Extreme weather event prediction method and system based on intelligent model, and storage medium

Abstract

The invention relates to the technical field of meteorological data processing and artificial intelligence prediction, and discloses an extreme weather event prediction method based on an intelligent model. The method comprises the steps of collecting multi-source meteorological data, and performing preprocessing to unify time and space references of the data; extracting spatio-temporal features reflecting meteorological element change rules based on a preprocessing result, and forming feature vectors for training a deep learning model; inputting fusion data in the real-time platform according to a sliding time window for prediction, generating early warning information according to a prediction result and a set threshold value, and issuing the early warning information through multiple channels; and early warning and monitoring results are fed back to a feature extraction and model updating link to realize online updating and closed-loop processing. According to the method, the timeliness and accuracy of extreme weather event recognition can be improved on the basis of ensuring data consistency through multi-source meteorological data fusion and spatial-temporal feature modeling in combination with conjoint analysis of time and spatial information by deep learning.

Description

Technical Field

[0001] The present invention relates to the field of meteorological data processing and artificial intelligence prediction technology, and in particular to an extreme weather event prediction method, system and storage medium based on an intelligent model. Background Art

[0002] As global climate change intensifies, the frequency and impact of extreme weather events are increasing. Traditional weather forecasting methods rely heavily on numerical weather prediction models, which use meteorological observation data to drive atmospheric dynamics equations for calculations. However, due to limitations in model resolution, boundary conditions, and computing resources, the accuracy and timeliness of forecasts struggle to meet the demands of real-time warnings.

[0003] In recent years, meteorological observation methods have continued to expand. Multi-source observation systems, including weather stations, satellite remote sensing, ground-based sensors, and radar, can provide a vast amount of meteorological data across diverse temporal and spatial scales. However, differences in sampling frequency, spatial resolution, and observation methods among these data sources lead to inconsistent data formats, inconsistent temporal and spatial benchmarks, and the presence of missing or outliers, thus increasing the complexity of data fusion and processing.

[0004] Artificial intelligence methods, particularly deep learning models, have demonstrated strong capabilities in feature extraction and pattern recognition in fields such as image recognition and time series forecasting. Some research has attempted to apply these methods to meteorological forecasting. However, in extreme weather scenarios, these methods still face challenges such as difficulty integrating multi-source data, insufficient spatiotemporal feature extraction, limited model adaptability to real-time data changes, and the need to effectively connect warning information with forecast results. Therefore, how to fully utilize multi-source meteorological data, effectively extract key spatiotemporal features, dynamically update models, and accurately distribute warning information has become a pressing technical challenge. Summary of the Invention

[0005] The present invention provides an extreme weather event prediction method based on an intelligent model to solve the problems in the prior art of difficulty in fusing multi-source meteorological data, insufficient extraction of spatiotemporal features, and poor adaptability of prediction models to real-time data.

[0006] In a first aspect, the present invention provides a method for predicting extreme weather events based on an intelligent model, the method comprising: Collect and preprocess multi-source meteorological data to obtain preprocessed meteorological data; Perform feature extraction and feature processing based on the preprocessed meteorological data to obtain a feature vector; Building and training an intelligent prediction model for extreme weather event prediction based on the feature vector; Performing reasoning on the data to be predicted by the intelligent prediction model to obtain a prediction result, and updating the model parameters of the intelligent prediction model based on the difference between the actual observed data and the prediction result; Whether to issue an early warning is determined based on the prediction result and the preset probability threshold. If it is determined to be issued, early warning information is generated and distributed.

[0007] Optionally, collecting and preprocessing multi-source meteorological data to obtain preprocessed meteorological data includes: Acquire multi-source meteorological data from weather stations, meteorological satellites, ground observation sensors and radars; Check the integrity of the multi-source meteorological data and perform data cleaning to process missing data, erroneous data or abnormal data; Align multi-source meteorological data from different sources by timestamp and spatial reference and fuse them into a unified data format; Normalization is performed on the fused data to obtain preprocessed meteorological data.

[0008] Optionally, performing feature extraction and feature processing based on the preprocessed meteorological data to obtain a feature vector includes: performing time series analysis based on the preprocessed meteorological data to identify time series features; performing spatial analysis based on the pre-processed meteorological data to identify spatial features, wherein the spatial analysis utilizes a geographic information system; Analyze the identified features using statistical or machine learning methods and select those associated with extreme weather events; Perform data conversion or dimensionality reduction on the filtered features to obtain feature vectors; Cross-validation techniques are used to verify feature stability and test feature recognition effectiveness in real-time data streams.

[0009] Optionally, the method further includes: The features obtained by the screening are subjected to feature conversion and feature encoding as training samples, and the training samples are divided into a training set, a validation set and a test set according to a predetermined ratio; When the ratio of the number of target class samples representing extreme weather events to the number of non-target class samples in the training set is lower than a predetermined threshold, a synthetic minority class oversampling technique is used to generate supplementary samples, and the supplementary samples are merged with the training samples for offline initial training of the model; The training sample division is verified by using a cross-validation or an equivalent validation method.

[0010] Optionally, constructing and training an intelligent prediction model for extreme weather event prediction based on the feature vector includes: Constructing an intelligent prediction model to be trained, wherein the intelligent prediction model to be trained adopts a network structure in which a convolutional neural network processes spatial information and a recurrent neural network processes time series information; Input the feature vector into the intelligent prediction model to be trained, select a loss function suitable for probability or classification output, and use a batch training strategy to optimize parameters; During the training process, the training status of the validation set is monitored, and the training process is adjusted until the training convergence conditions are met to obtain an intelligent prediction model for extreme weather event prediction.

[0011] Optionally, performing reasoning on the data to be predicted by the intelligent prediction model to obtain a prediction result, and updating model parameters of the intelligent prediction model based on a difference between actual observed data and the prediction result, includes: Receive actual observation data through a data stream processing system; Performing window processing on the pre-processed meteorological data according to a configured time window, inputting the windowed feature vector into the intelligent prediction model, and outputting a prediction result; Compare the output prediction results with the actual observed data and record the difference; Based on the difference value and the newly added actual observation data, the model parameters of the intelligent prediction model are updated through an online learning training method.

[0012] Optionally, determining whether to issue a warning based on the prediction result and a preset probability threshold, and if so, generating and distributing warning information, includes: Determine the warning level and warning information based on the prediction result and the preset probability threshold, wherein the warning information includes the warning level, event type, predicted impact area and corresponding map layer information; Distributing the warning information to a target receiving end through at least one distribution channel; The sending status of each distribution channel is recorded and the model parameters of the intelligent prediction model are updated through the feedback monitoring data.

[0013] In a second aspect, the present invention provides an extreme weather event prediction system based on an intelligent model, the system comprising: The data collection and preprocessing module is used to collect and preprocess multi-source meteorological data to obtain preprocessed meteorological data; A feature extraction and processing module, configured to perform feature extraction and feature processing based on the preprocessed meteorological data to obtain a feature vector; A training model module, configured to construct and train an intelligent prediction model for extreme weather event prediction based on the feature vector; A prediction and update module, configured to obtain a prediction result by performing reasoning on the prediction data through the intelligent prediction model, and to update the model parameters of the intelligent prediction model based on the difference between the actual observation data and the prediction result; The early warning module is used to determine whether to issue an early warning based on the prediction results and a preset probability threshold. If it is determined to be issued, the early warning information is generated and distributed.

[0014] In a third aspect, an embodiment of the present invention provides a terminal device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described above when executing the computer program.

[0015] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0016] The above technical solution provided by the embodiment of the present invention has the following advantages over the prior art: the present invention reduces information bias caused by data sampling differences by collecting meteorological data from multiple sources such as weather stations, satellites, ground sensors and radars, and performing fusion processing under the condition of unified time and space benchmarks. In the feature processing link, feature data that can reflect the spatiotemporal variation laws of meteorological elements are extracted based on the preprocessing results, and feature vectors are generated for model training, which helps to improve the model's ability to characterize the occurrence patterns of extreme weather events. By constructing a deep learning model that can simultaneously process spatial information and time series information, and performing incremental updates in combination with real-time data, the model's adaptability to changes in meteorological conditions is improved. In the prediction and warning link, warning information distributed through multiple channels is generated based on the prediction results and set thresholds, and the distribution and monitoring results are fed back to the feature extraction and model update link, realizing closed-loop processing of prediction, warning and model update, thereby technically improving the deficiencies of existing extreme weather prediction methods in data fusion, feature extraction and dynamic update. BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0017] 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, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] One or more embodiments are exemplarily illustrated by pictures in the corresponding drawings. These exemplifications do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings are represented as similar elements. Unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0019] Figure 1 A schematic diagram of a process for predicting extreme weather events based on an intelligent model provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of an extreme weather event prediction system based on an intelligent model provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of the structure of a computer-readable storage medium provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0021] The disclosure below provides many different embodiments or examples for implementing different configurations of the present invention. To simplify the disclosure of the present invention, the components and configurations of specific examples are described below. Of course, these are merely examples and are not intended to limit the present invention. In addition, the present invention may repeat reference numerals and / or letters in different examples. Such repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or configurations discussed.

[0022] Figure 1 An embodiment of the present invention provides a method for predicting extreme weather events based on an intelligent model, the method comprising: S100, collecting and preprocessing multi-source meteorological data to obtain preprocessed meteorological data.

[0023] Collect and preprocess multi-source real-time meteorological data, which includes meteorological station observation data, satellite remote sensing data, ground observation sensor data and radar echo data; the preprocessing includes performing data cleaning on the original observation data to identify and correct or eliminate missing, erroneous or abnormal data points, aligning the data from each source according to timestamp and spatial reference and fusing them into a unified data representation, and performing standardization or normalization on the fusion results to obtain preprocessed meteorological data for subsequent processing.

[0024] S200 , performing feature extraction and feature processing based on the preprocessed meteorological data to obtain a feature vector.

[0025] Based on the pre-processed meteorological data obtained in S100, feature extraction and feature processing are performed, time series analysis is performed on the data from the time dimension to obtain temporal features, and spatial analysis is performed on the data from the spatial dimension to obtain spatial features; statistical methods or machine learning methods are used to perform correlation analysis and importance evaluation on candidate features to screen features associated with extreme weather events; data conversion or dimensionality reduction processing is performed on the screened features according to predetermined rules to form feature vectors for model input.

[0026] S300: Constructing and training an intelligent prediction model for extreme weather event prediction based on the feature vector.

[0027] An intelligent prediction model is constructed and trained based on the feature vector obtained in S200. The model adopts a network structure that can simultaneously process spatial information and time series information. The feature vector is used as the model input and the model parameters are batch trained according to the training configuration. During the training process, the training state is monitored by dividing and using a training set and a validation set. The training configuration and sample division are verified by cross-validation or an equivalent method until the training convergence judgment is met. After the training is completed, the model is deployed on a real-time prediction platform.

[0028] S400: performing reasoning on the data to be predicted by the intelligent prediction model to obtain a prediction result, and updating the model parameters of the intelligent prediction model based on the difference between the actual observation data and the prediction result.

[0029] The deployed model in S300 performs reasoning on the predicted data to obtain a prediction result; the prediction result is compared with the actual observation data subsequently obtained to calculate the prediction difference and record the error log; when the model parameters need to be updated, the model parameters are updated by online learning or incremental training based on the newly added observation data, and the updated model is used for subsequent real-time reasoning.

[0030] S500: Determine whether to issue an early warning based on the prediction result and a preset probability threshold. If it is determined to be issued, generate and distribute early warning information.

[0031] Whether to issue an early warning is determined based on the prediction results of S400 and the preset probability threshold. If it is determined to be issued, an early warning package containing information such as the warning level, event type, and estimated impact area is generated and distributed to the target receiving end through distribution channels such as SMS, mobile application push, or network platform; the sending status of each distribution channel is recorded and the sending records and monitoring data are fed back to the feature extraction and model update links of S200 and S400 to support subsequent processing.

[0032] This method can ensure data consistency in time and space by uniformly preprocessing multi-source real-time meteorological data; combine feature extraction and feature processing to convert meteorological change trends and spatial distribution characteristics into feature vectors that can be used for model training, thereby achieving structuring and information compression of input data; use the feature vectors to train an intelligent prediction model, enabling the model to predict extreme weather events, and improve the model's adaptability to real-time meteorological changes by updating parameters based on actual observation data; automatically generate and distribute warning information based on prediction results and preset thresholds, realizing full-process automation from data acquisition, model prediction to information release.

[0033] Furthermore, the collecting and preprocessing of multi-source meteorological data to obtain preprocessed meteorological data includes: S101, obtains multi-source meteorological data from weather stations, meteorological satellites, ground observation sensors and radars; S102, checking the integrity of the multi-source meteorological data and performing data cleaning to process missing data, erroneous data or abnormal data; S103, aligning multi-source meteorological data from different sources according to timestamps and spatial references, and fusing them into a unified data format; S104: performing normalization processing on the fused data to obtain pre-processed meteorological data.

[0034] First, observational data on meteorological elements such as temperature, humidity, air pressure, wind speed, and wind direction are collected from weather stations. Missing items in the collected weather station data are detected and supplemented by temporal interpolation using historical observations of similar types. Furthermore, a statistical threshold range is set to remove abnormal data records that exceed this range, thereby obtaining a dataset that conforms to statistical rules.

[0035] Subsequently, satellite remote sensing data, including cloud maps, surface temperature distribution, and precipitation inversion results, were obtained and resampled using bilinear interpolation according to the preset latitude and longitude grids to keep their spatial resolution consistent with that of the meteorological station data, and then converted into a unified projection coordinate system.

[0036] Observation values ​​such as temperature, humidity, rainfall, and wind speed are collected from ground sensor nodes and synchronized with a unified timestamp to ensure the alignment of multi-point sensor data in the time dimension; the synchronized data is formatted and converted to form a record format consistent with the weather station data structure.

[0037] In terms of processing radar echo data, time interpolation is performed on data from different scanning periods in the order of observation time to fill in the missing time points; then the echo intensity values ​​are normalized so that the echo intensity values ​​of different radar sites are at the same level, which is convenient for subsequent analysis.

[0038] Finally, the processed weather station data, satellite remote sensing data, ground sensor data, and radar echo data are fused on a unified time axis and spatial grid to generate a multi-dimensional meteorological data tensor. The minimum-maximum normalization operation is performed on the tensor to form a preprocessed meteorological data set that can be directly input into the feature extraction step.

[0039] This embodiment specifically describes the implementation details of data collection and preprocessing. The system connects to meteorological stations to achieve continuous collection of point data, and periodically obtains satellite remote sensing images to obtain large-scale cloud field information. At the same time, it accesses ground and ocean sensors and radar echoes to supplement local environmental measurements. All accessed data are first timestamped and coordinate standardized, and the physical units are consistent at the aggregation layer. Then, in the quality control stage, missing points, noise, and outliers are identified.

[0040] Short-term missing data are filled by interpolation methods based on neighboring sites or previous and subsequent time series. Outliers are corrected by quantiles based on historical distribution or marked as pending data. Then, data from different sources are spatiotemporally aligned and fused with time and geographic location as key fields to obtain a unified data set. Finally, the fused numerical features are standardized to eliminate scale differences caused by different dimensions, ensuring consistency and comparability of the data used for subsequent feature extraction and model training.

[0041] By uniformly formatting, verifying the integrity and quality, aligning spatiotemporally, and standardizing the numerical values ​​of multi-source observation data, this method significantly improves the comparability and reliability of the data used for training and inference, reduces model misjudgments and instability caused by data noise or format differences, and thus provides a more solid basic data guarantee for subsequent feature engineering and model training.

[0042] Furthermore, performing feature extraction and feature processing based on the pre-processed meteorological data to obtain a feature vector includes: S201, performing time series analysis based on the preprocessed meteorological data to identify time series features; S202, performing spatial analysis based on the pre-processed meteorological data to identify spatial features, wherein the spatial analysis utilizes a geographic information system; S203, analyzing the identified features using a statistical method or a machine learning method, and screening features associated with extreme weather events; S204, performing data conversion or dimensionality reduction processing on the screened features to obtain feature vectors; S205, cross-validation technology is used to verify feature stability and test the effectiveness of feature recognition in real-time data streams.

[0043] First, based on the preprocessed meteorological data, the moving average of meteorological elements such as temperature, humidity, air pressure, wind speed and precipitation is calculated according to the dynamically adjusted time window to reflect the overall change trend of each meteorological element within the time window. At the same time, the rolling standard deviation is calculated to characterize the fluctuation amplitude within this time period.

[0044] For radar echo data, adjacent spatial grid cells are selected to perform differential operations to obtain spatial gradients, and differential operations are performed on continuous time series frames to obtain temporal gradients to describe the cloud structure and the movement direction and intensity changes of precipitation areas.

[0045] The frequency of similar extreme weather events in the corresponding seasons and months is statistically analyzed, and events in the recent time period are given higher weights to form a time-attenuated weighted event frequency feature.

[0046] After obtaining the above features, feature importance evaluation is performed on all candidate features to screen feature data with high correlation with extreme weather events and eliminate redundant features with low correlation.

[0047] Finally, the filtered high-dimensional feature data is input into the principal component analysis module to perform dimensionality reduction operations to generate a low-dimensional feature vector containing temporal and spatial variation features, and the feature vector is provided for subsequent model construction and training steps.

[0048] This embodiment focuses on feature extraction and processing. In the temporal dimension, time series analysis is performed on each measurement sequence to identify long-term trends, seasonal cycles, short-term fluctuations, and mutation points. In the spatial dimension, geographic information processing is used to map point observations to a regular grid and spatial analysis is performed on satellite or radar images to extract spatial features such as cloud morphology, precipitation coverage, and neighborhood gradients.

[0049] In the candidate feature screening stage, statistical correlation tests are used to evaluate the linear or nonlinear correlation between each feature and the target event, and an integrated learning method composed of multiple decision trees such as random forest is used to evaluate the importance of the feature. The so-called random forest feature importance is determined by observing the average contribution of a certain feature to the node purity when constructing multiple trees or the degree of degradation of model performance after disrupting the feature, thereby screening out the features that are most helpful for predicting extreme events.

[0050] For feature sets with significant collinearity or excessively high dimensions, linear dimensionality reduction methods such as principal component analysis are used to project the original features onto several principal components that can explain most of the variance to obtain a compact representation, or nonlinear representation learning methods such as autoencoders are used to reduce the dimension while retaining key information. Finally, the selected features are uniformly encoded and standardized and combined into feature vectors acceptable to the model. Subsequently, cross-validation is used to verify the stability and generalization ability of these features in different time periods and regions.

[0051] The use of parallel feature extraction in the time and space domains and a rigorous feature screening and verification process can extract highly relevant and stable signals related to extreme weather from massive heterogeneous data. Through statistical and machine learning methods, importance assessment and dimensionality reduction of redundant or collinear features can be performed, which significantly improves the model's efficiency in utilizing useful information and generalization ability, reduces the risk of overfitting and improves robustness in practical applications.

[0052] Furthermore, the method further comprises: S206, performing feature conversion and feature encoding on the screened features as training samples, and dividing the training samples into a training set, a validation set, and a test set according to a predetermined ratio; S207, when the ratio of the number of target class samples representing extreme weather events to the number of non-target class samples in the training set is lower than a predetermined threshold, generating supplementary samples using a synthetic minority class oversampling technique, and merging the supplementary samples with the training samples for offline initial training of the model; S208: Verify the training sample division using a cross-validation or equivalent validation method.

[0053] First, during the offline training phase, data augmentation is performed on categories with fewer extreme weather event samples in the historical meteorological dataset.

[0054] Data augmentation involves performing interpolation on the original time series data to generate additional sample sequences that conform to the original trend. This process then introduces small, statistically distributed noise into the generated sequence to create a diverse set of enhanced samples. These enhanced samples are then merged with the original sample set for initial model training.

[0055] After the model is put into real-time operation, the system continuously records information such as prediction results, input data characteristics, and actual observation values, and regularly analyzes the time distribution characteristics of the accumulated data. Based on the analysis results, the length and step size of the sliding time window and related parameter configurations in the feature extraction process are adjusted to maintain the stability of data processing and model prediction.

[0056] For feature sets with significant collinearity or too high dimensions, linear dimensionality reduction methods such as principal component analysis are used to project the original features onto several principal components that can explain most of the variance to obtain a compact representation, or nonlinear representation learning methods such as autoencoders are used to reduce the dimension while maintaining key information.

[0057] Finally, the selected features are uniformly encoded and standardized and combined into feature vectors acceptable to the model. Subsequently, the stability and generalization ability of these features in different time periods and regions are verified through cross-validation.

[0058] This embodiment illustrates the training sample construction and sample imbalance processing strategy. First, the samples after feature conversion and encoding are divided into training set, validation set and test set according to a predetermined ratio for model training and performance evaluation.

[0059] When the number of positive samples representing extreme events in historical samples is obviously insufficient, data augmentation methods are used to expand the positive sample set. One method is based on the sliding window expansion of time series, which generates multiple samples that still retain time dependence by intercepting the time window with different starting points.

[0060] Another commonly used method is the synthetic minority class oversampling technology. This technology generates new synthetic samples by interpolating between minority class samples and their neighbors in the feature space to increase the diversity and support of the minority class. The amplified samples are merged with the original samples for offline initial training. At the same time, cross-validation or equivalent resampling validation schemes are used in the process of sample division and enhancement to ensure the representativeness of each category in each fold and the reliability of the evaluation results, thereby improving the model's recognition performance of rare extreme events.

[0061] Introducing a supplementary strategy based on the expansion of time series sliding windows and the synthesis of minority class samples in the construction of training samples can effectively alleviate the class imbalance problem caused by the scarcity of extreme event samples, improve the model's ability to identify rare but high-risk extreme events, and enable the trained model to still maintain a good detection rate and stability in actual small-sample scenarios.

[0062] Furthermore, the intelligent prediction model for extreme weather event prediction is constructed and trained based on the feature vector, including: S301, constructing an intelligent prediction model to be trained, wherein the intelligent prediction model to be trained adopts a network structure in which a convolutional neural network processes spatial information and a recurrent neural network processes time series information; S302: Input the feature vector into the intelligent prediction model to be trained, select a loss function suitable for probability or classification output, and use a batch training strategy to optimize parameters; S303, monitoring the training status of the validation set during the training process, and adjusting the training process until the training convergence condition is met, thereby obtaining an intelligent prediction model for extreme weather event prediction.

[0063] First, based on the feature vector generated by the feature extraction step, a spatiotemporal hybrid deep learning model consisting of a convolutional neural network and a recurrent neural network structure is constructed.

[0064] In the spatial feature extraction stage, the spatial dimension data of the input feature vector is input into the two-dimensional convolution module, and convolution operations are performed in the spatial neighborhood through multi-layer convolution kernels to extract the spatial correlation features of the local area. Nonlinear activation and pooling operations are performed on the convolution output to reduce feature redundancy and retain key spatial patterns.

[0065] In the time feature extraction stage, the output results of the convolution processing are spliced ​​with the original time dimension features in the time series order and input into the bidirectional long short-term memory network module, which performs sequence state update operations in the forward and backward time directions respectively to capture the long-term dependencies and short-term change characteristics in the time series.

[0066] During the model training phase, the loss value between the predicted output and the true label is calculated based on the labeled categories of the training set samples, gradient backpropagation is performed, and the network weights are iteratively updated using the optimization algorithm.

[0067] During the training process, the prediction accuracy of the model in different training rounds is monitored through the validation set, and the hyperparameters such as the learning rate of the optimization algorithm are adjusted accordingly until the performance indicators of the model on the validation set reach the convergence conditions, obtaining a spatiotemporal hybrid deep learning model that can be used for extreme weather event prediction.

[0068] This embodiment provides a specific explanation of model construction and training, and adopts a deep learning design with spatiotemporal collaboration to model spatial information and time series information in parallel. The spatial input comes from satellite images, radar or gridded observation data and is sent to the convolutional network to learn local spatial patterns and texture features. The convolutional network extracts local feature representations with translation invariance on the input through sliding convolution kernels. The time series input is sent to the recurrent network branch to capture historical evolution and time correlation. When selecting a recurrent network, structures with gating mechanisms such as long short-term memory networks or gated recurrent units are preferably used. They maintain or forget information in long sequences through built-in gating units, thereby alleviating the gradient vanishing or forgetting problems that occur in ordinary recurrent networks when learning long-term dependencies.

[0069] The representations extracted from the spatial branches and the temporal branches are fused into a comprehensive representation in the backend and used for classification or probabilistic prediction. During the training phase, batch optimization is performed according to the pre-divided training set and validation set. A loss function matching the classification or probabilistic output is used, and the validation set indicators are monitored during training and an early stopping strategy is implemented to avoid overfitting. After training is completed, the model parameters and the corresponding feature processing pipeline are saved to ensure the reproducibility of online reasoning.

[0070] The spatiotemporal collaborative model structure uses a convolutional network to process spatial information and combines it with a recurrent network to process time series. This enables the model to simultaneously learn local spatial patterns and temporal evolution laws of cloud fields, radar echoes, etc., thereby more comprehensively characterizing the generation and evolution mechanism of extreme weather events, improving prediction accuracy and improving the ability to distinguish different types of extreme events and generalization performance.

[0071] Furthermore, performing reasoning on the data to be predicted by the intelligent prediction model to obtain a prediction result, and updating the model parameters of the intelligent prediction model based on the difference between the actual observed data and the prediction result, includes: S401, receiving actual observation data through the data stream processing system; S402, performing window processing on the pre-processed meteorological data according to a configured time window, inputting the windowed feature vector into the intelligent prediction model, and outputting a prediction result; S403, comparing the output prediction result with the actual observation data, and recording the difference value; S404: Based on the difference value and the newly added actual observation data, update the model parameters of the intelligent prediction model through an online learning training method.

[0072] First, the latest preprocessed meteorological data is divided into multiple batches based on a sliding time window. Each batch contains a temporally continuous segment of meteorological data. Each batch of data is sequentially fed into the trained prediction model, and inference calculations are performed to obtain a predicted probability of extreme weather events for that batch.

[0073] When a predicted value exceeds the set probability threshold, the prediction results of the batch are recorded. When the actual observation data received later arrives, the predicted results are compared with the actual observation values ​​and the difference between the two is calculated.

[0074] Based on the calculated difference values, the newly collected observations are weighted and proportionally integrated with the original training dataset to construct an updated dataset containing the latest observations. This dataset is then used for mini-batch training to update the model parameters. The updated prediction model is then used for inference calculations in subsequent batches, enabling the prediction capability to continuously adapt and improve with new data.

[0075] This embodiment illustrates the implementation method of online reasoning and dynamic model updating. During real-time operation, the system accesses and preprocesses real-time observation data through the data stream processing platform, and windowed the data according to the configured time window to form time segment features for reasoning. The so-called windowing is to cut continuous data segments of fixed length on the time axis to construct samples with time context. After windowing, each sample performs feature extraction and conversion consistent with offline and is sent to the deployed model for real-time prediction.

[0076] The system continuously compares the model output probabilities with the true labels subsequently retrieved from the observation system to calculate prediction errors and difference trends. Based on these differences, the model parameters are updated using online learning or incremental training. Online learning refers to updating the model weights in small steps with newly arrived labeled samples without retraining the entire model from scratch, enabling the model to quickly respond to changes in climate patterns or observation conditions. After the update, the model performance is verified to ensure that no degradation occurs, and then the verified updates are deployed to the online inference environment.

[0077] By implementing windowed processing in real-time data streams and combining it with online learning or incremental training mechanisms, the system can quickly respond to emerging climate patterns or changes in observation conditions while maintaining real-time reasoning capabilities, reducing the risk of model performance degradation over time and ensuring continued prediction accuracy and timeliness in dynamic environments.

[0078] Furthermore, the step of determining whether to issue an early warning based on the prediction result and a preset probability threshold, and if so, generating and distributing early warning information, includes: S501, determining a warning level and warning information based on the prediction result and a preset probability threshold, wherein the warning information includes the warning level, event type, predicted impact area, and corresponding map layer information; S502, distributing the warning information to a target receiving end through at least one distribution channel; S503, recording the sending status of each distribution channel and updating the model parameters of the intelligent prediction model through the feedback monitoring data.

[0079] First, based on the probability of extreme weather events output by the prediction model, the warning level is divided into multiple levels, and different levels correspond to different information content templates.

[0080] Subsequently, the warning information content is generated based on the event type, forecast level and geographic location information. The information includes event category description, warning level identification, geographic boundary data of the potentially affected area, and regional layers generated based on meteorological data.

[0081] The generated warning information is sent through multiple distribution channels in a predetermined order, including SMS gateways to registered user terminals, application push services to transmit graphic and text messages to mobile clients, and event summary information released through social platform interfaces.

[0082] During the information distribution process, the sending status of each channel is recorded, and entries that fail to be sent are retried or forwarded using alternative channels to ensure that the warning information can reach the intended recipients.

[0083] This embodiment concretizes the early warning determination and distribution process. The system determines whether to trigger an early warning based on the predicted probability given by the model and the preset probability threshold, and generates structured early warning information when the trigger is determined. This information clearly classifies the type and severity of the event and describes the spatial impact area covered by the prediction. It also provides a visual map layer for superimposing radar or satellite observation images and recommended disposal measures for emergency response.

[0084] The generated warning information is sent to the designated receiving end through one or more distribution channels such as SMS, email, in-app push or docking interface with the emergency management system. At the same time, the system records the sending status of each message in each channel and collects feedback data from the monitoring end and the audience end. The collected feedback data and the sending log are used as input for model retraining and warning threshold adjustment, forming a closed-loop system of information release, effect monitoring and strategy optimization.

[0085] Combining the model output with the preset probability threshold to generate a graded warning, and distributing the warning message through multiple channels with the spatial impact area and map layer, not only improves the comprehensibility and operability of the warning information, but also incorporates the warning dissemination effect into the monitoring loop by recording the sending status and collecting feedback, thereby achieving continuous optimization of the warning quality and dissemination strategy.

[0086] like Figure 2 As shown, Figure 2 A schematic diagram of the structure of an extreme weather event prediction system based on an intelligent model provided in an embodiment of the present invention, the system comprising: The data collection and preprocessing module 610 is used to collect and preprocess multi-source meteorological data to obtain preprocessed meteorological data; A feature extraction and processing module 620 is configured to perform feature extraction and feature processing based on the pre-processed meteorological data to obtain a feature vector; A training model module 630 is used to construct and train an intelligent prediction model for extreme weather event prediction based on the feature vector; The prediction and update module 640 is configured to obtain a prediction result by performing reasoning on the prediction data through the intelligent prediction model, and update the model parameters of the intelligent prediction model based on the difference between the actual observation data and the prediction result; The warning module 650 is used to determine whether to issue a warning based on the prediction result and a preset probability threshold, and if it is determined to be issued, generate and distribute warning information.

[0087] like Figure 3 As shown, Figure 3 Schematic diagram of the structure of the computer-readable storage medium provided by an embodiment of the present invention. The computer-readable storage medium 700 of this embodiment includes: a server 710 ( Figure 3 Only one is shown), a client 720 and a data recovery program 721 stored in the client 720 and executable on the at least one client 720, the client 720 executes the data recovery program 721 to send a request to the server 710, and the server 710 feeds back a result to implement the steps in the above method embodiment.

[0088] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0089] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0090] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0091] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0092] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0093] If the integrated module / unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, and software distribution medium. It should be noted that the content of the computer-readable medium can be appropriately increased or decreased based on the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, based on legislation and patent practice, computer-readable media does not include electric carrier signal.

[0094] The present invention may implement all or part of the processes in the above-mentioned method embodiments, and may also be completed through a computer program product. When the computer program product runs on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0095] The above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may be modified or some of the technical features thereof may be replaced by equivalents; and such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention and are therefore intended to be included within the scope of protection of the present invention.

Claims

1. A method for predicting extreme weather events based on an intelligent model, characterized in that: The method comprises: Collect and preprocess multi-source meteorological data to obtain preprocessed meteorological data; Perform feature extraction and feature processing based on the preprocessed meteorological data to obtain a feature vector; Building and training an intelligent prediction model for extreme weather event prediction based on the feature vector; Performing reasoning on the data to be predicted by the intelligent prediction model to obtain a prediction result, and updating the model parameters of the intelligent prediction model based on the difference between the actual observed data and the prediction result; Whether to issue an early warning is determined based on the prediction result and the preset probability threshold. If it is determined to be issued, early warning information is generated and distributed.

2. The method according to claim 1, characterized in that The collecting and preprocessing of multi-source meteorological data to obtain preprocessed meteorological data includes: Acquire multi-source meteorological data from weather stations, meteorological satellites, ground observation sensors and radars; Check the integrity of the multi-source meteorological data and perform data cleaning to process missing data, erroneous data or abnormal data; Align multi-source meteorological data from different sources by timestamp and spatial reference and fuse them into a unified data format; Normalization is performed on the fused data to obtain preprocessed meteorological data.

3. The method according to claim 2, characterized in that The performing feature extraction and feature processing based on the pre-processed meteorological data to obtain a feature vector includes: performing time series analysis based on the preprocessed meteorological data to identify time series features; performing spatial analysis based on the pre-processed meteorological data to identify spatial features, wherein the spatial analysis utilizes a geographic information system; Analyze the identified features using statistical or machine learning methods and select those associated with extreme weather events; Perform data conversion or dimensionality reduction on the filtered features to obtain feature vectors; Cross-validation techniques are used to verify feature stability and test feature recognition effectiveness in real-time data streams.

4. The method according to claim 3, characterized in that The method further comprises: The features obtained by the screening are subjected to feature conversion and feature encoding as training samples, and the training samples are divided into a training set, a validation set and a test set according to a predetermined ratio; When the ratio of the number of target class samples representing extreme weather events to the number of non-target class samples in the training set is lower than a predetermined threshold, a synthetic minority class oversampling technique is used to generate supplementary samples, and the supplementary samples are merged with the training samples for offline initial training of the model; The training sample division is verified by using a cross-validation or an equivalent validation method.

5. The method according to claim 4, characterized in that The intelligent prediction model for extreme weather event prediction is constructed and trained based on the feature vector, including: Constructing an intelligent prediction model to be trained, wherein the intelligent prediction model to be trained adopts a network structure in which a convolutional neural network processes spatial information and a recurrent neural network processes time series information; Input the feature vector into the intelligent prediction model to be trained, select a loss function suitable for probability or classification output, and use a batch training strategy to optimize parameters; During the training process, the training status of the validation set is monitored, and the training process is adjusted until the training convergence conditions are met to obtain an intelligent prediction model for extreme weather event prediction.

6. The method according to claim 5, characterized in that The step of performing reasoning on the data to be predicted by the intelligent prediction model to obtain a prediction result, and updating the model parameters of the intelligent prediction model based on the difference between the actual observation data and the prediction result, includes: Receive actual observation data through a data stream processing system; Performing window processing on the pre-processed meteorological data according to a configured time window, inputting the windowed feature vector into the intelligent prediction model, and outputting a prediction result; Compare the output prediction results with the actual observed data and record the difference; Based on the difference value and the newly added actual observation data, the model parameters of the intelligent prediction model are updated through an online learning training method.

7. The method according to claim 6, characterized in that The determining whether to issue an early warning based on the prediction result and a preset probability threshold, and if so, generating and distributing early warning information, includes: Determine the warning level and warning information based on the prediction result and the preset probability threshold, wherein the warning information includes the warning level, event type, predicted impact area and corresponding map layer information; Distributing the warning information to a target receiving end through at least one distribution channel; The sending status of each distribution channel is recorded and the model parameters of the intelligent prediction model are updated through the feedback monitoring data.

8. An extreme weather event prediction system based on an intelligent model, characterized in that: The system comprises: The data collection and preprocessing module is used to collect and preprocess multi-source meteorological data to obtain preprocessed meteorological data; A feature extraction and processing module, configured to perform feature extraction and feature processing based on the preprocessed meteorological data to obtain a feature vector; A training model module, configured to construct and train an intelligent prediction model for extreme weather event prediction based on the feature vector; A prediction and update module, configured to obtain a prediction result by performing reasoning on the data to be predicted through the intelligent prediction model, and to update the model parameters of the intelligent prediction model based on the difference between the actual observed data and the predicted result; The early warning module is used to determine whether to issue an early warning based on the prediction results and a preset probability threshold. If it is determined to be issued, the early warning information is generated and distributed.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

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

Images