A method and apparatus for intelligent modeling and prediction of meteorological environment

Abstract

This invention discloses an intelligent modeling and prediction method and apparatus for meteorological environment. The method includes: acquiring a historical meteorological environment dataset; training a preset meteorological environment prediction model based on the historical meteorological environment dataset to obtain a trained meteorological environment prediction model; and using the trained meteorological environment prediction model to process a real-time collected meteorological dataset to obtain a set of cloud and rain prediction information.

Description

Technical Field

[0001] This invention relates to the fields of meteorological forecasting and data modeling, specifically to an intelligent modeling and forecasting method and apparatus for meteorological environments. Background Technology

[0002] In the field of meteorology, accurate cloud and rain forecasts are crucial for agricultural production, traffic management, and disaster early warning. Existing meteorological forecasting methods largely rely on traditional numerical weather prediction models, which have several shortcomings when handling complex meteorological data. Firstly, traditional models have limited ability to extract features from meteorological data, making it difficult to effectively capture the intrinsic relationships between complex meteorological elements such as typical pressure layer parameters, multi-dimensional wind field data, and water vapor-related data, resulting in low prediction accuracy. Secondly, the structural design of traditional models is relatively fixed, lacking refined processing modules for different prediction targets such as precipitation, snowfall, and total cloud cover, failing to meet diverse prediction needs. Furthermore, during model training, traditional methods employ relatively simple training strategies, failing to fully leverage the value of historical meteorological data, leading to slow model convergence and a tendency to get trapped in local optima. This makes it difficult to quickly adapt to dynamic changes in the meteorological environment and achieve efficient and accurate meteorological forecasts. Summary of the Invention

[0003] This invention primarily addresses the problem that traditional models have limited ability to extract features from meteorological data, making it difficult to effectively capture the intrinsic relationships between complex meteorological elements such as typical pressure layer parameters, multi-dimensional wind field data, and water vapor-related data, resulting in low prediction accuracy. This invention discloses an intelligent modeling and prediction method and device for meteorological environment.

[0004] In a first aspect, this invention discloses an intelligent modeling and prediction method for meteorological environments, comprising:

[0005] S1, Obtain historical meteorological and environmental datasets;

[0006] S2, Based on the historical meteorological environment dataset, the preset meteorological environment prediction model is trained to obtain the trained meteorological environment prediction model.

[0007] S3. Using the trained meteorological environment prediction model, the real-time collected meteorological dataset is processed to obtain a set of cloud and rain prediction information.

[0008] The historical meteorological environment dataset includes a training dataset and a label dataset; the training dataset includes several training data sets; the label dataset includes several label data sets; each training data set in the training dataset has a corresponding label data set in the label dataset.

[0009] The training data is represented as a training vector, and the elements of the training vector include the parameter information of a typical pressure layer, 10-meter wind field, 2-meter dew point temperature, 2-meter temperature, sea level pressure, surface pressure, total precipitable water volume and total water vapor volume of the layer.

[0010] The tag data is represented as a tag vector, and the elements of the tag vector include precipitation, snowfall, and total cloud cover.

[0011] The parameter information of the typical pressure layer includes: temperature, specific humidity, relative humidity, east-west wind speed, north-south wind speed, vertical velocity, and geopotential height.

[0012] The meteorological environment prediction model includes: an input unit, a feature extraction unit, and a prediction unit group; the prediction unit group includes a precipitation prediction unit, a snowfall prediction unit, and a total cloud cover prediction unit.

[0013] The input unit is used to receive meteorological data; the meteorological data is a vector consisting of parameter information of a typical pressure layer, 10-meter wind field, 2-meter dew point temperature, 2-meter temperature, sea level pressure, surface pressure, total precipitable water content, and total water vapor content; the output of the input unit is connected to the input of the feature extraction unit; the output of the feature extraction unit is connected to the input of each prediction unit in the prediction unit group.

[0014] The feature extraction unit includes a first convolution module, a second convolution module, a first feature transformation module, a third convolution module, a second feature transformation module, a fourth convolution module, a third feature transformation module, a fifth convolution module, a fourth feature transformation module, a pyramid pooling module, and a channel attention module.

[0015] The input terminal of the first convolution module serves as the input terminal of the feature extraction unit;

[0016] The output of the first convolutional module is connected to the input of the second convolutional module; the output of the second convolutional module is connected to the input of the first feature transformation module; the output of the first feature transformation module is connected to the input of the third convolutional module; the output of the third convolutional module is connected to the input of the second feature transformation module; the output of the second feature transformation module is connected to the input of the fourth convolutional module; the output of the fourth convolutional module is connected to the input of the third feature transformation module; the output of the third feature transformation module is connected to the input of the fifth convolutional module; the output of the fifth convolutional module is connected to the input of the fourth feature transformation module; the output of the fourth feature transformation module is connected to the input of the pyramid pooling module; the output of the pyramid pooling module is connected to the input of the channel attention module; the output of the channel attention module serves as the output of the feature extraction unit.

[0017] Each prediction unit includes a first upsampling layer module, a first concatenation layer module, a first feature transformation module, a second upsampling layer module, a second concatenation layer module, and a fully connected layer module;

[0018] The output of the first upsampling layer module is connected to the input of the first stitching layer module; the output of the first stitching layer module is connected to the input of the first feature conversion module; the output of the first feature conversion module is connected to the input of the second upsampling layer module; the output of the second upsampling layer module is connected to the input of the second stitching layer module; the output of the second stitching layer module is connected to the input of the fully connected layer module; the input of the first upsampling layer module serves as the input of the prediction unit; and the output of the fully connected layer module serves as the output of the prediction unit.

[0019] The feature transformation module includes a first convolution submodule, a second convolution submodule, a segmentation layer submodule, a bottleneck layer unit combination, a splicing layer submodule, and a third convolution module; the bottleneck layer unit combination includes three bottleneck layer units connected in series.

[0020] The input terminal of the first convolutional submodule serves as the input terminal of the feature transformation module. In the feature transformation module, the output terminal of the first convolutional submodule is connected to the input terminals of the bottleneck layer unit combination, the segmentation layer submodule, and the concatenation layer submodule, respectively. The output terminal of the segmentation layer submodule is connected to the input terminal of the bottleneck layer unit combination. The output terminal of the bottleneck layer unit combination is connected to the input terminal of the concatenation layer submodule. The output terminal of the second convolutional submodule is connected to the input terminal of the concatenation layer submodule. The output terminal of the concatenation layer submodule is connected to the input terminal of the third convolutional submodule. The output terminal of the third convolutional submodule serves as the output terminal of the feature transformation module.

[0021] The bottleneck layer unit combination includes three cascaded 2D convolutional modules.

[0022] The step of training a pre-defined meteorological environment prediction model based on the historical meteorological environment dataset to obtain a trained meteorological environment prediction model includes:

[0023] S21, the historical meteorological and environmental dataset is evenly divided to obtain several training data sets;

[0024] S22, randomly select a training data subset;

[0025] S23, take each training data in the training data set as input data and input it into the preset meteorological environment prediction model;

[0026] S24, The input data is processed using the meteorological environment prediction model to obtain the corresponding predicted value;

[0027] S25, perform update information calculation processing on the predicted values ​​of the obtained training data and the label data corresponding to the training data to obtain the update vector;

[0028] S26. Based on the update vector, the prediction unit group parameters of the meteorological environment prediction model are updated using the first update model.

[0029] S27, determine whether to input all training data in the training data set into the preset meteorological environment prediction model, and obtain a first judgment result; if the first judgment result is yes, execute S28; if the first judgment result is no, execute S23;

[0030] S28, perform convergence discrimination on the predicted values ​​of all the training data and the corresponding label data of the training data to obtain the second discrimination result;

[0031] When the second discrimination result is negative, the model training state is determined to be that the termination training condition is not met.

[0032] When the second discrimination result is yes, the model training state is determined to meet the termination training condition;

[0033] S29, when the model training state does not meet the termination training condition, the parameters of the meteorological environment prediction model are updated using the second update model, triggering the execution of S22;

[0034] When the model training state meets the termination training condition, the training process of the meteorological environment prediction model is completed, and the trained meteorological environment prediction model is obtained.

[0035] The expression used for calculating and processing the updated information includes:

[0036]

[0037] Where, α i To update the i-th element of the vector, D i and P i Let i be the i-th predicted value of the training data and the i-th corresponding label data, where i = 1, 2, and 3, representing the predicted value or corresponding label data of precipitation, snowfall, and total cloud cover, respectively.

[0038] The expression for the first update model is:

[0039]

[0040] in, γ represents the updated parameters of the i-th prediction unit in the prediction unit group. i The parameters are for the i-th prediction unit in the prediction unit group. The first to third prediction units are the precipitation prediction unit, snowfall prediction unit, and total cloud cover prediction unit, respectively.

[0041] A second aspect of this invention discloses an intelligent modeling and prediction device for meteorological environments, the device comprising:

[0042] Memory containing executable program code;

[0043] A processor coupled to the memory;

[0044] The processor calls the executable program code stored in the memory to execute the intelligent modeling and prediction method for the meteorological environment.

[0045] In a third aspect of this invention, a computer-storable medium is disclosed, wherein the computer-storable medium stores computer instructions, and when the computer instructions are invoked by a computer, they are used to execute the intelligent modeling and prediction method for the meteorological environment.

[0046] In a fourth aspect of this invention, an information data processing terminal is disclosed, which is used to implement the intelligent modeling and prediction method for the meteorological environment.

[0047] The beneficial effects of this invention are as follows:

[0048] The intelligent meteorological environment modeling and prediction method provided in this invention has significant beneficial effects. By constructing a meteorological environment prediction model that includes an input unit, a feature extraction unit, and a prediction unit group, it is possible to perform systematic and refined processing of meteorological data. The synergistic effect of multiple convolutional modules, feature transformation modules, pyramid pooling modules, and channel attention modules in the feature extraction unit can fully extract the deep features of meteorological data and effectively capture the complex relationships between different meteorological elements, thereby improving the accuracy of prediction information such as precipitation, snowfall, and total cloud cover.

[0049] The independent prediction units set up in the prediction unit group of this invention for different prediction targets can combine the features output by the feature extraction unit to accurately predict precipitation, snowfall, and total cloud cover respectively, meeting diverse meteorological forecasting needs. In terms of model training, the strategy of uniformly dividing the historical meteorological environment dataset and training it randomly effectively avoids the model overfitting problem and improves the model's generalization ability. The method of updating model parameters based on update vectors and controlling the training process by judging the convergence of training data enables the model to converge quickly and reach the global optimum, significantly improving model training efficiency. This allows the model to quickly adapt to dynamic changes in the meteorological environment, achieving efficient and accurate meteorological forecasting and providing strong technical support for meteorological services and related fields. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention. Detailed Implementation

[0051] To better understand the content of this invention, an embodiment is provided here.

[0052] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0053] In a first aspect, this invention discloses an intelligent modeling and prediction method for meteorological environments, comprising:

[0054] S1, Obtain historical meteorological and environmental datasets;

[0055] S2, Based on the historical meteorological environment dataset, the preset meteorological environment prediction model is trained to obtain the trained meteorological environment prediction model.

[0056] S3. Using the trained meteorological environment prediction model, the real-time collected meteorological dataset is processed to obtain a set of cloud and rain prediction information.

[0057] The cloud and rain forecast information set includes predicted precipitation, predicted snowfall, and predicted total cloud cover.

[0058] The historical meteorological environment dataset includes a training dataset and a label dataset; the training dataset includes several training data sets; the label dataset includes several label data sets; each training data set in the training dataset has a corresponding label data set in the label dataset.

[0059] The training data is represented as a training vector, and the elements of the training vector include the parameter information of a typical pressure layer, 10-meter wind field, 2-meter dew point temperature, 2-meter temperature, sea level pressure, surface pressure, total precipitable water volume and total water vapor volume of the layer.

[0060] The tag data is represented as a tag vector, and the elements of the tag vector include precipitation, snowfall, and total cloud cover.

[0061] The parameters of the typical pressure layer include: temperature, specific humidity, relative humidity, east-west wind speed, north-south wind speed, vertical velocity, and geopotential height.

[0062] The typical pressure layers include altitude layers with pressures of 200hPa, 300hPa, 400hPa, 500hPa, 600hPa, 700hPa, 850hPa, 925hPa, and 1000hPa.

[0063] The real-time acquisition of the meteorological dataset can be achieved using micro-rain radar, raindrop spectrometer, fogdrop spectrometer, visibility meter, millimeter-wave radar, dual-polarization Doppler radar, wind profiler radar, microwave radiometer, active lightning positioning imaging system, and ground electric field meter.

[0064] The meteorological environment prediction model includes: an input unit, a feature extraction unit, and a prediction unit group; the prediction unit group includes a precipitation prediction unit, a snowfall prediction unit, and a total cloud cover prediction unit.

[0065] The input unit is used to receive meteorological data; the meteorological data is a vector consisting of parameter information of a typical pressure layer, 10-meter wind field, 2-meter dew point temperature, 2-meter temperature, sea level pressure, surface pressure, total precipitable water content, and total water vapor content; the output of the input unit is connected to the input of the feature extraction unit; the output of the feature extraction unit is connected to the input of each prediction unit in the prediction unit group.

[0066] The feature extraction unit includes a first convolution module, a second convolution module, a first feature transformation module, a third convolution module, a second feature transformation module, a fourth convolution module, a third feature transformation module, a fifth convolution module, a fourth feature transformation module, a pyramid pooling module, and a channel attention module.

[0067] The input terminal of the first convolution module serves as the input terminal of the feature extraction unit;

[0068] The output of the first convolutional module is connected to the input of the second convolutional module; the output of the second convolutional module is connected to the input of the first feature transformation module; the output of the first feature transformation module is connected to the input of the third convolutional module; the output of the third convolutional module is connected to the input of the second feature transformation module; the output of the second feature transformation module is connected to the input of the fourth convolutional module; the output of the fourth convolutional module is connected to the input of the third feature transformation module; the output of the third feature transformation module is connected to the input of the fifth convolutional module; the output of the fifth convolutional module is connected to the input of the fourth feature transformation module; the output of the fourth feature transformation module is connected to the input of the pyramid pooling module; the output of the pyramid pooling module is connected to the input of the channel attention module; the output of the channel attention module serves as the output of the feature extraction unit.

[0069] Each prediction unit includes a first upsampling layer module, a first concatenation layer module, a first feature transformation module, a second upsampling layer module, a second concatenation layer module, and a fully connected layer module;

[0070] The output of the first upsampling layer module is connected to the input of the first stitching layer module; the output of the first stitching layer module is connected to the input of the first feature conversion module; the output of the first feature conversion module is connected to the input of the second upsampling layer module; the output of the second upsampling layer module is connected to the input of the second stitching layer module; the output of the second stitching layer module is connected to the input of the fully connected layer module; the input of the first upsampling layer module serves as the input of the prediction unit; the output of the fully connected layer module serves as the output of the prediction unit.

[0071] The feature transformation module can be implemented using a CNN module;

[0072] The feature transformation module includes a first convolution submodule, a second convolution submodule, a segmentation layer submodule, a bottleneck layer unit combination, a splicing layer submodule, and a third convolution module; the bottleneck layer unit combination includes three bottleneck layer units connected in series.

[0073] The input terminal of the first convolutional submodule serves as the input terminal of the feature transformation module. In the feature transformation module, the output terminal of the first convolutional submodule is connected to the input terminals of the bottleneck layer unit combination, the segmentation layer submodule, and the concatenation layer submodule, respectively. The output terminal of the segmentation layer submodule is connected to the input terminal of the bottleneck layer unit combination. The output terminal of the bottleneck layer unit combination is connected to the input terminal of the concatenation layer submodule. The output terminal of the second convolutional submodule is connected to the input terminal of the concatenation layer submodule. The output terminal of the concatenation layer submodule is connected to the input terminal of the third convolutional submodule. The output terminal of the third convolutional submodule serves as the output terminal of the feature transformation module.

[0074] The bottleneck layer unit combination includes three cascaded 2D convolutional modules;

[0075] The segmentation layer submodule can be implemented using a U-Net network.

[0076] The step of training a pre-defined meteorological environment prediction model based on the historical meteorological environment dataset to obtain a trained meteorological environment prediction model includes:

[0077] S21, the historical meteorological and environmental dataset is evenly divided to obtain several training data sets;

[0078] S22, randomly select a training data subset;

[0079] S23, take each training data in the training data set as input data and input it into the preset meteorological environment prediction model;

[0080] S24, The input data is processed using the meteorological environment prediction model to obtain the corresponding predicted value;

[0081] S25, perform update information calculation processing on the predicted values ​​of the obtained training data and the label data corresponding to the training data to obtain the update vector;

[0082] S26. Based on the update vector, the prediction unit group parameters of the meteorological environment prediction model are updated using the first update model.

[0083] S27, determine whether to input all training data in the training data set into the preset meteorological environment prediction model, and obtain a first judgment result; if the first judgment result is yes, execute S28; if the first judgment result is no, execute S23;

[0084] S28, perform convergence discrimination on the predicted values ​​of all the training data and the corresponding label data of the training data to obtain the second discrimination result;

[0085] When the second discrimination result is negative, the model training state is determined to be that the termination training condition is not met.

[0086] When the second discrimination result is yes, the model training state is determined to meet the termination training condition;

[0087] S29, when the model training state does not meet the termination training condition, the parameters of the meteorological environment prediction model are updated using the second update model, triggering the execution of S22;

[0088] When the model training state meets the termination training condition, the training process of the meteorological environment prediction model is completed, and the trained meteorological environment prediction model is obtained.

[0089] In the training data set partitioning of S21, the historical meteorological environmental dataset is evenly divided into several training data sets. This approach allows the model to be exposed to different data combinations during training, avoiding over-reliance on any single data set and effectively preventing overfitting. Simultaneously, different training data sets can simulate the diversity and dynamic changes of meteorological data, enhancing the model's generalization ability and enabling it to make more accurate predictions when faced with real-time meteorological data.

[0090] The expression used for calculating and processing the updated information includes:

[0091]

[0092] Where, α i To update the i-th element of the vector, Di and P i Let i be the i-th predicted value of the training data and the i-th corresponding label data, where i = 1, 2, and 3, representing the predicted value or corresponding label data of precipitation, snowfall, and total cloud cover, respectively.

[0093] The update information calculation process involves calculating update vectors from the predicted values ​​and corresponding label data of the training data. This process quantifies the difference between the predicted and actual values, providing a clear direction and basis for adjusting model parameters. Updating model parameters based on the update vector allows the model to optimize its structure and parameters in a targeted manner based on feedback from the training data, quickly correcting prediction biases and thus improving the model's prediction accuracy and convergence speed.

[0094] The label data refers to the data in the label vector.

[0095] The expression for the first update model is:

[0096]

[0097] in, γ represents the updated parameters of the i-th prediction unit in the prediction unit group. i The parameters are for the i-th prediction unit in the prediction unit group. The first to third prediction units are the precipitation prediction unit, snowfall prediction unit, and total cloud cover prediction unit, respectively.

[0098] The first update model achieves dynamic adaptive adjustment of prediction unit parameters by introducing logarithmic transformation and ratio calculation. When the ratio of the predicted value to the actual label value deviates from 1, the logarithmic operation amplifies this deviation, resulting in a larger parameter update magnitude and accelerating the model's correction of erroneous predictions. Conversely, when the ratio is close to 1, the parameter update magnitude is smaller, which helps the model to make fine adjustments when approaching the optimal solution, improving convergence stability. The subscript i in the formula allows for setting independent parameter update strategies for different prediction units (precipitation, snowfall, total cloud cover). The parameters of each prediction unit can be customized according to the characteristics of its specific prediction target (such as data distribution and prediction difficulty), enabling the model to perform differentiated optimization for the prediction needs of different meteorological elements and improve overall prediction accuracy. The combination of logarithmic transformation and exponential operation ensures the smoothness of the parameter update process, avoiding the parameter oscillation problem that may occur in traditional gradient descent methods. This smooth update mechanism helps the model converge to the global optimum more stably, especially when dealing with complex datasets with high-dimensional and nonlinear characteristics such as meteorological data, exhibiting stronger robustness.

[0099] The first updated model compresses the original ratio through logarithmic operations. The increased dynamic range reduces the model's sensitivity to outlier data points, enhancing its robustness against noise. This is particularly important in weather forecasting scenarios, as meteorological data is frequently affected by instrument errors, sudden weather events, and other factors, leading to significant local fluctuations. Based on the mathematical properties of logarithms and exponentials, this formula is computationally more efficient than traditional gradient calculation methods, especially when dealing with large-scale meteorological datasets. It significantly reduces computational resource consumption and improves the real-time performance of model training.

[0100] The convergence criterion includes:

[0101] The fusion convergence value is calculated; the expression for the fusion convergence value is:

[0102]

[0103] Among them, a ji Let b be the predicted value of the i-th training data of the j-th training data subset. ji Let be the i-th label data of the j-th training data in the training data set, β be the preset adjustment factor, N be the total number of training data in the training data set, and i = 1, 2, 3, representing the predicted values ​​or corresponding label data of precipitation, snowfall and total cloud cover, respectively.

[0104] If the fusion convergence value is less than a preset first discrimination threshold, the second discrimination result is determined to be yes; if the fusion convergence value is not less than the preset first discrimination threshold, the second discrimination result is determined to be no.

[0105] The convergence determination process performs convergence assessment on the predicted values ​​and corresponding label data of all training data. By determining whether the model has reached a stable training state, it avoids the model stopping training due to insufficient or excessive training. This method ensures that the model stops training when it reaches the optimal solution, effectively improving the efficiency and quality of model training. The trained model can then more accurately predict meteorological data, reducing prediction errors caused by insufficient or excessive training.

[0106] The expression for the second update model is:

[0107]

[0108] In the formula, x j For the j-th training data in the training data subset, y j The label data for the j-th training data in the training data subset. Let v be the loss function, θ be the parameter update value, and θ be the parameters of the meteorological environment prediction model. This represents taking the partial derivative with respect to the variable θ, f(x) j;θ) represents the predicted value of the j-th training data in the training data set by the meteorological environment prediction model, f(.) is the calculation function corresponding to the meteorological environment prediction model, and exp represents the power operation of the constant e. This is a preset proportional adjustment factor.

[0109] The value can be 0.6.

[0110] A second aspect of this invention discloses an intelligent modeling and prediction device for meteorological environments, the device comprising:

[0111] Memory containing executable program code;

[0112] A processor coupled to the memory;

[0113] The processor calls the executable program code stored in the memory to execute the intelligent modeling and prediction method for the meteorological environment.

[0114] In a third aspect of this invention, a computer-storable medium is disclosed, wherein the computer-storable medium stores computer instructions, and when the computer instructions are invoked by a computer, they are used to execute the intelligent modeling and prediction method for the meteorological environment.

[0115] In a fourth aspect of this invention, an information data processing terminal is disclosed, which is used to implement the intelligent modeling and prediction method for the meteorological environment.

[0116] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for intelligent modeling and prediction of meteorological environment, characterized in that, include: S1, Obtain historical meteorological and environmental datasets; S2, based on the historical meteorological environment dataset, the preset meteorological environment prediction model is trained to obtain a trained meteorological environment prediction model, including: S21, the historical meteorological and environmental dataset is evenly divided to obtain several training data sets; S22, randomly select a training data subset; S23, take each training data in the training data set as input data and input it into the preset meteorological environment prediction model; S24, The input data is processed using the meteorological environment prediction model to obtain the corresponding predicted value; S25, perform update information calculation processing on the predicted values ​​of the obtained training data and the label data corresponding to the training data to obtain the update vector; S26. Based on the update vector, the prediction unit group parameters of the meteorological environment prediction model are updated using the first update model. S27, determine whether to input all training data in the training data set into the preset meteorological environment prediction model, and obtain a first judgment result; if the first judgment result is yes, execute S28; if the first judgment result is no, execute S23; S28, perform convergence discrimination on the predicted values ​​of all the training data and the corresponding label data of the training data to obtain the second discrimination result; The convergence criterion includes: The fusion convergence value is calculated; the expression for the fusion convergence value is: Among them, a ji Let b be the predicted value of the i-th training data of the j-th training data subset. ji Let be the i-th label data of the j-th training data in the training data set, β be the preset adjustment factor, N be the total number of training data in the training data set, and i = 1, 2, 3, representing the predicted values ​​or corresponding label data of precipitation, snowfall and total cloud cover, respectively. If the fusion convergence value is less than a preset first discrimination threshold, the second discrimination result is determined to be yes; if the fusion convergence value is not less than the preset first discrimination threshold, the second discrimination result is determined to be no. When the second discrimination result is negative, the model training state is determined to be that the termination training condition is not met. When the second discrimination result is yes, the model training state is determined to meet the termination training condition; S29, when the model training state does not meet the termination training condition, the parameters of the meteorological environment prediction model are updated using the second update model, triggering the execution of S22; When the model training state meets the termination training condition, the training process of the meteorological environment prediction model is completed, and the trained meteorological environment prediction model is obtained. The meteorological environment prediction model includes: an input unit, a feature extraction unit, and a prediction unit group; the prediction unit group includes a precipitation prediction unit, a snowfall prediction unit, and a total cloud cover prediction unit. The input unit is used to receive meteorological data; the meteorological data is a vector consisting of parameter information of a typical pressure layer, 10-meter wind field, 2-meter dew point temperature, 2-meter temperature, sea level pressure, surface pressure, total precipitable water content, and total water vapor content; the output of the input unit is connected to the input of the feature extraction unit; the output of the feature extraction unit is connected to the input of each prediction unit in the prediction unit group. The feature extraction unit includes a first convolution module, a second convolution module, a first feature transformation module, a third convolution module, a second feature transformation module, a fourth convolution module, a third feature transformation module, a fifth convolution module, a fourth feature transformation module, a pyramid pooling module, and a channel attention module. The input terminal of the first convolution module serves as the input terminal of the feature extraction unit; The output of the first convolutional module is connected to the input of the second convolutional module; the output of the second convolutional module is connected to the input of the first feature transformation module; the output of the first feature transformation module is connected to the input of the third convolutional module; the output of the third convolutional module is connected to the input of the second feature transformation module; the output of the second feature transformation module is connected to the input of the fourth convolutional module; the output of the fourth convolutional module is connected to the input of the third feature transformation module; the output of the third feature transformation module is connected to the input of the fifth convolutional module; the output of the fifth convolutional module is connected to the input of the fourth feature transformation module; the output of the fourth feature transformation module is connected to the input of the pyramid pooling module; the output of the pyramid pooling module is connected to the input of the channel attention module; the output of the channel attention module serves as the output of the feature extraction unit. Each prediction unit includes a first upsampling layer module, a first concatenation layer module, a first feature transformation module, a second upsampling layer module, a second concatenation layer module, and a fully connected layer module; The output of the first upsampling layer module is connected to the input of the first stitching layer module; the output of the first stitching layer module is connected to the input of the first feature conversion module; the output of the first feature conversion module is connected to the input of the second upsampling layer module; the output of the second upsampling layer module is connected to the input of the second stitching layer module; the output of the second stitching layer module is connected to the input of the fully connected layer module; the input of the first upsampling layer module serves as the input of the prediction unit; the output of the fully connected layer module serves as the output of the prediction unit; S3, using the trained meteorological environment prediction model, the real-time collected meteorological dataset is processed to obtain a set of cloud and rain prediction information.

2. The intelligent modeling and prediction method for meteorological environment as described in claim 1, characterized in that, The historical meteorological environment dataset includes a training dataset and a label dataset; the training dataset includes several training data sets; the label dataset includes several label data sets; each training data set in the training dataset has a corresponding label data set in the label dataset. The training data is represented as a training vector, and the elements of the training vector include the parameter information of a typical pressure layer, 10-meter wind field, 2-meter dew point temperature, 2-meter temperature, sea level pressure, surface pressure, total precipitable water volume and total water vapor volume of the layer. The tag data is represented as a tag vector, and the elements of the tag vector include precipitation, snowfall, and total cloud cover. The parameter information of the typical pressure layer includes: temperature, specific humidity, relative humidity, east-west wind speed, north-south wind speed, vertical velocity, and geopotential height.

3. The intelligent modeling and prediction method for meteorological environment as described in claim 1, characterized in that, The feature transformation module includes a first convolution submodule, a second convolution submodule, a segmentation layer submodule, a bottleneck layer unit combination, a splicing layer submodule, and a third convolution module; the bottleneck layer unit combination includes three bottleneck layer units connected in series. The input terminal of the first convolutional submodule serves as the input terminal of the feature transformation module. In the feature transformation module, the output terminal of the first convolutional submodule is connected to the input terminals of the bottleneck layer unit combination, the segmentation layer submodule, and the concatenation layer submodule, respectively. The output terminal of the segmentation layer submodule is connected to the input terminal of the bottleneck layer unit combination. The output terminal of the bottleneck layer unit combination is connected to the input terminal of the concatenation layer submodule. The output terminal of the second convolutional submodule is connected to the input terminal of the concatenation layer submodule. The output terminal of the concatenation layer submodule is connected to the input terminal of the third convolutional submodule. The output terminal of the third convolutional submodule serves as the output terminal of the feature transformation module. The bottleneck layer unit combination includes three cascaded 2D convolutional modules.

4. The intelligent modeling and prediction method for meteorological environment as described in claim 1, characterized in that, The expression used for calculating and processing the updated information includes: Where, α i To update the i-th element of the vector, D i and P i Let i be the i-th predicted value of the training data and the i-th corresponding label data, where i = 1, 2, and 3, representing the predicted value or corresponding label data of precipitation, snowfall, and total cloud cover, respectively.

5. The intelligent modeling and prediction method for meteorological environment as described in claim 1, characterized in that, The expression for the first update model is: in, γ represents the updated parameters of the i-th prediction unit in the prediction unit group. i The parameters are for the i-th prediction unit in the prediction unit group. The first to third prediction units are the precipitation prediction unit, snowfall prediction unit, and total cloud cover prediction unit, respectively.

6. A meteorological environment intelligent modeling and prediction device, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the intelligent modeling and prediction method for meteorological environment as described in any one of claims 1 to 5.

7. A computer-storable medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked by the computer, are used to execute the intelligent modeling and prediction method for meteorological environment as described in any one of claims 1 to 5.

8. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the intelligent modeling and prediction method for meteorological environment as described in any one of claims 1 to 5.

Images