Sun and rain forecasting method and system based on complex microtopographic condition

Abstract

The invention relates to the technical field of weather forecast element forecasting, in particular to a weather forecast method and system based on complex microtopography conditions, and the method comprises the steps: obtaining high-resolution NCEP / GFS forecast field data and MODIS satellite earth surface static data; setting a multi-layer nested grid and a parameterization scheme, and generating a WRFOUT numerical weather forecast file; interpolating the forecast data to the position of the weather station by using bilinear interpolation; performing feature engineering by using variational mode decomposition and principal component analysis methods; matching the forecast data and the observation data to serve as a training set of a machine learning model; optimal parameters are found out through Bayesian optimization; and predicting the precipitation of the weather station. The method further comprises the steps of spatial interpolation, terrain correction, icing early warning, risk assessment and the like. According to the method, the accuracy and practicability of weather forecast under the complex micro-topographic condition are remarkably improved, and powerful support is provided for weather forecast and disaster prevention and reduction.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological forecast element forecasting, in particular to a sunny-rainy forecast method and system based on complex microtopography conditions. Background Art

[0002] As an important support for the development of modern society, the accuracy and refinement degree of meteorological forecasting have always been the focus of research. With the development of numerical weather prediction models, especially the wide application of the mesoscale numerical model Weather Research and Forecasting (WRF), the meteorological forecasting ability has been significantly improved. However, under complex terrain conditions, especially in the microtopography environment, traditional numerical prediction models are often difficult to accurately capture the variation characteristics of local meteorological elements due to the limitations of resolution and parameterization schemes. In addition, the existing forecasting methods have problems of low integration efficiency in dealing with multi-source heterogeneous data, and it is difficult to make full use of multi-dimensional information such as satellite remote sensing and ground observations to optimize the forecasting results. Especially in precipitation forecasting, due to the high nonlinearity and uncertainty of the precipitation process, refined forecasting still faces great challenges.

[0003] To reduce the losses caused by icing disasters, it is necessary to accurately predict the icing thickness of conductors, and the occurrence and development of conductor icing are closely related to the changes in meteorological elements. In recent years, a new generation of numerical prediction model - the WRF model has been widely used in meteorological forecasting. However, the simulation effect of the WRF model is affected by terrain, underlying surface, resolution, driving field and physical processes. It is difficult to accurately predict meteorological elements based on a single WRF model, especially in areas with complex terrain, and the simulation accuracy of the WRF model often cannot meet the actual needs. Therefore, it is necessary to correct the simulation results of the WRF model to obtain meteorological data with higher accuracy. Secondly, the means of sunny and rainy forecasts in winter have been continuously updated. In order to better and more accurately predict the sunny and rainy forecasts of regions, first establish the relationship between statistical forecast rainfall and actual rainfall by downscaling to meteorological stations, and finally extend the statistical relationship to grid points. The accuracy of sunny and rainy forecasts has been improved, which also plays a good role in icing prediction. When the existing technology deals with sunny and rainy forecasts under complex micro-topographic conditions, the following deficiencies mainly exist: First, the forecast accuracy of the traditional WRF model in micro-topographic areas is insufficient, and it is difficult to effectively simulate the local circulation and precipitation distribution characteristics caused by the terrain. Second, the existing methods are inefficient in integrating multi-source data and fail to fully utilize high-resolution surface information and real-time observation data to correct the numerical forecast results. Thirdly, in terms of feature engineering and the application of machine learning algorithms, there is a lack of targeted optimization design, and key forecast factors related to micro-topography cannot be effectively extracted and utilized. Finally, the existing forecast systems have a single method in spatial interpolation and terrain correction, and it is difficult to accurately depict the impact of complex terrain on precipitation distribution. These problems lead to the spatio-temporal accuracy and reliability of sunny and rainy forecasts being difficult to meet the growing demand for refined forecasts under complex micro-topographic conditions, especially in the application in fields such as disaster prevention and mitigation and water resource management being restricted. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention lies in the fact that the simulation effect of the WRF model is affected by terrain, underlying surface, resolution, driving field and physical processes, and it is difficult to accurately predict meteorological elements based on a single WRF model, especially in areas with complex terrain, and the simulation accuracy of the WRF model often cannot meet the actual needs.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides a sunny and rainy forecast method based on complex micro-topographic conditions, which includes obtaining NCEP / GFS forecast field data with a preset resolution as the initial field and lateral boundary conditions of the WRF model;

[0008] Obtain the surface static data of terrain, soil data, and vegetation cover with a preset resolution provided by the MODIS satellite;

[0009] Set the grid nesting levels, grid numbers, and horizontal grid resolutions of the WRF model, select the parameterization scheme of the WRF model, and generate a WRFOUT numerical weather forecast file containing various meteorological elements;

[0010] Obtain meteorological grid data and meteorological station data with a preset resolution, as well as the geographical information of the meteorological stations;

[0011] Interpolate the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological stations using the bilinear interpolation method;

[0012] Use the variational mode decomposition and principal component analysis methods to perform feature engineering processing on the precipitation data predicted by the WRF;

[0013] Match the forecast data and the actual observed data in time as the training and test data sets of the machine learning model;

[0014] Use the Bayesian parameter optimization method to find the best parameters for the training of the machine learning model;

[0015] Use the optimized machine learning model to predict the precipitation at the meteorological stations on the test set.

[0016] As a preferred embodiment of the rain and shine forecast method based on complex micro-topographic conditions of the present invention, wherein: obtain the NCEP / GFS forecast field data with a preset resolution as the initial field and lateral boundary conditions of the WRF model; obtain the NCEP / GFS forecast field data with a daily data resolution of 0.25°×0.25°, a starting time of 18:00 UTC, and a 102-hour forecast every 3 hours as the initial field and lateral boundary conditions of the WRF model;

[0017] Obtain the surface static data of terrain with a resolution of 15s (500m), soil data, and vegetation cover provided by the MODIS satellite;

[0018] Combine the WRF grid settings with 2 layers of grid nesting, grid numbers of 600×500 and 967×535 respectively, horizontal grid resolutions of 9km and 3km respectively, and the grid center points;

[0019] Combined with the parameterization scheme named "CONUS": the microphysical scheme is the Thompson scheme, the cumulus parameterization scheme is the Tiedtke scheme, the long-wave and short-wave radiation schemes are both the RRTMG scheme, the boundary layer and surface parameterization schemes are both the MYJ scheme, and the land surface process scheme uses the Noah land surface process scheme to generate the WRFOUT numerical weather forecast file. The weather forecast file includes meteorological elements such as temperature, relative humidity, precipitation, air pressure, 10m wind speed, 10m wind direction, 2m dew point temperature, 10m meridional wind, 10m zonal wind, and icing thickness;

[0020] Obtain the 3km*3km meteorological grid data and the number of meteorological stations issued by the China Public Service Center, as well as obtain the longitude, latitude, altitude, and geographical information of the meteorological stations. The meteorological station data includes temperature, air pressure, relative humidity, wind speed, and precipitation data;

[0021] Use the bilinear interpolation method to interpolate the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological stations;

[0022] And use the variational mode decomposition and principal component analysis methods to perform feature engineering on the precipitation data predicted by WRF, and extract the precipitation components useful for the model;

[0023] Match the forecast data and the actual observed data in time as the training and test data sets of the LightGBM model;

[0024] Use the Bayesian parameter optimization method to find the best parameters for the training of the LightGBM model, and predict the precipitation of 410 meteorological stations on the test set;

[0025] The NCEP / GFS forecast field data is provided by the US Environmental Prediction Center.

[0026] As a preferred scheme of the rain and shine forecast method based on complex microtopography conditions of the present invention, wherein: the bilinear interpolation method is used to interpolate the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological stations. The bilinear interpolation method represents the linear interpolation extension of the interpolation function containing two variables, and performs linear interpolation in two directions respectively;

[0027] The bilinear interpolation assumes that the known function f is at four points Q 11 =(x1,y1), Q 12 =(x1,y2), Q 21 =(x2,y1) and Q 22 =(x2,y2). The value at the point P=(x, y) of the unknown function f is to be obtained. The specific operation steps are as follows:

[0028] First, perform linear interpolation in the x direction to obtain the following two expressions:

[0029]

[0030] Then, perform linear interpolation in the y direction according to the above two expressions to obtain the following expression:

[0031]

[0032] In this way, the result of the function f(x, y) is obtained, as shown in the following expression:

[0033]

[0034] As a preferred solution of the rain and shine forecasting method based on complex micro-topographic conditions according to the present invention, wherein: the Bayesian parameter optimization method is used to find the best parameters for training the LightGBM model, and the precipitation of 410 meteorological stations is predicted on the test set, wherein the LightGBM model is a lightweight gradient boosting algorithm.

[0035] As a preferred solution of the rain and shine forecasting method based on complex micro-topographic conditions according to the present invention, wherein: the predicted precipitation results are spatially interpolated to generate high-resolution precipitation grid data covering the entire forecast area. The spatial interpolation is based on the data of surrounding known meteorological stations to calculate the precipitation of unknown points, and spatial correlation is considered to predict more accurate values. The following expression can be used:

[0036]

[0037] Among them, Kriging is a statistical interpolation method that can consider the correlation in space and generate more accurate grid data. Among them, λ i is the weight coefficient, P i is the precipitation of the known point, and μ is the local average precipitation;

[0038] Use the terrain elevation data to correct the terrain of the interpolated precipitation, considering the influence of the terrain on precipitation; based on the corrected precipitation data, combined with other meteorological elements such as temperature and humidity, calculate the precipitation phase to distinguish between rainfall and snowfall; according to the precipitation phase and the amount of precipitation, divide the weather types of clear, cloudy, light rain, moderate rain, and heavy rain to generate a refined rain and shine forecast product;

[0039] Use the precipitation P krig obtained from Kriging interpolation as the input, and consider the influence of the terrain to adjust the precipitation; according to the influence of the terrain elevation h(x, y) on the precipitation, construct a terrain adjustment model, and the following expression can be used:

[0040]

[0041] Among them, α and β are model parameters, and this exponential adjustment takes into account the non-linear variation of precipitation with increasing altitude.

[0042] As a preferred embodiment of the rain and shine forecasting method based on complex micro-topographic conditions of the present invention, wherein: based on the corrected precipitation data and other meteorological data, a statistical relationship model between precipitation and ice coating thickness is established, and the following expression can be used:

[0043]

[0044] Among them, M i represents the decision tree prediction model. All decision tree prediction models use meteorological data as input, and the meteorological data includes adjusted precipitation, temperature, wind speed, and humidity; P adj is the precipitation of the terrain adjustment model; T is the temperature; V is the wind speed; H is the relative humidity; Θ i represents the parameter of the i-th model, and w i is the model weight, which is determined by cross-validation to optimize the overall prediction performance;

[0045] Using the predicted meteorological elements such as precipitation, temperature, and wind speed, combined with the statistical relationship model, predict the ice coating thickness of the transmission line;

[0046] According to the predicted ice coating thickness, divide the ice coating grades and generate ice coating warning information for the transmission line;

[0047] Visualize the rain and shine forecasting products and ice coating warning information, and push the forecasting and warning results to users;

[0048] According to different ice coating grades, a risk assessment can be constructed. The risk assessment can be carried out by a multi-label classification method. Each category corresponds to different ice coating thickness thresholds and possible risks. The risk assessment can adopt the following expression:

[0049] Risk Level=MultiLabelClassifier(I;Φ)

[0050] Among them, the multi-label classifier MultiLabelClassifier is used to classify the ice coating thickness I into multiple risk levels. Here, Φ represents the classifier parameters, and the risk levels can be divided into no risk, low risk, medium risk, and high risk.

[0051] As a preferred embodiment of the rain and shine forecasting method based on complex micro-topographic conditions of the present invention, wherein: the risk levels can be divided into no risk, low risk, medium risk, and high risk, and the conditions for each risk level can be:

[0052] Low risk: I < 5mm. Countermeasures: Judge that the risk level of the warning level is low risk, send warning information to relevant departments and the public that may be affected, generate and display low-intensity virtual disaster scenarios for risk awareness training;

[0053] Medium risk: 5mm ≤ I < 10mm. Countermeasures: Judge that the risk level of the warning level is medium risk, send warning information to relevant departments, generate and display medium-intensity virtual disaster scenarios for risk communication and emergency drills; At the same time, start the emergency response plan and prepare evacuation and rescue resources;

[0054] High risk: I ≥ 10mm. Countermeasures: Judge that the risk level of the warning level is high risk, issue an emergency warning to all relevant departments and the public; Generate and display high-intensity virtual disaster scenarios for evacuation route planning and rescue strategy formulation; At the same time, immediately start the evacuation procedure and emergency response measures;

[0055] According to the forecast and warning results pushed to users, a real-time verification and feedback mechanism for the model forecast results can be established, and actual observation data can be collected; Calculate the error between the forecast result and the actual observation, and evaluate the forecast accuracy; Based on the error analysis results, dynamically adjust the model parameters and forecast strategies; Regularly analyze historical forecast data, summarize forecast experience, and continuously optimize the forecast methods and processes.

[0056] In the second aspect, the embodiment of the present invention provides a sunny and rainy forecast system based on complex micro-topographic conditions, which includes an acquisition module that acquires NCEP / GFS forecast field data with a preset resolution as the initial field and side boundary conditions of the WRF model; Acquire surface static data of terrain, soil data, and vegetation cover with a preset resolution provided by the MODIS satellite;

[0057] A preset module that sets the number of grid nestings, the number of grids, and the horizontal grid resolution of the WRF model, selects the parameterization scheme of the WRF model, and generates a WRFOUT numerical weather forecast file containing various meteorological elements; Acquire meteorological grid data and meteorological station data with a preset resolution, as well as the geographical information of the meteorological station;

[0058] A processing module that interpolates the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological station using the bilinear interpolation method; Use variational mode decomposition and principal component analysis methods to perform feature engineering processing on the precipitation data forecast by the WRF;

[0059] A training module that matches the forecast data and the actual observation data in time as the training and test data sets of the machine learning model; Use the Bayesian parameter optimization method to find the best parameters for training the machine learning model;

[0060] An output module that uses an optimized machine learning model to predict the precipitation of a meteorological station on a test set.

[0061] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program instructions are executed by the processor, the steps of the rain and shine forecasting method based on complex microtopography conditions as described in the first aspect of the present invention are implemented.

[0062] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program instructions are executed by the processor, the steps of the rain and shine forecasting method based on complex microtopography conditions as described in the first aspect of the present invention are implemented.

[0063] The beneficial effects of the present invention are as follows: By obtaining meteorological station data and meteorological element data predicted by numerical models, and through a series of processes, the precipitation at the meteorological station can be corrected more accurately, or rather, the results of meteorological elements predicted by numerical models can be corrected better, providing scientific support for the prediction of transmission line icing, and having important scientific significance and application value.

[0064] By obtaining high-resolution NCEP / GFS forecast field data and surface static data provided by MODIS satellites, an accurate description of the initial field and boundary conditions under complex microtopography conditions is achieved. This provides high-quality input data for subsequent numerical simulations and effectively improves the accuracy of simulation results.

[0065] By setting up multi-level grid nesting and high-resolution horizontal grids, and selecting a suitable parameterization scheme, a refined simulation of complex microtopography areas is achieved. This setting can better capture the influence of local topography on meteorological elements and improve the spatial accuracy of simulation results.

[0066] Using the bilinear interpolation method to interpolate WRFOUT grid data to the longitude and latitude of the meteorological station realizes the spatial matching of model output and observed data. This step provides high-quality input data for subsequent machine learning model training and helps improve the accuracy of forecasting.

[0067] By using variational mode decomposition and principal component analysis methods to perform feature engineering processing on the precipitation data predicted by WRF, the extraction of key features in the precipitation data is achieved. This step effectively reduces the noise of the data and improves the training effect of subsequent machine learning models.

[0068] Adopting the LightGBM model and using the Bayesian parameter optimization method realizes the automatic optimization of model parameters. This method can make full use of the information of historical data, improve the generalization ability of the model, and thus improve the accuracy of forecasting.

[0069] By performing spatial interpolation and terrain correction on the predicted precipitation results, the generation of high-resolution grid precipitation data is achieved. This step takes into account the influence of terrain on precipitation, improving the spatial accuracy and reliability of the forecast results.

[0070] A statistical relationship model between precipitation and ice coating thickness is established to achieve the prediction of ice coating risk on transmission lines. This function expands the application scope of sunny-rainy forecasts and provides important support for the safe operation of power systems.

[0071] By constructing a risk assessment model and a hierarchical early warning system, the timely early warning of different levels of ice coating risk is achieved. This function provides a scientific basis for disaster prevention and mitigation and helps reduce economic losses caused by extreme weather events.

[0072] A real-time verification and feedback mechanism for forecast results is established to achieve the dynamic adjustment and continuous optimization of the model. This mechanism ensures the long-term effectiveness of the forecast system and the continuous improvement of forecast accuracy.

[0073] In summary, through the comprehensive utilization of multi-source data, high-resolution numerical simulation, machine learning algorithms, and risk assessment models, the present invention achieves the comprehensive optimization of sunny-rainy forecasts under complex micro-topographic conditions. Compared with the prior art, this method significantly improves the spatio-temporal accuracy of forecasts, expands the application scope, and has important practical value especially in the safe operation of power systems and disaster prevention and mitigation. In addition, the modular design and dynamic optimization mechanism of this method ensure the scalability and continuous improvement ability of the system, laying a foundation for further improving the meteorological forecast level under complex terrain conditions in the future. Brief Description of the Drawings

[0074] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0075] Figure 1 It is a flowchart of a sunny-rainy forecast method based on complex micro-topographic conditions;

[0076] Figure 2 It is a computer equipment diagram of a sunny-rainy forecast method based on complex micro-topographic conditions;

[0077] Figure 3 It is a density schematic diagram of the corrected value of the precipitation wrf forecast value and the test set and the monitored precipitation at meteorological stations of a sunny-rainy forecast method based on complex micro-topographic conditions. Detailed Embodiments

[0078] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification.

[0079] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0080] Secondly, as used herein, an "embodiment" or "an embodiment" refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in an embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is separate or mutually exclusive of other embodiments.

[0081] Embodiment 1

[0082] Referring to Figures 1 to 2 , which is the first embodiment of the present invention. This embodiment provides a sunny-rainy forecasting method based on complex micro-topographic conditions, including:

[0083] S100: Obtain NCEP / GFS forecast field data with a preset resolution as the initial field and lateral boundary conditions of the WRF model; obtain surface static data of terrain, soil, and vegetation cover with a preset resolution provided by the MODIS satellite.

[0084] S101: Based on the sunny-rainy forecasting under complex micro-topographic conditions, obtain NCEP / GFS forecast field data with a preset resolution as the initial field and lateral boundary conditions of the WRF model; obtain NCEP / GFS forecast field data with a daily data resolution of 0.25°×0.25°, a starting time of 18:00 UTC, and a 102-hour forecast period with a 3-hour interval as the initial field and lateral boundary conditions of the WRF model.

[0085] Obtain surface static data of terrain with a resolution of 500 m, soil, and vegetation cover provided by the MODIS satellite.

[0086] Combine the number of nested grid layers of 2 layers, with the number of grids being 600×500 and 967×535 respectively, the horizontal grid resolutions being 9 km and 3 km respectively, and the WRF grid settings at the grid center points.

[0087] Combined with the parameterization scheme named "CONUS": the microphysical scheme is the Thompson scheme, the cumulus parameterization scheme is the Tiedtke scheme, the long-wave and short-wave radiation schemes are both the RRTMG scheme, the boundary layer and surface parameterization schemes are both the MYJ scheme, and the surface process scheme uses the Noah surface process scheme to generate the WRFOUT numerical weather forecast file. The weather forecast file for sunny-rainy forecast under complex micro-topographic conditions includes meteorological elements such as temperature, relative humidity, precipitation, air pressure, 10m wind speed, 10m wind direction, 2m dew point temperature, 10m meridional wind, 10m zonal wind, and icing thickness;

[0088] Obtain the 3km*3km meteorological grid data and meteorological stations issued by the China Public Service Center, as well as the longitude, latitude, altitude, and geographical information of the meteorological stations. The meteorological station data for sunny-rainy forecast under complex micro-topographic conditions includes temperature, air pressure, relative humidity, wind speed, and precipitation data;

[0089] Use the bilinear interpolation method to interpolate the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological stations;

[0090] And use the variational mode decomposition and principal component analysis methods to perform feature engineering on the precipitation data predicted by WRF, and extract the precipitation components useful for the model;

[0091] Match the forecast data and the actual observed data in time as the training and test data sets of the LightGBM model;

[0092] Use the Bayesian parameter optimization method to find the best parameters for training the LightGBM model, and predict the precipitation of 410 meteorological stations on the test set;

[0093] For sunny-rainy forecast under complex micro-topographic conditions, use the Bayesian parameter optimization method to find the best parameters for training the LightGBM model, and predict the precipitation of 410 meteorological stations on the test set. Among them, the LightGBM model for sunny-rainy forecast under complex micro-topographic conditions is a lightweight gradient boosting algorithm.

[0094] For sunny-rainy forecast under complex micro-topographic conditions, the NCEP / GFS forecast field data is provided by the US Environmental Prediction Center.

[0095] S200: Set the grid nesting level, number of grids, and horizontal grid resolution of the WRF model, select the parameterization scheme of the WRF model, and generate a WRFOUT numerical weather forecast file containing various meteorological elements; Obtain the meteorological grid data and meteorological station data with a preset resolution, as well as the geographical information of the meteorological stations;

[0096] S300: Interpolate the WRF OUT grid meteorological forecast data to the longitude and latitude of the meteorological station using the bilinear interpolation method; perform feature engineering processing on the precipitation data forecasted by WRF using the variational mode decomposition and principal component analysis methods;

[0097] S301: For the sunny-rainy forecast under complex micro-topographic conditions, interpolate the WRF OUT grid meteorological forecast data to the longitude and latitude of the meteorological station using the bilinear interpolation method. The bilinear interpolation method for the sunny-rainy forecast under complex micro-topographic conditions represents the linear interpolation extension of an interpolation function with two variables, and performs linear interpolation in two directions respectively;

[0098] For the bilinear interpolation of the sunny-rainy forecast under complex micro-topographic conditions, assume that the known function f has values at four points Q 11 =(x1, y1), Q 12 =(x1, y2), Q 21 =(x2, y1) and Q 22 =(x2, y2). To find the value of the unknown function f at the point P=(x, y), the specific operation steps are as follows:

[0099] First, perform linear interpolation in the x direction to obtain the following two expressions:

[0100]

[0101] Then, perform linear interpolation in the y direction according to the above two expressions to obtain the following expression:

[0102]

[0103] In this way, the result of the function f(x, y) is obtained, as shown in the following expression:

[0104]

[0105] S302: Perform spatial interpolation on the predicted precipitation results for the sunny-rainy forecast under complex micro-topographic conditions to generate high-resolution precipitation grid data covering the entire forecast area. For the spatial interpolation of the sunny-rainy forecast under complex micro-topographic conditions, based on the data of surrounding known meteorological stations, calculate the precipitation at unknown points, and consider spatial correlation to predict more accurate values. The following expression can be used:

[0106]

[0107] Among them, Kriging is a statistical interpolation method that can consider the correlation in space and generate more accurate grid data. Among them, λ i is the weight coefficient, P i is the precipitation at the known point, and μ is the local average precipitation;

[0108] Use topographic elevation data to correct the interpolated precipitation considering the influence of terrain on precipitation. Based on the corrected precipitation data, combined with other meteorological elements such as temperature and humidity, calculate the precipitation phase to distinguish between rainfall and snowfall. According to the precipitation phase and the amount of precipitation, classify the weather types of clear, cloudy, light rain, moderate rain, and heavy rain, and generate refined rain and shine forecast products.

[0109] Use the precipitation P obtained from Kriging interpolation krig as input, and consider the influence of terrain to adjust the precipitation. According to the influence of topographic elevation h(x, y) on precipitation, construct a terrain adjustment model, which can be expressed as follows:

[0110]

[0111] where α and β are model parameters, and this exponential adjustment takes into account the non-linear change of precipitation with increasing height.

[0112] S400: Match the forecast data and actual observed data in time as the training and test data sets of the machine learning model; Use the Bayesian parameter optimization method to find the best parameters for training the machine learning model.

[0113] S500: Use the optimized machine learning model to predict the precipitation of meteorological stations on the test set.

[0114] S501: Based on the rain and shine forecast under complex micro-topographic conditions, establish a statistical relationship model between precipitation and ice coating thickness based on the corrected precipitation data and other meteorological data, which can be expressed as follows:

[0115]

[0116] where M i represents the decision tree prediction model. All decision tree prediction models use meteorological data as input. Based on the rain and shine forecast under complex micro-topographic conditions, the meteorological data includes the adjusted precipitation, temperature, wind speed, and humidity; P adj is the precipitation of the terrain adjustment model; T is the temperature; V is the wind speed; H is the relative humidity; Θ i represents the parameter of the i-th model, and w i is the model weight, which is determined by cross-validation to optimize the overall prediction performance.

[0117] Use the predicted meteorological elements such as precipitation, temperature, and wind speed, combined with the statistical relationship model, to predict the ice coating thickness of transmission lines.

[0118] According to the predicted ice coating thickness, classify the ice coating levels and generate ice coating warning information for transmission lines.

[0119] Visualize the rain and shine forecast products and icing warning information, and push the forecast and warning results to users;

[0120] Risk assessments can be constructed according to different icing levels. The risk assessment based on the rain and shine forecast under complex micro-topographic conditions can be carried out through a multi-label classification method. Each category corresponds to different icing thickness thresholds and possible risks. The risk assessment can adopt the following expression:

[0121] Risk Level=MultiLabelClassifier(I;Φ)

[0122] Among them, the multi-label classifier MultiLabelClassifier is used to classify the icing thickness I into multiple risk levels. Here, Φ represents the classifier parameters. The risk levels based on the rain and shine forecast under complex micro-topographic conditions can be divided into no risk, low risk, medium risk, and high risk.

[0123] S502: The risk levels based on the rain and shine forecast under complex micro-topographic conditions can be divided into no risk, low risk, medium risk, and high risk. The conditions for each risk level can be:

[0124] Low risk: I < 5mm, Countermeasure: Judge that the risk level of the warning level is low risk, send warning information to relevant departments and the public who may be affected, generate and display low-intensity virtual disaster scenarios for risk awareness training;

[0125] Medium risk: 5mm ≤ I < 10mm, Countermeasure: Judge that the risk level of the warning level is medium risk, send warning information to relevant departments, generate and display medium-intensity virtual disaster scenarios for risk communication and emergency drills; At the same time, start the emergency response plan and prepare evacuation and rescue resources;

[0126] High risk: I ≥ 10mm, Countermeasure: Judge that the risk level of the warning level is high risk, issue an emergency warning to all relevant departments and the public; Generate and display high-intensity virtual disaster scenarios for evacuation route planning and rescue strategy formulation; At the same time, immediately start the evacuation procedure and emergency response measures;

[0127] Based on the rain and shine forecast under complex micro-topographic conditions, according to the forecast and warning results pushed to users, a real-time verification and feedback mechanism for the model forecast results can be established, and actual observation data can be collected; Calculate the error between the forecast result and the actual observation, and evaluate the forecast accuracy; Based on the error analysis results, dynamically adjust the model parameters and forecast strategies; Regularly analyze historical forecast data, summarize forecast experience, and continuously optimize the forecast methods and processes.

[0128] Furthermore, this embodiment also provides a sunny-rainy weather forecasting system based on complex micro-topographic conditions, including:

[0129] An acquisition module that acquires NCEP / GFS forecast field data with a preset resolution as the initial field and lateral boundary conditions of the WRF model; and acquires static surface data of terrain, soil, and vegetation cover with a preset resolution provided by the MODIS satellite;

[0130] A preset module that sets the number of grid nesting layers, the number of grids, and the horizontal grid resolution of the WRF model, selects the parameterization scheme of the WRF model, and generates a WRFOUT numerical weather forecast file containing various meteorological elements; acquires meteorological grid data and meteorological station data with a preset resolution, as well as the geographical information of the meteorological stations;

[0131] A processing module that interpolates the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological station using the bilinear interpolation method; and performs feature engineering processing on the precipitation data predicted by the WRF using the variational mode decomposition and principal component analysis methods;

[0132] A training module that matches the forecast data and the actual observation data in time as the training and test data sets of the machine learning model; and uses the Bayesian parameter optimization method to find the optimal parameters for training the machine learning model;

[0133] An output module that uses the optimized machine learning model to predict the precipitation at the meteorological stations on the test set.

[0134] This embodiment also provides a computer device applicable to the sunny-rainy weather forecasting method based on complex micro-topographic conditions, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the sunny-rainy weather forecasting method based on complex micro-topographic conditions proposed in the above embodiment.

[0135] The computer device can be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0136] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for sunny-rainy weather forecasting based on complex micro-topographic conditions as proposed in the above embodiment.

[0137] In summary, by obtaining high-resolution NCEP / GFS forecast field data and surface static data provided by MODIS satellites, the accurate setting of the initial field and boundary conditions is achieved. Such high-precision input data provides a more accurate basis for the subsequent WRF model, which helps to improve the forecasting accuracy under complex micro-topographic conditions.

[0138] By setting up multi-level grid nesting and high-resolution horizontal grids, combined with an optimized parameterization scheme, a WRFOUT numerical weather forecast file containing various meteorological elements is generated. Such refined model settings can better capture the local meteorological characteristics under complex micro-topographic conditions, improving the spatial resolution and accuracy of the forecast.

[0139] Using the bilinear interpolation method to interpolate the WRFOUT grid point meteorological forecast data to the longitude and latitude of the meteorological station realizes the accurate matching of the model output and the observation site. This step effectively solves the problem of inconsistency between the model grid and the observation site, providing high-quality input data for the subsequent machine learning model training.

[0140] By using variational mode decomposition and principal component analysis methods to perform feature engineering processing on the precipitation data predicted by WRF, the precipitation components useful for the model are extracted. Such advanced data processing technologies can effectively reduce data noise and extract key features, thereby improving the training effect and prediction accuracy of the machine learning model.

[0141] By adopting the LightGBM model and combining it with the Bayesian parameter optimization method, the best training parameters are found, achieving high-precision prediction of precipitation at meteorological stations. This method that combines advanced machine learning algorithms and parameter optimization techniques can make full use of multi-source data, significantly improving the accuracy of sunny-rainy forecasts under complex micro-topographic conditions.

[0142] Through spatial interpolation and terrain correction of the prediction results, high-resolution precipitation grid data covering the entire forecast area is generated. This step takes into account the impact of terrain on precipitation, making the forecast results more in line with the actual terrain conditions and improving the spatial accuracy and reliability of the forecast.

[0143] Combining the predicted precipitation and other meteorological elements, a statistical relationship model for ice coating thickness is established, realizing ice coating warning for transmission lines. This method of comprehensive analysis of multiple elements not only improves the practicality of sunny-rainy forecasts but also expands the application scope of the forecast system, providing important support for the safe operation of the power system.

[0144] By constructing a risk assessment model, the prediction results are divided into different risk levels, and corresponding warning information and countermeasures are generated. The introduction of this risk assessment and warning mechanism greatly improves the practical value of the forecast results, providing a scientific basis for disaster prevention and mitigation.

[0145] A real-time verification and feedback mechanism for the forecast results is established. Through continuous error analysis and model optimization, dynamic adjustment and continuous improvement of the forecast method are realized. This closed-loop optimization mechanism ensures the long-term effectiveness and adaptability of the forecast system.

[0146] In summary, through multi-source data fusion, high-resolution numerical simulation, advanced feature engineering and machine learning algorithms, and post-processing techniques considering the impact of micro-topography, the invention comprehensively improves the accuracy, fineness and practicality of sunny-rainy forecasts under complex micro-topographic conditions. At the same time, by introducing ice coating warning and risk assessment mechanisms, the application scope of the forecast system is expanded, providing strong technical support for fields such as meteorological forecasting, disaster prevention and mitigation, and the safe operation of the power system.

[0147] Example 2

[0148] Refer to Figure 1 - Figure 3 , which is the second embodiment of the present invention. This embodiment provides a sunny-rainy forecast method based on complex micro-topographic conditions. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0149] Refer to Figure 1 As shown, the meteorological forecast element correction method for meteorological stations provided in this embodiment includes:

[0150] Obtain the NCEP / GFS forecast field data with a daily data resolution of 0.25°×0.25°, starting at 18:00 UTC, forecasting every 3 hours for a total of 102 hours as the initial field and lateral boundary conditions of the WRF model; obtain the surface static data such as terrain, soil data, and vegetation cover with a resolution of 15s (about 500m) based on the MODIS satellite.

[0151] Combine the 2-layer grid nesting levels. The grid numbers are 600×500 and 967×535 respectively, the horizontal grid resolutions are 9km and 3km respectively, and set the WRF grid at the grid center point.

[0152] Combine the parameterization scheme named "CONUS": the microphysical scheme is the Thompson scheme, the cumulus parameterization scheme is the Tiedtke scheme, the long-wave and short-wave radiation schemes are both the RRTMG scheme, the boundary layer and near-surface parameterization schemes are both the MYJ scheme, and the surface process scheme uses the Noah surface process scheme to generate the WRFOUT numerical weather forecast file (including meteorological elements such as temperature, relative humidity, precipitation, air pressure, 10m wind speed, 10m wind direction, 2m dew point temperature, 10m meridional wind, 10m zonal wind, and ice thickness).

[0153] Obtain the 3km*3km meteorological grid data and meteorological station data (including temperature, air pressure, relative humidity, wind speed, and precipitation data) issued by the China Public Service Center, and obtain the longitude, latitude, altitude, and geographical information of the meteorological stations.

[0154] Use the bilinear interpolation method to interpolate the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological stations; and use the variational mode decomposition and principal component analysis (VMD-PCA) method to perform feature engineering on the precipitation data predicted by WRF and extract the precipitation components useful for the model.

[0155] Match the forecast data and the actual observed data in time as the training and test data sets of the LightGBM model; use the Bayesian parameter optimization method to find the best parameters for training the LightGBM model and predict the precipitation of 410 meteorological stations on the test set.

[0156] Thus, it can be seen that by obtaining the meteorological station data and the meteorological element data predicted by the numerical model, and through a series of processes, it is possible to correct the precipitation at 410 meteorological stations more accurately or better correct the meteorological element results predicted by the numerical model, providing scientific support for the prediction of transmission line icing, which has important scientific significance and application value.

[0157] Specifically, the above meteorological station data includes temperature, air pressure, relative humidity, wind speed, and precipitation data. The WRF forecast data includes, but is not limited to, temperature, relative humidity, precipitation, air pressure, 10m wind speed, 10m wind direction, 2m dew point temperature, 10m meridional wind, 10m zonal wind, and ice thickness.

[0158] Specifically:

[0159] Mathematically, bilinear interpolation represents the linear interpolation extension of an interpolation function with two variables. The core idea of this method is to perform linear interpolation in two directions respectively.

[0160] Based on the bilinear interpolation of sunny-rainy forecasts under complex micro-topographic conditions, assume that the known function f has values at four points Q 11 =(x1,y1), Q 12 =(x1,y2), Q 21 =(x2,y1), and Q 22 =(x2,y2). To find the value of the unknown function f at the point P=(x, y), the specific operation steps are as follows:

[0161] First, perform linear interpolation in the x direction to obtain the following two expressions:

[0162]

[0163] Then, perform linear interpolation in the y direction according to the above two expressions to obtain the following expression:

[0164]

[0165] In this way, the result of the function f(x, y) is obtained, as shown in the following expression:

[0166]

[0167] In this way, through the above formula, the meteorological forecast elements at each point output by the WRF model can be interpolated to 410 meteorological stations with different longitudes and latitudes.

[0168] The following further elaborates on this method in combination with an application scenario example:

[0169] The temperature, relative humidity, precipitation, air pressure, 10m wind speed, 10m wind direction, 2m dew point temperature, 10m meridional wind, 10m zonal wind, ice thickness and other factors predicted by the WRF model from 8:00 on December 2, 2023 to 7:00 on March 2, 2024, with a latitude range of 17.3°N-32.1°N, a longitude range of 89.6°E-119.4°E, a unified time resolution of 1 hour, and a grid resolution of 3km were selected. At the same time, the variational mode decomposition and principal component analysis (VMD-PCA) algorithm was used to extract multiple characteristic factor components of precipitation. The bilinear interpolation method is used to interpolate these forecast meteorological elements to 410 meteorological stations with known longitude and latitude, and the monitoring data of 410 meteorological stations at the same time, including temperature, relative humidity, precipitation, air pressure, and 10m wind speed, are combined with the known information of these stations: longitude and latitude, altitude. The WRF forecast data interpolated to the station and the actual precipitation monitoring data of the station are combined according to the time dimension to form the training set and test set of the LightGBM algorithm, and outliers and null values ​​are eliminated. Here, we select 8:00 on December 2, 2023 to 7:00 on January 1, 2023 as the training set, and the three months after that, that is, 8:00 on January 1, 2024 to 7:00 on March 2, 2024, as the test set. The test set does not participate in model training. The training set conducts 5 cross-experiments, and uses the Bayesian parameter optimization method to find the parameter settings of the LightGBM algorithm with the highest training set accuracy. The specific parameters for precipitation correction are 'max_depth': 39, 'min_child_samples': 18, 'n_estimators': 331, 'num_leaves': 115. The specific steps are as follows:

[0170] Step A: Using the WRF model, the NCEP / GFS forecast field data with a daily data resolution of 0.25°×0.25°, a start time of 18:00 UTC, and a forecast every 3 hours, a total of 102 hours of forecast time, are used as the initial field and lateral boundary conditions of the WRF model; static surface data such as terrain, soil data, and vegetation coverage with a resolution of 15s (about 500m) provided by the MODIS satellite are obtained; combined with two layers of grid nesting, the grid numbers are 600×500 and 967×535 respectively, and the horizontal grid is divided into The resolutions are 9km and 3km respectively, and the grid center points are located at 29°N and 96°E. The WRF grid setting is combined with a parameterization scheme named "CONUS": the microphysics scheme is the Thompson scheme, the cumulus parameterization scheme is the Tiedtke scheme, the long- and short-wave radiation schemes are both the RRTMG scheme, the boundary layer and near-surface parameterization schemes are both the MYJ scheme, and the pavement process scheme uses the Noah pavement process scheme to generate the WRFOUT numerical weather forecast file (including temperature, humidity, precipitation and other meteorological elements).

[0171] Step B: Using the bilinear interpolation method, interpolate the meteorological elements of temperature, relative humidity, precipitation, air pressure, 10m wind speed, 10m wind direction, 2m dew point temperature, 10m meridional wind, 10m zonal wind, and icing thickness predicted by the WRF model with a grid resolution of 3km to 410 meteorological stations with known longitude and latitude.

[0172] Step C: The data of 410 meteorological stations within the same time (including temperature, air pressure, relative humidity, wind speed, and precipitation data). In addition, combined with the known information of these 410 meteorological stations: longitude, latitude, and geographical information, according to the time dimension, jointly combine the WRF forecast data interpolated to 410 meteorological stations and the actual monitoring data of 410 meteorological stations to form the training set and test set of the LightGBM algorithm.

[0173] Step D: Use the Bayesian parameter optimization method to find the parameter settings of the LightGBM algorithm with the highest accuracy in the training set. The specific parameters are'max_depth': 39,'min_child_samples': 18, 'n_estimators': 331, 'num_leaves': 115. The accuracy ACC of predicting the precipitation test set at 410 meteorological stations is 0.90, and the TS score of the sunny-rainy forecast accuracy has increased from 0.22 of the WRF forecast to 0.30, as Figure 3 . And the scores of 410 stations have all improved. In summary, the method provided by the embodiment of the present invention can correct and predict the precipitation of 410 meteorological stations on the transmission line more accurately, can increase the accuracy of the meteorological element forecast of 410 meteorological stations in the future to a certain extent, and has a good early warning effect on the probability of future disasters, with important scientific significance and application value.

[0174] Example 3

[0175] Refer to Figure 2 - Figure 3 , which is the third embodiment of the present invention. This embodiment provides a sunny-rainy forecast method based on complex micro-topographic conditions. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0176] To verify the effectiveness of the method of the present invention, Sichuan Province is selected as the research area. The terrain of this area is complex, including basins, mountains, and plateaus, which poses a huge challenge to sunny-rainy forecasting. This experiment selects the period from December 1, 2023 to February 29, 2024 as the research period, a total of 91 days.

[0177] First, obtain the NCEP / GFS forecast field data with a resolution of 0.25°×0.25° as the initial field and lateral boundary conditions of the WRF model. The starting time is 18:00 UTC, and the forecast is made every 3 hours for a duration of 102 hours. Meanwhile, obtain the terrain, soil, and vegetation cover data with a resolution of 15 seconds (about 500 meters) provided by the MODIS satellite as the surface static data.

[0178] Set the grid nesting of the WRF model to three layers. The outer layer covers the entire southwestern region, the middle layer covers Sichuan Province, and the inner layer focuses on covering the western Sichuan region with the most complex terrain. The number of grids for the three layers is 200×180, 300×270, and 400×360 respectively, and the corresponding horizontal grid resolutions are 9 km, 3 km, and 1 km. Select the Thompson microphysics scheme, Kain-Fritsch cumulus parameterization scheme, RRTMG shortwave and longwave radiation scheme, YSU boundary layer scheme, and Noah-MP land surface process scheme to generate the WRFOUT numerical weather forecast file containing 15 meteorological elements such as temperature, relative humidity, and precipitation.

[0179] Obtain the observation data of 410 automatic weather stations in Sichuan Province from the China Meteorological Administration, including elements such as temperature, relative humidity, air pressure, wind speed, and precipitation per hour. Use the kriging spatial interpolation method to interpolate these station data to generate a grid observation field with a resolution of 1 km as the true value field for subsequent verification and model training.

[0180] Use the bicubic spline interpolation method to interpolate the WRFOUT grid meteorological forecast data to a resolution of 1 km to match the observed true value field. To extract effective precipitation characteristics, use the variational mode decomposition (VMD) method to decompose the precipitation forecast by WRF into 5 intrinsic mode functions (IMFs), and then extract the first 3 principal components as the feature inputs for precipitation forecasting through the principal component analysis (PCA) method.

[0181] Randomly divide the 91-day data into a training set and a test set at a ratio of 7:3. Use the LightGBM model for training, and use the Bayesian optimization algorithm to optimize the model hyperparameters. The finally determined main parameters are: max_depth = 12, num_leaves = 128, learning_rate = 0.05, n_estimators = 500.

[0182] To evaluate the performance of this method, it was compared with the traditional WRF direct output (WRF-Raw) and the simple bias correction method (WRF-BC). The evaluation metrics include the root mean square error (RMSE), correlation coefficient (R), and critical success index (CSI). Among them, CSI was calculated for three precipitation levels: light rain (0.1 - 9.9 mm / 24h), moderate rain (10.0 - 24.9 mm / 24h), and heavy rain (≥25.0 mm / 24h). The experimental results are shown in the following table:

[0183] Method RMSE (mm / 24h) R CSI (light rain) CSI (moderate rain) CSI (heavy rain) WRF-Raw 8.76 0.62 0.45 0.32 0.18 WRF-BC 7.32 0.71 0.52 0.38 0.23 This method 5.14 0.85 0.68 0.53 0.41

[0184] By analyzing the data in the above table, it can be clearly seen that this method is significantly superior to the traditional method in all evaluation metrics. Specifically: in terms of the overall accuracy of precipitation prediction, the RMSE of this method is 5.14 mm / 24h, which is 41.3% and 29.8% lower than that of WRF-Raw and WRF-BC respectively. This indicates that this method has greatly improved the accuracy of precipitation prediction, especially its prediction ability under complex terrain conditions.

[0185] The correlation coefficient R reflects the consistency between the predicted value and the actual observed value. The R value of this method reaches 0.85, which is 37.1% higher than that of WRF-Raw and 19.7% higher than that of WRF-BC. This means that this method can better capture the spatio-temporal variation characteristics of precipitation, and the prediction results are closer to the actual observations. In terms of the forecasting skills for different precipitation levels, the advantages of this method are more obvious. For light rain, the CSI of this method is 0.68, which is 51.1% and 30.8% higher than that of WRF-Raw and WRF-BC respectively; for moderate rain, the improvement amplitudes of CSI are 65.6% and 39.5% respectively; for heavy rain, the improvement amplitudes of CSI reach 127.8% and 78.3% respectively. This shows that this method has significant advantages in predicting precipitation events of different intensities, especially for the improvement of the forecasting ability of stronger precipitation. The superiority of this method is mainly attributed to the following aspects: First, the high-resolution multi-level nested grid setting can better depict the impact of complex terrain on weather systems. Second, the VMD-PCA feature extraction method effectively extracts the key features of precipitation and reduces the noise impact. Third, the LightGBM model has a strong non-linear fitting ability and can make full use of multi-source data for learning. Finally, the Bayesian optimization algorithm ensures the optimal selection of model parameters.

[0186] These innovative points work together to enable this method to more accurately predict precipitation distribution and intensity under complex terrain conditions. Especially the significant improvement in heavy rain forecasting has important practical application value in fields such as flood control and drought relief, and water resource management. In addition, the calculation efficiency of this method is also relatively high. The entire forecasting process (including WRF simulation and post-processing) only takes about 4 hours for a 91-day forecast (using a 32-core CPU), meeting the requirements of operational operation.

[0187] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A sunny-rainy weather forecasting method based on complex micro-topographic conditions, characterized in that: including, obtaining NCEP / GFS forecast field data with a preset resolution as the initial field and lateral boundary conditions of the WRF model; obtaining surface static data of terrain, soil data, and vegetation cover with a preset resolution provided by MODIS satellite; setting the grid nesting level, number of grids, and horizontal grid resolution of the WRF model, selecting the parameterization scheme of the WRF model, and generating a WRFOUT numerical weather forecast file containing various meteorological elements; obtaining meteorological grid data and meteorological station data with a preset resolution, as well as the geographical information of the meteorological station; interpolating the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological station using the bilinear interpolation method; performing feature engineering processing on the precipitation data predicted by WRF using variational mode decomposition and principal component analysis methods; matching the forecast data and actual observation data in time as the training and test data sets of the machine learning model; using the Bayesian parameter optimization method to find the best parameters for training the machine learning model; using the optimized machine learning model to predict the precipitation of meteorological stations on the test set.

2. The rain and shine forecasting method based on complex microtopography conditions as claimed in claim 1, wherein: The step of obtaining NCEP / GFS forecast field data with a preset resolution as the initial field and lateral boundary conditions of the WRF model; obtaining NCEP / GFS forecast field data with a daily data resolution of 0.25°×0.25°, starting time at 18:00 UTC, forecasting every 3 hours for a total of 102 hours as the initial field and lateral boundary conditions of the WRF model; obtaining surface static data of terrain with a resolution of 15s (500m), soil data, and vegetation cover provided by MODIS satellite; combining the WRF grid settings with 2 layers of grid nesting, the number of grids being 600×500 and 967×535 respectively, and the horizontal grid resolutions being 9km and 3km respectively, with the grid center point; combining the parameterization scheme named "CONUS": the microphysical scheme is the Thompson scheme, the cumulus parameterization scheme is the Tiedtke scheme, the long-wave and short-wave radiation schemes are both the RRTMG scheme, the boundary layer and surface parameterization schemes are both the MYJ scheme, and the surface process scheme adopts the Noah surface process scheme to generate the WRFOUT numerical weather forecast file. The weather forecast file includes meteorological elements such as temperature, relative humidity, precipitation, air pressure, 10m wind speed, 10m wind direction, 2m dew point temperature, 10m meridional wind, 10m zonal wind, and icing thickness; obtaining 3km*3km meteorological grid data and meteorological station data issued by the China Public Service Center, as well as obtaining the longitude, latitude, altitude, and geographical information of the meteorological station. The meteorological station data includes temperature, air pressure, relative humidity, wind speed, and precipitation data; using the bilinear interpolation method to interpolate the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological station; and using variational mode decomposition and principal component analysis methods to perform feature engineering processing on the precipitation data predicted by WRF, and extracting the precipitation components useful for the model; matching the forecast data and actual observation data in time as the training and test data sets of the LightGBM model; Use the Bayesian parameter optimization method to find the optimal parameters for LightGBM model training, and predict the precipitation of 410 meteorological stations on the test set; The NCEP / GFS forecast field data is provided by the US Environmental Prediction Center.

3. The rain and shine forecasting method based on complex microtopography conditions according to claim 2, wherein: Use the bilinear interpolation method to interpolate the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological station. The bilinear interpolation method represents the linear interpolation extension of the interpolation function with two variables, and performs linear interpolation in two directions respectively; The bilinear interpolation assumes that the values of the known function \(f\) at four points \(Q 11 =(x_1,y_1)\), \(Q 12 =(x_1,y_2)\), \(Q 21 =(x_2,y_1)\) and \(Q 22 =(x_2,y_2)\) are known, and it is to find the value of the unknown function \(f\) at the point \(P=(x,y)\). The specific operation steps are as follows: First, perform linear interpolation in the x direction to obtain the following two expressions: Then, perform linear interpolation in the y direction according to the above two expressions to obtain the following expression: In this way, the result of the function f(x, y) is obtained, as shown in the following expression:

4. The rain and shine forecasting method based on complex micro-topography conditions according to claim 3, characterized in that: Use the Bayesian parameter optimization method to find the optimal parameters for LightGBM model training, and predict the precipitation of 410 meteorological stations on the test set, where the LightGBM model is a lightweight gradient boosting algorithm.

5. The rain and shine forecasting method based on complex micro-topographic conditions according to claim 4, wherein: Perform spatial interpolation on the predicted precipitation results to generate high-resolution precipitation grid data covering the entire forecast area. The spatial interpolation is based on the data of surrounding known meteorological stations, calculates the precipitation of unknown points, and considers spatial correlation to predict more accurate values. The following expression can be used: Among them, Kriging is a statistical interpolation method that can consider the correlation in space and generate more accurate grid data, where λ i is the weight coefficient, P i is the precipitation of known points, and μ is the local average precipitation; Use the terrain elevation data to correct the terrain of the interpolated precipitation, considering the influence of terrain on precipitation; based on the corrected precipitation data, combine other meteorological elements such as temperature and humidity to calculate the precipitation phase, distinguish between rainfall and snowfall; according to the precipitation phase and the magnitude of precipitation, divide the weather types of clear, cloudy, light rain, moderate rain, and heavy rain to generate refined clear and rain forecast products; Using the precipitation P obtained from Kriging interpolation krig As the input, consider the influence of terrain to adjust the precipitation; according to the influence of terrain elevation h(x, y) on precipitation, construct a terrain adjustment model, which can be expressed as follows: Among them, α and β are model parameters, and this exponential adjustment considers the non-linear change of precipitation with increasing height.

6. The rain and shine forecasting method based on complex micro-topographic conditions according to claim 5, characterized in that: Based on the corrected precipitation data and other meteorological data, establish a statistical relationship model between precipitation and ice coating thickness. The following expression can be used: Among them, M i represents a decision tree prediction model. All decision tree prediction models use meteorological data as input, and the meteorological data includes adjusted precipitation, temperature, wind speed, and humidity; P adj is the precipitation of the terrain adjustment model; T is the temperature; V is the wind speed; H is the relative humidity; Θ i represents the parameter of the i-th model, w i is the model weight, which is determined by cross-validation to optimize the overall prediction performance; Use the predicted meteorological elements such as precipitation, temperature, and wind speed, and combine the statistical relationship model to predict the ice coating thickness of the transmission line; According to the predicted ice coating thickness, divide the ice coating grades and generate ice coating warning information for the transmission line; Visualize the clear and rain forecast products and ice coating warning information, and push the forecast and warning results to users; A risk assessment can be constructed according to different ice coating grades. The risk assessment can be carried out by the multi-label classification method. Each category corresponds to different ice coating thickness thresholds and possible risks. The following expression can be used for the risk assessment: Risk Level = MultiLabelClassifier(I; Φ) Among them, the multi-label classifier MultiLabelClassifier is used to classify the ice coating thickness I into multiple risk levels. Here, Φ represents the classifier parameters, and the risk levels can be divided into no risk, low risk, medium risk, and high risk.

7. The method for sunny-rainy weather forecasting based on complex micro-topography conditions according to claim 6, wherein: The risk levels can be divided into no risk, low risk, medium risk, and high risk. The conditions for each risk level can be: Low risk: I < 5 mm. Response measures: Determine that the risk level of the warning level is low risk, send warning messages to relevant departments and the public that may be affected, generate and display low-intensity virtual disaster scenarios for risk awareness training; Medium risk: 5 mm ≤ I < 10 mm. Response measures: Determine that the risk level of the warning level is medium risk, send warning messages to relevant departments, generate and display medium-intensity virtual disaster scenarios for risk communication and emergency drills; At the same time, initiate the emergency response plan and prepare evacuation and rescue resources; High risk: I ≥ 10 mm. Response measures: Determine that the risk level of the warning level is high risk, issue an emergency warning to all relevant departments and the public; Generate and display high-intensity virtual disaster scenarios for evacuation route planning and rescue strategy formulation; At the same time, immediately initiate the evacuation procedure and emergency response measures; According to the forecast and warning results pushed to users, a real-time verification and feedback mechanism for the model forecast results can be established to collect actual observation data; Calculate the error between the forecast result and the actual observation, and evaluate the forecast accuracy; Based on the error analysis results, dynamically adjust the model parameters and forecast strategies; Regularly analyze historical forecast data, summarize forecast experience, and continuously optimize the forecast methods and processes.

8. A sunny-rainy weather forecasting system based on complex micro-topographical conditions, based on the sunny-rainy weather forecasting method based on complex micro-topographical conditions according to any one of claims 1 to 7, characterized in that: It also includes, An acquisition module that acquires NCEP / GFS forecast field data with a preset resolution as the initial field and lateral boundary conditions of the WRF model; Acquire static surface data of terrain, soil, and vegetation cover with a preset resolution provided by the MODIS satellite; A preset module that sets the number of grid nestings, the number of grids, and the horizontal grid resolution of the WRF model, selects the parameterization scheme of the WRF model, and generates a WRFOUT numerical weather forecast file containing various meteorological elements; Acquire meteorological grid data and meteorological station data with a preset resolution, as well as the geographical information of the meteorological station; A processing module that interpolates the WRFOUT grid meteorological forecast data to the longitude and latitude of the meteorological station using the bilinear interpolation method; Use variational mode decomposition and principal component analysis methods to perform feature engineering processing on the precipitation data predicted by WRF; A training module that matches the forecast data and the actual observation data in time as the training and test data sets of the machine learning model; Use the Bayesian parameter optimization method to find the best parameters for training the machine learning model; An output module that uses the optimized machine learning model to predict the precipitation at meteorological stations on the test set.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the rain and shine forecast method based on complex micro-topographic conditions according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the rain and shine forecast method based on complex micro-topographic conditions according to any one of claims 1 to 7.

Images