Multi-modal fusion weather forecasting method and apparatus

Abstract

The application provides a multi-mode fused weather forecast method and device, relates to the meteorological technical field, and solves the problem of poor forecast effect caused by the difficulty in adaptive fusion according to the actual weather situation during multi-mode fusion by processing the original grid output data of N numerical modes in an adaptive weight fusion manner. The adaptive weight fusion manner comprises the following steps: independently calculating a target diagnostic parameter field by using the original output grid data of each numerical mode; classifying the original grid output data of each numerical mode according to a preset weather situation category to determine a dominant weather situation; obtaining a target historical performance score of a target meteorological element predicted by each numerical mode under a target prediction time limit and the dominant weather situation; normalizing the target historical performance scores corresponding to the N numerical modes to obtain a weight vector; and weighting and fusing the target diagnostic parameter fields corresponding to the N numerical modes according to the weight vector.

Description

Technical Field

[0001] This application relates to the field of meteorological technology, and more specifically to a multi-model fusion meteorological forecasting method, apparatus, equipment, medium, and program product. Background Technology

[0002] Weather forecasting is crucial for many aspects, such as agricultural planting, transportation, and disaster prevention. Taking aviation weather forecasting as an example, aircraft move at high speeds in three-dimensional space and are extremely sensitive to weather conditions. In particular, small-to-medium-scale weather phenomena such as low-level wind shear, clear-air turbulence, and aircraft icing have become major meteorological factors threatening aviation safety, necessitating the most accurate weather forecasts possible. A widely accepted technical approach to improving the accuracy of weather forecasts is to post-process the output of numerical models, such as by integrating and fusing forecast results from multiple numerical models to overcome the biases or inaccuracies of forecasts from a single numerical model.

[0003] However, during the conceptualization of this invention, it was discovered that existing multi-model fusion methods suffer from insufficient adaptability, failing to achieve true "intelligent selection." Specifically, existing multi-model fusion methods (including some simple machine learning fusion models) often employ fixed weights or weights calculated based on long-term (e.g., one-year) average performance. However, numerical models exhibit significant weather system dependence. For example, numerical model A excels in forecasting typhoon-related weather, but its forecasts of Meiyu frontal precipitation may be systematically weak; numerical model B may be superior in global mid-latitude system forecasts, but its ability to capture high-altitude weather is poor. Current multi-model fusion methods typically cannot adaptively adjust which models to trust based on the current actual weather conditions, resulting in the fusion effect potentially being inferior to a single dominant model under specific weather conditions. Summary of the Invention

[0004] In view of the above problems, embodiments of this application provide a multi-model fusion weather forecasting method, apparatus, equipment, medium, and program product that can fuse forecast results from multiple models through adaptive weights.

[0005] A first aspect of this application provides a multi-model fusion weather forecasting method. The method includes: acquiring raw grid output data from N numerical models, each based on a first forecast time point and according to a target forecast lead time, to generate N target raw grid output data, where N is an integer greater than 1; processing the N target raw grid output data using an adaptive weighted fusion method to obtain a first fusion forecast product; and generating a weather forecast based on the first fusion forecast product. The adaptive weight fusion method includes: calculating a target diagnostic parameter field using the original grid output data of each numerical model according to the calculation method of the target diagnostic parameters, wherein the target diagnostic parameters are used to forecast target meteorological elements; wherein, N sets of the target diagnostic parameter fields are obtained for each of the N numerical models; classifying the original grid output data of each numerical model according to a preset weather situation category to obtain the dominant weather situation; obtaining the target historical performance score of each numerical model for forecasting the target meteorological elements under the target forecast lead time and the dominant weather situation; normalizing the target historical performance scores obtained for each of the N numerical models to obtain a weight vector; and weighting and fusing the N sets of target diagnostic parameter fields according to the weight vector and then outputting the result.

[0006] According to this embodiment, the determined fusion weights can be adapted to the dominant weather conditions corresponding to each numerical model, thereby dynamically selecting whether to trust the model's output more based on the actual performance of the model under the dominant weather conditions, thus improving the accuracy of the final fusion forecast product.

[0007] According to an embodiment of this application, obtaining the target historical performance score of each numerical model for forecasting the target meteorological element under the target forecast lead time and the dominant weather situation includes: for any first numerical model, using the identifier of the first numerical model, the dominant weather situation corresponding to the first numerical model, the target forecast lead time, and the target meteorological element as query inputs, querying a knowledge base and obtaining the target historical performance score returned by the knowledge base; wherein the knowledge base records mapping relationships, the inputs of the mapping relationships include the dominant weather situation, forecast lead time, the identifier of the numerical model, and the meteorological element, and the output of the mapping relationships is the historical performance score obtained by mapping based on the inputs.

[0008] According to this embodiment, by storing mapping relationships in a knowledge base and then quickly calculating weight vectors by querying the knowledge base in real time, efficient and fast adaptive weight fusion is achieved, thus improving fusion efficiency.

[0009] According to an embodiment of this application, the knowledge base is constructed as follows: For any second numerical model among N numerical models, historical raw grid output data of the second numerical model are collected; the historical raw grid output data are classified according to the weather situation category to obtain historical dominant weather situations; a historical diagnostic parameter field is calculated using the historical raw grid output data according to the calculation method of each diagnostic parameter, wherein different diagnostic parameters are used to forecast different types of meteorological elements; the historical diagnostic parameter fields corresponding to the same diagnostic parameters calculated from the historical raw grid output data with the same historical dominant weather situation and forecast lead time are aggregated together; based on the comparison and statistics of the forecast results of the corresponding meteorological elements by the aggregated historical diagnostic parameter fields and the actual meteorological observation results, the historical performance score of the second numerical model for forecasting the corresponding meteorological elements under the historical dominant weather situation and forecast lead time is obtained; the identifier of the numerical model, dominant weather situation, meteorological elements and forecast lead time in the calculation process of the historical performance score are extracted to form the mapping relationship; the mapping relationship is stored in the knowledge base.

[0010] According to this embodiment, a knowledge base can be constructed that stores the historical performance scores of each numerical model for specific meteorological elements under different dominant weather conditions and within different forecast timeframes, providing a decision-making basis for adaptive weight fusion.

[0011] According to an embodiment of this application, the method further includes: periodically or in response to an event triggering to collect and add the historical original grid output data; and in response to the addition of the historical original grid output data, updating the mapping relationship.

[0012] According to this embodiment, the constructed knowledge base is capable of automatically tracking changes in the performance of various numerical patterns and updating the mapping relationship, and has strong adaptability and rapid response capabilities.

[0013] According to an embodiment of this application, the historical performance score is characterized by a key performance index.

[0014] According to this embodiment, using the Threat Score (TS score) to represent the historical performance score can reduce subjective bias in quantifying historical performance.

[0015] According to an embodiment of this application, the weather forecasting based on the first fusion forecast product includes: acquiring a second fusion forecast product, wherein the second fusion forecast product is obtained by processing the original grid output data of N numerical models, each at a second reporting time point according to the target forecast lead time, for the target spatial region, using the adaptive weight fusion method, wherein the second reporting time point is earlier than the first reporting time point; extracting grid data at multiple actual observation station locations from the second fusion forecast product to obtain a discrete grid dataset; calculating the error between the grid data in the discrete grid dataset and the observation point data at the same location and time to obtain a discrete grid error set, wherein the observation point data is obtained based on the observation data of the actual observation stations; processing the data in the discrete grid error set using an interpolation method to obtain a grid error field covering the grid field corresponding to the target spatial region; correcting the first fusion forecast product using the grid error field to obtain a target forecast product; and performing weather forecasting based on the target forecast product.

[0016] According to this embodiment, the systematic error of the fusion forecast product is estimated by using the previously obtained second fusion forecast product, which is then used to correct the current first fusion forecast product. This can quickly eliminate the systematic position and intensity biases of the numerical model, achieve high-precision weather forecasts, and use only recent data without the need for long-term training.

[0017] According to an embodiment of this application, the weather forecasting based on the target forecast product includes: obtaining a preset hazardous weather level threshold configuration table; dividing the target spatial region into different risk levels based on a comparison between the thresholds of the target diagnostic parameters in the hazardous weather level threshold configuration table and the grid data of the target diagnostic parameters in the target forecast product, so as to obtain the forecast results of the target meteorological elements; and outputting the forecast results of the target meteorological elements in a visual manner.

[0018] According to this embodiment, the visual output of forecast results can bridge the "last mile" of weather forecasting, achieving operational automation and efficiency improvement.

[0019] According to an embodiment of this application, classifying the raw grid output data of each numerical model according to a preset weather situation category includes: classifying by clustering or according to preset expert rules.

[0020] According to this embodiment, a relatively objective classification of weather conditions can be achieved, and the range of dominant weather conditions can be limited to the scope of user concern by preset weather condition categories.

[0021] According to embodiments of this application, the weather situation categories include one or more of the following: typhoon, cold front, warm-sector precipitation, stationary front, cold high pressure, southwest vortex, or clear sky. These weather situation categories are of great interest in aviation meteorological forecasting, thus enabling embodiments of this application to be applied to the field of aviation meteorological forecasting.

[0022] According to embodiments of this application, the target diagnostic parameter is one of a plurality of diagnostic parameters, including at least one of the following: turbulence index or icing index. This allows embodiments of this application to be used for predicting small- to medium-scale weather phenomena such as clear-sky turbulence and aircraft icing.

[0023] A second aspect of this application provides a multi-model fusion weather forecasting device. The device includes an acquisition module, a fusion module, and a forecasting module. The acquisition module acquires raw grid output data from N numerical models, each forecasting a target spatial area based on a first reporting time point and according to the target forecast lead time, resulting in N target raw grid output data, where N is an integer greater than 1. The fusion module processes the N target raw grid output data through an adaptive weighted fusion submodule to obtain a first fused forecast product. The forecasting module performs weather forecasting based on the first fused forecast product. The adaptive weight fusion submodule is used to process data according to an adaptive weight fusion method, specifically including: calculating a target diagnostic parameter field using the original grid output data of each numerical model according to the calculation method of the target diagnostic parameters, wherein the target diagnostic parameters are used to forecast target meteorological elements; wherein, N sets of the target diagnostic parameter fields are obtained for each of the N numerical models; classifying the original grid output data of each numerical model according to a preset weather situation category to obtain the dominant weather situation; obtaining the target historical performance score of each numerical model for forecasting the target meteorological elements under the target forecast lead time and the dominant weather situation; normalizing the target historical performance scores obtained for each of the N numerical models to obtain a weight vector; and weighting and fusing the N sets of target diagnostic parameter fields according to the weight vector and then outputting the result.

[0024] The apparatus according to this embodiment can adapt the determined fusion weights to the dominant weather conditions corresponding to each numerical model, thereby adaptively selecting whether to trust the output of the model more based on the actual performance of the model under the dominant weather conditions, thus improving the accuracy of the final fusion forecast product.

[0025] According to a third aspect of this application, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0026] According to a fourth aspect of this application, a computer-readable storage medium is also provided, on which a computer program or instructions are stored, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0027] According to a fifth aspect of this application, a computer program product is also provided, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0028] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0029] Figure 1 A schematic diagram illustrating the overall flowchart of a multi-model fusion weather forecasting method according to an embodiment of this application is shown.

[0030] Figure 2 The flowchart illustrating the adaptive weight fusion method in the multi-mode fusion weather forecasting method according to an embodiment of this application is shown in the illustration.

[0031] Figure 3 This illustration schematically shows a flowchart of constructing a knowledge base in a multi-mode fusion weather forecasting method according to an embodiment of this application;

[0032] Figure 4 This illustration schematically shows a flowchart of error correction for a first fused forecast product in a multi-model fusion weather forecasting method according to an embodiment of this application;

[0033] Figure 5 The diagram illustrates a process flow chart for visually outputting forecast results of meteorological elements in a multi-mode fusion weather forecasting method according to an embodiment of this application.

[0034] Figure 6 A flowchart illustrating the processing of a multi-mode fusion weather forecasting method according to another embodiment of this application is shown.

[0035] Figure 7 A block diagram of a multi-mode fusion weather forecasting device according to an embodiment of this application is shown schematically;

[0036] Figure 8 A block diagram of an electronic device suitable for implementing the multi-mode fusion weather forecasting method of the embodiments of this application is shown schematically. Detailed Implementation

[0037] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0038] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0039] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0040] This application provides a multi-model fusion weather forecasting method, device, electronic device, medium, and program product. The core of this method is to dynamically allocate fusion weights based on the historical performance scores of different numerical models for specific meteorological elements under specific dominant weather conditions and forecast lead times, rather than using fixed weights, so as to achieve adaptive fusion of forecast results from multiple numerical models.

[0041] The key terms used in this article are explained below:

[0042] Numerical models are scientific computational models based on physical laws, using mathematical equations to describe atmospheric and oceanic motions, and solving them through supercomputers to predict future weather and climate change.

[0043] Diagnostic parameters refer to indices that directly characterize hazardous weather, calculated from basic meteorological elements (such as wind, temperature, and humidity) using specific physical formulas, such as the turbulence index, icing index, or low-level wind shear intensity.

[0044] Figure 1 A schematic diagram illustrates the overall flowchart of a multi-model fusion weather forecasting method according to an embodiment of this application.

[0045] like Figure 1 As shown, the multi-model fusion weather forecasting method according to this embodiment generally includes three main steps: steps S110 to S130.

[0046] Step S110: Obtain the original grid output data of N numerical models for meteorological forecasting of the target spatial area based on the first reporting time point and according to the target forecast lead time, and obtain N target original grid output data, where N is an integer greater than 1.

[0047] Step S120: Process the output data of N target original grid points according to the adaptive weight fusion method to obtain the first fused forecast product.

[0048] Step S130: Make a weather forecast based on the first fused forecast product.

[0049] Step S120 is the core of the method in this embodiment, enabling adaptive weighted fusion of the meteorological element diagnostic results obtained from each of the N numerical models. For a detailed implementation of step S120, please refer to [link to relevant documentation]. Figure 2 .

[0050] Specifically, such as Figure 2 As shown, step S120 may include steps S201 to S205.

[0051] In step S201, the target diagnostic parameter field is calculated using the original grid output data of each numerical model according to the calculation method of the target diagnostic parameters. Specifically, N sets of target diagnostic parameter fields are obtained for each of the N numerical models.

[0052] The target diagnostic parameter field is composed of the target diagnostic parameter data at each grid point in the grid field corresponding to the target spatial region, reflecting the numerical distribution of the target diagnostic parameters in the grid field corresponding to the target spatial region.

[0053] Target diagnostic parameters are used to forecast target meteorological elements. Specifically, the same meteorological element can often be forecasted using different diagnostic parameters. In this embodiment, the target diagnostic parameters are pre-selected parameters used to characterize the intensity of the target meteorological element, and therefore can be used to forecast target meteorological elements (such as turbulence, icing, or low-level wind shear). According to this embodiment, the target diagnostic parameter is one of multiple diagnostic parameters, including at least one of the following: turbulence index or icing index. Different types of diagnostic parameters are also pre-selected parameters used to forecast different meteorological elements (such as turbulence, icing, low-level wind shear, etc.). This embodiment can be applied in the field of aviation meteorology for the prediction of small- and medium-scale weather phenomena such as clear-air turbulence and aircraft icing.

[0054] In step S202, the raw grid output data of each numerical model are classified according to preset weather situation categories to obtain the dominant weather situation. A relatively objective weather situation classification can be achieved through clustering or classification according to preset expert rules.

[0055] Classifying weather conditions according to preset weather patterns can limit the range of selectable dominant weather patterns to those of interest to the user. Preset weather pattern categories can include one or more of the following: typhoon, cold front, warm-sector precipitation, stationary front, cold high-pressure system, southwest vortex, or clear sky. These weather pattern categories are of great interest in aviation meteorological forecasting, which is why the embodiments of this application can be applied to the field of aviation meteorological forecasting.

[0056] In step S203, the historical performance score of each numerical model for forecasting target meteorological elements under the target forecast lead time and the aforementioned dominant weather conditions is obtained.

[0057] The historical performance score of a target is used to quantify the reliability or accuracy of the corresponding numerical model's forecasts of target meteorological elements under the target's preset lead time and the corresponding dominant weather patterns in history. In one embodiment, the historical performance score of the target can be represented by the Key Performance Index (TS score). The TS score has advantages such as objective quantification, comparability, and stability, and can reduce subjective bias in quantifying historical performance.

[0058] In step S204, the target historical performance scores obtained for each of the N numerical modes are normalized to obtain a weight vector.

[0059] In step S205, the N sets of target diagnostic parameter fields are weighted and fused according to the above weight vector and then output. This yields a fused forecast product corresponding to the target diagnostic parameters.

[0060] The embodiments of this application can overcome the problem in related technologies where multi-model fusion cannot adaptively adjust which models to trust based on the current weather conditions, leading to inaccurate forecasts. Therefore, the determined fusion weights are adapted to the dominant weather conditions corresponding to each numerical model. In this way, if a numerical model has a strong ability to predict target meteorological elements under its corresponding dominant weather conditions, its weight will be higher. This allows for dynamic selection of whether to trust the output of a model more based on its actual performance under the dominant weather conditions, thereby improving the accuracy of the final fusion forecast product.

[0061] In one embodiment, step S203 can obtain the target historical performance score by querying a specially constructed knowledge base. Specifically, any numerical mode from N numerical modes is used as the first numerical mode. The identifier of the first numerical mode, the dominant weather situation corresponding to the first numerical mode, the target forecast lead time, and the target meteorological elements are used as query inputs to query the knowledge base and obtain the target historical performance score returned by the knowledge base. The knowledge base records mapping relationships. The inputs of the mapping relationships include the dominant weather situation, forecast lead time, the identifier of the numerical mode, and meteorological elements. The output of the mapping relationships is the historical performance score obtained by mapping based on the inputs. This embodiment of the application achieves efficient and fast adaptive weight fusion by storing mapping relationships in the knowledge base and querying the knowledge base when making weather forecasts, thereby improving fusion efficiency. The mapping relationships stored in the knowledge base can be updated offline (e.g., periodically or after calculating the mapping relationships offline based on user operations and then updating them to the knowledge base) or can be updated online in real time (e.g., continuously accumulating new data and synchronously updating the mapping relationships).

[0062] In one embodiment, the knowledge base construction process used in step S203 can be found in [reference needed]. Figure 3 . Specifically, Figure 3 The flowchart illustrating the construction of a knowledge base in a multi-mode fusion weather forecasting method according to an embodiment of this application is shown.

[0063] like Figure 3 As shown, the construction process of this knowledge base may include steps S310 to S360.

[0064] In step S310, for any second numerical mode among the N numerical modes, collect the historical original grid output data of the second numerical mode.

[0065] In step S320, the historical raw grid output data is classified according to weather situation categories to obtain the historical dominant weather situations. The classification method is the same as in step S202 above, such as classification by clustering or expert rules.

[0066] In step S330, a historical diagnostic parameter field is calculated using the historical raw grid output data according to the calculation method of each diagnostic parameter, wherein different diagnostic parameters are used to forecast different types of meteorological elements.

[0067] In step S340, the historical diagnostic parameter fields corresponding to the same diagnostic parameters calculated from the historical original grid output data with the same historical dominant weather patterns and forecast lead time are aggregated together. Based on the comparison and statistics of the forecast results of the corresponding meteorological elements and the actual meteorological observation results of the aggregated historical diagnostic parameter fields, the historical performance score of the second numerical model for forecasting the corresponding meteorological elements under the historical dominant weather patterns and forecast lead time is obtained (such as calculating the TS score of the forecasting corresponding meteorological elements).

[0068] In step S350, the identifiers of numerical patterns, dominant weather conditions, meteorological elements, and forecast lead time in the calculation process of historical performance scores are extracted to form a mapping relationship.

[0069] In step S360, the above mapping relationship is stored in the knowledge base.

[0070] This embodiment can construct a knowledge base by collecting and statistically analyzing historical data. It stores historical performance scores of various numerical models for specific meteorological elements under different dominant weather conditions and within different forecast periods, providing a basis for decision-making for adaptive weight fusion.

[0071] According to embodiments of this application, step S310 can be executed periodically or in response to an event to collect and add historical raw grid output data. Then, in response to the addition of historical raw grid output data, steps S320 to S360 can be re-executed to update the mapping relationship in the knowledge base. Events triggering step S310 may include, but are not limited to, user operations or the operation in step S203 above to query the knowledge base to obtain the target historical performance score. For example, when querying the knowledge base in step S203 above, new historical raw grid output data can be collected through step S310. If collected, the mapping relationship in the knowledge base is updated through steps S320 to S360, thus ensuring high real-time performance of the target historical performance score obtained in step S203.

[0072] The knowledge base constructed in this application embodiment can be an online learning system that can automatically track changes in the performance of various numerical modes and update the mapping relationship in a rolling manner. In this way, if the performance of a certain numerical mode improves, its weight will automatically increase during fusion. The system can enjoy the benefits brought by the progress of the numerical mode without any manual intervention, and has strong adaptability and fast response capability.

[0073] According to some embodiments of this application, when making a weather forecast based on the first fused forecast product in step S130 above, error correction and / or visualization processing can also be performed on the first fused forecast product obtained in step S120 to form a forecast result that can be provided to the end user.

[0074] Figure 4 The illustration shows a flowchart of error correction for a first fused forecast product in a multi-model fusion weather forecasting method according to an embodiment of this application.

[0075] like Figure 4 As shown, according to this embodiment, step S130 may further include steps S401 to S406.

[0076] In step S401, at least one second fusion forecast product is obtained. The second fusion forecast product is the original grid output data obtained by each of the N numerical models performing meteorological forecasts for the target spatial region at the second forecast time point according to the target forecast lead time, and fused according to an adaptive weighting method (see details). Figure 2 The processing flow shown is as follows, in which the second reporting time is earlier than the first reporting time.

[0077] In step S402, grid data from multiple actual observation station locations are extracted from the second fused forecast product to obtain a discrete grid dataset. These multiple actual observation stations are meteorological observation stations distributed across the target spatial region.

[0078] In step S403, the error between the grid data in the discrete grid dataset and the observation point data at the same location and time is calculated to obtain the discrete grid error set, wherein the aforementioned observation point data is obtained based on the observation data of the actual observation station.

[0079] In one embodiment, when multiple second fusion forecast products are obtained in step S401, such as fusion forecast products from the previous 2 to 3 days, the error can be calculated based on the multiple second fusion forecast products in step S403, and then the average value can be taken.

[0080] In step S404, the data in the discrete grid error set is processed using an interpolation method to obtain the grid error field of the grid field corresponding to the target spatial region. The interpolation method used can be the optimal interpolation method (Optimal Interpolation, abbreviated as OI) or any other interpolation method.

[0081] The grid error field is composed of the fitting error data of the target diagnostic parameters at each grid point in the grid field corresponding to the target spatial region. It reflects the spatial distribution of the fitting error data of the target diagnostic parameters in the grid field corresponding to the target spatial region and is used to estimate the systematic errors of the fused forecast products (such as systematic position bias and intensity bias).

[0082] In step S405, the first fused forecast product is corrected using the grid error field to obtain the target forecast product. Specifically, the first fused forecast product is added to the grid data of the same grid points in the grid error field to obtain the target forecast product.

[0083] Finally, in step S406, a weather forecast is made based on the target forecast product.

[0084] In one embodiment, based on the logical assumption that the systematic errors of the data model in the near term (e.g., 1-3 days) will remain persistent in the near future, a gridded error field can be obtained using second fusion forecast products obtained in the near term (e.g., several forecast times prior to the first forecast time). This gridded error field can then be used to correct the current required first fusion forecast product. In another embodiment, the average error of all second fusion forecast products within a relatively long preset period (e.g., one month) prior to the first forecast time can be selected to calculate the gridded error field. This method can reduce or eliminate the randomness of errors corresponding to a single or small number of second fusion forecast products. By using at least one second fusion forecast product to calculate the gridded error field and correcting the current first fusion forecast product accordingly, the systematic positional and intensity biases of the numerical model can be quickly eliminated, thereby achieving high-precision weather forecasting.

[0085] In some embodiments of this application, a correction process can be implemented by utilizing recent errors, using the OI algorithm to generate a gridded error field, and applying the error field to the current forecast. The "fast" aspect is reflected in the fact that only recent data can be used, without the need for long-term training.

[0086] The error correction scheme in this application has significant advantages in the field of aviation meteorology, meeting the needs for rapid response and immediate effectiveness. Specifically, traditional statistical correction methods such as Model Output Statistics (MOS) require years of historical "model-real-time" paired data for model training. When the numerical model version is upgraded (e.g., physical process adjustments, resolution improvements), the model's systematic error characteristics change, rendering the original MOS equations invalid. Reconstructing the MOS model requires accumulating several years of new version model data, a very long period, meaning that effective correction products will be unavailable for a considerable time after the model update. Furthermore, for sudden and rare severe weather events (such as certain extreme convective weather), there may be no similar cases in historical samples, causing models based on historical statistics to completely fail. In contrast, some embodiments of this application can rely solely on recent (e.g., the output data of the last 1-3 days) numerical models. This means that after a numerical model version update, the ability to correct errors for the new model can be established in a maximum of 3 days, solving to some extent the problem of the MOS method's months-long lag. In the event of sudden weather events, the embodiments of this application can quickly capture the latest error characteristics of the model and correct them.

[0087] In one embodiment, step S406 above can visualize and publish the target forecast product obtained after error correction, such as... Figure 5 As shown.

[0088] Specifically Figure 5 The diagram illustrates a process flow chart for visually outputting meteorological element forecast results in a multi-mode fusion meteorological forecasting method according to an embodiment of this application.

[0089] like Figure 5 As shown, according to this embodiment, step S406 may include steps S501 to S503.

[0090] In step S501, a preset hazardous weather level threshold configuration table is obtained.

[0091] In step S502, based on the comparison between the threshold of the target diagnostic parameters in the hazardous weather level threshold configuration table and the grid data of the target diagnostic parameters in the target forecast product, the target spatial area is divided into different risk levels to obtain the forecast results of the target meteorological elements.

[0092] In step S503, the forecast results of the target meteorological elements are output in a visual manner.

[0093] According to the embodiments of this application, the forecast results can be output visually, which can bridge the "last mile" of weather forecasting and realize business automation and efficiency improvement.

[0094] The following is passed Figure 6 An exemplary embodiment of the multi-mode fusion weather forecasting method according to the embodiments of this application is described in the aviation field.

[0095] like Figure 6 As shown, the multi-mode fusion weather forecasting method of this embodiment may include steps S1 to S4.

[0096] Step S1: Calculate the diagnostic parameter field independently using the original grid output data of each numerical model, realizing the first transformation from general elements to aviation-specific parameters.

[0097] First, input data is acquired. Specifically, after determining the target forecast lead time (e.g., the next 24 hours) and the target spatial region (e.g., an airport or a complete spatial region including the departure and arrival airports and the entire airway), raw gridded output data from various global and regional numerical models (such as global or regional forecasting systems for different regions, Weather Research and Forecasting Models (WRFs)) are obtained from the data server. The required variables in the raw gridded output data include at least: multi-level U / V wind components, temperature, relative humidity, geometrical altitude, vertical velocity, cloud water content, and ice water content.

[0098] The next step is to process the input data. This process involves the system initiating parallel computing tasks to independently calculate the diagnostic parameter field using the raw grid output data for each numerical model. The computation process eliminates the need for inter-model interaction, significantly improving efficiency.

[0099] A diagnostic parameter field is the distribution data of a diagnostic parameter in a gridded field within a target spatial region. The diagnostic parameters selected are specialized diagnostic parameters from the aerospace field, such as turbulence index and icing index. Different diagnostic parameter fields are calculated for different diagnostic parameters.

[0100] Regarding the Turbulence Index (TI index): it can be indicated by the Ellrod index, which is a dynamically related turbulence index, obtained by simplifying the frontogeneity function. It has two forms: Ellrod1 (…). ) and Ellrod2 ( The Ellrod index is calculated as follows:

[0101]

[0102]

[0103] in

[0104] This refers to the total deformation caused by strong winds (including horizontal shear and tensile deformation). , Let x be the component of the horizontal wind speed in the x-direction. This represents the component of the horizontal wind speed in the y-direction;

[0105] For divergence term, ;

[0106] This is vertical wind shear. .

[0107] Calculations can be performed on multiple isobaric surfaces (such as 300hPa and 250hPa isobaric surfaces, used to forecast clear-sky turbulence).

[0108] The icing index can be calculated using the Icing Index (IC) method. Aircraft are most prone to icing when encountering large supercooled water droplets while flying in the 0°C to −14°C range, with the strongest icing zone occurring between −5°C and −9°C. The IC index method is based on modeling and calculating the temperature and humidity ranges conducive to icing. The formula is as follows:

[0109]

[0110] Where RH is relative humidity (%) and T is temperature (°C).

[0111] The final output is N sets of diagnostic parameter fields for each diagnostic parameter. For example, the N sets of turbulence fields include the turbulence field for numerical mode A, the turbulence field for numerical mode B, and so on; or the N sets of icing fields include the icing field for numerical mode A, the icing field for numerical mode B, and so on. In other words, the diagnostic parameter fields for each of the N numerical modes are calculated independently.

[0112] This marks the first successful transformation of raw grid output data from N numerical models from general features to aerospace-specific parameters.

[0113] Step S2: Process the N sets of diagnostic parameter fields corresponding to the target diagnostic parameters according to the adaptive weight fusion method.

[0114] First, input the N sets of diagnostic parameter fields generated in step S1, which are the target diagnostic parameters (such as the turbulence index).

[0115] The processing procedure for these N sets of diagnostic parameter fields is as follows:

[0116] 1) Objective Weather Situation Identification: Based on the raw grid output data of each numerical model, the system obtains a multi-model analysis field (such as sea level pressure field, 500 hPa geopotential height field, and 850 hPa temperature field). Then, a rule-based expert system or clustering algorithm is used to process this multi-model analysis field, automatically classifying the dominant weather situation currently affecting the target spatial region. The weather situation categories in the classification system are predefined, and may include, for example, typhoons, cold fronts, warm-sector precipitation, stationary fronts, cold high pressure systems, southwest vortices, and clear skies. Determining the dominant weather situation to which each numerical model forecast belongs is a prerequisite for dynamic decision-making.

[0117] 2) Query the specially constructed knowledge base.

[0118] This knowledge base can be used as a reference. Figure 3The process described is as follows. Specifically, for each numerical model, the identifier of the numerical model, the classification label of the dominant weather pattern identified for that numerical model (e.g., "cold front"), the forecast lead time (e.g., "24 hours"), and the target meteorological element (e.g., clear-sky turbulence) are used as query elements to initiate a query to a knowledge base based on "numerical model - dominant weather pattern - forecast lead time - meteorological element". This knowledge base, acting as an intelligent decision-making center, outputs a historical performance score for each set of query elements.

[0119] This allows us to retrieve N historical performance scores for each of the N numerical models. Then, by normalizing these N historical performance scores, we obtain a weight vector optimized for the current forecast scenario. , where any weight w i This represents the proportion of the historical performance of the i-th numerical model under the "target meteorological element-target forecast lead time" combination, and the sum of all weights is 1, thereby achieving the optimal model combination configuration.

[0120] The construction and maintenance of the aforementioned knowledge base can be an online learning process with real-time updates, or it can be an offline, periodic self-learning process. Historical data can be continuously collected over a configurable period (e.g., one year), including forecast data from various numerical models under different weather pattern classifications and observed field data (usually based on high-precision reanalysis data or multi-source fusion analysis fields).

[0121] For each combination of "numerical model - dominant weather pattern - forecast lead time - meteorological elements", a historical performance score is obtained by comparing and statistically analyzing historical forecasts with actual conditions. For example, the key score index (TS score) can be used as a historical performance score for diagnostic parameters such as turbulence index or icing index.

[0122] Example of TS score calculation: Based on the historical turbulence indices of each numerical model and the turbulence indices calculated from observational data, the TS scores of the turbulence indices of each numerical model under different dominant weather conditions and different forecast lead times are calculated as follows:

[0123]

[0124] Where a represents the number of times that the forecast was correct and the observation was also correct; b represents the number of times that the forecast was correct but the observation was incorrect; and c represents the number of times that the forecast was incorrect but the observation was correct.

[0125] A mapping relationship can be formed based on the numerical model, dominant weather pattern, forecast lead time, and meteorological elements in the calculation process of each historical performance score (such as TS score): f(numerical model, dominant weather pattern, forecast lead time, meteorological elements) = historical forecast score.

[0126] As you can see, the knowledge base itself is equivalent to a multi-dimensional rating index table. The core content is to map the input numerical patterns, dominant weather conditions, forecast lead time and meteorological elements to historical performance ratings (such as TS ratings). When querying, you can search according to the mapping relationship.

[0127] After retrieving N historical performance ratings from the knowledge base, these ratings are normalized to obtain a weight vector W. Taking TS ratings as an example, the formula for calculating the weight vector W is: Weight of each numerical pattern = TS rating corresponding to that numerical pattern / Sum of TS ratings corresponding to all numerical patterns.

[0128] 3) Weighted fusion: Using the queried weight vector W, the N sets of diagnostic parameter fields (such as model1, model2, ..., model...) are weighted together. n A weighted average calculation is performed to generate the first fusion forecast product. ,as follows:

[0129]

[0130] For example, historical statistics show that during the past 30 cold front weather events, the average TS scores for the turbulence index of numerical models A, B, and C were 0.6, 0.3, and 0.1, respectively. After normalization, the weight vectors corresponding to numerical models A, B, and C are W = (0.6, 0.3, 0.1). Then, the turbulence fields of numerical model A, numerical model B, and numerical model C can be weighted and fused based on this weight vector.

[0131] The adaptive weight fusion mechanism of this application embodiment can perform intelligent weight allocation, ensuring that it can dynamically reflect the performance differences of each model under different dominant weather conditions.

[0132] Finally, the first fused forecast product is output. This first fusion forecast product has incorporated the advantages of various models under the current weather conditions, and is theoretically a relatively optimized initial estimate.

[0133] Step S3: Construct a grid error field and use the grid error field to quickly correct the biases of the fused forecast products.

[0134] First, input the first fused forecast product. And actual observation data from multiple actual observation stations.

[0135] Actual observation data can come from meteorological observations from platforms such as airport automatic weather stations, aircraft messages, wind profiler radar, radiosondes, and microwave radiometers.

[0136] Next, we construct the lattice error field.

[0137] The first step is to calculate the recent error. This involves acquiring one or more second fused forecast products. Such as fusion forecast products with the same forecast lead time within the last 1-3 days. Among them, the second fusion forecast product... With the first fusion forecast product The calculation process is the same, but the starting time points for the two are different. Second fusion forecast product It can be regarded as the first fusion forecast product Historical values.

[0138] Through each second fusion forecast product Based on the actual analysis field obtained from the observation data of each actual observation station at the corresponding time, the error in the position of each actual observation station is calculated: Where, obs is the actual analysis field obtained based on observation data from various actual observation stations. If there are multiple In this case, the errors can be calculated separately and then averaged.

[0139] The second step is to generate the gridded error field. Since actual observation stations are sparse, therefore... The error data in the first fused forecast product is discrete, and the error data needs to be corrected. Since the entire grid field is covered, an interpolation method (such as optimal interpolation) is used to convert the discrete error data obtained in the first step. This constructs a continuous grid-point error field E covering the entire region. The optimal interpolation method (OI algorithm) can take into account the spatial distribution density of actual observation stations and the correlation between errors, and can give statistically optimal analysis results.

[0140] Finally, the grid error field E is applied to correct the first fused forecast product. The grid error field E obtained from the analysis is used as a correction factor and applied to the first fused forecast product. Above, generate target fusion forecast products ,in The logic here is: assuming that recent systematic errors in the data model will persist for a short period, past errors can be used to correct future forecasts in a rapidly corrected target fusion forecast product. It can achieve high-precision aviation hazardous weather forecasting.

[0141] Step S4: Final product generation and release.

[0142] First, input the target fusion forecast product. .

[0143] Then, the target fusion forecast product Threshold determination and objective forecasting are performed.

[0144] Specifically, a pre-set aviation hazardous weather level threshold configuration table can be read. This configuration table may be formulated with reference to relevant technical specifications of various meteorological organizations, civil aviation organizations, or aviation authorities. Based on this configuration table, target fusion forecast products can be... The gridded diagnostic parameters are used to perform fully automatic and objective level judgment.

[0145] Example 1 (Bumpy): Based on a preset bump index threshold, the grid field is divided into different risk levels. For example, when the value of the bump index TI is 3 < TI ≤ 6, the grid point is determined to be a "moderate bumpy area"; when TI > 6, it is determined to be a "severe bumpy area".

[0146] Example 2 (Ice Accumulation): Based on the preset threshold of the ice accumulation index IC, when IC > 50, the grid point is determined as a "severe ice accumulation area".

[0147] Based on the above judgment results, a corresponding binary risk flag field can be automatically generated (e.g., moderate turbulence = 1, severe turbulence = 2, no risk = 0), which can be used as a direct input source for graphic and message products.

[0148] Finally, the forecast results, after thresholding and risk level classification, can be formatted and output as products. For example, graphical products can be generated, such as automatically drawing turbulence warning maps and icing potential forecast maps, and can also overlay terrain and route information, outputting in various image formats. Another example is generating digital message products; automatically compiling and sending digital products conforming to relevant standards, as well as telegrams conforming to aviation data link requirements, which can be distributed through dedicated aviation networks. Yet another example is providing Application Programming Interface (API) services: providing API interfaces in JavaScript Object Notation (JSON) format to provide real-time warning information for direct use by airline flight decision-making systems or air traffic management systems. This can output various forms of forecast products that can be directly used for aviation decision-making.

[0149] Step S4 above bridges the "last mile" of weather forecasting, enabling operational automation and efficiency improvements. Forecast products can be delivered directly to users through visualization, such as automatically generating standardized, operational final products like graphical warning maps and directly distributable professional aviation messages. This largely solves the pain point of forecasters manually interpreting model data and manually compiling reports. Furthermore, the entire process is automated, achieving unattended automation from data to usable reports. This frees forecasters from hours of manual labor, allowing products to be produced within minutes, significantly reducing subjective errors, improving forecast timeliness and consistency, and enabling efficient operations.

[0150] The embodiments of this application can be used for aviation weather forecasting, which is a key technology for ensuring flight safety, improving operational efficiency, or saving economic costs. The method of the embodiments of this application not only absorbs the advantages of multi-mode forecasting but also maintains physical interpretability, and can largely meet the requirements of high-precision aviation weather forecasting that requires rapid adaptive capabilities.

[0151] and Figure 6 The example weather forecasting method is highly interpretable, with high operational trust and acceptance: the entire process is based on a clear physical-statistical framework that aligns with forecasters' thinking habits: "weather situation identification - historical performance query - optimal interpolation correction." Forecasters can fully trace the reasons behind each forecast conclusion (e.g., "Because the current situation is a cold front, numerical model A has historically performed best, so its weight is given to 0.7; and based on yesterday's forecast error, the system determines that the turbulence index for this region needs to be generally increased by 0.5 units"). This white-box nature makes its output easily understood and trusted by forecasters and decision-makers, thus truly integrating it into the operational workflow, offering advantages that black-box machine learning models cannot match.

[0152] The method in this application embodiment is based on a post-processing framework of adaptive weight fusion and fast bias correction, which has many advantages in terms of forecast accuracy, interpretability, adaptability, operational cost and / or efficiency. It can improve the accuracy of weather forecasts and provide favorable support for aviation weather warnings, etc.

[0153] Based on the multi-model fusion weather forecasting method of the above embodiments, embodiments of this application also provide a multi-model fusion weather forecasting device. The following will be combined with... Figure 7 The device is described in detail.

[0154] Figure 7 A block diagram of a multi-mode fusion weather forecasting device 1000 according to an embodiment of this application is shown schematically.

[0155] like Figure 7As shown, the multi-mode fusion weather forecasting device 1000 of this embodiment includes an acquisition module 1010, a fusion module 1020, and a forecasting module 1030.

[0156] The acquisition module 1010 is used to acquire the original grid output data of N numerical models, each based on the first reporting time point and according to the target forecast lead time, for meteorological forecasting of the target spatial area, thus obtaining N target original grid output data, where N is an integer greater than 1. In one embodiment, the acquisition module 1010 can be used to execute step S110 described above, which will not be repeated here.

[0157] The fusion module 1020 includes an adaptive weight fusion submodule 1021. The fusion module 1020 processes the output data of N target raw grid points through the adaptive weight fusion submodule 1021 to obtain a first fused forecast product. In one embodiment, the fusion module 1020 can be used to perform step S120 described above, which will not be repeated here.

[0158] The adaptive weight fusion submodule 1021 is used to process data according to an adaptive weight fusion method, specifically including: calculating a target diagnostic parameter field using the original grid output data of each numerical model according to the calculation method of the target diagnostic parameters, wherein the target diagnostic parameters are used to forecast target meteorological elements; wherein, N sets of target diagnostic parameter fields are obtained for each of the N numerical models; classifying the original grid output data of each numerical model according to a preset weather situation category to obtain the dominant weather situation; obtaining the target historical performance score of each numerical model under the target forecast lead time and the dominant weather situation; normalizing the target historical performance scores obtained for each of the N numerical models to obtain a weight vector; and weighting and fusing the N sets of target diagnostic parameter fields according to the above weight vector and then outputting the result. In one embodiment, the adaptive weight fusion submodule 1021 can be used to execute steps S201 to S205 described above.

[0159] The forecast module 1030 is used to make weather forecasts based on the first fused forecast product. In one embodiment, the forecast module 1030 can be used to perform the step S130 described above, which will not be repeated here.

[0160] In one embodiment, the forecast module 1030 includes a deviation correction submodule 1031. The deviation correction submodule 1031 is used to: acquire at least one second fusion forecast product, which is obtained by processing the raw grid output data of N numerical models, each at a second forecast time point according to the target forecast lead time, using an adaptive weighted fusion method; extract grid data from multiple actual observation station locations from the second fusion forecast product to obtain a discrete grid dataset; calculate the error between the grid data in the discrete grid dataset and the observation point data at the same location and time to obtain a discrete grid error set, where the observation point data is obtained based on the observation data from actual observation stations; process the data in the discrete grid error set using interpolation to obtain a grid error field covering the grid field corresponding to the target spatial region; and correct the first fusion forecast product using the grid error field to obtain the target forecast product. In one embodiment, the deviation correction submodule 1031 can be used to execute steps S401 to S405 described above.

[0161] As an optional embodiment, the multi-model fusion weather forecasting device 1000 may further include a knowledge base construction module 1040. This knowledge base construction module 1040 is used for: collecting historical raw grid output data of any second numerical model among N numerical models; classifying the historical raw grid output data according to weather situation categories to obtain historical dominant weather situations; calculating a historical diagnostic parameter field using the historical raw grid output data according to the calculation method of each diagnostic parameter, wherein different diagnostic parameters are used to forecast different types of meteorological elements; aggregating the historical diagnostic parameter fields corresponding to the same diagnostic parameters calculated from historical raw grid output data with the same historical dominant weather situation and forecast lead time; obtaining a historical performance score of the second numerical model for forecasting corresponding meteorological elements under historical dominant weather situations and forecast lead time based on the comparison and statistics of the forecast results of the aggregated historical diagnostic parameter fields for the corresponding meteorological elements with the actual meteorological observation results; extracting the numerical model identifier, dominant weather situation, meteorological elements, and forecast lead time from the calculation process of the historical performance score to form a mapping relationship; and storing the above mapping relationship in the knowledge base. In one embodiment, the knowledge base construction module 1040 is used to perform steps S310 to S360 described above, which will not be repeated here.

[0162] According to embodiments of this application, any multiple modules among the acquisition module 1010, fusion module 1020, prediction module 1030, and knowledge base construction module 1040 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the acquisition module 1010, fusion module 1020, prediction module 1030, and knowledge base construction module 1040 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays, programmable logic arrays, systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits, or implemented by any other reasonable means of integrating or packaging circuits, or implemented by any one of software, hardware, and firmware, or by any appropriate combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 1010, fusion module 1020, forecast module 1030 and knowledge base construction module 1040 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0163] Figure 8 A block diagram of an electronic device suitable for implementing the multi-mode fusion weather forecasting method of the embodiments of this application is shown schematically.

[0164] like Figure 8 As shown, an electronic device 1200 according to an embodiment of this application includes a processor 1201, which can perform various appropriate actions and processes according to a program stored in a read-only memory 1202 or a program loaded from a storage portion 1208 into a random access memory 1203. The processor 1201 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a dedicated microprocessor. The processor 1201 may also include onboard memory for caching purposes. The processor 1201 may include a single processing unit or multiple processing units for executing different steps of the method flow according to an embodiment of this application.

[0165] Random access memory 1203 stores various programs and data required for the operation of electronic device 1200. Processor 1201, read-only memory 1202, and random access memory 1203 are interconnected via bus 1204. Processor 1201 executes various steps of the method flow according to embodiments of this application by executing programs stored in read-only memory 1202 and / or random access memory 1203. It should be noted that the program may also be stored in one or more memories other than read-only memory 1202 and random access memory 1203. Processor 1201 may also execute various steps of the method flow according to embodiments of this application by executing programs stored in the aforementioned one or more memories.

[0166] According to embodiments of this application, the electronic device 1200 may further include an input / output interface 1205, which is also connected to the bus 1204. The electronic device 1200 may also include one or more of the following components connected to the input / output interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including a cathode ray tube, liquid crystal display, etc., and a speaker, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card, such as a local area network card, modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the input / output interface 1205 as needed. A removable medium 1211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1210 as needed so that computer programs read from it can be installed into the storage section 1208 as needed.

[0167] Embodiments of this application also provide a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0168] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include the read-only memory 1202, and / or random access memory 1203, and / or one or more memories other than read-only memory 1202 and random access memory 1203 described above.

[0169] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this application.

[0170] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1209, and / or installed from the removable medium 1211. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0171] In embodiments of this application, the computer program can be downloaded and installed from a network via communication section 1209, and / or installed from removable medium 1211. When the computer program is executed by processor 1201, it performs the functions defined in the system of this application embodiment. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0172] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0173] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0174] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A multi-model fusion meteorological forecasting method, characterized in that, The method includes: Obtain the original grid output data of N numerical models for meteorological forecasting of the target spatial area based on the first reporting time point and according to the target forecast lead time, and obtain N target original grid output data, where N is an integer greater than 1; The output data of the N target original grid points are processed according to an adaptive weight fusion method to obtain a first fused forecast product; and Weather forecasts are made based on the first fused forecast product; The adaptive weight fusion method mentioned above includes: Using the raw grid output data of each numerical model, a target diagnostic parameter field is calculated according to the calculation method of the target diagnostic parameters. The target diagnostic parameters are at least one of the diagnostic parameters used to forecast target meteorological elements. N sets of the target diagnostic parameter fields are obtained for each of the N numerical models. The target meteorological elements include one of turbulence, icing, or low-level wind shear. The raw grid output data of each numerical model are classified according to the preset weather situation categories to obtain the dominant weather situation corresponding to each numerical model. The weather situation categories include one or more of the following: typhoon, cold front, warm sector precipitation, stationary front, southwest vortex or clear sky. Obtain the target historical performance score of each numerical model for forecasting the target meteorological elements under the target forecast lead time and the dominant weather pattern; the target historical performance score corresponding to each numerical model is used to quantify the historical forecast reliability or accuracy of the numerical model for the target meteorological elements under the target preset lead time and the dominant weather pattern. The target historical performance scores obtained corresponding to N numerical patterns are normalized to obtain a weight vector; and The N sets of target diagnostic parameter fields are weighted and fused according to the weight vector and then output. The weather forecasting based on the first fused forecast product includes: At least one second fusion forecast product is obtained. The second fusion forecast product is obtained by processing the raw grid output data of N numerical models for meteorological forecasting of the target spatial area at the second reporting time point according to the target forecast lead time according to the adaptive weight fusion method. The second reporting time point is earlier than the first reporting time point. Grid data from multiple actual observation station locations are extracted from the second fused forecast product to obtain a discrete grid dataset. The error between the grid data in the discrete grid dataset and the observation point data at the same location and time is calculated to obtain the discrete grid error set, wherein the observation point data is obtained based on the observation data of the actual observation station. The data in the discrete grid error set are processed by interpolation to obtain a grid error field covering the grid field corresponding to the target spatial region; The first fused forecast product is corrected using the grid error field to obtain the target forecast product; and Weather forecasts are made based on the target forecast products; The weather forecasting based on the target forecast product includes: Obtain the preset hazardous weather level threshold configuration table; Based on the comparison between the threshold of the target diagnostic parameter in the hazardous weather level threshold configuration table and the grid data of the target diagnostic parameter in the target forecast product, the target spatial area is divided into different risk levels to obtain the forecast results of the target meteorological elements. The forecast results of the target meteorological elements are output in a visual manner.

2. The method according to claim 1, characterized in that, The acquisition of the target historical performance score for each numerical model in forecasting the target meteorological elements under the target forecast lead time and the dominant weather situation includes: For any first numerical model, the first numerical model identifier, the dominant weather situation corresponding to the first numerical model, the target forecast timeliness, and the target meteorological elements are used as query inputs to query the knowledge base and obtain the target historical performance score returned by the knowledge base. The knowledge base contains mapping relationships, the inputs of which include dominant weather patterns, forecast lead times, numerical model identifiers, and meteorological elements, and the output of which is a historical performance score obtained based on the input mapping.

3. The method according to claim 2, characterized in that, The knowledge base is constructed in the following way: For any second numerical mode among N numerical modes, collect the historical raw grid output data of the second numerical mode; The historical raw grid output data are classified according to the weather situation category to obtain the historical dominant weather situation; Using the historical raw grid output data, a historical diagnostic parameter field is calculated according to the calculation method of each diagnostic parameter, wherein different diagnostic parameters are used to forecast different types of meteorological elements; The historical diagnostic parameter fields corresponding to the same diagnostic parameters calculated from the historical original grid output data with the same historical dominant weather conditions and forecast lead time are aggregated together. Based on the comparison and statistics of the forecast results of the corresponding meteorological elements and the actual meteorological observation results of the aggregated historical diagnostic parameter fields, the historical performance score of the second numerical model for forecasting the corresponding meteorological elements under the historical dominant weather conditions and forecast lead time is obtained. Extract the identifiers of numerical patterns, dominant weather conditions, meteorological elements, and forecast lead time from the calculation process of the historical performance scores to form the mapping relationship; The mapping relationship is stored in the knowledge base.

4. The method according to claim 3, characterized in that, The method further includes: Periodically or in response to event triggers, collect and add the aforementioned historical raw grid output data; and In response to the addition of the historical original grid point output data, the mapping relationship is updated.

5. The method according to any one of claims 2 to 4, characterized in that, The historical performance score is characterized by a key performance index.

6. The method according to claim 1, characterized in that, The classification of the raw grid output data of each numerical model according to the preset weather situation category includes: classification by clustering or according to preset expert rules.

Images