Lightning approaching trend forecasting method and system

Abstract

The invention discloses a lightning approaching trend forecasting method and system. The method comprises the steps of obtaining cloud cover space distribution and a time sequence, extracting a first cloud cover texture feature vector, and generating a cloud cover hierarchical feature matrix; performing dimensionality reduction on the first cloud cover texture feature vector to generate a second cloud cover texture feature vector, and generating a regionalized cloud cover distribution feature set through clustering in combination with the cloud cover hierarchical feature matrix; extracting humidity, temperature and wind field data to generate a lightning trigger factor data set, and analyzing a regionalized cloud cover distribution feature set by adopting a time sequence analysis method and an optical flow method to form a cloud cover dynamic feature set; and according to the cloud cover dynamic feature set and the cloud cover hierarchical feature matrix, in combination with a pre-established lightning association cloud type label, generating a lightning intensity grade classification feature set and the like. According to the invention, through deep fusion and dynamic optimization of spatial-temporal characteristics, the precision and timeliness of lightning prediction are significantly improved, and efficient technical support is provided for meteorological disaster early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of lightning imminent trend forecasting, and in particular to a lightning imminent trend forecasting method and system. Background Art

[0002] Lightning approach trend forecasting is an important research direction in the field of meteorological forecasting. It is of key significance for ensuring the safe operation of industries such as aviation, electricity, and transportation. Accurately predicting lightning activity can effectively reduce disaster losses and improve social and economic benefits. However, current lightning forecasting methods have significant limitations when dealing with complex meteorological scenarios. Existing technologies often have difficulty adapting to the diversity of cloud distribution in different regions, especially when dealing with high-humidity cloud clusters on the coast and sparse cloud systems inland. The lack of accurate capture of dynamic changes in cloud cover leads to insufficient timeliness and reliability of forecast results. In addition, existing methods have bottlenecks in the efficiency of processing massive meteorological data, making it difficult to quickly extract features directly related to lightning activity, limiting the level of refinement of forecasts.

[0003] The core challenge lies in extracting key dynamic features from the complex and ever-changing cloud distribution and achieving accurate predictions of lightning activity trends. First, the spatiotemporal variations in cloud distribution are complex. Clouds in coastal areas are dense and rapidly evolving, while inland clouds are sparse and unevenly distributed. This difference makes a single feature extraction method difficult to apply. For example, in coastal areas, a lightning event may occur due to the rapid accumulation of high-humidity clouds, but existing systems struggle to identify this rapidly changing pattern from massive cloud data in a short period of time. Second, due to the complexity of cloud dynamic characteristics, existing methods lack universality across different meteorological conditions, making it difficult to accurately distinguish the triggering conditions for lightning activity through a unified framework. This leads to frequent classification biases or inaccurate prediction time windows when forecasting systems face regional differences. For example, a coastal city may miss the optimal warning opportunity due to rapidly moving clouds, while inland areas may experience frequent false alarms due to sparse cloud systems.

[0004] Therefore, how to efficiently analyze the spatiotemporal characteristics of cloud cover changes and construct adaptive feature extraction and classification methods for multiple cloud scenarios has become a key issue in lightning trend forecasting. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to propose a method for forecasting the approaching trend of lightning, which can significantly improve the accuracy and timeliness of lightning prediction through deep fusion and dynamic optimization of spatiotemporal features, and provide efficient technical support for meteorological disaster warning.

[0006] According to one aspect of the present invention, a method for predicting an approaching lightning trend is provided, the method comprising: Obtain the cloud amount spatial distribution and time series, extract the first cloud amount texture feature vector, and generate a cloud amount layer feature matrix; reduce the dimension of the first cloud amount texture feature vector to generate a second cloud amount texture feature vector, and combine it with the cloud amount layer feature matrix to generate a regionalized cloud amount distribution feature set through clustering; Humidity, temperature, and wind field data were extracted to generate a lightning trigger factor dataset. Time series analysis and optical flow methods were used to analyze the regional cloud distribution feature set to form a cloud dynamic feature set. Based on the cloud cover dynamic feature set and cloud cover layer feature matrix, combined with the pre-established lightning-related cloud type labels, a lightning intensity level classification feature set is generated; Based on the lightning intensity level classification feature set and the cloud cover stratification feature matrix, the coupling weight coefficient of cloud cover vertical stratification and lightning intensity classification is calculated to generate the cloud cover-lightning vertical stratification coupling feature set; Based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, the time series analysis method is used to generate the lightning occurrence probability distribution and lightning spatial location coordinate sequence, forming a lightning probability distribution feature set; Based on the lightning probability distribution feature set and cloud periodicity feature vector, the time series analysis method is used to generate the lightning occurrence probability time series and the lightning time window range series, forming the lightning time window prediction feature set; Based on the cloud cover periodicity feature vector and the lightning time window range sequence, the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window is calculated to generate the cloud cover-lightning time rhythm coupling feature set. Based on the lightning time window prediction feature set and the lightning trigger factor dataset, the time series analysis method is used to generate the lightning movement path sequence and lightning duration sequence to form the lightning trend prediction feature set; Based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning temporal rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, spatial and temporal features are extracted, and a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends is generated through weighted fusion. By performing cluster analysis on the spatiotemporal coupling feature set, the dynamic evolution pattern of the lightning trend is obtained; based on the dynamic evolution pattern, the lightning trend is predicted using the time series analysis method to obtain the final lightning trend prediction result.

[0007] In the above technical solution, the spatial distribution and time series of cloud cover are first obtained. A first cloud texture feature vector is then extracted from this vector, and a cloud layer feature matrix is ​​generated. This process fully exploits the key features of cloud cover at different spatial and temporal levels. Dimensionality reduction is then performed on the first cloud texture feature vector to obtain a second cloud texture feature vector. This second cloud texture feature vector, combined with the cloud layer feature matrix, is then clustered to generate a regionalized cloud distribution feature set. Dimensionality reduction not only reduces data dimensionality and computational complexity, but also removes redundant information, retains core features, and enables regionalization of cloud distribution. Data on key meteorological factors such as humidity, temperature, and wind are extracted to generate a lightning trigger factor dataset. These meteorological factors are closely related to lightning formation, and their changes play a role in inducing lightning. The regionalized cloud distribution feature set is analyzed using time series analysis and optical flow methods to form a cloud dynamic feature set. The time series analysis method captures the temporal trends of cloud cover, while the optical flow method reflects the motion and evolution of clouds. Combining these two methods comprehensively characterizes the dynamic characteristics of cloud cover. Based on the cloud cover dynamic feature set and the cloud cover stratification feature matrix, combined with pre-established lightning-related cloud type labels, a lightning intensity classification feature set is generated. By associating cloud type labels, cloud cover characteristics are directly linked to lightning intensity levels, enabling lightning intensity classification and prediction. The coupling weight coefficients between cloud cover vertical stratification and lightning intensity classification are calculated to generate a cloud cover-lightning vertical stratification coupled feature set. This process fully considers the interaction between the vertical distribution of clouds and lightning intensity, providing a basis for more accurate lightning prediction. Combined with lightning monitoring data, a time series analysis method is used to generate a lightning occurrence probability distribution and a lightning spatial location coordinate sequence, forming a lightning probability distribution feature set. By analyzing historical lightning monitoring data, the spatial and temporal distribution patterns of lightning occurrence probability can be determined. Based on the lightning probability distribution feature set and the cloud cover periodicity feature vector, a time series analysis method is used to generate a lightning occurrence probability time series and a lightning time window range series, forming a lightning time window prediction feature set. Periodic changes in cloud cover often correlate with the temporal patterns of lightning activity. By analyzing the cloud cover periodicity feature vector, it is possible to predict the time window in which lightning is likely to occur. The degree of rhythmic matching between the periodic fluctuations in cloud cover and the lightning time window is calculated to generate a cloud cover-lightning temporal rhythm coupling feature set. This further reveals the inherent rhythmic relationship between cloud cover changes and the temporal distribution of lightning, providing more comprehensive temporal information for lightning trend forecasting. Based on the lightning time window prediction feature set and a dataset of lightning triggering factors, a time series analysis method is used to generate lightning movement path sequences and lightning duration sequences, forming a lightning trend prediction feature set. This enables the prediction of lightning movement paths and durations in space.A comprehensive analysis of lightning trend prediction feature sets, cloud cover-lightning vertical layer coupling feature sets, and cloud cover-lightning temporal rhythm coupling feature sets was conducted. Combined with lightning-associated cloud type labels, spatial and temporal features were extracted. A weighted fusion was then used to generate a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends. This process organically combines the spatiotemporal characteristics of lightning to form a comprehensive feature set that comprehensively reflects the spatiotemporal evolution of lightning trends. Cluster analysis of the spatiotemporal coupling feature set revealed the dynamic evolution patterns of lightning trends. Different clustering results represent different lightning trend evolution paths, and these patterns reflect the changing characteristics of lightning activity under different spatiotemporal conditions. Based on the dynamic evolution patterns, a time series analysis method was used to predict lightning trends, resulting in the final lightning trend forecast, providing a scientific basis for lightning disaster prevention and mitigation decision-making.

[0008] This lightning approach trend forecasting method utilizes multi-dimensional feature extraction, analysis, and fusion, leveraging multi-source data such as cloud cover and meteorological elements to construct a comprehensive lightning prediction system. This method can provide meteorological departments with more accurate and timely lightning approach trend forecasts, helping them to implement appropriate lightning protection measures in advance and reduce the loss of life and property caused by lightning disasters. It has broad application prospects and significant practical value.

[0009] In some embodiments, humidity, temperature, and wind field data are extracted to generate a lightning trigger factor dataset, and a regionalized cloud distribution feature set is analyzed using a time series analysis method and an optical flow method to form a cloud dynamic feature set, including: Humidity, temperature, and wind field data are obtained from multi-source meteorological data, and a lightning trigger factor dataset is generated through data cleaning and standardization. The long short-term memory network is used to analyze the time series in the regional cloud distribution feature set, extract the temporal variation characteristics of cloud distribution, and obtain the cloud time series feature set. The cloud time series feature set is analyzed by optical flow method, the periodic change and movement speed of cloud distribution are calculated, and the cloud periodic feature vector and cloud movement speed sequence are generated. Based on the cloud cover periodic characteristic vector and cloud cover movement speed sequence, humidity, temperature and wind field data are integrated to construct a cloud cover dynamic evolution feature set.

[0010] In the aforementioned technical solution, accurately capturing lightning triggering conditions and cloud cover dynamics is key to improving forecast accuracy in the field of lightning approach trend forecasting. The aforementioned steps focus on a method for generating a cloud cover dynamic feature set based on humidity, temperature, and wind field data, combining time series analysis with optical flow methods.

[0011] Humidity, temperature, and wind data are obtained from multiple meteorological sources, including ground-based meteorological station observations, satellite remote sensing, and radar detection. The raw data is cleaned to remove erroneous, outlier, and missing values. Then, through normalization, data with different dimensions and ranges are converted to a unified scale to generate a lightning trigger factor dataset. A long short-term memory (LSTM) network is used to analyze the time series within the regionalized cloud distribution feature set. An LSTM is a special recurrent neural network that effectively handles long-term dependencies in time series data. Using a structure consisting of memory cells, input gates, forget gates, and output gates, it learns the cloud distribution time series, extracting the temporal variation characteristics of the cloud distribution. This generates a cloud time series feature set that captures the evolution of cloud cover at different time scales. The cloud time series feature set is analyzed using the optical flow method. This method calculates the periodic variation and movement speed of cloud distribution based on the motion information of pixels in an image sequence. First, the cloud time series feature set is treated as an image sequence, with the cloud distribution feature at each time step as an image. Then, an optical flow algorithm is used to estimate the motion vectors of pixels between adjacent images, thereby determining the direction and speed of cloud cover movement. Furthermore, the periodic variations in cloud cover distribution are analyzed to generate cloud cover periodicity feature vectors and cloud cover movement speed sequences, providing key information for understanding the dynamic evolution of cloud cover. Based on these cloud cover periodicity feature vectors and cloud cover movement speed sequences, humidity, temperature, and wind field data are integrated to construct a cloud cover dynamic evolution feature set. Humidity, temperature, and wind field, as important meteorological factors, are closely related to the formation, development, and movement of clouds. Combining these data with the cloud cover periodicity and movement speed features comprehensively reflects the dynamic evolution of cloud cover under different meteorological conditions, forming a comprehensive feature set that provides rich information for subsequent lightning trend forecasts.

[0012] This method constructs a lightning trigger factor dataset, uses LSTM and optical flow to extract cloud cover dynamic features, and integrates multi-source meteorological data to construct a cloud cover dynamic evolution feature set. This method accurately captures lightning trigger conditions and cloud cover dynamics. This process provides key data support and feature descriptions for lightning trend forecasts, helping to improve the accuracy and timeliness of lightning predictions, and has important practical significance for early warning and prevention of lightning disasters.

[0013] In some embodiments, based on the cloud cover dynamic feature set and the cloud cover layer feature matrix, combined with pre-established lightning-related cloud type labels, a lightning intensity level classification feature set is generated, including: Obtain cloud cover dynamic feature set and cloud cover layer feature matrix, and obtain standardized feature data set through data preprocessing; if the standardized feature data set contains missing values, use mean interpolation method to obtain a complete feature data set; Based on the complete feature dataset and pre-established lightning-related cloud type labels, a classification model is trained using the random forest algorithm to obtain a lightning intensity level classifier. The lightning intensity level classifier is then used to predict the complete feature dataset to generate an initial lightning intensity level classification result. If the confidence level of the initial lightning intensity level classification result is lower than a threshold, the classification result is adjusted using an integrated evaluation method to obtain an optimized lightning intensity level classification result. Based on the optimized lightning intensity level classification results, high-confidence classification features are extracted to generate a lightning intensity level classification feature set. Through the lightning intensity level classification feature set, combined with the spatiotemporal distribution of cloud cover dynamic characteristics and stratification characteristics, the spatiotemporal feature mapping of lightning intensity level is generated.

[0014] In this technical solution, a cloud cover dynamic feature set and a cloud cover layer feature matrix are obtained. These data contain rich information about cloud cover across time, space, and different layers. Data preprocessing, including noise removal and normalization, is performed on the raw data to eliminate dimensional differences and inconsistent data distribution, resulting in a standardized feature dataset. During data preprocessing, if missing values ​​are found, they are filled using mean interpolation, estimating the missing values ​​using the mean of the existing data to ensure data integrity, thereby obtaining a complete feature dataset. Based on the complete feature dataset and pre-established lightning-related cloud type labels, a classification model is trained using the random forest algorithm. Random forest is an ensemble learning algorithm that improves classification accuracy and stability by constructing multiple decision trees and combining their results. After sufficient training, a lightning intensity level classifier is obtained. This classifier is used to make predictions on the complete feature dataset, generating initial lightning intensity level classification results. This initial result includes a lightning intensity level prediction for each sample, but due to factors such as model uncertainty, the confidence level of some predictions is relatively low. In response to the situation where the confidence level in the initial lightning intensity level classification results is lower than the set threshold, an integrated evaluation method is introduced to adjust the classification results. The integrated evaluation method comprehensively considers the opinions of multiple models or evaluation indicators, optimizes the classification results through weighted voting and other methods, and finally obtains the optimized lightning intensity level classification results. The optimized classification results have higher accuracy and reliability. Based on the optimized classification results, high-confidence classification features are extracted. These features are the essence of cloud cover dynamics and stratification characteristics that are closely related to the lightning intensity level, thereby generating a lightning intensity level classification feature set. The lightning intensity level classification feature set is combined with the spatiotemporal distribution of cloud cover dynamics and stratification characteristics to generate a spatiotemporal feature map of the lightning intensity level. This process links the classification results with the temporal and spatial distribution characteristics of cloud cover, and intuitively displays the distribution of different lightning intensity levels in different time periods and spatial locations.

[0015] This method effectively transforms cloud cover dynamic and layered features into a set of lightning intensity classification features through steps including data preprocessing, classification model training, classification result optimization, and spatiotemporal feature map generation. Leveraging a random forest algorithm and ensemble evaluation methods, the accuracy and reliability of the classification results are improved. The generated spatiotemporal feature maps provide an intuitive basis for lightning trend forecasting.

[0016] In some embodiments, based on the lightning intensity level classification feature set and the cloud cover layer feature matrix, a coupling weight coefficient between cloud cover vertical layer and lightning intensity classification is calculated to generate a cloud cover-lightning vertical layer coupling feature set, including: From the lightning intensity level classification feature set and cloud cover layer feature matrix, the data matrix of lightning intensity level and cloud cover layer feature is obtained, and the main feature components are extracted using the principal component analysis method to obtain the feature dimension reduction set; Based on the feature dimensionality reduction set, the Pearson correlation coefficient between the lightning intensity level and the cloud cover layer characteristics is calculated to determine the correlation analysis result. If the absolute value of the Pearson correlation coefficient in the correlation analysis result is greater than a preset range, the corresponding lightning intensity level and cloud cover layer feature pair is marked as a high-correlation feature pair, and a high-correlation feature pair set is obtained. For a set of highly correlated feature pairs, a weighted linear combination method is used to calculate the coupling weight coefficients between lightning intensity levels and cloud cover stratification features, obtaining a set of coupling weight coefficients. From this set of coupling weight coefficients, feature pairs with weight coefficients greater than a preset threshold are obtained to generate a set of cloud cover-lightning vertical stratification coupling features. According to the cloud-lightning vertical stratification coupling feature set, the K-means clustering algorithm is used to classify the feature set to obtain the vertical stratification structure of cloud-lightning coupling features; from the vertical stratification structure, the correspondence between cloud distribution pattern and lightning activity intensity is extracted to generate the final cloud-lightning vertical stratification coupling feature set.

[0017] In this technical solution, a data matrix is ​​obtained from a set of lightning intensity classification features and a cloud cover stratification feature matrix. These data represent the characteristics of lightning intensity and cloud cover at different vertical layers, respectively. Principal component analysis (PCA) is used to reduce the data's dimensionality, extracting the key feature components and generating a reduced-dimensionality feature set. PCA effectively removes redundant information from the data, retaining its core features while reducing its dimensionality and improving the efficiency of subsequent computations. Based on the reduced-dimensionality feature set, the Pearson correlation coefficient is calculated between the lightning intensity level and cloud cover stratification features to quantify the linear correlation between them. Within a preset range, when the absolute value of the Pearson correlation coefficient exceeds this range, the corresponding lightning intensity level and cloud cover stratification feature is considered to have a strong correlation and is labeled as a highly correlated feature pair, forming a highly correlated feature pair set. This step identifies feature combinations with potentially significant value for lightning intensity prediction. For this highly correlated feature pair set, a weighted linear combination method is used to calculate the coupling weight coefficient between the lightning intensity level and cloud cover stratification features. The coupling weight coefficients reflect the relative influence of different cloud layer characteristics on lightning intensity, resulting in a set of coupling weight coefficients. Feature pairs with weight coefficients greater than a preset threshold are selected from this set. These feature pairs play a dominant role in cloud-lightning vertical layer coupling, generating a set of cloud-lightning vertical layer coupling features, focusing on key features. The K-means clustering algorithm is used to classify the set of cloud-lightning vertical layer coupling features, yielding a vertical hierarchical structure of cloud-lightning coupling features. K-means clustering can group similar feature sets into the same category, revealing the vertical distribution patterns of cloud-lightning coupling features. The correspondence between cloud distribution patterns and lightning activity intensity is extracted from the vertical hierarchical structure, ultimately generating an optimized set of cloud-lightning vertical layer coupling features, providing a clearer and more regular feature representation for lightning trend forecasting.

[0018] This method effectively couples cloud vertical stratification with lightning intensity classification through steps including data acquisition and feature extraction, correlation analysis and screening of highly correlated feature pairs, calculation of coupling weight coefficients and generation of coupled feature sets, and cluster analysis and extraction of vertical stratified structures. The use of principal component analysis and the Pearson correlation coefficient ensures the scientific nature of feature extraction and screening, while weighted linear combinations and K-means clustering effectively reveal the vertical stratified relationship between cloud cover and lightning intensity. The resulting cloud cover-lightning vertical stratification coupled feature set provides high-quality feature data for lightning approach trend forecasts, helping to improve the refinement and accuracy of lightning predictions.

[0019] In some embodiments, based on the lightning intensity level classification feature set, the cloud cover dynamic feature set, and the cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, a time series analysis method is used to generate a lightning occurrence probability distribution and a lightning spatial location coordinate sequence, forming a lightning probability distribution feature set, including: Obtain lightning intensity level classification feature sets, cloud cover dynamic feature sets, and cloud cover-lightning vertical layer coupling feature sets. Extract raw spatiotemporal data from lightning monitoring data and use data preprocessing methods to obtain standardized feature data sets. By standardizing the feature data set and using the long short-term memory network, a lightning occurrence probability prediction model is trained to obtain the lightning occurrence probability distribution. According to the probability distribution of lightning occurrence and combined with spatial positioning coordinates, a convolutional neural network is used to extract the spatial distribution characteristics of lightning and obtain the lightning spatial positioning coordinate sequence; If the probability distribution of lightning occurrence exceeds the preset range, the lightning probability distribution feature set is generated by integrating the lightning spatial positioning coordinate sequence through spatiotemporal sequence analysis.

[0020] In the above technical solution, in the lightning nowcasting system, accurately generating the probability distribution of lightning occurrence and the spatial positioning coordinate sequence is of vital importance for early warning of lightning activities and reducing the risk of lightning disasters.

[0021] A lightning intensity classification feature set, a cloud cover dynamic feature set, and a cloud cover-lightning vertical layer coupling feature set are collected. Raw spatiotemporal data is also extracted from lightning monitoring data. This raw data is preprocessed through data cleaning and normalization to eliminate noise and dimensionality differences, resulting in a standardized feature dataset. Using this standardized feature dataset as input, a long short-term memory (LSTM) network is used to train a lightning occurrence probability prediction model. LSTMs effectively capture long-term dependencies in time series data. Through their memory units and gating mechanism, they learn the temporal evolution of lightning-related features to predict a lightning occurrence probability distribution. This probability distribution reflects the likelihood of lightning occurrence at different time points and spatial locations. The lightning occurrence probability distribution is combined with spatial location coordinates, and a convolutional neural network (CNN) is used to extract lightning spatial distribution features, resulting in a sequence of lightning spatial location coordinates. CNNs excel at processing grid-like data, such as images, and can automatically learn both local and global features of lightning spatial distribution, enabling precise localization and serialized representation of the spatial location of lightning activity. If the lightning probability distribution exceeds the preset range, it indicates that lightning activity is relatively active and the probability of occurrence is high. In this case, a lightning probability distribution feature set is generated through spatiotemporal sequence analysis and the integration of the lightning spatial location coordinate sequence. Spatiotemporal sequence analysis comprehensively considers the temporal and spatial trends of lightning probability, incorporating the lightning spatial location coordinate sequence into the probability distribution. The resulting lightning probability distribution feature set not only contains information on lightning occurrence probability but also reflects the spatial evolution of lightning activity, providing a more comprehensive and detailed feature description for lightning trend prediction.

[0022] This method effectively transforms a multi-source feature set into a lightning probability distribution feature set through steps including data acquisition and preprocessing, lightning probability prediction model training, lightning spatial distribution feature extraction, and lightning probability distribution feature set generation. The application of LSTM and CNN leverages their respective strengths in time series analysis and spatial feature extraction, ensuring the accurate generation of lightning probability distributions and spatial location coordinate sequences. The generated lightning probability distribution feature set provides high-quality, information-rich feature data for lightning approach trend forecasting, helping to improve the refinement and accuracy of lightning predictions.

[0023] In some embodiments, based on the lightning probability distribution feature set and the cloud cover periodicity feature vector, a time series analysis method is used to generate a lightning occurrence probability time series and a lightning time window range series, forming a lightning time window prediction feature set, including: The input data is obtained from the lightning probability distribution feature set and the cloud periodicity feature vector, and the standardized feature data input is obtained through data preprocessing; Long short-term memory network is used to train the standardized feature data input to generate the lightning occurrence probability time series; According to the lightning occurrence probability time series, the high probability lightning occurrence time point is determined through the preset threshold judgment, and the lightning time window range sequence is obtained; For the lightning time window range sequence, cluster analysis method is used to extract the periodic characteristics of the time window and obtain the periodic feature set of the time window; Obtain feature vectors from the time window periodic feature set, perform secondary training through the long short-term memory network, and generate a lightning time window prediction feature set; If the matching degree between the lightning time window prediction feature set and the lightning occurrence probability time series is lower than the preset range, the long short-term memory network parameters are adjusted through iterative optimization to obtain the optimized prediction feature set; According to the optimized prediction feature set, the sliding window method is used to smooth the time series to obtain the final lightning time window prediction feature set.

[0024] In this technical solution, input data is obtained from a set of lightning probability distribution features and a cloud cover periodicity feature vector. These data incorporate key information about lightning probability and cloud cover periodicity. Data preprocessing, including noise removal and normalization, eliminates dimensionality differences and noise interference, resulting in standardized feature data input to ensure the accuracy of subsequent analysis. A long short-term memory (LSTM) network is then trained on this standardized feature input. LSTM effectively captures long-term dependencies in time series data. Through its unique memory units and gating mechanism, it learns the temporal evolution of lightning probability and cloud cover periodicity, generating a lightning probability time series. This time series details the temporal trend of lightning probability. Based on this lightning probability time series, a preset threshold is used to identify high-probability lightning occurrence times. Specifically, when the lightning probability exceeds the threshold, lightning is considered likely to occur at that time. This results in a lightning time window sequence. This sequence preliminarily delineates the time intervals during which lightning is likely to occur, providing a foundation for subsequent analysis. Cluster analysis is used to extract the periodic characteristics of lightning time window range sequences, generating a set of periodic features. Cluster analysis can group similar time windows together, revealing the temporal periodic distribution of lightning time windows and further deepening our understanding of the temporal patterns of lightning activity. Feature vectors are extracted from the set of periodic features and retrained using a long-short-term memory (LSTM) network to generate a set of lightning time window prediction features. This retraining aims to further explore the deep correlation between the periodic characteristics of the time window and lightning occurrence times, thereby improving the accuracy and relevance of the prediction feature set. If the match between the lightning time window prediction feature set and the lightning occurrence probability time series falls below a preset range, the LSTM network parameters are adjusted through iterative optimization. This iterative optimization process continuously fine-tunes the model parameters to improve the consistency of the prediction feature set with the actual lightning probability time series, resulting in an optimized prediction feature set. Based on this optimized prediction feature set, a sliding window method is used to smooth the time series. The sliding window method can effectively reduce short-term fluctuations and noise interference in time series, making the prediction results smoother and more stable, and ultimately obtaining a high-quality lightning time window prediction feature set, providing a reliable basis for lightning time prediction.

[0025] This method effectively transforms lightning probability distribution feature sets and cloud cover periodic feature vectors into lightning time window prediction feature sets through steps such as data acquisition and preprocessing, lightning occurrence probability time series generation, lightning time window range sequence determination, time window periodic feature extraction, lightning time window prediction feature set generation, prediction feature set optimization, and time series smoothing. The application of LSTM and cluster analysis leverages their respective strengths in time series analysis and periodic feature extraction, ensuring the accuracy and reliability of lightning time window predictions. The generated lightning time window prediction feature set provides high-quality, information-rich feature data for lightning approaching trend forecasts, helping to improve the refinement and accuracy of lightning predictions.

[0026] In some embodiments, based on the cloud cover periodicity feature vector and the lightning time window range sequence, the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window is calculated to generate a cloud cover-lightning time rhythm coupling feature set, including: Obtain cloud cover periodic characteristic vectors and lightning time window range series, and use data preprocessing methods to normalize and denoise the data to obtain cloud cover characteristic series and lightning time series; By using signal processing technology, the periodic fluctuation characteristics are extracted from the cloud cover feature sequence to generate a cloud cover periodic feature set. The cross-correlation analysis method is used to calculate the temporal correlation between the cloud cover periodic feature set and the lightning time series to obtain a rhythm matching degree sequence. If there are values ​​in the rhythm matching degree sequence that exceed the preset range, the corresponding cloud cover periodicity characteristics are paired with the lightning time window to generate a preliminary coupling feature set. For this preliminary coupling feature set, the principal component analysis method is used to extract the main feature components to obtain a simplified cloud cover-lightning time rhythm coupling feature set. Through time series analysis, periodic patterns in the simplified coupled feature set are detected to generate the final cloud cover-thunderstorm temporal rhythm coupled feature set.

[0027] In the above technical solution, a cloud cover periodic feature vector and a lightning time window range sequence are obtained. These data respectively contain the periodic variation patterns of cloud cover over time and the time intervals during which lightning may occur. Data preprocessing methods are used to normalize the raw data and scale the data to a uniform scale. A denoising algorithm is also used to remove noise interference from the data, resulting in smooth and dimensionally consistent cloud cover feature sequences and lightning time series, providing a high-quality data foundation for subsequent analysis. Signal processing techniques are used to extract periodic fluctuation characteristics from the cloud cover feature sequence. Methods such as fast Fourier transform are used to identify the main periodic components of cloud cover variation and generate a cloud cover periodic feature set. Cross-correlation analysis is then used to calculate the temporal correlation between the cloud cover periodic feature set and the lightning time series. Cross-correlation analysis can quantify the delayed correlation between the two time series, thereby obtaining a rhythm matching degree sequence that reflects the degree of correlation between the periodic fluctuations of cloud cover and the lightning time window at different time lags. If any value in the rhythm matching sequence exceeds a preset range, the corresponding cloud cover periodicity feature is considered to have a significant correlation with the lightning time window. These features are then paired to generate a preliminary coupled feature set. Principal component analysis is used to extract the main characteristic components from this preliminary coupled feature set. Principal component analysis effectively reduces data dimensionality, removes redundant information, and retains the core features within the coupled feature set, resulting in a streamlined cloud cover-lightning time rhythm coupled feature set, improving the data's compactness and representativeness. Periodic pattern detection is performed on the streamlined coupled feature set through time series analysis. Stable periodic patterns within the coupled feature set are identified using methods such as autocorrelation analysis and spectral analysis. The feature set is then further optimized and refined to generate the final cloud cover-lightning time rhythm coupled feature set. This final feature set not only retains the key correlation features between cloud cover periodic fluctuations and lightning time windows but also reflects their periodic variation patterns, providing a more accurate and efficient feature representation for lightning trend forecasting.

[0028] This method calculates the rhythmic matching degree between cloud cover periodic fluctuations and lightning time windows and generates a coupled feature set through steps including data preprocessing, feature extraction and correlation analysis, generation and simplification of a preliminary coupled feature set, and final coupled feature set generation. Signal processing techniques and cross-correlation analysis ensure the scientific nature of feature extraction and correlation analysis, while principal component analysis and time series analysis effectively enhance the quality and representativeness of the feature set. The resulting cloud cover-lightning time rhythm coupled feature set provides high-quality, information-rich feature data for lightning approaching trend forecasts, helping to improve the refinement and accuracy of lightning predictions.

[0029] In some embodiments, based on the lightning time window prediction feature set and the lightning trigger factor dataset, a time series analysis method is used to generate a lightning movement path sequence and a lightning duration sequence to form a lightning trend prediction feature set, including: Obtain a lightning time window prediction feature set and a lightning trigger factor dataset, perform standardization processing through a data preprocessing module, and obtain a standardized feature set and a standardized trigger factor dataset; Long short-term memory network is used to perform time series analysis on the standardized feature set to generate lightning movement path sequence and obtain path sequence data; Based on the standardized trigger factor dataset, the long short-term memory network is used to predict the duration of lightning, generate the lightning duration series, and obtain the duration series data. Through feature fusion, the path sequence data and duration series data are merged and processed to construct a lightning trend prediction feature set.

[0030] In the above technical solution, a lightning time window prediction feature set and a lightning trigger factor dataset are obtained. These data contain information about the temporal patterns of lightning occurrence and the meteorological conditions that trigger lightning, respectively. The raw data is standardized using a data preprocessing module, including outlier removal and normalization, to eliminate dimensional differences and noise interference, resulting in a standardized feature set and a standardized trigger factor dataset. Time series analysis is then performed on the standardized feature set using a long short-term memory (LSTM) network. The LSTM effectively captures long-term dependencies in time series data. Through its memory units and gating mechanism, it learns the spatiotemporal information in the lightning time window prediction feature set, thereby generating a lightning path sequence and obtaining path sequence data. This path sequence details the spatial trajectory of lightning activity and provides a basis for predicting lightning propagation direction. Based on the standardized trigger factor dataset, the LSTM is again applied to duration prediction. The LSTM analyzes the time series of meteorological elements in the trigger factor dataset, exploring the intrinsic correlations between factors such as humidity, temperature, and wind field and lightning duration, thereby generating a lightning duration sequence and obtaining duration sequence data. This sequence reflects the temporal continuity of lightning activity and helps assess the duration and intensity of lightning processes. Using feature fusion technology, path sequence data and duration sequence data are combined. The feature fusion process comprehensively considers two key aspects: lightning movement path and duration. By integrating these two data points, a feature set is constructed that comprehensively reflects the trends in lightning activity—the lightning trend prediction feature set. This feature set not only includes information on lightning movement in space but also incorporates its temporal persistence, providing a more comprehensive and accurate feature description for lightning trend prediction.

[0031] This method effectively transforms lightning time window prediction feature sets and lightning trigger factor datasets into lightning trend prediction feature sets through steps such as data acquisition and preprocessing, lightning path sequence generation, lightning duration sequence generation, and the construction of a lightning trend prediction feature set. The two applications of long-short-term memory networks fully leverage their advantages in time series analysis, accurately capturing the characteristics of lightning path and duration, respectively. The resulting lightning trend prediction feature set provides high-quality, information-rich feature data for impending lightning trend forecasts, helping to improve the refinement and accuracy of lightning predictions.

[0032] In some embodiments, based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning temporal rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, spatial features and temporal features are extracted, and a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends is generated through weighted fusion, including: Obtain lightning trend prediction feature sets, cloud cover-lightning vertical layer coupling feature sets, and cloud cover-lightning time rhythm coupling feature sets, and normalize the feature sets using a standardization processing method to obtain a normalized feature set; Based on the normalized feature set, a convolutional neural network is used to process the vertical layered cloud cover data, extract the spatial distribution features, and obtain the spatial feature set. According to the normalized feature set, the long short-term memory network is used to process the temporal rhythm coupling data, extract the temporal dynamic features, and obtain the temporal feature set; If the dimensions of the spatial feature set and the temporal feature set are consistent, the weighted fusion method is used to integrate the spatial feature set and the temporal feature set to obtain a fused feature set; if the dimensions are inconsistent, linear interpolation is performed on the feature set with lower dimensions to obtain a fused feature set; Based on the fusion feature set and combined with the pre-established lightning-related cloud type labels, a fully connected neural network is used for classification processing to obtain the spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends.

[0033] In this technical solution, a lightning trend prediction feature set, a cloud cover-lightning vertical layer coupling feature set, and a cloud cover-lightning temporal rhythm coupling feature set are obtained. These feature sets reflect the patterns of lightning activity and the correlation between cloud cover and lightning from different perspectives. A normalization process is used to normalize the feature set, eliminating dimensional differences and inconsistent data distribution between different features, resulting in a normalized feature set. Based on this normalized feature set, a convolutional neural network (CNN) is used to process the cloud cover vertical layer data. CNNs excel at processing grid-like data and can automatically learn the spatial distribution characteristics of cloud cover within vertical layers. Through operations such as convolutional and pooling layers, local and global spatial distribution patterns of cloud cover at different vertical layers are extracted, resulting in a spatial feature set. This provides key information for the spatial location of lightning activity and the spatial patterns of cloud cover evolution. Similarly, based on this normalized feature set, a long short-term memory network (LSTM) is used to process the temporal rhythm coupling data. LSTMs effectively capture long-term dependencies in time series data. Through their memory units and gating mechanisms, they learn the coupled characteristics of lightning temporal rhythms, extracting the dynamic temporal patterns of lightning activity and generating a temporal feature set. This set reflects the temporal evolution of lightning probability, duration, and other characteristics, providing support for predicting temporal patterns of lightning activity. If the spatial and temporal feature sets have the same dimensionality, they are integrated using a weighted fusion method. This weighted fusion assigns different weights to spatial and temporal features based on their importance in lightning prediction, linearly combining them to form a fused feature set. If the dimensionality differs, the feature set with the lower dimensionality is first linearly interpolated to match the dimensionality of the other feature set, and then weighted fusion is performed to generate the fused feature set. This fused feature set retains the spatial distribution of cloud cover and lightning activity while incorporating temporal evolution patterns, forming a comprehensive feature representation. Classification is performed using a fully connected neural network based on the fused feature set and pre-established lightning-related cloud type labels. The fully connected neural network is capable of in-depth learning and abstraction of the fused features. Through multi-layer neuron transformations, it ultimately outputs a collection of spatiotemporal coupled features of cloud cover dynamics and lightning trends. This collection comprehensively reflects the complex relationship between cloud cover and lightning activity in terms of vertical stratification and temporal rhythms, providing high-quality, information-rich feature data for lightning trend forecasting.

[0034] This method effectively integrates lightning trend prediction features, cloud cover-lightning vertical layer coupling features, and cloud cover-lightning temporal rhythm coupling features through steps such as data acquisition and standardization, spatial feature extraction, temporal feature extraction, feature fusion, and generation of a spatiotemporal coupling feature set. The combination of convolutional neural networks and long-short-term memory networks leverages their respective strengths in spatial feature extraction and time series analysis, ensuring the high-quality generation of the spatiotemporal coupling feature set. The resulting spatiotemporal coupling feature set provides more accurate and comprehensive feature data for lightning approaching trend forecasts, helping to improve the refinement and accuracy of lightning forecasts.

[0035] According to another aspect of the present invention, a lightning imminent trend forecasting system is provided, the system comprising: The first feature set module is used to obtain the cloud space distribution and time series, extract the first cloud texture feature vector, and generate a cloud layer feature matrix; reduce the dimension of the first cloud texture feature vector to generate a second cloud texture feature vector, and combine it with the cloud layer feature matrix to generate a regionalized cloud distribution feature set through clustering; The second feature set module is used to extract humidity, temperature, and wind field data to generate a lightning trigger factor dataset. It uses time series analysis and optical flow methods to analyze the regional cloud distribution feature set to form a cloud dynamic feature set. The third feature set module is used to generate a lightning intensity level classification feature set based on the cloud cover dynamic feature set and the cloud cover layer feature matrix, combined with the pre-established lightning-related cloud type labels; The fourth feature set module is used to calculate the coupling weight coefficient of cloud vertical stratification and lightning intensity classification based on the lightning intensity level classification feature set and the cloud layer feature matrix, and generate a cloud-lightning vertical stratification coupling feature set; The fifth feature set module is used to generate a lightning occurrence probability distribution and a lightning spatial location coordinate sequence using a time series analysis method based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, to form a lightning probability distribution feature set; The sixth feature set module is used to generate a lightning occurrence probability time series and a lightning time window range series using a time series analysis method based on the lightning probability distribution feature set and the cloud cover periodicity feature vector, thereby forming a lightning time window prediction feature set; The seventh feature set module is used to calculate the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window based on the cloud cover periodic feature vector and the lightning time window range sequence, and generate the cloud cover-lightning time rhythm coupling feature set; An eighth feature set module is used to generate a lightning movement path sequence and a lightning duration sequence using a time series analysis method based on the lightning time window prediction feature set and the lightning trigger factor dataset, thereby forming a lightning trend prediction feature set; The ninth feature set module is used to extract spatial and temporal features based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning time rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, and generate a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends through weighted fusion; The prediction module is used to obtain the dynamic evolution pattern of lightning trends by performing cluster analysis on the spatiotemporal coupling feature set; based on the dynamic evolution pattern, the lightning trend is predicted using the time series analysis method to obtain the final lightning trend prediction result.

[0036] In the above technical solution, in order to better utilize the above method, the present application proposes a lightning approach trend forecasting system, wherein each module corresponds to each step of the above method, and its specific principles have been described above and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is a flow chart of an embodiment of a method for predicting an approaching lightning trend according to the present invention; Figure 2 It is a structural diagram of an embodiment of a lightning approach trend forecasting system of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be described in further detail below with reference to the accompanying drawings and examples. It is particularly noted that the following examples are intended only to illustrate the present invention and are not intended to limit the scope of the present invention. Similarly, the following examples are only some embodiments of the present invention and are not intended to be exhaustive. All other embodiments obtained by those of ordinary skill in the art without creative effort are intended to fall within the scope of protection of the present invention.

[0040] Example 1 See also Figure 1 , a method for predicting an approaching lightning trend, the method comprising: S1. Obtain cloud space distribution and time series, extract the first cloud texture feature vector, and generate a cloud layer feature matrix; reduce the dimension of the first cloud texture feature vector to generate a second cloud texture feature vector, and combine it with the cloud layer feature matrix to generate a regionalized cloud distribution feature set through clustering; In this embodiment, S1, obtaining the cloud amount spatial distribution and time series, extracting a first cloud amount texture feature vector, and generating a cloud amount layered feature matrix; reducing the dimension of the first cloud amount texture feature vector to generate a second cloud amount texture feature vector, and combining the cloud amount layered feature matrix to generate a regionalized cloud amount distribution feature set through clustering, including: S11. Obtain cloud cover spatial distribution and time series from multi-source meteorological data, use convolutional neural network to extract the first cloud cover texture feature vector, combine gradient operator to generate cloud cover layered feature matrix, and generate cloud cover spatiotemporal feature set through feature splicing; S12. Performing principal component analysis on the first cloud texture feature vector in the cloud spatiotemporal feature set, retaining features whose principal component dimension is lower than a preset threshold, generating a second cloud texture feature vector, combining the cloud stratification feature matrix based on cloud height, and generating a regionalized cloud distribution feature set through K-means clustering. For example, cloud cover spatial distribution and time series are obtained from multi-source meteorological data. First, cloud cover grid data with a resolution of 0.25° × 0.25° is obtained by fusing satellite remote sensing data (such as MODIS) with ground-based observation data (such as the ERA5 reanalysis dataset). The grid data spans January 1 to December 31, 2023, with a three-hourly interval for 2920 hours. Data preprocessing uses a weighted average algorithm, with a weight of 0.7 for satellite data and 0.3 for ground-based data, ensuring high resolution combined with ground-based accuracy. This generates a cloud cover spatial distribution matrix (720 × 1440 pixels) and a time series vector (2920 pixels). Next, a convolutional neural network (CNN) is used to extract the first cloud cover texture feature vector. The network structure consists of three convolutional layers (kernel size 3 × 3, stride 1, padding 1), each with a Reluctant Unit (ReLU) activation function, outputting a 32-dimensional feature vector. The specific process is as follows: the cloud cover spatial distribution matrix is ​​input into a CNN. Convolution operations are then performed to extract local texture features (such as cloud boundary roughness). Dimensionality reduction is performed through maximum pooling (2×2), ultimately resulting in a 32-dimensional texture feature vector at each time point, reflecting the spatial heterogeneity of cloud cover. Next, the Sobel gradient operator is used to calculate the cloud cover layered feature matrix. Horizontal and vertical convolution kernels ([-1, 0, 1; -2, 0, 2; -1, 0, 1] and their transposes) are applied to the cloud cover grid data to generate a gradient magnitude matrix (720×1440) representing cloud edge intensity. The maximum gradient value is normalized to 1.0, reflecting the intensity of cloud boundary changes. Finally, a cloud cover spatiotemporal feature set is generated through feature concatenation. The 32-dimensional texture feature vector is concatenated with the 16-dimensional principal components (extracted via PCA, with a variance contribution of 95%) after dimensionality reduction of the gradient matrix, resulting in a 48-dimensional spatiotemporal feature vector at each time point. After feature splicing, K-means clustering (K=5) can be used to divide the spatiotemporal patterns of cloud cover. For example, high cloud cover areas are concentrated in the intertropical convergence zone, and low cloud cover areas appear in the subtropical high region. The time series shows that the fluctuation amplitude of cloud cover in summer (standard deviation 0.15) is greater than that in winter (0.08).

[0041] S2. Extract humidity, temperature, and wind field data to generate a lightning trigger factor dataset. Use time series analysis and optical flow methods to analyze the regional cloud distribution feature set to form a cloud dynamic feature set. In this embodiment, S2 extracts humidity, temperature, and wind field data to generate a lightning trigger factor dataset. The regionalized cloud distribution feature set is analyzed using the time series analysis method and the optical flow method to form a cloud dynamic feature set, including: S21. Obtain humidity, temperature, and wind field data from multi-source meteorological data, and generate a lightning trigger factor dataset through data cleaning and standardization. S22. Use a long short-term memory network to analyze the time series in the regionalized cloud distribution feature set, extract the temporal variation characteristics of cloud distribution, and obtain a cloud time series feature set; S23. Analyze the cloud time series feature set by the optical flow method, calculate the periodic change and movement speed of cloud distribution, and generate cloud periodic feature vectors and cloud movement speed series; S24. Based on the cloud cover periodic characteristic vector and cloud cover movement speed sequence, humidity, temperature and wind field data are integrated to construct a cloud cover dynamic evolution feature set.

[0042] For example, humidity, temperature, and wind data were extracted from multi-source meteorological data to generate a lightning trigger factor dataset. Data preprocessing was performed to obtain relative humidity (%), 2-meter temperature (°C), and 10-meter wind (u and v components, m / s) at a resolution of 0.25°×0.25° from the ECMWF ERA5 dataset. The temporal resolution was 1 hour, covering the region 30°N-40°N and 110°E-120°E. Assuming data from 00:00-23:00 on July 1, 2025, relative humidity (range 60%-90%), temperature (20-35°C), and wind speed (0-15 m / s) were extracted. Dimensionality reduction was performed using principal component analysis (PCA), retaining 90% of the variance. The resulting lightning trigger factor dataset consisted of a dominant humidity factor (weighted 0.6), a temperature factor (weighted 0.3), and a wind factor (weighted 0.1). A long short-term memory (LSTM) network was used to analyze the time series of regional cloud distribution characteristics. The input cloud cover data was MODIS cloud cover (0-100%) with a 24-hour time step. A three-layer LSTM model (with 128, 64, and 32 hidden units) was constructed. The model was trained for 100 epochs using the Adam optimizer (learning rate 0.001) with a mean squared error (MSE) loss function. The output was a predicted cloud cover series (with an error <5%). The optical flow method was used to generate cloud cover periodicity feature vectors and motion velocity series. The cloud cover motion vectors were calculated from GOES-16 satellite cloud images (2 km resolution) using the Farneback optical flow algorithm. Assuming cloud cluster velocities ranged from 0 to 10 m / s, the periodic features were extracted using a fast Fourier transform (FFT) with dominant periods of 12 and 24 hours. This generated a feature vector (10 dimensions, including periodicity strengths of 0.2-0.8) and a velocity series (mean 5 m / s, standard deviation 1.2 m / s). Ultimately, a set of cloud cover dynamic evolution features was formed, integrating triggering factors, periodic characteristics, and velocity sequences. K-means clustering (K=3) was used to classify cloud cover evolution patterns. The resulting feature sets were divided into three categories: high dynamics (speed >7 m / s), medium dynamics (3-7 m / s), and low dynamics (<3 m / s), with a classification accuracy of 85%. Triggering factors provide the foundation for the lightning environment, while LSTM captures the temporal patterns of cloud cover. Optical flow quantifies motion characteristics, and clustering integrates dynamic features, providing data support for lightning prediction.

[0043] S3. Generate a lightning intensity classification feature set based on the cloud cover dynamic feature set and the cloud cover layer feature matrix, combined with pre-established lightning-related cloud type labels; In this embodiment, S3 generates a lightning intensity level classification feature set based on the cloud cover dynamic feature set and the cloud cover layer feature matrix, combined with pre-established lightning-related cloud type labels, including: S31. Obtain a cloud cover dynamic feature set and a cloud cover layer feature matrix, and obtain a standardized feature data set through data preprocessing; if the standardized feature data set contains missing values, use the mean interpolation method to obtain a complete feature data set; S32. Based on the complete feature data set and pre-established lightning-associated cloud type labels, a classification model is trained using a random forest algorithm to obtain a lightning intensity level classifier. The complete feature data set is predicted using the lightning intensity level classifier to generate an initial lightning intensity level classification result. If the confidence level of the initial lightning intensity level classification result is lower than a threshold, the classification result is adjusted using an integrated evaluation method to obtain an optimized lightning intensity level classification result. S33. Based on the optimized lightning intensity level classification results, high-confidence classification features are extracted to generate a lightning intensity level classification feature set; through the lightning intensity level classification feature set, combined with the spatiotemporal distribution of cloud cover dynamic characteristics and stratification characteristics, a spatiotemporal feature map of the lightning intensity level is generated.

[0044] As an example, the following implementation method is used to generate a lightning intensity classification feature set using a random forest algorithm based on a cloud cover dynamic feature set and a cloud cover layer feature matrix, combined with pre-established lightning-related cloud type labels. Assume that the cloud cover dynamic feature set includes cloud cover change rate and cloud movement speed, with 1000 sample points, each of which includes cloud cover change rate (0.1 to 0.5, unit: % / minute) and cloud movement speed (5 to 20, unit: m / s). The first cloud cover layer feature matrix contains the cloud cover percentages of low-level clouds (0-2 km), mid-level clouds (2-6 km), and high-level clouds (6-12 km). The matrix dimensions are 1000×3, and the values ​​range from 0 to 1.

[0045] For example, the data for a sample point is [0.3, 0.4, 0.2], representing the percentages of low, medium, and high-level cloud cover. Lightning-related cloud type labels are based on historical data and are categorized as cumulonimbus (Cb), stratocumulus (Sc), and others. The label set consists of 1000 items, labeled 0 (no lightning), 1 (weak lightning), and 2 (strong lightning). First, dynamic features and hierarchical features are integrated to construct a feature vector with a dimension of 1000 × 5 (2 dynamic features + 3 hierarchical features). The feature vector is normalized using the Min-Max method to map values ​​to [0, 1]. For example, a cloud cover change rate of 0.3 is normalized to (0.3 - 0.1) / (0.5 - 0.1) = 0.5. A random forest algorithm is used with 100 decision trees and a maximum depth of 10. Feature selection is performed using the Gini index. The model is trained after the training set (80%) and the test set (20%) are split. The model outputs the probability of lightning intensity level for each sample. For example, if the prediction results for a sample are [0.1, 0.6, 0.3], the maximum probability corresponds to level 1 (weak lightning). To evaluate model performance, we calculated the test set accuracy (e.g., 85%) and analyzed feature importance. We found that the cloud cover change rate and the proportion of high-level clouds contributed most to classification, with importance scores of 0.4 and 0.35, respectively. Finally, we generated a lightning intensity level classification feature set containing the predicted levels and probability distributions for 1,000 samples.

[0046] S4. Calculate the coupling weight coefficient of cloud vertical stratification and lightning intensity classification based on the lightning intensity level classification feature set and the cloud layer feature matrix to generate a cloud-lightning vertical stratification coupling feature set; In this embodiment, S4, based on the lightning intensity level classification feature set and the cloud cover layer feature matrix, calculates the coupling weight coefficient of cloud cover vertical layer and lightning intensity classification, and generates a cloud cover-lightning vertical layer coupling feature set, including: S41. Obtaining a data matrix of lightning intensity level and cloud layer characteristics from the lightning intensity level classification feature set and the cloud layer feature matrix, extracting main feature components using a principal component analysis method, and obtaining a feature dimension reduction set; S42. Calculate the Pearson correlation coefficient between the lightning intensity level and the cloud cover layer characteristics based on the feature dimensionality reduction set to determine a correlation analysis result; if the absolute value of the Pearson correlation coefficient in the correlation analysis result is greater than a preset range, mark the corresponding lightning intensity level and cloud cover layer feature pair as a high-correlation feature pair, thereby obtaining a high-correlation feature pair set; S43. For the set of highly correlated feature pairs, a weighted linear combination method is used to calculate the coupling weight coefficients between the lightning intensity level and the cloud cover layer characteristics to obtain a set of coupling weight coefficients; from the set of coupling weight coefficients, feature pairs whose weight coefficients are greater than a preset threshold are obtained to generate a cloud cover-lightning vertical layer coupling feature set; S44. Based on the cloud-lightning vertical stratification coupling feature set, the K-means clustering algorithm is used to classify the feature set to obtain the vertical stratification structure of the cloud-lightning coupling features; from the vertical stratification structure, the correspondence between the cloud distribution pattern and the lightning activity intensity is extracted to generate the final cloud-lightning vertical stratification coupling feature set.

[0047] For example, assume the first lightning intensity classification feature set includes lightning intensity levels (classified from 1 to 5, corresponding to current intensities of 10kA, 20kA, 30kA, 40kA, and 50kA, respectively). The first cloud cover layer feature matrix contains three cloud cover data layers (low-level clouds 0-2 km, mid-level clouds 2-5 km, and high-level clouds 5-10 km), with the cloud cover percentage for each layer expressed from 0 to 100%. The data sample consists of 100 meteorological observation points, each containing the lightning intensity level and the corresponding three-layer cloud cover values. For example, sample 1: lightning intensity level 3 (30kA), low-level clouds 80%, mid-level clouds 60%, and high-level clouds 20%. First, construct a feature matrix, numerically converting the lightning intensity level and the three-layer cloud cover into a matrix X, where each row of X represents a sample, with columns consisting of [lightning intensity, low-level cloud cover, mid-level cloud cover, high-level cloud cover], for example, [30, 80, 60, 20]. To calculate the coupling weight coefficient between vertical cloud cover stratification and lightning intensity classification, we used the Pearson correlation coefficient analysis formula: r = cov(Xi, Y) / [std(Xi)·std(Y)], where Xi represents the cloud cover at a specific layer and Y represents the lightning intensity. For the calculation of low-level cloud cover and lightning intensity, assuming sample means μ_X1 = 70 (low-level cloud cover mean) and μ_Y = 25 (lightning intensity mean), covariance cov(X1, Y) = 150, standard deviations std(X1) = 15, and std(Y) = 10, we obtain r1 = 150 / (15·10) = 1, indicating a strong positive correlation between low-level cloud cover and lightning intensity. Similarly, for mid-level cloud cover, r2 = 0.8, and for high-level cloud cover, r3 = 0.4. The weight coefficient is obtained by normalizing the correlation coefficient using the formula wi = ri / Σri, where Σri = 1 + 0.8 + 0.4 = 2.2. Thus, w1 = 1 / 2.2 ≈ 0.455, w2 = 0.8 / 2.2 ≈ 0.364, and w3 = 0.4 / 2.2 ≈ 0.182. The cloud cover-lightning vertical layered coupled feature set is generated by weighted fusion of cloud cover data using the formula F = Σ(wi·Xi). For sample 1, F = 0.455·80 + 0.364·60 + 0.182·20 = 36.4 + 21.84 + 3.64 = 61.88, generating a new feature vector [30, 61.88]. This calculation is repeated for 100 samples to form the coupled feature set.

[0048] S5. Based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, a time series analysis method is used to generate a lightning occurrence probability distribution and a lightning spatial location coordinate sequence to form a lightning probability distribution feature set; In this embodiment, S5, based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, a time series analysis method is used to generate a lightning occurrence probability distribution and a lightning spatial location coordinate sequence, forming a lightning probability distribution feature set, including: S51. Obtain a lightning intensity level classification feature set, a cloud cover dynamic feature set, and a cloud cover-lightning vertical layer coupling feature set, extract original spatiotemporal data from lightning monitoring data, and use data preprocessing methods to obtain a standardized feature data set; S52. Using a standardized feature data set and a long short-term memory network, a lightning occurrence probability prediction model is trained to obtain a lightning occurrence probability distribution. S53. Based on the probability distribution of lightning occurrence and the spatial positioning coordinates, a convolutional neural network is used to extract the spatial distribution characteristics of lightning to obtain a sequence of lightning spatial positioning coordinates; S54. If the probability distribution of lightning occurrence exceeds a preset range, a lightning probability distribution feature set is generated by fusing the lightning spatial positioning coordinate sequence through spatiotemporal sequence analysis.

[0049] For example, based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layered coupling feature set, combined with lightning monitoring data, a long short-term memory network (LSTM) is used to generate a first lightning occurrence probability distribution and a first lightning spatial location coordinate sequence. The implementation method for forming the first lightning probability distribution feature set is as follows. Assuming the lightning intensity level classification feature set includes lightning intensity levels (classified from 1 to 5, corresponding to current intensities of 10kA, 20kA, 30kA, 40kA, and 50kA, respectively), classification is performed using radar reflectivity data (measured in dBZ, ranging from 20 to 60). Using the support vector machine (SVM) algorithm, the radial basis function (RBF) kernel function is selected, the penalty parameter C is set to 1.0, and the kernel parameter γ is set to 0.01. Training on 1,000 sets of sample data (each set containing radar reflectivity, wind speed, and humidity) yields a classification accuracy of 85%. The cloud cover dynamic feature set extracts cloud coverage (0-100%) and cloud top height (2-15 km) from meteorological satellite data. A convolutional neural network (CNN) is used to extract dynamic features with a convolution kernel size of 3×3 and a stride of 1. The pooling layer uses maximum pooling and outputs a 64-dimensional feature vector. Ten consecutive frames of cloud cover images (with a time interval of 5 minutes) are analyzed to calculate the cloud cover change rate.

[0050] For example, an increase in cloud cover from 70% to 90% has a rate of change of 4% / min. The cloud cover-lightning vertical layer coupling feature set is derived by analyzing the coupling relationship between cloud height (0-3 km for the lower layer, 3-8 km for the middle layer, and 8-15 km for the upper layer) and the location of lightning occurrence. Principal component analysis (PCA) is used for dimensionality reduction, retaining 95% variance, to produce a 5-dimensional coupling feature vector. Lightning monitoring data includes the spatial coordinates (latitude, longitude, and altitude) and timestamps of 1,000 lightning events, with a temporal resolution of 1 second and a spatial resolution of 0.1°. The above feature sets (64-dimensional intensity features, 64-dimensional cloud cover dynamic features, and 5-dimensional coupling features) are concatenated into a 133-dimensional input vector and fed into an LSTM network. The network structure consists of two layers of LSTM units (128 neurons per layer), with a time step of 10, a tanh activation function, an Adam optimizer, a learning rate of 0.001, and 100 epochs of training. The loss function is the mean squared error (MSE), and the MSE converges to 0.015. The LSTM outputs a lightning probability distribution (ranging from 0 to 1, for example, a probability of 0.75 for a certain area) and a sequence of spatial coordinates. For example, latitude 39.9°, longitude 116.3°, and altitude 5 km. The probabilities are normalized using a softmax function to generate a lightning probability distribution feature set, consisting of probability values ​​and corresponding coordinates.

[0051] S6. Based on the lightning probability distribution feature set and the cloud cover periodic feature vector, a time series analysis method is used to generate a lightning occurrence probability time series and a lightning time window range series to form a lightning time window prediction feature set; In this embodiment, S6, based on the lightning probability distribution feature set and the cloud cover periodic feature vector, a time series analysis method is used to generate a lightning occurrence probability time series and a lightning time window range series to form a lightning time window prediction feature set, including: S61. Obtain input data from a lightning probability distribution feature set and a cloud cover periodicity feature vector, and obtain standardized feature data input through data preprocessing; S62. Using a long short-term memory network to train the standardized feature data input to generate a time series of lightning occurrence probabilities; S63. Determine a high-probability lightning occurrence time point based on a lightning occurrence probability time series and a preset threshold value, and obtain a lightning time window range sequence; S64. For the lightning time window range sequence, a cluster analysis method is used to extract the periodic characteristics of the time window to obtain a set of periodic characteristics of the time window; S65. Obtain a feature vector from the time window periodic feature set, perform secondary training through a long short-term memory network, and generate a lightning time window prediction feature set; S66. If the matching degree between the lightning time window prediction feature set and the lightning occurrence probability time series is lower than a preset range, adjusting the long short-term memory network parameters through iterative optimization to obtain an optimized prediction feature set; S67. Based on the optimized prediction feature set, a sliding window method is used to smooth the time series to obtain a final lightning time window prediction feature set.

[0052] For example, based on the lightning probability distribution feature set and the cloud cover periodicity feature vector, a lightning occurrence probability time series and a lightning time window range series are first generated to form a lightning time window prediction feature set. Assume that the lightning probability distribution feature set contains the lightning occurrence probability for each hour within a 24-hour period, for example, [0.1, 0.15, 0.2, …, 0.05], a total of 24 values, representing the lightning probability for each hour of the day in a certain area. The cloud cover periodicity feature vector is the cloud cover percentage for a 12-hour period, for example, [0.3, 0.5, 0.7, …, 0.2], reflecting the periodic trend of cloud cover changes. First, these two feature sets are normalized using the minimum-maximum normalization method to map the lightning probability and cloud cover values ​​to the range [0, 1]. For example, a lightning probability of 0.2 is normalized to (0.2-0.05) / (0.3-0.05) = 0.6. Next, a long short-term memory (LSTM) network was constructed. The network structure consists of two layers of LSTM units, with 64 neurons in each layer. The input dimension is 2 (lightning probability and cloud cover), and the time step is 12 hours. The training data consists of normalized features from the past 30 days, with a batch size of 32. The optimizer uses Adam with a learning rate of 0.001 and a loss function of mean squared error. After training, the LSTM predicts a time series of lightning probability for the next 24 hours, such as [0.12, 0.18, 0.25, ..., 0.08]. Subsequently, based on a probability threshold of 0.15, time points with a probability greater than 0.15 are extracted to form a time window range sequence, such as [14:00-16:00, 20:00-22:00]. To ensure forecast accuracy, we analyze the wind speed data within the time window (assuming a mean of 5 m / s) based on meteorological business logic. If the wind speed is greater than 4 m / s, the probability of lightning is increased. We then adjust the time window weights to generate the final forecast feature set, for example, {[14:00-16:00, 0.8], [20:00-22:00, 0.9]}.

[0053] S7. Based on the cloud cover periodicity feature vector and the lightning time window range sequence, the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window is calculated to generate a cloud cover-lightning time rhythm coupling feature set; In this embodiment, S7, based on the cloud cover periodicity feature vector and the lightning time window range sequence, calculates the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window, and generates a cloud cover-lightning time rhythm coupling feature set, including: S71. Obtain cloud cover periodic characteristic vectors and lightning time window range sequences, and perform data preprocessing normalization and denoising on the data to obtain cloud cover characteristic sequences and lightning time series. S72. Using signal processing technology, extract periodic fluctuation characteristics from the cloud cover feature sequence to generate a cloud cover periodic feature set; using a cross-correlation analysis method, calculate the temporal correlation between the cloud cover periodic feature set and the lightning time series to obtain a rhythm matching degree sequence; S73. If a value in the rhythm matching degree sequence exceeds a preset range, the corresponding cloud cover periodicity feature is paired with the lightning time window to generate a preliminary coupling feature set; principal component analysis is used to extract the main feature components of the preliminary coupling feature set to obtain a simplified cloud cover-lightning time rhythm coupling feature set; S74. Detect periodic patterns in the simplified coupled feature set through time series analysis, and generate the final cloud cover-lightning time rhythm coupled feature set.

[0054] For example, assume the input data is a cloud cover periodicity feature vector, representing the periodic variation of cloud cover within a 24-hour period. The data is expressed as hourly cloud cover percentages, for example: [10, 15, 20, 30, 40, 50, 60, 70, 65, 55, 45,35, 30, 25, 20, 15, 10, 10, 15, 20, 25, 30, 25, 20], expressed in percentages, reflecting the fluctuation of cloud cover within a day. The lightning time window range sequence is the time period in which lightning occurs, for example: [3, 4, 5, 15, 16, 17], indicating that lightning occurs within the 3rd, 4th, 5th, 15th, 16th, and 17th hours, with a time resolution of 1 hour. First, we generated a cloud cover periodic feature vector. We used a fast Fourier transform (FFT) to extract the periodic features, and calculated the spectrum using Python's numpy.fft module. We found a dominant period of 12 hours and an amplitude of 30. We normalized the cloud cover data to a mean of 35 and a variance of 400, resulting in a data range of [-1.25, 1.75]. The lightning time window sequence was converted to a binary sequence with the same length as the cloud cover vector. The lightning hour was set to 1, and all other values ​​were set to 0, resulting in the following sequence: [0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0]. Cross-correlation analysis was then performed using the numpy.correlate function with the "full" mode to calculate the cross-correlation between the normalized cloud cover vector and the lightning binary sequence. The maximum cross-correlation coefficient was 0.65, with a lag time of 2 hours, indicating that the cloud cover peak lagged the lightning event by 2 hours. The rhythm matching degree was defined as the maximum cross-correlation coefficient, i.e., 0.65. A cloud cover-lightning temporal rhythm coupling feature set was generated, including a cross-correlation coefficient of 0.65, a lag time of 2 hours, a main period of 12 hours, a cloud cover amplitude of 30, and a lightning time window ratio of 6 / 24 (0.25). Cross-correlation analysis established a temporal association between the periodic fluctuations in cloud cover and the lightning time window. This coupled feature set can be used in subsequent meteorological forecast models to optimize the accuracy of lightning warnings.

[0055] S8. Based on the lightning time window prediction feature set and the lightning trigger factor dataset, a time series analysis method is used to generate a lightning movement path sequence and a lightning duration sequence to form a lightning trend prediction feature set; In this embodiment, S8, based on the lightning time window prediction feature set and the lightning trigger factor dataset, a time series analysis method is used to generate a lightning movement path sequence and a lightning duration sequence to form a lightning trend prediction feature set, including: S81, obtaining a lightning time window prediction feature set and a lightning trigger factor dataset, and performing standardization processing through a data preprocessing module to obtain a standardized feature set and a standardized trigger factor dataset; S82. Use a long short-term memory network to perform time series analysis on the standardized feature set to generate a lightning movement path sequence and obtain path sequence data; S83. Based on the standardized trigger factor dataset, a long short-term memory network is used to predict the duration of lightning, generate a lightning duration series, and obtain duration series data; S84. By feature fusion, the path sequence data and the duration sequence data are merged and processed to construct a lightning trend prediction feature set.

[0056] As an example, based on a lightning time window prediction feature set and a lightning triggering factor dataset, a long short-term memory network is used to generate a lightning path sequence and a lightning duration sequence, thereby forming a lightning trend prediction feature set. Assume that the lightning time window prediction feature set contains meteorological data from 18:00 to 20:00 on July 22, 2025, including a temperature of 30.5°C, a humidity of 75%, a wind speed of 5.2m / s, and an air pressure of 1002hPa, with a spatial resolution of 1km×1km. The lightning triggering factor dataset includes historical lightning occurrence times, locations (latitude and longitude, such as 120.5°E, 30.2°N), intensity (current 50kA), and cloud thickness of 500m. First, the prediction feature set is preprocessed using Z-score normalization to transform features like temperature and humidity into distributions with a mean of 0 and a standard deviation of 1. The calculation formula is z = (x - μ) / σ, where μ is the mean and σ is the standard deviation. For example, after normalization, temperature is (30.5 - 28) / 2.5 = 1.0. Next, a long short-term memory network model is constructed with three LSTM layers, 64 units per layer, an input dimension of 4 (temperature, humidity, wind speed, and air pressure), and a time step of 12 (one data point every 10 minutes, covering 2 hours). The Adam optimizer is used with a learning rate of 0.001 and a mean squared error loss function. During training, the normalized feature set and the triggering factor dataset are input. The model learns the relationship between lightning location and duration, outputting a sequence of movement paths (for example, from 120.5°E, 30.2°N to 120.6°E, 30.3°N with a step size of 0.1°) and a duration sequence. For example, lightning may last 30 minutes or 45 minutes. Path prediction uses a two-dimensional Gaussian distribution to estimate movement probability, with a probability density function of f(x,y)=1 / (2πσ²)exp(-((x-μx)²+(y-μy)²) / (2σ²)), where σ=0.05°. Duration prediction uses an LSTM to output regression values. Based on the historical average lightning duration of 40 minutes, the predicted value is adjusted to a range of 20-60 minutes. Finally, the path and duration sequences are combined to generate a trend prediction feature set, such as {(120.6°E, 30.3°N, 30 minutes), (120.7°E, 30.4°N, 45 minutes)}.

[0057] S9. Based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning temporal rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, spatial and temporal features are extracted, and a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends is generated through weighted fusion. In this embodiment, S9 extracts spatial and temporal features based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning temporal rhythm coupling feature set, combined with pre-established lightning-related cloud type labels. A spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trend is generated through weighted fusion, including: S91. Obtain a lightning trend prediction feature set, a cloud cover-lightning vertical layer coupling feature set, and a cloud cover-lightning time rhythm coupling feature set, and normalize the feature sets using a standardization processing method to obtain a normalized feature set. S92. Based on the normalized feature set, a convolutional neural network is used to process the cloud cover vertical layer data, extract spatial distribution features, and obtain a spatial feature set; S93. Processing the temporal rhythm coupling data using a long short-term memory network based on the normalized feature set, extracting temporal dynamic features, and obtaining a temporal feature set; S94. If the dimensions of the spatial feature set and the temporal feature set are consistent, a weighted fusion method is used to integrate the spatial feature set and the temporal feature set to obtain a fused feature set; if the dimensions are inconsistent, linear interpolation is performed on the feature set with the lower dimension to obtain a fused feature set; S95. Based on the fusion feature set and the pre-established lightning-related cloud type labels, a fully connected neural network is used for classification processing to obtain the spatiotemporal coupling feature set of the dynamic evolution of cloud cover and lightning trends.

[0058] As an example, based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning temporal rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, a convolutional neural network (CNN) is used to extract spatial features, and a long short-term memory network (LSTM) is used to extract temporal features. Through weighted fusion, a spatiotemporal coupling feature set for cloud cover dynamic evolution and lightning trends is generated. Assume that the input lightning trend prediction feature set includes lightning occurrence probability (0.7), lightning frequency (3 times per minute), and location coordinates (100×100 latitude and longitude grid). The cloud cover-lightning vertical layer coupling feature set includes cloud height (5000 meters), cloud cover (80%), and vertical charge distribution (positive charge is concentrated at the cloud top). The cloud cover-lightning temporal rhythm coupling feature set includes time series of cloud cover change rate (10% per hour) and lightning activity period (peak every 2 hours). Pre-established lightning-related cloud type labels are based on meteorological radar data and are categorized as cumulonimbus (80%), stratocumulus (15%), and other cloud types (5%). First, a CNN is used to process spatial features, taking a 100×100 grid of lightning probability and cloud cover as input. The convolution kernel size is set to 3×3 with a stride of 1. A 2×2 max pooling layer is used in the pooling layer to extract spatial correlation features and output a 64-dimensional feature map. The activation function is ReLU, and the calculation formula is f(x)=max(0,x). Next, an LSTM is used to process time series data, taking the cloud cover change rate and lightning activity period as input. The number of hidden layer units is set to 128, the time step is 6 (representing 6 hours), the forget gate weight is initialized to 0.5, and the update formula is ft=σ(Wf·[ht-1,xt]+bf), generating a 32-dimensional temporal feature vector. To fuse features, a weighted fusion algorithm was used, with a spatial feature weight of 0.6 and a temporal feature weight of 0.4. The fusion formula was F = 0.6·Fs + 0.4·Ft, where Fs is the CNN output and Ft is the LSTM output. The fused feature dimension was 96. This was fed into a fully connected layer, which output a set of spatiotemporal coupled features (10 dimensions) representing the dynamic evolution of cloud cover and lightning trends, corresponding to 10 lightning intensity levels.

[0059] S10. By performing cluster analysis on the spatiotemporal coupling feature set, a dynamic evolution pattern of the lightning trend is obtained; based on the dynamic evolution pattern, the lightning trend is predicted using a time series analysis method to obtain a final lightning trend prediction result.

[0060] This embodiment discloses a lightning prediction and analysis method based on the fusion of multi-source meteorological data. It aims at the business scenario problems of lightning occurrence probability, spatial positioning, time window and trend prediction under complex meteorological conditions, and realizes high-precision lightning prediction through multi-level feature extraction and coupling analysis. The present invention first extracts the spatiotemporal distribution characteristics of cloud cover from multi-source meteorological data, uses convolutional neural networks and gradient operators to generate texture and edge features, and optimizes the cloud cover feature set through principal component analysis and K-means clustering; combines humidity, temperature, and wind field data, and uses long-short-term memory networks and optical flow methods to extract dynamic evolution characteristics of cloud cover; constructs cloud cover-lightning coupling features through random forest algorithm and correlation analysis; and finally generates lightning probability distribution, time window and movement path prediction through weighted fusion of convolutional neural networks and long-short-term memory networks. The present invention significantly improves the accuracy and timeliness of lightning prediction through deep fusion and dynamic optimization of spatiotemporal features, providing efficient technical support for meteorological disaster warning.

[0061] Example 2 See also Figure 2 A lightning approach trend forecasting system, based on the method described in one embodiment, comprises: The first feature set module is used to obtain the cloud space distribution and time series, extract the first cloud texture feature vector, and generate a cloud layer feature matrix; reduce the dimension of the first cloud texture feature vector to generate a second cloud texture feature vector, and combine it with the cloud layer feature matrix to generate a regionalized cloud distribution feature set through clustering; The second feature set module is used to extract humidity, temperature, and wind field data to generate a lightning trigger factor dataset. It uses time series analysis and optical flow methods to analyze the regional cloud distribution feature set to form a cloud dynamic feature set. The third feature set module is used to generate a lightning intensity level classification feature set based on the cloud cover dynamic feature set and the cloud cover layer feature matrix, combined with the pre-established lightning-related cloud type labels; The fourth feature set module is used to calculate the coupling weight coefficient of cloud vertical stratification and lightning intensity classification based on the lightning intensity level classification feature set and the cloud layer feature matrix, and generate a cloud-lightning vertical stratification coupling feature set; The fifth feature set module is used to generate a lightning occurrence probability distribution and a lightning spatial location coordinate sequence using a time series analysis method based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, to form a lightning probability distribution feature set; The sixth feature set module is used to generate a lightning occurrence probability time series and a lightning time window range series using a time series analysis method based on the lightning probability distribution feature set and the cloud cover periodicity feature vector, thereby forming a lightning time window prediction feature set; The seventh feature set module is used to calculate the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window based on the cloud cover periodic feature vector and the lightning time window range sequence, and generate the cloud cover-lightning time rhythm coupling feature set; An eighth feature set module is used to generate a lightning movement path sequence and a lightning duration sequence using a time series analysis method based on the lightning time window prediction feature set and the lightning trigger factor dataset, thereby forming a lightning trend prediction feature set; The ninth feature set module is used to extract spatial and temporal features based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning time rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, and generate a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends through weighted fusion; The prediction module is used to obtain the dynamic evolution pattern of lightning trends by performing cluster analysis on the spatiotemporal coupling feature set; based on the dynamic evolution pattern, the lightning trend is predicted using the time series analysis method to obtain the final lightning trend prediction result.

[0062] In this embodiment, in order to better utilize the method described in one of the embodiments, the present application proposes a lightning approach trend forecasting system, wherein each module corresponds to each step of the above method, and its specific principles have been described above and will not be repeated here.

[0063] The above descriptions are only some embodiments of the present invention and do not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting the approaching trend of lightning, characterized in that: The method comprises: Obtain the cloud amount spatial distribution and time series, extract the first cloud amount texture feature vector, and generate a cloud amount layer feature matrix; reduce the dimension of the first cloud amount texture feature vector to generate a second cloud amount texture feature vector, and combine it with the cloud amount layer feature matrix to generate a regionalized cloud amount distribution feature set through clustering; Humidity, temperature, and wind field data were extracted to generate a lightning trigger factor dataset. Time series analysis and optical flow methods were used to analyze the regional cloud distribution feature set to form a cloud dynamic feature set. Based on the cloud cover dynamic feature set and cloud cover layer feature matrix, combined with the pre-established lightning-related cloud type labels, a lightning intensity level classification feature set is generated; Based on the lightning intensity level classification feature set and the cloud cover stratification feature matrix, the coupling weight coefficient of cloud cover vertical stratification and lightning intensity classification is calculated to generate the cloud cover-lightning vertical stratification coupling feature set; Based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, the time series analysis method is used to generate the lightning occurrence probability distribution and lightning spatial location coordinate sequence, forming a lightning probability distribution feature set; Based on the lightning probability distribution feature set and cloud periodicity feature vector, the time series analysis method is used to generate the lightning occurrence probability time series and the lightning time window range series, forming the lightning time window prediction feature set; Based on the cloud cover periodicity feature vector and the lightning time window range sequence, the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window is calculated to generate the cloud cover-lightning time rhythm coupling feature set. Based on the lightning time window prediction feature set and the lightning trigger factor dataset, the time series analysis method is used to generate the lightning movement path sequence and lightning duration sequence to form the lightning trend prediction feature set; Based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning temporal rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, spatial and temporal features are extracted, and a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends is generated through weighted fusion. By performing cluster analysis on the spatiotemporal coupling feature set, the dynamic evolution pattern of the lightning trend is obtained; based on the dynamic evolution pattern, the lightning trend is predicted using the time series analysis method to obtain the final lightning trend prediction result.

2. A method for predicting the approaching trend of lightning according to claim 1, characterized in that: Humidity, temperature, and wind field data were extracted to generate a lightning trigger factor dataset. Time series analysis and optical flow methods were used to analyze the regional cloud distribution feature set to form a cloud dynamic feature set, including: Humidity, temperature, and wind field data are obtained from multi-source meteorological data, and a lightning trigger factor dataset is generated through data cleaning and standardization. The long short-term memory network is used to analyze the time series in the regional cloud distribution feature set, extract the temporal variation characteristics of cloud distribution, and obtain the cloud time series feature set. The cloud time series feature set is analyzed by optical flow method, the periodic change and movement speed of cloud distribution are calculated, and the cloud periodic feature vector and cloud movement speed sequence are generated. Based on the cloud cover periodic characteristic vector and cloud cover movement speed sequence, humidity, temperature and wind field data are integrated to construct a cloud cover dynamic evolution feature set.

3. A method for predicting the approaching trend of lightning according to claim 1, characterized in that: Based on the cloud cover dynamic feature set and cloud cover layer feature matrix, combined with the pre-established lightning-related cloud type labels, a lightning intensity level classification feature set is generated, including: Obtain cloud cover dynamic feature set and cloud cover layer feature matrix, and obtain standardized feature data set through data preprocessing; if the standardized feature data set contains missing values, use mean interpolation method to obtain a complete feature data set; Based on the complete feature dataset and pre-established lightning-related cloud type labels, a classification model is trained using the random forest algorithm to obtain a lightning intensity level classifier. The lightning intensity level classifier is then used to predict the complete feature dataset to generate an initial lightning intensity level classification result. If the confidence level of the initial lightning intensity level classification result is lower than a threshold, the classification result is adjusted using an integrated evaluation method to obtain an optimized lightning intensity level classification result. Based on the optimized lightning intensity level classification results, high-confidence classification features are extracted to generate a lightning intensity level classification feature set. Through the lightning intensity level classification feature set, combined with the spatiotemporal distribution of cloud cover dynamic characteristics and stratification characteristics, the spatiotemporal feature mapping of lightning intensity level is generated.

4. A method for predicting the approaching trend of lightning according to claim 1, characterized in that: Based on the lightning intensity level classification feature set and the cloud cover stratification feature matrix, the coupling weight coefficient of cloud cover vertical stratification and lightning intensity classification is calculated to generate the cloud cover-lightning vertical stratification coupling feature set, including: From the lightning intensity level classification feature set and cloud cover layer feature matrix, the data matrix of lightning intensity level and cloud cover layer feature is obtained, and the main feature components are extracted using the principal component analysis method to obtain the feature dimension reduction set; Based on the feature dimensionality reduction set, the Pearson correlation coefficient between the lightning intensity level and the cloud cover layer characteristics is calculated to determine the correlation analysis result. If the absolute value of the Pearson correlation coefficient in the correlation analysis result is greater than a preset range, the corresponding lightning intensity level and cloud cover layer feature pair is marked as a high-correlation feature pair, and a high-correlation feature pair set is obtained. For a set of highly correlated feature pairs, a weighted linear combination method is used to calculate the coupling weight coefficients between lightning intensity levels and cloud cover stratification features, obtaining a set of coupling weight coefficients. From this set of coupling weight coefficients, feature pairs with weight coefficients greater than a preset threshold are obtained to generate a set of cloud cover-lightning vertical stratification coupling features. According to the cloud-lightning vertical stratification coupling feature set, the K-means clustering algorithm is used to classify the feature set to obtain the vertical stratification structure of cloud-lightning coupling features; from the vertical stratification structure, the correspondence between cloud distribution pattern and lightning activity intensity is extracted to generate the final cloud-lightning vertical stratification coupling feature set.

5. A method for predicting the approaching trend of lightning according to claim 1, characterized in that: Based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, the time series analysis method is used to generate the lightning occurrence probability distribution and lightning spatial location coordinate sequence, forming a lightning probability distribution feature set, including: Obtain lightning intensity level classification feature sets, cloud cover dynamic feature sets, and cloud cover-lightning vertical layer coupling feature sets. Extract raw spatiotemporal data from lightning monitoring data and use data preprocessing methods to obtain standardized feature data sets. By standardizing the feature data set and using the long short-term memory network, a lightning occurrence probability prediction model is trained to obtain the lightning occurrence probability distribution. According to the probability distribution of lightning occurrence and combined with spatial positioning coordinates, a convolutional neural network is used to extract the spatial distribution characteristics of lightning and obtain the lightning spatial positioning coordinate sequence; If the probability distribution of lightning occurrence exceeds the preset range, the lightning probability distribution feature set is generated by integrating the lightning spatial positioning coordinate sequence through spatiotemporal sequence analysis.

6. A method for predicting the approaching trend of lightning according to claim 1, characterized in that: Based on the lightning probability distribution feature set and cloud cover periodic feature vector, the time series analysis method is used to generate the lightning occurrence probability time series and the lightning time window range series, forming the lightning time window prediction feature set, including: The input data is obtained from the lightning probability distribution feature set and the cloud periodicity feature vector, and the standardized feature data input is obtained through data preprocessing; Long short-term memory network is used to train the standardized feature data input to generate the lightning occurrence probability time series; According to the lightning occurrence probability time series, the high probability lightning occurrence time point is determined through the preset threshold judgment, and the lightning time window range sequence is obtained; For the lightning time window range sequence, cluster analysis method is used to extract the periodic characteristics of the time window and obtain the periodic feature set of the time window; Obtain feature vectors from the time window periodic feature set, perform secondary training through the long short-term memory network, and generate a lightning time window prediction feature set; If the matching degree between the lightning time window prediction feature set and the lightning occurrence probability time series is lower than the preset range, the long short-term memory network parameters are adjusted through iterative optimization to obtain the optimized prediction feature set; According to the optimized prediction feature set, the sliding window method is used to smooth the time series to obtain the final lightning time window prediction feature set.

7. A method for predicting the approaching trend of lightning according to claim 1, characterized in that: Based on the cloud cover periodic feature vector and the lightning time window range sequence, the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window is calculated to generate the cloud cover-lightning time rhythm coupling feature set, including: Obtain cloud cover periodic characteristic vectors and lightning time window range series, and use data preprocessing methods to normalize and denoise the data to obtain cloud cover characteristic series and lightning time series; By using signal processing technology, the periodic fluctuation characteristics are extracted from the cloud cover feature sequence to generate a cloud cover periodic feature set. The cross-correlation analysis method is used to calculate the temporal correlation between the cloud cover periodic feature set and the lightning time series to obtain a rhythm matching degree sequence. If there are values ​​in the rhythm matching degree sequence that exceed the preset range, the corresponding cloud cover periodicity characteristics are paired with the lightning time window to generate a preliminary coupling feature set. For this preliminary coupling feature set, the principal component analysis method is used to extract the main feature components to obtain a simplified cloud cover-lightning time rhythm coupling feature set. Through time series analysis, periodic patterns in the simplified coupled feature set are detected to generate the final cloud cover-thunderstorm temporal rhythm coupled feature set.

8. A method for predicting the approaching trend of lightning according to claim 1, characterized in that: Based on the lightning time window prediction feature set and the lightning trigger factor dataset, the time series analysis method is used to generate the lightning movement path sequence and lightning duration sequence, forming a lightning trend prediction feature set, including: Obtain a lightning time window prediction feature set and a lightning trigger factor dataset, perform standardization processing through a data preprocessing module, and obtain a standardized feature set and a standardized trigger factor dataset; Long short-term memory network is used to perform time series analysis on the standardized feature set to generate lightning movement path sequence and obtain path sequence data; Based on the standardized trigger factor dataset, the long short-term memory network is used to predict the duration of lightning, generate the lightning duration series, and obtain the duration series data. Through feature fusion, the path sequence data and duration series data are merged and processed to construct a lightning trend prediction feature set.

9. A method for predicting the approaching trend of lightning according to claim 1, characterized in that: Based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning temporal rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, spatial and temporal features are extracted. Through weighted fusion, a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends is generated, including: Obtain lightning trend prediction feature sets, cloud cover-lightning vertical layer coupling feature sets, and cloud cover-lightning time rhythm coupling feature sets, and normalize the feature sets using a standardization processing method to obtain a normalized feature set; Based on the normalized feature set, a convolutional neural network is used to process the vertical layered cloud cover data, extract the spatial distribution features, and obtain the spatial feature set. According to the normalized feature set, the long short-term memory network is used to process the temporal rhythm coupling data, extract the temporal dynamic features, and obtain the temporal feature set; If the dimensions of the spatial feature set and the temporal feature set are consistent, the weighted fusion method is used to integrate the spatial feature set and the temporal feature set to obtain a fused feature set; if the dimensions are inconsistent, linear interpolation is performed on the feature set with lower dimensions to obtain a fused feature set; Based on the fusion feature set and combined with the pre-established lightning-related cloud type labels, a fully connected neural network is used for classification processing to obtain the spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends.

10. A lightning approach trend forecasting system, characterized in that: Based on the method according to any one of claims 1 to 9, the system comprises: The first feature set module is used to obtain the cloud space distribution and time series, extract the first cloud texture feature vector, and generate a cloud layer feature matrix; reduce the dimension of the first cloud texture feature vector to generate a second cloud texture feature vector, and combine it with the cloud layer feature matrix to generate a regionalized cloud distribution feature set through clustering; The second feature set module is used to extract humidity, temperature, and wind field data to generate a lightning trigger factor dataset. It uses time series analysis and optical flow methods to analyze the regional cloud distribution feature set to form a cloud dynamic feature set. The third feature set module is used to generate a lightning intensity level classification feature set based on the cloud cover dynamic feature set and the cloud cover layer feature matrix, combined with the pre-established lightning-related cloud type labels; The fourth feature set module is used to calculate the coupling weight coefficient of cloud vertical stratification and lightning intensity classification based on the lightning intensity level classification feature set and the cloud layer feature matrix, and generate a cloud-lightning vertical stratification coupling feature set; The fifth feature set module is used to generate a lightning occurrence probability distribution and a lightning spatial location coordinate sequence using a time series analysis method based on the lightning intensity level classification feature set, cloud cover dynamic feature set, and cloud cover-lightning vertical layer coupling feature set, combined with lightning monitoring data, to form a lightning probability distribution feature set; The sixth feature set module is used to generate a lightning occurrence probability time series and a lightning time window range series using a time series analysis method based on the lightning probability distribution feature set and the cloud cover periodicity feature vector, thereby forming a lightning time window prediction feature set; The seventh feature set module is used to calculate the rhythm matching degree between the cloud cover periodic fluctuation and the lightning time window based on the cloud cover periodic feature vector and the lightning time window range sequence, and generate the cloud cover-lightning time rhythm coupling feature set; An eighth feature set module is used to generate a lightning movement path sequence and a lightning duration sequence using a time series analysis method based on the lightning time window prediction feature set and the lightning trigger factor dataset, thereby forming a lightning trend prediction feature set; The ninth feature set module is used to extract spatial and temporal features based on the lightning trend prediction feature set, the cloud cover-lightning vertical layer coupling feature set, and the cloud cover-lightning time rhythm coupling feature set, combined with pre-established lightning-related cloud type labels, and generate a spatiotemporal coupling feature set of cloud cover dynamic evolution and lightning trends through weighted fusion; The prediction module is used to obtain the dynamic evolution pattern of lightning trends by performing cluster analysis on the spatiotemporal coupling feature set; based on the dynamic evolution pattern, the lightning trend is predicted using the time series analysis method to obtain the final lightning trend prediction result.

Images