A lightning disaster early warning method based on meteorological data

Abstract

The application discloses a thunder and lightning disaster early warning method based on meteorological data. The method comprises the following steps: fusing meteorological data and terrain data, generating a time-space distribution data set of atmospheric parameters and terrain features in a complex mountainous environment through a weighted superposition and space-time interpolation method; analyzing electromagnetic wave multi-path propagation effect according to the time-space distribution data set, calculating the attenuation coefficient of high-frequency components greater than 1 gigahertz and the propagation loss of low-frequency components less than 1 gigahertz, and generating a preliminary quantitative data set of spectrum degradation; if the high-frequency component attenuation coefficient in the preliminary quantitative data set of spectrum degradation exceeds an attenuation threshold value, then the propagation path parameters are optimized through a genetic algorithm to generate a corrected spectrum degradation model, and the like. The application can significantly improve the accuracy and real-time performance of the thunder and lightning disaster early warning in the complex mountainous area, and effectively reduce the disaster risk.

Description

Technical Field

[0001] The present invention relates to the technical field of lightning disaster early warning, and in particular to a lightning disaster early warning method based on meteorological data. Background Art

[0002] Lightning disaster warning is a key research direction in the field of disaster prevention and mitigation, directly related to the safety of life and property, and economic and social stability. Especially in complex mountainous environments, where lightning activity is frequent and highly destructive, accurate warnings are crucial for ensuring the safety of transportation, power facilities, and residents in mountainous areas. Currently, lightning warning methods based on meteorological data have achieved certain results in flat terrain or stable meteorological conditions. However, when faced with complex mountainous scenarios, existing methods are clearly insufficiently adaptable. Traditional technologies often ignore the combined effects of terrain undulations, vegetation cover, and dynamic meteorological conditions on lightning signal propagation, resulting in reduced warning accuracy. This is especially true in severe convective weather, where short-term meteorological changes further exacerbate the difficulty of prediction. These limitations make existing methods difficult to meet the actual needs of lightning disaster warnings in mountainous areas.

[0003] In complex mountainous environments, spectrum degradation of electromagnetic wave propagation is a core factor affecting the accuracy of early warnings. Due to terrain obstruction and vegetation absorption, the high-frequency components of electromagnetic waves are significantly attenuated, while the low-frequency components are relatively stable, resulting in signal spectrum distortion, which in turn affects the identification and positioning accuracy of lightning types. More importantly, the dynamic changes in atmospheric humidity and rainfall intensity have a particularly prominent nonlinear effect on high-frequency signals. For example, during localized heavy rainfall, short-term high humidity environments will accelerate the attenuation of high-frequency signals, making it difficult for early warning algorithms based on fixed parameters to accurately distinguish between cloud-to-ground lightning, resulting in positioning errors or missed reports. The complexity of spectrum degradation stems from the multipath effect and the transient nature of meteorological conditions. Existing technologies have difficulty in quantifying the attenuation patterns of signals in different frequency bands in real time, and are unable to dynamically adjust algorithm parameters to adapt to the rapidly changing mountain meteorological environment.

[0004] Therefore, how to integrate multi-source meteorological data such as atmospheric humidity and rainfall intensity in complex mountainous scenarios, accurately quantify the differentiated impact of spectrum degradation on signals in different frequency bands, and dynamically optimize the warning algorithm parameters has become a key issue in improving the accuracy of lightning disaster warnings. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to propose a lightning disaster warning method based on meteorological data, which can significantly improve the accuracy and real-time performance of lightning disaster warning in complex mountainous areas and effectively reduce disaster risks.

[0006] According to one aspect of the present invention, a lightning disaster early warning method based on meteorological data is provided, the method comprising:

[0007] The meteorological data and the terrain data are fused, a time-space distribution data set of atmospheric parameters and terrain features in a complex mountainous environment is generated by weighted superposition and time-space interpolation method;

[0008] According to the time-space distribution data set, the multi-path propagation effect of electromagnetic waves is analyzed, the attenuation coefficient of high-frequency components greater than 1 GHz and the propagation loss of low-frequency components less than 1 GHz are calculated, and a preliminary quantitative data set of spectral degradation is generated;

[0009] If the high-frequency component attenuation coefficient in the preliminary quantitative data set of spectral degradation exceeds the attenuation threshold, the propagation path parameters are optimized by a genetic algorithm, and a corrected spectral degradation model is generated;

[0010] According to the corrected spectral degradation model, the atmospheric humidity change rate and the rainfall intensity change rate in the dynamic meteorological data are extracted, the nonlinear attenuation coefficient of high-frequency signals is calculated, and a frequency band differentiation impact data set is generated;

[0011] According to the frequency band differentiation impact data set, combined with the signal waveform characteristics, the lightning types are classified, the cloud flash and ground flash feature distributions are distinguished, and a lightning type recognition data set and a confidence level are generated;

[0012] If the confidence level of the lightning type recognition data set is lower than the confidence threshold, the real-time meteorological transient characteristic data is fused, the support vector machine kernel function and the penalty coefficient are adjusted, and an optimized lightning type recognition model is generated;

[0013] According to the lightning type and feature distribution generated by the optimized lightning type recognition model, combined with the dynamic meteorological data, the state of the lightning signal propagation path is estimated, and a lightning positioning accuracy data set is generated;

[0014] According to the lightning positioning accuracy data set, the real-time meteorological data and the terrain data are updated, the dynamic parameters of the corrected spectral degradation model are adjusted, and a lightning disaster warning parameter data set is generated; according to the lightning disaster warning parameter data set, it is transmitted to the warning system, and the lightning disaster warning in the complex mountainous environment is completed.

[0015] In the above technical scheme, the scheme aims to improve the accuracy and effectiveness of lightning disaster warning in complex mountainous environment.

[0016] Firstly, meteorological data and terrain data are fused, and weighted superposition and spatio-temporal interpolation methods are used to generate a spatio-temporal distribution dataset of atmospheric parameters and terrain features in complex mountainous environments. This step can fully consider the influence of complex mountainous terrain on meteorological elements and construct a more realistic spatio-temporal distribution. Based on the spatio-temporal distribution dataset, the electromagnetic wave multi-path propagation effect is analyzed, and the attenuation coefficient of high-frequency components greater than 1 gigahertz and the propagation loss of low-frequency components less than 1 gigahertz are calculated to generate a preliminary quantitative dataset of spectral degradation. Such analysis helps to understand the changes in the propagation characteristics of electromagnetic waves of different frequency bands in complex mountainous environments. When the high-frequency component attenuation coefficient in the preliminary quantitative dataset of spectral degradation exceeds the attenuation threshold, the genetic algorithm is used to optimize the propagation path parameters, thereby generating a corrected spectral degradation model. The introduction of genetic algorithm can effectively search for better propagation path parameters, further improve the description accuracy of the model for spectral degradation, and make it more suitable for the actual complex and variable mountain lightning propagation environment. Based on the corrected spectral degradation model, the atmospheric humidity change rate and rainfall intensity change rate are extracted from the dynamic meteorological data, and then the nonlinear attenuation coefficient of high-frequency signals is calculated to generate a frequency-differentiated impact dataset. This step combines dynamic meteorological changes to further explore the differentiation of different frequency band signals affected by meteorological factors, providing more detailed data basis for subsequent lightning type recognition. According to the frequency-differentiated impact dataset and signal waveform characteristics, lightning types are classified, and the characteristic distribution of cloud flashes and ground flashes is determined to generate a lightning type recognition dataset and confidence. If the confidence is lower than the set threshold, the real-time meteorological transient characteristic data is fused, the support vector machine kernel function and penalty coefficient are adjusted, and an optimized lightning type recognition model is generated. This confidence-based feedback optimization mechanism can continuously improve the accuracy and reliability of lightning type recognition, ensuring the scientificity of lightning type judgment. Relying on the lightning type and characteristic distribution obtained by the optimized lightning type recognition model, and combining dynamic meteorological data, the state of lightning signal propagation path is estimated to generate a lightning positioning accuracy dataset. This step helps to accurately determine the propagation trajectory of lightning signals and locate the lightning occurrence position, providing more accurate spatial information for subsequent warning. According to the lightning positioning accuracy dataset, the real-time meteorological data and terrain data are updated, the dynamic parameters of the corrected spectral degradation model are adjusted, a lightning disaster warning parameter dataset is generated, and it is transmitted to the warning system, finally completing the lightning disaster warning in complex mountainous environments. The whole process forms a closed-loop dynamic updating and warning mechanism, which can timely issue accurate lightning disaster warning according to the latest data and model optimization results, and strive for valuable time for disaster prevention and reduction.

[0017] The weather data-based lightning disaster early warning method is closely connected and mutually coordinated from data fusion to model optimization to the final early warning implementation, and fully considers many challenges faced by lightning disaster early warning in complex mountainous environments, such as the influence of terrain on weather and electromagnetic wave propagation, and the difficulty of lightning type identification. By using various advanced algorithms and data processing techniques, it is expected to significantly improve the accuracy and timeliness of lightning disaster early warning, provide strong protection for lightning protection in complex mountainous areas, and reduce the loss caused by lightning disasters.

[0018] In some embodiments, the weather data and the terrain data are fused, and a spatiotemporal distribution dataset of atmospheric parameters and terrain features in a complex mountainous environment is generated by weighted superposition and spatiotemporal interpolation methods, including:

[0019] Atmospheric humidity, rainfall intensity, and temperature in the weather data and terrain undulations and vegetation coverage in the terrain data are obtained, and standardized weather and terrain datasets are generated by preprocessing;

[0020] A weighted superposition method is used to assign different weights to atmospheric humidity, rainfall intensity, and temperature, and a comprehensive weather influence factor is generated by linear combination to obtain a comprehensive weather dataset;

[0021] A spatiotemporal interpolation method is used to extract spatiotemporal distribution features from the comprehensive weather dataset and the terrain data to generate continuous weather and terrain spatiotemporal distribution datasets;

[0022] If the rainfall intensity in the comprehensive weather dataset exceeds the intensity threshold, the rainfall intensity is subjected to Gaussian smoothing to generate a smoothed weather spatiotemporal distribution dataset;

[0023] According to the smoothed weather spatiotemporal distribution dataset and the terrain spatiotemporal distribution dataset, a Kriging interpolation method is used to generate a spatiotemporal distribution dataset of atmospheric parameters and terrain features in a complex mountainous environment.

[0024] In the above technical solution, the weather data and the terrain data are fused to generate a spatiotemporal distribution dataset of atmospheric parameters and terrain features in a complex mountainous environment. Various methods are used to process, analyze, and optimize the data throughout the process, aiming to provide accurate and spatiotemporally continuous data basis for subsequent related research or application.

[0025] First, the atmospheric humidity, rainfall intensity, temperature in the meteorological data and the terrain relief and vegetation coverage in the terrain data are obtained, and the key elements are preprocessed to generate standardized meteorological and terrain data sets. This step is the basis of the whole process. Through standardization processing, data of different sources and formats can be unified, preparing for subsequent fusion and analysis. By using the weighted superposition method, different weights are assigned to atmospheric humidity, rainfall intensity and temperature, and a comprehensive meteorological influence factor is generated by linear combination to obtain a comprehensive meteorological data set. Here, weights are assigned according to the importance of different meteorological elements in the actual environmental impact, which can more scientifically consider the combined action of each meteorological factor, making the generated comprehensive meteorological data set more representative and valuable. The spatio-temporal distribution characteristics are extracted from the comprehensive meteorological data set and the terrain data, and the continuous meteorological and terrain spatio-temporal distribution data set is generated by using the spatio-temporal interpolation method. The application of spatio-temporal interpolation method can make up for the lack of data in time and space, and generate continuous spatio-temporal distribution data set, which provides more complete data support for the subsequent research on the spatio-temporal variation law of atmospheric parameters and terrain characteristics. If the rainfall intensity in the comprehensive meteorological data set exceeds the intensity threshold, the rainfall intensity is processed by Gaussian smoothing to generate a smoothed meteorological spatio-temporal distribution data set. This is an optimization processing for rainfall intensity data. When the rainfall intensity is too high, Gaussian smoothing can smooth the mutation part of the data and reduce the influence of extreme values, making the meteorological spatio-temporal distribution data set more reasonable and stable. According to the smoothed meteorological spatio-temporal distribution data set and the terrain spatio-temporal distribution data set, the Kriging interpolation method is used to generate the spatio-temporal distribution data set of atmospheric parameters and terrain characteristics in complex mountainous environment. The Kriging interpolation method can fully consider the spatial correlation and other characteristics of the data, further improve the accuracy and reliability of the final spatio-temporal distribution data set, and make it more consistent with the actual characteristics of complex mountainous environment.

[0026] The above steps effectively fuse meteorological data and terrain data, and reasonably process and optimize the data, finally generating the spatio-temporal distribution data set of atmospheric parameters and terrain characteristics in complex mountainous environment.

[0027] In some embodiments, according to the spatio-temporal distribution data set, the electromagnetic wave multi-path propagation effect is analyzed, the attenuation coefficient of high frequency component greater than 1 gigahertz and the propagation loss of low frequency component less than 1 gigahertz are calculated, and a preliminary quantitative data set of spectral degradation is generated, including:

[0028] The spatio-temporal distribution data set is obtained, and the data is cleaned by using the preprocessing method to eliminate noise points and outliers, and a standardized data set is obtained;

[0029] The electromagnetic wave propagation path of the standardized data set is simulated by ray tracing method to determine the reflection and refraction and diffraction paths in the multi-path effect, and a propagation path set is obtained;

[0030] For the set of propagation paths, the attenuation coefficient of electromagnetic waves with high frequency components greater than 1 GHz is calculated, and if the number of path reflections is greater than a preset threshold, the geometric optics model is used to calculate the attenuation to obtain a set of high-frequency attenuation coefficients;

[0031] According to the set of propagation paths, the propagation loss of electromagnetic waves with low frequency components less than 1 GHz is analyzed, and if the path diffraction angle is greater than a preset threshold, the uniform geometric diffraction theory is used to calculate the loss to obtain a set of low-frequency propagation loss;

[0032] From the set of high-frequency attenuation coefficients and the set of low-frequency propagation losses, extract the spectral degradation features, and use the fast Fourier transform algorithm to convert the features to the frequency domain to obtain spectral degradation data;

[0033] Through the spectral degradation data, combined with the space-time distribution characteristics, a quantitative data model is constructed, the least squares method is used to fit the spectral degradation trend, and a quantitative data set is obtained;

[0034] For the quantitative data set, perform data verification, and if the data points deviate from the fitted curve by more than a preset threshold, remove the abnormal points to obtain the final spectral degradation quantitative data set.

[0035] In the above technical solution, focus on how to analyze the multi-path propagation effect of electromagnetic waves in complex mountainous environments based on the space-time distribution data set, and accurately calculate the attenuation and loss of electromagnetic waves in different frequency bands, and finally generate a reliable preliminary quantitative data set of spectral degradation.

[0036] After obtaining the spatio-temporal distribution dataset, the data is cleaned by preprocessing methods to remove noise points and outliers, obtaining the standardized dataset. Through preprocessing, the data quality can be effectively improved to ensure the accuracy and reliability of the input data. The ray tracing method is used to simulate the electromagnetic wave propagation path of the standardized dataset, determine the reflection and refraction and diffraction paths in the multipath effect, and obtain the propagation path set. The ray tracing method can simulate the multiple propagation paths of electromagnetic waves according to the topography and atmospheric parameter changes in complex mountainous environments, and can visualize the abstract electromagnetic wave propagation process, providing a clear object set for subsequent electromagnetic wave attenuation and loss calculation for different paths. For the propagation path set, the electromagnetic wave attenuation coefficient of the high-frequency component greater than 1 gigahertz is calculated. If the number of reflections is greater than the preset threshold, the geometric optics model is used to calculate the attenuation, and the high-frequency attenuation coefficient set is obtained. The calculation method is clearly distinguished under different conditions. When the number of reflections exceeds the threshold, the geometric optics model can more accurately describe the attenuation law of high-frequency electromagnetic waves in the multiple reflection process, making the calculation results more consistent with the actual physical situation. According to the propagation path set, the propagation loss of the low-frequency component less than 1 gigahertz is analyzed. If the diffraction angle is greater than the preset threshold, the uniform geometric diffraction theory is used to calculate the loss, and the low-frequency propagation loss set is obtained. Similarly, for the special propagation characteristics of low-frequency electromagnetic waves, the uniform geometric diffraction theory is introduced for loss calculation, fully considering the diffraction effect of low-frequency electromagnetic waves in complex mountainous terrain, so that the low-frequency propagation loss set can accurately reflect the actual propagation loss of low-frequency electromagnetic waves. From the high-frequency attenuation coefficient set and the low-frequency propagation loss set, the spectral degradation features are extracted, and the fast Fourier transform algorithm is used to convert the features to the frequency domain, obtaining the spectral degradation data. The fast Fourier transform algorithm can convert the spectral degradation features in the time domain to the frequency domain, making the spectral degradation features more intuitive and more suitable for further analysis, providing a convenient data basis for subsequent quantitative data modeling. Through the spectral degradation data, combined with the spatio-temporal distribution characteristics, a quantitative data model is constructed, and the least squares method is used to fit the spectral degradation trend to obtain the quantitative dataset. The extracted spectral degradation data is combined with the spatio-temporal distribution characteristics, and the least squares method is used to fit the overall trend of spectral degradation, quantifying the spectral degradation phenomenon in the form of a mathematical model, making the description of spectral degradation more systematic and regular, and improving the usability and interpretability of the data. For the quantitative dataset, data verification is performed. If the data points deviate from the fitted curve by more than the preset threshold, the abnormal points are removed, and the final spectral degradation quantitative dataset is obtained. The data verification step can timely detect and remove abnormal data points that do not conform to the overall law due to various factors, further improving the accuracy and reliability of the spectral degradation quantitative dataset, and ensuring that the dataset can truly and reliably reflect the spectral degradation of electromagnetic waves in complex mountainous environments.

[0037] The above steps revolve around the core goal of accurately analyzing the multi-path propagation effect of electromagnetic waves in complex mountainous environments. By using various algorithms, the attenuation and loss of electromagnetic waves at different frequency bands are processed, and the final spectrum degradation quantitative data set has high accuracy and reliability.

[0038] In some embodiments, if the high-frequency component attenuation coefficient in the preliminary spectrum degradation quantitative data set exceeds the attenuation threshold, the propagation path parameters are optimized by a genetic algorithm to generate a corrected spectrum degradation model, including:

[0039] The high-frequency component is obtained from the spectrum degradation data set, and the attenuation coefficient is calculated. If the attenuation coefficient exceeds the attenuation threshold, the propagation path parameters are initialized by a genetic algorithm to generate an initial parameter set.

[0040] According to the terrain data and vegetation coverage data, a propagation environment model is constructed, the reflection angle and refractive index are adjusted, and an optimized parameter set is obtained. The parameter set is iteratively optimized by a genetic algorithm, and the propagation environment model is fused to generate a candidate path set.

[0041] The optimal propagation path is obtained from the candidate path set, the corrected high-frequency component attenuation coefficient is calculated, and the updated spectrum degradation data is obtained.

[0042] The updated spectrum degradation data is used to construct a corrected spectrum degradation model to determine the model parameters. The high-frequency component attenuation coefficient is verified by the corrected spectrum degradation model to determine the model accuracy.

[0043] In the above technical solution, the focus is on the specific problem of how to use a genetic algorithm to optimize the propagation path parameters when the high-frequency component attenuation coefficient in the preliminary spectrum degradation quantitative data set exceeds the standard, to generate a corrected spectrum degradation model, and a complete optimization process is provided.

[0044] High-frequency components are extracted from the spectrum degradation dataset, their attenuation coefficients are calculated, and compared with an attenuation threshold. If the attenuation coefficient exceeds the threshold, a genetic algorithm is used to initialize the propagation path parameters to generate an initial parameter set. A propagation environment model is then constructed based on terrain relief and vegetation cover data, and the reflection angle and refractive index are adjusted to obtain an optimized parameter set. The genetic algorithm is introduced here. The generation of the initial parameter set provides a diverse starting point for subsequent optimization. The construction of the propagation environment model fully considers the impact of complex mountain environments on electromagnetic wave propagation, making the optimization process more realistic. The parameter set is iteratively optimized using the genetic algorithm and integrated with the propagation environment model to generate a set of candidate paths. This iterative optimization process continuously selects better parameter combinations, gradually approaching the optimal solution. The integrated propagation environment model ensures the rationality and feasibility of the optimization parameters, resulting in a set of candidate paths that encompasses a wide range of possible propagation paths, providing a richer selection for subsequent optimal path selection. The optimal propagation path is then obtained from the candidate path set, and the revised high-frequency component attenuation coefficient is calculated to obtain updated spectrum degradation data. This step corrects the spectral degradation data by selecting the optimal path and recalculating the attenuation coefficient, ensuring that the data more accurately reflects actual electromagnetic wave propagation conditions. Using the updated spectral degradation data, a revised spectral degradation model is constructed, model parameters are determined, and the high-frequency component attenuation coefficient is verified against this model to determine model accuracy. Rebuilding the model using the corrected data and verifying its accuracy effectively evaluates the effectiveness of the optimization operation and ensures the revised spectral degradation model has higher accuracy and reliability.

[0045] In some embodiments, based on the modified spectrum degradation model, the atmospheric humidity change rate and rainfall intensity change rate in the dynamic meteorological data are extracted, the nonlinear attenuation coefficient of the high-frequency signal is calculated, and a frequency band differential impact data set is generated, including:

[0046] Obtain the atmospheric humidity change rate and rainfall intensity change rate from dynamic meteorological data, and use time series analysis method to obtain the time series data of the change rate;

[0047] Based on the acquired time series data, the Fourier transform method is used to extract the high-frequency signal component and determine the initial amplitude of the high-frequency signal;

[0048] According to the exponential decay formula ,in is the amplitude after attenuation, is the initial amplitude, is the attenuation coefficient, For the propagation distance, the nonlinear attenuation coefficient is calculated in combination with the preset propagation distance;

[0049] If the calculated nonlinear attenuation coefficient exceeds the preset range, the least squares method is used to optimize the attenuation coefficient to obtain an optimized attenuation coefficient;

[0050] From the optimized attenuation coefficient, the attenuation characteristics corresponding to different frequency ranges are extracted to generate a frequency band differentiation impact data set;

[0051] Through cluster analysis on the frequency band differentiation impact data set, the K-means algorithm is used to determine the attenuation mode of each frequency range; and according to the determined attenuation mode, a frequency band differentiation impact data set containing the attenuation coefficient and the frequency range is generated.

[0052] In the above technical solution, the scheme aims to extract the key change rate from dynamic meteorological data based on the corrected spectral degradation model, and then accurately calculate the nonlinear attenuation coefficient of high-frequency signals, and generate a frequency band differentiation impact data set with important reference value.

[0053] The atmospheric humidity change rate and the rainfall intensity change rate are obtained from the dynamic meteorological data, and the time series analysis method is used to obtain the time series data thereof. Through time series analysis, the dynamic change trend of atmospheric humidity and rainfall intensity over time can be clearly shown. Based on the obtained time series data, the Fourier transform method is used to extract the high-frequency signal component to determine the initial amplitude of the high-frequency signal. Fourier transform converts time series data from time domain to frequency domain, so that the high-frequency signal component is highlighted, and then the initial amplitude thereof can be accurately determined. According to the exponential attenuation formula , and a nonlinear attenuation coefficient is calculated in combination with a preset propagation distance. The formula can quantitatively calculate the nonlinear attenuation coefficient by substituting known parameters such as the initial amplitude and the propagation distance, thereby quantitatively evaluating the degree of attenuation of the high-frequency signal during the propagation process. If the calculated nonlinear attenuation coefficient exceeds the preset range, the least squares method is used to optimize the attenuation coefficient to obtain an optimized attenuation coefficient. When the initial calculation result does not meet the expectation or exceeds the reasonable range, the attenuation coefficient is optimized and adjusted through the least squares method, so that the result is more in line with the actual physical law and the expected model requirement, thereby improving the credibility and usability of the data. From the optimized attenuation coefficient, the attenuation characteristics corresponding to different frequency ranges are extracted to generate a frequency band differentiation impact dataset. This step classifies and organizes the attenuation characteristics in different frequency ranges through in-depth mining of the optimized attenuation coefficient, thereby forming a dataset that can intuitively reflect the attenuation differences of signals in each frequency band. Through clustering analysis on the frequency band differentiation impact dataset, the K-means algorithm is used to determine the attenuation mode of each frequency range, and a frequency band differentiation impact dataset containing attenuation coefficients and frequency ranges is generated. The application of clustering analysis in this case enables the attenuation characteristics of different frequency ranges to be systematically summarized and classified, and the K-means algorithm can divide each frequency range into different attenuation mode categories according to the inherent similarity of the data, thereby forming a frequency band differentiation impact dataset that not only contains detailed attenuation coefficient information but also clearly identifies the attenuation mode corresponding to each frequency range.

[0054] The above steps start with the processing of dynamic meteorological data, and through a series of methods such as Fourier transform, exponential decay formula calculation, and least squares optimization, the nonlinear attenuation characteristics of high-frequency signals in different frequency ranges are gradually and deeply mined, and finally a frequency band differentiation impact dataset containing rich information is generated.

[0055] In some embodiments, according to the frequency band differentiation impact dataset, in combination with signal waveform characteristics, lightning types are classified, cloud flash and ground flash characteristic distributions are distinguished, a lightning type recognition dataset and a confidence level are generated, including:

[0056] The peak amplitude and the rise time are separated from the initial feature set to obtain a quantitative feature vector; if the peak amplitude of the quantitative feature vector exceeds a preset range, then a fast Fourier transform is used to analyze the frequency band difference to obtain a frequency domain feature distribution;

[0057] According to the frequency domain feature distribution and the quantitative feature vector, an input dataset for a support vector machine model is constructed to obtain training data; a support vector machine algorithm is used to classify the training data to distinguish cloud flashes and ground flashes, thereby obtaining a classification result; a confidence value is calculated from the classification result through a cross-validation method to obtain a lightning type confidence level;

[0058] According to the classification result and the confidence value, a lightning type classification data set containing cloud flashes and ground flashes is generated, and a final data set is obtained.

[0059] In the above technical solution, the focus is on how to accurately classify lightning types based on the influence of frequency band differentiation on data sets and signal waveform features, distinguish cloud flash and ground flash feature distribution, and generate lightning type recognition data set and confidence. Through multi-step feature extraction, model construction and verification, the accuracy and reliability of lightning type recognition are improved, and key support is provided for lightning disaster warning.

[0060] From the initial feature set, the peak amplitude and rise time are separated to form a quantitative feature vector. If the peak amplitude exceeds the preset range, the frequency band difference is analyzed using fast Fourier transform to obtain the frequency domain feature distribution. By fast Fourier transform, the time domain features are converted into frequency domain features, and the frequency characteristics of lightning signals are fully captured, making the feature vector more distinguishable. According to the frequency domain feature distribution and the quantitative feature vector, the input data set of the support vector machine model is constructed. The support vector machine algorithm is used to classify the training data to distinguish cloud flashes and ground flashes, and the classification result is obtained. The support vector machine algorithm finds the optimal classification hyperplane in a high-dimensional feature space, which can effectively handle complex nonlinear classification problems and ensure the accuracy of the classification result. The confidence value is calculated from the classification result by cross-validation method to obtain the confidence of lightning type. Cross-validation can fully utilize the data set to divide the training set and test set multiple times for model evaluation to obtain stable confidence estimation, enhancing the quantification of the credibility of the classification result. According to the classification result and the confidence value, a lightning type classification data set containing cloud flashes and ground flashes is generated. The final data set not only contains the classification result of lightning type, but also has the corresponding confidence value, providing comprehensive and reliable information for subsequent lightning disaster warning and analysis.

[0061] The above steps start with feature extraction, combine frequency domain analysis and time series analysis, construct a support vector machine model for classification, and evaluate the confidence through cross-validation to finally generate a detailed lightning type recognition data set.

[0062] In some embodiments, if the confidence of the lightning type recognition data set is lower than the confidence threshold, the real-time meteorological transient characteristic data is fused, the support vector machine kernel function and the penalty coefficient are adjusted, and an optimized lightning type recognition model is generated, including:

[0063] If the confidence of the lightning type recognition data set is lower than the confidence threshold, the humidity mutation rate and rainfall intensity mutation value are obtained from the real-time meteorological data to generate a fusion feature data set;

[0064] According to the fusion feature data set, the principal component analysis method is used to reduce the dimension of the humidity mutation rate and rainfall intensity mutation value to obtain a reduced dimension feature set;

[0065] If the feature dimension of the dimensionality reduction feature set meets the preset requirements, the kernel function type and penalty coefficient of the support vector machine are adjusted through the grid search method to obtain the optimized parameter combination;

[0066] According to the optimized parameter combination, the support vector machine algorithm is used to train the reduced dimension feature set to generate the initial lightning type recognition model;

[0067] Obtain new humidity mutation rate and rainfall intensity mutation values ​​from real-time meteorological data, make predictions using the initial lightning type recognition model, and obtain prediction confidence;

[0068] If the prediction confidence is lower than the confidence threshold, the initial lightning type recognition model is iteratively optimized by the gradient boosting method to obtain the optimized lightning type recognition model.

[0069] In this technical solution, when the confidence level of a lightning type recognition dataset falls below a threshold, a method is proposed to optimize the support vector machine model by integrating real-time meteorological transient characteristic data. Through feature extraction, dimensionality reduction, parameter optimization, and model iteration, an optimized lightning type recognition model is generated to improve the accuracy and confidence of lightning type recognition.

[0070] When the confidence of the lightning type identification dataset is lower than the confidence threshold, the humidity mutation rate and the rainfall intensity mutation value are obtained from the real-time meteorological data to generate a fusion feature dataset. The real-time meteorological data can reflect the environmental changes when lightning occurs, and these mutation features are closely related to the lightning type, and after fusion, more rich information can be provided for the model. Principal component analysis method is used to reduce the dimension of the humidity mutation rate and the rainfall intensity mutation value to obtain a reduced dimension feature set. Dimension reduction can remove redundant information and noise, retain main features, improve model training efficiency and performance, and make the model more focused on key features. If the feature dimension of the reduced dimension feature set meets the preset requirements, the kernel function type and the penalty coefficient of the support vector machine are adjusted by the grid search method to obtain the optimal parameter combination. Grid search can search the parameter space comprehensively to find the parameter combination that makes the model have the best performance on the cross-validation set, thereby improving the classification performance of the model. According to the optimal parameter combination, the support vector machine algorithm is used to train the reduced dimension feature set to generate an initial lightning type identification model. Training the model with the optimized parameters can make the model learn the patterns and rules in the data better and improve the performance of the initial model. New humidity mutation rate and rainfall intensity mutation value are obtained from the real-time meteorological data, and the initial lightning type identification model is used for prediction to obtain a prediction confidence. The prediction performance on new data can be used to evaluate the confidence of the model, and the reliability of the model in actual application can be judged. If the prediction confidence is lower than the confidence threshold, the initial lightning type identification model is iteratively optimized by the gradient boosting method to obtain an optimized lightning type identification model. Gradient boosting can gradually correct the errors of the model, improve the prediction ability and confidence of the model, and further improve the accuracy of lightning type identification.

[0071] The above steps effectively solve the problem of low confidence of the lightning type identification dataset by fusing real-time meteorological transient feature data, combining principal component analysis dimension reduction, grid search parameter optimization, and gradient boosting iterative optimization methods, and improve the performance and reliability of the lightning type identification model.

[0072] In some embodiments, according to the lightning type and feature distribution generated by the optimized lightning type identification model, the state of the lightning signal propagation path is estimated in combination with dynamic meteorological data to generate a lightning positioning accuracy dataset, including:

[0073] Real-time signal data of lightning monitoring stations are obtained, and a pre-established lightning type identification model is used to classify signal features to obtain lightning types and feature distributions;

[0074] According to the feature distribution, in combination with dynamic meteorological data, a spatio-temporal distribution matrix of lightning signals is constructed to determine the initial state of the signal propagation path;

[0075] The Kalman filtering algorithm is used to estimate the state of the signal propagation path to obtain a path estimation result; if the path estimation result deviates from the wind field distribution of the dynamic meteorological data by more than a deviation threshold, the path estimation result is corrected to obtain a corrected propagation path;

[0076] The positioning coordinates of the lightning signal are calculated through the corrected propagation path to obtain a position error data set; the signal arrival time difference is analyzed according to the position error data set to generate a time resolution data set;

[0077] The least square method is used to optimize the positioning coordinates for the time resolution data set to obtain a final lightning positioning accuracy data set.

[0078] In the above technical solution, the purpose is to accurately estimate the state of the lightning signal propagation path based on the optimized lightning type recognition model and dynamic meteorological data, and then generate a high-precision lightning positioning accuracy data set. Through multiple steps of signal processing, state estimation, path correction and data optimization, the accuracy and reliability of lightning positioning are ensured, providing key support for lightning disaster warning.

[0079] The real-time signal data of the lightning monitoring station is the basic data source of the entire positioning process. Through the optimized identification model, the signal characteristics are accurately classified, and the type and characteristic distribution of lightning are determined. This provides key prior information for subsequent propagation path estimation. According to the characteristic distribution, combined with dynamic meteorological data, a time-space distribution matrix of lightning signals is constructed to determine the initial state of the signal propagation path. The time-space distribution matrix can comprehensively describe the distribution of lightning signals in time and space. The integration of dynamic meteorological data further considers the influence of meteorological conditions on signal propagation, making the determination of the initial state more close to the actual propagation environment. The Kalman filter algorithm is used to estimate the state of the signal propagation path, and the path estimation result is obtained. If the deviation between the path estimation result and the wind field distribution of the dynamic meteorological data exceeds the deviation threshold, the path estimation result is corrected to obtain the corrected propagation path. The Kalman filter algorithm can update the path state estimation in real time, effectively handling noise and uncertainty in signal propagation. When the estimated result deviates too much from the actual meteorological conditions, timely correction can significantly improve the accuracy of path estimation. Through the corrected propagation path, the positioning coordinates of the lightning signal are calculated to obtain the position error data set. According to the position error data set, the signal arrival time difference is analyzed to generate the time resolution data set. The calculation of the positioning coordinates is a key step in determining the location of the lightning occurrence. The position error data set reflects the accuracy level of the positioning. The time resolution data set further analyzes the characteristics of the signal arrival time difference, providing a basis for subsequent optimization. For the time resolution data set, the least squares method is used to optimize the positioning coordinates to obtain the final lightning positioning accuracy data set. The least squares method optimizes the positioning coordinates by minimizing the sum of squares of errors, which can effectively improve the positioning accuracy. The final lightning positioning accuracy data set not only contains the position information of the lightning, but also reflects the accuracy level of the positioning, providing high-quality data support for lightning disaster warning and analysis.

[0080] The above steps fuse the optimized lightning type identification model and dynamic meteorological data, use Kalman filtering, path correction and least squares optimization techniques, and realize accurate state estimation of the lightning signal propagation path and improvement of lightning positioning accuracy.

[0081] In some embodiments, according to the lightning positioning accuracy data set, the real-time meteorological data and the terrain data are updated, the dynamic parameters of the corrected spectral degradation model are adjusted, and the lightning disaster warning parameter data set is generated, including:

[0082] The humidity and rainfall intensity in the real-time meteorological data and the relief and vegetation in the terrain data are obtained, and are fused into a unified multi-dimensional feature data set.

[0083] The spectrum degradation model is optimized by least squares method to adjust the attenuation coefficient and path loss, and the optimized model parameters are obtained;

[0084] According to the optimized model parameters, the propagation characteristics of lightning signals are calculated, and the preliminary warning parameters of lightning disasters are generated; if the signal intensity in the preliminary warning parameters exceeds the preset threshold, the potential impact range of lightning disasters is determined in combination with humidity, rainfall intensity, and relief and vegetation data;

[0085] The random forest algorithm is used to classify the meteorological and terrain features in the potential impact range, and the intensity level of lightning disasters is judged; through the intensity level and the impact range, the lightning disaster warning parameter data set is generated, including the warning range and the intensity level;

[0086] For the generated warning parameter data set, a spatial interpolation algorithm is applied to optimize the boundary of the warning range, and the final lightning disaster warning data set is obtained.

[0087] In the above technical solutions, starting from the lightning positioning accuracy data set, multi-source data is fused, and through model optimization, characteristic calculation, disaster analysis and spatial interpolation, a precise lightning disaster warning parameter data set is generated, which provides key support for disaster prevention and mitigation.

[0088] Integrate lightning positioning accuracy data set, real-time weather data and terrain data to build a multi-dimensional feature data set. Fuse multi-source data to capture lightning disaster related physical phenomena and provide a comprehensive perspective for subsequent analysis. Use least squares method to optimize spectrum degradation model to accurately adjust attenuation coefficient and path loss and other parameters to improve model performance. Calculate the propagation characteristics based on the optimized model to generate warning parameters. Combine signal intensity with weather and terrain data to determine the potential impact range and consider disaster related factors. Use random forest algorithm to classify features within the potential impact range to determine disaster intensity level. Use spatial interpolation algorithm to optimize warning range boundary to obtain final warning data set. Improve the spatial continuity and accuracy of the warning data set to enhance the practical application value.

[0089] In some embodiments, according to the lightning disaster warning parameter data set, it is transmitted to the warning system to complete the lightning disaster warning in complex mountainous environment, including:

[0090] Obtain lightning disaster warning parameter data set, and obtain structured data set through data cleaning and standardization processing; according to the structured data set, calculate second-level time resolution data by using interpolation algorithm to generate high time resolution data set;

[0091] For high time resolution data set, grid division technology is applied to generate hundred-meter spatial resolution data, and obtain space-time data set; if the lightning activity parameter in the space-time data set exceeds the preset activity threshold, the lightning disaster risk level is judged through the logistic regression algorithm, and the early warning signal data set is generated;

[0092] According to the early warning signal data set, JSON serialization technology is adopted to generate standard JSON format data, and formatted early warning signal is obtained; through data transmission protocol, the formatted early warning signal is transmitted to the early warning system to complete signal distribution;

[0093] For complex mountainous terrain characteristics, terrain correction algorithm is adopted to optimize the early warning signal, and disaster early warning signal suitable for complex mountainous area is generated.

[0094] In the above technical scheme, based on the lightning disaster early warning parameter data set, through data processing, time resolution improvement, spatial resolution refinement, risk judgment, signal format conversion and transmission, and terrain correction, lightning disaster early warning in complex mountainous environment is realized. The whole process closely surrounds the accurate processing and efficient transmission of data, fully considers the influence of complex mountainous terrain on early warning signal, and ensures the timeliness, accuracy and adaptability of early warning information.

[0095] The lightning disaster early warning parameter data set is obtained, and a structured data set is obtained through data cleaning and standardization processing. The data cleaning removes the noise and abnormal values in the data, and the standardization processing unifies the data from different sources to the same scale. According to the structured data set, an interpolation algorithm is used to calculate the second-level time resolution data, and a high time resolution data set is generated. The application of the interpolation algorithm makes the data more detailed in the time dimension, which can more accurately capture the transient changes of lightning activity, and is of great significance for timely discovery and early warning of lightning disasters. For the high time resolution data set, a grid division technology is applied to generate a hundred-meter spatial resolution data, and a space-time data set is obtained. The grid division technology divides the research area into a hundred-meter grid unit, so that the data is also refined in the spatial dimension, which can more accurately locate the position and range of lightning activity. If the lightning activity parameters in the space-time data set exceed the preset activity threshold, a logic regression algorithm is used to judge the lightning disaster risk level, and an early warning signal data set is generated. The logic regression algorithm is used here to establish the quantitative relationship between the lightning activity parameters and the disaster risk level, which can quickly and accurately judge the risk level according to the real-time data, and realize the conversion from data to early warning signal. According to the early warning signal data set, a JSON serialization technology is used to generate standard JSON format data, and a formatted early warning signal is obtained. Through a data transmission protocol, the formatted early warning signal is transmitted to the early warning system to complete the signal distribution. The JSON format has good readability and easy parsing, which is suitable as the standard format of the early warning signal. The data transmission protocol ensures the stability and efficiency of the signal in the transmission process, so that the early warning information can be timely and accurately transmitted to the early warning system, realizing the comprehensive coverage of the complex mountainous area. According to the terrain characteristics of the complex mountainous area, a terrain correction algorithm is used to optimize the early warning signal, and a disaster early warning signal suitable for the complex mountainous area is generated. The terrain correction algorithm fully considers the influence of the mountainous terrain on the propagation of the early warning signal, and through correction, the early warning signal can better adapt to the complex terrain, improving the propagation effect and coverage range of the early warning information in the mountainous area.

[0096] The above steps start from the acquisition and preprocessing of the lightning disaster early warning parameter data set, and realize the lightning disaster early warning in the complex mountainous area through the steps of time resolution improvement, spatial resolution refinement, risk level judgment, signal format conversion and transmission, and terrain correction. BRIEF DESCRIPTION OF DRAWINGS

[0097] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0098] Figure 1 is a flowchart of an embodiment of a lightning disaster early warning method based on meteorological data. DETAILED DESCRIPTION

[0099] The application will be further described below in conjunction with the drawings and embodiments. It is particularly pointed out that the following embodiments are only used to illustrate the application, but do not limit the scope of the application. Similarly, the following embodiments are only part of the embodiments of the application, not all embodiments, and all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of the application.

[0100] The application provides a lightning disaster early warning method based on meteorological data, which is aimed at the business scene problems of complex electromagnetic wave propagation path, difficult lightning type identification and insufficient positioning accuracy caused by the undulating terrain, vegetation coverage and dynamic meteorological conditions in mountainous areas. By fusing the spatial and temporal distribution characteristics of atmospheric humidity, rainfall intensity, temperature in meteorological data and terrain data, a high-precision data set is generated by using weighted superposition and spatial and temporal interpolation. The ray tracing method is used to analyze the multi-path propagation effect of electromagnetic waves and calculate the high and low frequency attenuation coefficients. If the high frequency attenuation exceeds the standard, the path parameters are optimized by genetic algorithm, the lightning type identification is optimized by combining support vector machine and grid search, the path state estimation is performed by fusing Kalman filter, and a high spatial and temporal resolution lightning positioning accuracy data set is generated. The least square method is used to dynamically adjust the spectral degradation model parameters, and finally a lightning disaster early warning signal data set with second-level time resolution and hundred-meter spatial resolution is generated.

[0101] One of the embodiments

[0102] Please refer to Figure 1 A lightning disaster early warning method based on meteorological data, the method comprises:

[0103] S1, fuse meteorological data and terrain data, generate spatial and temporal distribution data set of atmospheric parameters and terrain characteristics in complex mountainous environment by weighted superposition and spatial and temporal interpolation method;

[0104] In this embodiment, S1, fuse meteorological data and terrain data, generate spatial and temporal distribution data set of atmospheric parameters and terrain characteristics in complex mountainous environment by weighted superposition and spatial and temporal interpolation method, including:

[0105] S11, obtain atmospheric humidity, rainfall intensity, temperature in meteorological data and terrain undulation and vegetation coverage in terrain data, generate standardized meteorological and terrain data set by preprocessing;

[0106] S12. Using the weighted superposition method, different weights are assigned to atmospheric humidity, rainfall intensity, and temperature, and a comprehensive meteorological impact factor is generated through linear combination to obtain a comprehensive meteorological data set;

[0107] S13, extracting spatiotemporal distribution features from the comprehensive meteorological dataset and terrain data by spatiotemporal interpolation method to generate a continuous meteorological and terrain spatiotemporal distribution dataset;

[0108] S14. If the rainfall intensity in the comprehensive meteorological dataset exceeds the intensity threshold, Gaussian smoothing is performed on the rainfall intensity to generate a smoothed meteorological spatiotemporal distribution dataset;

[0109] S15. Based on the smoothed meteorological spatiotemporal distribution dataset and the terrain spatiotemporal distribution dataset, a Kriging interpolation method is used to generate a spatiotemporal distribution dataset of atmospheric parameters and terrain characteristics in a complex mountainous environment.

[0110] For example, an initial dataset is first obtained by fusing meteorological data on atmospheric humidity, rainfall intensity, and temperature with terrain relief and vegetation cover from topographic data. Assume that the atmospheric humidity data is hourly relative humidity (%). For example, at 12:00 PM on July 22, 2025, the humidity at a mountainous site is 75%, the rainfall intensity is 5 mm per hour, and the temperature is 25°C. The terrain data uses a digital elevation model (DEM) with a resolution of 30 meters, a point elevation of 1500 meters, a slope of 20 degrees, and a vegetation cover of 0.6 (normalized value). To fuse these data, a weighted overlay method is used, with weights determined based on the parameters' impact on electromagnetic wave propagation: humidity weight 0.4, rainfall intensity weight 0.3, temperature weight 0.2, terrain relief weight 0.2, and vegetation cover weight 0.1. The calculation formula is: Composite value = 0.4 × humidity + 0.3 × rainfall intensity + 0.2 × temperature + 0.2 × terrain relief + 0.1 × vegetation cover. For example, the normalized data for a given point is humidity 0.75, rainfall intensity 0.5 (normalized to 0-1), temperature 0.625 (upper bound of 40°C), terrain relief 0.4 (upper bound of 50-degree slope), and vegetation cover 0.6. The calculated composite value is 0.4 × 0.75 + 0.3 × 0.5 + 0.2 × 0.625 + 0.2 × 0.4 + 0.1 × 0.6 = 0.615. Next, spatiotemporal interpolation is performed to generate a continuously distributed dataset. The inverse distance weighted (IDW) interpolation algorithm is used, with the formula: ,in For known points Parameter value, For distance, For the power (take 2). Assume that the distance between an unknown point and three known points is 100 meters, 200 meters, and 300 meters, and the comprehensive values are 0.615, 0.58, and 0.55, respectively. The interpolation point comprehensive value is (0.615 / 100^2 + 0.58 / 200^2 + 0.55 / 300^2) / (1 / 100^2 + 1 / 200^2 + 1 / 300^2) ≈ 0.602. Time interpolation uses linear interpolation. Assuming that the comprehensive values of 12:00 and 13:00 are 0.602 and 0.620, respectively, the interpolation of 12:30 is 0.602 + (0.620-0.602)×0.5 = 0.611. The final generated data set contains the comprehensive values of each grid point (30 meter resolution), and the time resolution is 30 minutes.

[0111] S2, according to the spatiotemporal distribution data set, analyze the multi-path propagation effect of electromagnetic waves, calculate the attenuation coefficient of high-frequency components greater than 1 gigahertz and the propagation loss of low-frequency components less than 1 gigahertz, and generate a preliminary quantitative data set of spectral degradation;

[0112] In this embodiment, S2, according to the spatiotemporal distribution data set, analyzes the multi-path propagation effect of electromagnetic waves, calculates the attenuation coefficient of high-frequency components greater than 1 gigahertz and the propagation loss of low-frequency components less than 1 gigahertz, and generates a preliminary quantitative data set of spectral degradation, including:

[0113] S21, obtain the spatiotemporal distribution data set, clean the data by using a preprocessing method, eliminate noise points and outliers, and obtain a standardized data set;

[0114] S22, simulate the electromagnetic wave propagation path of the standardized data set by ray tracing method, determine the reflection and refraction and diffraction path in the multi-path effect, and obtain a propagation path set;

[0115] S23, for the propagation path set, calculate the attenuation coefficient of electromagnetic waves with high-frequency components greater than 1 gigahertz, if the number of path reflections is greater than a preset threshold, use geometric optics model to calculate the attenuation, and obtain a high-frequency attenuation coefficient set;

[0116] S24, according to the propagation path set, analyze the propagation loss of electromagnetic waves with low-frequency components less than 1 gigahertz, if the diffraction angle of the path is greater than a preset threshold, use uniform geometric diffraction theory to calculate the loss, and obtain a low-frequency propagation loss set;

[0117] S25, extract the spectral degradation features from the high-frequency attenuation coefficient set and the low-frequency propagation loss set, and use fast Fourier transform algorithm to convert the features to frequency domain to obtain spectral degradation data;

[0118] S26, construct a quantitative data model by combining the spectral degradation data with the spatial and temporal distribution characteristics, fit the spectral degradation trend using the least squares method, and obtain a quantitative data set;

[0119] S27, perform data verification on the quantitative data set, if the data points deviate from the fitted curve by more than a preset threshold, eliminate the abnormal points, and obtain the final spectral degradation quantitative data set.

[0120] For example, based on the spatial and temporal distribution data set analysis of electromagnetic wave multi-path propagation effect, first construct a three-dimensional environment model, assume the scene is a 1000m x 1000m x 100m area, containing 50 randomly distributed buildings, height range 20-80m, material is concrete (dielectric constant εr=7, conductivity σ=0.01 S / m). Using ray tracing method, set the transmitting source at (500m, 500m, 10m), transmitting power is 1W, frequency covers 0.1GHz to 10GHz. The ray tracing algorithm is based on geometric optics, calculates direct, reflected and diffracted paths, the maximum reflection order is 3, and the diffraction uses UTD model. The propagation loss of each path is calculated by the free space loss formula , where is the path length (m), is the frequency (GHz). For high frequency components (>1GHz), the attenuation coefficient is calculated by multi-path superposition, considering reflection loss (each reflection attenuation is about 3-5dB, depending on the incident angle, assuming an average of 4dB) and diffraction loss (based on knife-edge diffraction model, typical value is 6-10dB). Taking 5GHz as an example, assuming that the receiving point (600m, 600m, 5m) has 3 effective paths: the direct path is 141.42m long, the loss is 20log10(141.42)+20log10(5)+32.44=86.98dB; the first reflection path is 150m long, the loss is 88.52dB+4dB=92.52dB; the diffraction path is 160m long, the loss is 90.08dB+8dB=98.08dB. The total received power is synthesized by the RSSI formula, and the attenuation coefficient is about 0.9dB / m. For low frequency components (<1GHz), taking 0.5GHz as an example, the loss is mainly affected by free space propagation and building penetration, and the penetration loss is calculated as 10dB / wall, assuming 2 walls, the loss is 20dB, and the total path loss is 20log10(141.42)+20log10(0.5)+32.44+20=66.98dB. The spectral degradation data set is generated by simulation and stored in CSV format, containing fields such as frequency, path loss, attenuation coefficient, etc.

[0121] S3, if the high-frequency component attenuation coefficient in the preliminary quantification data of spectral degradation exceeds the attenuation threshold, the propagation path parameters are optimized by a genetic algorithm to generate a corrected spectral degradation model;

[0122] In this embodiment, S3, if the high-frequency component attenuation coefficient in the preliminary quantification data of spectral degradation exceeds the attenuation threshold, the propagation path parameters are optimized by a genetic algorithm to generate a corrected spectral degradation model, comprising:

[0123] S31, obtain the high-frequency component from the spectral degradation data set, calculate the attenuation coefficient; if the attenuation coefficient exceeds the attenuation threshold, initialize the propagation path parameters by a genetic algorithm to generate an initial parameter set;

[0124] S32, construct a propagation environment model according to the terrain data and vegetation cover data, adjust the reflection angle and refractive index to obtain an optimized parameter set; iteratively optimize the parameter set by a genetic algorithm, fuse the propagation environment model to generate a candidate path set;

[0125] S33, obtain the optimal propagation path from the candidate path set, calculate the corrected high-frequency component attenuation coefficient to obtain the updated spectral degradation data;

[0126] S34, use the updated spectral degradation data to construct a corrected spectral degradation model and determine the model parameters; verify the high-frequency component attenuation coefficient by the corrected spectral degradation model to judge the model accuracy.

[0127] For example, if the spectral degradation preliminary quantification dataset shows that the attenuation coefficient of high-frequency components exceeds 0.5 dB / km, the propagation path parameters need to be optimized by a genetic algorithm, and the terrain undulation and vegetation coverage data are fused to generate a corrected spectral degradation model. First, assume that high-frequency components (such as 2.4 GHz) are propagating in a certain area, and preliminary quantification data shows that the attenuation coefficient is 0.6 dB / km, which exceeds the threshold of 0.5 dB / km, triggering the optimization process. Collect terrain data, use a digital elevation model to represent terrain undulation at a resolution of 10 meters, and obtain the height variation range (such as 0-200 meters). Vegetation coverage data is obtained by analyzing remote sensing images, quantifying coverage (such as 40% for dense forest areas, and the initial refractive index value is set to 1.0003). The genetic algorithm initializes the population size to 100, and the individual contains the reflection angle (0°-90°) and the refractive index (1.0001-1.0005), and the fitness function is defined as the minimum difference between the attenuation coefficient and the target value of 0.4 dB / km. During the iteration process, the population is updated by a crossover probability of 0.8 and a mutation probability of 0.01, the loss of each propagation path is calculated, the reflection angle is evaluated using the Fresnel equation, and the path loss is adjusted considering the terrain slope (for example, 5°). After 50 iterations, the reflection angle is optimized to 45°, the refractive index is adjusted to 1.0004, and the attenuation coefficient is reduced to 0.45 dB / km. After fusing the terrain and vegetation data, the corrected model simulates the propagation path by a ray tracing algorithm, considers the multipath effect and vegetation absorption loss (for example, 0.1 dB / m), generates a spectral degradation curve, and verifies that the attenuation coefficient is stable at 0.4-0.45 dB / km. The optimized path parameters effectively reduce the attenuation of high-frequency components and meet the design requirements of the communication system. If the terrain complexity increases (for example, the slope exceeds 10°), the mutation probability can be further adjusted to 0.02 to improve the convergence speed.

[0128] S4、According to the corrected spectral degradation model, extract the atmospheric humidity change rate and rainfall intensity change rate from the dynamic meteorological data, calculate the nonlinear attenuation coefficient of high-frequency signals, and generate a frequency band differentiation influence dataset;

[0129] In this embodiment, S4, according to the corrected spectral degradation model, extract the atmospheric humidity change rate and rainfall intensity change rate from the dynamic meteorological data, calculate the nonlinear attenuation coefficient of high-frequency signals, and generate a frequency band differentiation influence dataset, including:

[0130] S41, obtain the atmospheric humidity change rate and rainfall intensity change rate from the dynamic meteorological data, and obtain the time series data of the change rate by using time series analysis method;

[0131] S42, based on the obtained time series data, extract the high-frequency signal component by using Fourier transform method, and determine the initial amplitude of the high-frequency signal;

[0132] S43, according to the exponential decay formula wherein is the amplitude after attenuation, is the initial amplitude, is the attenuation coefficient, is the propagation distance, in combination with the preset propagation distance, the non-linear attenuation coefficient is calculated;

[0133] S44, if the calculated non-linear attenuation coefficient exceeds the preset range, the least squares method is used to optimize the attenuation coefficient to obtain the optimized attenuation coefficient;

[0134] S45, from the optimized attenuation coefficient, the attenuation characteristics corresponding to different frequency ranges are extracted to generate a frequency band differentiation impact data set;

[0135] S46, by clustering analysis on the frequency band differentiation impact data set, the K-means algorithm is used to determine the attenuation mode of each frequency range; according to the determined attenuation mode, a frequency band differentiation impact data set containing attenuation coefficients and frequency ranges is generated.

[0136] For example, based on the corrected spectral degradation model, the atmospheric humidity change rate and the rainfall intensity change rate in the dynamic meteorological data are extracted, and the high-frequency signal non-linear attenuation coefficient is calculated, and the implementation method of generating the frequency band differentiation impact data set is as follows. Assuming that the dynamic meteorological data comes from a real-time meteorological monitoring station in a certain area, the time resolution is 1 hour, and contains atmospheric humidity (%) and rainfall intensity (mm / h). First, through time series analysis, the humidity change rate and the rainfall intensity change rate are extracted from the meteorological data.

[0137] For example, the humidity data sequence is [60, 62, 65, 63]%, the time interval is 1 hour, and the humidity change rate is calculated by using the difference method: (62-60) / 1=2% / h, (65-62) / 1=3% / h, (63-65) / 1=-2% / h, and the change rate sequence [2, 3, -2]% / h is obtained. Similarly, the rainfall intensity sequence [0, 2, 5, 3] mm / h is calculated to obtain the change rate [2, 3, -2] mm / h. Next, based on the exponential decay formula the high-frequency signal non-linear attenuation coefficient is calculated. Assuming that the signal initial amplitude =100, the propagation distance =10km, the frequency band is divided into low frequency (1-5GHz), medium frequency (5-10GHz), and high frequency (10-20GHz). For the humidity change rate 2% / h and the rainfall intensity change rate 2 mm / h, it is assumed that there is a linear relationship between the humidity change rate and the rainfall intensity change rate: = 0.01 x |humidity change rate| + 0.02 x |rainfall intensity change rate|, substituting the numerical values = 0.01 x 2 + 0.02 x 2 = 0.06. Then the amplitude after attenuation = 100e^(-0.06x10) = 54.88. For different frequency bands, adjust the impact factor of, for example, the high frequency band increases by 20%, then = 0.072, = 100e^(-0.072x10) = 48.66. The data set includes frequency bands, values, and attenuation amplitudes: [(1-5GHz, 0.06, 54.88), (5-10GHz, 0.066, 51.61), (10-20GHz, 0.072, 48.66)].

[0138] S5, according to the frequency band differentiation impact data set, combined with the signal waveform characteristics, classify the lightning type, distinguish the cloud flash and the ground flash characteristic distribution, generate the lightning type recognition data set and the confidence;

[0139] In this embodiment, S5, according to the frequency band differentiation impact data set, combined with the signal waveform characteristics, classify the lightning type, distinguish the cloud flash and the ground flash characteristic distribution, generate the type recognition data set and the confidence, including:

[0140] S51, separate the peak amplitude and the rise time from the initial feature set to obtain a quantized feature vector; if the peak amplitude of the quantized feature vector exceeds a preset range, then analyze the frequency band difference using fast Fourier transform to obtain a frequency domain feature distribution;

[0141] S52, according to the frequency domain feature distribution and the quantized feature vector, construct an input data set of a support vector machine model to obtain training data; use a support vector machine algorithm to classify the training data to distinguish cloud flashes and ground flashes, and obtain a classification result; calculate a confidence value from the classification result by a cross-validation method to obtain a lightning type confidence;

[0142] S53, according to the classification result and the confidence value, generate a lightning type classification data set containing cloud flashes and ground flashes to obtain a final data set.

[0143] For example, based on the frequency band differentiation impact dataset, the lightning signal is first preprocessed, the original signal waveform of cloud flash and ground flash is collected, it is assumed that the sampling frequency is 10 MHz, the time window is 1 ms, and the peak amplitude and rise time characteristics are included. The peak amplitude is calculated by detecting the maximum value of the signal, for example, the peak amplitude of the cloud flash signal ranges from 0.5 to 2.0 V, and the ground flash ranges from 1.5 to 5.0 V; the rise time is defined as the time from 10% to 90% of the peak value of the signal, and the typical value of the cloud flash is 2-5 μs, and the ground flash is 0.5-2 μs. Then, the frequency band features are extracted, the fast Fourier transform (FFT) is used to convert the time domain signal to the frequency domain, the energy distribution of the 0-1 MHz low frequency band and the 1-5 MHz high frequency band is analyzed, and the cloud flash low frequency energy ratio is about 70%, and the ground flash high frequency energy ratio is about 60%. In order to enhance the feature discrimination, the ratio of peak amplitude to rise time is calculated as a combined feature, for example, the cloud flash ratio ranges from 0.1 to 0.4 V / μs, and the ground flash ratio ranges from 0.75 to 10 V / μs. Support vector machine (SVM) algorithm is used for classification, radial basis function (RBF) kernel is selected, parameter C is set to 1.0, and γ is set to 0.1, which is optimized by grid search. The training dataset contains 1000 samples (600 cloud flashes, 400 ground flashes), the test set contains 200 samples, 5-fold cross validation is used, and the classification accuracy is 92%. The confidence is output by the decision function of SVM, which is mapped to the interval [0, 1], for example, the confidence of the cloud flash sample is 0.85, and the confidence of the ground flash is 0.95. The recognition dataset is generated, which includes feature vectors (peak amplitude, rise time, frequency band energy ratio, combined feature) and classification labels (cloud flash / ground flash), and the confidence column is attached.

[0144] S6, if the confidence of the lightning type recognition dataset is lower than the confidence threshold, fuse the real-time meteorological transient characteristic data, adjust the kernel function and the penalty coefficient, and generate an optimized lightning type recognition model;

[0145] In this embodiment, S6, if the confidence of the lightning type recognition dataset is lower than the confidence threshold, fuse the real-time meteorological transient characteristic data, adjust the support vector machine kernel function and the penalty coefficient, and generate an optimized lightning type recognition model, comprising:

[0146] S61, if the confidence of the lightning type recognition dataset is lower than the confidence threshold, obtain the humidity mutation rate and the rainfall intensity mutation value from the real-time meteorological data, and generate a fusion feature dataset;

[0147] S62, according to the fusion feature dataset, the humidity mutation rate and the rainfall intensity mutation value are processed by principal component analysis method, and the dimensionality reduction feature set is obtained;

[0148] S63, if the feature dimension of the dimensionality reduction feature set meets the preset requirement, the kernel function type and the penalty coefficient of the support vector machine are adjusted by the grid search method, and the optimized parameter combination is obtained;

[0149] S64, train the reduced feature set using the support vector machine algorithm according to the optimized parameter combination to generate an initial lightning type recognition model;

[0150] S65, obtain new humidity mutation rate and rainfall intensity mutation value from real-time meteorological data, and predict through the initial lightning type recognition model to obtain a prediction confidence;

[0151] S66, if the prediction confidence is lower than the confidence threshold, iteratively optimize the initial lightning type recognition model through the gradient boosting method to obtain an optimized lightning type recognition model.

[0152] For example, after receiving the lightning type recognition data set, the system first evaluates the confidence of each sample. Assuming that the data set contains 1000 lightning event samples, the confidence is calculated by the softmax function and ranges between 0 and 1. If the sample confidence is lower than 0.9, for example, the confidence of a certain sample is 0.85, the data fusion process is triggered. The system extracts the humidity mutation rate and rainfall intensity mutation value from the real-time meteorological database. Assuming that the humidity mutation rate is 2.5% change per minute and the rainfall intensity mutation value is 3.2 mm change per hour. These data are obtained in JSON format through the API interface with a time resolution of 1 minute, ensuring alignment with the lightning event timestamp. The fusion process uses a weighted average method to concatenate the lightning feature vector (including voltage and current features) and the meteorological feature vector (humidity mutation rate and rainfall intensity mutation value) with weights of 0.6 and 0.4, respectively, to form a new feature vector after normalization. Then, the system uses grid search to optimize the support vector machine (SVM) model, with kernel function candidates being linear kernel, RBF kernel and polynomial kernel, and penalty coefficient C being searched in the range of [0.1, 1, 10, 100]. The grid search is based on 5-fold cross-validation, with F1 score as the evaluation index. Assuming that the RBF kernel has the highest F1 score of 0.92 when C=10. Finally, the system generates an optimized SVM model saved as a.pkl file, and makes classification prediction for new lightning events, outputting the type (such as positive flash, negative flash) and confidence. If the F1 score does not meet the expectation (such as lower than 0.9), the system automatically triggers a second grid search, narrows the C range to [5, 15] with a step of 1, and repeats the optimization until the threshold is met.

[0153] S7, according to the lightning type and feature distribution generated by the optimized lightning type recognition model, combined with dynamic meteorological data, the state of the lightning signal propagation path is estimated to generate a lightning positioning accuracy data set;

[0154] In the embodiment, S7, according to the lightning type and feature distribution generated by the optimized lightning type identification model, the state of the lightning signal propagation path is estimated in combination with dynamic meteorological data, and a lightning positioning accuracy data set is generated, including:

[0155] S71, real-time signal data of a lightning monitoring station is acquired, a pre-established lightning type identification model is used to classify signal features, and lightning types and feature distributions are obtained;

[0156] S72, according to the feature distribution, in combination with dynamic meteorological data, a time-space distribution matrix of lightning signals is constructed, and an initial state of the signal propagation path is determined;

[0157] S73, a Kalman filtering algorithm is used to estimate the state of the signal propagation path, and a path estimation result is obtained; if the path estimation result deviates from the wind field distribution of the dynamic meteorological data by more than a deviation threshold, the path estimation result is corrected, and a corrected propagation path is obtained;

[0158] S74, through the corrected propagation path, the positioning coordinates of the lightning signal are calculated, and a position error data set is obtained; according to the position error data set, the signal arrival time difference is analyzed, and a time resolution data set is generated;

[0159] S75, for the time resolution data set, the least square method is used to optimize the positioning coordinates, and the final lightning positioning accuracy data set is obtained.

[0160] For example, based on the optimized lightning type identification model, first, the lightning electromagnetic signal is processed by a convolutional neural network (CNN), and 1000 groups of lightning signal samples are input, each group of signal including time series (sampling rate 10 MHz, duration 1 ms) and spectral features (0-500 kHz). The model uses ReLU activation function and Dropout (0.3) regularization, and after training, the accuracy rate of identifying intra-cloud lightning, cloud-to-ground lightning and positive polarity lightning reaches 95%. Feature distribution analysis shows that the peak current of cloud-to-ground lightning is 30 kA on average, and the spectral energy is concentrated in 50-100 kHz. Combined with dynamic meteorological data such as wind speed (5 m / s), humidity (80%) and temperature (25°C), a correlation model between lightning type and meteorological parameters is established by multi-dimensional linear regression, with a correlation coefficient R² of 0.85, indicating that humidity has a significant impact on lightning type. Then, the Kalman filter algorithm is used to estimate the state of the lightning signal propagation path, with an initial position error of 100 m, a velocity vector of 0.1 m / s, a process noise covariance Q of 0.01, and an observation noise covariance R of 0.1. After 50 iterations, the position state estimation converges, and the root mean square error (RMSE) of the path trajectory is 15 m. Finally, a lightning positioning accuracy dataset is generated, containing 1000 groups of data, each group including longitude and latitude (resolution 0.001°), height (error ±10 m) and timestamp (resolution 0.1 ms). Analysis shows that the time resolution is limited by the signal sampling rate, and the positioning error is related to the meteorological conditions. The higher the humidity, the error increases slightly (about 5%). Through data set statistics, 90% of the positioning errors are less than 20 m, and the time resolution is 0.12 ms, meeting the real-time monitoring requirements. To ensure logical rigor, after the positioning data and meteorological data are fused, the stability of the path estimation is further verified by support vector machine (SVM), with a classification accuracy of 92%, indicating that the Kalman filter result is reliable.

[0161] S8, according to the lightning positioning accuracy dataset, update real-time meteorological data and terrain data, adjust the dynamic parameters of the corrected spectral degradation model, generate lightning disaster warning parameter dataset; according to the lightning disaster warning parameter dataset, transmit to the warning system, complete the lightning disaster warning in complex mountainous environment.

[0162] In this embodiment, S81, according to the lightning positioning accuracy dataset, update real-time meteorological data and terrain data, adjust the dynamic parameters of the corrected spectral degradation model, generate disaster warning parameter dataset, including:

[0163] S811, obtain the lightning positioning accuracy dataset, humidity in real-time meteorological data, rainfall intensity and terrain data, and fuse them into a unified multi-dimensional feature dataset;

[0164] S812, optimize the spectral degradation model by least squares method, adjust the attenuation coefficient and path loss, get the optimized model parameters;

[0165] S813, calculate the propagation characteristics of lightning signals according to the optimized model parameters, generate preliminary warning parameters of lightning disaster; if the signal intensity in the preliminary warning parameters exceeds the preset threshold, combine humidity, rainfall intensity and relief and vegetation data to determine the potential impact range of lightning disaster;

[0166] S814, classify the meteorological and terrain features in the potential impact range by using the random forest algorithm, and judge the intensity level of lightning disaster; generate lightning disaster warning parameter data set including warning range and intensity level through intensity level and impact range;

[0167] S815, for the generated warning parameter data set, apply spatial interpolation algorithm to optimize the boundary of the warning range, and get the final lightning disaster warning data set.

[0168] For example, based on the lightning positioning accuracy data set, first obtain the lightning occurrence position through the high-precision lightning positioning system, assuming that the positioning accuracy is 100 meters, the data includes latitude and longitude coordinates (such as north latitude 39.9°, east longitude 116.4°) and time stamp. The humidity data is extracted from the real-time meteorological database, assuming that the humidity in a certain area is 75%, the humidity is gridded by interpolation algorithm (linear interpolation), the grid resolution is 1 kilometer, the calculation formula is , where , is the humidity value of the adjacent site. The rainfall intensity is obtained by Doppler radar, assuming that the rainfall intensity at a certain point is 10 mm / h, the continuous rainfall intensity field is generated by using Kriging interpolation method, the formula is , the weight is calculated by the variation function. The terrain relief is extracted from the digital elevation model (DEM), assuming that the maximum relief in a certain area is 200 meters, the vegetation coverage is calculated by remote sensing NDVI data, assuming that it is 0.6, and the normalized data is used for model input. Adjust the spectral degradation model based on the least squares method, the initial value of the attenuation coefficient α is 0.02, the path loss is 80 decibels, the model is , where is the distance, is the reference loss, is the path loss index (set to 2.5). Minimize the error by least squares method, update α to 0.025 and L to 82 decibels. Generate lightning disaster warning parameters, the warning range is extended 5 kilometers radius based on the lightning positioning point, the intensity level is calculated according to the rainfall intensity and the attenuation coefficient, the formula is , assuming that I=5.5, it is divided into medium warning.

[0169] In the present embodiment, S82, according to the lightning disaster warning parameter data set, transmits to the warning system, completes the lightning disaster warning in complex mountainous environment, including:

[0170] S821, obtain the lightning disaster warning parameter data set, obtain the structured data set through data cleaning and standardization processing; according to the structured data set, adopt the interpolation algorithm to calculate the second-level time resolution data, generate the high time resolution data set;

[0171] S822, for the high time resolution data set, apply the grid division technology, generate the hundred-meter spatial resolution data, obtain the space-time data set; if the lightning activity parameter in the space-time data set exceeds the preset activity threshold, judge the lightning disaster risk grade through the logistic regression algorithm, generate the warning signal data set;

[0172] S823, according to the warning signal data set, adopt the JSON serialization technology, generate the standard JSON format data, obtain the formatted warning signal; through the data transmission protocol, transmit the formatted warning signal to the warning system, complete the signal distribution;

[0173] S824, for the complex mountainous terrain characteristics, adopt the terrain correction algorithm to optimize the warning signal, generate the disaster warning signal suitable for complex mountainous area.

[0174] For example, based on the lightning disaster warning parameter data set, a warning signal data set with second-level time resolution and hundred-meter spatial resolution is generated, which needs to realize complex mountainous environment lightning disaster warning through multi-step processing and transmit in standard JSON format. First, assuming that the input data set contains lightning occurrence time (accurate to milliseconds, such as 2025-07-22 18:39:45.123), latitude and longitude coordinates (such as 114.1234°E, 22.5678°N), lightning intensity (in current peak kA, such as 30.5 kA) and terrain height (such as 500.3 meters). To achieve second-level time resolution, a time interpolation algorithm is used to aggregate millisecond-level data to second-level. The method is to take the average value of all lightning events in each second, for example, the time stamps of 3 lightning events in a second are 18:39:45.123, 18:39:45.456 and 18:39:45.789, and the average time is 18:39:45.456, and the intensity is the average value (if they are 30.5 kA, 32.1 kA and 31.2 kA respectively, then the average value is 31.27 kA). The spatial resolution processing adopts the gridding method, which divides the study area into 100m x 100m grids, and distributes the lightning events to the corresponding grid based on latitude and longitude. If there is no event in the grid, the intensity is marked as 0, and if there are multiple events, the maximum value is taken, for example, the lightning intensity of a certain grid is 31.27 kA and 33.4 kA, and the record is 33.4 kA. To adapt to the complex mountainous environment, a terrain correction factor is introduced, and the slope is calculated based on the digital elevation model (DEM) (for example, the slope arctan(Δh / Δd), Δh is the height difference of 50 meters, Δd is the horizontal distance of 100 meters, the slope is 26.57°), and the lightning risk coefficient of the grid with slope greater than 15° is increased by 20%, for example, 33.4 kA is adjusted to 40.08 kA. The warning signal generation adopts threshold method, and sets the lightning intensity greater than 25 kA and the terrain adjusted greater than 30 kA as high risk, generates warning level (high / medium / low), such as 40.08 kA corresponding to high risk. The final data set is packaged in JSON format, for example, {“timestamp”:“2025-07-22 18:39:45”,“grid_id”:“114.1234_22.5678”,“intensity”:40.08,“risk_level”:“high”}, which is transmitted to the warning system through API interface, and the system parses the JSON to trigger regional alarm.

[0175] The above only describes some embodiments of the present application, and does not limit the protection scope of the present application, and any equivalent device or equivalent process transformation using the content of the specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A lightning disaster early warning method based on meteorological data, characterized in that: The method comprises: By integrating meteorological data with topographic data and using weighted overlay and spatiotemporal interpolation methods, we can generate a spatiotemporal distribution dataset of atmospheric parameters and topographic characteristics in complex mountainous environments. Based on the spatiotemporal distribution dataset, the multipath propagation effects of electromagnetic waves were analyzed, the attenuation coefficients of high-frequency components greater than 1 GHz and the propagation losses of low-frequency components less than 1 GHz were calculated, and a preliminary quantitative dataset of spectral degradation was generated. If the attenuation coefficient of the high-frequency component in the preliminary quantified data set of spectrum degradation exceeds the attenuation threshold, the propagation path parameters are optimized by genetic algorithm to generate a modified spectrum degradation model; Based on the modified spectrum degradation model, the atmospheric humidity change rate and rainfall intensity change rate in the dynamic meteorological data are extracted, the nonlinear attenuation coefficient of the high-frequency signal is calculated, and a frequency band differential impact data set is generated; Based on the frequency band differential impact dataset and signal waveform characteristics, lightning types are classified, the characteristic distribution of cloud-to-ground lightning is distinguished, and a lightning type identification dataset and confidence level are generated. If the confidence level of the lightning type recognition dataset is lower than the confidence threshold, the real-time meteorological transient characteristic data is integrated, the support vector machine kernel function and penalty coefficient are adjusted, and an optimized lightning type recognition model is generated; Based on the lightning type and characteristic distribution generated by the optimized lightning type identification model and combined with dynamic meteorological data, the state of the lightning signal propagation path is estimated to generate a lightning location accuracy dataset; Based on the lightning location accuracy dataset, real-time meteorological data and terrain data are updated, the dynamic parameters of the corrected spectrum degradation model are adjusted, and a lightning disaster warning parameter dataset is generated. Based on the lightning disaster warning parameter dataset, it is transmitted to the warning system to complete lightning disaster warning in complex mountainous environments.

2. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: By integrating meteorological data with terrain data and using weighted overlay and spatiotemporal interpolation methods, we can generate a spatiotemporal distribution dataset of atmospheric parameters and terrain characteristics in complex mountainous environments, including: Obtain atmospheric humidity, rainfall intensity, temperature from meteorological data, and terrain relief and vegetation cover from terrain data, and generate standardized meteorological and terrain datasets through preprocessing; A weighted superposition method is used to assign different weights to atmospheric humidity, rainfall intensity, and temperature, and a comprehensive meteorological impact factor is generated through linear combination to obtain a comprehensive meteorological dataset. By using the spatiotemporal interpolation method, the spatiotemporal distribution characteristics are extracted from the comprehensive meteorological dataset and terrain data to generate a continuous meteorological and terrain spatiotemporal distribution dataset; If the rainfall intensity in the comprehensive meteorological dataset exceeds the intensity threshold, the rainfall intensity is Gaussian smoothed to generate a smoothed meteorological spatiotemporal distribution dataset; Based on the smoothed meteorological and terrain spatiotemporal distribution datasets, the Kriging interpolation method is used to generate the spatiotemporal distribution datasets of atmospheric parameters and terrain characteristics in complex mountainous environments.

3. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: Based on the spatiotemporal distribution dataset, the multipath propagation effects of electromagnetic waves are analyzed, the attenuation coefficients of high-frequency components greater than 1 GHz and the propagation losses of low-frequency components less than 1 GHz are calculated, and a preliminary quantitative dataset of spectrum degradation is generated, including: Obtain a spatiotemporal distribution data set, use preprocessing methods to clean the data, remove noise points and outliers, and obtain a standardized data set; The electromagnetic wave propagation path is simulated on the standardized data set by ray tracing method, and the reflection, refraction and diffraction paths in the multipath effect are determined to obtain the propagation path set; For a set of propagation paths, the attenuation coefficient of electromagnetic waves with high-frequency components greater than 1 GHz is calculated. If the number of path reflections is greater than a preset threshold, the attenuation is calculated using a geometric optics model to obtain a set of high-frequency attenuation coefficients. Based on the propagation path set, the electromagnetic wave propagation loss of low-frequency components less than 1 GHz is analyzed. If the path diffraction angle is greater than a preset threshold, the uniform geometric diffraction theory is used to calculate the loss to obtain the low-frequency propagation loss set; Extract spectrum degradation features from the set of high-frequency attenuation coefficients and the set of low-frequency propagation losses, and use the fast Fourier transform algorithm to convert the features into the frequency domain to obtain spectrum degradation data; By combining spectrum degradation data with spatiotemporal distribution characteristics, a quantitative data model is constructed, and the spectrum degradation trend is fitted using the least squares method to obtain a quantitative data set. For the quantized data set, data verification is performed. If the data point deviates from the fitting curve by more than a preset threshold, the outlier is removed to obtain the final spectrum degradation quantized data set.

4. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: If the attenuation coefficient of the high-frequency component in the preliminary quantified spectral degradation data set exceeds the attenuation threshold, the propagation path parameters are optimized using a genetic algorithm to generate a modified spectral degradation model, including: Obtain high-frequency components from the spectrum degradation dataset and calculate the attenuation coefficient; if the attenuation coefficient exceeds the attenuation threshold, initialize the propagation path parameters through a genetic algorithm to generate an initial parameter set; Based on terrain data and vegetation coverage data, a propagation environment model is constructed, and the reflection angle and refractive index are adjusted to obtain an optimized parameter set. The parameter set is iteratively optimized using a genetic algorithm, and the propagation environment model is integrated to generate a set of candidate paths. Obtain the optimal propagation path from the candidate path set, calculate the corrected high-frequency component attenuation coefficient, and obtain updated spectrum degradation data; The updated spectrum degradation data is used to construct a revised spectrum degradation model and determine the model parameters. The high-frequency component attenuation coefficient is verified through the revised spectrum degradation model to determine the model accuracy.

5. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: Based on the modified spectrum degradation model, the atmospheric humidity change rate and rainfall intensity change rate in the dynamic meteorological data are extracted, the nonlinear attenuation coefficient of the high-frequency signal is calculated, and a frequency band differential impact data set is generated, including: Obtain the atmospheric humidity change rate and rainfall intensity change rate from dynamic meteorological data, and use time series analysis method to obtain the time series data of the change rate; Based on the acquired time series data, the Fourier transform method is used to extract the high-frequency signal component and determine the initial amplitude of the high-frequency signal; According to the exponential decay formula ,in is the amplitude after attenuation, is the initial amplitude, is the attenuation coefficient, For the propagation distance, the nonlinear attenuation coefficient is calculated in combination with the preset propagation distance; If the calculated nonlinear attenuation coefficient exceeds the preset range, the attenuation coefficient is optimized using the least square method to obtain an optimized attenuation coefficient; From the optimized attenuation coefficient, the attenuation characteristics corresponding to different frequency ranges are extracted to generate a frequency band differential impact data set; By performing cluster analysis on the frequency band differential impact data set and adopting the K-means algorithm, the attenuation pattern of each frequency range is determined; according to the determined attenuation pattern, a frequency band differential impact data set containing the attenuation coefficient and frequency range is generated.

6. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: Based on the frequency band differential impact dataset and signal waveform characteristics, lightning types are classified, cloud-to-ground lightning distribution characteristics are distinguished, and a lightning type identification dataset and confidence level are generated, including: The peak amplitude and rise time are separated from the initial feature set to obtain a quantized feature vector. If the peak amplitude of the quantized feature vector exceeds a preset range, the frequency band difference is analyzed using a fast Fourier transform to obtain the frequency domain feature distribution. Based on the frequency domain feature distribution and quantized feature vectors, the input data set of the support vector machine model was constructed to obtain training data. The support vector machine algorithm was used to classify the training data to distinguish between cloud-to-ground lightning and lightning flashes, and the classification results were obtained. The confidence value was calculated from the classification results through cross-validation to obtain the confidence level of the lightning type. According to the classification results and confidence values, a lightning type classification dataset containing cloud-to-ground lightning is generated to obtain the final dataset.

7. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: If the confidence level of the lightning type identification dataset is lower than the confidence threshold, the real-time meteorological transient characteristic data is integrated, the support vector machine kernel function and penalty coefficient are adjusted, and an optimized lightning type identification model is generated, including: If the confidence level of the lightning type identification dataset is lower than the confidence threshold, the humidity mutation rate and rainfall intensity mutation values ​​are obtained from the real-time meteorological data to generate a fusion feature dataset; According to the fusion feature data set, the principal component analysis method is used to reduce the dimension of the humidity mutation rate and rainfall intensity mutation value to obtain the reduced dimension feature set; If the feature dimension of the dimensionality reduction feature set meets the preset requirements, the kernel function type and penalty coefficient of the support vector machine are adjusted through the grid search method to obtain the optimized parameter combination; According to the optimized parameter combination, the support vector machine algorithm is used to train the reduced dimension feature set to generate the initial lightning type recognition model; Obtain new humidity mutation rate and rainfall intensity mutation values ​​from real-time meteorological data, make predictions using the initial lightning type recognition model, and obtain prediction confidence; If the prediction confidence is lower than the confidence threshold, the initial lightning type recognition model is iteratively optimized by the gradient boosting method to obtain the optimized lightning type recognition model.

8. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: Based on the lightning types and characteristic distribution generated by the optimized lightning type identification model and combined with dynamic meteorological data, the state of the lightning signal propagation path is estimated to generate a lightning location accuracy dataset, including: Acquire real-time signal data from lightning monitoring stations, use a pre-established lightning type recognition model to classify signal features, and obtain lightning types and characteristic distributions; Based on the characteristic distribution and combined with dynamic meteorological data, the spatiotemporal distribution matrix of lightning signals is constructed to determine the initial state of the signal propagation path; The Kalman filter algorithm is used to perform state estimation on the signal propagation path to obtain a path estimation result. If the deviation between the path estimation result and the wind field distribution of the dynamic meteorological data exceeds the deviation threshold, the path estimation result is corrected to obtain a corrected propagation path. The location coordinates of the lightning signal are calculated using the corrected propagation path to obtain a position error dataset. Based on the position error dataset, the signal arrival time difference is analyzed to generate a time resolution dataset. For the time resolution dataset, the least squares method is used to optimize the positioning coordinates to obtain the final lightning location accuracy dataset.

9. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: Based on the lightning location accuracy dataset, real-time meteorological and terrain data are updated, and the dynamic parameters of the corrected spectrum degradation model are adjusted to generate a lightning disaster warning parameter dataset, including: Obtain lightning location accuracy datasets, humidity and rainfall intensity from real-time meteorological data, and undulation and vegetation from terrain data, and fuse them into a unified multidimensional feature dataset; The spectrum degradation model is optimized by the least square method, the attenuation coefficient and path loss are adjusted, and the optimized model parameters are obtained; Based on the optimized model parameters, the propagation characteristics of lightning signals are calculated to generate preliminary warning parameters for lightning disasters. If the signal strength in the preliminary warning parameters exceeds a preset threshold, the potential impact range of the lightning disaster is determined by combining humidity, rainfall intensity and undulation, and vegetation data. A random forest algorithm is used to classify meteorological and topographical features within the potential impact range to determine the intensity level of lightning disasters. Based on the intensity level and impact range, a lightning disaster warning parameter dataset is generated, including the warning range and intensity level. For the generated warning parameter dataset, the spatial interpolation algorithm is applied to optimize the boundary of the warning range and obtain the final lightning disaster warning dataset.

10. The lightning disaster early warning method based on meteorological data according to claim 1, characterized in that: Based on the lightning disaster warning parameter data set, it is transmitted to the warning system to complete the lightning disaster warning in complex mountainous environments, including: Obtain a lightning disaster warning parameter dataset, and obtain a structured dataset through data cleaning and standardization. Based on the structured dataset, use an interpolation algorithm to calculate second-level time resolution data to generate a high-time resolution dataset. For high-temporal-resolution datasets, gridding technology is applied to generate data with a spatial resolution of 100 meters, resulting in a spatiotemporal dataset. If the lightning activity parameters in the spatiotemporal dataset exceed the preset activity threshold, a logistic regression algorithm is used to determine the lightning disaster risk level and generate an early warning signal dataset. Based on the early warning signal dataset, JSON serialization technology is used to generate standard JSON format data to obtain formatted early warning signals. The formatted early warning signals are transmitted to the early warning system through the data transmission protocol to complete signal distribution. In view of the complex mountainous terrain characteristics, a terrain correction algorithm is used to optimize the warning signal to generate a disaster warning signal suitable for complex mountainous areas.

Images