Method and system for selecting and forecasting optimal factor of high influence weather event

Abstract

The invention discloses a high influence weather event optimal factor selecting and forecasting method and system, and the method comprises the steps: collecting first object data, and carrying out the preprocessing; and constructing a multi-dimensional factor library, and screening and optimizing the factor library. Coupling the factors, inputting a calculation model, and outputting a forecast result. According to the invention, a multi-scale factor coupling analysis method is introduced, a large-scale circulation factor is combined with a small and medium-scale weather system factor, and an interaction mechanism between different meteorological factors is comprehensively considered. The method can more accurately predict high-influence weather events under complex climate conditions in Yunnan regions. A dynamic factor library updating mechanism is established, and the content of the factor library can be adjusted in real time so as to adapt to climate changes and diversity under different seasons and climate years. According to the mechanism, the adaptability of the forecasting model is greatly improved, so that the forecasting model can flexibly deal with the complexity and the variability of the climate in the Yunnan region, and the forecasting reliability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological forecasting, and in particular to a method and system for selecting and forecasting optimal factors for high-impact weather events. Background Art

[0002] High-impact weather events are extreme weather phenomena that can have a significant impact on key infrastructure such as power grids, transportation, agriculture, and people's daily lives, including heavy precipitation, blizzards, freezing rain, typhoons, etc. In Yunnan, due to its complex geographical and climatic characteristics, high-impact weather events are more frequent, and due to the diverse terrain and variable regional climate conditions, these events have significant differences in spatial distribution and intensity. These weather events can cause line breakage, equipment damage, large-scale power outages, etc., causing operational risks and economic losses to the power grid. Therefore, accurate forecasting of high-impact weather has become an important means to ensure the safety of the power grid.

[0003] In the field of forecasting high-impact weather events, existing numerical weather forecasting, statistical regression models, machine learning models and other technologies have their own applications, but there are still significant deficiencies, which are specifically reflected in the following aspects:

[0004] Factor selection relies on manual experience and lacks systematicity: The existing high-impact weather forecast models mainly rely on expert experience to select meteorological factors, usually based on the correlation between certain known factors (such as temperature, humidity, wind speed, etc.) and high-impact weather events. However, this method often lacks systematicity, ignores the complex multidimensional correlations between factors, and is difficult to capture implicit highly correlated factors. Therefore, when faced with complex and changeable meteorological systems, these models often cannot effectively select factors, resulting in low model prediction accuracy.

[0005] Lack of comprehensive analysis of multi-scale factors: The formation of high-impact weather usually involves the interaction of large-scale circulation factors (such as subtropical high pressure, southern trough, etc.) and small and medium-scale weather system factors (such as southwest vortex, shear line, etc.), and the configuration on the ground and in the air is different. However, the existing forecast models are mostly focused on a certain scale or a certain combination of factors, lacking the coordinated analysis of multi-scale factors, which makes it difficult to accurately forecast extreme weather under the climate conditions of the Yunnan Plateau.

[0006] Poor dynamic adaptability of forecast models: Traditional forecast models lack flexibility in updating factor libraries and adjusting model parameters, making it difficult for models to dynamically adjust with climate change and meteorological system changes. However, climate change in Yunnan is significant, and high-impact weather events in different seasons and different climate years are quite different. Existing models have poor adaptability and are unable to cope with the diversity of high-impact weather events.

[0007] Insufficient model error correction and precision optimization: In existing technologies, many models rely on regular manual parameter adjustment or use fixed regression coefficients, lacking automatic error correction and dynamic optimization mechanisms. When dealing with complex meteorological patterns and changing climate conditions, model errors are difficult to eliminate effectively, resulting in inaccurate and in-time forecasts of high-impact weather. Summary of the invention

[0008] In view of the above-mentioned problems, the present invention is proposed.

[0009] Therefore, the technical problem solved by the present invention is: in the field of forecasting high-impact weather events, existing numerical weather forecasting, statistical regression models, machine learning models and other technologies have their own applications, but there are still significant deficiencies, including reliance on manual experience for factor selection, lack of systematicity, lack of comprehensive analysis of multi-scale factors, poor dynamic adaptability of forecasting models, and insufficient model error correction and precision optimization.

[0010] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for selecting and forecasting optimal factors of high-impact weather events, comprising: collecting first object data and performing preprocessing.

[0011] A multidimensional factor library is constructed, and the factor library is screened and optimized.

[0012] The factors are coupled, input into the calculation model, and the prediction results are output.

[0013] As a preferred solution of the method for selecting and forecasting optimal factors for high-impact weather events described in the present invention, the preprocessing includes standardizing the first object data and converting them to the same interval.

[0014] As a preferred solution of the method for selecting and forecasting the optimal factors of high-impact weather events described in the present invention, the multidimensional factor library is constructed by calculating the first object data to generate a series of derived factors. The derived factors and spatial region division are combined to further expand the factor library based on factors of different levels and different physical quantities, and the composite factors are generated by interactive combination.

[0015] As a preferred solution of the method for selecting and forecasting the optimal factors for high-impact weather events described in the present invention, the screening and optimization of the factor library includes preliminary screening and in-depth screening.

[0016] Preliminary screening and evaluation of the correlation between each factor and the forecast target is performed, and factors with clear physical meaning and significant contribution to the forecast are retained.

[0017] Deep screening uses deep learning models to extract and screen the factor library to extract the most critical factor features for high-impact weather events.

[0018] As a preferred solution of the method for selecting and forecasting the optimal factors for high-impact weather events described in the present invention, the optimization includes automatically optimizing the factor library through a genetic algorithm, iteratively optimizing the combination of factors and weight parameters, and obtaining the optimal factor combination.

[0019] As a preferred solution of the method for selecting and forecasting the optimal factors for high-impact weather events described in the present invention, the first object data includes but is not limited to meteorological station data, and the dimensions of the multidimensional factor library include but are not limited to multi-level meteorological variables.

[0020] As a preferred scheme of the method for selecting and forecasting the optimal factors for high-impact weather events described in the present invention, the factors are coupled, input into the calculation model, and the forecast results are output, including coupling long-term factors and short-term factors, inputting into a multi-scale association learning model to form a composite factor, inputting the composite factor into the forecast model, and outputting the forecast results.

[0021] A high-impact weather event optimal factor selection and forecasting system, characterized by: comprising:

[0022] The preprocessing module collects the first object data and performs preprocessing.

[0023] Construct a factor library module, construct a multidimensional factor library, and screen and optimize the factor library.

[0024] The calculation module couples the factors, inputs the calculation model, and outputs the prediction results.

[0025] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0026] A computer-readable storage medium stores a computer program, which implements the steps of the method described above when executed by a processor.

[0027] The beneficial effects of the present invention are as follows: a coupling analysis method of multi-scale factors is introduced, large-scale circulation factors are combined with medium- and small-scale weather system factors, and the interaction mechanism between different meteorological factors is fully considered. It is possible to more accurately predict high-impact weather events under complex climatic conditions in Yunnan. A dynamic factor library update mechanism is established, which can adjust the content of the factor library in real time to adapt to climate change and the diversity under different seasons and climate years. This mechanism greatly improves the adaptability of the forecast model, enabling it to flexibly respond to the complexity and variability of the climate in Yunnan, thereby improving the reliability of the forecast. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0029] Figure 1 An overall flow chart of a method and system provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0031] Example 1, reference Figure 1 , which is an embodiment of the present invention, provides a method for selecting and forecasting optimal factors for high-impact weather events, comprising:

[0032] S1: Collect first object data and perform preprocessing.

[0033] In the present invention, the first object data is the meteorological station data, which collects the ERA5 data released by the European Center for Medium-Range Weather Forecasts, including temperature, humidity, geopotential height and wind speed. The dimensions and value ranges of various meteorological variables are different. The preprocessing is to convert these data to the same interval through standardization to ensure the balance of various factors in the model calculation and avoid affecting the prediction results due to differences in data ranges.

[0034] In the present invention, the first object data may also be satellite remote sensing data and numerical weather prediction (NWP) data.

[0035] In an optional embodiment of the present invention, the first object data is satellite remote sensing data, and various meteorological parameters of the high altitude and ground atmosphere are obtained through satellite observation, including cloud top temperature (unit: K): This data can reflect the temperature information of the cloud layer in the atmosphere, which is helpful to identify the formation of high-impact weather such as cumulonimbus clouds. Surface albedo (unit: %): Albedo refers to the ability of the surface to reflect solar radiation. This parameter is of great significance for analyzing surface temperature changes and precipitation probability.

[0036] The preprocessing process includes that satellite remote sensing data usually has a high resolution and contains a large amount of meteorological information at different levels. The focus of preprocessing is on data noise reduction and standardization. In the preprocessing process, spatial resampling is first required to keep the spatial resolution of the remote sensing data consistent with the resolution of the meteorological station observation data, so as to ensure the consistency of the data in the subsequent multi-factor coupling. At the same time, for data such as cloud top temperature and surface albedo, standardization is applied to convert their value range to a unified range of 0 to 1. In addition, remote sensing data will be affected by factors such as clouds and climate interference during the actual acquisition process, so filtering is also required to remove random noise and improve data accuracy.

[0037] In an optional embodiment of the present invention, the first object data is numerical weather forecast (NWP) data, which is forecast information calculated by a meteorological model, such as data of a global or regional meteorological model, including ground pressure (unit: hPa): ground pressure data is used to analyze changes in the atmospheric pressure system and can help identify upcoming weather changes. Humidity (unit: %): Humidity is an important parameter for predicting precipitation and cloud cover. By analyzing changes in humidity, the probability of future precipitation can be inferred.

[0038] The preprocessing process includes interpolation and standardization of numerical weather forecast data. The grid distribution of numerical weather forecast data is usually sparse, so interpolation methods are needed to convert it into more detailed grid data to meet the needs of refined forecast models. Secondly, due to the large differences in the value ranges of variables such as air pressure and humidity, standardization is also essential. All numerical data are scaled to a uniform interval of 0 to 1 to ensure the balance of data when multiple factors are fused. In addition, since numerical weather forecast data usually have errors, data assimilation technology is also used in actual operations to fuse observation data with forecast data to improve the accuracy and representativeness of the data.

[0039] S2: Construct a multidimensional factor library and screen and optimize the factor library.

[0040] The first object data at multiple levels such as 500hPa, 700hPa, and 850hPa are selected to cover the key dynamic features in the atmosphere.

[0041] Based on the basic physical quantities, a series of derived factors are generated by calculation to reveal the characteristics of complex meteorological systems, enabling the model to identify factors that are critical to high-impact weather and further enhance the expressive power of the factor library. For example, vorticity and divergence are calculated using wind speed data to capture the rotation and divergence characteristics of airflow. The temperature and humidity are used to calculate the temperature dew point difference to reflect the air humidity level. The water vapor flux divergence can characterize the transport and aggregation of water vapor and is suitable for precipitation forecasting.

[0042] Factor combination and key area selection: To enhance the multi-dimensional characteristics of the factor library, the descriptive ability of the factor library is further expanded by combining factors and dividing spatial regions, so that it can more accurately describe the characteristics of different weather systems:

[0043] Composite factors are generated by combining factors based on different levels and physical quantities. Variables such as humidity, temperature, and vorticity in different pressure layers are combined to generate coupling factors to improve the ability to express complex meteorological phenomena. These combined factors can not only express multivariate interactions that are difficult to reveal with a single factor, but also capture key weather phenomena such as convection and the intersection of cold and warm air.

[0044] In order to enhance the accuracy of the factor library, composite factors are generated by interactively combining different physical quantities and hierarchical factors, such as convection terms, cold and warm air intersection terms, etc. In addition, key grid data (such as ice-covered areas) are selected in combination with the impact areas of different weather events, and the combined factors are weighted and optimized according to regional characteristics to adapt to the spatial characteristics of specific weather events.

[0045] After the factor library is built, the optimal factor combination for high-impact weather events is screened out through automated factor screening and combination optimization to improve the performance of the forecast model.

[0046] Preliminary screening: Use the PRESS (prediction sum of squares) criterion to preliminarily screen the candidate factors in the factor library to evaluate the correlation between each factor and the forecast target, and retain factors with clear physical meaning and significant contribution to the forecast.

[0047] Deep learning screening factors: After the initial construction of the factor library is completed, the deep learning model is used to extract and screen the factor library to extract the most critical factor features for high-impact weather events. The addition of deep learning can effectively handle complex factor relationships, especially spatial and temporal associations.

[0048] The convolutional neural network (CNN) is used to extract spatial features of the combined factors, automatically identifying the spatial structure and mutual influence between grid points, thereby extracting the spatial patterns that have a significant impact on weather events. The CNN model can reduce redundant data and compress data dimensions, making the model more focused on significant factors in terms of spatial structure, thereby improving the sensitivity of the forecast model to spatial changes.

[0049] For time series features, the LSTM network is used to extract dependencies in the time dimension. The LSTM model can automatically learn the temporal variation patterns of factors, which is particularly suitable for the temporal prediction of extreme weather events. The temporal features extracted by LSTM can retain the most time-sensitive factors, enabling the model to more accurately predict short-term sudden weather events or long-term weather trends.

[0050] Dynamic optimization and adaptive adjustment: In order to enhance the long-term adaptability of the factor library, the automated optimization algorithm is combined to enable the factor library to adaptively adjust with climate change and seasonal fluctuations, further improving the forecast effect:

[0051] The factor library is automatically optimized through genetic algorithms, which simulate the evolutionary process of "survival of the fittest" and iteratively optimize the combination of factors and weight parameters to obtain the optimal factor combination. This method effectively avoids the limitations of traditional factor screening and makes the factor library more adaptable to changing meteorological conditions.

[0052] S3: Couple the factors, input the calculation model, and output the prediction results.

[0053] Selection and combination of short-term and long-term factors.

[0054] Short-term factors: Short-term scale factors are factors that have a significant impact on rapid weather changes that occur in a short period of time (such as sudden drops in temperature, sudden increases in humidity, etc.). These factors usually manifest as short-term violent fluctuations, reflecting the rapid response process of weather events. Typical short-term factors include: near-surface temperature, relative humidity, wind speed and wind direction changes, etc.

[0055] Long-term factors: Long-term scale factors usually have a more stable impact on the overall trend of the weather system, describing the atmospheric changes over a longer period of time (such as seasonal pressure fields, high and low pressure systems, etc.). These factors change relatively slowly and provide important information for longer-term weather forecasts. Typical long-term factors include: geopotential height, large-scale wind field, etc.

[0056] When coupling multi-scale factors, short-term factors and long-term factors are combined to capture the spatiotemporal characteristics of meteorological phenomena. For example, combining short-term temperature changes with long-term pressure fields can reveal the trend and short-term influence of extreme weather events caused by the convergence of cold and warm air.

[0057] The coupling analysis between factors includes:

[0058] Coupling factor screening based on correlation matrix: First, the correlation matrix between short-term and long-term factors is constructed to screen out high-correlation factor combinations. These highly correlated coupling factors reflect the multi-scale driving forces of weather events. For example, the coupling factor between 500hPa potential height change and near-surface temperature can usually be used to predict heavy precipitation or sudden cooling events.

[0059] Eigenvalue decomposition and weight allocation: The weights of each factor are analyzed using eigenvalue decomposition for the coupling factor combination, and optimized with historical data of weather events. The main coupling factors are extracted through principal component analysis (PCA), and the weights of factors of different time scales are allocated according to their contribution to the forecast events, ensuring that both short-term changes and long-term trends are responded to.

[0060] In order to more accurately capture the spatiotemporal relationship of multi-scale factors, a deep learning model is used to train the combined effects of short-term and long-term factors based on the coupling factors, thereby extracting complex nonlinear features, including using CNN to extract spatial features and using LSTM to capture temporal dependencies. The spatial features extracted by CNN are combined with the temporal features extracted by LSTM to form a multi-scale association learning model. Finally, the composite factors formed by multi-scale coupling analysis are input into the BP neural network for weather event prediction. The BP neural network uses nonlinear mapping capabilities to efficiently fit the coupling factors and output accurate weather event prediction results.

[0061] The computer device may be a server. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data cluster data of a power monitoring system. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method is implemented.

[0062] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0063] Example 2 is an embodiment of the present invention, which provides a method and system for selecting and forecasting optimal factors for high-impact weather events. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0064] In order to verify the actual effect of a method for selecting optimal factors and forecasting high-impact weather events, the experiment will construct a multidimensional factor library and screen and optimize it based on data sources such as meteorological observation data, satellite remote sensing data, and numerical weather forecast (NWP) data, so as to compare the accuracy and effectiveness of the method of the present invention with traditional weather forecasting methods.

[0065] Subjects:

[0066] Existing technology: Traditional high-impact weather forecasting methods rely on single meteorological observation data and empirical models. The factor selection only includes basic meteorological parameters, the factor library is limited and the update frequency is low.

[0067] The present invention uses multiple data sources to establish a composite factor library, and through preprocessing, screening and optimization, uses a multi-scale association learning model to fuse short-term and long-term factors, dynamically generates and updates the composite factor library, and improves the prediction accuracy of high-impact weather events.

[0068] Experimental steps:

[0069] Data collection and preprocessing: First, various meteorological data are collected, including meteorological station observations, satellite remote sensing and numerical weather forecast data. In the prior art, each data source is usually processed separately, and standardization is not achieved. Through the preprocessing steps of the present invention, all data are converted to the 0-1 interval and noise is filtered out, making them suitable for unified model input.

[0070] Factor library construction and screening optimization: Based on traditional factors, the present invention constructs a multidimensional factor library, including basic factors such as temperature, humidity, wind speed, and further generates composite factors such as vorticity, divergence, and dew point difference. The initial screening evaluates the correlation between each factor and weather events through the PRESS criterion, and then further screens out key characteristic factors through deep learning models (CNN and LSTM). Existing methods lack systematic factor screening methods and are usually based on simple correlation analysis, while the composite factors and dynamic optimization of the present invention ensure high expressivity of the factors.

[0071] Coupling analysis of short-term and long-term factors: Short-term factors and long-term factors are coupled to generate a composite factor combination. The weight distribution is optimized by PCA principal component analysis, and the spatial and temporal dependencies are extracted by combining the deep learning model. The coupling strategy of the present invention improves the sensitivity of the prediction model to meteorological changes, while the traditional method is insufficient in extreme weather prediction due to the lack of multi-scale fusion. The experimental results are shown in Table 1.

[0072] Table 1 Experimental results

[0073]

[0074] By comparing the table data, it can be clearly seen that the present invention is significantly superior to the prior art in predicting high-impact weather events, as shown in the following:

[0075] In the prior art, the prediction accuracy is only 75% due to the single factor library and the lack of multidimensional data support in the model. The present invention improves the accuracy to 92% through multi-data source composite factor library, dynamic optimization and other methods, which is an increase of 22.67%, significantly enhancing the prediction accuracy.

[0076] Traditional technologies rely on manual updates, and the factor library is updated only once a year, which makes it difficult to adapt to seasonal weather changes. The factor library of the present invention is automatically updated up to 12 times a month, ensuring the real-time adaptability of the prediction model, and the frequency is increased by 1100%.

[0077] Traditional technologies take a long time to process and analyze factors, resulting in delayed forecast response. The present invention shortens the response time to 10 minutes through automated factor screening and efficient calculation models, with an optimization margin of 83.33%.

[0078] The alarm accuracy of the traditional method is 70%, and there are false alarms and missed alarms. The high-precision factor screening and optimization of the present invention improves the alarm accuracy to 95%, an improvement of 35.71%, and reduces the waste of resources caused by false alarms.

[0079] The existing method takes a long time to update and optimize the model, and the cost of a single training is 12 hours. The present invention combines deep learning and automatic optimization algorithms to shorten the training cost to 6 hours, a reduction of 50%.

[0080] The number of factors in the factor library in traditional technology is only 50, which is difficult to fully express the complex weather system. The factor library of the present invention is expanded to 200 composite factors, covering more physical quantities and spatial information, with a capacity increase of 300%, which enhances the predictive expression ability of the model.

[0081] The high maintenance cost of traditional technology is about 40,000 yuan per year. The present invention reduces the maintenance cost to 25,000 yuan through automation system optimization, saving 37.5%.

[0082] In summary, the experimental results show that the optimal factor selection and forecasting method for high-impact weather events is significantly superior to existing technologies in terms of prediction accuracy, response speed, dynamic adaptability of factor library, etc. through multidimensional data sources, automatic factor screening and optimization, coupled factor analysis, etc., providing reliable technical support for the accurate prediction of high-impact weather events.

[0083] Embodiment 3 is an embodiment of the present invention, which provides a high-impact weather event optimal factor selection and forecasting system, specifically:

[0084] The preprocessing module collects the first object data and performs preprocessing.

[0085] Construct a factor library module, construct a multidimensional factor library, and screen and optimize the factor library.

[0086] The calculation module couples the factors, inputs the calculation model, and outputs the prediction results.

[0087] 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 preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for selecting and forecasting optimal factors for high-impact weather events, characterized in that: include: collecting first object data and performing preprocessing; constructing a multidimensional factor library and screening and optimizing the factor library; The factors are coupled, input into the calculation model, and the prediction results are output.

2. The method for selecting and forecasting optimal factors for high-impact weather events according to claim 1, wherein: The preprocessing includes standardizing the first object data and converting them into the same interval.

3. The method for selecting and forecasting optimal factors for high-impact weather events according to claim 2, wherein: The construction of the multidimensional factor library includes calculating the first object data to generate a series of derived factors; combining the derived factors and spatial region division to further expand the factor library based on factors at different levels and different physical quantities, and interactively combining to generate composite factors.

4. The method for selecting and forecasting optimal factors for high-impact weather events according to claim 3, wherein: The screening and optimization of the factor library includes preliminary screening and in-depth screening; Preliminary screening and evaluation of the correlation between each factor and the forecast target, retaining factors with clear physical meaning and significant contribution to the forecast; Deep screening uses deep learning models to extract and screen the factor library to extract the most critical factor features for high-impact weather events.

5. The method for selecting and forecasting optimal factors for high-impact weather events according to claim 4, wherein: The optimization includes automatically optimizing the factor library through a genetic algorithm, iteratively optimizing the combination of factors and weight parameters, and obtaining the optimal factor combination.

6. The method for selecting and forecasting optimal factors for high-impact weather events according to claim 1, wherein: The first object data includes but is not limited to meteorological station data, and the dimensions of the multidimensional factor library include but are not limited to multi-level meteorological variables.

7. The method for selecting and forecasting optimal factors for high-impact weather events according to claim 6, wherein: The coupling of factors, inputting into a calculation model, and outputting a forecast result includes coupling long-term factors and short-term factors, inputting into a multi-scale association learning model to form a composite factor, inputting the composite factor into a forecast model, and outputting a forecast result.

8. A high-impact weather event optimal factor selection and forecasting system using the method according to any one of claims 1 to 7, characterized in that: A preprocessing module collects first object data and performs preprocessing; Constructing a factor library module, constructing a multidimensional factor library, and screening and optimizing the factor library; The calculation module couples the factors, inputs the calculation model, and outputs the prediction results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images