Short-term intelligent numerical weather prediction method and system based on environmental perception

Abstract

This invention discloses a short-term intelligent numerical weather prediction method and system based on environmental perception, belonging to the field of meteorological management technology. The system includes a data acquisition and fusion module, an intelligent zoning and feature mining module, a short-term forecasting and risk assessment module, and a visualization and focusing module. The data acquisition and fusion module collects and fuses environmental data from meteorological radar and weather stations, as well as GIS maps. The intelligent zoning and feature mining module divides the data into grids based on the GIS map, analyzes the rainfall synergy between grids based on historical environmental data to construct analysis groups, and selects and constructs an impact set for each analysis group. The short-term forecasting and risk assessment module fits near-term precipitation intensity relationships and short-term precipitation intensity relationships based on the impact sets of each analysis group. The visualization and focusing module calculates and outputs the near-term and short-term predicted precipitation intensities for each analysis group based on the relationships, identifies key areas of concern using topographic and hydrological models, and enhances visualization.

Description

Technical Field

[0001] This invention relates to the field of meteorological management technology, specifically to a method and system for short-term intelligent numerical weather prediction based on environmental perception. Background Technology

[0002] Short-duration heavy rainfall, thunderstorms, and strong winds are characterized by their sudden onset and high destructive potential, posing a severe threat to urban operations and public safety. Achieving high-precision, high-timeliness short-term and nowcasting, and transforming it into intuitive disaster risk warnings, is a core requirement and a major technological challenge in modern meteorological disaster prevention and mitigation.

[0003] Currently, existing technological systems have certain shortcomings. On the one hand, fixed forecast zones fail to adapt to the dynamic evolution and regional characteristics of weather systems. Traditional forecasts are typically based on fixed administrative regions or static geographic grids, which do not adequately reflect the inherent characteristics of severe convective weather systems—rapid movement and dynamically changing impact areas. Existing forecast models often perform averaging calculations within fixed areas with inconsistent physical processes, failing to accurately characterize localized and sudden convective cells, resulting in forecast bias and intensity distortion. On the other hand, the selection of forecast factors is coarse, leading to weak model interpretability and localization adaptability. Mainstream machine learning forecast models often employ a unified, empirical set of forecast factors, ignoring the significant differences that key driving factors for heavy precipitation processes may vary under different underlying surfaces, topography, and climate backgrounds. This fixed factor scheme not only reduces the physical interpretability of the model but also makes it difficult for the model to achieve accurate localization adaptability in special scenarios such as complex terrain areas. These bottlenecks collectively restrict the accuracy, timeliness, and operational efficiency of early warning systems. Summary of the Invention

[0004] The purpose of this invention is to provide a short-term intelligent numerical weather prediction method and system based on environmental perception, so as to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides a short-term intelligent numerical weather prediction method based on environmental perception, comprising:

[0006] S100: Collect environmental data and map information for the designated airspace.

[0007] Environmental data includes various monitoring indicators and precipitation intensity distribution data, as well as historical rainfall event records and TS scores for each rainfall event.

[0008] Forecast score (TS) is a method used to quantitatively test the accuracy of forecasts. Its purpose is to objectively evaluate forecast quality and improve forecasting techniques by comparing different methods. It involves comprehensive research in meteorology, mathematical statistics, and information theory.

[0009] Monitoring indicators refer to a series of physical parameters acquired by meteorological radar and automatic weather stations to characterize the atmospheric thermal, dynamic and water vapor states in order to describe and predict weather systems. Map information is a GIS map that includes topographic elevation information.

[0010] Construct a unified, high-quality data foundation for the forecasting system. Solve the problems of real-time access, standardization, and fusion of multi-source heterogeneous meteorological data, providing a spatiotemporally continuous and physically consistent initial field for subsequent analysis.

[0011] It overcomes the limitations of single observation methods and provides rich and reliable training and input data for machine learning models.

[0012] S200. Divide the specified airspace into grids based on the GIS map, and analyze the synergy of precipitation intensity changes between grids based on historical environmental data. Merge grids to construct analysis groups, and select and construct an impact set for each analysis group. Specifically, this includes:

[0013] S201. Mark the area of ​​the specified airspace covered by the GIS map, and establish a grid array in the area according to the preset two-dimensional geometric segmentation rules to divide it into uniform grids.

[0014] S202. Analyze all rainfall events contained in the rainfall event records and extract the sequence curve of precipitation intensity changing with time for each grid during each rainfall event.

[0015] S203. Calculate the Pearson correlation coefficient between the precipitation intensity change curves of any two grids, and use it as the synergy coefficient to characterize the synergy of the precipitation intensity changes between the two.

[0016] S204. Calculate the synergy coefficient between any two grids across all rainfall events, and use the average value as the synergy index between the two grids. Associate grid pairs whose synergy index is not less than a preset threshold.

[0017] S205. The community detection algorithm is adopted to merge interconnected grids into an analysis group based on the association relationship. Grids in the same analysis group are not necessarily adjacent in space.

[0018] By calculating the Pearson correlation coefficient of historical rainfall curves between grids, it is possible to identify areas that are affected by topography and experience both rain and sunshine, even if they are geographically discontinuous.

[0019] Subsequently, the community discovery algorithm clustered these highly correlated grids into analysis groups with clear physical meaning.

[0020] S206. From the rainfall event records, select rainfall events with TS scores higher than a preset threshold as valid events and count their number. During the extraction of each valid event, the changes in various monitoring indicators and precipitation intensity within each analysis group were analyzed.

[0021] S207. Establish precipitation intensity curves and measured value curves for various monitoring indicators for each analysis group under each valid event. After comparing the differences in the curves, select several monitoring indicators to construct the impact set for each analysis group. Specifically, this includes:

[0022] S2071, Extraction and Analysis Group In valid events Duration period Within the region, the measured values ​​of various monitoring indicators and the changes in precipitation intensity are observed.

[0023] S2072, Calculate the analysis groups at consecutive identical time points respectively. The average precipitation intensity across all grids, and the average measured values ​​of the same monitoring index, are calculated, along with the total number of all monitoring indexes. .

[0024] S2073, Analysis of Duration Period Within the region, the changes in the average precipitation intensity and the average measured values ​​of various monitoring indicators provide the basis for the analysis group. In valid events Establish a precipitation intensity curve and A measured numerical curve.

[0025] S2074, respectively, analysis group Precipitation intensity curves and measurement numerical curves were established for each effective event, and uniform settings were applied within the duration of each effective event. A specific point in time.

[0026] S2075. Analyze the average precipitation intensity at each time point, and the monitoring indicators. The average value of the measured values ​​was used to calculate the analytical group. monitoring indicators The contribution index. Specifically, it includes:

[0027] S2075-1, Valid Events Arrange all time points in chronological order and calculate the remaining time points excluding the first time point. Rate of change of precipitation intensity at each time point The formula is: .

[0028] in, and The first The and the first The average precipitation intensity at each time point in the precipitation intensity curve; .

[0029] S2075-2, Calculate the remaining time points excluding the first time point. Monitoring indicators at each time point rate of change The formula is: .

[0030] in, and The first The and the first At each time point, in the monitoring indicators The average value of the measured values ​​in the curve.

[0031] S2075-3, Through the formula: Calculate the difference at each time point And then according to The difference at each time point Calculate the standard deviation as a valid event. The fluctuation coefficient.

[0032] S2075-4. Calculate the fluctuation coefficient for each valid event, and the remaining fluctuation coefficients excluding the first time point. At each time point, the rate of change of precipitation intensity and monitoring indicators Rate of change.

[0033] Therefore, the monitoring indicators are calculated. Contribution Index The calculation formula is:

[0034] ;

[0035] In the formula, and These are preset constants, For the first The volatility coefficient of a valid event; For the first The first valid event The difference between each time point.

[0036] For the first The first valid event Rate of change of precipitation intensity at each time point For the first The first valid event Monitoring indicators at each time point Rate of change.

[0037] For a sign function, when and When they are the same sign, The value is 1. and When any is zero, The value is 0. and When the signs are different, The value is -2.

[0038] When constructing the impact set, the contribution index not only measures the synchronicity between the indicator and rainfall changes, but also assesses the stability of the relationship through the fluctuation coefficient and the sign function. We use weighted differentiation to distinguish between positive, negative, and ineffective relationships, ensuring that the selected indicators are both sensitive and reliable.

[0039] S2076, Calculate and analyze the groups respectively The contribution index of each monitoring indicator is calculated. All monitoring indicators with contribution indices greater than a preset threshold are selected to construct an impact set. Monitoring indicators are then selected and impact sets are constructed for each analysis group.

[0040] This enables dynamic and refined zoning of forecast objects and data-driven selection of forecast factors. It breaks away from the rigid model of traditional forecasting based on administrative regions or fixed geographical areas, allowing forecast units to match the range of meteorological activity.

[0041] At the same time, a set of the most relevant forecast indicators is tailored for each region, abandoning fixed factor combinations, which significantly improves the physical interpretability and localized forecasting potential of the model.

[0042] S300. For each analysis group, based on the historical environmental data of monitoring indicators and precipitation intensity in its impact set, fit the relationships between near-term precipitation intensity and short-term precipitation intensity to obtain the following:

[0043] S301, Preset Proximity Duration and short duration Obtain the duration of each valid event. and end time Time period As a nearby reference period, the time period Used as a short-term reference period.

[0044] in, ; = ; ; = ;

[0045] S302, uniformly set the adjacent reference time period and short-term reference time period for each valid event. At each time point, the average measured values ​​of the monitoring indicators within the influence set of each analysis group, as well as the time delay duration, are analyzed. and duration The average precipitation intensity at that time.

[0046] S303. Each analysis group used the monitoring indicators at each time point as independent variables and the precipitation intensity at the corresponding time delay as the dependent variable to fit the relationships between near-term precipitation intensity and short-term precipitation intensity. Specifically, this included:

[0047] S3031, Analysis Group The average measured values ​​of each monitoring indicator at the same time point are used as independent variables, with the time delay at that time point being... The average precipitation intensity is used as the dependent variable.

[0048] S3032. At each time point, all independent and dependent variables are packaged into samples, and the LightGBM machine learning algorithm is used to analyze these samples. Training was performed using a sample set, and the fitting was used to obtain the analysis group. The relationship between the intensity of nearby precipitation.

[0049] S3033, and then the analysis group The average measured values ​​of each monitoring indicator at the same time point are used as independent variables, with the time delay at that time point being... The average precipitation intensity is used as the dependent variable.

[0050] S3034. At each time point, all independent and dependent variables are packaged into samples, and the LightGBM machine learning algorithm is used to analyze these samples. Training was performed using a sample set, and the fitting was used to obtain the analysis group. The relationship between short-term precipitation intensity and precipitation intensity.

[0051] Tree models like LightGBM can automatically capture the complex nonlinear relationship between monitoring indicators and future rainfall intensity with high efficiency.

[0052] The formulas for fitting the near-term precipitation intensity and the short-term precipitation intensity are respectively for each analysis group.

[0053] Construct high-precision machine learning forecasting models with time-series and regional characteristics to achieve a leap from weather forecasting to risk warning.

[0054] Models were established for different dominant physical processes in short-term and short-term forecasts, which improved the forecast accuracy for each time period.

[0055] S400: Based on the relational formula, calculate and output the near-term and short-term predicted precipitation intensity for each analysis group. Combined with topographic factors in the GIS map, a topographic-hydrological model is used to identify key areas of interest and enhance their display. Specifically, this includes:

[0056] S401. Real-time acquisition of the average measured values ​​of all monitoring indicators within the influence set of each analysis group at the current moment, using these values ​​as input, and substituting them into the two relational formulas of the corresponding analysis group to calculate and output the near-term and short-term predicted precipitation intensity for that analysis group. Generate a gridded forecast field covering all analysis groups.

[0057] S402. Analyze the near-term and short-term predicted precipitation intensity, and use the SCS-CN hydrological model to calculate the surface runoff coefficient and runoff volume for each grid. Extract the topographic slope and aspect data for each grid based on the GIS map, simulate the confluence path of surface runoff, use the D8 algorithm to determine the flow direction of each grid, and calculate its upstream catchment area.

[0058] During the risk identification phase, the SCS-CN model calculates the runoff coefficient based on soil, surface cover, and pre-existing humidity, while the D8 algorithm determines the flow direction of each grid based on the digital elevation model, simulating the confluence network.

[0059] By combining these two methods, it is possible to accurately calculate which low-lying areas will become hotspots for water accumulation under specific rainfall conditions.

[0060] S403, superimposed runoff and catchment area, identifies the period in the near term and short duration Within the forecasting system, grids whose accumulated runoff exceeds a preset safety threshold are designated as areas of high concern. These areas of high concern are graphically highlighted on the forecasting visualization platform, using methods including highlighting boundaries and overlaying special color layers.

[0061] When a key focus area is identified, the following actions are automatically triggered:

[0062] Increase the inference frequency of machine learning intelligent forecasting models for areas with high predicted precipitation intensity near key areas of concern.

[0063] The command radar performs more frequent volume scans on areas near the key area of ​​interest where the predicted precipitation intensity is high.

[0064] A local rapid numerical model with higher spatial resolution will be launched for areas near key areas of concern with high predicted precipitation intensity.

[0065] By coupling hydro-topographic models, the predicted rainfall is transformed into waterlogging risk information, making the forecast products more valuable for decision support and achieving a service-oriented upgrade.

[0066] The present invention also provides a short-term intelligent numerical weather prediction system based on environmental perception, including a data acquisition and fusion module, an intelligent zoning and feature mining module, a short-term forecast and risk assessment module, and a visualization and focusing module.

[0067] The data acquisition and fusion module is used to collect and fuse environmental data from weather radar, automatic weather stations, and GIS map information.

[0068] The system collects real-time data from meteorological radar, temperature, pressure, humidity, wind, and precipitation intensity from automatic weather stations via data interfaces, and simultaneously accesses a GIS map database containing topographic elevation information.

[0069] The raw observation data is subjected to quality control and standardization, and the station data is gridded using methods such as Kriging interpolation. Finally, it is integrated with radar and other products to generate a multi-element real-time analysis field with high spatiotemporal resolution.

[0070] It forms the data foundation for the forecasting system. Through real-time fusion and quality control, it provides high-quality, uniformly formatted, and spatially continuous real-time data products, providing reliable and rich input for subsequent intelligent zoning and accurate forecasting. It also solves the problem of incomplete initial field information caused by heterogeneous multi-source data and sparse observation stations.

[0071] The intelligent zoning and feature mining module is used to divide the grid according to the GIS map, analyze the rainfall synergy between the grids based on historical environmental data to construct analysis groups, and select and construct an impact set for each analysis group.

[0072] First, the target area is divided into two-dimensional grids based on the GIS map.

[0073] Secondly, the Pearson correlation coefficient of precipitation intensity variation curves between any two grids in historical rainfall events is calculated, and based on the synergy index threshold, a community detection algorithm is used to merge grids with high synergy into analysis groups.

[0074] Finally, for each analysis group, from the valid historical events with high TS scores, the synchronicity between monitoring indicators and changes in precipitation intensity was analyzed, and quantitative evaluation was carried out using the contribution index formula. The monitoring indicators that are most indicative of rainfall in the region were then selected to construct an impact set.

[0075] Breaking away from the limitations of traditional fixed geographical divisions, the analysis units are dynamically divided according to the actual impact range of weather systems, making the forecast model closer to physical reality.

[0076] Meanwhile, by using a data-driven approach to select the most relevant forecasting factors for each region, the forecasting model was localized and customized, thereby improving the interpretability and forecasting accuracy of the model from the source.

[0077] The short-term forecast and risk assessment module is used to fit the near-term precipitation intensity relationship and the short-term precipitation intensity relationship based on the impact set of each analysis group.

[0078] In terms of forecasting, near-term and short-term forecast models were constructed for each analysis group:

[0079] Using the monitoring indicators in the impact set of the analysis group as independent variables, respectively, with delay and The precipitation intensity after the duration is the dependent variable, and the LightGBM machine learning algorithm is used to fit two independent prediction relationships.

[0080] In terms of risk assessment, the predicted precipitation intensity is used in conjunction with GIS topographic data to drive the SCS-CN hydrological model to calculate surface runoff, and the D8 confluence algorithm is used to simulate runoff accumulation, thereby intelligently identifying grids that may accumulate water in the short-term future and designating them as key areas of concern.

[0081] This enables a leap from meteorological element forecasting to disaster risk early warning. Its time-series machine learning model can more accurately predict rainfall in different lead times.

[0082] By introducing topographic and hydrological models for secondary analysis, the output is no longer an abstract amount of rainfall, but a direct representation of the location of water accumulation risk. This enables the forecast to be implemented and applied, directly serving disaster prevention and mitigation decision-making.

[0083] The visualization and focusing module is used to calculate and output the near-term and short-term precipitation intensity of each analysis group based on the relational formula, identify key areas of concern by combining topographic and hydrological models, and enhance visualization.

[0084] First, substitute the real-time monitoring data into the forecast formulas of each analysis group to calculate and generate a gridded precipitation intensity forecast field covering the entire region.

[0085] The coordinates of the key areas of interest are then transmitted to the visualization platform, where they are graphically highlighted on the electronic map by methods such as highlighting boundaries and overlaying special color layers.

[0086] Transform complex numerical forecast results and risk analysis conclusions into intuitive and easy-to-understand graphical products.

[0087] By visually enhancing high-risk areas, the readability and operational usability of forecast products are greatly improved. This enables forecasters to quickly focus their attention, achieve targeted focus of early warning resources, and ensure that critical information is efficiently identified and responded to.

[0088] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0089] Intelligent data fusion and dynamic meteorological zoning: This technical solution constructs analysis groups with clear physical meaning by fusing multi-source data and merging dynamic grids based on historical synergy. This overcomes the problems of rigid data utilization, fixed forecast zoning and inability to adapt to the dynamic evolution of weather systems in existing technologies, and achieves adaptive matching between forecast units and the actual weather impact range.

[0090] Data-driven localized forecast factor selection: This technical solution uses contribution index to quantitatively select the influence set specific to each analysis group, achieving accurate and adaptive selection of forecast factors. It solves the shortcomings of existing technologies such as coarse factor selection, weak model interpretability and localization adaptability, and builds the most indicative personalized forecast model for each analysis group.

[0091] Seamless coupling of forecasting and risk assessment: The technical solution deeply integrates machine learning forecasting models with topographic and hydrological models (SCS-CN, D8 algorithm) to realize an automated process from predicting rainfall to identifying waterlogging risk areas. This overcomes the bottleneck of existing technologies where forecast products are disconnected from risk assessment and lack practicality, and directly produces impact forecasts for disaster prevention decision-making. Attached Figure Description

[0092] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0093] Figure 1 This is a flowchart illustrating the short-term intelligent numerical weather prediction method based on environmental perception according to the present invention.

[0094] Figure 2 This is a schematic diagram of the structure of the short-term intelligent numerical weather prediction system based on environmental perception according to the present invention. Detailed Implementation

[0095] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0096] Example 1: Please refer to Figure 1 This invention provides a short-term intelligent numerical weather prediction method based on environmental perception, comprising:

[0097] S100: Collect environmental data and map information for the designated airspace.

[0098] In the specific implementation process, environmental data includes various monitoring indicators and precipitation intensity distribution data, as well as historical rainfall event records and TS scores for each rainfall event.

[0099] Forecast score (TS) is a method used to quantitatively test the accuracy of forecasts. Its purpose is to objectively evaluate forecast quality and improve forecasting techniques by comparing different methods. It involves comprehensive research in meteorology, mathematical statistics, and information theory.

[0100] Monitoring indicators refer to a series of physical parameters acquired by meteorological radar and automatic weather stations to characterize the atmospheric thermal, dynamic and water vapor states in order to describe and predict weather systems. Map information is a GIS map that includes topographic elevation information.

[0101] Construct a unified, high-quality data foundation for the forecasting system. Solve the problems of real-time access, standardization, and fusion of multi-source heterogeneous meteorological data, providing a spatiotemporally continuous and physically consistent initial field for subsequent analysis.

[0102] It overcomes the limitations of single observation methods (such as sparse stations and radar blind spots) and provides rich and reliable training and input data for machine learning models.

[0103] S200. Divide the specified airspace into grids based on the GIS map, and analyze the synergy of precipitation intensity changes between grids based on historical environmental data. Merge grids to construct analysis groups, and select and construct an impact set for each analysis group. Specifically, this includes:

[0104] S201. Mark the area of ​​the specified airspace covered by the GIS map, and establish a grid array in the area according to the preset two-dimensional geometric segmentation rules to divide it into uniform grids.

[0105] S202. Analyze all rainfall events contained in the rainfall event records and extract the sequence curve of precipitation intensity changing with time for each grid during each rainfall event.

[0106] S203. Calculate the Pearson correlation coefficient between the precipitation intensity change curves of any two grids, and use it as the synergy coefficient to characterize the synergy of the precipitation intensity changes between the two.

[0107] S204. Calculate the synergy coefficient between any two grids across all rainfall events, and use the average value as the synergy index between the two grids. Associate grid pairs whose synergy index is not less than a preset threshold.

[0108] S205. The community detection algorithm is adopted to merge interconnected grids into an analysis group based on the association relationship. Grids in the same analysis group are not necessarily adjacent in space.

[0109] In practice, by calculating the Pearson correlation coefficient of historical rainfall curves between grids, it is possible to identify areas that are affected by topography and experience both rain and sunshine, even if they are geographically discontinuous (e.g., two leeward slope areas affected by the topographic uplift of the same mountain range).

[0110] Subsequently, community discovery algorithms (such as the Louvain algorithm) cluster these highly correlated grids into analysis groups with clear physical meaning.

[0111] S206. From the rainfall event records, select rainfall events with TS scores higher than a preset threshold as valid events and count their number. During the extraction of each valid event, the changes in various monitoring indicators and precipitation intensity within each analysis group were analyzed.

[0112] S207. Establish precipitation intensity curves and measured value curves for various monitoring indicators for each analysis group under each valid event. After comparing the differences in the curves, select several monitoring indicators to construct the impact set for each analysis group. Specifically, this includes:

[0113] S2071, Extraction and Analysis Group In valid events Duration period Within the region, the measured values ​​of various monitoring indicators and the changes in precipitation intensity are observed.

[0114] S2072, Calculate the analysis groups at consecutive identical time points respectively. The average precipitation intensity across all grids, and the average measured values ​​of the same monitoring index, are calculated, along with the total number of all monitoring indexes. .

[0115] S2073, Analysis of Duration Period Within the region, the changes in the average precipitation intensity and the average measured values ​​of various monitoring indicators provide the basis for the analysis group. In valid events Establish a precipitation intensity curve and A measured numerical curve.

[0116] S2074, respectively, analysis group Precipitation intensity curves and measurement numerical curves were established for each effective event, and uniform settings were applied within the duration of each effective event. A specific point in time.

[0117] S2075. Analyze the average precipitation intensity at each time point, and the monitoring indicators. The average value of the measured values ​​was used to calculate the analytical group. monitoring indicators The contribution index. Specifically, it includes:

[0118] S2075-1, Valid Events Arrange all time points in chronological order and calculate the remaining time points excluding the first time point. Rate of change of precipitation intensity at each time point The formula is: .

[0119] in, and The first The and the first The average precipitation intensity at each time point in the precipitation intensity curve; .

[0120] S2075-2, Calculate the remaining time points excluding the first time point. Monitoring indicators at each time point rate of change The formula is: .

[0121] in, and The first The and the first At each time point, in the monitoring indicators The average value of the measured values ​​in the curve.

[0122] S2075-3, Through the formula: Calculate the difference at each time point And then according to The difference at each time point Calculate the standard deviation as a valid event. The fluctuation coefficient.

[0123] S2075-4. Calculate the fluctuation coefficient for each valid event, and the remaining fluctuation coefficients excluding the first time point. At each time point, the rate of change of precipitation intensity and monitoring indicators Rate of change.

[0124] Therefore, the monitoring indicators are calculated. Contribution Index The calculation formula is:

[0125] ;

[0126] In the formula, and These are preset constants, For the first The volatility coefficient of a valid event; For the first The first valid event The difference between each time point.

[0127] For the first The first valid event Rate of change of precipitation intensity at each time point For the first The first valid event Monitoring indicators at each time point Rate of change.

[0128] For a sign function, when and When they are the same sign, The value is 1. and When any is zero, The value is 0. and When the signs are different, The value is -2.

[0129] When constructing the impact set, the contribution index not only measures the synchronicity between the indicator and rainfall changes (through the difference in the rate of change) but also... Furthermore, the stability of the relationship is assessed through the volatility coefficient and through the sign function. We use weighted differentiation to distinguish between positive, negative, and ineffective relationships, ensuring that the selected indicators are both sensitive and reliable.

[0130] S2076, Calculate and analyze the groups respectively The contribution index of each monitoring indicator is calculated. All monitoring indicators with contribution indices greater than a preset threshold are selected to construct an impact set. Monitoring indicators are then selected and impact sets are constructed for each analysis group.

[0131] This enables dynamic and refined zoning of forecast objects and data-driven selection of forecast factors. It breaks away from the rigid model of traditional forecasting based on administrative regions or fixed geographical areas, allowing forecast units to match the range of meteorological activity.

[0132] At the same time, a set of the most relevant forecast indicators (influence set) is tailored for each region, abandoning fixed factor combination schemes, which significantly improves the physical interpretability and localized forecasting potential of the model.

[0133] S300. For each analysis group, based on the historical environmental data of monitoring indicators and precipitation intensity in its impact set, fit the relationships between near-term precipitation intensity and short-term precipitation intensity to obtain the following:

[0134] S301, Preset Proximity Duration and short duration Obtain the duration of each valid event. and end time Time period As a nearby reference period, the time period Used as a short-term reference period.

[0135] in, ; = ; ; = ;

[0136] S302, uniformly set the adjacent reference time period and short-term reference time period for each valid event. At each time point, the average measured values ​​of the monitoring indicators within the influence set of each analysis group, as well as the time delay duration, are analyzed. and duration The average precipitation intensity at that time.

[0137] S303. Each analysis group used the monitoring indicators at each time point as independent variables and the precipitation intensity at the corresponding time delay as the dependent variable to fit the relationships between near-term precipitation intensity and short-term precipitation intensity. Specifically, this included:

[0138] S3031, Analysis Group The average measured values ​​of each monitoring indicator at the same time point are used as independent variables, with the time delay at that time point being... The average precipitation intensity is used as the dependent variable.

[0139] S3032. At each time point, all independent and dependent variables are packaged into samples, and the LightGBM machine learning algorithm is used to analyze these samples. Training was performed using a sample set, and the fitting was used to obtain the analysis group. The relationship between the intensity of nearby precipitation.

[0140] S3033, and then the analysis group The average measured values ​​of each monitoring indicator at the same time point are used as independent variables, with the time delay at that time point being... The average precipitation intensity is used as the dependent variable.

[0141] S3034. At each time point, all independent and dependent variables are packaged into samples, and the LightGBM machine learning algorithm is used to analyze these samples. Training was performed using a sample set, and the fitting was used to obtain the analysis group. The relationship between short-term precipitation intensity and precipitation intensity.

[0142] In practical implementation, tree models such as LightGBM can be used to automatically capture the complex nonlinear relationship between monitoring indicators and future rainfall intensity, and are highly efficient.

[0143] The formulas for fitting the near-term precipitation intensity and the short-term precipitation intensity are respectively for each analysis group.

[0144] Construct high-precision machine learning forecasting models with time-series and regional characteristics to achieve a leap from weather forecasting to risk warning.

[0145] Models were established for different dominant physical processes in short-term (0-2 hours) and short-term (2-12 hours) forecasts, which improved the forecast accuracy for each time period.

[0146] S400: Based on the relational formula, calculate and output the near-term and short-term predicted precipitation intensity for each analysis group. Combined with topographic factors in the GIS map, a topographic-hydrological model is used to identify key areas of interest and enhance their display. Specifically, this includes:

[0147] S401. Real-time acquisition of the average measured values ​​of all monitoring indicators within the influence set of each analysis group at the current moment, using these values ​​as input, and substituting them into the two relational formulas of the corresponding analysis group to calculate and output the near-term and short-term predicted precipitation intensity for that analysis group. Generate a gridded forecast field covering all analysis groups.

[0148] S402. Analyze the near-term and short-term predicted precipitation intensity, and use the SCS-CN hydrological model to calculate the surface runoff coefficient and runoff volume for each grid. Extract the topographic slope and aspect data for each grid based on the GIS map, simulate the confluence path of surface runoff, use the D8 algorithm to determine the flow direction of each grid, and calculate its upstream catchment area.

[0149] In the specific implementation process, during the risk identification phase, the SCS-CN model calculates the runoff coefficient based on soil, surface cover, and pre-existing humidity, while the D8 algorithm determines the water flow direction of each grid based on the digital elevation model, simulating the confluence network.

[0150] By combining these two methods, it is possible to accurately calculate which low-lying areas will become hotspots for water accumulation under specific rainfall conditions.

[0151] S403, superimposed runoff and catchment area, identifies the period in the near term and short duration Within the forecasting system, grids whose accumulated runoff exceeds a preset safety threshold are designated as areas of high concern. These areas of high concern are graphically highlighted on the forecasting visualization platform, using methods including highlighting boundaries and overlaying special color layers.

[0152] In the specific implementation process, when a key focus area is designated, the following actions are automatically triggered:

[0153] Increase the inference frequency of machine learning intelligent forecasting models for areas with high predicted precipitation intensity near key areas of concern.

[0154] The command radar performs more frequent volume scans on areas near the key area of ​​interest where the predicted precipitation intensity is high.

[0155] For areas near key areas of concern with high predicted precipitation intensity, a local rapid numerical model with a higher spatial resolution (e.g., 1 km instead of 5 km) will be launched.

[0156] By coupling hydro-topographic models, the predicted rainfall is transformed into waterlogging risk information, making the forecast products more valuable for decision support and achieving a service-oriented upgrade.

[0157] Example 2: Please refer to Figure 2 The present invention also provides a short-term intelligent numerical weather forecasting system based on environmental perception, including a data acquisition and fusion module, an intelligent zoning and feature mining module, a short-term forecasting and risk assessment module, and a visualization and focusing module.

[0158] The data acquisition and fusion module is used to collect and fuse environmental data from weather radar, automatic weather stations, and GIS map information.

[0159] In the specific implementation process, the data interface is used to collect meteorological radar base data, monitoring indicators such as temperature, pressure, humidity, and wind uploaded by automatic weather stations, as well as precipitation intensity data, and is simultaneously connected to a GIS map database containing topographic elevation information.

[0160] The raw observation data is subjected to quality control and standardization, and the station data is gridded using methods such as Kriging interpolation. Finally, it is integrated with radar and other products to generate a multi-element real-time analysis field with high spatiotemporal resolution.

[0161] It forms the data foundation for the forecasting system. Through real-time fusion and quality control, it provides high-quality, uniformly formatted, and spatially continuous real-time data products, providing reliable and rich input for subsequent intelligent zoning and accurate forecasting. It also solves the problem of incomplete initial field information caused by heterogeneous multi-source data and sparse observation stations.

[0162] The intelligent zoning and feature mining module is used to divide the grid according to the GIS map, analyze the rainfall synergy between the grids based on historical environmental data to construct analysis groups, and select and construct an impact set for each analysis group.

[0163] In the specific implementation process, the target area is first divided into two-dimensional grids based on the GIS map.

[0164] Secondly, the Pearson correlation coefficient (synergy coefficient) of precipitation intensity change curves between any two grids in historical rainfall events is calculated, and based on the synergy index threshold, the community detection algorithm is used to merge grids with high synergy (which may not be adjacent) into analysis groups.

[0165] Finally, for each analysis group, from the valid historical events with high TS scores, the synchronicity between monitoring indicators and changes in precipitation intensity was analyzed, and quantitative evaluation was carried out using the contribution index formula. The monitoring indicators that are most indicative of rainfall in the region were then selected to construct an impact set.

[0166] Breaking away from the limitations of traditional fixed geographical divisions, the analysis units are dynamically divided according to the actual impact range of weather systems, making the forecast model closer to physical reality.

[0167] Meanwhile, by using a data-driven approach to select the most relevant forecast factors (influence set) for each region, the forecast model was localized and customized, improving the interpretability and forecast accuracy of the model from the source.

[0168] The short-term forecast and risk assessment module is used to fit the near-term precipitation intensity relationship and the short-term precipitation intensity relationship based on the impact set of each analysis group.

[0169] In the specific implementation process, for forecasting, near-term (e.g., 0-2 hours) and short-term (e.g., 2-12 hours) forecasting models are constructed for each analysis group:

[0170] Using the monitoring indicators in the impact set of the analysis group as independent variables, respectively, with delay and The precipitation intensity after the duration is the dependent variable, and the LightGBM machine learning algorithm is used to fit two independent prediction relationships.

[0171] In terms of risk assessment, the predicted precipitation intensity is used in conjunction with GIS topographic data to drive the SCS-CN hydrological model to calculate surface runoff, and the D8 confluence algorithm is used to simulate runoff accumulation, thereby intelligently identifying grids that may accumulate water in the short-term future and designating them as key areas of concern.

[0172] This enables a leap from meteorological element forecasting to disaster risk early warning. Its time-series machine learning model can more accurately predict rainfall in different lead times.

[0173] By introducing topographic and hydrological models for secondary analysis, the output is no longer an abstract amount of rainfall, but a direct representation of the location of water accumulation risk. This enables the forecast to be implemented and applied, directly serving disaster prevention and mitigation decision-making.

[0174] The visualization and focusing module is used to calculate and output the near-term and short-term precipitation intensity of each analysis group based on the relational formula, identify key areas of concern by combining topographic and hydrological models, and enhance visualization.

[0175] In the specific implementation process, the real-time monitoring data is first substituted into the forecast formula of each analysis group to calculate and generate a gridded precipitation intensity forecast field covering the entire region.

[0176] The coordinates of the key areas of interest are then transmitted to the visualization platform, where they are graphically highlighted on the electronic map by methods such as highlighting boundaries and overlaying special color layers.

[0177] Transform complex numerical forecast results and risk analysis conclusions into intuitive and easy-to-understand graphical products.

[0178] By visually enhancing high-risk areas, the readability and operational usability of forecast products are greatly improved. This enables forecasters to quickly focus their attention, achieve targeted focus of early warning resources, and ensure that critical information is efficiently identified and responded to.

[0179] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0180] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An environmental perception based nowcasting intelligent numerical weather prediction method, characterized in that: The method includes: S100: Collect environmental data and map information for the designated airspace; S200. Divide the specified airspace into grids based on the GIS map, analyze the synergy of precipitation intensity changes between grids based on historical environmental data; merge grids to construct analysis groups, and select and construct an impact set for each analysis group; S300. For each analysis group, based on the historical environmental data of monitoring indicators and precipitation intensity in its impact set, fit the relationship between near-term precipitation intensity and short-term precipitation intensity. S400: Based on the relational formula, calculate and output the near-term and short-term precipitation intensity of each analysis group. Combined with the topographic factors in the GIS map, use the topographic and hydrological model to identify key areas of concern and enhance their display.

2. The short-term intelligent numerical weather prediction method based on environmental perception according to claim 1, characterized in that: In S100, environmental data includes various monitoring indicators and precipitation intensity distribution data, as well as historical rainfall event records and TS scores for each rainfall event; Monitoring indicators refer to a series of physical parameters acquired by meteorological radar and automatic weather stations to characterize the atmospheric thermal, dynamic and water vapor states in order to describe and predict weather systems. Map information is a GIS map that includes topographic elevation information. 3.The short-term intelligent numerical weather prediction method based on environment perception according to claim 2, wherein: S200 includes: S201. Mark the area of ​​the specified airspace covered by the GIS map, and establish a grid array in the area according to the preset two-dimensional geometric segmentation rules to divide it into uniform grids; S202. Analyze all rainfall events contained in the rainfall event records and extract the sequence curve of precipitation intensity changing with time for each grid during each rainfall event; S203. Calculate the Pearson correlation coefficient between the precipitation intensity change curves of any two grids, and use it as the synergy coefficient characterizing the synergy of the precipitation intensity changes between the two grids. S204. Calculate the synergy coefficient between any two grids in all rainfall events, and calculate the average value as the synergy index between the two grids; associate grid pairs whose synergy index is not less than a preset threshold. S205. The community detection algorithm is adopted to merge interconnected grids into an analysis group based on the association relationship. Grids in the same analysis group are not necessarily adjacent in space. S206、From the rainfall event records, filter out the rainfall events with TS score higher than a preset threshold as effective events and count the number During each effective event, extract the changes of each monitoring index and the changes of rainfall intensity in each analysis group; S207. Establish precipitation intensity curves and measured value curves of various monitoring indicators for each analysis group under each valid event; after comparing the differences between the curves, select several monitoring indicators to construct the influence set of each analysis group.

4. The short-term intelligent numerical weather prediction method based on environmental perception according to claim 3, characterized in that: S207 includes: S2071, Extraction and Analysis Group In valid events Duration period Within, the measured values ​​of various monitoring indicators and the changes in precipitation intensity; S2072, Calculate the analysis groups at consecutive identical time points respectively. The average precipitation intensity across all grids, and the average measured values ​​of the same monitoring index, are calculated, along with the total number of all monitoring indexes. ; S2073, Analysis of Duration Period Within the region, the changes in the average precipitation intensity and the average measured values ​​of various monitoring indicators provide the basis for the analysis group. In valid events Establish a precipitation intensity curve and A measured numerical curve; S2074, respectively, analysis group Precipitation intensity curves and measurement numerical curves were established for each effective event, and uniform settings were applied within the duration of each effective event. A point in time; S2075. Analyze the average precipitation intensity at each time point, and the monitoring indicators. The average value of the measured values ​​was used to calculate the analytical group. monitoring indicators Contribution index; S2076, Calculate and analyze the groups respectively The contribution index of each monitoring indicator is calculated; all monitoring indicators with contribution indices greater than a preset threshold are selected to construct an impact set; monitoring indicators are selected for each analysis group and an impact set is constructed.

5. The short-term intelligent numerical weather prediction method based on environmental perception according to claim 4, characterized in that: S2075 includes: S2075-1, Valid Events Arrange all time points in chronological order and calculate the remaining time points excluding the first time point. Rate of change of precipitation intensity at each time point The formula is: ; in, and The first The and the first The average precipitation intensity at each time point in the precipitation intensity curve; ; S2075-2, Calculate the remaining time points excluding the first time point. Monitoring indicators at each time point rate of change The formula is: ; in, and The first The and the first At each time point, in the monitoring indicators The average value of the measured values ​​in the curve of measured values; S2075-3, Through the formula: Calculate the difference at each time point And then according to The difference at each time point Calculate the standard deviation as a valid event. The fluctuation coefficient; S2075-4. Calculate the fluctuation coefficient for each valid event, and the remaining fluctuation coefficients excluding the first time point. At each time point, the rate of change of precipitation intensity and monitoring indicators Rate of change; thus calculating monitoring indicators Contribution Index .

6. The short-term intelligent numerical weather prediction method based on environmental perception according to claim 5, characterized in that: Contribution Index The calculation formula is: ； In the formula, and These are preset constants, For the first The volatility coefficient of a valid event; For the first The first valid event The difference between each time point; For the first The first valid event Rate of change of precipitation intensity at each time point For the first The first valid event Monitoring indicators at each time point Rate of change; For a sign function, when and When they are the same sign, The value is 1; when and When any is zero, The value is 0; when and When the signs are different, The value is -2.

7. The short-term intelligent numerical weather prediction method based on environmental perception according to claim 3, characterized in that: The S300 includes: S301, Preset Proximity Duration and short duration Obtain the duration of each valid event. and end time ; Time period As a nearby reference period, the time period As a short-term reference period; in, ; = ; ; = ; S302, uniformly set the adjacent reference time period and short-term reference time period for each valid event. At each time point, the average measured values ​​of the monitoring indicators within the influence set of each analysis group, as well as the time delay duration, are analyzed. and duration The average precipitation intensity at that time; S303. Each analysis group used the monitoring indicators at each time point as independent variables and the precipitation intensity at the corresponding time point delay as dependent variables to fit the relationship between the near-term precipitation intensity and the short-term precipitation intensity.

8. The short-term intelligent numerical weather prediction method based on environmental perception according to claim 7, characterized in that: S303 includes: S3031, Analysis Group The average measured values ​​of each monitoring indicator at the same time point are used as independent variables, with the time delay at that time point being... The average precipitation intensity is used as the dependent variable; S3032. At each time point, all independent and dependent variables are packaged into samples, and the LightGBM machine learning algorithm is used to analyze these samples. Training was performed using a sample set, and the fitting was used to obtain the analysis group. The relationship between near-term precipitation intensity; S3033, and then the analysis group The average measured values ​​of each monitoring indicator at the same time point are used as independent variables, with the time delay at that time point being... The average precipitation intensity is used as the dependent variable; S3034. At each time point, all independent and dependent variables are packaged into samples, and the LightGBM machine learning algorithm is used to analyze these samples. Training was performed using a sample set, and the fitting was used to obtain the analysis group. The relationship between short-duration precipitation intensity; The formulas for fitting the near-term precipitation intensity and the short-term precipitation intensity are respectively for each analysis group.

9. The short-term intelligent numerical weather prediction method based on environmental perception according to claim 7, characterized in that: The S400 includes: S401. Real-time acquisition of the average measured values ​​of various monitoring indicators within the influence set of each analysis group at the current moment, using these values ​​as input, and substituting them into the two relational formulas of the corresponding analysis group to calculate and output the near-term and short-term predicted precipitation intensity of that analysis group; generating a gridded forecast field covering all analysis groups; S402. Analyze the near-term and short-term predicted precipitation intensity, and use the SCS-CN hydrological model to calculate the surface runoff coefficient and runoff volume of each grid. Extract the topographic slope and aspect data of each grid based on the GIS map, simulate the confluence path of surface runoff, use the D8 algorithm to determine the water flow direction of each grid, and calculate its upstream catchment area. S403, superimposed runoff and catchment area, identifies the period in the near term and short duration Within the forecasting system, grids whose accumulated runoff exceeds a preset safety threshold are designated as key areas of concern. On the forecasting visualization platform, these key areas of concern are highlighted graphically, with display methods including highlighting boundaries and overlaying special color layers.

10. A short-term intelligent numerical weather prediction system based on environmental perception, applied to the short-term intelligent numerical weather prediction method based on environmental perception as described in claim 1, characterized in that: The system includes a data acquisition and fusion module, an intelligent zoning and feature mining module, a short-term forecasting and risk assessment module, and a visualization and focusing module; The data acquisition and fusion module is used to collect and fuse environmental data from weather radar, automatic weather stations, and GIS map information; The intelligent partitioning and feature mining module is used to divide the grid according to the GIS map, analyze the rainfall synergy between the grids based on historical environmental data to construct analysis groups, and select and construct an impact set for each analysis group. The short-term forecast and risk assessment module is used to fit the near-term precipitation intensity relationship and the short-term precipitation intensity relationship based on the impact set of each analysis group; The visualization and focusing module is used to calculate and output the near-term and short-term predicted precipitation intensities of each analysis group based on the relational formula, identify key areas of concern in combination with the topographic and hydrological model, and perform visualization enhancement.

Images