Objective forecasting method for rainstorm early warning information

Abstract

The invention provides an objective forecasting method for rainstorm early warning information, and the method comprises the steps: collecting the precipitation data of a plurality of weather stations, and obtaining the rainfall numerical model forecasting data; performing data preprocessing on the rainfall numerical model forecast data and the rainfall data of the plurality of weather stations to obtain a plurality of processed target rainfall data and target rainfall numerical model forecast data, and calculating the rainfall numerical model forecast data of the plurality of weather stations according to the processed target rainfall data and the target rainfall numerical model forecast data at first time intervals. Calculating a forecast precipitation sequence of the first time interval; according to a preset multistage rainfall early warning standard and a forecast rainfall amount sequence, determining and judging whether forecast of rainstorm early warning information needs to be carried out or not; and if the rainstorm early warning information needs to be forecasted, generating alarm information, and pushing the alarm information to a target user display interface in the jurisdiction corresponding to the alarm information. And objective forecasting of real-time updated rainstorm early warning information is realized.

Description

Technical Field

[0001] The present application relates to the technical field of rainfall forecasting, and in particular to an objective forecasting method for rainstorm warning information. Background Art

[0002] Heavy rain warning signals are issued by meteorological authorities before heavy rain arrives based on weather monitoring. They are categorized into four levels: blue, yellow, orange, and red. In China, a red warning is the highest level, indicating that rainfall in some areas has reached over 100 mm in the past three hours and is likely to continue. Issuing heavy rain warning signals plays a vital role in improving early warning capabilities and reducing property losses and casualties, which is crucial for production and daily life.

[0003] However, the output of objective forecast products for rainstorm warning information, and the objective forecast and targeted reminder release functions of rainstorm warning information have not been realized. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide an objective forecasting method for heavy rain warning information, which can at least realize the output of objective forecast products of heavy rain warning information updated at preset time intervals, objective forecast of heavy rain warning information and targeted reminder release functions through precipitation data and precipitation numerical model forecast data of multiple meteorological stations.

[0005] An embodiment of the present application provides an objective forecasting method for rainstorm warning information, characterized in that the method includes: collecting precipitation data from multiple meteorological stations and obtaining precipitation numerical model forecast data, wherein the precipitation numerical model forecast data is precipitation forecast data at every second time interval, and the precipitation numerical model forecast data covers precipitation forecast information within a first time period; performing data preprocessing on the precipitation numerical model forecast data and the precipitation data from the multiple meteorological stations to obtain a plurality of processed target precipitation data and target precipitation numerical model forecast data, wherein the plurality of target precipitation data is precipitation data at every first time interval; calculating a forecast precipitation sequence for the first time interval based on the processed target precipitation data and the precipitation numerical model forecast data at every first time interval; determining whether it is necessary to forecast rainstorm warning information based on pre-set multi-level precipitation warning standards and the forecast precipitation sequence; and generating an alarm message if it is necessary to forecast rainstorm warning information, and pushing the alarm message to a target user display interface within the jurisdiction corresponding to the alarm message.

[0006] Optionally, the step of performing data preprocessing on the precipitation data of multiple meteorological stations includes: performing a completeness check on the precipitation data of multiple meteorological stations, removing abnormal data, and obtaining multiple third intermediate precipitation data; performing a data consistency check and / or a threshold check on the multiple third intermediate precipitation data, correcting or removing the multiple third intermediate precipitation data, and obtaining second intermediate precipitation data; and forming multiple target precipitation data according to the standard accumulated precipitation at each first time interval.

[0007] Optionally, the step of performing data preprocessing on the precipitation numerical model forecast data includes: determining a target precipitation numerical model based on pre-established evaluation indicators, according to the actual precipitation conditions in the second time period and a plurality of precipitation numerical models; checking the timeliness and spatial coverage of the precipitation numerical model forecast data to ensure the reliability of the data; performing format conversion on the precipitation numerical model forecast data, extracting accumulated precipitation element data, and processing the data into a precipitation sequence at a third time interval; obtaining geographic information surface data of each city, district, and county, determining a representative precipitation value for each region based on the numerical model precipitation values ​​in the surface data, and extracting the precipitation of the city, district, and county corresponding to the meteorological station in each region; converting the precipitation sequence of the third time interval of the precipitation numerical model forecast data into a precipitation sequence of the first time interval based on a pre-established spatiotemporal mapping relationship and a conditional diffusion model, wherein the resolution of the precipitation sequence of the first time interval is higher than that of the precipitation sequence of the third time interval.

[0008] Optionally, the step of determining the target precipitation numerical model includes: calculating multiple forecast skill scores for multiple rainstorm scoring indicators pre-set in the second time period; ranking multiple forecast skill scores for multiple rainstorm scoring indicators, and calculating the score value of the rainstorm scoring indicator based on the ranking of each forecast skill score; calculating the time periods before the preset time period to obtain multiple reference total scores, and selecting the numerical model forecast with the highest score after weighted averaging as the target precipitation numerical model.

[0009] Optionally, a spatiotemporal mapping relationship is established by the following steps: establishing a coarse-fine-grained time mapping matrix and defining physical conservation constraints; wherein, the third time interval is 3 hours and the first time interval is 10 minutes, and the coarse-fine-grained time mapping matrix is ​​established by the following formula: assuming that the original 3-hour time point sequence is: ; The 10-minute time series is: ;in .

[0010] Optionally, a conditional diffusion model is established by the following steps: designing a conditional encoder and a noise prediction network based on UNet; adding a physical constraint mechanism to obtain a conditional diffusion model.

[0011] Optionally, the conditional encoder uses a 5-layer temporal convolutional network to train the original 3-hour precipitation sequence For feature extraction, the calculation process is: ; Among them, the position code After being processed by the Transformer Encoder, it is added to the features extracted by convolution to obtain the final conditional feature representation:

[0012]

[0013] Optionally, the UNet-based noise prediction network includes a downsampling module, an upsampling module and a skip connection, and a conditional attention module; wherein the downsampling module adopts 4-level downsampling, each level uses 1D convolution with a step size of 2, and the number of channels is doubled in the downsampling process; the upsampling module and the skip connection: corresponding to the downsampling module, a transposed convolution or upsampling operation is adopted, while the low-level feature information is retained through the skip connection; the conditional attention module introduces the conditional attention module in the second layer of the decoder, and its calculation formula is:

[0014] ;

[0015] Among them, the query Derived from the current decoder features, key With numerical value Derived from the conditional encoder output .

[0016] Optionally, the physical constraint mechanism includes a training phase and a generation phase; wherein the training phase adds a conservation regularization term to the total loss function, namely:

[0017] ;

[0018] In the generation stage, after each denoising step, a physical projection operation is performed on the generated 10-minute precipitation, and each 10-minute forecast is normalized so that the accumulated value satisfies the following formula:

[0019] ;

[0020] To ensure that the total precipitation in 10 minutes in each 3-hour period is strictly equal to the original forecast .

[0021] Optionally, calculating the forecast precipitation sequence for the first time interval includes: determining a forecast precipitation sequence for the third time period every first time interval based on the forecast precipitation sequence; and splicing a plurality of target precipitation data with the forecast precipitation sequence for the third time period every first time interval at each first time interval to form a forecast precipitation sequence for the third time period every first time interval in the future first time interval.

[0022] The method further includes: updating the forecast precipitation sequence at every first time interval.

[0023] The coronary artery evaluation and prediction methods, devices, equipment, media and products provided in the embodiments of the present application can at least determine the multi-level topological information of the coronary arteries by constructing image sub-blocks and a hierarchical topological perception network, and determine the changing trend of the plaque component of the stenotic coronary artery based on the multi-level topological information of the coronary artery and multiple time-series images.

[0024] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 A flow chart of an objective forecasting method for rainstorm warning information provided in an embodiment of the present application;

[0027] Figure 2 Schematic diagram of the targeted push and display of objective warning and forecast information of urban and county-level rainstorms provided in the embodiment of this application Figure 1 ;

[0028] Figure 3 Schematic diagram of the targeted push and display of objective warning and forecast information of urban and county-level rainstorms provided in the embodiment of this application Figure 2 . DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for which protection is claimed, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work falls within the scope of protection of the present application.

[0030] First, the application scenarios to which this application is applicable are introduced. This application can be applied in the field of rainfall warning technology.

[0031] Heavy rain warning signals are issued by meteorological authorities before heavy rain arrives based on weather monitoring. They are categorized into four levels: blue, yellow, orange, and red. In China, a red warning is the highest level, indicating that rainfall in some areas has reached over 100 mm in the past three hours and is likely to continue. Issuing heavy rain warning signals plays a vital role in improving early warning capabilities and reducing property losses and casualties, which is crucial for production and daily life.

[0032] However, the output of objective forecast products for rainstorm warning information, and the objective forecast and targeted reminder release functions of rainstorm warning information have not been realized.

[0033] Based on this, an embodiment of the present application provides an objective forecasting method for heavy rain warning information, which can at least realize the output of objective forecast products of heavy rain warning information updated at preset time intervals, and the objective forecast and targeted reminder release functions of heavy rain warning information through precipitation data and precipitation numerical model forecast data of multiple meteorological stations.

[0034] See also Figure 1 , Figure 1 This is a flow chart of an objective forecasting method for rainstorm warning information provided by an embodiment of the present application. Figure 1 As shown in , the objective forecasting method of rainstorm warning information provided by the embodiment of the present application includes:

[0035] S101. Collect precipitation data from multiple weather stations and obtain precipitation numerical model forecast data.

[0036] The precipitation numerical model forecast data is precipitation forecast data at every second time interval, and the precipitation numerical model forecast data covers precipitation forecast information within the first time period.

[0037] As an example, the precipitation data of a weather station may be 1-minute automatic station observation data: meteorological element data such as minute-by-minute precipitation is collected from each automatic weather station.

[0038] The precipitation numerical model forecast data can be precipitation forecast data with a time interval of 3 hours, covering precipitation forecast information with a time interval of 0-72 hours.

[0039] S102 , performing data preprocessing on the precipitation numerical model forecast data and precipitation data of multiple meteorological stations to obtain processed multiple target precipitation data and target precipitation numerical model forecast data.

[0040] The multiple target precipitation data are precipitation data at first time intervals.

[0041] Among them, the step of preprocessing the precipitation data of multiple meteorological stations includes: performing a completeness check on the precipitation data of multiple meteorological stations, removing abnormal data, and obtaining multiple third intermediate precipitation data; performing a data consistency check and / or a threshold check on the multiple third intermediate precipitation data, correcting or removing the multiple third intermediate precipitation data, and obtaining second intermediate precipitation data; and forming multiple target precipitation data according to the standard accumulated precipitation at each first time interval.

[0042] Among them, the step of data preprocessing of the precipitation numerical model forecast data includes: based on pre-established evaluation indicators, according to the actual precipitation conditions in the second time period and multiple precipitation numerical models, determining the target precipitation numerical model; checking the timeliness and spatial coverage of the precipitation numerical model forecast data to ensure the reliability of the data; converting the format of the precipitation numerical model forecast data, and extracting the cumulative precipitation element data, and processing it into a precipitation sequence according to a third time interval; obtaining the geographic information surface data of each city, district and county, determining the representative value of precipitation in each area according to the numerical model precipitation values ​​in the surface data, and extracting the precipitation of the city, district and county corresponding to the meteorological station in each area; based on the pre-established spatiotemporal mapping relationship and conditional diffusion model, converting the precipitation sequence of the third time interval of the precipitation numerical model forecast data into the precipitation sequence of the first time interval, and the resolution of the precipitation sequence of the first time interval is higher than that of the precipitation sequence of the third time interval.

[0043] Specifically, the step of determining the target precipitation numerical model includes: calculating multiple forecast skill scores of multiple rainstorm scoring indicators pre-set in the second time period; ranking multiple forecast skill scores of multiple rainstorm scoring indicators, and calculating the score value of the rainstorm scoring indicator according to the ranking of each forecast skill score; calculating the time periods before the preset time period to obtain multiple reference total scores, and selecting the numerical model forecast with the highest score after weighted averaging as the target precipitation numerical model.

[0044] As an example, data preprocessing for precipitation data from multiple weather stations primarily involves checking data completeness, quality control, and accumulating precipitation data on a 10-minute basis. This processing involves checking data integrity and removing records with excessive missing values ​​or obvious anomalies; performing data quality control to correct or remove erroneous data through methods such as data consistency checks and threshold checks; and, after data quality control, accumulating precipitation data on a 10-minute basis to produce 10-minute cumulative precipitation data.

[0045] For example, the formula for the 10-minute cumulative precipitation is as follows:

[0046] ;

[0047] in, Indicates the accumulated precipitation in 10 minutes. Indicates the The 1-minute precipitation value is expressed in mm. Ten 1-minute precipitation values ​​are accumulated every 10 minutes starting at the hour. For example, the 10-minute precipitation at 10:20 is the accumulation of 10 1-minute precipitation values ​​at 10:11, 10:12, 10:13…10:18, 10:19, and 10:20.

[0048] As an example, the precipitation numerical model forecast data preprocessing process includes numerical model forecast optimization (target precipitation numerical model), data format conversion, district and county station forecast data extraction, and 3-hour precipitation time downscaling.

[0049] Among them, the steps of preprocessing the precipitation numerical model forecast data include: the optimization of the model forecast based on the actual precipitation situation mainly involves establishing evaluation indicators to select a suitable numerical model for forecasting.

[0050] Calculate the error index between the actual 24-hour precipitation and the forecast of each precipitation numerical model to evaluate the accuracy and reliability of each model forecast. According to the needs of heavy rain warning, focus on the forecast ability of heavy precipitation, set heavy rain (≥50mm) scoring indicators, including hit rate (HIT), TS score (TS), false alarm rate, BIAS score, and ETS score. Score each numerical model forecast Relative CCTV forecast score Calculating Forecast Skill Score :

[0051] ;

[0052] According to different seasons, regions, weather conditions and other factors, the score weights of the evaluation indicators are dynamically adjusted to improve the targeting and adaptability of the selection. The total score is 20, that is, 100 points. The numerical model forecast skill values ​​of each individual scoring index are ranked from high to low. If the individual ranking is , calculate the corresponding score value according to the sorting:

[0053] ;

[0054] In the ranking screening period, four total scores were calculated for the past three days, the past week, the past month and the past three months respectively. After weighted average, the numerical model forecast with the highest score was selected as the preferred numerical model forecast.

[0055] Among them, the preprocessing steps of precipitation numerical model forecast data also include: checking the timeliness (0-72 hours) and spatial coverage (ensuring complete coverage of provinces, cities, districts and counties) of the forecast data to ensure the availability of the data.

[0056] Among them, the preprocessing steps of the precipitation numerical model forecast data also include: converting the format of the numerical model forecast data, extracting the cumulative precipitation element data, processing it into a 3-hour precipitation sequence at 3-hour intervals, obtaining the geographic information surface data of each city, district and county, extracting the numerical model precipitation value in the surface data, obtaining the representative precipitation value representing the area through aggregation analysis and weighted averaging, extracting the precipitation of the corresponding city, district and county of the automatic observation station, and unifying the data format for subsequent processing.

[0057] Among them, this application also designs a generative downscaling method that integrates physical conservation laws to convert the numerical forecast 3-hour precipitation data into a 10-minute high-resolution sequence, while strictly ensuring the conservation of time integral to ensure that the accumulated value of 10-minute precipitation every 3 hours and the total 3-hour precipitation remain consistent.

[0058] Specifically, the space-time mapping relationship is established through the following steps: establishing a coarse-fine grained time mapping matrix and defining physical conservation constraints.

[0059] The third time interval is 3 hours, the first time interval is 10 minutes, and the coarse-fine granularity time mapping matrix is ​​established by the following formula:

[0060] Assume the original 3-hour time point sequence is:

[0061] ;

[0062] The 10-minute time series is:

[0063] ;

[0064] in, .

[0065] Specifically, the conditional diffusion model is established through the following steps: designing a conditional encoder and a noise prediction network based on UNet; adding a physical constraint mechanism to obtain the conditional diffusion model.

[0066] Specifically, the conditional encoder uses a 5-layer temporal convolutional network to train the original 3-hour precipitation sequence For feature extraction, the calculation process is:

[0067] ;

[0068] Among them, position encoding After being processed by the Transformer Encoder, it is added to the features extracted by convolution to obtain the final conditional feature representation:

[0069]

[0070] Specifically, the UNet-based noise prediction network includes a downsampling module, an upsampling module with skip connections, and a conditional attention module.

[0071] Among them, the downsampling module adopts 4-level downsampling, and each level uses 1D convolution with a stride of 2. The number of channels is doubled in the downsampling process. The upsampling module and skip connection: corresponding to the downsampling module, transposed convolution or upsampling operation is used, while the low-level feature information is retained through skip connection. The conditional attention module introduces the conditional attention module in the second layer of the decoder, and its calculation formula is:

[0072] ;

[0073] Among them, the query Derived from the current decoder features, key With numerical value Derived from the conditional encoder output .

[0074] Specifically, the physical constraint mechanism includes a training phase and a generation phase; wherein, the training phase adds a conservation regularization term to the total loss function, namely:

[0075] ;

[0076] In the generation stage, after each denoising step, a physical projection operation is performed on the generated 10-minute precipitation, and each 10-minute forecast is normalized so that the accumulated value satisfies the following formula:

[0077] ;

[0078] To ensure that the total precipitation in 10 minutes in each 3-hour period is strictly equal to the original forecast .

[0079] Optionally, the method further comprises: data preprocessing, processing the original 3-hour precipitation data Normalize and construct the corresponding 10-minute time map ; Conditional encoding, extracting features from input data through 5 layers of Conv1D and TransformerEncoder to generate conditional features Diffusion process: Using a diffusion model with a preset number of steps (e.g., 1000 steps), a high-resolution 10-minute precipitation series is generated by iteratively adding and denoising noise, while injecting conditional information and physical constraints at each step. Physical projection: After each denoising step, a normalized projection operation is performed to ensure the conservation of the total amount of precipitation in the local area (within a 3-hour interval). Joint loss optimization: During the training phase, the total loss function is used:

[0080] ;

[0081] in, To balance the hyperparameters of the reconstruction loss and the physical conservation regularization term, the model parameters are optimized.

[0082] S103 . Calculate a forecast precipitation sequence for the first time interval at every first time interval based on the processed target precipitation data and the precipitation numerical model forecast data.

[0083] Specifically, calculating the forecast precipitation sequence for the first time interval includes: determining a forecast precipitation sequence for the third time period every first time interval based on the forecast precipitation sequence; and splicing a plurality of target precipitation data with the forecast precipitation sequence for the third time period every first time interval at each first time interval to form a forecast precipitation sequence for the third time period every first time interval in the future first time interval.

[0084] The method further includes: updating the forecast precipitation sequence at every first time interval.

[0085] As an example, consider the calculation of one-hour rainfall: Since the precipitation forecast data is provided every 10 minutes, the hourly precipitation data must be accumulated for every six consecutive 10-minute periods to obtain the hourly precipitation. For example, the sum of the precipitation data for the first six 10-minute periods is the precipitation for the first hour.

[0086] As an example, for 6-hour rainfall calculation: starting from each 10-minute time point, accumulate the precipitation for the next 6 hours (i.e. 36 10-minute periods) to obtain a 6-hour cumulative precipitation series with 10-minute intervals.

[0087] As an example, for 24-hour rainfall calculation: starting from each 10-minute time point, accumulate the precipitation for the next 24 hours (i.e. 144 10-minute periods) to obtain a 24-hour cumulative precipitation series with 10-minute intervals.

[0088] As an example, for 48-hour rainfall calculation: starting from each 10-minute time point, accumulate the precipitation for the next 48 hours (i.e. 288 10-minute periods) to obtain a 48-hour cumulative precipitation sequence with 10-minute intervals.

[0089] Every 10 minutes, the actual precipitation observed by the automatic station is combined with the interpolated 10-minute forecast precipitation to form the 1-hour precipitation, 6-hour precipitation and 24-hour precipitation at 10-minute intervals for the next 72 hours. The weight of the forecast accumulation is dynamically adjusted according to the deviation and accuracy score of the forecast. is the accumulated precipitation in a certain 10 minutes from the current moment to the past (i=1,2,3…6), is the predicted precipitation accumulated from the current moment to the next 10 minutes, 、 、 、 is the precipitation after splicing for 1 hour, 6 hours, 24 hours and 48 hours, and the weight coefficient is (0≤ ≤1).

[0090] The precipitation splicing formula for the next n 10 minutes is:

[0091] ;

[0092] Where j = mn, where m is the number of accumulated 10-minute precipitation events. For 1-hour precipitation, m = 6; for 6-hour precipitation, m = 36; for 24-hour precipitation, m = 144; and for 48-hour precipitation, m = 288. Time continuity must be ensured during splicing.

[0093] Weight This can be determined based on statistical analysis of historical data, real-time forecast error assessment, etc. For example, when the forecast accuracy is high, the forecast weight can be appropriately increased; when it is closer to the actual observation, the forecast weight can be increased.

[0094] As an example, every 10 minutes, new live observation data from an automatic station is added to the splicing process, along with the precipitation forecast. The specific steps are as follows: When new 10-minute live observation data arrives, it is first quality-controlled and preprocessed. Based on the latest 1-hour live precipitation product, the precipitation pattern forecast is re-optimized and interpolated to 10-minute precipitation values. Following the splicing method, the new live observation precipitation values ​​are spliced ​​with the updated forecast precipitation values ​​to produce new 10-minute spliced ​​precipitation values. This process is repeated to achieve a rolling 10-minute update of the 1-hour precipitation.

[0095] As an example, consider the calculation of hourly rainfall: Since the precipitation forecast data is provided every 10 minutes, the hourly rainfall data must be accumulated for every six consecutive 10-minute periods to obtain the hourly rainfall data. For example, the sum of the rainfall data for the first six 10-minute periods is the rainfall data for the first hour.

[0096] As an example, for 6-hour rainfall calculation: starting from each 10-minute time point, accumulate the precipitation for the next 6 hours (i.e. 36 10-minute periods) to obtain a 6-hour cumulative precipitation series with 10-minute intervals.

[0097] As an example, for 24-hour rainfall calculation: starting from each 10-minute time point, accumulate the precipitation for the next 24 hours (i.e. 144 10-minute periods) to obtain a 24-hour cumulative precipitation series with 10-minute intervals.

[0098] As an example, for 48-hour rainfall calculation: starting from each 10-minute time point, accumulate the precipitation for the next 48 hours (i.e. 288 10-minute periods) to obtain a 48-hour cumulative precipitation sequence with 10-minute intervals.

[0099] S104: Determine whether a rainstorm warning information forecast is required based on pre-set multi-level precipitation warning standards and a predicted precipitation sequence.

[0100] For example, the early warning and warning signal standards at a provincial level are as follows: Early warning judgments are divided into four levels: blue, yellow, orange, and red. They are mainly used to provide early warning guidance and forecasts for cities, districts, and counties. The judgment generation rules are as follows:

[0101] Blue Alert: For each provincial region, the 24-hour cumulative precipitation sequence is checked and calculated. If the 24-hour cumulative precipitation at a certain moment is greater than or equal to 50 mm, the provincial region is judged to have met the blue alert standard at that moment, and the blue alert alarm is triggered.

[0102] Yellow warning: For 24-hour rainfall conditions, check the 24-hour cumulative precipitation sequence. If the 24-hour rainfall at a certain moment is greater than or equal to 100 mm, the yellow warning alarm is triggered.

[0103] For hourly rainfall conditions, check the hourly precipitation sequence. If the precipitation in a certain hour is greater than or equal to 50 mm, a yellow warning alarm is triggered.

[0104] For the combined conditions, the 24-hour cumulative precipitation sequence is checked. If the 24-hour rainfall at a certain moment is greater than or equal to 50 mm, and the cumulative precipitation 48 hours before that moment is greater than or equal to 150 mm, a yellow warning alarm is triggered.

[0105] Orange warning: For 24-hour rainfall conditions, check the 24-hour cumulative precipitation sequence. If the 24-hour rainfall at a certain moment is greater than or equal to 150 mm, the orange warning alarm is triggered.

[0106] For hourly rainfall conditions, the hourly precipitation sequence is checked. If the precipitation in a certain hour is greater than or equal to 75 mm, an orange warning alarm is triggered.

[0107] For the combined conditions, the 24-hour cumulative precipitation sequence is checked. If the 24-hour rainfall at a certain moment is greater than or equal to 50 mm, and the cumulative precipitation 48 hours before that moment is greater than or equal to 200 mm, the orange warning alarm is triggered.

[0108] Red warning: For 24-hour rainfall conditions, check the 24-hour cumulative precipitation sequence. If the 24-hour rainfall at a certain moment is greater than or equal to 200 mm, a red warning alarm is triggered.

[0109] For hourly rainfall conditions, the hourly precipitation sequence is checked. If the precipitation in a certain hour is greater than or equal to 100 mm, a red warning alarm is triggered.

[0110] For the combined conditions, the 24-hour cumulative precipitation sequence is checked. If the 24-hour rainfall at a certain moment is greater than or equal to 50 mm, and the cumulative precipitation 48 hours before that moment is greater than or equal to 250 mm, a red warning alarm is triggered.

[0111] Specifically, the warning signals are divided into four levels: blue, yellow, orange, and red. They are mainly used to objectively forecast the warning signals for the jurisdiction. The judgment and generation rules are as follows:

[0112] Blue Warning Signal: For each city or county, the calculated 6-hour cumulative precipitation sequence is checked. If the 6-hour cumulative precipitation at a certain moment is greater than or equal to 50 mm, the city or county is judged to have met the blue warning standard at that moment, and the blue warning signal alarm is triggered.

[0113] Yellow warning signal: For 24-hour rainfall conditions, check the 24-hour cumulative precipitation sequence. If the 24-hour rainfall at a certain moment is greater than or equal to 150 mm, the yellow warning signal alarm is triggered.

[0114] For the 6-hour rainfall condition, the 6-hour cumulative precipitation sequence is checked. If the 6-hour rainfall at a certain moment is greater than or equal to 100 mm, a yellow warning signal alarm is triggered.

[0115] For hourly rainfall conditions, check the hourly precipitation sequence. If the precipitation in a certain hour is greater than or equal to 50 mm, a yellow warning signal alarm is triggered.

[0116] Orange warning signal: For 24-hour rainfall conditions, check the 24-hour cumulative precipitation sequence. If the 24-hour rainfall at a certain moment is greater than or equal to 200 mm, the orange warning signal alarm is triggered.

[0117] For the 6-hour rainfall condition, the 6-hour cumulative precipitation sequence is checked. If the 6-hour rainfall at a certain moment is greater than or equal to 150 mm, the orange early warning signal alarm is triggered.

[0118] For hourly rainfall conditions, check the hourly precipitation sequence. If the precipitation in a certain hour is greater than or equal to 75 mm, the orange warning signal alarm is triggered.

[0119] Red warning signal: For 24-hour rainfall conditions, check the 24-hour cumulative precipitation sequence. If the 24-hour rainfall at a certain moment is greater than or equal to 250 mm, a red warning signal alarm is triggered.

[0120] For the 6-hour rainfall condition, the 6-hour cumulative precipitation sequence is checked. If the 6-hour rainfall at a certain moment is greater than or equal to 200 mm, a red warning signal alarm is triggered.

[0121] For hourly rainfall conditions, check the hourly precipitation sequence. If the precipitation in a certain hour is greater than or equal to 100 mm, a red warning signal alarm is triggered.

[0122] S105: If a rainstorm warning information forecast is required, an alarm message is generated and pushed to a target user display interface within the jurisdiction corresponding to the alarm message.

[0123] Specifically, the data transmission method for pushing and displaying objective rainstorm warning and forecast information is to organize the alarm information into a structured data format, such as JSON or XML, to facilitate docking with subsequent information release systems. The alarm information can also be output to a local storage device for subsequent query and analysis.

[0124] Specifically, the system for pushing and displaying objective rainstorm warning and forecast information adopts a distributed architecture, deploying functional modules such as data acquisition, preprocessing, interpolation calculations, splicing and updating, warning product generation, and alarm notifications on different servers to improve system stability and scalability. A database system is established to store information such as automatic station observation data, precipitation model forecast data, intermediate calculation results, and warning products.

[0125] Specifically, once a region is determined to have reached the warning criteria, detailed alarm information will be generated, including the warning level (blue, yellow, orange, or red), the affected region (specific provincial, municipal, or county administrative district name), and the estimated time when the warning criteria will be reached.

[0126] When provincial users log in, targeted push and display of objective rainstorm warning and forecast information for each city, district and county, such as Figure 2 When users at the city, district, and county levels log in, targeted push and display of objective warning and forecast information on heavy rain in their respective areas are available, such as Figure 3 shown.

[0127] The objective forecasting method for rainstorm warning information provided in the embodiment of the present application can at least realize the output of objective forecast products of rainstorm warning information updated at preset time intervals, objective forecast of rainstorm warning information and targeted reminder release functions through precipitation data and precipitation numerical model forecast data of multiple meteorological stations.

[0128] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0129] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0130] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0131] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. An objective forecasting method for rainstorm warning information, characterized in that: The method comprises: Collecting precipitation data from multiple weather stations and obtaining precipitation numerical model forecast data, wherein the precipitation numerical model forecast data is precipitation forecast data at every second time interval, and the precipitation numerical model forecast data covers precipitation forecast information within the first time period; performing data preprocessing on the precipitation numerical model forecast data and precipitation data of multiple meteorological stations to obtain processed multiple target precipitation data and target precipitation numerical model forecast data, wherein the multiple target precipitation data are precipitation data at every first time interval; At every first time interval, a forecast precipitation series for the first time interval is calculated based on the processed target precipitation data and the precipitation numerical model forecast data; Determine whether a rainstorm warning information forecast is necessary based on the pre-set multi-level precipitation warning standards and forecast precipitation sequence; If a rainstorm warning information forecast is required, an alarm message is generated and pushed to a target user display interface within the jurisdiction corresponding to the alarm message.

2. The method according to claim 1, characterized in that The steps for data preprocessing of precipitation data from multiple weather stations include: Perform a completeness check on the precipitation data of multiple weather stations, remove abnormal data, and obtain multiple third intermediate precipitation data; performing a data consistency check and / or a threshold check on the plurality of third intermediate precipitation data, and correcting or removing the plurality of third intermediate precipitation data to obtain second intermediate precipitation data; A plurality of target precipitation data are generated according to the standard accumulated precipitation at each first time interval.

3. The method according to claim 1, characterized in that The step of performing data preprocessing on the precipitation numerical model forecast data includes: Based on pre-established evaluation indicators, a target precipitation numerical model is determined according to the actual precipitation conditions in the second time period and multiple precipitation numerical models; Check the timeliness and spatial coverage of the precipitation numerical model forecast data to ensure the reliability of the data; The precipitation numerical model forecast data is formatted, and accumulated precipitation element data is extracted and processed into a precipitation sequence according to a third time interval; Obtain geographic information surface data for each city, district, and county, determine the representative precipitation value for each region based on the numerical model precipitation values ​​in the surface data, and extract the precipitation for the city, district, and county corresponding to the meteorological station in each region; Based on a pre-established spatiotemporal mapping relationship and a conditional diffusion model, the precipitation sequence of the third time interval of the precipitation numerical model forecast data is converted into a precipitation sequence of the first time interval, where the resolution of the precipitation sequence of the first time interval is higher than that of the precipitation sequence of the third time interval.

4. The method according to claim 3, characterized in that The steps to determine the target precipitation numerical model include: Calculating multiple forecast skill scores for multiple pre-set rainstorm scoring indicators within the second time period; Ranking multiple forecast skill scores of multiple heavy rain scoring indicators, and calculating the score value of the heavy rain scoring indicator according to the ranking of each forecast skill score; Multiple reference total scores are calculated for each time period before the preset time period, and after weighted averaging, the numerical model forecast with the highest score is selected as the target precipitation numerical model.

5. The method according to claim 4, characterized in that The spatiotemporal mapping relationship is established through the following steps: Establish a coarse-fine grained time mapping matrix and define physical conservation constraints; The third time interval is 3 hours, the first time interval is 10 minutes, and the coarse-fine granularity time mapping matrix is ​​established by the following formula: Assume the original 3-hour time point sequence is: ； The 10-minute time series is: ； in .

6. The method according to claim 5, characterized in that The conditional diffusion model is established by following the steps below: Design a conditional encoder and a noise prediction network based on UNet; By adding physical constraint mechanism, we get the conditional diffusion model.

7. The method according to claim 6, characterized in that The conditional encoder uses a 5-layer temporal convolutional network to train the original 3-hour precipitation sequence For feature extraction, the calculation process is: ； Among them, position encoding After being processed by the Transformer Encoder, it is added to the features extracted by convolution to obtain the final conditional feature representation:

8. The method according to claim 7, characterized in that The noise prediction network based on UNet includes a downsampling module, an upsampling module with skip connections, and a conditional attention module; Among them, the downsampling module adopts 4-level downsampling, each level uses 1D convolution with a stride of 2, and the number of channels is doubled in the downsampling process; Upsampling module and skip connection: Corresponding to the downsampling module, it uses transposed convolution or upsampling operations, while retaining low-level feature information through skip connections; The conditional attention module is introduced in the second layer of the decoder, and its calculation formula is: ； Among them, the query Derived from the current decoder features, key With numerical value Derived from the conditional encoder output .

9. The method according to claim 6, characterized in that The physical constraint mechanism includes a training phase and a generation phase; In the training phase, a conservation regularization term is added to the total loss function, namely: ； In the generation stage, after each denoising step, a physical projection operation is performed on the generated 10-minute precipitation, and each 10-minute forecast is normalized so that the accumulated value satisfies the following formula: ； To ensure that the total precipitation in 10 minutes in each 3-hour period is strictly equal to the original forecast .

10. The method according to claim 1, characterized in that Calculate the forecast precipitation series for the first time interval, including: Determining a forecast precipitation sequence for a third time period every first time interval according to the forecast precipitation sequence; The plurality of target precipitation data are spliced ​​with the forecast precipitation sequence of the third time period every first time interval at each first time interval to form a forecast precipitation sequence of the third time period every first time interval in the future first time period. The method further comprises: The forecast precipitation sequence is updated at every first time interval.

Images