to achieve spatial consistency, and output a grid field with uniform resolution; based on the spatially standardized grid field, respectively counting the cumulative frequencies of model forecasts and reference data at different precipitation levels, and generating a nonlinear correction curve to eliminate systematic deviations and retain random errors; The feature analysis module is used to perform uncertainty feature analysis on standardized precipitation data based on the generalized tricorne theory to obtain uncertainty analysis results for each numerical weather forecast model, including: expanding the spatial fields of multiple numerical weather forecast models into a one-dimensional sequence; randomly selecting a numerical weather forecast model as a reference field; calculating the precipitation difference sequence of the remaining models and the reference field in the same time and space range to generate a difference matrix; and calculating the covariance matrix of the precipitation difference sequence based on the difference matrix, the formula is: S=J·R·J T , where S is the covariance matrix of the difference matrix, J is the structure matrix, and J T represents the transpose of J, R is the noise covariance matrix; based on the covariance matrix, the random error term corresponding to each mode is parsed to obtain the uncertainty analysis result of each mode; The weight calculation module is used to assign weight coefficients to the corresponding numerical weather forecast modes based on the uncertainty analysis results. The formula is: ,in, is the weight coefficient of the i-th mode, r ii is the i-th diagonal element in the noise covariance matrix R; The forecast output module is used to perform weighted integration on the standardized precipitation data based on the weight coefficient and output a multi-mode integrated precipitation forecast product. The integration formula is: ,in is the standardized precipitation data under the i-th mode.

7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images