Severe weather warning system based on big data

Through the combination of local mean decomposition and spherical harmonic neural operators, the prediction problems caused by inaccurate data and spherical characteristics in traditional meteorological early warning systems are solved, and high-accurate weather warning and accurate prediction of catastrophic weather events are achieved.

CN118656792BActive Publication Date: 2024-11-12STATE GRID SIJI FEITIAN (LANZHOU) CLOUD TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411141014.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-12
Estimated Expiration
2044-08-20

AI Technical Summary

Technical Problem

Traditional meteorological early warning systems face problems such as inaccurate data, fuzzy warning range, and insufficient warning timeliness, and ignore the spherical characteristics of meteorological data, resulting in poor prediction accuracy.

Method used

The empirical modal decomposition algorithm of fully adaptive noise ensemble based on local mean decomposition is used to depolate real-time meteorological data, and the spherical harmonic nerve operator based on gating residual spectrum is used to despherical processing to eliminate the influence of spherical characteristics and improve prediction accuracy.

Benefits of technology

It achieves high-accurate weather forecasts, meets the accurate warning needs of disaster weather events, and improves the warning timeliness and the accuracy of data processing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118656792B_ABST
    Figure CN118656792B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of intelligent weather warning, specifically, a disastrous weather warning system based on big data, including a data collection and integration module, a data cleaning and preprocessing module, an intelligent prediction module, a real-time monitoring and warning module, a geographic information system integration module and a user interaction and control module; the present invention adopts a fully adaptive noise set empirical mode decomposition algorithm based on local mean decomposition, decomposes meteorological data in a time series dimension, performs local mean decomposition prediction fusion on high-frequency components, directly predicts low-frequency components, and performs weighted fusion on the prediction results to obtain accurate results; the present invention adopts a spherical harmonic neural operator based on gated residual spectral attention, uses spherical harmonic functions to convert and process signals, introduces a projection matrix, performs despherical characterization, obtains modal components without spherical features, eliminates the distortion effect of the spherical characteristics of meteorological data on data processing, and increases the accuracy of prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of intelligent weather warning, and specifically relates to a disastrous weather warning system based on big data. Background Art

[0002] The disastrous weather warning system is an important measure to protect the safety of public life and property. The disastrous weather warning system can monitor and warn of extreme weather events in advance, thereby effectively reducing the harm of disastrous weather events to people's lives, property and lives, and helping to reduce casualties and losses. Traditional meteorological warning systems often face problems such as inaccurate data, unclear warning range, and insufficient warning timeliness. They cannot meet the needs for accurate warnings of disastrous weather events. The existing prediction systems based on big data and artificial intelligence ignore the spherical characteristics of meteorological data when predicting and analyzing the weather. When processing meteorological data, the spherical characteristics of the data will cause distortion in data processing, and the prediction accuracy is poor. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the traditional meteorological warning system often faces problems such as inaccurate data, vague warning range, insufficient warning timeliness, etc., and cannot meet the problem of accurate warning needs for disastrous weather events. The present invention adopts a fully adaptive noise set empirical mode decomposition algorithm based on local mean decomposition, decomposes real-time meteorological data in the time series dimension, and classifies the intrinsic mode components obtained by the decomposition, performs local mean decomposition prediction fusion on the high-frequency components, directly predicts the low-frequency components, and weightedly fuses the prediction results to obtain high-accuracy prediction results; in view of the existing intelligent prediction system in When forecasting and analyzing the weather, the spherical characteristics of meteorological data distributed on the earth's spherical surface are ignored, resulting in data processing distortion and poor prediction accuracy. The present invention adopts a spherical harmonic neural operator based on gated residual spectral attention. On the basis of fully adaptive ensemble empirical mode decomposition of noise (CEEMDAN), spherical harmonic functions are used to transform and process the signal, and a projection matrix is ​​introduced to remove the spherical influence to obtain the spherical modal component. The local mean decomposition is then used to further decompose and fuse the modal components without spherical characteristics, thereby eliminating the distortion caused by the spherical characteristics of meteorological data on data processing and increasing the accuracy of the prediction.

[0004] The disastrous weather warning system based on big data provided by the present invention includes a data collection and integration module, a data cleaning and preprocessing module, an intelligent prediction module, a real-time monitoring and warning module, a geographic information system integration module and a user interaction and control module;

[0005] The data collection and integration module collects real-time meteorological data time series diagrams and geographic information data from the meteorological station and sends them to the data cleaning and preprocessing module;

[0006] The data cleaning and preprocessing module performs preprocessing work such as cleaning, denoising, and filling missing values ​​on the meteorological data time series graph and geographic information data, and sends the preprocessed meteorological data time series graph and geographic information data to the intelligent prediction module and the geographic information system integration module;

[0007] The intelligent prediction module uses a fully adaptive noise set empirical mode decomposition algorithm to decompose the preprocessed meteorological data time series diagram into n intrinsic mode components and a residual, and uses a spherical harmonic neural operator based on gated residual spectral attention to remove the spherical influence of the intrinsic mode components to obtain n despherical mode components, and finally performs local mean decomposition processing and neural network prediction on the despherical mode components and residuals to obtain weather forecast information, and sends the weather forecast information to the real-time monitoring and early warning module;

[0008] The real-time monitoring and early warning module monitors the meteorological data time series diagram and geographic information data in real time, analyzes the weather forecast information, and issues early warning information in a timely manner if there is a disastrous weather forecast;

[0009] The geographic information system integration module integrates the warning information with the pre-processed geographic information data to visualize the space of the disastrous weather event;

[0010] The user interaction and control module provides a user interface and data storage functions for users to view warning information, receive warning information and update it in real time, and also allows users to control all modules.

[0011] Furthermore, in the intelligent prediction module, the preprocessed meteorological data time series graph is decomposed to obtain n intrinsic mode components and a residual, and the spherical harmonic neural operator based on gated residual spectral attention is used to remove the spherical influence of the intrinsic mode components to obtain the removed spherical mode components, which specifically includes the following steps:

[0012] Step S1: Empirical mode decomposition, adding an adaptive white noise sequence to the meteorological data time series diagram to generate multiple groups of random time series diagrams, and performing empirical mode decomposition on the multiple groups of random time series diagrams to obtain a group of intrinsic mode components, and setting convergence conditions at the same time;

[0013] Step S2: weighted processing, weighted superposition of a group of eigenmode components to obtain a total eigenmode component;

[0014] Step S3: multi-stage processing, repeating steps S1 and S2 until convergence, and obtaining n intrinsic mode components and a residual;

[0015] Step S4: frequency domain conversion, defining the intrinsic mode component as the initial signal, using an encoder to lift the initial signal to a high-dimensional channel space, and using spherical harmonic functions to convert it to obtain a spectral domain signal;

[0016] Step S5: multi-layer processing, performing multi-layer network processing on the spectral domain signal to obtain a multi-layer processing vector, the formula used is as follows:

[0017] ;

[0018] ;

[0019] ;

[0020] ;

[0021] In the formula, represents the number of layers in a multi-layer network, represents the spectral domain signal, represents the learnable projection matrix, represents the learnable bias, It is processed by a multi-layer perceptron. is the activation function, Mask generation function for the lower triangular matrix, represents the conjugate transpose calculation, represent The matrix of represents the hadamard product operation, for The identity matrix of is the preset parameter, , and is the intermediate variable, represents a multi-layer processing vector;

[0022] Step S6: Multi-head attention processing: multi-head attention processing is performed on the spectral domain signal, and the processing result is activated and fused with the multi-layer processing vector to obtain the despherical modal component. The formula used is as follows:

[0023] ;

[0024] ;

[0025] ;

[0026] In the formula, represents the spectral domain signal, and represents the learned projection matrix, and represents the learnable bias, , is the intermediate variable processed by multi-head attention, Represents the smooth maximum unit activation function, represents the despherical modal component.

[0027] Furthermore, in the intelligent prediction module, the despherical modal components and residuals are subjected to local mean decomposition processing and neural network prediction to obtain weather forecast information, which specifically includes the following steps:

[0028] Step Q1: After decoding the desphericalized modal component corresponding to the highest frequency eigenmodal component, perform local mean decomposition processing to obtain k product function components and a residual;

[0029] Step Q2: Use a neural network to predict and output k product function components and a residual respectively, and obtain k component prediction values ​​and a residual prediction value;

[0030] Step Q3: Perform weighted summation on the k component prediction values ​​and a residual prediction value to obtain a high-frequency prediction result;

[0031] Step Q4: Decode n-1 desphericalized modal components other than the intrinsic modal components of the high-frequency prediction results, and after decoding, use a neural network to predict and output the n-1 decoded results respectively to obtain n-1 low-frequency prediction results;

[0032] Step Q5: Perform weighted summation on the high-frequency prediction results and n-1 low-frequency prediction results to obtain weather forecast information;

[0033] Step Q6: Compare the weather forecast information and the actual weather information to obtain a comparison result, and feed the comparison result back to the intelligent prediction module.

[0034] The beneficial results achieved by the present invention using the above scheme are as follows:

[0035] (1) In view of the fact that traditional meteorological warning systems often face problems such as inaccurate data, vague warning range, and insufficient warning timeliness, and are unable to meet the demand for accurate warning of disastrous weather events, the present invention adopts a fully adaptive noise ensemble empirical mode decomposition algorithm based on local mean decomposition to decompose real-time meteorological data in the time series dimension, and classifies the intrinsic mode components obtained by the decomposition, performs local mean decomposition prediction fusion on high-frequency components, directly predicts low-frequency components, and weightedly fuses the prediction results to obtain high-accuracy prediction results;

[0036] (2) In view of the fact that the existing intelligent prediction system ignores the spherical characteristics of meteorological data distributed on the earth's spherical surface when performing weather forecast analysis, resulting in data processing distortion and poor prediction accuracy, the present invention adopts a spherical harmonic neural operator based on gated residual spectral attention. On the basis of fully adaptive ensemble empirical mode decomposition of noise (CEEMDAN), spherical harmonic functions are used to transform the signal, and a projection matrix is ​​introduced to remove the spherical influence to obtain the spherical modal component. The modal component without spherical characteristics is further decomposed and fused for prediction using local mean decomposition, thereby eliminating the distortion caused by the spherical characteristics of meteorological data on data processing and increasing the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A module diagram of a disastrous weather warning system based on big data provided by the present invention;

[0038] Figure 2 Schematic diagram of the data processing flow of the intelligent prediction module.

[0039] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0041] Example 1, see Figure 1 The disastrous weather warning system based on big data provided by the present invention includes a data collection and integration module, a data cleaning and preprocessing module, an intelligent prediction module, a real-time monitoring and warning module, a geographic information system integration module and a user interaction and control module;

[0042] The data collection and integration module collects real-time meteorological data time series diagrams and geographic information data from the meteorological station and sends them to the data cleaning and preprocessing module;

[0043] The data cleaning and preprocessing module performs preprocessing work such as cleaning, denoising, and filling missing values ​​on the meteorological data time series graph and geographic information data, and sends the preprocessed meteorological data time series graph and geographic information data to the intelligent prediction module and the geographic information system integration module;

[0044] The intelligent prediction module uses a fully adaptive noise ensemble empirical mode decomposition algorithm (CEEMDAN) to decompose the preprocessed meteorological data time series diagram to obtain 10 intrinsic mode components and one residual, and uses a spherical harmonic neural operator based on gated residual spectral attention to remove the spherical influence of the intrinsic mode components to obtain 10 despherical mode components. Finally, the despherical mode components and residuals are subjected to local mean decomposition processing and neural network prediction to obtain weather forecast information, and the weather forecast information is sent to the real-time monitoring and early warning module;

[0045] The real-time monitoring and early warning module monitors the meteorological data time series diagram and geographic information data in real time, analyzes the weather forecast information, and issues early warning information in a timely manner if there is a disastrous weather forecast;

[0046] The geographic information system integration module integrates the warning information with the pre-processed geographic information data to visualize the space of the disastrous weather event;

[0047] The user interaction and control module provides a user-friendly interface and data storage function, allowing users to easily view warning information, receive warning information and update it in real time, and allowing users to control all modules.

[0048] By performing the above operations, traditional meteorological warning systems often face problems such as inaccurate data, vague warning range, insufficient warning timeliness, and are unable to meet the needs of accurate warnings for disastrous weather events. The present invention adopts a fully adaptive noise set empirical mode decomposition algorithm based on local mean decomposition to decompose real-time meteorological data in the time series dimension, classify the intrinsic mode components obtained by the decomposition, perform local mean decomposition prediction fusion on high-frequency components, directly predict low-frequency components, and weightedly fuse the prediction results to obtain high-accuracy prediction results.

[0049] Example 2, see Figure 2 Based on the above embodiment, in the intelligent prediction module, the preprocessed meteorological data time series diagram is decomposed to obtain 10 intrinsic mode components and one residual, and the spherical harmonic neural operator based on gated residual spectrum attention is used to remove the spherical surface effect on the intrinsic mode components to obtain the spherical surface removed mode components, which specifically includes the following steps:

[0050] Step S1: EMD decomposition, adding an adaptive white noise sequence to the meteorological data time series diagram to generate multiple groups of random time series diagrams, and performing EMD decomposition on the multiple groups of random time series diagrams to obtain a group of intrinsic mode components, and setting convergence conditions at the same time;

[0051] Step S2: weighted processing, weighted superposition of a group of eigenmode components to obtain a total eigenmode component;

[0052] Step S3: multi-level processing, repeating steps S1 and S2 until convergence, and obtaining 10 intrinsic mode components and one residual;

[0053] Step S4: frequency domain conversion, defining the intrinsic mode component as the initial signal, using an encoder to lift the initial signal to a high-dimensional channel space, and using spherical harmonic functions to convert it to obtain a spectral domain signal;

[0054] Step S5: multi-layer processing, performing multi-layer network processing on the spectral domain signal to obtain a multi-layer processing vector, the formula used is as follows:

[0055] ;

[0056] ;

[0057] ;

[0058] ;

[0059] In the formula, represents the number of layers in a multi-layer network, represents the spectral domain signal, represents the learnable projection matrix, represents the learnable bias, It is processed by a multi-layer perceptron. is the activation function, Mask generation function for the lower triangular matrix, represents the conjugate transpose calculation, represent The matrix of represents the hadamard product operation, for The identity matrix of is the preset parameter, , and is the intermediate variable, represents a multi-layer processing vector;

[0060] Step S6: Multi-head attention processing: multi-head attention processing is performed on the spectral domain signal, and the processing result is activated and fused with the multi-layer processing vector to obtain the despherical modal component. The formula used is as follows:

[0061] ;

[0062] ;

[0063] ;

[0064] In the formula, represents the spectral domain signal, and represents the learned projection matrix, and represents the learnable bias, , is the intermediate variable processed by multi-head attention, represents the SMU activation function, represents the despherical modal component.

[0065] Example 3, see Figure 2 Based on the above embodiment, this embodiment performs local mean decomposition processing and neural network prediction on the despherical modal component and the residual in the intelligent prediction module to obtain weather forecast information, which specifically includes the following steps:

[0066] Step Q1: After decoding the desphericalized modal component corresponding to the highest frequency intrinsic modal component, perform local mean decomposition processing to obtain four PF components and a residual;

[0067] Step Q2: Use a neural network to predict and output the four PF components and one residual respectively, and obtain four component prediction values ​​and one residual prediction value;

[0068] Step Q3: Perform weighted summation on the four component prediction values ​​and one residual prediction value to obtain a high-frequency prediction result;

[0069] Step Q4: Decode the 9 desphericalized modal components other than the highest frequency intrinsic modal component, and after decoding, use a neural network to predict and output the 9 decoding results respectively to obtain 9 low-frequency prediction results;

[0070] Step Q5: Perform weighted summation on the high-frequency prediction results and the nine low-frequency prediction results to obtain weather forecast information;

[0071] Step Q6: Optimize the intelligent prediction module according to the accuracy of weather forecast information.

[0072] The weather data of Lanzhou in the past two months were collected as a test set, and the prediction ability of the neural network corresponding to different numbers of intrinsic mode components and PF components was trained:

[0073] The number of PF components is set to 6, and the prediction capabilities of the neural network corresponding to different numbers of intrinsic mode components are calculated:

[0074] Number: 5; RMSE (℃): 0.987; MAPE (%): 0.037; MAE (℃): 0.997;

[0075] Number: 10; RMSE (℃): 0.354; MAPE (%): 0.010; MAE (℃): 0.348;

[0076] Number: 15; RMSE (℃): 0.458; MAPE (%): 0.021; MAE (℃): 0.449;

[0077] The number of intrinsic mode components is set to 10, and the prediction ability of the neural network corresponding to different numbers of PF components is calculated:

[0078] Number: 2; RMSE (℃): 1.013; MAPE (%): 0.031; MAE (℃): 0.985;

[0079] Number: 3; RMSE (℃): 0.713; MAPE (%): 0.025; MAE (℃): 0.701;

[0080] Number: 4; RMSE (℃): 0.332; MAPE (%): 0.011; MAE (℃): 0.326;

[0081] Number: 5; RMSE (℃): 0.368; MAPE (%): 0.014; MAE (℃): 0.385;

[0082] Number: 6; RMSE (℃): 0.354; MAPE (%): 0.010; MAE (℃): 0.348;

[0083] Subsequently, the neural model is iteratively optimized by combining different numbers of eigenmode components and PF components.

[0084] In view of the fact that the existing intelligent prediction system ignores the spherical characteristics of meteorological data distributed on the earth's spherical surface when performing weather forecast analysis, resulting in data processing distortion and poor prediction accuracy, the present invention adopts a spherical harmonic neural operator based on gated residual spectral attention. On the basis of fully adaptive ensemble empirical mode decomposition of noise (CEEMDAN), spherical harmonic functions are used to transform and process the signal, and a projection matrix is ​​introduced to remove the spherical influence to obtain the spherical modal component. Then, local mean decomposition is used to further decompose and process the modal component without spherical characteristics and fusion prediction, so as to eliminate the distortion caused by the spherical characteristics of meteorological data on data processing and increase the accuracy of prediction.

[0085] Embodiment 4: This embodiment is based on the above embodiment. In step S4, the spherical harmonic function is used as follows:

[0086] Language selection: Python, install scipy and add dependencies;

[0087] Importing multi-modal components: Importing 10 intrinsic mode components, where the intrinsic mode components are time series modes, the frequency represents the frequency of real-time collection of meteorological data, and the amplitude represents the value of meteorological data. Here, temperature is used as meteorological data.

[0088] Core implementation code:

[0089] def sht(mode_number,natural_frequency,mode_amplitude):

[0090] # Input is the mode number, natural frequency and mode amplitude

[0091] # Output is the amplitude and phase of the corresponding mode and the spectral domain signal image described in step S4

[0092] amplitude=mode_amplitude*np.sqrt(2)#amplitude calculation

[0093] phase=2*np.pi mode_number / len(natural_frequency)#Phase calculation

[0094] return amplitude,phase

[0095] # Example input

[0096] mode_number =1#Mode number

[0097] natural_frequency=np.array([10, 20, 30])#Natural sampling frequency of meteorological data

[0098] mode_amplitude =np.array([40,30,20])#Mode amplitude, that is, the mean temperature within the sampling frequency period

[0099] #Call sht (spherical harmonics)

[0100] amplitude,phase=sht(mode_number,natural_frequency,mode_amplitude)

[0101] # Output results

[0102] print(f"Amplitude:{amplitude},Phase:{phase}").

[0103] Embodiment 5: This embodiment is based on the above embodiment. In step Q1 and step Q4, the neural network details are as follows:

[0104] Neural network selection: Long short-term memory network;

[0105] Input sequence: The LSTM model receives input sequence data, and the input of each time step is represented as a vector;

[0106] Gating unit: Initialize three LSTM gating units, namely input gate, forget gate, output gate, and a cell state to control the flow and memory of information;

[0107] Input gate: The input gate controls the degree to which the new input information updates the cell state. It contains a sigmoid activation function, which takes the input data and the output of the previous time step as input and outputs a value between 0 and 1, indicating the importance of each unit.

[0108] Forget gate: The forget gate determines which information needs to be forgotten or remembered. It uses a sigmoid activation function and the output of the previous time step to output a value between 0 and 1 to determine how much cell state needs to be retained.

[0109] Cell state update: Update through the operation of input gate and forget gate to retain relevant information;

[0110] Output gate: Outputs a value between 0 and 1 through a sigmoid activation function and the output of the previous time step, indicating the importance of the output;

[0111] Hidden state output: The LSTM model finally outputs the hidden state controlled by the output gate, which will also serve as the input for the next time step;

[0112] Backpropagation: Through the backpropagation algorithm, the LSTM model learns to update the weights to minimize the loss function.

[0113] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0114] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0115] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. Severe weather warning system based on big data, Features: It includes data collection and integration module, data cleaning and preprocessing module, intelligent prediction module, real-time monitoring and early warning module, geographic information system integration module and user interaction and control module; The data collection and integration module collects real-time meteorological data time series diagrams and geographic information data from the meteorological station and sends them to the data cleaning and preprocessing module; The data cleaning and preprocessing module performs preprocessing work such as cleaning, denoising, and filling missing values ​​on the meteorological data time series graph and geographic information data, and sends the preprocessed meteorological data time series graph and geographic information data to the intelligent prediction module and the geographic information system integration module; The intelligent prediction module uses a fully adaptive noise set empirical mode decomposition algorithm to decompose the preprocessed meteorological data time series diagram into n intrinsic mode components and a residual, and uses a spherical harmonic neural operator based on gated residual spectral attention to remove the spherical influence of the intrinsic mode components to obtain n despherical mode components, and finally performs local mean decomposition processing and neural network prediction on the despherical mode components and residuals to obtain weather forecast information, and sends the weather forecast information to the real-time monitoring and early warning module; The real-time monitoring and early warning module monitors the meteorological data time series diagram and geographic information data in real time, analyzes the weather forecast information, and issues early warning information in a timely manner if there is a disastrous weather forecast; The geographic information system integration module integrates the warning information with the pre-processed geographic information data to visualize the space of the disastrous weather event; The user interaction and control module provides a user interface and data storage functions for users to view warning information, receive warning information and update it in real time, and also allows users to control all modules.

2. According to the big data-based severe weather warning system of claim 1, Features: In the intelligent prediction module, the preprocessed meteorological data time series graph is decomposed into n intrinsic mode components and a residual, and the spherical harmonic neural operator based on gated residual spectral attention is used to remove the spherical influence of the intrinsic mode components to obtain the removed spherical mode components, which specifically includes the following steps: Step S1: Empirical mode decomposition, adding an adaptive white noise sequence to the meteorological data time series diagram to generate multiple groups of random time series diagrams, and performing empirical mode decomposition on the multiple groups of random time series diagrams to obtain a group of intrinsic mode components, and setting convergence conditions at the same time; Step S2: weighted processing, weighted superposition of a group of eigenmode components to obtain a total eigenmode component; Step S3: multi-stage processing, repeating steps S1 and S2 until convergence, and obtaining n intrinsic mode components and a residual; Step S4: frequency domain conversion, defining the intrinsic mode component as the initial signal, using an encoder to lift the initial signal to a high-dimensional channel space, and using spherical harmonic functions to convert it to obtain a spectral domain signal; Step S5: multi-layer processing, performing multi-layer network processing on the spectral domain signal to obtain a multi-layer processing vector, the formula used is as follows: ; ; ; ; In the formula, represents the number of layers in a multi-layer network, represents the spectral domain signal, represents the learnable projection matrix, represents the learnable bias, It is processed by a multi-layer perceptron. is the activation function, Mask generation function for the lower triangular matrix, represents the conjugate transpose calculation, represent The matrix of represents the hadamard product operation, for The identity matrix of is the preset parameter, , and is the intermediate variable, represents a multi-layer processing vector, represent The multi-layer processing vector of the previous network; Step S6: Multi-head attention processing: multi-head attention processing is performed on the spectral domain signal, and the processing result is activated and fused with the multi-layer processing vector to obtain the despherical modal component. The formula used is as follows: ; ; ; In the formula, represents the spectral domain signal, and represents the learnable projection matrix, and represents the learnable bias, , is the intermediate variable processed by multi-head attention, Represents the smooth maximum unit activation function, represents the despherical modal component.

3. According to the disastrous weather warning system based on big data as claimed in claim 2, Features: In the intelligent prediction module, the despherical modal components and residuals are subjected to local mean decomposition and neural network prediction to obtain weather forecast information, which specifically includes the following steps: Step Q1: After decoding the desphericalized modal component corresponding to the highest frequency eigenmodal component, perform local mean decomposition processing to obtain k product function components and a residual; Step Q2: Use a neural network to predict and output k product function components and a residual respectively, and obtain k component prediction values ​​and a residual prediction value; Step Q3: Perform weighted summation on the k component prediction values ​​and a residual prediction value to obtain a high-frequency prediction result; Step Q4: Decode n-1 desphericalized modal components other than the intrinsic modal components of the high-frequency prediction results, and after decoding, use a neural network to predict and output the n-1 decoded results respectively to obtain n-1 low-frequency prediction results; Step Q5: Perform weighted summation on the high-frequency prediction results and n-1 low-frequency prediction results to obtain weather forecast information; Step Q6: Compare the weather forecast information and the actual weather information to obtain a comparison result, and feed the comparison result back to the intelligent prediction module.