A meteorological data acquisition system for meteorological disaster warning

Abstract

The invention discloses a meteorological data collection system for meteorological disaster warning, which belongs to the field of meteorological data technology, and specifically comprises: selecting a plurality of monitoring images of different severe convective weather events in a meteorological database, inputting the monitoring image set into a preset ResNet model, encoding into a public vector space, and outputting a monitoring image vector; collecting meteorological basic elements under current weather conditions, inputting the basic element set into a preset Transformer model, encoding into a public vector space, and outputting a basic element vector; calculating weather grade values ​​of a plurality of monitoring image vectors that are closest to the current basic element vector, and if the variance of the grade values ​​is greater than a preset threshold, adjusting the number of basic elements until the variance of the grade values ​​is less than or equal to the preset threshold, and taking the average value of the grade values ​​at this time as the severe convective weather grade of the warning; the invention simplifies the data processing flow of the meteorological warning.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological data, and in particular to a meteorological data collection system for meteorological disaster early warning. Background Art

[0002] In recent years, the rapid development of remote sensing technology has provided new means for meteorological monitoring. Radar and satellite images can provide observation data with large-scale and high temporal and spatial resolution, which is of great significance for capturing and analyzing severe convective weather systems. Similarly, complex monitoring indicators collected in advance include atmospheric instability index, vertical wind shear data, water vapor content in the atmosphere and humidity in the upper troposphere, as well as radar echo intensity and cloud top temperature observed by satellites. These indicators are crucial for capturing extreme weather conditions. By analyzing these complex meteorological parameters, the possibility and intensity of short-term heavy rainfall can be predicted more accurately, providing a scientific basis for the timely issuance of warnings.

[0003] However, processing these image data and monitoring indicators to extract useful meteorological information requires powerful computing resources and professional image processing software, which requires high operating costs. For the collection of some complex monitoring indicators, high-precision equipment such as radar and acoustic wave detectors are required, which requires professional technical support and irregular upgrades, which further increases the operating costs of the system. Summary of the invention

[0004] The purpose of the present invention is to provide a meteorological data acquisition system for meteorological disaster warning to solve the following technical problems:

[0005] Processing these image data and monitoring indicators to extract useful meteorological information requires powerful computing resources and professional image processing software, high-precision monitoring equipment, and high operating costs.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A meteorological data collection system for meteorological disaster warning, comprising:

[0008] The sample training module is used to select several monitoring images of different severe convective weather events in the meteorological database and perform preprocessing, wherein the monitoring images include radar images and satellite images, and the monitoring images of the same event are merged into a monitoring image set, and the corresponding severe convection level is marked as a label, and the monitoring image set is input into a preset ResNet model, so that the monitoring image set is encoded into a common vector space, and the monitoring image vector is output;

[0009] An indicator training module is used to collect n1 meteorological monitoring indicators under current weather conditions, select n2 meteorological monitoring indicators therefrom, where n2<n1, and mark them as meteorological basic elements, merge the meteorological basic elements into a basic element set, input the basic element set into a preset Transformer model, encode the basic elements into a common vector space, and output a basic element vector;

[0010] The result optimization module is used to calculate several monitoring image vectors that are closest to the current basic element vector, mark them as pending vectors, and obtain the severe convective weather level values ​​of all pending vectors. If the variance of the level values ​​is greater than the preset threshold, the number of basic meteorological elements is increased one by one, and the above steps are repeated until the variance of the level values ​​is less than or equal to the preset threshold. The average value of the level values ​​at this time is used as the severe convective weather level for warning.

[0011] As a further solution of the present invention: in the sample training module, the preprocessing process is:

[0012] Use median filtering, Gaussian filtering or bilateral filtering to reduce random noise in the image. For periodic noise, use Fourier transform to convert it to the frequency domain, then attenuate specific frequency components and return to the spatial domain through inverse transform. Enhance the contrast of the image through nonlinear stretching, adjust the grayscale distribution of the image so that it is evenly distributed on all possible grayscale values, and process the satellite image based on the dehazing algorithm.

[0013] As a further solution of the present invention: in the indicator training module:

[0014] The meteorological monitoring indicators include but are not limited to atmospheric instability, water vapor conditions, vertical wind shear, thermal and dynamic conditions; the initial indicators of the basic meteorological elements are temperature, humidity, wind speed and wind direction.

[0015] As a further solution of the present invention: the training process of the ResNet model and the Transformer model is:

[0016] Obtain monitoring images and monitoring data of n severe convective events, pair the monitoring images and monitoring data of the same severe convective event to generate n pairs of samples; input the monitoring images in all samples into the ResNet model, and the ResNet model outputs image feature vectors P1, ..., P n , the monitoring data is input into the Transformer model, and the Transformer model inputs the data feature vector W1,...,W n , pair the image feature vector and the data feature vector in turn to obtain the matrix R:

[0017] ;

[0018] Among them, P j ·W j As a positive sample pair, j belongs to 1,...,n, and P j ·W u As a negative sample pair, u also belongs to 1,...,n, but j is not equal to u;

[0019] The similarity Sim of the sample pair and the model loss function Loss are calculated by the following formula:

[0020] ;

[0021] ;

[0022] Among them, A i and B i Represent the components of the image feature vector and the data feature vector, i∈1,...,n; coefficients is the set hyperparameter, e is a natural constant, q and k represent the data feature vector and image feature vector respectively, where k + represents a positive sample, k i represents the i-th sample; the numerator represents the similarity between positive examples, and the denominator represents the similarity between positive examples and negative examples;

[0023] Calculate the similarity and loss function of the samples, and adjust the model parameters according to the calculation results so that the similarity of positive samples approaches 1 and the similarity of negative samples approaches 0. When the output value of the loss function Loss is less than or equal to the preset threshold, the model converges and the trained ResNet model and Transformer model are obtained.

[0024] As a further solution of the present invention: in the result optimization module, the process of calculating the vector distance is:

[0025] Label the current basic element vector as , the collected monitoring image vector is , the basic element vector is calculated by the following formula Vector image with monitor The distance d between:

[0026] ;

[0027] Among them, M and N represent the dimensions corresponding to the basic element vector and the monitoring image vector respectively.

[0028] As a further solution of the present invention: in the sample training module, the severe convective weather level includes wind speed, rainfall per unit time and hail diameter.

[0029] Beneficial effects of the present invention:

[0030] By combining the monitoring images processed by the ResNet model and the basic meteorological elements processed by the Transformer model, the present invention simplifies the data processing process, improves the processing efficiency, and significantly reduces the manpower and material costs by optimizing key meteorological monitoring indicators and automatically processing a large amount of data using deep learning technology compared to the traditional method that relies on high-cost satellite cloud images and complex indicators; at the same time, this method utilizes the visual information of the image and the quantitative information of the meteorological data, thereby providing a more comprehensive perspective than a single data source; based on the deep learning model, the system of the present invention has good scalability, and can adjust the model parameters according to different regions and different meteorological conditions to adapt to various complex meteorological environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be further described below in conjunction with the accompanying drawings.

[0032] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 creative work are within the scope of protection of the present invention.

[0034] See also Figure 1 As shown, the present invention is a meteorological data collection system for meteorological disaster warning, comprising:

[0035] The sample training module is used to select several monitoring images of different severe convective weather events in the meteorological database and perform preprocessing, wherein the monitoring images include radar images and satellite images, and the monitoring images of the same event are merged into a monitoring image set, and the corresponding severe convection level is marked as a label, and the monitoring image set is input into a preset ResNet model, so that the monitoring image set is encoded into a common vector space, and the monitoring image vector is output;

[0036] An indicator training module is used to collect n1 meteorological monitoring indicators under current weather conditions, select n2 meteorological monitoring indicators therefrom, where n2<n1, and mark them as meteorological basic elements, merge the meteorological basic elements into a basic element set, input the basic element set into a preset Transformer model, encode the basic elements into a common vector space, and output a basic element vector;

[0037] The result optimization module is used to calculate several monitoring image vectors that are closest to the current basic element vector, mark them as pending vectors, and obtain the severe convective weather level values ​​of all pending vectors. If the variance of the level values ​​is greater than the preset threshold, the number of basic meteorological elements is increased one by one, and the above steps are repeated until the variance of the level values ​​is less than or equal to the preset threshold. The average value of the level values ​​at this time is used as the severe convective weather level for warning.

[0038] In another preferred embodiment of the present invention, in the sample training module, the preprocessing process is:

[0039] Filtering: Reduce random noise in the image using techniques such as median filtering, Gaussian filtering, or bilateral filtering. These filters are effective in removing noise while maintaining edge clarity.

[0040] Fourier transform: For periodic noise, Fourier transform can be used to convert it to the frequency domain, then attenuate specific frequency components, and then return to the spatial domain through inverse transform.

[0041] Contrast stretching: Enhances the contrast of an image through linear or nonlinear stretching, making the details of the image more obvious.

[0042] Histogram equalization: Adjust the grayscale distribution of the image so that it is evenly distributed across all possible grayscale values, thereby enhancing the contrast of the image.

[0043] Dehazing algorithms: Especially when processing satellite images, removing the haze effect caused by atmospheric scattering can significantly improve the visual effect and analysis accuracy of the image.

[0044] In another preferred embodiment of the present invention, in the indicator training module:

[0045] The meteorological monitoring indicators include but are not limited to atmospheric instability, water vapor conditions, vertical wind shear, thermal and dynamic conditions; the initial indicators of the basic meteorological elements are temperature, humidity, wind speed and wind direction.

[0046] Conventional monitoring indicators that need to be collected in advance before short-term heavy rainfall include atmospheric instability, water vapor conditions, lifting mechanism, vertical wind shear, etc. The following is a detailed analysis:

[0047] Temperature and humidity distribution: The vertical distribution of temperature and humidity in the atmosphere is measured by radiosondes or radiosondes to assess the instability of the atmosphere. High instability is conducive to the occurrence of heavy rainfall.

[0048] Convective index: Calculates the lifting index, severe convection index, etc. These indices comprehensively reflect the instability of the ambient atmosphere, vertical wind shear and lifting effects, and are used to assess the probability of short-term heavy rainfall.

[0049] Water vapor content of the entire atmosphere: Monitor the water vapor content of the entire atmosphere from the ground to high altitudes, because sufficient water vapor is the key to the development of convective clouds and ultimately heavy rainfall. It can usually be obtained through radiosonde measurements or satellite remote sensing inversion.

[0050] Orographic lift: Monitor topographic features and their lifting effect on airflow, especially in mountainous and hilly areas. Orographic lift is an important factor in triggering short-term heavy rainfall.

[0051] Frontal Systems and Low-Level Convergence: Analyzes the location and movement of weather systems such as cold fronts, warm fronts, and low-level convergence zones that can significantly lift air and trigger heavy rainfall.

[0052] Wind speed and direction changes: Monitor changes in wind speed and direction at different altitudes, especially wind shear in the lower and middle atmosphere. Strong wind shear is conducive to the organization and development of strong convective systems, thus affecting the occurrence of short-term heavy rainfall.

[0053] High Wind Areas and Jet Streams: Identify and monitor the location and strength of high wind areas and jet streams, which can affect the development and movement of convective systems.

[0054] Ground heat flux and surface temperature: Collect ground heat flux and surface temperature data. High surface temperature can enhance the thermal uplift of the ground and promote the development of convection.

[0055] Boundary layer height: Monitors the height and structure of the boundary layer. A lower boundary layer restricts the development of convective clouds, while a higher boundary layer favors the vertical development of convective clouds.

[0056] Numerical Prediction Model: Use high-resolution numerical prediction models to predict precipitation in the next few days and evaluate the accuracy and reliability of the model in predicting short-term heavy rainfall.

[0057] Historical data analysis: Analyze precipitation data for the same period in history to understand and establish the climate background and patterns of short-term heavy rainfall.

[0058] Radar data: Doppler weather radar is used to monitor the echo structure and development of convective clouds, providing information on the location, intensity, movement and internal airflow structure of convective clouds.

[0059] Satellite data: Satellite cloud images are used to monitor the development and evolution of clouds, estimate the height and temperature of cloud tops, and provide information on the direction and speed of cloud movement.

[0060] In another preferred embodiment of the present invention, the training process of the ResNet model and the Transformer model is:

[0061] Obtain monitoring images and monitoring data of n severe convective events, pair the monitoring images and monitoring data of the same severe convective event to generate n pairs of samples; input the monitoring images in all samples into the ResNet model, and the ResNet model outputs image feature vectors P1, ..., P n , the monitoring data is input into the Transformer model, and the Transformer model inputs the data feature vector W1,...,W n , pair the image feature vector and the data feature vector in turn to obtain the matrix R:

[0062] ;

[0063] Among them, P j ·W j As a positive sample pair, j belongs to 1,...,n, and P j ·W u As a negative sample pair, u also belongs to 1,...,n, but j is not equal to u;

[0064] The similarity Sim of the sample pair and the model loss function Loss are calculated by the following formula:

[0065] ;

[0066] ;

[0067] Among them, A i and B i Represent the components of the image feature vector and the data feature vector, i∈1,...,n; coefficients is the set hyperparameter, e is a natural constant, q and k represent the data feature vector and image feature vector respectively, where k + represents a positive sample, k i represents the i-th sample; the numerator represents the similarity between positive examples, and the denominator represents the similarity between positive examples and negative examples;

[0068] Calculate the similarity and loss function of the samples, and adjust the model parameters according to the calculation results so that the similarity of positive samples approaches 1 and the similarity of negative samples approaches 0. When the output value of the loss function Loss is less than or equal to the preset threshold, the model converges and the trained ResNet model and Transformer model are obtained.

[0069] In another preferred embodiment of the present invention, in the result optimization module, the process of calculating the vector distance is:

[0070] Label the current basic element vector as , the collected monitoring image vector is , the basic element vector is calculated by the following formula Vector image with monitor The distance d between:

[0071] ;

[0072] Among them, M and N represent the dimensions corresponding to the basic element vector and the monitoring image vector respectively.

[0073] By acquiring monitoring images and monitoring data of multiple severe convective events for model training, the adaptability and generalization ability of the model under different meteorological conditions are increased. This method allows the model to learn a wider range of meteorological patterns and characteristics, thereby improving the accuracy and reliability of the early warning system in practical applications.

[0074] The ResNet model and the Transformer model are used to process monitoring images and monitoring data respectively, and then the two different types of data (image feature vectors and data feature vectors) are effectively fused to achieve in-depth analysis and learning of comprehensive features. This fusion strategy can more accurately capture and express the complex characteristics of severe convective weather, further improving the accuracy of early warning.

[0075] By calculating the similarity of sample pairs and the model loss function, the model parameters are continuously adjusted and optimized until the model converges. This iterative optimization process helps the model better fit the training data, thereby showing higher accuracy and lower error in actual predictions.

[0076] The result optimization module evaluates the reliability of the forecast results by calculating the distance between the basic element vector and the monitoring image vector. When more basic meteorological elements are introduced and their similarity with historical monitoring data is calculated, the system can effectively identify and filter abnormal data and reduce false positives and negatives.

[0077] The present invention can dynamically adjust the number of input basic meteorological elements according to the variance of the output warning level value. This dynamic adjustment mechanism enables the system to provide real-time feedback and optimize the warning results according to actual conditions, thereby improving the timeliness and accuracy of the warning.

[0078] In another preferred embodiment of the present invention, in the sample training module, the severe convective weather level includes wind speed, rainfall per unit time and hail diameter.

[0079] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A meteorological data collection system for meteorological disaster warning, characterized in that: include: The sample training module is used to select several monitoring images of different severe convective weather events in the meteorological database and perform preprocessing, wherein the monitoring images include radar images and satellite images, and the monitoring images of the same event are merged into a monitoring image set, and the corresponding severe convective weather level is marked as a label, and the monitoring image set is input into a preset ResNet model, so that the monitoring image set is encoded into a common vector space, and the monitoring image vector is output; An indicator training module is used to collect n1 meteorological monitoring indicators under current weather conditions, select n2 meteorological monitoring indicators therefrom, where n2<n1, and mark them as meteorological basic elements, merge the meteorological basic elements into a basic element set, input the basic element set into a preset Transformer model, encode the basic elements into a common vector space, and output a basic element vector; The result optimization module is used to calculate several monitoring image vectors that are closest to the current basic element vector, mark them as pending vectors, and obtain the severe convective weather level values ​​of all pending vectors. If the variance of the level values ​​is greater than the preset threshold, the number of basic meteorological elements is increased one by one, and the above steps are repeated until the variance of the level values ​​is less than or equal to the preset threshold. The average value of the level values ​​at this time is used as the severe convective weather level for warning.

2. A meteorological data collection system for meteorological disaster warning according to claim 1, characterized in that: In the sample training module, the preprocessing process is as follows: Use median filtering, Gaussian filtering or bilateral filtering to reduce random noise in the image. For periodic noise, use Fourier transform to convert it to the frequency domain, then attenuate specific frequency components and return to the spatial domain through inverse transform. Enhance the contrast of the image through nonlinear stretching, adjust the grayscale distribution of the image so that it is evenly distributed on all possible grayscale values, and process the satellite image based on the dehazing algorithm.

3. A meteorological data collection system for meteorological disaster warning according to claim 1, characterized in that: In the indicator training module: The meteorological monitoring indicators include but are not limited to atmospheric instability, water vapor conditions, vertical wind shear, thermal and dynamic conditions; the initial indicators of the basic meteorological elements are temperature, humidity, wind speed and wind direction.

4. A meteorological data collection system for meteorological disaster warning according to claim 1, characterized in that: The training process of the ResNet model and the Transformer model is: Obtain monitoring images and monitoring data of n severe convective weather events, pair the monitoring images and monitoring data of the same severe convective weather event to generate n pairs of samples; input the monitoring images in all samples into the ResNet model, and the ResNet model outputs image feature vectors P1, ..., P n , the monitoring data is input into the Transformer model, and the Transformer model inputs the data feature vector W1,...,W n , pair the image feature vector and the data feature vector in turn to obtain the matrix R: ； Among them, P j ·W j As a positive sample pair, j belongs to 1,...,n, and P j ·W u As a negative sample pair, u also belongs to 1,...,n, but j is not equal to u; The similarity Sim of the sample pair and the model loss function Loss are calculated by the following formula: ； ； Among them, A i and B i Represent the components of the image feature vector and the data feature vector, i∈1,...,n; coefficients is the set hyperparameter, e is a natural constant, q and k represent the data feature vector and image feature vector respectively, where k + represents a positive sample, k i represents the i-th sample; the numerator in the model loss function Loss represents the similarity between positive examples, and the denominator represents the similarity between positive examples and negative examples; Calculate the similarity and loss function of the samples, and adjust the model parameters according to the calculation results so that the similarity of positive samples approaches 1 and the similarity of negative samples approaches 0. When the output value of the loss function Loss is less than or equal to the preset threshold, the model converges and the trained ResNet model and Transformer model are obtained.

5. A meteorological data collection system for meteorological disaster warning according to claim 4, characterized in that: In the result optimization module, the process of calculating the vector distance is: Label the current basic element vector as , the collected monitoring image vector is , the basic element vector is calculated by the following formula Vector with monitoring image The distance d between: ； Among them, M and N represent the dimensions corresponding to the basic element vector and the monitoring image vector respectively.

6. A meteorological data collection system for meteorological disaster warning according to claim 1, characterized in that: In the sample training module, the severe convective weather level includes wind speed, rainfall per unit time and hail diameter.

Images