Multi-source data disaster automatic early warning system and method for rural meteorological safety

Abstract

The invention relates to the technical field of automatic early warning of disasters, and particularly discloses a multi-source data automatic early warning system and method for rural meteorological safety, which can monitor and collect multi-source meteorological data of a rural predetermined area in real time, and can perform early warning on disasters when early warning the disasters. Multi-source meteorological time sequence data and remote sensing image data of a rural predetermined area are input into a data processing and image semantic comprehension algorithm based on artificial intelligence and deep learning together for analysis, so that cross-modal response features between multi-source meteorological time sequence association semantics and remote sensing image semantics are captured; the method is used for performing intelligent disaster early warning in a rural predetermined area. Therefore, the rural meteorological safety data can be monitored and analyzed in real time, and the multi-source data is utilized to automatically carry out ground disaster early warning, so that the defects in the aspects of spatial distribution, time resolution, real-time performance and the like caused by the adoption of a single data source in a traditional disaster early warning scheme are overcome.

Description

Technical Field

[0001] The present application relates to the technical field of automatic disaster warning, and more specifically, to a multi-source data automatic disaster warning system and method for rural meteorological safety. Background Art

[0002] As an important part of the power system, the operating efficiency and safety of thermal power plants are directly related to the stability and reliability of power supply. As one of the core equipment of thermal power plants, the speed control of steam turbines is crucial. Accurate speed measurement not only affects the power generation efficiency, but also has a decisive impact on the safe operation of the equipment. In actual operation, changes in ambient temperature will have a significant impact on the speed measurement of steam turbines, resulting in measurement errors, which may cause unnecessary adjustments or failures.

[0003] Traditional temperature compensation methods mostly rely on fixed parameters or empirical formulas, which are difficult to adapt to complex and changeable actual operating conditions, especially under extreme temperature conditions, and the effects of these methods are often poor. Specifically, traditional methods are usually based on fixed temperature compensation coefficients, which are measured under specific conditions (such as laboratory environments). During the operation of the steam turbine, the ambient temperature may change rapidly, and the fixed parameters cannot be adjusted in real time to cope with such dynamic changes. Secondly, empirical formulas are usually derived from historical data or theoretical models and are suitable for specific types of equipment or specific operating conditions. When applied to different equipment or different operating conditions, these formulas may no longer be valid.

[0004] Therefore, an optimized auxiliary control solution for thermal power plants is desired, which can perform temperature-compensated speed measurement correction in a more intelligent way to adapt to complex and changeable actual operating conditions. Summary of the invention

[0005] The present application provides a multi-source data disaster automatic warning system and method for rural meteorological safety, which can monitor and analyze rural meteorological safety data in real time, and use multi-source data to automatically issue geological disaster warnings, thereby overcoming the deficiencies in spatial distribution, temporal resolution, real-time performance, etc. caused by the use of a single data source in traditional disaster warning schemes, thereby helping to increase the real-time response capability and intelligence level of the automatic disaster warning system.

[0006] In the first aspect, a multi-source data disaster automatic early warning system for rural meteorological safety is provided, comprising:

[0007] A meteorological data collection module, used to obtain a time queue of multi-source meteorological data of a predetermined area collected by a sensor network, wherein the multi-source meteorological data includes wind direction, wind speed value, temperature value, air pressure value, rainfall, river water level and reservoir water level;

[0008] A remote sensing image data acquisition module, used for extracting remote sensing image data of the predetermined area from a remote sensing image database;

[0009] A meteorological time series coding module, used for performing time series coding on the time queue of the multi-source meteorological data to obtain multi-source meteorological time series associated implicit coding features;

[0010] A geomorphic feature extraction module is used to extract geomorphic features from the remote sensing image data of the predetermined area to obtain semantic coding features of the geomorphic remote sensing image of the predetermined area;

[0011] A geomorphic-meteorological cross-modal interaction analysis module is used to perform cross-modal interaction analysis on the multi-source meteorological time series associated implicit coding features and the semantic coding features of the geomorphic remote sensing images of the predetermined area to obtain geomorphic-meteorological cross-modal response semantic coding features, wherein the geomorphic-meteorological cross-modal interaction analysis module includes: a meteorological time series autocorrelation coding unit, used to perform autocorrelation coding on the multi-source meteorological time series associated implicit coding features to obtain multi-source meteorological time series semantic autocorrelation coding features; a geomorphic remote sensing image semantic decomposition unit, used to perform feature decoupling on the semantic coding features of the geomorphic remote sensing images of the predetermined area to obtain a set of semantic local features of the geomorphic remote sensing images of the predetermined area; a geomorphic-meteorological semantic cross-modal response unit, used to perform cross-modal interaction optimization on the set of the multi-source meteorological time series semantic autocorrelation coding features and the semantic local features of the geomorphic remote sensing images of the predetermined area to obtain the geomorphic-meteorological cross-modal response semantic coding features;

[0012] The geological disaster safety warning module is used to perform intelligent disaster warning based on the landform-meteorological cross-modal response semantic coding features to determine whether to generate a geological disaster warning prompt.

[0013] In the second aspect, a multi-source data disaster automatic early warning method for rural meteorological safety is provided, comprising:

[0014] Acquire a time queue of multi-source meteorological data of a predetermined area collected by a sensor network, wherein the multi-source meteorological data includes wind direction, wind speed value, temperature value, air pressure value, rainfall, river water level and reservoir water level;

[0015] Extracting remote sensing image data of the predetermined area from a remote sensing image database;

[0016] Performing time series coding on the time queue of the multi-source meteorological data to obtain multi-source meteorological time series associated implicit coding features;

[0017] Extracting geomorphic features from the remote sensing image data of the predetermined area to obtain semantic coding features of the geomorphic remote sensing image of the predetermined area;

[0018] The multi-source meteorological time series associated implicit coding features and the predetermined area landform remote sensing image semantic coding features are cross-modal interactively analyzed to obtain landform-meteorological cross-modal response semantic coding features, including: performing autocorrelation coding on the multi-source meteorological time series associated implicit coding features to obtain multi-source meteorological time series semantic autocorrelation coding features; performing feature decoupling on the predetermined area landform remote sensing image semantic coding features to obtain a set of predetermined area landform remote sensing image semantic local features; performing cross-modal interactive optimization on the multi-source meteorological time series semantic autocorrelation coding features and the predetermined area landform remote sensing image semantic local features to obtain the landform-meteorological cross-modal response semantic coding features;

[0019] Based on the landform-meteorological cross-modal response semantic coding features, intelligent disaster warning is performed to determine whether a geological disaster warning prompt is generated.

[0020] The present application provides a multi-source data disaster automatic early warning system and method for rural meteorological safety, which can collect multi-source meteorological data of a predetermined rural area through real-time monitoring. When warning of disasters, these multi-source meteorological time series data and remote sensing image data of the predetermined rural area are input together into a data processing and image semantic understanding algorithm based on artificial intelligence and deep learning for analysis, so as to capture the cross-modal response characteristics between the multi-source meteorological time series association semantics and the remote sensing image semantics, so as to carry out intelligent early warning of disasters in the predetermined rural area. In this way, rural meteorological safety data can be monitored and analyzed in real time, and multi-source data can be used to automatically warn of geological disasters, thereby overcoming the deficiencies in spatial distribution, temporal resolution, real-time performance, etc. caused by the use of a single data source in traditional disaster early warning schemes, thereby helping to increase the real-time response capability and intelligence level of the automatic disaster early warning system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0022] Figure 1 This is a schematic block diagram of a multi-source data disaster automatic warning system for rural meteorological safety according to an embodiment of the present application.

[0023] Figure 2 This is a data flow diagram of a multi-source data disaster automatic warning system for rural meteorological safety according to an embodiment of the present application.

[0024] Figure 3 This is a schematic block diagram of a geomorphic-meteorological cross-modal interactive analysis module in a multi-source data disaster automatic warning system for rural meteorological safety in an embodiment of the present application.

[0025] Figure 4 This is a schematic block diagram of a geomorphic-meteorological semantic cross-modal response unit in a multi-source data disaster automatic warning system for rural meteorological safety in an embodiment of the present application.

[0026] Figure 5 This is a schematic block diagram of a multi-source data disaster automatic warning method for rural meteorological safety according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0028] In view of the above technical problems, in the technical solution of this application, a multi-source data disaster automatic early warning system for rural meteorological safety is proposed, which can collect multi-source meteorological data of a predetermined rural area through real-time monitoring. When warning of disasters, these multi-source meteorological time series data and remote sensing image data of the predetermined rural area are input into the data processing and image semantic understanding algorithm based on artificial intelligence and deep learning for analysis, so as to capture the cross-modal response characteristics between the multi-source meteorological time series association semantics and the remote sensing image semantics, so as to carry out intelligent early warning of disasters in the predetermined rural area. In this way, rural meteorological safety data can be monitored and analyzed in real time, and multi-source data can be used to automatically warn of geological disasters, thereby overcoming the deficiencies in spatial distribution, temporal resolution, real-time performance, etc. caused by the use of a single data source in traditional disaster early warning schemes, and thus helping to increase the real-time response capability and intelligence level of the automatic disaster early warning system.

[0029] Specifically, in the technical solution of the present application, if Figure 1 and Figure 2As shown, the multi-source data disaster automatic early warning system for rural meteorological safety includes: a meteorological data acquisition module 10, which is used to obtain a time queue of multi-source meteorological data of a predetermined area collected by a sensor network, wherein the multi-source meteorological data includes wind direction, wind speed value, temperature value, air pressure value, rainfall, river water level and reservoir water level; a remote sensing image data acquisition module 20, which is used to extract remote sensing image data of the predetermined area from a remote sensing image database; a meteorological time series encoding module 30, which is used to time series encode the time queue of the multi-source meteorological data to obtain a multi-source meteorological time series associated implicit encoding. code features; a geomorphic feature extraction module 40, used to extract geomorphic features from the remote sensing image data of the predetermined area to obtain semantic coding features of the geomorphic remote sensing images of the predetermined area; a geomorphic-meteorological cross-modal interaction analysis module 50, used to perform cross-modal interaction analysis on the multi-source meteorological time series associated implicit coding features and the semantic coding features of the geomorphic remote sensing images of the predetermined area to obtain geomorphic-meteorological cross-modal response semantic coding features; a geological disaster safety warning module 60, used to perform intelligent disaster warning based on the geomorphic-meteorological cross-modal response semantic coding features to determine whether to generate a geological disaster warning prompt.

[0030] Exemplarily, in the meteorological data acquisition module 10, a time queue of multi-source meteorological data of a predetermined area collected by a sensor network is obtained, wherein the multi-source meteorological data includes wind direction, wind speed value, temperature value, air pressure value, rainfall, river water level and reservoir water level. It should be understood that these multi-source meteorological data can provide detailed information about weather changes, which helps to more accurately predict possible natural disasters such as floods, landslides, etc. By monitoring the changes of these key meteorological parameters in real time, potential dangerous situations can be quickly identified and alarms can be issued in time, thereby improving the real-time response capability of the system. Traditional disaster warning schemes usually rely on a single data source and have deficiencies in spatial distribution, temporal resolution and real-time performance. Multi-source data fusion can make up for these defects and provide more comprehensive and accurate information support. Using artificial intelligence and deep learning technology to process these multi-source time series data can dig out complex patterns and associations, provide a scientific basis for intelligent warning, and assist relevant departments in making faster and better response measures.

[0031] In one embodiment, obtaining a time queue of multi-source meteorological data of a predetermined area collected by a sensor network includes: first, multiple types of sensors need to be deployed in the target area to form a dense monitoring network. These sensors can measure different environmental parameters, such as wind direction, wind speed, temperature, air pressure, precipitation, etc., and special water level monitoring stations are set up at key locations of rivers and reservoirs. Such a layout can ensure comprehensive and detailed monitoring of meteorological conditions in the entire predetermined area. The selection and layout of sensors should take into account the coverage, installation location, and factors that may affect the measurement accuracy, such as terrain features or human activity interference. Next, it is crucial to establish a stable and reliable data transmission infrastructure. Each sensor node should have wireless communication capabilities and send the collected data to a central processing center or cloud platform through the Internet of Things (IoT) technology or other appropriate communication protocols. Redundant design and technical means can be used here to ensure that the continuity of data transmission can be maintained even in network failures or other abnormal situations. In addition, considering that there may be unstable network signals in rural areas, satellite communications can also be considered as a supplementary solution to ensure that data can be effectively uploaded even in remote areas. Then there is data management work. Since a large amount of historical and real-time data is involved, a powerful database management system is required for storage and maintenance. This is not only to save the original observations, but more importantly to provide support for subsequent data analysis. The design of the database should take into account factors such as query efficiency, scalability, and security, and should also be equipped with necessary backup mechanisms to prevent data loss. For the incoming data stream, preliminary quality control is also required, such as removing obvious errors or unreasonable values ​​to improve the reliability of subsequent analysis results.

[0032] Exemplarily, in the remote sensing image data acquisition module 20, the remote sensing image data of the predetermined area is extracted from the remote sensing image database. It should be understood that by combining remote sensing image data with multi-source meteorological time series data, a more comprehensive environmental model can be constructed, which not only takes into account the spatial distribution, but also takes into account the changes in the time dimension. This helps to deeply understand the complex interaction between meteorological conditions and geographical environment, such as phenomena such as mountains guiding airflow or lakes affecting local humidity, which together determine the meteorological changes and extreme weather disaster events that may occur in a specific area. Specifically, remote sensing images provide static feature information such as surface features and terrain undulations, while meteorological time series association semantics include dynamic changes in meteorological data in the time dimension. Combining and analyzing these two different modal data can capture more details, improve the depth of understanding of complex environmental phenomena, and provide support for more accurate disaster prediction. In addition, remote sensing images can also help identify potential risk areas, such as places prone to landslides or low-lying areas in cities prone to waterlogging, so as to take preventive measures in advance.

[0033] Exemplarily, in the meteorological time series encoding module 30, the time queue of the multi-source meteorological data is time-series encoded to obtain multi-source meteorological time series associated implicit coding features. It should be understood that, considering that multi-source meteorological data are essentially time series data, they have temporal continuity and correlation. RNN and its variant LSTM are very suitable for processing such data because they can remember the information of previous moments and the correlation relationship in the time dimension when processing the data at the current moment. For meteorological forecasting, past weather conditions and time series correlation have an important impact on future weather. For example, a continuous rainfall may be affected by the accumulation of humidity in the previous few days. Therefore, the use of RNN-LSTM hybrid model can better utilize this historical information. Based on this, in one embodiment of the present application, the meteorological time series encoding module is used to: input the time queue of the multi-source meteorological data into a multi-source meteorological time series encoder based on the RNN-LSTM hybrid model to obtain a multi-source meteorological time series associated implicit coding feature vector as the multi-source meteorological time series associated implicit coding feature. Through the processing of the multi-source meteorological time series encoder based on the RNN-LSTM hybrid model, the dynamic change characteristics and time series dependencies of each meteorological data source in the time dimension can be captured, and the time series correlation semantics between different meteorological data sources can be understood to provide a more comprehensive multi-source meteorological data time series understanding, and enable the model processing process to achieve a better performance balance, thereby enhancing the flexibility and adaptability of the model.

[0034] Exemplarily, in the geomorphic feature extraction module 40, the remote sensing image data of the predetermined area is subjected to geomorphic feature extraction to obtain semantic coding features of the geomorphic remote sensing image of the predetermined area. It should be understood that the remote sensing image provides static feature information about surface features, terrain undulations, etc., while the meteorological time series association semantics contains the dynamic changes of meteorological data in the time dimension. Combining and analyzing the data of these two different modes can capture more details, improve the depth of understanding of complex environmental phenomena, and provide support for more accurate disaster prediction. In addition, by extracting geomorphic features from remote sensing images, it can also help identify potential risk areas, such as places prone to landslides or low-lying areas in cities prone to waterlogging, so as to take preventive measures in advance. In order to achieve this goal, it is necessary to use a specially designed geomorphic feature extraction algorithm (for example, based on a void convolutional neural network model) to mine geomorphic features of the selected remote sensing image data, extract the semantic implicit feature information of the geomorphic remote sensing image of the predetermined area, and obtain the semantic coding feature map of the geomorphic remote sensing image of the predetermined area. This process can effectively remove redundant information, reduce noise interference, and ensure that the generated feature representation is both compact and representative, thereby improving the performance of the model.

[0035] In one embodiment, the geomorphic feature extraction module is used to: input the remote sensing image data of the predetermined area into the geomorphic feature extractor of the predetermined area based on the hole convolution neural network model to obtain the semantic coding feature map of the geomorphic remote sensing image of the predetermined area as the semantic coding feature of the geomorphic remote sensing image of the predetermined area. That is, the remote sensing image data of the predetermined area is input into the geomorphic feature extractor of the predetermined area based on the hole convolution neural network model to perform feature mining, so as to extract the semantic implicit feature information of the geomorphic remote sensing image of the predetermined area from the remote sensing image data of the predetermined area, thereby obtaining the semantic coding feature map of the geomorphic remote sensing image of the predetermined area.

[0036] In a specific embodiment, the remote sensing image data of the predetermined area is input into a predetermined area geomorphic feature extractor based on a hole convolutional neural network model to obtain a semantic coding feature map of the predetermined area geomorphic remote sensing image as the semantic coding feature of the predetermined area geomorphic remote sensing image, including: constructing and training a geomorphic feature extractor based on ACNN. Hollow convolution is a special convolution method that expands the receptive field without increasing the number of parameters by inserting "holes" (i.e. skipping some pixels) between standard convolution kernels. This method is particularly suitable for tasks that require maintaining a high spatial resolution while capturing a wider range of contextual information. When designing a network architecture, in addition to using a hole convolution layer, other types of convolution layers, pooling layers, activation functions and other components can also be combined to form a multi-layer cascade structure. In order to improve generalization ability and robustness, data enhancement techniques can be introduced during the training process, such as random flipping, rotation, scaling and other transformations, and supervised learning can be performed using a large number of labeled samples. When the model training is completed, the preprocessed remote sensing image can be sent to the geomorphic feature extractor. Inside the network, each layer gradually abstracts a higher level of geomorphic feature representation until it finally outputs a semantically encoded feature map of a geomorphic remote sensing image containing rich semantic information. This feature map not only retains the important visual elements of the original image, but also incorporates the spatial patterns and structural relationships automatically learned by the deep learning algorithm.

[0037] Exemplarily, in the landform-meteorological cross-modal interaction analysis module 50, the multi-source meteorological time series associated implicit coding features and the predetermined area landform remote sensing image semantic coding features are cross-modal interactively analyzed to obtain landform-meteorological cross-modal response semantic coding features. It should be understood that since the predetermined area landform remote sensing image semantic coding feature map and the multi-source meteorological time series associated implicit coding feature vector respectively contain the landform remote sensing image semantic features and multi-source meteorological time series associated features of the predetermined area, the landform remote sensing image semantics provide static feature information such as surface features and terrain undulations, while the meteorological time series associated semantics contain weather conditions about the dynamic changes in the time series of meteorological data in the time dimension. In addition, there are complex interactions between meteorological conditions and geographical environment. For example, mountains may guide airflow, and lakes may affect local humidity. These factors jointly determine the meteorological changes and extreme weather disaster events that may occur in a specific area. Therefore, by combining the data of these two modes, a more comprehensive environmental model can be constructed, which not only takes into account the spatial distribution, but also takes into account the changes in the time dimension. Based on this, in order to make full use of the complementary information between these two different modal semantic features about rural meteorology, so as to improve the understanding of complex environmental phenomena and the ability to predict geological disasters, in the technical solution of the present application, the multi-source meteorological time series associated implicit coding features and the semantic coding features of the predetermined area geomorphic remote sensing images are further subjected to cross-modal interactive analysis to obtain geomorphic-meteorological cross-modal response semantic coding features. In particular, the process of cross-modal interactive analysis, by introducing a cross-modal prompt gate mechanism, can deeply explore and model the complex interaction patterns and correlation relationships between multi-source meteorological time series associated semantics and geomorphic remote sensing image semantics, and realize efficient feature interaction and optimized coding between data features from different modalities, so as to improve the effect of multi-modal data fusion and provide support for more accurate disaster warnings.

[0038] In one embodiment, Figure 3 As shown, the landform-meteorological cross-modal interaction analysis module 50 includes: a meteorological time series autocorrelation coding unit 51, which is used to perform autocorrelation coding on the multi-source meteorological time series associated implicit coding features to obtain multi-source meteorological time series semantic autocorrelation coding features; a landform remote sensing image semantic decomposition unit 52, which is used to perform feature decoupling on the semantic coding features of the landform remote sensing images of the predetermined area to obtain a set of semantic local features of the landform remote sensing images of the predetermined area; a landform-meteorological semantic cross-modal response unit 53, which is used to perform cross-modal interaction optimization on the set of the multi-source meteorological time series semantic autocorrelation coding features and the semantic local features of the landform remote sensing images of the predetermined area to obtain the landform-meteorological cross-modal response semantic coding features.

[0039] In one embodiment, the meteorological time series autocorrelation coding unit 51 is used to calculate the product between the multi-source meteorological time series associated implicit coding feature vector and the transposed vector of the multi-source meteorological time series associated implicit coding feature vector to obtain a multi-source meteorological time series associated semantic autocorrelation coding matrix as the multi-source meteorological time series semantic autocorrelation coding feature. Specifically, the process can be expressed as follows:

[0040]

[0041] Wherein, V1 is the implicit coding feature vector of the multi-source meteorological time series association, is matrix multiplication, V1 T is the transposed vector of V1, M z It is the semantic autocorrelation coding matrix of multi-source meteorological time series association.

[0042] Specifically, the process of calculating the product between the implicitly encoded feature vector of multi-source meteorological time series association and its transpose to generate the semantic autocorrelation coding matrix of multi-source meteorological time series association is essentially an autocorrelation analysis. The effect of this process is to reveal the correlation or similarity between different time points within the same set of meteorological time series data, helping the present application to understand the trends and patterns of these data over time. Specifically, this autocorrelation analysis can capture the inherent laws of meteorological conditions evolving over time. In this way, the present application can identify which time periods have higher similarity and continuity in meteorological conditions, which may indicate that certain potential risk factors are accumulating. Using autocorrelation coding can also help filter out some noise interference and retain those truly meaningful time dependencies. Doing so can make subsequent cross-modal interaction optimization more focused on key features and improve the efficiency and accuracy of the system.

[0043] In one embodiment, the landform remote sensing image semantic decomposition unit 52 is used to: perform feature decoupling along the channel dimension on the semantic coding feature map of the landform remote sensing image of the predetermined area to obtain a set of semantic local feature matrices of the landform remote sensing image of the predetermined area as the set of semantic local features of the landform remote sensing image of the predetermined area. Specifically, the process can be expressed as follows:

[0044] Decouple(F2)={M1,...,M i , ..., M n}

[0045] Wherein, F2 is the semantic coding feature map of the topographic remote sensing image of the predetermined area, Decouple(F2) is the feature decoupling operation of F2, M1, M i and M nThey are respectively the 1st, ith and nth semantic local feature matrices of topographic remote sensing images of the predetermined area in the set of semantic local feature matrices of topographic remote sensing images of the predetermined area.

[0046] It should be understood that through feature decoupling, the originally complex geomorphic features are refined into multiple local feature matrices, each of which represents an aspect of different surface properties, such as vegetation coverage, soil moisture, building density, etc. This method can better distinguish small but critical geomorphic differences. For example, in flood warnings, it can accurately identify which areas are more prone to water accumulation because they may be located in low-lying areas or close to rivers; in landslide warnings, it can clearly point out which hillsides are more vulnerable due to sparse vegetation. At the same time, the decoupled local feature matrix reduces the impact of noise and makes the model less sensitive to outliers, which means that even if there are errors in some sensor readings, the system can still make judgments based on other reliable features, thereby improving the stability and reliability of the entire early warning system.

[0047] In one embodiment, Figure 4 As shown, the geomorphology-meteorology semantic cross-modal response unit 53 includes: a multi-source meteorology semantic linear transformation subunit 531, which is used to linearly transform the multi-source meteorology time series association semantic autocorrelation coding matrix to obtain a multi-source meteorology time series association semantic query coding matrix and a multi-source meteorology time series association semantic value coding matrix; a geomorphology-meteorology semantic cross-modal prompt information coding subunit 532, which is used to input each predetermined area geomorphology remote sensing image semantic local feature matrix in the set of the multi-source meteorology time series association semantic query coding matrix, the multi-source meteorology time series association semantic value coding matrix and the predetermined area geomorphology remote sensing image semantic local feature matrix into a cross-modal prompt information encoder based on a converter structure to obtain a geomorphology-meteorology semantic cross-modal A set of prompt information semantic coding matrices; a landform-meteorological cross-modal semantic decoding processing subunit 533, used to perform information gating-based decoding processing on the set of landform-meteorological semantic cross-modal prompt information semantic coding matrices to obtain a set of landform-meteorological cross-modal semantic interaction attention weights; a landform-meteorological cross-modal interaction optimization coding subunit 534, used to perform cross-modal interaction optimization coding on the set of multi-source meteorological temporal correlation semantic autocorrelation coding matrices and the semantic local feature matrices of the landform remote sensing images of the predetermined area based on the set of landform-meteorological cross-modal semantic interaction attention weights to obtain a landform-meteorological cross-modal response semantic coding feature map as the rendering style to guide the semantic coding modulation features of the landform remote sensing images of the predetermined area.

[0048] Exemplarily, in the multi-source meteorological semantic linear transformation subunit 531, the multi-source meteorological time series associated semantic autocorrelation coding matrix is ​​linearly transformed to obtain a multi-source meteorological time series associated semantic query coding matrix and a multi-source meteorological time series associated semantic value coding matrix. Specifically, the process can be expressed as follows:

[0049]

[0050] Among them, W q is the query embedding matrix, M q is the multi-source meteorological time series association semantic query encoding matrix, W v is the value embedding matrix, M v It is the semantic value encoding matrix of multi-source meteorological time series association.

[0051] It should be understood that by linearly transforming the autocorrelation coding matrix, the intrinsic connection between meteorological conditions at different time points can be effectively captured and decomposed into two complementary parts: the query coding matrix and the value coding matrix. The query coding matrix is ​​mainly used to represent how the meteorological characteristics at each time point interact with other time points, that is, the "question" relationship between them; while the value coding matrix reflects the actual meteorological state or value at these time points, that is, the content of the "answer". In the application scenario of disaster warning, this transformation enables the system to more accurately identify which historical meteorological patterns have similarities with current or future weather conditions. For example, when predicting extreme events such as heavy rain or floods, query coding can help the model find disaster cases that have occurred under similar conditions in history, so as to better assess the risk level in the current situation. At the same time, value coding provides specific meteorological parameters (such as rainfall, wind speed, etc.), so that the system can make more accurate forecasts based on this information. In addition, by decomposing the autocorrelation coding matrix into query and value coding matrices, the model's ability to understand long-term series data can also be enhanced. It not only takes into account immediate meteorological changes, but also takes into account the cumulative effects over a period of time in the past, which is particularly important for understanding disaster trends that take a long time to appear. For example, a few days of light rain may not immediately raise alarm bells, but when combined with factors such as previous drought conditions and soil saturation, it may reveal potential flooding risks.

[0052] Exemplarily, in the landform-meteorological semantic cross-modal prompt information encoding subunit 532, each predetermined area landform remote sensing image semantic local feature matrix in the set of the multi-source meteorological temporal association semantic query encoding matrix, the multi-source meteorological temporal association semantic value encoding matrix and the predetermined area landform remote sensing image semantic local feature matrix is ​​respectively input into the cross-modal prompt information encoder based on the converter structure to obtain a set of landform-meteorological semantic cross-modal prompt information semantic encoding matrices. Specifically, the process can be expressed by the formula:

[0053]

[0054] Among them, M j T It is M i The transposed matrix of M i The scale is the matrix width multiplied by the matrix height, softmax(·) is the softmax function, I i It is M q and M i Semantic encoding matrix of geomorphic-meteorological semantic cross-modal cue information.

[0055] That is, the multi-source meteorological time series associated semantic query encoding matrix, the multi-source meteorological time series associated semantic value encoding matrix, and the set of semantic local feature matrices of geomorphic remote sensing images in the predetermined area are respectively input into the cross-modal prompt information encoder based on the converter structure. This operation enables the model to consider information in both time and space dimensions at the same time. The query encoding matrix represents the interaction or "inquiry" relationship between meteorological conditions at different time points, while the value encoding matrix provides the specific meteorological parameters at these time points. At the same time, the semantic local feature matrix of geomorphic remote sensing images contains spatial information about surface features such as topography and vegetation cover. When these data are fed into the cross-modal prompt information encoder based on the converter structure, the encoder will dynamically adjust and integrate information from both meteorological and geomorphic aspects according to the attention mechanism guided by the query encoding matrix, and generate a series of new comprehensive feature representations, namely, the geomorphic-meteorological semantic cross-modal prompt information semantic encoding matrix. In the actual application of disaster warning, the effect of this method is particularly significant. For example, when predicting floods, it is necessary not only to know the current and next few days' rainfall (provided by meteorological data), but also to understand geographical factors such as the distribution of rivers in the area, soil types and their water absorption capacity (provided by geomorphological data). Through the above steps, the system can identify which areas face a higher risk of flooding due to specific terrain conditions (such as low-lying areas) plus continuous heavy rainfall. Similarly, in terms of landslide warning, combining meteorological changes such as wind speed and humidity with geomorphological features such as slope stability and the ability of vegetation roots to fix, it is possible to more accurately determine which places are more prone to geological disasters.

[0056] Exemplarily, in the landform-meteorology cross-modal semantic decoding processing subunit 533, the set of landform-meteorology semantic cross-modal prompt information semantic encoding matrices is subjected to information gating-based decoding processing to obtain a set of landform-meteorology cross-modal semantic interaction attention weights.

[0057] a i =softmax{decoder(I i , W a )}

[0058] Among them, W a is the weight matrix, decoder is the decoder, a i It is M i The corresponding landform-meteorological cross-modal semantic interaction attention weight. Specifically, the process can be expressed as follows:

[0059] That is, the data processed by the cross-modal cue information encoder already contains rich information about the association between meteorological conditions (such as wind speed and rainfall) and geomorphic features (such as topography and vegetation cover). However, not all of this information is equally important; data at certain time points or geographical locations may be more critical than others. Therefore, the information gating mechanism is introduced to perform decoding processing, aiming to screen and strengthen the information that best reflects potential risk factors, while suppressing irrelevant or redundant data. The effect of the information gating decoding processing is to generate a set of geomorphic-meteorological cross-modal semantic interaction attention weights. These weights represent the degree of importance of meteorological and geomorphic features at different times and spaces, helping the model focus on those factors that really affect the occurrence of disasters. For example, in flood warnings, if a low-lying area has abnormally high rainfall for several consecutive days and the soil moisture in the area is close to saturation, the corresponding attention weight will increase significantly, indicating that this is a high-risk area that requires special vigilance. Similarly, in landslide warnings, if the vegetation coverage of a hillside suddenly decreases and is accompanied by continuous strong winds, then the location will also be given a higher attention weight, indicating that there may be a landslide risk. In addition, this information-gated decoding method also enhances the flexibility and adaptability of the system. It allows the model to dynamically adjust its focus according to different types of disaster events, ensuring that reasonable judgments can be made even in the face of a complex and changing natural environment. For example, during droughts, the system may pay more attention to changes in groundwater levels in areas with little precipitation; and during typhoon season, it will strengthen monitoring of storm surge threats in coastal areas.

[0060] In one embodiment, the landform-meteorological cross-modal interaction optimization coding subunit 534 is used to: input the set of the multi-source meteorological temporal association semantic autocorrelation coding matrix and the set of the semantic local feature matrix of the predetermined area landform remote sensing image into the cross-modal interaction unit to obtain the set of landform-meteorological cross-modal interaction local feature matrix; input the set of the landform-meteorological cross-modal semantic interaction attention weights and the set of the landform-meteorological cross-modal interaction local feature matrix into the cross-modal interaction optimization unit to obtain the landform-meteorological cross-modal response semantic coding feature map. Specifically, the process can be expressed as follows:

[0061] F 1-2 =couple{a1·M1⊙M z , ..., a i ·M i ⊙M z , ..., a n ·M n ⊙M z}

[0062] Among them, a1, a i 、a n They are M1, M i 、M n The corresponding landform-meteorological cross-modal semantic interaction attention weight, ⊙ is the point product by location, couple{·,·,...,·} is the feature coupling along the channel dimension, F 1-2 It is the semantic encoding feature map of the landform-meteorological cross-modal response.

[0063] That is, the set of geomorphic-meteorological cross-modal interaction local feature matrices processed by the cross-modal interaction unit can capture the complex relationship between meteorological conditions (such as wind speed and rainfall) and geographical environment (such as topography and vegetation cover). For example, in flood warning, this step can identify which low-lying areas face higher flood risks due to continuous heavy rainfall; in landslide warning, it can determine which hillsides are prone to landslides due to reduced vegetation and continuous strong winds. This fusion not only takes into account immediate meteorological changes, but also takes into account long-term cumulative effects, such as soil saturation, so that the model can understand potential risk factors more comprehensively. Next, the introduction of a set of geomorphic-meteorological cross-modal semantic interaction attention weights further enhances the focus on key information. These weights reflect the importance of meteorological and geomorphic features in different time and space, helping the model focus on those factors that really affect the occurrence of disasters. For example, if data at a specific time point or geographical location is given a higher attention weight, it means that it is crucial to the current warning decision. In this way, the system can more accurately screen and strengthen the information that best reflects potential risks, while suppressing irrelevant or redundant data to improve the accuracy of early warnings. Finally, the above results are input into the cross-modal interactive optimization unit to generate a geomorphic-meteorological cross-modal response semantic encoding feature map, which is a comprehensive expression of the entire process. This feature map not only integrates information from both meteorological and geomorphic aspects, but also highlights the decisive temporal and spatial feature combinations through optimization processing.

[0064] Specifically, the processing of cross-modal interaction analysis can more accurately capture the correlation between multi-source meteorological time series correlation features and geomorphic remote sensing image features by introducing the intra-modal autocorrelation matrix and the deep interaction mechanism between modalities, thereby overcoming the challenges brought by inter-modal heterogeneity and improving the quality of multi-modal data fusion. At the same time, by performing feature decoupling and subsequent information gating processing on the semantic coding feature map of the geomorphic remote sensing image of the predetermined area, it is possible to effectively remove redundant information, reduce noise interference, and ensure that the generated feature representation is both compact and representative, thereby improving the performance of the model. In addition, through the cross-modal prompt information encoder and the cross-modal interaction optimization unit, the information importance between multi-source meteorological time series correlation features and geomorphic remote sensing image features can be dynamically adjusted, and the complementary effect between modalities can be strengthened, so that the generated geomorphic-meteorological cross-modal response semantic coding feature information is more comprehensive and accurate, and is suitable for complex disaster monitoring and early warning tasks.

[0065] Exemplarily, in the geological disaster safety warning module 60, intelligent disaster warning is performed based on the landform-meteorological cross-modal response semantic coding features to determine whether a geological disaster warning prompt is generated. In one embodiment, the geological disaster safety warning module is used to: input the landform-meteorological cross-modal response semantic coding feature map into the classifier-based disaster intelligent warning module to obtain a warning analysis result, and the warning analysis result is used to indicate whether a geological disaster warning prompt is generated. In other words, the cross-modal responsiveness interaction characteristics between the semantics of the geomorphic remote sensing image of the predetermined area and the multi-source meteorological time series association semantics are used to perform intelligent disaster warning in the rural predetermined area. In this way, rural meteorological safety data can be monitored and analyzed in real time, and geological disaster warnings can be automatically performed by combining and analyzing multi-source data, thereby overcoming the deficiencies in spatial distribution, temporal resolution, real-time performance, etc. caused by the use of a single data source in traditional disaster warning schemes, thereby helping to increase the real-time response capability and intelligence level of the automatic disaster warning system.

[0066] In a specific embodiment, in the process of inputting the landform-meteorological cross-modal response semantic coding feature map into the classifier-based disaster intelligent warning module to obtain the warning analysis result, a support vector machine (SVM) can be used as a specific classifier. Support vector machine is a supervised learning model that maximizes the interval between different categories by finding a hyperplane, thereby achieving effective classification of data. For multidimensional and complex feature data such as the landform-meteorological cross-modal response semantic coding feature map, SVM can effectively identify the decision boundary in the feature space and distinguish different disaster risk levels or types. For example, when facing warnings of different types of geological disasters such as floods and landslides, the trained support vector machine can judge whether the current situation indicates that a certain type of disaster is about to occur based on the input landform-meteorological feature map, and output the corresponding warning level or directly give whether a geological disaster warning prompt is generated.

[0067] Here, when the semantic coding feature map of the topographic remote sensing image of the predetermined area and the multi-source meteorological time series associated implicit coding feature vector respectively represent the image semantic coding features and multi-source meteorological time series associated coding features of the remote sensing image data of the predetermined area, after performing meteorological-topographic cross-modal analysis on them, the topographic-meteorological cross-modal response semantic coding feature map will also have significant interactive optimization distribution spatial structure differences due to the differences in cross-modal semantic suggestiveness, affecting the convergence consistency of the classifier, and thus affecting the accuracy of the warning analysis results obtained by the classifier-based intelligent disaster warning module.

[0068] Therefore, in one example, the landform-meteorological cross-modal response semantic encoding feature map is optimized, and the optimization process includes:

[0069] The square root of the sum of the absolute values ​​of all feature values ​​of the landform-meteorological cross-modal response semantic coding feature map and the sum of the squares is calculated to obtain the first landform-meteorological cross-modal response semantic coding spatial structure value and the second landform-meteorological cross-modal response semantic coding spatial structure value, that is:

[0070] w1=∑ l |f i |

[0071]

[0072] Among them, f i represents the i-th eigenvalue of the landform-meteorological cross-modal response semantic coding feature map, W1 represents the first landform-meteorological cross-modal response semantic coding spatial structure value, and w2 represents the second landform-meteorological cross-modal response semantic coding spatial structure value;

[0073] Determine the total number n of eigenvalues ​​of all eigenvalues ​​of the geomorphic-meteorological cross-modal response semantic encoding feature map;

[0074] For each eigenvalue of the landform-meteorological cross-modal response semantic coding feature map, the first landform-meteorological cross-modal response semantic coding spatial structure value minus the product of the eigenvalue and the total number of eigenvalues ​​is calculated to obtain the first landform-meteorological cross-modal response semantic coding long-range dependency value x i =w1-f i ×n1, where w1 represents the spatial structure value of the semantic encoding of the first geomorphological-meteorological cross-modal response, f i represents the ith eigenvalue of the landform-meteorological cross-modal response semantic coding feature map, n1 represents the total number of eigenvalues ​​of all eigenvalues ​​of the landform-meteorological cross-modal response semantic coding feature map, x i It represents the long-range dependency value of the semantic encoding of the first geomorphological-meteorological cross-modal response;

[0075] Calculate the second landform-meteorological cross-modal response semantic coding long-range dependency value obtained by multiplying the square root of the total number of eigenvalues ​​by the product of the eigenvalues ​​minus the second landform-meteorological cross-modal response semantic coding spatial structure value Among them, w2 represents the spatial structure value of the second geomorphological-meteorological cross-modal response semantic encoding, f i represents the ith eigenvalue of the landform-meteorological cross-modal response semantic coding feature map, n1 represents the total number of eigenvalues ​​of all eigenvalues ​​of the landform-meteorological cross-modal response semantic coding feature map, y i It represents the long-range dependency value of the semantic encoding of the second geomorphological-meteorological cross-modal response;

[0076] The index value calculated by taking the first landform-meteorological cross-modal response semantic coding long-range dependency value as the exponent of the natural constant and the inverse of the second landform-meteorological cross-modal response semantic coding long-range dependency value are weighted summed to obtain the optimized eigenvalue corresponding to each eigenvalue Among them, x i represents the long-range dependency value of the semantic encoding of the first geomorphological-meteorological cross-modal response, y i represents the long-range dependency value of the semantic encoding of the second geomorphological-meteorological cross-modal response, e represents a natural constant, α and β represent weighted hyperparameters, and f i Indicates the optimized eigenvalue corresponding to each eigenvalue;

[0077] The optimized feature values ​​are combined into an optimized geomorphic-meteorological cross-modal response semantic encoding feature map.

[0078] That is, in view of the possible spatial structure loss in the high-dimensional space of the feature set of the landform-meteorological cross-modal response semantic coding feature map, which leads to the weight of the classifier implicitly inferring the spatial structure information based on the features, resulting in inconsistent convergence. A long-distance feature dependency relationship is established based on the spatial structure representation of the landform-meteorological cross-modal response semantic coding feature map based on the overall feature scale of the landform-meteorological cross-modal response semantic coding feature map, so as to establish the feature local connectivity of the landform-meteorological cross-modal response semantic coding feature map, and capture the spatial ambiguous information of the object feature value through the unstructured feature value point prediction of the landform-meteorological cross-modal response semantic coding feature map, thereby improving the spatial inductive deviation perception ability of the feature set of the landform-meteorological cross-modal response semantic coding feature map, improving the convergence consistency of the classifier, and improving the accuracy of the warning analysis results obtained by the disaster intelligent warning module based on the classifier of the landform-meteorological cross-modal response semantic coding feature map.

[0079] In another example of the present application, a multi-source data disaster automatic early warning system for rural meteorological safety also includes: In order to build an efficient disaster early warning system, it is first necessary to create a "risk map" that integrates factors such as population distribution, economic activities and geological conditions. By combining with the meteorological risk level zoning, the system can accurately screen out high-risk areas for sudden disasters and determine the geographical scope of early warning release. Using the real-time number-taking technology of the communication base station, a list of all users in the area to be released can be generated to ensure that every online user is covered and comprehensive information is conveyed. Data monitoring and alarm functions are one of the core parts of the system. It relies on the formulated data collection strategy to capture data from various sources such as automatic stations, API interfaces, special detection equipment and superior forecast systems at regular intervals. These data are collected and stored in an integrated data environment to provide support for subsequent processing. Once an abnormal situation is detected, the system will immediately analyze the relevant data and send warning information to relevant personnel through multiple communication channels (such as text messages and voice calls) to ensure that they can take preventive measures in time. In addition, the system also has functions such as notification mode management, site configuration, real-time parameter setting and event lock to improve operational efficiency and intelligence level. The emergency plan management system is designed to ensure rapid response in emergency situations. It uses atomic components and graph structure design, allowing users to adjust plans according to actual needs, ensuring that the emergency process is efficient and forms a closed loop. Plans can be edited according to actual conditions, and the execution status is displayed in real time, so as to flexibly respond to different types of emergencies. When facing severe convective weather or other emergency meteorological conditions, the channel fusion call service will use multiple channels such as SMS, voice outbound calls and instant messaging tools to transmit warning information to ensure that flood control responsible persons can receive key information in the first time. This service has the ability to operate 24 hours a day, 7 days a week, and can quickly transmit large-scale information in a short period of time without being restricted by operators, so as to ensure that the information quickly covers everyone who needs it. The multi-scenario cockpit editing system provides functions such as chart drawing, data display and map presentation, supports access to multiple data sources, and realizes smooth communication and collaboration between different sections. This system not only supports data analysis, information sharing and integration services, but also users can quickly generate new interfaces based on the system's editing functions to meet the needs of more different scenarios.

[0080] In terms of key technologies, the early warning system of this application also integrates data collection and fusion technology, and realizes real-time collection and fusion of multi-source data through API interface, database connection and real-time data stream processing technology. At the same time, the system expands the monitoring and alarm data, providing basic data basis for later rainstorm monitoring, forecasting and early warning analysis. For the refined forecast products of heavy rainfall, the system can automatically process and automatically generate relevant business products that meet the regional early warning needs, and remind the situation of reaching the early warning threshold. The numerical discovery and monitoring module is responsible for collecting, sorting and processing various types of data, and alarming and warning according to specific indicators and thresholds to ensure that abnormal data is processed in a timely manner. The full-process data monitoring combines GIS technology, integrates geographic location annotation and application scenario access, and realizes the one-screen presentation and decision-making implementation of meteorological, hydrological and other data in the whole region. Finally, the rainfall threshold model is established based on the research results of historical rainfall-induced geological disasters, taking into account the susceptibility index and physical process simulation, realizing the calculation of rainfall thresholds for different slopes, and providing warning signals for geological disaster elements. When the monitoring data exceeds the set threshold, the system triggers an alarm reminder and notifies relevant personnel in various ways to ensure timely action.

[0081] In summary, according to the embodiment of the present application, the multi-source data disaster automatic early warning system for rural meteorological safety is explained, which can collect multi-source meteorological data of a predetermined rural area through real-time monitoring. When warning of disasters, these multi-source meteorological time series data and the remote sensing image data of the predetermined rural area are input together into the data processing and image semantic understanding algorithm based on artificial intelligence and deep learning for analysis, so as to capture the cross-modal response characteristics between the multi-source meteorological time series association semantics and the remote sensing image semantics, so as to carry out intelligent early warning of disasters in the predetermined rural area. In this way, rural meteorological safety data can be monitored and analyzed in real time, and multi-source data can be used to automatically warn of geological disasters, thereby overcoming the deficiencies in spatial distribution, temporal resolution, real-time performance, etc. caused by the use of a single data source in traditional disaster early warning schemes, thereby helping to increase the real-time response capability and intelligence level of the automatic disaster early warning system.

[0082] Figure 5 Schematic diagram of the multi-source data disaster automatic warning method for rural meteorological safety in the embodiment of the present application. Figure 5As shown, the multi-source data disaster automatic warning method 100 for rural meteorological safety includes: S1, obtaining a time queue of multi-source meteorological data of a predetermined area collected by a sensor network, wherein the multi-source meteorological data includes wind direction, wind speed value, temperature value, air pressure value, rainfall, river water level and reservoir water level; S2, extracting remote sensing image data of the predetermined area from a remote sensing image database; S3, performing time series coding on the time queue of the multi-source meteorological data to obtain multi-source meteorological time series associated implicit coding features; S4, performing geomorphic feature extraction on the remote sensing image data of the predetermined area to obtain semantic coding features of geomorphic remote sensing images of the predetermined area; S5, performing cross-modal interactive analysis on the multi-source meteorological time series associated implicit coding features and the semantic coding features of geomorphic remote sensing images of the predetermined area to obtain geomorphic-meteorological cross-modal response semantic coding features; S6, performing intelligent disaster warning based on the geomorphic-meteorological cross-modal response semantic coding features to determine whether to generate a geological disaster alarm prompt.

[0083] In one embodiment, the time queue of the multi-source meteorological data is time-series encoded to obtain multi-source meteorological time series associated implicit coding features, including: inputting the time queue of the multi-source meteorological data into a multi-source meteorological time series encoder based on an RNN-LSTM hybrid model to obtain a multi-source meteorological time series associated implicit coding feature vector as the multi-source meteorological time series associated implicit coding feature.

[0084] In one embodiment, a cross-modal interactive analysis is performed on the multi-source meteorological time series associated implicit coding features and the predetermined area landform remote sensing image semantic coding features to obtain landform-meteorological cross-modal response semantic coding features, including: autocorrelation coding is performed on the multi-source meteorological time series associated implicit coding features to obtain multi-source meteorological time series semantic autocorrelation coding features; feature decoupling is performed on the predetermined area landform remote sensing image semantic coding features to obtain a set of semantic local features of the predetermined area landform remote sensing images; cross-modal interactive optimization is performed on the multi-source meteorological time series semantic autocorrelation coding features and the set of semantic local features of the predetermined area landform remote sensing images to obtain the landform-meteorological cross-modal response semantic coding features.

[0085] Here, those skilled in the art will appreciate that the specific operations of each step in the multi-source data disaster automatic early warning method for rural meteorological safety have been described in the above reference. Figures 1 to 4 The multi-source data disaster automatic early warning system for rural meteorological safety has been introduced in detail, and therefore, its repeated description will be omitted.

[0086] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0087] It should be understood that the specific examples in this article are only intended to help those skilled in the art better understand the embodiments of the present application, rather than to limit the scope of the embodiments of the present application.

[0088] It should also be understood that in the various embodiments of the present application, the size of the serial number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0089] It should also be understood that the various implementation methods described in this specification can be implemented individually or in combination, and the embodiments of the present application are not limited to this.

[0090] Unless otherwise stated, all technical and scientific terms used in the embodiments of the present application are the same as the meanings generally understood by those skilled in the art of the technical field of the present application. The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present application. The term "and / or" used in the present application includes any and all combinations of one or more related listed items. The singular forms of "a kind of", "above" and "the" used in the embodiments of the present application and the appended claims are also intended to include majority forms, unless the context clearly indicates other meanings. In addition, the terms "first", "second" etc. are only used for description purposes and cannot be understood as indicating or suggesting relative importance.

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

[0092] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0093] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0094] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0095] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A multi-source data disaster automatic early warning system for rural meteorological safety, characterized in that: include: A meteorological data collection module, used to obtain a time queue of multi-source meteorological data of a predetermined area collected by a sensor network, wherein the multi-source meteorological data includes wind direction, wind speed value, temperature value, air pressure value, rainfall, river water level and reservoir water level; A remote sensing image data acquisition module is used to extract remote sensing image data of a predetermined area from a remote sensing image database; A meteorological time series coding module, used for performing time series coding on the time queue of the multi-source meteorological data to obtain multi-source meteorological time series associated implicit coding features; A geomorphic feature extraction module is used to extract geomorphic features from the remote sensing image data of the predetermined area to obtain semantic coding features of the geomorphic remote sensing image of the predetermined area; A geomorphic-meteorological cross-modal interaction analysis module is used to perform cross-modal interaction analysis on the multi-source meteorological time series associated implicit coding features and the semantic coding features of the geomorphic remote sensing images of the predetermined area to obtain geomorphic-meteorological cross-modal response semantic coding features, wherein the geomorphic-meteorological cross-modal interaction analysis module includes: a meteorological time series autocorrelation coding unit, used to perform autocorrelation coding on the multi-source meteorological time series associated implicit coding features to obtain multi-source meteorological time series semantic autocorrelation coding features; a geomorphic remote sensing image semantic decomposition unit, used to perform feature decoupling on the semantic coding features of the geomorphic remote sensing images of the predetermined area to obtain a set of semantic local features of the geomorphic remote sensing images of the predetermined area; a geomorphic-meteorological semantic cross-modal response unit, used to perform cross-modal interaction optimization on the set of the multi-source meteorological time series semantic autocorrelation coding features and the semantic local features of the geomorphic remote sensing images of the predetermined area to obtain the geomorphic-meteorological cross-modal response semantic coding features; The geological disaster safety warning module is used to perform intelligent disaster warning based on the landform-meteorological cross-modal response semantic coding features to determine whether to generate a geological disaster warning prompt.

2. The multi-source data disaster automatic early warning system for rural meteorological safety according to claim 1 is characterized in that: The meteorological time series encoding module is used to: input the time queue of the multi-source meteorological data into a multi-source meteorological time series encoder based on the RNN-LSTM hybrid model to obtain a multi-source meteorological time series associated implicit coding feature vector as the multi-source meteorological time series associated implicit coding feature.

3. The multi-source data disaster automatic early warning system for rural meteorological safety according to claim 2 is characterized in that: The landform feature extraction module is used to: input the remote sensing image data of the predetermined area into a predetermined area landform feature extractor based on a hole convolutional neural network model to obtain a semantic coding feature map of the predetermined area landform remote sensing image as the semantic coding feature of the predetermined area landform remote sensing image.

4. The multi-source data disaster automatic early warning system for rural meteorological safety according to claim 3 is characterized in that: The meteorological time series autocorrelation coding unit is used to calculate the product between the multi-source meteorological time series associated implicit coding feature vector and the transposed vector of the multi-source meteorological time series associated implicit coding feature vector to obtain a multi-source meteorological time series associated semantic autocorrelation coding matrix as the multi-source meteorological time series semantic autocorrelation coding feature.

5. The multi-source data disaster automatic early warning system for rural meteorological safety according to claim 4 is characterized in that: The landform remote sensing image semantic decomposition unit is used to: perform feature decoupling along the channel dimension on the semantic coding feature map of the landform remote sensing image of the predetermined area to obtain a set of semantic local feature matrices of the landform remote sensing image of the predetermined area as a set of semantic local features of the landform remote sensing image of the predetermined area.

6. The multi-source data disaster automatic early warning system for rural meteorological safety according to claim 5 is characterized in that: The geomorphology-meteorology semantic cross-modal response unit includes: A multi-source meteorological semantic linear transformation subunit, used for linearly transforming the multi-source meteorological time series associated semantic autocorrelation coding matrix to obtain a multi-source meteorological time series associated semantic query coding matrix and a multi-source meteorological time series associated semantic value coding matrix; A geomorphic-meteorological semantic cross-modal prompt information encoding subunit is used to input each predetermined area geomorphic remote sensing image semantic local feature matrix in the set of the multi-source meteorological time series associated semantic query encoding matrix, the multi-source meteorological time series associated semantic value encoding matrix and the predetermined area geomorphic remote sensing image semantic local feature matrix into a cross-modal prompt information encoder based on a converter structure to obtain a set of geomorphic-meteorological semantic cross-modal prompt information semantic encoding matrices; A landform-meteorological cross-modal semantic decoding processing subunit, used for performing information gating-based decoding processing on the set of landform-meteorological semantic cross-modal prompt information semantic encoding matrices to obtain a set of landform-meteorological cross-modal semantic interaction attention weights; The landform-meteorology cross-modal interaction optimization coding subunit is used to perform cross-modal interaction optimization coding on the set of the multi-source meteorological time series correlation semantic autocorrelation coding matrix and the semantic local feature matrix of the landform remote sensing image of the predetermined area based on the set of the landform-meteorology cross-modal semantic interaction attention weights to obtain the landform-meteorology cross-modal response semantic coding feature map as the rendering style to guide the semantic coding modulation feature of the landform remote sensing image of the predetermined area.

7. The multi-source data disaster automatic early warning system for rural meteorological safety according to claim 6 is characterized in that: The landform-meteorology cross-modal interaction optimization coding subunit is used for: Inputting the set of the multi-source meteorological time series associated semantic autocorrelation coding matrix and the semantic local feature matrix of the predetermined area landform remote sensing image into a cross-modal interaction unit to obtain a set of landform-meteorological cross-modal interaction local feature matrices; The set of the landform-meteorological cross-modal semantic interaction attention weights and the set of the landform-meteorological cross-modal interaction local feature matrices are input into a cross-modal interaction optimization unit to obtain the landform-meteorological cross-modal response semantic encoding feature map.

8. The multi-source data disaster automatic early warning system for rural meteorological safety according to claim 7 is characterized in that: The geological disaster safety warning module is used to: input the landform-meteorological cross-modal response semantic coding feature map into the classifier-based disaster intelligent warning module to obtain a warning analysis result, and the warning analysis result is used to indicate whether a geological disaster warning prompt is generated.

9. A multi-source data disaster automatic early warning method for rural meteorological safety, characterized in that: include: Acquire a time queue of multi-source meteorological data of a predetermined area collected by a sensor network, wherein the multi-source meteorological data includes wind direction, wind speed value, temperature value, air pressure value, rainfall, river water level and reservoir water level; Extracting remote sensing image data of the predetermined area from a remote sensing image database; Performing time series coding on the time queue of the multi-source meteorological data to obtain multi-source meteorological time series associated implicit coding features; Extracting geomorphic features from the remote sensing image data of the predetermined area to obtain semantic coding features of the geomorphic remote sensing image of the predetermined area; The multi-source meteorological time series associated implicit coding features and the predetermined area landform remote sensing image semantic coding features are cross-modal interactively analyzed to obtain landform-meteorological cross-modal response semantic coding features, including: performing autocorrelation coding on the multi-source meteorological time series associated implicit coding features to obtain multi-source meteorological time series semantic autocorrelation coding features; performing feature decoupling on the predetermined area landform remote sensing image semantic coding features to obtain a set of predetermined area landform remote sensing image semantic local features; performing cross-modal interactive optimization on the multi-source meteorological time series semantic autocorrelation coding features and the predetermined area landform remote sensing image semantic local features to obtain the landform-meteorological cross-modal response semantic coding features; Based on the geomorphic-meteorological cross-modal response semantic coding features, intelligent disaster warning is performed to determine whether a geological disaster warning prompt is generated.

10. The multi-source data disaster automatic early warning method for rural meteorological safety according to claim 9 is characterized in that: The time queue of the multi-source meteorological data is time-series encoded to obtain multi-source meteorological time series associated implicit coding features, including: inputting the time queue of the multi-source meteorological data into a multi-source meteorological time series encoder based on an RNN-LSTM hybrid model to obtain a multi-source meteorological time series associated implicit coding feature vector as the multi-source meteorological time series associated implicit coding feature.

Images