Severe weather prediction method and system based on satellite-ground collaboration

Abstract

The embodiment of the present application provides a disastrous weather prediction method and system based on satellite-ground collaboration, which converts the collaborative observation data through a climate factor conversion unit to generate a climate factor steering vector, and blends the climate factor steering vector with meteorological activity information to generate a blended steering vector, which can achieve the organic integration of data from different sources and fully explore the complementarity and correlation between them. The disastrous weather vector of the reference satellite-ground monitoring station is parsed from the blended steering vector according to the disastrous vector positioning unit, which can accurately identify and locate the key features of the disastrous weather, optimize the network configuration parameters in the disastrous weather prediction network according to the disastrous weather vector, and generate a deployable disastrous weather prediction network, which can significantly improve the performance and accuracy of the disastrous weather prediction network and better adapt to various complex and changeable disastrous weather conditions.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for predicting disastrous weather based on satellite-ground collaboration. Background Art

[0002] With the impact of global climate change, the frequency and intensity of disastrous weather are on the rise, posing a huge threat to human society and the natural environment. Therefore, accurately predicting disastrous weather and taking preventive measures in advance are of great significance for reducing disaster losses and protecting people's lives and property.

[0003] Traditional disastrous weather forecasting methods mainly rely on data from ground-based meteorological observation stations. However, due to the limited distribution of ground observation stations and the influence of various factors such as terrain and landforms, the temporal and spatial resolution and coverage of their observation data are limited, making it difficult to meet the needs of high-precision and timely disastrous weather forecasting.

[0004] In recent years, with the development of satellite remote sensing technology, satellite-ground coordinated observation has become a new means of disastrous weather prediction. Through the coordinated observation of satellites and ground monitoring stations, more extensive and comprehensive meteorological data can be obtained, improving the accuracy and timeliness of disastrous weather prediction. However, how to effectively integrate the observation data from satellites and ground monitoring stations, extract key information related to disastrous weather, and build a high-performance disastrous weather prediction model is still a problem that needs to be solved urgently. Summary of the invention

[0005] In order to at least overcome the above-mentioned deficiencies in the prior art, an object of the embodiments of the present application is to provide a method and system for predicting disastrous weather based on satellite-ground collaboration.

[0006] According to one aspect of the present application, a method for forecasting disastrous weather based on satellite-ground collaboration is provided, characterized in that the method comprises:

[0007] Acquiring collaborative observation data and meteorological activity information of reference satellite and ground monitoring stations; the meteorological activity information is obtained based on real-time meteorological activity data captured by the reference satellite and ground monitoring stations for each disaster risk area;

[0008] The collaborative observation data is converted by a climate factor conversion unit in the disastrous weather prediction network with initialized network parameters to generate a climate factor steering vector;

[0009] Blending the climate factor steering vector with the meteorological activity information to generate a blended steering vector;

[0010] According to the disastrous vector positioning unit in the disastrous weather prediction network of the initialized network parameters, the disastrous weather vector of the reference satellite-ground monitoring station is obtained by parsing from the blended steering vector;

[0011] The network configuration parameters in the disastrous weather prediction network with initialized network parameters are optimized according to the disastrous weather vector, and the network parameter learning of the disastrous weather prediction network with initialized network parameters is iteratively performed according to the optimized network configuration parameters until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

[0012] According to one aspect of an embodiment of the present application, a disaster weather prediction system is provided, which includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the disaster weather prediction method based on satellite-ground collaboration in any one of the possible implementation modes described above.

[0013] According to one aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in various optional implementations of the above three aspects.

[0014] In the technical solutions provided in some embodiments of the present application, the present application can make full use of the advantages of satellite-ground collaborative observation by acquiring the collaborative observation data and meteorological activity information of the reference satellite-ground monitoring station, realize the real-time and comprehensive meteorological data acquisition of the disaster risk area, and provide an accurate and reliable data basis for the subsequent disastrous weather forecast. Secondly, by converting the collaborative observation data through the climate factor conversion unit in the disastrous weather forecast network with initialized network parameters, a climate factor steering vector is generated, which can effectively extract the key climate factor information in the collaborative observation data, and provide a convenient and efficient data form for the subsequent blending processing. Thirdly, the climate factor steering vector is blended with the meteorological activity information to generate a blending steering vector, which can realize the organic fusion of data from two different sources, fully explore the complementarity and correlation between them, and improve the accuracy and reliability of the disastrous weather forecast. In addition, according to the disaster vector positioning unit in the disastrous weather forecast network with initialized network parameters, the disastrous weather vector of the reference satellite-ground monitoring station is parsed from the blending steering vector, which can accurately identify and locate the key features of disastrous weather, and provide clear and targeted guidance for the subsequent network parameter optimization and learning. Finally, the network configuration parameters of the disaster weather prediction network with initialized network parameters are optimized according to the disaster weather vector, and the network parameters of the disaster weather prediction network with initialized network parameters are learned iteratively according to the optimized network configuration parameters until the preset network convergence requirements are met, and a deployable disaster weather prediction network is generated. This process can significantly improve the performance and accuracy of the disaster weather prediction network, so that it can better adapt to various complex and changeable disaster weather conditions in practical applications, and provide strong support for disaster warning and prevention.

[0015] That is, the present application can accurately and quickly predict disastrous weather conditions, improving the accuracy and timeliness of early warnings. At the same time, by introducing climate factor guidance vectors, the understanding of environmental factors by the disastrous weather prediction network is enhanced, further improving the accuracy of the prediction. In addition, by iterative learning and optimizing network parameters, the disastrous weather prediction network can adapt to different climate environments and weather conditions, and has stronger universality and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be extracted in combination with these drawings without creative work.

[0017] Figure 1A schematic diagram of a process flow of a disastrous weather prediction method based on satellite-ground collaboration provided in an embodiment of the present application;

[0018] Figure 2 A schematic block diagram of the architecture of a disastrous weather prediction system for implementing the above-mentioned disastrous weather prediction method based on satellite-ground collaboration provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The following description is intended to enable one of ordinary skill in the art to implement and combine the present application, and the description is provided in the context of a specific application scenario and its requirements. It will be apparent to one of ordinary skill in the art that various changes may be made to the disclosed embodiments, and that the general principles defined in the present application may be applied to other embodiments and application scenarios without departing from the principles and scope of the present application. Therefore, the present application is not limited to the described embodiments, but should be given the broadest scope consistent with the claims.

[0020] Figure 1 This is a flow chart of a method for predicting disastrous weather based on satellite-ground collaboration provided by an embodiment of the present application. The method for predicting disastrous weather based on satellite-ground collaboration is introduced in detail below.

[0021] Step S110, obtaining the coordinated observation data and meteorological activity information of the reference satellite and ground monitoring stations.

[0022] The meteorological activity information is obtained based on the real-time meteorological activity data captured by the reference satellite-ground monitoring station for each disaster risk area.

[0023] In this embodiment, the disaster weather prediction system is located in a data center as a server and can be connected to multiple reference satellite and ground monitoring stations distributed in various places through high-speed Internet connections. These reference satellite and ground monitoring stations are equipped with advanced meteorological sensors and observation equipment, which can capture various meteorological data in real time, such as wind speed, wind direction, temperature, humidity, atmospheric pressure, etc. For example, a reference satellite and ground monitoring station is set up at the top of a mountainous area, and the reference satellite and ground monitoring station is equipped with anemometers, thermometers, hygrometers and other equipment. When the wind blows over the top of the mountain, the anemometer will record the wind speed and wind direction data; at the same time, the thermometer and hygrometer will also record the current temperature and humidity information respectively. These data will then be transmitted to the server in the data center for analysis and processing.

[0024] For example, at a specific time point, such as 2 a.m. every day, the server will automatically send a data request to all connected reference satellite and ground monitoring stations. After receiving the data request, the reference satellite and ground monitoring station will immediately package the latest collected collaborative observation data and send it to the server through an encrypted transmission protocol. After receiving these collaborative observation data, the server will first decrypt and verify them to ensure the integrity and accuracy of the data. At the same time, the server will also obtain meteorological activity information generated from meteorological agencies based on the real-time meteorological activity data captured by these monitoring stations. This meteorological activity information may include the path, intensity, and estimated arrival time of the storm, which is crucial for the prediction of disastrous weather.

[0025] In detail, the collaborative observation data refers to a collection of meteorological data collected by multiple reference satellite and ground monitoring stations at the same time or at a similar time. After integration and calibration, these meteorological data can provide more comprehensive and accurate meteorological information, which is helpful for analyzing and predicting weather changes. For example, suppose there are three reference satellite and ground monitoring stations A, B and C located in different cities. At a certain point in time, these three reference satellite and ground monitoring stations simultaneously collected wind speed, temperature and humidity data in their respective areas, which were aggregated to form a set of collaborative observation data, which can reflect the meteorological conditions of the three cities at the same time. By analyzing these data, we can understand the spatial distribution and changing trends of meteorological elements.

[0026] The meteorological activity information refers to information about weather phenomena and meteorological processes obtained based on the analysis and processing of collaborative observation data, which may include the path, intensity, expected arrival time, etc. of the storm, which is crucial for the prediction and warning of disastrous weather. The disaster risk area refers to an area prone to natural disasters that is delineated based on historical meteorological records, topography, population density and other factors. These areas are more likely to suffer disasters when facing specific meteorological conditions, so they need special attention and monitoring. For example, a coastal area has suffered flood disasters caused by typhoons and heavy rains many times in history due to factors such as low-lying terrain and imperfect drainage systems. Therefore, the area is designated as a disaster risk area. In order to monitor and warn of possible disasters in a timely manner, the local area has set up multiple reference satellite-ground monitoring stations in the area to collect meteorological data in real time and transmit it to the data center for analysis and processing. When abnormal meteorological conditions are monitored, relevant departments can take quick action to reduce the losses caused by the disaster.

[0027] Step S120, converting the collaborative observation data through the climate factor conversion unit in the disastrous weather prediction network with initialized network parameters to generate a climate factor steering vector.

[0028] In detail, the disastrous weather prediction network is a deep learning or machine learning model that can be trained to predict the probability, type, intensity, etc. of disastrous weather. This disastrous weather prediction network inputs various meteorological data, and after complex internal calculations and processing, outputs the prediction results about disastrous weather. For example, suppose the task of the disastrous weather prediction network is to predict the probability of heavy rain in a certain area within the next 24 hours. This disastrous weather prediction network will receive meteorological data of the area as input, such as temperature, humidity, wind speed, etc. Then, the algorithm inside the disastrous weather prediction network will process and analyze this data, and finally output a probability value, indicating the possibility of heavy rain in the next 24 hours. This probability value can help people take corresponding preventive measures.

[0029] Initializing network parameters refers to the process of assigning initial values ​​to parameters such as weights and biases in the disaster weather prediction network before training the disaster weather prediction network. These initial values ​​have an important impact on the training process and final prediction performance of the disaster weather prediction network. For example, when training a neural network, it is necessary to assign an initial value to the connection weights and biases between each neuron in the disaster weather prediction network. These initial values ​​can be random or selected according to a certain strategy. Then, through the forward propagation and back propagation algorithms of the disaster weather prediction network, these parameters are continuously updated so that the network can gradually learn the mapping relationship from input to output. In this process, initializing network parameters is a very important step, which directly affects the training speed of the network and the final convergence result.

[0030] Therefore, in this embodiment, the server runs a disastrous weather prediction network, which includes a climate factor conversion unit, whose function is to convert the original collaborative observation data into a form that is easier for the network to process. For example, suppose there is a set of original collaborative observation data, including the values ​​of meteorological elements such as temperature, humidity, and wind speed. These collaborative observation data may be difficult to be effectively processed and analyzed if they are directly input into the disastrous weather prediction network. Therefore, a climate factor conversion unit is needed to convert these collaborative observation data. This climate factor conversion unit can first denoise and normalize the collaborative observation data to eliminate outliers and dimensional differences. Then, methods such as principal component analysis can be used to reduce the dimensionality of the data and extract the main climate factor features. Finally, these climate factor features are combined into a high-dimensional climate factor guidance vector as the input of the network.

[0031] The climate factor steering vector is the output result of the climate factor conversion unit. It is a high-dimensional vector that contains the key climate information and features in the original collaborative observation data. This climate factor steering vector is the basis for the subsequent processing and analysis of the disastrous weather prediction network. Let's continue with the above-mentioned climate factor conversion unit as an example. Suppose that after the conversion, a climate factor steering vector containing 10 features is obtained. Each element in this climate factor steering vector represents the characteristic value of a certain climate factor in the original data. These characteristic values ​​may be the average temperature, the maximum humidity, the standard deviation of the wind speed, etc. Through this climate factor steering vector, we can have a more comprehensive understanding of the climate information in the original data, and provide strong support for subsequent disastrous weather forecasts.

[0032] In an example, suppose that the following collaborative observation data are collected by a reference satellite-ground monitoring station in a certain area: temperature, humidity, wind speed, air pressure, and precipitation. These data can form a raw meteorological data vector, such as:

[0033] Original meteorological data vector = [temperature, humidity, wind speed, air pressure, precipitation]

[0034] The task of the climate factor conversion unit is to convert these raw data into climate factor steering vectors. This process may include steps such as data standardization, normalization, and feature extraction. For example, after conversion, the following climate factor steering vector may be obtained:

[0035] Climate factor steering vector = [standardized temperature, normalized humidity, wind speed level, pressure change rate, precipitation probability]

[0036] Here, the standardized temperature may be obtained by subtracting a fixed value (such as the average temperature of the area) from the original temperature value and dividing it by a standard deviation to obtain a dimensionless value. The normalized humidity may be converted to a range of 0 to 1. The wind speed level may be a discrete level divided according to the wind speed value. The pressure change rate may be the difference between the current pressure and the pressure at the previous time point. The probability of precipitation may be a probability value calculated based on historical data and current meteorological conditions.

[0037] Step S130, blending the climate factor steering vector with the meteorological activity information to generate a blended steering vector.

[0038] In this embodiment, after obtaining the climate factor steering vector, the server integrates it with the meteorological activity information obtained from the meteorological agency. The integration process may involve weight allocation, feature fusion and dimensionality reduction of the two types of data to ensure that the two types of information can be fully combined and complement each other.

[0039] Specifically, the server can first normalize the climate factor steering vector and meteorological activity information so that they are in the same dimension and range. Then, the feature fusion technology is used to fuse the features of the two data. This may involve operations such as weighted summation, series connection or parallel connection of features. The fused data retains the key information of the original data and combines the complementarity of the two data. Finally, the server uses dimensionality reduction techniques, such as principal component analysis or autoencoders, to reduce the dimensionality of the fused data. The reduced dimensionality data can be used as a fusion steering vector, which integrates the key features of the original observation data and meteorological activity information, while reducing the dimension and complexity of the data, making it easier for subsequent network processing and analysis.

[0040] That is, the blending steering vector is the result of blending the climate factor steering vector and the meteorological activity information. The blending process may involve weight assignment, feature fusion, and dimensionality reduction of the two types of data to ensure that the two types of information can be fully combined and complement each other. The blending steering vector integrates the key features of the original observation data and meteorological activity information, providing richer input information for the severe weather prediction network.

[0041] For example, suppose there is a climate factor steering vector and a meteorological activity information containing storm path information. In order to combine these two kinds of information, a feature fusion method can be used. First, the climate factor steering vector and the meteorological activity information are normalized so that they are in the same dimension and range. Then, the features of the two data are weighted summed or concatenated to generate a new high-dimensional vector. This high-dimensional vector is the fusion steering vector, which contains both the climate information of the original observation data and the key features of the meteorological activity information. Through this fusion steering vector, the disastrous weather prediction network can more accurately predict the disastrous weather conditions that may occur in the future.

[0042] In one example, the blended steering vector is the result of blending the climate factor steering vector and the meteorological activity information. The meteorological activity information may include key features such as the storm's path, intensity, and estimated arrival time. Assume there is a climate factor steering vector and a meteorological activity information vector containing storm information, as follows:

[0043] Climate factor steering vector = [standardized temperature, normalized humidity, wind speed level, pressure change rate, precipitation probability]

[0044] Meteorological activity information vector = [storm intensity, longitude of storm path, latitude of storm path, estimated arrival time]

[0045] The fusion process may involve weighted summation of the two data, feature concatenation, or other complex fusion methods. For example, the climate factor guidance vector and the meteorological activity information vector can be concatenated to form a longer vector:

[0046] Blending Steering Vector = [normalized temperature, normalized humidity, wind speed level, pressure change rate, precipitation probability, storm intensity, longitude of storm path, latitude of storm path, estimated time of arrival]

[0047] Alternatively, the two data can be fused in a more complex way, such as using a neural network or other machine learning algorithm to learn the nonlinear relationship between them and generate a new fusion guidance vector. This new vector will contain the key features of both the climate information of the original observation data and the meteorological activity information, providing richer input information for the severe weather prediction network.

[0048] Step S140, according to the disastrous weather vector positioning unit in the disastrous weather prediction network with the initialized network parameters, parses and obtains the disastrous weather vector of the reference satellite-ground monitoring station from the blended steering vector.

[0049] In this embodiment, the disastrous weather prediction network also includes a disastrous vector positioning unit, whose main function is to parse and extract vector information related to disastrous weather from the blending guidance vector. This information is usually expressed in the form of a vector, which contains key features such as the type, intensity, location, and direction of movement of the disastrous weather. For example, it is assumed that the blending guidance vector is a complex vector that contains a variety of meteorological information and disastrous weather characteristics. The task of the disastrous vector positioning unit is to identify and extract features directly related to disastrous weather from this complex vector, such as the intensity, path, and estimated arrival time of the storm. These features are organized into a new vector, the disastrous weather vector, which more directly reflects the key information of disastrous weather and helps the network make more accurate predictions.

[0050] When the server obtains the blending guidance vector, it inputs the blending guidance vector into the disaster vector positioning unit, so as to further extract and select the features of the blending guidance vector to find the features most relevant to the disastrous weather. Then, the extracted features are learned and predicted using classification or regression algorithms, such as support vector machines, random forests or neural networks. Finally, the disaster vector positioning unit will output a disastrous weather vector, which contains key information about the type, intensity, and impact range of possible disastrous weather. Taking a typhoon as an example, a disastrous weather vector may contain information such as the center position coordinates, maximum wind speed, moving direction and speed of the typhoon. This information is encoded into a vector form, such as [longitude, latitude, maximum wind speed, moving direction, moving speed]. Through this vector, the disastrous weather prediction network can understand the current state of the typhoon and predict its future development trend based on historical data and models, so as to issue early warning information in a timely manner.

[0051] In practical applications, the severe weather prediction network will continuously receive new collaborative observation data and process and analyze the data through components such as the climate factor conversion unit and the severe vector positioning unit. In this way, the network can track and predict the development and changes of severe weather in real time, providing strong support for disaster prevention and mitigation work.

[0052] Step S150, optimizing the network configuration parameters in the disastrous weather prediction network with initialized network parameters according to the disastrous weather vector, and iterating network parameter learning for the disastrous weather prediction network with initialized network parameters according to the optimized network configuration parameters until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

[0053] In this embodiment, after obtaining the disastrous weather vector, the server will use it to optimize the network configuration parameters in the disastrous weather prediction network. The optimization process may involve adjusting and improving the structure of the network layer, the connection weights between nodes, the activation function, etc.

[0054] Specifically, the server can first calculate the back propagation algorithm of the disaster weather prediction network according to the disaster weather vector to find out the errors and deficiencies in the disaster weather prediction network. Then, the configuration parameters in the disaster weather prediction network are iteratively updated and optimized using the gradient descent method, the stochastic gradient descent method or the Adam algorithm. The goal of the optimization is to enable the disaster weather prediction network to output the disaster weather vector more accurately when a new blending guidance vector is input. Through repeated iterations of this process, the server continuously improves the performance of the disaster weather prediction network and gradually reduces the prediction error. When the performance of the disaster weather prediction network reaches the preset convergence requirements (such as the error rate is lower than a certain threshold or the number of iterations reaches the upper limit, etc.), the server will stop iterating and save the optimized network configuration parameters. At this point, the disaster weather prediction network has been trained and can be used in practical applications. The server can deploy this disaster weather prediction network on the cloud or edge devices to provide users with real-time and accurate disaster weather prediction services.

[0055] Based on the above steps, this application can make full use of the advantages of satellite-ground collaborative observation by acquiring the collaborative observation data and meteorological activity information of the reference satellite-ground monitoring station, realize real-time and comprehensive meteorological data acquisition of disaster risk areas, and provide accurate and reliable data basis for subsequent disastrous weather forecasts. Secondly, by converting the collaborative observation data through the climate factor conversion unit in the disastrous weather prediction network with initialized network parameters, a climate factor guidance vector is generated, which can effectively extract the key climate factor information in the collaborative observation data and provide a convenient and efficient data form for subsequent blending processing. Thirdly, the climate factor guidance vector is blended with the meteorological activity information to generate a blending guidance vector, which can realize the organic fusion of data from two different sources, fully explore the complementarity and correlation between them, and improve the accuracy and reliability of disastrous weather forecasts. In addition, according to the disaster vector positioning unit in the disastrous weather prediction network with initialized network parameters, the disastrous weather vector of the reference satellite-ground monitoring station is parsed from the blending guidance vector, which can accurately identify and locate the key features of disastrous weather, and provide clear and targeted guidance for subsequent network parameter optimization and learning. Finally, the network configuration parameters of the disaster weather prediction network with initialized network parameters are optimized according to the disaster weather vector, and the network parameters of the disaster weather prediction network with initialized network parameters are learned iteratively according to the optimized network configuration parameters until the preset network convergence requirements are met, and a deployable disaster weather prediction network is generated. This process can significantly improve the performance and accuracy of the disaster weather prediction network, so that it can better adapt to various complex and changeable disaster weather conditions in practical applications, and provide strong support for disaster warning and prevention.

[0056] That is, the present application can accurately and quickly predict disastrous weather conditions, improving the accuracy and timeliness of early warnings. At the same time, by introducing climate factor guidance vectors, the understanding of environmental factors by the disastrous weather prediction network is enhanced, further improving the accuracy of the prediction. In addition, by iterative learning and optimizing network parameters, the disastrous weather prediction network can adapt to different climate environments and weather conditions, and has stronger universality and robustness.

[0057] In a possible implementation, step S120 may include:

[0058] Step S121, performing self-attention feature extraction on the collaborative observation data to generate a self-attention climate vector of the collaborative observation data.

[0059] In this embodiment, the server receives collaborative observation data from multiple meteorological monitoring stations, including multiple meteorological elements such as temperature, humidity, wind speed, air pressure and precipitation, etc. These data exist in the form of time series, and each time point has a corresponding set of observation values.

[0060] The server will first preprocess these collaborative observation data, such as denoising and missing value filling, to ensure the quality and integrity of the data. Then, the server will use the self-attention mechanism to extract features from these data.

[0061] The self-attention mechanism is a method that allows the model to learn the internal dependencies of the data. In this scenario, the server will build a self-attention model, take the collaborative observation data as input, and extract the key climate features in the data by calculating the similarity and correlation between different time points in the data.

[0062] For example, the server can find that there is a strong correlation between the temperature and humidity values ​​at one point in time and the wind speed and air pressure values ​​at another point in time, which may indicate the occurrence of some kind of severe weather. In this way, the server can generate a vector that contains the key climate features in the collaborative observation data, namely the self-attention climate vector.

[0063] Step S122, loading the self-attention climate vector into the climate factor conversion unit in the disastrous weather prediction network with initialized network parameters for conversion to generate a climate factor steering vector.

[0064] After the server generates the self-attention climate vector, it will load it into the severe weather prediction network with initialized network parameters. This severe weather prediction network contains multiple layers and nodes, each of which corresponds to the processing and conversion of one or more climate factors.

[0065] The climate factor conversion unit is an important part of the network, responsible for further conversion and processing of the input self-attention climate vector. In this process, the climate factor conversion unit can use some complex mathematical operations and neural network models to transform and map each element in the self-attention climate vector one by one.

[0066] For example, the climate factor conversion unit can perform a nonlinear transformation on the temperature value in the self-attention climate vector and map it into a higher-dimensional space to better capture the complex relationship between temperature and other climate factors. At the same time, the climate factor conversion unit can also perform similar processing on other elements in the self-attention climate vector.

[0067] After being processed by the climate factor conversion unit, the server can obtain a new vector, namely the climate factor steering vector. This vector contains the key climate information and features in the original collaborative observation data, and is the basis for subsequent processing and analysis of the severe weather prediction network.

[0068] Therefore, we can more clearly understand how to transform the collaborative observation data through the climate factor conversion unit in the severe weather prediction network that initializes the network parameters and generate the climate factor steering vector.

[0069] In a possible implementation, the collaborative observation data includes multiple items.

[0070] Step S122 may include:

[0071] Step S1221, based on the self-attention climate vectors corresponding to each of the collaborative observation data, the self-attention climate vectors are arranged in the spatiotemporal domain to generate a corresponding self-attention climate vector set.

[0072] In this embodiment, these self-attention climate vectors are different in time and space because they represent the weather conditions at different places and times. In order to effectively integrate these self-attention climate vectors, they need to be arranged in the time and space domains.

[0073] Temporal and spatial arrangement refers to organizing the self-attention climate vectors of different time and space into a set according to a certain order and rules. Each element in this set corresponds to a self-attention climate vector of a specific time point and geographical location.

[0074] For example, the server can arrange the self-attention climate vectors in chronological order to form a time series set. In this set, the self-attention climate vectors of earlier time points are arranged in front, and the self-attention climate vectors of later time points are arranged in the back. At the same time, the server can also spatially arrange the self-attention climate vectors according to the distance relationship of the geographical locations to form a spatial sequence set. In this set, the self-attention climate vectors of adjacent geographical locations are organized together.

[0075] Through the arrangement of time and space domains, the server can generate a set of self-attention climate vectors corresponding to multiple collaborative observation data, namely the self-attention climate vector set. This set provides a rich and orderly data basis for the subsequent climate factor conversion.

[0076] Step S1222: Load the self-attention climate vector set into the climate factor conversion unit in the disastrous weather prediction network with initialized network parameters for conversion to generate a climate factor steering vector. The climate factor steering vector includes a preset number of climate factor steering elements.

[0077] After the self-attention climate vector set is generated, it can be loaded into the severe weather prediction network with initialized network parameters. The climate factor conversion unit is responsible for further conversion and processing of the input self-attention climate vector set. In this process, the climate factor conversion unit uses some complex mathematical operations and neural network models to transform and map each self-attention climate vector in the set one by one.

[0078] For example, the climate factor conversion unit can perform a convolution operation on each self-attention climate vector in the set to extract its local features; or use a recurrent neural network to model the time series data in the set to capture temporal dependencies; and can also use a fully connected layer to perform nonlinear transformation and mapping on each element in the set.

[0079] Through these processes, the climate factor conversion unit can convert the self-attention climate vector set into a new vector, namely the climate factor oriented vector. This vector contains the key climate information and features in the original collaborative observation data and represents them in a more abstract and advanced form. These climate factor oriented elements can be regarded as a highly summarized and abstract representation of the original meteorological data.

[0080] The generation of climate factor guidance vectors is an important step in the severe weather prediction network. It provides more abundant and useful input information for subsequent network layers, which helps the network to better learn and predict the occurrence and development trend of severe weather. At the same time, by setting a preset number of climate factor guidance elements, the network can reduce the complexity of calculation and resource consumption while maintaining a certain complexity.

[0081] In a possible implementation, step S140 may include:

[0082] Step S141, loading the blended steering vector into the disaster vector positioning unit in the disaster weather prediction network with initialized network parameters to extract features and generate a disaster vector knowledge network.

[0083] In this embodiment, the server has generated a blended steering vector in the previous step, which combines the climate factor steering vector and meteorological activity information and contains rich information required for disastrous weather forecasting. Now, the server needs to load this blended steering vector into the disastrous vector positioning unit in the disastrous weather forecasting network for processing.

[0084] The disaster vector positioning unit is an important part of the disaster weather prediction network, responsible for extracting characteristic information related to the disaster weather from the blending guidance vector. In order to complete this task, the disaster vector positioning unit will use a series of complex neural network models and algorithms to perform in-depth feature extraction and processing on the blending guidance vector.

[0085] Specifically, the server will input the blending guidance vector into the disaster vector positioning unit, which will first pre-process the blending guidance vector, such as normalization and denoising, to ensure the accuracy and consistency of the data. Then, the disaster vector positioning unit will use deep learning models such as convolutional neural networks and recurrent neural networks to extract features from the blending guidance vector layer by layer. , so that complex patterns such as spatial structure information, time dependency, and interaction between different features in the blending guidance vector can be captured.

[0086] After multiple layers of feature extraction and processing, the disastrous vector positioning unit will output a vector containing the key disastrous weather features in the blending guidance vector, namely the disastrous vector knowledge network. This disastrous vector knowledge network represents the disastrous weather information in the blending guidance vector in a more abstract and advanced form, providing strong support for the subsequent disastrous weather vector space mapping.

[0087] Step S142: the disastrous weather vector positioning unit maps the disastrous weather vector space at least once according to the disastrous vector knowledge network to generate the disastrous weather vector of the reference satellite-to-ground monitoring station.

[0088] After obtaining the disaster vector knowledge network, the server needs to map the disaster weather vector space based on the disaster vector knowledge network. This mapping process is the process of converting the characteristic information in the disaster vector knowledge network into a specific disaster weather vector.

[0089] In order to achieve this mapping process, the disaster vector location unit will use some complex mathematical operations and neural network models to further transform and process the disaster vector knowledge network. For example, the disaster vector location unit can use the fully connected layer to perform nonlinear transformation and mapping on each element in the disaster vector knowledge network; or use methods such as principal component analysis to reduce the dimension of the disaster vector knowledge network and extract the main characteristic components; it can also use methods such as clustering algorithms to group and classify the features in the disaster vector knowledge network.

[0090] Through these processes, the disaster vector positioning unit can map the disaster vector knowledge network into a specific disaster weather vector space and generate the disaster weather vector of the reference satellite-ground monitoring station. This disaster weather vector represents the type, intensity, duration and other key information of the possible future disaster weather at the reference satellite-ground monitoring station in a numerical form, providing an important reference for subsequent disaster weather warnings and decision-making.

[0091] It should be noted that the server can perform multiple disastrous weather vector space mappings to obtain more accurate and comprehensive disastrous weather information. Each mapping can be adjusted and optimized based on different feature extraction methods, mathematical models or parameter settings to improve the accuracy and reliability of the mapping. At the same time, the server can also customize the design and implementation of the mapping process according to actual needs and application scenarios.

[0092] In a possible implementation manner, step S150 includes at least one of the following:

[0093] Step S151, based on the disastrous weather vector and the sample data sequence formed according to each disaster risk area, determine the disastrous weather vector error parameter, optimize the network weight bias information of the climate factor conversion unit according to the disastrous weather vector error parameter, and iterate the network parameter learning according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

[0094] In this embodiment, after obtaining the disastrous weather vector, the server needs to compare it with the sample data sequence composed of known disaster risk areas to determine the error parameter of the disastrous weather vector. These sample data sequences are collected and processed in advance, and contain various meteorological data and related information when disastrous weather occurred in different disaster risk areas in history.

[0095] Specifically, the server will first calculate the numerical difference between the disastrous weather vector and the corresponding position in the sample data sequence. This difference can be measured by indicators such as Euclidean distance and cosine similarity. Then, the server will aggregate and average these difference values ​​to obtain an overall disastrous weather vector error parameter. This parameter reflects the degree of deviation between the current disastrous weather vector and the actual disastrous weather vector, providing an important reference for subsequent network optimization.

[0096] After determining the vector error parameters of severe weather, the server needs to optimize the network weight bias information of the climate factor conversion unit in the severe weather prediction network based on the vector error parameters of severe weather. The climate factor conversion unit is the key component in the network responsible for converting the collaborative observation data into the climate factor guidance vector, and its performance directly affects the prediction accuracy of the entire network.

[0097] In order to optimize the network weight bias information of the climate factor conversion unit, the server can use gradient descent method, stochastic gradient descent method, etc. These algorithms can calculate the gradient value of each weight and bias in the network according to the disastrous weather vector error parameter, and then fine-tune the weight and bias in the direction of the gradient to make the network output closer to the actual disastrous weather situation.

[0098] Through multiple iterations of optimization and network parameter learning, the performance of the climate factor conversion unit will gradually improve, and the prediction accuracy of the severe weather prediction network will also improve accordingly. When the preset network convergence requirements are met, the server can generate a deployable severe weather prediction network.

[0099] Step S152, based on the disastrous weather vector and the sample data sequence formed according to each disaster risk area, determine the disastrous weather vector error parameter, optimize the network weight bias information of the climate factor conversion unit and the network weight bias information of the disastrous vector positioning unit according to the disastrous weather vector error parameter, and iterate the network parameter learning according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

[0100] Similar to step S151, after determining the disastrous weather vector error parameter, the server needs to optimize the network weight bias information of the climate factor conversion unit and the disastrous vector positioning unit in the disastrous weather prediction network according to the disastrous weather vector error parameter. The difference is that in step S152, the server will optimize these two units at the same time to further improve the prediction performance of the network.

[0101] Specifically, the server will first calculate the gradient values ​​of each weight and bias in the climate factor conversion unit and the disaster vector positioning unit, and then fine-tune them according to the direction of the gradient. In this process, the server can use some more advanced optimization algorithms, such as Adam, RMSprop, etc., to speed up the optimization speed and improve the optimization effect.

[0102] By optimizing the climate factor conversion unit and the disaster vector positioning unit and learning the network parameters at the same time, the prediction performance of the disaster weather prediction network will be comprehensively improved. When the preset network convergence requirements are met, the server can generate a more accurate and reliable disaster weather prediction network that can be deployed. This disaster weather prediction network can provide strong support for subsequent disaster weather warnings and decision-making.

[0103] In a possible implementation, the method further includes:

[0104] Step A110, according to the disaster coordination vector extraction unit in the disastrous weather prediction network with the initialized network parameters, parses the coordinated observation data to obtain the disaster coordination vector of the reference satellite-ground monitoring station.

[0105] In this embodiment, when the server performs the disaster weather prediction task, it first uses the disaster coordination vector extraction unit in the initialization network parameters to process the coordinated observation data. The coordinated observation data usually includes meteorological observation data and environmental data from multiple satellite and ground monitoring stations, which are important inputs for disaster weather prediction.

[0106] The disaster synergy vector extraction unit is an important part of the severe weather prediction network. It is responsible for extracting disaster synergy vectors related to severe weather from the collaborative observation data. The disaster synergy vector reflects the mutual relationship and synergy of meteorological factors between different monitoring stations, which is of great significance for accurately predicting severe weather.

[0107] Specifically, the server will input the collaborative observation data into the disaster collaborative vector extraction unit. This disaster collaborative vector extraction unit will use a series of neural network models and algorithms to process and analyze the collaborative observation data to extract key features and collaborative patterns. After processing, the disaster collaborative vector extraction unit will output a disaster collaborative vector, which represents the key information related to disastrous weather in the collaborative observation data in a numerical form.

[0108] Step A120, integrating the disaster coordination vector and the disastrous weather vector to generate an integrated disastrous weather vector.

[0109] After obtaining the disaster coordination vector and the severe weather vector, the server needs to integrate the disaster coordination vector and the severe weather vector to generate a more comprehensive and accurate integrated severe weather vector. The integration process is to fuse and integrate the key information in the disaster coordination vector and the severe weather vector to form a more comprehensive and complete severe weather description.

[0110] Specifically, the server can use some mathematical operations or neural network models to achieve vector integration. For example, the server can perform operations such as splicing or weighted summation of the disaster coordination vector and the disastrous weather vector to obtain a longer integrated disastrous weather vector. This integrated disastrous weather vector contains both the key information in the disastrous weather vector and the coordination patterns and relationships in the coordinated observation data, providing a more comprehensive and accurate data basis for subsequent disastrous weather forecasts.

[0111] Step S150 may include:

[0112] Step S153, determining an integrated disastrous weather vector error parameter based on the integrated disastrous weather vector and a sample data sequence formed according to each disaster risk area.

[0113] After obtaining the integrated severe weather vector, the server needs to compare it with the sample data sequence consisting of known disaster risk areas to determine the error parameter of the integrated severe weather vector. Similar to the previous steps, this comparison process is achieved by calculating the numerical difference between the integrated severe weather vector and the corresponding position in the sample data sequence.

[0114] Specifically, the server can calculate the Euclidean distance, cosine similarity and other indicators between the integrated disastrous weather vector and the sample data sequence to measure the degree of deviation between them. Then, the server will aggregate and average these difference values ​​to obtain an overall integrated disastrous weather vector error parameter. This integrated disastrous weather vector error parameter reflects the degree of deviation between the current integrated disastrous weather vector and the real disastrous weather vector, providing an important reference for subsequent network optimization.

[0115] Step S154, based on the integrated disastrous weather vector error parameter, the network weight bias information of the climate factor conversion unit and the network weight bias information of the disaster coordination vector extraction unit are optimized respectively, and network parameter learning is performed iteratively based on the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

[0116] After determining the integrated severe weather vector error parameter, the server needs to optimize the network weight bias information of the climate factor conversion unit and the disaster coordination vector extraction unit in the severe weather prediction network based on this parameter. This optimization process is achieved by adjusting the weights and biases in the network, with the aim of making the network output closer to the actual severe weather conditions.

[0117] Specifically, the server can use some classic neural network optimization algorithms to calculate the gradient values ​​of each weight and bias in the network, and fine-tune them according to the direction of the gradient. Through multiple iterations of optimization and network parameter learning, the performance of the climate factor conversion unit and the disaster coordination vector extraction unit will gradually improve, and the prediction accuracy of the disastrous weather prediction network will also improve accordingly. When the preset network convergence requirements are met, the server can generate a deployable disastrous weather prediction network.

[0118] Alternatively, in step S154, based on the integrated disastrous weather vector error parameters, the network weight bias information of the climate factor conversion unit, the network weight bias information of the disastrous vector positioning unit, and the network weight bias information of the disaster coordination vector extraction unit are optimized respectively, and network parameter learning is performed iteratively based on the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

[0119] Similar to step S153, after determining the integrated disastrous weather vector error parameter, the server can also optimize the network weight bias information of the climate factor conversion unit, the disastrous vector positioning unit, and the disaster coordination vector extraction unit at the same time. By adjusting the weights and biases in these three units at the same time, the server can further improve the prediction performance of the disastrous weather prediction network.

[0120] Specifically, the server will first calculate the gradient values ​​of each weight and bias in the three units, and then fine-tune them according to the direction of the gradient. In this process, the server can use some more advanced optimization algorithms to speed up the optimization speed and improve the optimization effect. Through multiple iterations of optimization and network parameter learning, when the preset network convergence requirements are met, the server can generate a more accurate and reliable disaster weather prediction network that can be deployed.

[0121] In a possible implementation, the training step of the disaster vector positioning unit includes:

[0122] Step B110, obtaining a disaster vector positioning network with initialized network parameters.

[0123] When the server performs the training task of the disaster vector positioning unit, it first needs to obtain a disaster vector positioning network with initialized network parameters. This disaster vector positioning network is pre-designed and contains a series of neural network layers and initialized weight bias parameters, which constitute the basic structure of the disaster vector positioning network and provide a starting point for subsequent training and optimization.

[0124] Specifically, the server can load the disaster vector localization network with initialized network parameters from a pre-trained model library or model storage location. The loading process may involve reading network structure files, weight files, and bias files to ensure the integrity and accuracy of the network. Once the loading is complete, the server has a disaster vector localization network that can be trained and optimized.

[0125] Step B120, according to the disaster vector positioning network with initialized network parameters, parse and obtain the disaster positioning vector from the meteorological activity information.

[0126] After obtaining the disaster vector positioning network with initialized network parameters, the server needs to use this network to process meteorological activity information. Meteorological activity information usually includes data related to meteorological disasters such as wind speed, wind direction, air pressure, temperature, etc. These data are important inputs for disaster vector positioning.

[0127] Specifically, the server will input the meteorological activity information into the disaster vector location network. This network will use a series of neural network models and algorithms to process and analyze the meteorological activity information, extracting key features and disaster location information. After processing, the disaster vector location network will output a disaster location vector, which represents the key information related to the location of the disaster weather in the meteorological activity information in a numerical form.

[0128] Step B130, determining a disaster location vector error parameter based on the disaster location vector and a sample data sequence formed according to each disaster risk area.

[0129] After obtaining the disaster location vector, the server needs to compare it with the sample data sequence consisting of known disaster risk areas to determine the error parameter of the disaster location vector. This comparison process is achieved by calculating the numerical difference between the disaster location vector and the corresponding position in the sample data sequence.

[0130] Specifically, the server will calculate the Euclidean distance, cosine similarity and other indicators between the disaster location vector and the sample data sequence to measure the degree of deviation between them. Then, the server will summarize and average these difference values ​​to obtain an overall disaster location vector error parameter. This parameter reflects the degree of deviation between the current disaster location vector and the actual disaster risk area data, providing an important reference for subsequent network optimization.

[0131] Step B140, optimizing the network weight bias information of the disaster vector positioning network of the initialized network parameters according to the disaster positioning vector error parameter, and iterating network parameter learning according to the optimized disaster vector positioning network until the preset network convergence requirement is met, thereby generating the disaster vector positioning unit.

[0132] After determining the disaster location vector error parameter, the server needs to optimize the network weight bias information in the disaster vector location network based on this parameter. This optimization process is achieved by adjusting the weights and biases in the network, with the aim of making the network output closer to the actual location of the disaster weather.

[0133] Specifically, the server uses some classic neural network optimization algorithms to calculate the gradient values ​​of each weight and bias in the network, and fine-tune them in the direction of the gradient. Through multiple iterations of optimization and network parameter learning, the performance of the disastrous vector positioning network will gradually improve, and its prediction accuracy of the location of disastrous weather will also improve accordingly. When the preset network convergence requirements are met (such as reaching the preset number of iterations, the error is reduced to below a certain threshold, etc.), the server can generate a trained disastrous vector positioning unit that can be used for actual deployment. This unit can accurately extract the location information of disastrous weather from meteorological activity information, providing strong support for subsequent disaster warning and response.

[0134] In a possible implementation, the method further includes:

[0135] Step C110, obtaining a shielded blending steering vector generated by performing feature shielding processing on the blending steering vector.

[0136] In the server's processing flow, the blended guidance vector is an important data input, which usually contains weather, climate and environmental information from multiple sources and types. In order to more accurately extract the characteristics of severe weather in a specific area, the server will perform feature masking on the blended guidance vector.

[0137] Feature shielding is a data processing technology that shields features in the blending guidance vector that are irrelevant to or have little impact on disastrous weather in a specific area, thereby highlighting key features that are closely related to disastrous weather. This process may involve steps such as data cleaning, feature selection, and weight adjustment.

[0138] After the feature shielding process is completed, the server will generate a shielded blending guidance vector. Compared with the original blending guidance vector, this shielded blending guidance vector focuses more on the key features of disastrous weather, which helps to improve the accuracy and efficiency of subsequent disastrous weather forecasts.

[0139] Step C120: Analyze and obtain the regional disaster vector of the reference satellite-to-ground monitoring station from the shielding blending guidance vector according to the disaster vector positioning unit.

[0140] After obtaining the shielding blending guidance vector, the server will use the previously trained disaster vector positioning unit to process the shielding blending guidance vector. The disaster vector positioning unit is a neural network module specially used to extract the characteristics of disastrous weather from complex data.

[0141] Through the processing of the disaster vector positioning unit, the server can parse the regional disaster vector of the reference satellite-ground monitoring station from the shielding blending guidance vector. This regional disaster vector represents the key characteristics and potential risks related to severe weather in a specific area in a numerical form.

[0142] This step is crucial for severe weather forecasting because it helps the server transform complex intertwined steering vectors into more targeted and actionable regional disaster vectors, providing strong data support for subsequent network optimization and forecasting.

[0143] Step S150 may include: optimizing the network configuration parameters in the disastrous weather prediction network of the initialized network parameters according to the disastrous weather vector, the regional disastrous vector and the regional disastrous vectors of other reference satellite-to-ground monitoring stations except the reference satellite-to-ground monitoring station, and iteratively learning the network parameters according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

[0144] After obtaining the disastrous weather vector, the regional disastrous vector of the reference satellite-ground monitoring station, and the regional disastrous vectors of other reference satellite-ground monitoring stations except the reference satellite-ground monitoring station, the server needs to comprehensively optimize the disastrous weather prediction network with initialized network parameters.

[0145] The optimization process first calculates the difference between the actual output of the network and the expected output based on these vector data, that is, the error. Then, the server uses optimization algorithms (such as gradient descent algorithms) to adjust configuration parameters such as weights and biases in the network to minimize this error.

[0146] Through multiple iterations of optimization and network parameter learning, the performance of the severe weather prediction network will gradually improve. When the network output error drops below the preset threshold or reaches the preset number of iterations, the server will determine that the network has converged and generate a deployable severe weather prediction network.

[0147] This disastrous weather prediction network can make accurate disastrous weather predictions by comprehensively utilizing multi-source data and multi-type features, providing strong support for meteorological disaster warning and prevention. At the same time, because the network has been fully optimized and trained, it has high stability and reliability in practical applications.

[0148] In a possible implementation, step C110 may include at least one of the following:

[0149] Step C111, performing feature shielding processing on the climate factor steering vector to generate a climate factor steering shielding vector, blending the climate factor steering shielding vector with the meteorological activity information to generate a shielding blending steering vector.

[0150] In the data processing flow of the server, the climate factor steering vector is a data set that includes a variety of climate factors (such as long-term average values ​​of temperature, humidity, wind speed, etc.). These data have an important impact on the formation and development of disastrous weather. However, not all climate factors are directly related to disastrous weather in a specific area. Some factors may play a smaller role or even cause interference.

[0151] In order to more accurately focus on the climate factors that are closely related to disastrous weather, the server will perform feature shielding on the climate factor guidance vector. In this process, the server will use pre-set rules or machine learning algorithms to identify and shield those climate factor features that are less relevant to disastrous weather. After this processing, the server generates a climate factor guidance shielding vector, which only retains the key climate factor features that are highly correlated with disastrous weather.

[0152] After obtaining the climate factor-guided shielding vector, the server needs to integrate it with meteorological activity information. Meteorological activity information usually includes real-time observation data such as wind speed, wind direction, air pressure, and temperature, which reflect the current state and behavior of the atmosphere.

[0153] During the fusion process, the server will use certain algorithms or models (such as neural networks, decision trees, etc.) to organically combine the climate factor-guided shielding vector and meteorological activity information. This combination is not a simple data superposition, but through complex calculations and processing, the information in the two can complement and enhance each other. The shielding fusion guidance vector generated after fusion contains both the long-term climate factor characteristics and the short-term meteorological activity status, providing a more comprehensive and accurate data basis for subsequent severe weather forecasts.

[0154] Step C112, performing feature shielding processing on the meteorological activity information to generate meteorological activity shielding information, blending the climate factor steering vector with the meteorological activity shielding information to generate a shielding blending steering vector.

[0155] Similar to the climate factor-oriented vector, there are also some features in the meteorological activity information that are less relevant to the disastrous weather in a specific area. In order to improve the accuracy and efficiency of the prediction, the server also needs to perform feature shielding processing on the meteorological activity information.

[0156] During the processing, the server will use statistical methods, machine learning techniques or expert knowledge to analyze and evaluate the features in the meteorological activity information. Features that are judged to have a low correlation with disastrous weather will be blocked or downgraded to generate a meteorological activity shielding information. This shielding information mainly retains the key meteorological features that are closely related to disastrous weather.

[0157] After obtaining the weather activity shielding information, the server will perform another blending operation, which combines the original climate factor steering vector (unshielded) with the weather activity shielding information that has undergone feature shielding.

[0158] Similar to the previous blending process, the server will use appropriate algorithms or models to process the two data sets. Through the blending process, the server can generate a new shielded blending guidance vector. This vector contains both long-term stable climate factor information and highlights the real-time meteorological characteristics that are closely related to disastrous weather. Such a data structure provides more powerful support for subsequent disastrous weather forecasts.

[0159] In a possible implementation, the optimizing the network configuration parameters in the disastrous weather prediction network of the initialized network parameters according to the disastrous weather vector, the regional disastrous vector, and the regional disastrous vectors of other reference satellite-to-ground monitoring stations except the reference satellite-to-ground monitoring station, and iterating the network parameter learning according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used, including:

[0160] Step D110, determining a disastrous weather vector error parameter based on the disastrous weather vector and a sample data sequence formed according to each disaster risk area.

[0161] The server first receives and analyzes the disastrous weather vector, which is a numerical representation of various disastrous weather characteristics (such as strong winds, heavy rains, hail, etc.). At the same time, the server also stores sample data sequences of various disaster risk areas. These sample data sequences are records of historical disaster events and have important reference value for predicting future disastrous weather.

[0162] The server will compare the disastrous weather vector with the sample data sequence and calculate the difference or error between them. This process may involve calculation methods such as mean square error and absolute error in statistics. Through comparison and analysis, the server can determine the disastrous weather vector error parameter, which reflects the degree of deviation between the current disastrous weather vector and the historical disaster event data.

[0163] Step D120, generating a comparison error parameter according to characteristic distances between the disastrous weather vector and the regional disastrous vector and the regional disastrous vectors of other reference satellite-to-ground monitoring stations except the reference satellite-to-ground monitoring station.

[0164] The server then compares the severe weather vector with two different data sources: the regional severe weather vector, which is a numerical representation of severe weather characteristics for a specific region, and the regional severe weather vectors of other reference satellite-ground monitoring stations in addition to the reference satellite-ground monitoring station. The regional severe weather vector is a numerical representation of severe weather characteristics for a specific region, while the regional severe weather vectors of other reference satellite-ground monitoring stations provide severe weather information over a wider geographical range.

[0165] The server calculates the feature distance between the severe weather vector and the two data sources, which usually involves distance calculation in high-dimensional space, such as Euclidean distance, cosine similarity, etc. Feature distance reflects the similarity or difference between different data in feature space.

[0166] By comparing the characteristic distances, the server can generate a comparison error parameter. This parameter measures the degree of match or difference between the disastrous weather vector and the regional disastrous vectors from different sources. If the comparison error parameter is large, it means that there is a large inconsistency between the current disastrous weather vector and these regional disastrous vectors, and further data correction or model adjustment may be required.

[0167] Step D130, obtaining a global error parameter based on the disastrous weather vector error parameter and the comparison error parameter.

[0168] After obtaining the disastrous weather vector error parameter and the comparative error parameter, the server will perform comprehensive processing to obtain the global error parameter. The global error parameter is a comprehensive indicator that combines the error of the disastrous weather vector itself and the comparative error between the disastrous weather vectors in different regions.

[0169] The server can use weighted average, linear combination or other mathematical methods to calculate the global error parameter. This parameter reflects the overall ability of the current severe weather forecast network to capture and predict severe weather characteristics. The smaller the global error parameter, the better the performance of the forecast network and the more accurate the forecast of severe weather.

[0170] Step D140, optimizing the network configuration parameters in the disastrous weather prediction network with the initialized network parameters according to the global error parameters, and iterating network parameter learning according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

[0171] Finally, the server will use the global error parameters to optimize the network configuration parameters in the severe weather prediction network. The optimization process may involve machine learning techniques such as gradient descent algorithm and back propagation algorithm, as well as the adjustment of neural network structure, weight and bias update and other operations.

[0172] Through continuous iterative optimization and network parameter learning, the performance of the disaster weather prediction network will gradually improve. The server will monitor the output error or performance index of the network and compare it with the preset network convergence requirements. When the network meets the convergence requirements (such as the error is reduced to below a certain threshold or the performance index reaches the preset standard), the server will determine that the network has been trained to maturity and generate a deployable disaster weather prediction network.

[0173] This severe weather prediction network can accurately predict future severe weather based on input severe weather vectors and regional severe weather vectors. By continuously verifying and adjusting network parameters in practical applications, the accuracy and stability of the severe weather prediction network can be further improved.

[0174] In one possible implementation, the method includes:

[0175] Step S160, obtaining the coordinated observation data and meteorological activity information of the target satellite-ground monitoring station. The meteorological activity information is obtained based on the real-time meteorological activity data captured by the target satellite-ground monitoring station for each disaster risk area.

[0176] Step S160, converting the collaborative observation data to obtain a climate factor steering vector, blending the climate factor steering vector with the meteorological activity information to generate a blended steering vector, and obtaining the disastrous weather vector of the target satellite-to-ground monitoring station based on the blended steering vector analysis.

[0177] Step S170, determining disastrous weather forecast information according to the disastrous weather vector, and issuing an early warning of the disastrous weather forecast information to the disaster monitoring station of the target satellite-to-ground monitoring station.

[0178] In this embodiment, the server first establishes a data connection with the target satellite-ground monitoring station. The target satellite-ground monitoring station is a ground station equipped with advanced observation equipment, which can capture various real-time data related to the weather. These data include but are not limited to temperature, humidity, wind speed, wind direction, air pressure, etc.

[0179] The server receives the collaborative observation data from the target satellite and ground monitoring station in real time through a dedicated data transmission protocol. These data have been preliminarily processed and analyzed and have a certain structure and format, which is convenient for subsequent processing and analysis.

[0180] At the same time, the server also receives real-time meteorological activity data captured by target satellite and ground monitoring stations for each disaster risk area. These data are real-time records of meteorological phenomena occurring in a specific area and have important reference value for predicting future disastrous weather.

[0181] After receiving the collaborative observation data, the server will perform a series of data conversion and processing operations. The purpose of these operations is to convert the original observation data into a data form that is easier to analyze and process, namely, the climate factor-oriented vector.

[0182] The climate factor steering vector is a numerical representation of multiple climate factors (such as long-term average temperature, humidity, wind speed, etc.). The server will use a preset algorithm or model to extract features, normalize, reduce dimensions, and other processing on the collaborative observation data to generate a climate factor steering vector. This process may involve complex mathematical calculations and data processing techniques.

[0183] After obtaining the climate factor steering vector and meteorological activity information, the server will perform a fusion process. The purpose of the fusion is to organically combine the data from these two different sources to generate a more comprehensive and accurate data representation, namely the fusion steering vector.

[0184] During the fusion process, the server will use specific algorithms or models (such as neural networks, decision trees, etc.) to perform fusion calculations on the climate factor guidance vector and meteorological activity information. This calculation is not a simple data superposition, but through complex mathematical operations and data processing technology, the information in the two can complement and enhance each other. The fusion guidance vector generated after fusion contains both the long-term climate factor characteristics and the short-term meteorological activity status.

[0185] After obtaining the blended guidance vector, the server will further perform analysis to obtain the disastrous weather vector of the target satellite-ground monitoring station. The disastrous weather vector is a numerical representation that includes a variety of disastrous weather characteristics (such as heavy rain, strong wind, hail, etc.).

[0186] During the analysis process, the server will use preset rules or models to analyze and identify the convergence guidance vector. These rules or models may be based on historical disaster data, expert knowledge, or machine learning algorithms. By analyzing the characteristic patterns and associations in the convergence guidance vector, the server can extract key information related to disastrous weather and generate a disastrous weather vector.

[0187] Finally, the server will determine the disastrous weather forecast information based on the disastrous weather vector. This information may include the type of disaster, expected time of occurrence, expected impact range, possible degree of harm, etc. The server will use the preset threshold or model to evaluate and judge the disastrous weather vector to generate the corresponding forecast information.

[0188] Once the disastrous weather forecast information is determined, the server will immediately send it to the disaster monitoring station of the target satellite-ground monitoring station for early warning. The early warning information may be sent through a dedicated data transmission channel, SMS, email or other communication methods to ensure that relevant personnel can receive the early warning information in time and take corresponding response measures. In this way, the server completes a complete process from data acquisition to disaster early warning.

[0189] Figure 2 The hardware structure of the disaster weather prediction system 100 provided in the embodiment of the present application for implementing the above-mentioned disaster weather prediction method based on satellite-ground collaboration is shown. Figure 2 As shown, the severe weather prediction system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0190] In an alternative embodiment, the disaster weather prediction system 100 can be a single server or a server group. The server group can be centralized or distributed (for example, the disaster weather prediction system 100 can be a distributed system). In an alternative embodiment, the disaster weather prediction system 100 can be local or remote. For example, the disaster weather prediction system 100 can access information and / or data stored in a machine-readable storage medium 120 via a network. For another example, the disaster weather prediction system 100 can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In an alternative embodiment, the disaster weather prediction system 100 can be implemented on a cloud platform. By way of example only, the cloud platform can include a private cloud, a public cloud, an aggregated cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any integration thereof.

[0191] The machine-readable storage medium 120 can store data and / or instructions. In an alternative embodiment, the machine-readable storage medium 120 can store data obtained from an external terminal. In an alternative embodiment, the machine-readable storage medium 120 can store data and / or instructions used by the disaster weather prediction system 100 to execute or use to complete the exemplary method described in this application. In an alternative embodiment, the machine-readable storage medium 120 may include a large-capacity memory, a removable memory, a volatile read-write memory, a read-only memory, etc. or any integration thereof. Exemplary large-capacity memory may include a magnetic disk, an optical disk, a solid-state disk, etc. Exemplary removable memory may include a flash drive, a floppy disk, an optical disk, a memory card, a compressed disk, a magnetic tape, etc.

[0192] During the specific implementation process, multiple processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the disastrous weather prediction method based on satellite-ground collaboration in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0193] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned disaster weather prediction system 100. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0194] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are pre-installed. When the processor executes the computer executable instructions, the above-mentioned disastrous weather prediction method based on satellite-ground collaboration is implemented.

[0195] Similarly, it should be noted that in order to simplify the description disclosed in this application, thereby helping to understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description disclosed in this application, thereby helping to understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof.

Claims

1. A method for forecasting disastrous weather based on satellite-ground collaboration, characterized in that: The method comprises: Acquiring collaborative observation data and meteorological activity information of reference satellite and ground monitoring stations; the meteorological activity information is obtained based on real-time meteorological activity data captured by the reference satellite and ground monitoring stations for each disaster risk area; The collaborative observation data is converted by a climate factor conversion unit in the disastrous weather prediction network with initialized network parameters to generate a climate factor steering vector; The climate factor steering vector is blended with the meteorological activity information to generate a blended steering vector, wherein the climate factor steering vector and the meteorological activity information are first normalized so that the climate factor steering vector and the meteorological activity information are in the same dimension and range, and the climate factor steering vector and the meteorological activity information are fused using a feature fusion technique, wherein the feature fusion technique includes weighted summation, series or parallel operation of features, and a dimensionality reduction technique is used to reduce the dimensionality of the fused data, and the reduced dimensionality data is used as the blended steering vector; According to the disastrous vector positioning unit in the disastrous weather prediction network of the initialized network parameters, the disastrous weather vector of the reference satellite-ground monitoring station is obtained by parsing from the blended steering vector; Optimizing the network configuration parameters in the disastrous weather prediction network with the initialized network parameters according to the disastrous weather vector, and iteratively learning the network parameters of the disastrous weather prediction network with the initialized network parameters according to the optimized network configuration parameters until a preset network convergence requirement is met, thereby generating a disastrous weather prediction network that can be deployed for use; The climatic factor conversion unit in the disastrous weather prediction network that initializes the network parameters converts the collaborative observation data to generate a climatic factor steering vector, including: Performing self-attention feature extraction on the collaborative observation data to generate a self-attention climate vector for the collaborative observation data; Loading the self-attention climate vector into a climate factor conversion unit in a disastrous weather prediction network that initializes network parameters for conversion to generate a climate factor steering vector; The disastrous weather vector positioning unit in the disastrous weather prediction network according to the initialized network parameters obtains the disastrous weather vector of the reference satellite-ground monitoring station from the blended steering vector by parsing, including: Loading the blended guidance vector into the disaster vector positioning unit in the disaster weather prediction network with initialized network parameters to extract features and generate a disaster vector knowledge network; The disastrous weather vector positioning unit maps the disastrous weather vector space at least once according to the disastrous vector knowledge network to generate the disastrous weather vector of the reference satellite-ground monitoring station; The step of optimizing the network configuration parameters in the disastrous weather prediction network with initialized network parameters according to the disastrous weather vector, and iteratively learning the network parameters of the disastrous weather prediction network with initialized network parameters according to the optimized network configuration parameters until a preset network convergence requirement is met, thereby generating a disastrous weather prediction network that can be deployed and used, includes at least one of the following: Based on the disastrous weather vector and the sample data sequence formed according to each disaster risk area, a disastrous weather vector error parameter is determined, the network weight bias information of the climate factor conversion unit is optimized according to the disastrous weather vector error parameter, and network parameter learning is performed iteratively according to the optimized disastrous weather prediction network until a preset network convergence requirement is met, thereby generating a disastrous weather prediction network that can be deployed and used; Based on the disastrous weather vector and a sample data sequence formed according to each disaster risk area, a disastrous weather vector error parameter is determined, and the network weight bias information of the climate factor conversion unit and the network weight bias information of the disastrous vector positioning unit are optimized respectively according to the disastrous weather vector error parameter, and network parameter learning is iterated according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

2. The method for forecasting disastrous weather based on satellite-ground coordination according to claim 1, characterized in that: The collaborative observation data includes multiple; The step of loading the self-attention climate vector into a climate factor conversion unit in a disastrous weather prediction network that initializes network parameters for conversion to generate a climate factor steering vector comprises: Based on the self-attention climate vectors corresponding to each of the collaborative observation data, the self-attention climate vectors are arranged in the spatiotemporal domain to generate a corresponding set of self-attention climate vectors; The self-attention climate vector set is loaded into the climate factor conversion unit in the disastrous weather prediction network with initialized network parameters for conversion to generate a climate factor steering vector; the climate factor steering vector includes a preset number of climate factor steering elements.

3. The method for forecasting disastrous weather based on satellite-ground coordination according to claim 1, characterized in that: The method further comprises: The disaster coordination vector extraction unit in the disastrous weather prediction network of the initialized network parameters parses and obtains the disaster coordination vector of the reference satellite-ground monitoring station from the coordinated observation data, wherein the disaster coordination vector reflects the mutual relationship and synergy of meteorological factors between different monitoring stations; Integrating the disaster coordination vector with the disastrous weather vector to generate an integrated disastrous weather vector; The method of optimizing the network configuration parameters in the disastrous weather prediction network with initialized network parameters according to the disastrous weather vector, and iteratively learning the network parameters of the disastrous weather prediction network with initialized network parameters according to the optimized network configuration parameters until a preset network convergence requirement is met, thereby generating a disastrous weather prediction network that can be deployed and used, includes: Determining an integrated disastrous weather vector error parameter based on the integrated disastrous weather vector and a sample data sequence formed according to each disaster risk area; According to the integrated disastrous weather vector error parameter, the network weight bias information of the climate factor conversion unit and the network weight bias information of the disaster coordination vector extraction unit are optimized respectively, and network parameter learning is performed iteratively according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used; Alternatively, based on the integrated disastrous weather vector error parameters, the network weight bias information of the climate factor conversion unit, the network weight bias information of the disastrous vector positioning unit, and the network weight bias information of the disaster coordination vector extraction unit are optimized respectively, and network parameter learning is performed iteratively based on the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a deployable disastrous weather prediction network.

4. The method for forecasting disastrous weather based on satellite-ground coordination according to claim 1, characterized in that: The training steps of the disaster vector positioning unit include: Obtaining a disaster vector positioning network with initialized network parameters; According to the disaster vector positioning network with initialized network parameters, a disaster positioning vector is obtained by parsing from the meteorological activity information; Determining a disaster location vector error parameter based on the disaster location vector and a sample data sequence formed according to each disaster risk area; The network weight bias information of the disaster vector positioning network of the initialized network parameters is optimized according to the disaster positioning vector error parameter, and the network parameter learning is performed iteratively according to the optimized disaster vector positioning network until the preset network convergence requirement is met, thereby generating the disaster vector positioning unit.

5. The method for forecasting disastrous weather based on satellite-ground collaboration according to any one of claims 1 to 4, characterized in that: The method further comprises: Acquire a shielded blending guidance vector generated by performing feature shielding processing on the blending guidance vector; According to the disaster vector positioning unit, the regional disaster vector of the reference satellite-ground monitoring station is obtained by analyzing from the shielding blending guidance vector; The method of optimizing the network configuration parameters in the disastrous weather prediction network with initialized network parameters according to the disastrous weather vector, and iteratively learning the network parameters of the disastrous weather prediction network with initialized network parameters according to the optimized network configuration parameters until a preset network convergence requirement is met, thereby generating a disastrous weather prediction network that can be deployed and used, includes: According to the disastrous weather vector, the regional disastrous vector and the regional disastrous vectors of other reference satellite-to-ground monitoring stations except the reference satellite-to-ground monitoring station, the network configuration parameters in the disastrous weather prediction network of the initialized network parameters are optimized, and network parameter learning is iterated according to the optimized disastrous weather prediction network until a preset network convergence requirement is met, thereby generating a disastrous weather prediction network that can be deployed for use; The step of obtaining a shielded blending steering vector generated by performing feature shielding processing on the blending steering vector includes at least one of the following: Performing feature shielding processing on the climate factor steering vector to generate a climate factor steering shielding vector; Blending the climate factor-oriented shielding vector with the meteorological activity information to generate a shielding-blending steering vector; Performing feature shielding processing on the meteorological activity information to generate meteorological activity shielding information; Blending the climate factor steering vector with the meteorological activity shielding information to generate a shielding blending steering vector; The method includes optimizing the network configuration parameters in the disastrous weather prediction network of the initialized network parameters according to the disastrous weather vector, the regional disastrous vector and the regional disastrous vectors of other reference satellite-to-ground monitoring stations except the reference satellite-to-ground monitoring station, and iteratively learning the network parameters according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used, including: Determining a disastrous weather vector error parameter based on the disastrous weather vector and a sample data sequence formed according to each disaster risk area; Generate a comparison error parameter according to characteristic distances between the disastrous weather vector and the regional disastrous vector and regional disastrous vectors of other reference satellite-to-ground monitoring stations except the reference satellite-to-ground monitoring station; Obtaining a global error parameter based on the disastrous weather vector error parameter and the comparison error parameter; The network configuration parameters in the disastrous weather prediction network of the initialized network parameters are optimized according to the global error parameters, and the network parameter learning is iteratively performed according to the optimized disastrous weather prediction network until the preset network convergence requirements are met, thereby generating a disastrous weather prediction network that can be deployed and used.

6. The method for forecasting disastrous weather based on satellite-ground collaboration according to claim 1, characterized in that: The method comprises: Acquiring coordinated observation data and meteorological activity information of target satellite-ground monitoring stations; the meteorological activity information is obtained based on real-time meteorological activity data captured by the target satellite-ground monitoring stations for each disaster risk area; Converting the collaborative observation data to obtain a climate factor steering vector; Blending the climate factor steering vector with the meteorological activity information to generate a blended steering vector; Obtaining the disastrous weather vector of the target satellite-to-ground monitoring station according to the blended steering vector analysis; Disaster weather forecast information is determined based on the disastrous weather vector, and the disastrous weather forecast information is warned to the disaster monitoring station of the target satellite-to-ground monitoring station.

7. A disaster weather prediction system, characterized in that: The disastrous weather prediction system includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the disastrous weather prediction method based on satellite-ground collaboration as described in any one of claims 1-6.

Images