Physical constraint minute hectometer gust fusion nowcasting method

Abstract

The invention discloses a physical constraint minute hectometer gust fusion nowcasting method. The method comprises the following steps: establishing a forecasting grid; establishing a radar combined reflectivity data set, a ground instantaneous wind speed data set and a lightning density data set; constructing a model framework for convective gust nowcasting, performing model training, and generating a convective gust optimal nowcasting model in combination with the gust coefficient; a future convective gust nowcasting result is obtained; adopting a multi-grid variation assimilation method to fuse the numerical model gust forecast result and the convective gust nowcasting result to generate a physically constrained fused gust forecast result; and generating a meteorological initial value, and obtaining a corrected fusion forecast wind field by using a wind field rapid diagnosis model. According to the method, a hectometer-level fusion gust nowcasting technology integrating a high-resolution mode and deep learning nowcasting is adopted, the mixed gust forecasting capability is enhanced through mode physical constraint, and the spatial fineness level of a forecasting result is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of meteorological forecast and warning, and in particular relates to a physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method. Background Art

[0002] Severe convective gale refers to weather phenomena with gusts reaching 17.2m / s and accompanied by thunder and lightning. Sometimes the wind speed reaches 24.5m / s or above, forming strong gusts of 100 meters per minute. Since the gusty weather caused by severe convective weather has nonlinear characteristics such as localization, suddenness, and concurrency, it has always been a difficult and important point in meteorological business to carry out detailed and accurate forecasts and warnings.

[0003] The high-temporal-resolution radar echoes obtained by weather radar observations characterize the precipitation particles inside the meteorological target and can be used to measure the intensity and distribution of weather phenomena. Currently, the 0-2 hour nowcast based on high-temporal-resolution weather radar data is the main way to carry out severe convective gust warnings. However, traditional 0-2 hour nowcasting methods are based on the assumption of linear evolution of radar echoes, which can only estimate the moving direction of convective storms and cannot describe the nonlinear evolution processes such as the initiation, enhancement, and organization of convection. Moreover, it is difficult to directly carry out gust element forecasts using only radar echoes.

[0004] With the rapid development of artificial intelligence technology in recent years, its subset deep learning technology has gradually played a role in the field of severe convective nowcasting due to its powerful ability to handle nonlinear problems, and has the potential to predict the evolution of convective storm intensity. In addition, the input end of the deep learning model architecture can simultaneously introduce multi-source forecasting factors such as radar echoes and ground wind speed through multi-channel superposition, which makes it possible to achieve direct forecasting of gust elements.

[0005] Although deep learning-based nowcasting of severe convective gusts can be carried out based on radar echoes and ground automatic station observations, there are still two limitations. First, under the constraint of strong echoes, the model can only forecast convective gusts, and its forecasting ability for mixed strong winds mixed with systematic strong winds is limited; second, for strong gusts of 24.5 m / s and above, it is difficult to extract their complex spatiotemporal characteristics through deep learning due to sample size limitations. Summary of the invention

[0006] The present invention aims to provide a physically constrained minute-by-minute 100-meter-level disaster-causing gust fusion nowcasting method to solve the problem that the current severe convective gust nowcasting based on weather radar cannot effectively forecast mixed gales and strong gusts above 24.5m / s.

[0007] In order to achieve the above-mentioned object, the present invention provides a physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method, comprising:

[0008] S1. Determine the spatial range and horizontal resolution of the forecast method and establish the forecast grid;

[0009] S2, based on the basic reflectivity factor data extracted by the weather radar, the instantaneous wind speed data extracted by the ground station, and the lightning observation frequency data extracted by the lightning locator in the forecast area, after preprocessing, align them to the forecast grid, and establish the radar combined reflectivity data set, the ground instantaneous wind speed data set, and the lightning density data set;

[0010] S3, constructing a model framework for convective gale nowcasting, and performing model training with each data set established in step S2, and generating an optimal convective gust nowcasting model in combination with the gust coefficient;

[0011] S4, inputting the real-time radar combined reflectivity data, the ground instantaneous wind speed data and the lightning density data into the convective gust optimal nowcasting model to obtain the future convective gust nowcasting result;

[0012] S5. Obtain the numerical model forecast data for the corresponding period within the spatial range of the forecast method, and generate the numerical model gust forecast results in combination with the gust coefficient; adopt the multi-grid variational assimilation method to fuse the numerical model gust forecast results and the convective gust nowcast results to generate the physically constrained fusion gust forecast results;

[0013] S6. Based on the fused gust forecast results, the meteorological initial values ​​are generated according to the terrain data within the spatial range of the forecast method and combined with the WRF parameterization scheme. Then, the wind field rapid diagnosis model is used to obtain the revised fused forecast wind field after adjustment based on the mass conservation law constraints.

[0014] A further preferred technical solution of the present invention is that step S2 comprises:

[0015] (1) Extract the basic reflectivity factor data within the spatial range of the forecast method from the weather radar, perform quality control, and interpolate to the equidistant forecast grid through three-dimensional meshing. The radar combined reflectivity data set is obtained by taking the maximum value in the vertical direction.

[0016] (2) Extract instantaneous wind speed data from ground automatic stations within the spatial range of the forecast method, interpolate to equidistant forecast grids using the inverse distance weighting method, and establish a ground instantaneous wind speed dataset;

[0017] (3) The lightning observation frequency data are extracted from the lightning locators within the spatial range of the forecast method, interpolated to the equidistant forecast grid using the nearest neighbor method, and combined with the spatial neighborhood method to establish a lightning density dataset.

[0018] Preferably, the model framework for convective gale nowcasting constructed in step S3 adopts the PhyDNet spatiotemporal convolutional neural network that introduces an attention mechanism.

[0019] Preferably, when performing model training with each data set in step S3, each data set is divided into time series, and in a data sequence of M consecutive time periods, the first M / 2 frames are used as input of the model, and the last M / 2 frames are used as labels;

[0020] In step S4, the real-time radar combined reflectivity data, ground instantaneous wind speed data and lightning density data are input into the convective gust optimal nowcasting model to obtain the convective gust sequence data of M / 2 consecutive time periods in the future, and the convective gust nowcasting result is obtained based on this.

[0021] Preferably, when performing model training in step S3, the designed loss function is a combined loss function, which is expressed as:

[0022] ;

[0023] in, represents the weight-balanced L1 loss function, represents the balanced loss function of strong wind and few sample data, represents the lightning probability loss function, represents the frequency domain adversarial loss function;

[0024] Weight-balanced joint L1 loss function , expressed as:

[0025] ;

[0026] in, and denote the observed field and model forecast field of convective gusts, respectively. and They represent the observation field and forecast field of radar combined reflectivity respectively, N represents the total number of samples of all grid points at all forecast times, and Respectively represent the weighted functions of the two variables, and automatically assign weights by counting the number of variables falling into different threshold intervals;

[0027] Balanced loss function for strong wind and few sample data , expressed as:

[0028] ;

[0029] in, For control The balance weight of , a and c are the contraction control parameters;

[0030] Lightning Occurrence Probability Loss Function , expressed as:

[0031] ;

[0032] in, is the binary cross entropy loss function, expressed as:

[0033] ;

[0034] in, , The observation field and model prediction field represent whether lightning occurs or not;

[0035] Based on the Dice coefficient, it is a measure of the similarity between lightning forecast results and observations, expressed as:

[0036] ;

[0037] Frequency Domain Adversarial Loss Function , expressed as:

[0038] ;

[0039] Among them, G is the generator, is the discriminator, is the observation data, is the model prediction data, F(x) is the DCT transform, expressed as:

[0040] ;

[0041] in is the radar echo field, N is the length and width of the radar echo field data, It is expressed as:

[0042] ;

[0043] An integer from 0 to N-1.

[0044] Preferably, step S5 comprises:

[0045] Obtain the 10m wind speed and direction, 2m temperature, 2m relative humidity, surface air pressure, 700hPa wind speed and direction, air temperature lapse rate and dew point temperature lapse rate forecasts of the numerical model within the spatial range of the forecast method;

[0046] The above background field is integrated with the convective gust nowcast results, which can be expressed as:

[0047] ;

[0048] ;

[0049] ;

[0050] ;

[0051] in, represents the grid background field; represents the incremental field used in multigrid solving, represents the site observation value, Indicates that when solving multiple grids, each layer needs to calculate the grid point background field correction increment. represents the final corrected background field, represents the observation operator that interpolates the grid background field to the station coordinates, and the superscript k represents the number of different levels of the multigrid. Represents the number of multi-grid levels set, which is determined according to the actual grid dimension; represents the ratio of the empirical observation error covariance to the background field error covariance, usually between 0 and 1; T represents transposition; is the name of the cost function of the kth layer.

[0052] Preferably, the horizontal resolution of the forecast grid established in step S1 is 1 km;

[0053] The convective gust nowcast result obtained in step S4 has a horizontal grid spacing of 1 km;

[0054] The numerical model forecast data obtained in step S5 and the fused gust forecast results generated with physical constraints have a horizontal grid spacing of 1 km;

[0055] In step S6, the terrain data are terrain height, slope aspect and land use type data with a resolution of 100 m, and the corrected fused forecast wind field is obtained with a horizontal grid spacing of 100 m.

[0056] Another aspect of the present invention provides a non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions enabling a computer to execute the above-mentioned physically constrained minute-by-minute 100-meter gust fusion nowcasting method.

[0057] Yet another aspect of the present invention provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the processor calls the logic instructions in the memory to execute the above-mentioned physically constrained minute-by-minute 100-meter gust fusion nowcasting method.

[0058] On the other hand, the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer executes the above-mentioned physically constrained minute-by-hundred-meter gust fusion nowcasting method.

[0059] Beneficial effect: The physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method provided by the present invention integrates the 100-meter-level fusion gust nowcasting technology of high-resolution model and deep learning nowcasting, strengthens the forecasting ability of mixed gusts through model physical constraints, and improves the spatial precision level of forecast results.

[0060] The present invention establishes a multi-observation factor data set of radar combined reflectivity, ground wind speed and lightning density as model input; at the same time, a comprehensive loss function that integrates multiple features is designed, among which the strong wind few-sample data balance loss function can effectively enhance the learning ability of small-sample strong convective gust events above 24.5m / s (level 10), and the frequency domain adversarial loss function can focus on the small and medium-scale structures in convective storms for training, thereby improving the model's convective storm structure and intensity evolution forecasting ability; the present invention solves the problem of forecast accuracy of strong convective gusts through the design of disaster-causing strong wind loss function under the deep learning framework and the use of high-resolution model physical constraints, providing important technical support for severe convective weather forecast and warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The present invention is a flow chart of the physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method.

[0062] Figure 2 This is a comparison chart of forecast results of the physical-constrained minute-hundred-meter-level gust fusion nowcasting method of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. 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. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0064] Combine the following Figure 1-Figure 2 The present invention describes the physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method provided by the present invention.

[0065] Embodiment 1: This embodiment provides a physically constrained one-hundred-meter-level gust fusion nowcasting method, such as Figure 1 As shown, the specific steps are as follows:

[0066] S1. Determine the spatial range and horizontal resolution of the forecast method and establish the forecast grid;

[0067] In this embodiment, Jiangsu Province is used as the spatial range of the forecast method to collect data and explain the gust forecast in detail. The horizontal grid spacing of the forecast grid is 1 km.

[0068] S2. Establish a data set. This step requires establishing a data set consisting of three different data types, including:

[0069] (1) Extract the basic reflectivity factor data of Jiangsu area from April to September 2019 to 2022 from the weather radar, perform quality control, and interpolate to the equidistant forecast grid through three-dimensional network mosaic. The radar combined reflectivity data set is obtained by taking the maximum value in the vertical direction;

[0070] (2) The instantaneous wind speed data in Jiangsu Province from April to September 2019 to 2022 were extracted from the ground automatic station and interpolated to the equidistant forecast grid using the inverse distance weighted method to establish a ground instantaneous wind speed dataset;

[0071] (3) The lightning observation frequency data in Jiangsu Province from April to September 2019 to 2022 were extracted from the lightning locator, interpolated to an equidistant forecast grid using the nearest neighbor method, and combined with the spatial neighborhood method to establish a lightning density dataset.

[0072] S3. Construct a model framework for convective gale nowcasting, and perform model training using the data sets established in step S2, and generate an optimal convective gust nowcasting model in combination with the gust coefficient.

[0073] This step first requires the design of a framework for a convective gale nowcasting model based on deep learning. In this embodiment, the PhyDNet spatiotemporal convolutional neural network with an attention mechanism is used as the model framework, and the attention mechanism acts on time features. When the PhyDNet model forecasts the gusts at the next moment, it is generated moment by moment, and it considers the actual weather conditions in the near future more. The spatiotemporal convolutional neural network with an attention mechanism can consider the actual weather conditions at all historical input moments, allocate appropriate attention (weights) to all historical input moments, and combine all attention (weights) to obtain the best forecast effect at the next moment.

[0074] Based on this model framework, before training, each data set needs to be divided into time series. In the data sequence of 40 consecutive time periods totaling 4 hours, the first 20 frames are model inputs and the last 20 frames are labels.

[0075] Under the algorithm framework of deep learning, the weighted balanced L1 loss function, the small sample data balance loss function of disastrous strong winds, the lightning occurrence probability loss function and the frequency domain adversarial loss function are designed to carry out model training and hyper-parameter adjustment, and the optimal nowcasting model for convective gusts is generated in combination with the gust coefficient.

[0076] The designed loss function is a combined loss function, expressed as:

[0077] ;

[0078] in, represents the weight-balanced L1 loss function, represents the balanced loss function of strong wind and few sample data, represents the lightning probability loss function, represents the frequency domain adversarial loss function.

[0079] Each loss function is described as follows:

[0080] Weight-balanced joint L1 loss function , expressed as:

[0081] ;

[0082] in, and denote the observation field and forecast field of convective gusts, respectively. and They represent the observation field and forecast field of radar combined reflectivity respectively, N represents the total number of samples of all grid points at all forecast times, and Respectively represent the weighted functions of the two variables, and automatically assign weights by counting the number of variables falling into different threshold intervals;

[0083] Balanced loss function for strong wind and few sample data , expressed as:

[0084] ;

[0085] in, For control The balance weight of , a and c are the contraction control parameters;

[0086] Lightning Occurrence Probability Loss Function , expressed as:

[0087] ;

[0088] in, is the binary cross entropy loss function, expressed as:

[0089] ;

[0090] in, , They respectively represent the observation field and model prediction field of whether lightning occurs.

[0091] Based on the Dice coefficient, it is a measure of the similarity between lightning forecast results and observations, expressed as:

[0092] ;

[0093] Frequency Domain Adversarial Loss Function , expressed as:

[0094] ;

[0095] Among them, G is the generator, is the discriminator, is the observation data, is the model prediction data, F(x) is the DCT transform, expressed as:

[0096] ;

[0097] in is the radar echo field, N is the length and width of the radar echo field data, It is expressed as:

[0098] ;

[0099] An integer from 0 to N-1.

[0100] S4. After the model training is completed, the real-time radar combined reflectivity data, ground instantaneous wind speed data and lightning density data are input into the optimal nowcasting model for convective gusts to generate a convective gust sequence for the next 20 consecutive hours, totaling 2 hours, and obtain the nowcasting results of convective gusts with a grid interval of 0-2 hours and 1 km.

[0101] S5. Obtain the forecast data of 10m wind speed and direction, 2m temperature, 2m relative humidity, ground pressure, 700hPa wind speed and direction, temperature lapse rate and dew point temperature lapse rate of the numerical model with a horizontal resolution of 1km in Jiangsu Province, and generate the gust forecast results of the numerical model in combination with the gust coefficient as the background field.

[0102] The multi-grid variational assimilation method is used to fuse the background field and convective gust nowcast results provided by the numerical model with a horizontal resolution of 1 km in Jiangsu Province to generate the physically constrained fusion gust forecast results, which are expressed as:

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] in, represents the grid background field; represents the incremental field used in multigrid solving, represents the site observation value, Indicates that when solving multiple grids, each layer needs to calculate the grid point background field correction increment. represents the final corrected background field, represents the observation operator that interpolates the grid background field to the station coordinates, and the superscript k represents the number of different levels of the multigrid. Represents the number of multi-grid levels set, which is determined according to the actual grid dimension; represents the ratio of the empirical observation error covariance to the background field error covariance, usually between 0 and 1; T represents transposition; is the name of the cost function of the kth layer.

[0108] S6. Based on the fused gust forecast results with a horizontal grid spacing of 1 km, the terrain height, slope aspect and land use type data with a resolution of 100 m are used, and combined with the WRF parameterization scheme (such as empirical parameterization schemes such as atmospheric stability), three-dimensional temperature, three-dimensional humidity, three-dimensional air pressure and three-dimensional wind field meteorological initial values ​​that are more in line with the actual situation are generated, and then input into the wind field rapid diagnosis model. After adjustment by the mass conservation law, the revised fused forecast wind field is obtained, and the horizontal grid spacing reaches 100 m.

[0109] The forecast results obtained by the forecast method of the present invention are compared with those obtained by the traditional old method. Figure 2 The first column in the figure shows the gusts observed by the automatic station, the second column shows the 1h and 2h forecast results provided by the traditional method, and the third column shows the 1h and 2h forecast results provided by the present invention.

[0110] Embodiment 2: This embodiment provides a non-transitory computer-readable storage medium, on which computer instructions are stored, the computer instructions causing the computer to execute a physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method, the method comprising the following steps:

[0111] S1. Determine the spatial range and horizontal resolution of the forecast method and establish the forecast grid;

[0112] S2, based on the basic reflectivity factor data extracted by the weather radar, the instantaneous wind speed data extracted by the ground station, and the lightning observation frequency data extracted by the lightning locator in the forecast area, after preprocessing, align them to the forecast grid, and establish the radar combined reflectivity data set, the ground instantaneous wind speed data set, and the lightning density data set;

[0113] S3, constructing a model framework for convective gale nowcasting, and performing model training with each data set established in step S2, and generating an optimal convective gust nowcasting model in combination with the gust coefficient;

[0114] S4, inputting the real-time radar combined reflectivity data, the ground instantaneous wind speed data and the lightning density data into the convective gust optimal nowcasting model to obtain the future convective gust nowcasting result;

[0115] S5. Obtain the numerical model forecast data for the corresponding period within the spatial range of the forecast method, and generate the numerical model gust forecast results in combination with the gust coefficient; adopt the multi-grid variational assimilation method to fuse the numerical model gust forecast results and the convective gust nowcast results to generate the physically constrained fusion gust forecast results;

[0116] S6. Based on the fused gust forecast results, the meteorological initial values ​​are generated according to the terrain data within the spatial range of the forecast method and combined with the WRF parameterization scheme. Then, the wind field rapid diagnosis model is used to obtain the revised fused forecast wind field after adjustment based on the mass conservation law constraints.

[0117] Embodiment 3: This embodiment provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. The processor may call the logic instructions in the memory to execute a physically constrained minute-by-minute 100-meter gust fusion nowcasting method, which includes the following steps:

[0118] S1. Determine the spatial range and horizontal resolution of the forecast method and establish the forecast grid;

[0119] S2, based on the basic reflectivity factor data extracted by the weather radar, the instantaneous wind speed data extracted by the ground station, and the lightning observation frequency data extracted by the lightning locator in the forecast area, after preprocessing, align them to the forecast grid, and establish the radar combined reflectivity data set, the ground instantaneous wind speed data set, and the lightning density data set;

[0120] S3, constructing a model framework for convective gale nowcasting, and performing model training with each data set established in step S2, and generating an optimal convective gust nowcasting model in combination with the gust coefficient;

[0121] S4, inputting the real-time radar combined reflectivity data, the ground instantaneous wind speed data and the lightning density data into the convective gust optimal nowcasting model to obtain the future convective gust nowcasting result;

[0122] S5. Obtain the numerical model forecast data for the corresponding period within the spatial range of the forecast method, and generate the numerical model gust forecast results in combination with the gust coefficient; adopt the multi-grid variational assimilation method to fuse the numerical model gust forecast results and the convective gust nowcast results to generate the physically constrained fusion gust forecast results;

[0123] S6. Based on the fused gust forecast results, the meteorological initial values ​​are generated according to the terrain data within the spatial range of the forecast method and combined with the WRF parameterization scheme. Then, the wind field rapid diagnosis model is used to obtain the revised fused forecast wind field after adjustment based on the mass conservation law constraints.

[0124] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0125] Embodiment 4: This embodiment provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method, which includes the following steps:

[0126] S1. Determine the spatial range and horizontal resolution of the forecast method and establish the forecast grid;

[0127] S2, based on the basic reflectivity factor data extracted by the weather radar, the instantaneous wind speed data extracted by the ground station, and the lightning observation frequency data extracted by the lightning locator in the forecast area, after preprocessing, align them to the forecast grid, and establish the radar combined reflectivity data set, the ground instantaneous wind speed data set, and the lightning density data set;

[0128] S3, constructing a model framework for convective gale nowcasting, and performing model training with each data set established in step S2, and generating an optimal convective gust nowcasting model in combination with the gust coefficient;

[0129] S4, inputting the real-time radar combined reflectivity data, the ground instantaneous wind speed data and the lightning density data into the convective gust optimal nowcasting model to obtain the future convective gust nowcasting result;

[0130] S5. Obtain the numerical model forecast data for the corresponding period within the spatial range of the forecast method, and generate the numerical model gust forecast results in combination with the gust coefficient; adopt the multi-grid variational assimilation method to fuse the numerical model gust forecast results and the convective gust nowcast results to generate the physically constrained fusion gust forecast results;

[0131] S6. Based on the fused gust forecast results, the meteorological initial values ​​are generated according to the terrain data within the spatial range of the forecast method and combined with the WRF parameterization scheme. Then, the wind field rapid diagnosis model is used to obtain the revised fused forecast wind field after adjustment based on the mass conservation law constraints.

[0132] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, 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 scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0133] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this 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 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 described in each embodiment or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method, characterized in that: include: S1. Determine the spatial range and horizontal resolution of the forecast method and establish the forecast grid; S2, based on the basic reflectivity factor data extracted by the weather radar, the instantaneous wind speed data extracted by the ground station, and the lightning observation frequency data extracted by the lightning locator in the forecast area, after preprocessing, align them to the forecast grid, and establish the radar combined reflectivity data set, the ground instantaneous wind speed data set, and the lightning density data set; S3, constructing a model framework for convective gale nowcasting, and performing model training with each data set established in step S2, and generating an optimal convective gust nowcasting model in combination with the gust coefficient; S4, inputting the real-time radar combined reflectivity data, the ground instantaneous wind speed data and the lightning density data into the convective gust optimal nowcasting model to obtain the future convective gust nowcasting result; S5, obtaining numerical model forecast data for a corresponding period within the spatial range of the forecast method, and generating a numerical model gust forecast result in combination with the gust coefficient; The multi-grid variational assimilation method is used to fuse the gust forecast results of the numerical model and the convective gust nowcast results to generate physically constrained fused gust forecast results. S6. Based on the fused gust forecast results, the meteorological initial values ​​are generated according to the terrain data within the spatial range of the forecast method and combined with the WRF parameterization scheme. Then, the wind field rapid diagnosis model is used to obtain the revised fused forecast wind field after adjustment based on the mass conservation law constraints.

2. The physical constraint minute-by-minute 100-meter-level gust fusion nowcasting method according to claim 1 is characterized in that: Step S2 includes: (1) Extract the basic reflectivity factor data within the spatial range of the forecast method from the weather radar, perform quality control, and interpolate to the equidistant forecast grid through three-dimensional meshing. The radar combined reflectivity data set is obtained by taking the maximum value in the vertical direction. (2) Extract instantaneous wind speed data from ground automatic stations within the spatial range of the forecast method, interpolate to equidistant forecast grids using the inverse distance weighting method, and establish a ground instantaneous wind speed dataset; (3) The lightning observation frequency data are extracted from the lightning locators within the spatial range of the forecast method, interpolated to the equidistant forecast grid using the nearest neighbor method, and combined with the spatial neighborhood method to establish a lightning density dataset.

3. The physical constraint minute-by-minute 100-meter-level gust fusion nowcasting method according to claim 1 is characterized in that: The model framework for convective gale nowcasting constructed in step S3 adopts the PhyDNet spatiotemporal convolutional neural network with an attention mechanism.

4. The physical constraint minute-by-minute 100-meter-level gust fusion nowcasting method according to claim 3 is characterized in that: When performing model training with each data set in step S3, each data set is divided into time series. In a data sequence of M consecutive time periods, the first M / 2 frames are used as input of the model, and the last M / 2 frames are used as labels. In step S4, the real-time radar combined reflectivity data, ground instantaneous wind speed data and lightning density data are input into the convective gust optimal nowcasting model to obtain the convective gust sequence data of M / 2 consecutive time periods in the future, and the convective gust nowcasting result is obtained based on this.

5. The physical constraint minute-by-minute 100-meter gust fusion nowcasting method according to claim 4 is characterized in that: When the model is trained in step S3, the designed loss function is a combined loss function, which is expressed as: ； in, represents the weight-balanced L1 loss function, represents the balanced loss function of strong wind and few sample data, represents the lightning probability loss function, represents the frequency domain adversarial loss function; Weight-balanced joint L1 loss function , expressed as: ； in, and denote the observed field and model forecast field of convective gusts, respectively. and They represent the observation field and forecast field of radar combined reflectivity respectively, N represents the total number of samples of all grid points at all forecast times, and Respectively represent the weighted functions of the two variables, and automatically assign weights by counting the number of variables falling into different threshold intervals; Balanced loss function for strong wind and few sample data , expressed as: ； in, For control The balance weight of , a and c are the contraction control parameters; Lightning Occurrence Probability Loss Function , expressed as: ； in, is the binary cross entropy loss function, expressed as: ； in, , The observation field and model prediction field represent whether lightning occurs or not; Based on the Dice coefficient, it is a measure of the similarity between lightning forecast results and observations, expressed as: ； Frequency Domain Adversarial Loss Function , expressed as: ； Among them, G is the generator, is the discriminator, is the observation data, is the model prediction data, F(x) is the DCT transform, expressed as: ； in is the radar echo field, N is the length and width of the radar echo field data, It is expressed as: ； An integer from 0 to N-1.

6. The physical constraint minute-by-minute 100-meter gust fusion nowcasting method according to claim 1 is characterized in that: Step S5 includes: Obtain the 10m wind speed and direction, 2m temperature, 2m relative humidity, surface air pressure, 700hPa wind speed and direction, air temperature lapse rate and dew point temperature lapse rate forecasts of the numerical model within the spatial range of the forecast method; The above background field is integrated with the convective gust nowcast results, which can be expressed as: ； ； ； ； in, represents the grid background field; represents the incremental field used in multigrid solving, represents the site observation value, Indicates that when solving multiple grids, each layer needs to calculate the grid point background field correction increment. represents the final corrected background field, represents the observation operator that interpolates the grid background field to the station coordinates, and the superscript k represents the number of different levels of the multigrid. Represents the number of multi-grid levels set, which is determined according to the actual grid dimension; represents the ratio of the empirical observation error covariance to the background field error covariance, usually between 0 and 1; T represents transposition; is the name of the cost function of the kth layer.

7. The physical constraint minute-by-minute 100-meter gust fusion nowcasting method according to claim 1 is characterized in that: The horizontal resolution of the forecast grid established in step S1 is 1 km; The convective gust nowcast result obtained in step S4 has a horizontal grid spacing of 1 km; The numerical model forecast data obtained in step S5 and the fused gust forecast results generated with physical constraints have a horizontal grid spacing of 1 km; In step S6, the terrain data are terrain height, slope aspect and land use type data with a resolution of 100 m, and the corrected fused forecast wind field is obtained with a horizontal grid spacing of 100 m.

8. A non-transitory computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and the computer instructions enable the computer to execute the physical constraint minute-by-minute 100-meter-level gust fusion nowcasting method as described in any one of claims 1-7.

9. An electronic device, characterized in that: include: A processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the processor calls the logic instructions in the memory to execute the physically constrained minute-by-minute 100-meter-level gust fusion nowcasting method as described in any one of claims 1-7.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer executes the physically constrained minute-by-minute 100-meter gust fusion nowcasting method as described in any one of claims 1 to 7.

Images