Three-dimensional wind field forecasting method and system based on live reanalysis data

Abstract

The present invention provides a three-dimensional wind field forecasting method and system based on live reanalysis data. The method comprises: obtaining an atmospheric three-dimensional element analysis field for a forecast area within a first historical time period from live data; obtaining a forecast result of the atmospheric three-dimensional element for the forecast area at the most recent time from background data; generating an initial field and lateral boundary conditions for a WRF model based on the atmospheric three-dimensional element forecast result at the most recent time period; using the initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field as inputs to the WRF model, and using the WRF model to generate a forecast result of the atmospheric three-dimensional element for the forecast area within a future time period; and sequentially performing three-dimensional element correction, terrain downscaling extraction, and attention focus correction on the atmospheric three-dimensional element forecast result predicted by the WRF model to obtain a three-dimensional wind field forecast result. The present invention improves the model's forecasting capability for three-dimensional wind fields.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological forecasting, and in particular to a three-dimensional wind field forecasting method and system based on live reanalysis data. Background Art

[0002] The three-dimensional wind field in the boundary layer can reflect the environmental characteristics before convection occurs (including convergence and divergence characteristics, vertical wind shear, etc.) as well as the wind field evolution characteristics during convection (including outflow, inflow, etc.). The establishment of a high-temporal and spatial precision short-term forecast system for the three-dimensional wind field in the boundary layer can not only directly provide wind field forecasts and quickly capture wind field changes, but also provide a variety of convection indices, which is conducive to improving the monitoring, forecasting and early warning of the occurrence and development of storms.

[0003] Refined short-term nowcasting methods for three-dimensional wind fields include downscaling numerical forecast results, extrapolation forecasting, fusion of numerical model results, and multi-model integrated forecasting techniques. Directly downscaling model results to obtain refined forecasts relies on model forecasts, and downscaling methods interpolate coarse-resolution data to a fine grid, so the forecast performance is also affected by the interpolation method. Extrapolation and fusion of model results are more commonly used in operational applications, but the actual data has a limited time range of influence. In the later stages of the forecast, the results are essentially the result of model downscaling. Therefore, this method is also affected by the forecast performance of the model. Multi-model integrated forecasting technology uses statistical and artificial intelligence methods to construct an optimal model based on multi-model results to conduct refined wind field forecasts. This method can integrate the advantages of each model, but the forecast performance of the model is also crucial to its forecast performance. Therefore, improving model forecast performance is key to enhancing the forecast capability of three-dimensional boundary layer wind fields. Summary of the Invention

[0004] The present invention provides a three-dimensional wind field forecasting method and system based on live reanalysis data, which is used to solve the defect of poor wind field model forecasting effect in the existing technology and improve the model forecasting capability of the wind field.

[0005] The present invention provides a three-dimensional wind field forecasting method based on live reanalysis data, comprising:

[0006] Obtain the three-dimensional atmospheric element analysis field of the forecast area in the first historical time period from the real-time data;

[0007] Obtaining the forecast results of the three-dimensional atmospheric elements in the forecast area at the nearest moment from the background data, and generating the initial field and lateral boundary conditions for the WRF model according to the forecast results of the three-dimensional atmospheric elements at the nearest moment;

[0008] The initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field are used as inputs of the WRF model, and the WRF model is used to generate a forecast result of the atmospheric three-dimensional elements in the forecast area in a future time period;

[0009] The atmospheric three-dimensional element forecast results predicted by the WRF model are sequentially subjected to three-dimensional element correction, terrain downscaling extraction and attention focus correction to obtain a three-dimensional wind field forecast result.

[0010] According to a three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention, the three-dimensional atmospheric element analysis field includes a wind field, a temperature field, a humidity field and an air pressure field.

[0011] According to a three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention, before the initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field are used as inputs of the WRF model, the method further includes:

[0012] performing sparse processing on the atmospheric three-dimensional element analysis field;

[0013] Wherein, jump point sampling is performed in the horizontal direction of the atmospheric three-dimensional element analysis field;

[0014] Performing thinning in the vertical direction of the atmospheric three-dimensional element analysis field based on a piecewise function; and / or,

[0015] The atmospheric three-dimensional element analysis field in the second historical time period is layered in the vertical direction;

[0016] Statistically analyze the three-dimensional atmospheric elements in different layers and obtain the probability density curves of the three-dimensional atmospheric elements in different layers;

[0017] According to the probability density curves of the atmospheric three-dimensional element analysis fields in different layers, the extreme values of the atmospheric three-dimensional element analysis fields in different layers are determined;

[0018] If the atmospheric three-dimensional element analysis field of any layer at each moment of the grid point in the forecast area during the first historical time period is greater than the corresponding extreme value, the atmospheric three-dimensional element analysis field of the grid point at that moment is eliminated; and / or,

[0019] Determining the difference between the atmospheric three-dimensional element analysis field at each moment observed at the grid point in the forecast area during the first historical time period and the atmospheric three-dimensional element analysis field predicted at the same moment at the grid point in the background data;

[0020] Calculating a probability density curve of the difference, and determining a preset threshold value according to the probability density curve of the difference;

[0021] When the difference is greater than a preset threshold, the atmospheric three-dimensional element analysis field of the grid point at the moment in the first historical time period is eliminated; and / or,

[0022] The atmospheric three-dimensional element analysis field within the first historical time period is normalized.

[0023] According to a three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention, the forecast results of the three-dimensional atmospheric elements in the forecast area at the most recent time are obtained from background data, and the initial field and lateral boundary conditions are generated for the WRF model based on the forecast results of the three-dimensional atmospheric elements at the most recent time, including:

[0024] Select the atmospheric three-dimensional element forecast results of the WRF framework model system or the global forecast model at the nearest moment;

[0025] If the atmospheric three-dimensional element forecast results of the WRF framework model system are selected, downscaling is performed based on the WPS module and NDOWN module in the WRF model to generate the initial field and lateral boundary conditions required for the forecast;

[0026] If the atmospheric three-dimensional element forecast results of the global forecast model are selected, downscaling is performed based on the WPS module and REAL module in the WRF model to generate the initial field and lateral boundary conditions required for the forecast.

[0027] According to a three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention, the initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field are used as inputs of the WRF model, and the WRF model is used to generate a forecast result of the atmospheric three-dimensional elements in the forecast area in a future time period, including:

[0028] Assimilate the atmospheric three-dimensional element analysis field using the OBS-Nudging module in the WRF model;

[0029] The initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field are used as inputs of the WRF model for assimilation integration for several hours. After the assimilation integration is completed, the WRF model begins formal free integration to generate the atmospheric three-dimensional element forecast results for the forecast area in the future time period.

[0030] According to a three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention, three-dimensional element correction is performed on the atmospheric three-dimensional element forecast result predicted by the WRF model, including:

[0031] Matching the grid of the atmospheric three-dimensional element analysis field in the third historical time period with the grid of the WRF model forecast;

[0032] Determine the residual of the atmospheric three-dimensional element forecast result predicted by the WRF model using the XGBoost model based on the atmospheric three-dimensional element analysis field within the third historical time period corresponding to the matched grid and the atmospheric three-dimensional element forecast result at the same time predicted by the WRF model;

[0033] A final model forecast result is determined based on the residual and the forecast result of the three-dimensional atmospheric elements in the forecast area in the future time period predicted by the WRF model.

[0034] According to a three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention, terrain downscaling is performed on the three-dimensional atmospheric element forecast results predicted by the WRF model, comprising:

[0035] The final model forecast results are downscaled for complex terrain using the CALMET model, and three-dimensional wind field forecast results are extracted therefrom.

[0036] According to a three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention, a three-dimensional atmospheric element forecast result predicted by the WRF model is corrected with attention focus, comprising:

[0037] The pre-acquired 3D wind farm forecast results are used as training samples, and the corresponding live wind farm observation data are used as labels. The LSTM model with attention mechanism is trained to obtain a wind farm rapid revision model.

[0038] The three-dimensional wind field forecast result extracted by the CALMET model is input into the wind field revision model to obtain a revised three-dimensional wind field forecast result.

[0039] The present invention also provides a three-dimensional wind field forecasting system based on live reanalysis data, comprising:

[0040] An extraction module is used to obtain the atmospheric three-dimensional element analysis field of the forecast area within the first historical time period from the real-time data;

[0041] A generation module is used to obtain the forecast results of the three-dimensional atmospheric elements in the forecast area at the most recent time from the background data, and generate the initial field and lateral boundary conditions for the WRF model according to the forecast results of the three-dimensional atmospheric elements at the most recent time;

[0042] A prediction module, configured to use the initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field as inputs to the WRF model, and generate a forecast result of the atmospheric three-dimensional elements in the forecast area within a future time period using the WRF model;

[0043] The post-processing module is used to perform three-dimensional element correction, terrain downscaling extraction and attention concentration correction on the atmospheric three-dimensional element forecast results predicted by the WRF model in sequence to obtain a three-dimensional wind field forecast result.

[0044] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the three-dimensional wind field forecasting method based on live reanalysis data as described above is implemented.

[0045] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the three-dimensional wind field forecasting method based on live reanalysis data as described above is implemented.

[0046] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described three-dimensional wind field forecasting methods based on live reanalysis data.

[0047] The present invention provides a three-dimensional wind field forecasting method and system based on live reanalysis data. From the perspective of directly improving the forecasting capability of the model, the method and system adopt nudging to assimilate the atmospheric three-dimensional element analysis field in the live reanalysis element data, thereby providing a high-update frequency and high-temporal-spatial resolution short-term forecast of the boundary layer three-dimensional wind field. The high-resolution live reanalysis data integrates multi-type and multi-source observation data, including three-dimensional thermodynamic information of the live atmosphere, which can optimize the quality of the model's initial thermodynamic field, thereby improving the model's forecasting capability for meteorological elements, especially the three-dimensional wind field. The method and system can achieve rapid cyclic updates, continuously absorb live three-dimensional reanalysis wind field data, optimize the model forecast effect, and generate terrain-associated three-dimensional wind field data, which is conducive to improving the forecast accuracy of the atmospheric three-dimensional wind field under complex terrain. The atmospheric three-dimensional wind field forecast results with high temporal-spatial resolution can be used for gale forecasts, and the refined three-dimensional wind field and ground element short-term forecast products can provide technical and data support for the forecast and early warning of sudden severe convective gale. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 This is one of the flow charts of the three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention;

[0050] Figure 2This is the second flow chart of the three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention;

[0051] Figure 3 Schematic diagram of assimilation and integration operation settings in the three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention;

[0052] Figure 4 Schematic diagram of the processing flow of the live data preprocessing module in the three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention;

[0053] Figure 5 Schematic diagram of the processing flow of the model data preprocessing module in the three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention;

[0054] Figure 6 1. It is a schematic diagram of the processing flow of the forecast result post-processing module in the three-dimensional wind field forecasting method based on live reanalysis data provided by the present invention;

[0055] Figure 7 Schematic diagram of the structure of a three-dimensional wind field forecasting system based on live reanalysis data provided by the present invention;

[0056] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0058] The following combination Figure 1 The present invention describes a three-dimensional wind field forecasting method based on live reanalysis data, comprising:

[0059] Step 101, obtaining an atmospheric three-dimensional element analysis field of a forecast area within a first historical time period from real-time data;

[0060] Step 102: Obtaining the forecast results of the three-dimensional atmospheric elements in the forecast area at the nearest time from the background data, and generating the initial field and lateral boundary conditions for the WRF model based on the forecast results of the three-dimensional atmospheric elements at the nearest time;

[0061] Step 103, using the initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field as inputs to the WRF model, and using the WRF model to generate a forecast result of atmospheric three-dimensional elements in the forecast area in a future time period;

[0062] Step 104 , performing three-dimensional element correction, terrain downscaling extraction, and attention focus correction on the atmospheric three-dimensional element forecast result predicted by the WRF model in sequence to obtain a three-dimensional wind field forecast result.

[0063] High-resolution 3D live reanalysis data includes a 3D atmospheric feature analysis field derived from the fusion of radar and ground-based automatic station observations. This field objectively reflects the 3D atmospheric characteristics at each moment. Obtain high-resolution 3D live reanalysis data and read the wind, temperature, humidity, and corresponding pressure field data.

[0064] The low-resolution three-dimensional atmospheric element forecast results are obtained from the background data, and the background field data are optimized to provide the optimal initial field and lateral boundary conditions for the WRF model.

[0065] Based on the WRF model, the OBS-Nudging method is used to introduce the real-time atmospheric three-dimensional element analysis field into the WRF model to obtain better weather forecast results.

[0066] In this embodiment, from the perspective of directly improving the model's forecasting capabilities, nudging is used to assimilate the atmospheric three-dimensional element analysis fields in the live reanalysis element data to provide a short-term forecast of the boundary layer three-dimensional wind field with high update frequency and high temporal and spatial resolution. The high-resolution live reanalysis data integrates multi-type and multi-source observation data, including three-dimensional thermodynamic information of the live atmosphere, which can optimize the quality of the model's initial thermodynamic field, thereby improving the model's forecasting capabilities for meteorological elements, especially the three-dimensional wind field. It can achieve rapid cyclic updates, continuously absorb live three-dimensional reanalysis wind field data, and optimize the model's forecast effect. At the same time, the terrain-associated three-dimensional wind field data generated is conducive to improving the forecast accuracy of the atmospheric three-dimensional wind field under complex terrain. The atmospheric three-dimensional wind field forecast results with high temporal and spatial resolution can be used for gale forecasts. The refined three-dimensional wind field and short-term forecast products can provide technical and data support for the forecast and warning of sudden severe convective gale.

[0067] On the basis of the above embodiment, the three-dimensional atmospheric element analysis field in this embodiment includes wind field, temperature field, humidity field and pressure field.

[0068] Based on the above embodiment, this embodiment further includes, before using the initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field as inputs to the WRF model:

[0069] performing sparse processing on the atmospheric three-dimensional element analysis field;

[0070] Wherein, jump point sampling is performed in the horizontal direction of the atmospheric three-dimensional element analysis field;

[0071] Performing thinning in the vertical direction of the atmospheric three-dimensional element analysis field based on a piecewise function; and / or,

[0072] The atmospheric three-dimensional element analysis field in the second historical time period is layered in the vertical direction;

[0073] Statistically analyze the three-dimensional atmospheric elements in different layers and obtain the probability density curves of the three-dimensional atmospheric elements in different layers;

[0074] According to the probability density curves of the atmospheric three-dimensional element analysis fields in different layers, the extreme values of the atmospheric three-dimensional element analysis fields in different layers are determined;

[0075] If the atmospheric three-dimensional element analysis field of any layer at each moment of the grid point in the forecast area during the first historical time period is greater than the corresponding extreme value, the atmospheric three-dimensional element analysis field of the grid point at that moment is eliminated; and / or,

[0076] Determining the difference between the atmospheric three-dimensional element analysis field at each moment observed at the grid point in the forecast area during the first historical time period and the atmospheric three-dimensional element analysis field predicted at the same moment at the grid point in the background data;

[0077] Calculating a probability density curve of the difference, and determining a preset threshold value according to the probability density curve of the difference;

[0078] When the difference is greater than a preset threshold, the atmospheric three-dimensional element analysis field of the grid point at the moment in the first historical time period is eliminated; and / or,

[0079] The atmospheric three-dimensional element analysis field within the first historical time period is normalized.

[0080] For high-temporal-resolution three-dimensional live reanalysis data, due to its high temporal-spatial resolution, if the data is not thinned to a certain extent, it will cause serious data redundancy, induce false noise, and easily cause model instability. During the thinning process, this embodiment adopts jump point sampling in the horizontal direction, sets a reasonable sampling distance based on the set forecast grid and grid spacing, and performs horizontal thinning. In the vertical direction, considering that the vertical layer of the WRF model is terrain-associated and the vertical layer distribution exhibits the characteristics of dense boundary layer and sparse high-level layer when the short-term system is designed, the thinning scheme in the vertical direction adopts a thinning scheme based on piecewise function, and searches for the optimal thinning grid below 3km to generate diluted live analysis grid data.

[0081] Because the OBS-Nudging module directly incorporates the forcing field into the governing equations, data quality control is particularly important. This implementation, in addition to traditional basic quality control (such as critical value checks and spatial and temporal consistency checks), incorporates climatological verification based on historical soundings and quality control based on the deviation between background field data and observational data, tailored to the characteristics of 3D data.

[0082] The climatological verification method based on historical soundings involves collecting historical sounding data and stratifying the atmosphere vertically. The historical sounding data within each layer is then statistically analyzed to obtain probability density curves for the data within each layer. A 99% confidence value is then determined as the climatological extreme value for each layer. The climatological extreme values for each layer are then used to further screen the extreme values of each vertical layer against the data that have already undergone extreme value quality control, eliminating any information at points that exceed the climatological threshold.

[0083] Based on the deviation between background field data and observation data: Considering that the difference between the data and the background field is too large, and forced observation data may cause model instability at this time, after completing the climatological verification based on historical soundings, quality control based on the deviation between background data (Background) and observation data (Observation) is carried out. When |Observation-Background| is greater than the set threshold, all data at that point are eliminated. At the same time, the deviation between background data and observation data is counted in real time, and the judgment threshold is continuously updated.

[0084] Since this embodiment performs forecasting based on the OBS-Nudging module of the WRF model, the extracted and thinned data needs to be normalized as required.

[0085] like Figure 2 As shown, the 3D wind farm forecast model system includes a model data preprocessing module, a real-time data preprocessing module, a model integration module, and a boundary layer 3D wind farm product generation module. This embodiment generates 0-6 hour boundary layer 3D wind farm forecast results with a resolution of 1 km and updated hourly. System operation settings ( Figure 3 ) is as follows: Taking into account the spin-up time of the model and the time of observation data, the three-dimensional wind field forecast system starts to perform obs-nudging of the observation data 3 hours in advance for each forecast, and then starts a 6-hour free integration. The forecast product comes from this 6-hour free integration, and the system starts one hour at a time.

[0086] like Figure 4 As shown in Figure 1, the live data processing module is responsible for acquiring high-resolution live 3D reanalysis data and converting them into model-readable files. The functional submodules involved include the live data automatic retrieval submodule, the data sparsification submodule, and the grid data conversion to station data submodule.

[0087] The live data automatic retrieval submodule is executed by two control scripts, get_vars.csh and check.csh. get_vars.csh regularly acquires high-resolution live 3D reanalysis data; the check.csh script is responsible for automatically retrieving live fusion analysis data near the system's start time. The high-resolution live 3D reanalysis data acquired in this embodiment is a 3D atmospheric feature analysis field derived from the fusion of radar and ground-based automatic station observation data. The analysis features include temperature, humidity, wind field, and air pressure, with a temporal resolution of 10 minutes and a spatial resolution of 3 km horizontally and 200 m vertically. The live data automatic retrieval submodule, implemented by get_vars.csh and check.csh, acquires analysis data hourly. If analysis data for the hour is not available, the previous 10 minutes of data is used as a replacement. Each activation of the short-term forecast system requires the previous three hours of live analysis data. Therefore, a check module is configured to verify the completeness of the previous three hours of analysis data before initiating observation data processing. If incomplete, a supplementary procedure is initiated.

[0088] Data Sparsification Submodule: After successful data retrieval, incoming observational data first enters the Data Format Conversion and Sparsification module. To avoid data redundancy, this module employs the sparsification scheme described in the present invention, namely, skip-point sampling in the horizontal direction and a piecewise function-based approach in the vertical direction. This module is primarily implemented by two scripts: readnc2txt.csh and readnc2txt.ncl. readnc2txt.csh controls readnc2txt.ncl in real time, reading high-resolution, live 3D reanalysis data while performing data sparsification and converting it into intermediate text-based data.

[0089] The Grid Data to Site Data submodule converts the grid data obtained by the Sparsification submodule into pattern data in the form of sites that are recognizable. This module is primarily controlled by two programs: nc2litter.csh and nc2litter.exe. nc2litter.csh controls nc2litter.exe in real time, reading the text-formatted intermediate data generated by the Data Sparsification submodule and writing it into a data format readable by the OBS-Nudging module.

[0090] The development of reanalysis data also needs to satisfy the atmospheric thermodynamic equations. During the generation process, multi-source data have been screened and quality controlled to a certain extent. This data is used for assimilation to reduce information redundancy caused by the assimilation of different data, as well as conflicts between different data during assimilation.

[0091] Based on the above embodiment, this embodiment obtains the forecast results of the three-dimensional atmospheric elements in the forecast area at the most recent time from background data, and generates the initial field and lateral boundary conditions for the WRF model based on the forecast results of the three-dimensional atmospheric elements at the most recent time, including:

[0092] Select the forecast results of atmospheric three-dimensional elements from the local model or global forecast model at the nearest moment;

[0093] If the atmospheric three-dimensional element forecast result of the local model is selected, downscaling is performed based on the WPS module and NDOWN module in the WRF model, and the initial field and lateral boundary conditions are generated according to the selected atmospheric three-dimensional element forecast result;

[0094] If the atmospheric three-dimensional element forecast results of the global forecast model are selected, downscaling is performed based on the WPS module and the REAL module in the WRF model, and the initial field and lateral boundary conditions are generated according to the selected atmospheric three-dimensional element forecast results.

[0095] For optimal background data selection, multiple criteria can be set to select the best background data. It is generally believed that the most detailed local forecast results at the nearest moment are the best, so the local model forecast results at the nearest moment are prioritized as input, with data at the next nearest moment and global model forecast results as alternatives.

[0096] During the generation of the initial field and lateral boundary conditions, different downscaling modules are selected based on the selected background field data. For local forecast results, the WRF WPS and NDOWN modules are used; for global model forecast results, the WRF WPS and REAL modules are used.

[0097] like Figure 5 As shown in Figure 1, the model data preprocessing module generates the initial fields and lateral boundary conditions required for the forecast based on the specified forecast area. It then downscales the background data to the required grid, serving as the driving field for the boundary layer 3D wind field forecast model. This module consists of two processing submodules: the automatic retrieval submodule and the downscaling submodule.

[0098] Automatic Retrieval Submodule: This module is controlled by two automated scripts, get_background.csh and check_bg.csh. get_background.csh is scheduled to retrieve multi-model results that can be used as WRF driver data. In this example, local regional numerical forecast data (3km forecast results based on the WRF model) and GFS global forecast results are selected. check_bg.csh is the main process for optimizing background field data, automatically searching for forecast results from various models at the near-term time, prioritizing the local model forecast results at the nearest time as input, and using the global forecast model forecast results as a backup.

[0099] Downscaling Submodule: This module generates background fields and lateral boundaries for the short-term forecast system based on the automatic retrieval submodule. It consists of two modules: the NDOWN module and the REAL module, controlled by the ndown.csh and initial.csh scripts, respectively. ndown.csh downscales local regional numerical forecast results using the WRF-based WPS module and the NDOWN (ndown.exe) module to generate initial fields and lateral boundaries for a 1km grid of the model's 1km boundary layer three-dimensional wind field forecast system. Initial.csh directly utilizes the WRF model's WPS module and the REAL (real.exe) module to generate initial fields and lateral boundaries for a 1km grid of global forecast model results.

[0100] Based on the above embodiment, in this embodiment, the initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field are used as inputs of the WRF model, and the WRF model is used to generate atmospheric three-dimensional element forecast results for the forecast area in the future time period, including:

[0101] Assimilate the atmospheric three-dimensional element analysis field using the OBS-Nudging module in the WRF model;

[0102] The initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field are used as inputs of the WRF model for assimilation integration for several hours. After the assimilation integration is completed, the WRF model begins formal free integration to generate the atmospheric three-dimensional element forecast results for the forecast area in the future time period.

[0103] Taking into account the spin-up time of the model and the time of observation data, OBS-Nudging is used to carry out four-dimensional relaxation approximation of observation data for several hours before the forecast time, and the analysis data of the observations are absorbed hour by hour.

[0104] After the assimilation is completed, free integration based on the WRF model is started to obtain fine-grained high-resolution numerical model deterministic forecasts.

[0105] The model integration forecast module assimilates the diluted live reanalysis data into the model in real time based on the OBS-Nudging module of the WRF model, and uses the WRF model to carry out forecasts. During the assimilation period of the model integration, a linear forcing term is added to the forecast equation to make the model forecast gradually approach the observation. The elements assimilated by this module include temperature, humidity, air pressure and wind field. The initial field, lateral boundary data and observation data generated by the model data preprocessing module and the observation data preprocessing module are used as input data, and the integration time is 9 hours, of which the first 3 hours are assimilation time and the last 6 hours are free integration ( Figure 3), and finally obtain the deterministic forecast result.

[0106] Based on the above embodiment, in this embodiment, three-dimensional element correction is performed on the atmospheric three-dimensional element forecast result predicted by the WRF model, including:

[0107] Matching the grid of the atmospheric three-dimensional element analysis field in the third historical time period with the grid of the WRF model forecast;

[0108] Determine the residual of the atmospheric three-dimensional element forecast result predicted by the WRF model using the XGBoost model based on the atmospheric three-dimensional element analysis field within the third historical time period corresponding to the matched grid and the atmospheric three-dimensional element forecast result at the same time predicted by the WRF model;

[0109] A final model forecast result is determined based on the residual and the forecast result of the three-dimensional atmospheric elements in the forecast area in the future time period predicted by the WRF model.

[0110] like Figure 6 As shown in the figure, the boundary layer three-dimensional wind field product generation module includes a forecast result post-processing module: first, a multi-factor correction model based on machine learning is established; then, complex terrain downscaling is performed based on CALMET to extract high temporal and spatial resolution three-dimensional wind field forecast results; finally, online wind field correction based on actual data is further carried out to form the corrected three-dimensional wind field data.

[0111] The atmospheric system is extremely complex, and the occurrence, development and evolution of various meteorological elements are connected in time and space. Therefore, the WRF forecast effect correction technology based on machine learning and multi-factor fusion is optimized.

[0112] This embodiment targets the three-dimensional atmospheric wind field, selects historical high-temporal-spatial resolution three-dimensional real-time reanalysis data, and carries out multi-factor coordinated three-dimensional wind field correction. The temperature, humidity, pressure, and wind in the historical reanalysis data are extracted and spatially matched with the WRF model results, and further multi-factor joint machine learning of temperature, humidity, air pressure, and wind components based on the XGBoost model is carried out to establish a correction model and optimize the simulation effect. When training the XGBoost model, the input is the historical high-temporal-spatial resolution real-time reanalysis data and the WRF model forecast results of the corresponding time of the model historical forecast. Based on the XGBoost model, the temperature, humidity, pressure, wind extracted from the historical reanalysis data and the predicted temperature, humidity, air pressure, and wind components are subjected to multi-factor joint machine learning, and the output is the result of the correction of each field. The corrected forecast results are used for CALMET analysis.

[0113] The steps for building a machine learning-based multi-factor forecast correction model include: ① Temporally and spatially aligning the grid of high-resolution live 3D reanalysis data with the WRF forecast grid, extracting temperature, humidity, wind, and pressure from both the live and model data, and bilinearly interpolating the WRF forecast results onto the 3D reanalysis data. ② Using the XGBoost model, a multi-factor optimization model is constructed to simultaneously predict the residuals (ΔT, ΔRH, ΔP, ΔU, and ΔV) of temperature, humidity, pressure, and U / V winds in a single model. ③ Based on the trained model, the WRF forecast results are corrected.

[0114] Based on the above embodiment, in this embodiment, terrain downscaling is performed on the three-dimensional atmospheric element forecast results predicted by the WRF model, including:

[0115] The final model forecast results are downscaled for complex terrain using the CALMET model, and three-dimensional wind field forecast results are extracted therefrom.

[0116] The CALMET model uses methods such as terrain dynamics and overland flow parameters to effectively analyze three-dimensional wind field characteristics under complex underlying surface conditions. Based on CALMET, the WRF results corrected by machine learning are used to extract three-dimensional wind field data on contour surfaces.

[0117] Downscaling is performed based on the CALMET model. The CALMET model is used to convert the corrected model output into terrain-dependent three-dimensional wind data. CALMET generates three-dimensional wind data with a horizontal resolution of 500 meters and a vertical resolution of 200 meters (0-4000 meters).

[0118] Based on the above embodiment, in this embodiment, the three-dimensional atmospheric element forecast results predicted by the WRF model are corrected with focused attention, including:

[0119] The pre-acquired 3D wind farm forecast results are used as training samples, and the corresponding actual wind farm observation data are used as labels. The LSTM model with attention mechanism is trained to obtain the wind farm revision model.

[0120] The three-dimensional wind field forecast result extracted by the CALMET model is input into the wind field revision model to obtain a revised three-dimensional wind field forecast result.

[0121] Real-time online wind field correction technology based on live wind profile data: For three-dimensional wind fields, an LSTM-attention mechanism is developed. Based on the initial forecast results of the forecast system, the corresponding time of the live wind profile data is matched to quickly establish a wind field revision model, correct the three-dimensional wind field forecast results at the current moment, and solve the performance degradation problem of traditional offline models caused by climate change.

[0122] All wind profile observation results are extracted and time-tagged, and LSTM-based attention training is carried out at the same time. For example, since this system is a 0-6 hour forecast system that reports hourly, the forecast result at 0 o'clock lags behind all processes before online correction by about 40 minutes. Therefore, the online correction is mainly based on the actual wind profile of the previous 30 minutes and the forecast results of the previous 30 minutes of the simulated forecast, and the correction parameters are quickly corrected to correct the three-dimensional wind field in the next 30 minutes to 6 hours.

[0123] This embodiment integrates machine learning and the CALMET model to construct a multi-factor and multi-scale machine learning correction process for three-dimensional wind fields in complex terrains, thereby improving the forecasting capability of three-dimensional wind fields.

[0124] The following describes the three-dimensional wind field forecasting system based on live reanalysis data provided by the present invention. The three-dimensional wind field forecasting system based on live reanalysis data described below and the three-dimensional wind field forecasting method based on live reanalysis data described above can be referenced to each other.

[0125] like Figure 7 As shown, the system includes an extraction module 701, a generation module 702 and a prediction module 703, wherein:

[0126] The extraction module 701 is used to obtain the atmospheric three-dimensional element analysis field of the forecast area within the first historical time period from the real-time data;

[0127] The generation module 702 is used to obtain the forecast results of the three-dimensional atmospheric elements in the forecast area at the most recent time from the background data, and generate the initial field and lateral boundary conditions for the WRF model based on the forecast results of the three-dimensional atmospheric elements at the most recent time;

[0128] The prediction module 703 is used to use the initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field as inputs of the WRF model, and generate a forecast result of the atmospheric three-dimensional elements in the forecast area in the future time period using the WRF model;

[0129] The post-processing module 704 is used to perform three-dimensional element correction, terrain downscaling extraction and attention concentration correction on the atmospheric three-dimensional element forecast results predicted by the WRF model in sequence to obtain a three-dimensional wind field forecast result.

[0130] This embodiment directly starts from the perspective of improving the model's forecasting capability, adopts the cyclic assimilation of the atmospheric three-dimensional element analysis field in the actual reanalysis element data, provides a high update frequency, and performs short-term forecasts of the boundary layer three-dimensional wind field with high temporal and spatial resolution; the high-resolution actual reanalysis data integrates multi-type and multi-source observation data, including three-dimensional thermodynamic information of the actual atmosphere, which can optimize the quality of the model's initial thermodynamic field, thereby improving the model's forecasting capability for meteorological elements, especially the three-dimensional wind field; it can achieve rapid cyclic updates, continuously absorb real-time three-dimensional reanalysis wind field data, optimize the model forecast effect, and the terrain-associated three-dimensional wind field data generated at the same time is conducive to improving the forecast accuracy of the atmospheric three-dimensional wind field under complex terrain; the atmospheric three-dimensional wind field forecast results with high temporal and spatial resolution can be used for gale forecasts, and the refined three-dimensional wind field and ground element short-term forecast products can provide technical and data support for the forecast and warning of sudden severe convective gale.

[0131] Figure 8 An example of a physical structure diagram of an electronic device is shown below. Figure 8 As shown, the electronic device may include: a processor (processor) 810, a communication interface (Communications Interface) 820, a memory (memory) 830 and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logic instructions in the memory 830 to execute a three-dimensional wind field forecasting method based on live reanalysis data, which includes obtaining an atmospheric three-dimensional element analysis field of the forecast area within a first historical time period from live data; obtaining the atmospheric three-dimensional element forecast result of the forecast area at the most recent moment from background data, and generating an initial field and lateral boundary conditions for the WRF model based on the atmospheric three-dimensional element forecast result at the most recent moment; using the initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field as inputs of the WRF model, and using the WRF model to generate the atmospheric three-dimensional element forecast result of the forecast area in a future time period; and performing three-dimensional element correction, terrain downscaling extraction and attention concentration correction on the atmospheric three-dimensional element forecast result predicted by the WRF model in sequence to obtain a three-dimensional wind field forecast result.

[0132] Furthermore, the logic instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0133] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the three-dimensional wind field forecasting method based on real-time reanalysis data provided by the above methods, the method including: obtaining an atmospheric three-dimensional element analysis field of the forecast area in a first historical time period from real-time data; obtaining the atmospheric three-dimensional element forecast result of the forecast area at the most recent moment from background data, and generating an initial field and lateral boundary conditions for a WRF model based on the atmospheric three-dimensional element forecast result at the most recent moment; using the initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field as inputs of the WRF model, and using the WRF model to generate the atmospheric three-dimensional element forecast result of the forecast area in a future time period; performing three-dimensional element correction, terrain downscaling extraction and attention concentration correction on the atmospheric three-dimensional element forecast result predicted by the WRF model in sequence to obtain a three-dimensional wind field forecast result.

[0134] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the three-dimensional wind field forecasting method based on live reanalysis data provided by the above-mentioned methods, the method comprising: obtaining an atmospheric three-dimensional element analysis field of the forecast area within a first historical time period from live data; obtaining an atmospheric three-dimensional element forecast result of the forecast area at the most recent moment from background data, and generating an initial field and lateral boundary conditions for a WRF model based on the atmospheric three-dimensional element forecast result at the most recent moment; using the initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field as inputs of the WRF model, and using the WRF model to generate an atmospheric three-dimensional element forecast result of the forecast area in a future time period; and performing three-dimensional element correction, terrain downscaling extraction and attention concentration correction on the atmospheric three-dimensional element forecast result predicted by the WRF model in sequence to obtain a three-dimensional wind field forecast result.

[0135] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0136] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A three-dimensional wind field forecasting method based on live reanalysis data, characterized in that: include: Obtaining a three-dimensional atmospheric element analysis field of the forecast area within a first historical time period from live data, wherein the live data is high temporal and spatial resolution three-dimensional live reanalysis data, including a three-dimensional atmospheric element analysis field obtained by fusion of radar and ground automatic station observation data; Obtaining a forecast result of the three-dimensional atmospheric elements in the forecast area at the nearest moment from background data, and generating an initial field and lateral boundary conditions for the WRF model based on the forecast result of the three-dimensional atmospheric elements at the nearest moment, wherein the forecast result of the three-dimensional atmospheric elements obtained from the background data is a low-resolution forecast result; The initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field are used as inputs of the WRF model, and the WRF model is used to generate atmospheric three-dimensional element forecast results for the forecast area in the future time period. Based on the WRF model, the OBS-Nudging method is used to introduce the actual atmospheric three-dimensional element analysis field into the WRF model to obtain a better weather forecast result; The atmospheric three-dimensional element forecast results predicted by the WRF model are sequentially subjected to three-dimensional element correction, terrain downscaling extraction and attention focus correction to obtain a three-dimensional wind field forecast result; Determining the difference between the atmospheric three-dimensional element analysis field at each moment observed at the grid point in the forecast area during the first historical time period and the atmospheric three-dimensional element analysis field predicted at the same moment at the grid point in the background data; Calculating a probability density curve of the difference, and determining a preset threshold value according to the probability density curve of the difference; When the difference is greater than a preset threshold, the atmospheric three-dimensional element analysis field of the grid point at the moment in the first historical time period is eliminated.

2. The three-dimensional wind field forecasting method based on live reanalysis data according to claim 1, characterized in that: The three-dimensional atmospheric element analysis field includes wind field, temperature field, humidity field and air pressure field.

3. The three-dimensional wind field forecasting method based on live reanalysis data according to claim 1, characterized in that: Before the initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field are used as inputs of the WRF model, the method further includes: performing sparse processing on the atmospheric three-dimensional element analysis field; Wherein, jump point sampling is performed in the horizontal direction of the atmospheric three-dimensional element analysis field; Performing thinning in the vertical direction of the atmospheric three-dimensional element analysis field based on a piecewise function; and / or, The atmospheric three-dimensional element analysis field in the second historical time period is layered in the vertical direction; Statistically analyze the three-dimensional atmospheric elements in different layers and obtain the probability density curves of the three-dimensional atmospheric elements in different layers; According to the probability density curves of the atmospheric three-dimensional element analysis fields in different layers, the extreme values of the atmospheric three-dimensional element analysis fields in different layers are determined; If the atmospheric three-dimensional element analysis field of any layer at each moment of the grid point in the forecast area during the first historical time period is greater than the corresponding extreme value, the atmospheric three-dimensional element analysis field of the grid point at that moment is eliminated; and / or, The atmospheric three-dimensional element analysis field within the first historical time period is normalized.

4. The three-dimensional wind field forecasting method based on live reanalysis data according to claim 1, characterized in that: Obtaining the forecast results of the three-dimensional atmospheric elements in the forecast area at the nearest moment from the background data, and generating the initial field and lateral boundary conditions for the WRF model based on the forecast results of the three-dimensional atmospheric elements at the nearest moment, including: Select the atmospheric three-dimensional element forecast results of the WRF framework model system or the global forecast model at the nearest moment; If the atmospheric three-dimensional element forecast results of the WRF framework model system are selected, downscaling is performed based on the WPS module and NDOWN module in the WRF model to generate the initial field and lateral boundary conditions required for the forecast; If the atmospheric three-dimensional element forecast results of the global forecast model are selected, downscaling is performed based on the WPS module and REAL module in the WRF model to generate the initial field and lateral boundary conditions required for the forecast.

5. The three-dimensional wind field forecasting method based on live reanalysis data according to claim 1, characterized in that: The initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field are used as inputs of the WRF model, and the WRF model is used to generate a forecast result of atmospheric three-dimensional elements in the forecast area in a future time period, including: Assimilate the atmospheric three-dimensional element analysis field using the OBS-Nudging module in the WRF model; The initial field, lateral boundary conditions and atmospheric three-dimensional element analysis field are used as inputs of the WRF model for assimilation integration for several hours. After the assimilation integration is completed, the WRF model begins formal free integration to generate the atmospheric three-dimensional element forecast results for the forecast area in the future time period.

6. The three-dimensional wind field forecasting method based on live reanalysis data according to claim 1, characterized in that: The three-dimensional atmospheric element forecast results predicted by the WRF model are corrected by three-dimensional elements, including: Matching the grid of the atmospheric three-dimensional element analysis field in the third historical time period with the grid of the WRF model forecast; Determine the residual of the atmospheric three-dimensional element forecast result predicted by the WRF model using the XGBoost model based on the atmospheric three-dimensional element analysis field within the third historical time period corresponding to the matched grid and the atmospheric three-dimensional element forecast result at the same time predicted by the WRF model; A final model forecast result is determined based on the residual and the forecast result of the three-dimensional atmospheric elements in the forecast area in the future time period predicted by the WRF model.

7. The three-dimensional wind field forecasting method based on live reanalysis data according to claim 6, characterized in that: Perform terrain downscaling extraction on the atmospheric three-dimensional element forecast results predicted by the WRF model, including: The final model forecast results are downscaled for complex terrain using the CALMET model, and three-dimensional wind field forecast results are extracted therefrom.

8. The three-dimensional wind field forecasting method based on live reanalysis data according to claim 7, characterized in that: Focus on correcting the three-dimensional atmospheric elements forecast results of the WRF model, including: The pre-acquired 3D wind farm forecast results are used as training samples, and the corresponding actual wind farm observation data are used as labels. The LSTM model with attention mechanism is trained to obtain the wind farm revision model. The three-dimensional wind field forecast result extracted by the CALMET model is input into the wind field revision model to obtain a revised three-dimensional wind field forecast result.

9. A three-dimensional wind field forecast system based on live reanalysis data, characterized in that: The three-dimensional wind field forecasting method based on live reanalysis data as described in any one of claims 1 to 8 comprises: An extraction module is used to obtain the atmospheric three-dimensional element analysis field of the forecast area within the first historical time period from the real-time data; A generation module is used to obtain the forecast results of the three-dimensional atmospheric elements in the forecast area at the most recent time from the background data, and generate the initial field and lateral boundary conditions for the WRF model according to the forecast results of the three-dimensional atmospheric elements at the most recent time; A prediction module, configured to use the initial field, lateral boundary conditions, and atmospheric three-dimensional element analysis field as inputs to the WRF model, and generate a forecast result of the atmospheric three-dimensional elements in the forecast area within a future time period using the WRF model; The post-processing module is used to perform three-dimensional element correction, terrain downscaling extraction and attention concentration correction on the atmospheric three-dimensional element forecast results predicted by the WRF model in sequence to obtain a three-dimensional wind field forecast result.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the three-dimensional wind field forecasting method based on live reanalysis data as described in any one of claims 1 to 8 is implemented.

Images