Grain and cotton crop meteorological risk grading early warning method and system oriented to multi-disaster-type linkage

Abstract

The invention provides a multi-disaster-type linkage-oriented grain and cotton crop meteorological risk grading early warning method and system, and relates to the technical field of risk grading early warning. The first information comprises soil salinity change data, frost meteorological data, drought meteorological data, wind disaster meteorological data and crop monitoring data in a preset time period; performing disaster type space-time relation analysis on the first information to generate a multi-disaster type linkage space-time model; performing fusion processing on the multi-disaster linkage spatial-temporal model through spectral clustering and a preset gradient boosting tree to obtain a disaster influence prediction model; carrying out adaptive filtering and grading judgment processing on the disaster influence prediction model to obtain a disaster influence grade; and generating an adversarial network based on the disaster influence levels and historical early warning coping strategies, and determining coping strategies corresponding to all disaster influence levels of each crop. According to the invention, targeted disaster grading can be carried out according to soil types and crop planting characteristics in different areas.

Description

Technical Field

[0001] The present invention relates to the field of risk classification and early warning technology, and in particular to a method and system for early warning of meteorological risks of grain and cotton crops for multi-disaster linkage. Background Art

[0002] Current meteorological risk assessment and disaster warning technologies primarily focus on predictive models for single disasters. Many traditional techniques often overlook the spatiotemporal linkages and compounding effects of different disasters. In the field of agrometeorological risk assessment, in particular, most methods assess the probability of a single disaster (such as frost or drought), ignoring the interactions and synergistic effects between disasters. With the intensification of climate change, multiple disasters (such as drought and windstorms, frost and saline-alkali soils) may occur simultaneously, resulting in compound disasters that have a far greater impact on crop growth than a single disaster alone.

[0003] Therefore, effectively capturing the correlation between single and multiple disasters, and accurately predicting the impact of different disasters, has become a key issue in current agrometeorological risk management. While existing technologies provide some early warning capabilities for single disasters through spatiotemporal models and disaster forecasting, they still have significant shortcomings in addressing the interplay of multiple disasters. Consequently, existing technologies struggle to provide accurate disaster assessments and real-time warnings in complex disaster scenarios, especially those involving multiple disasters and their combined effects. Furthermore, existing technologies lack the ability to provide targeted, graded warnings for various crops and regions, resulting in a significant waste of manpower and resources.

[0004] Therefore, there is an urgent need for a multi-disaster linkage grain and cotton crop meteorological risk classification warning method and system to solve the above technical problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-disaster-linked grain and cotton crop meteorological risk classification early warning method and system to improve the above-mentioned problems. To achieve the above-mentioned purpose, the technical solution adopted by the present invention is as follows: First, this application provides a multi-disaster linkage-based early warning method for grain and cotton crop meteorological risk, including: Acquiring first information, the first information including soil salinity change data, frost meteorological data, drought meteorological data, wind disaster meteorological data, and crop monitoring data within a preset time period; Analyzing the spatiotemporal relationship of disaster types using the first information, and building a model based on the spatiotemporal relationship obtained from the analysis to generate a spatiotemporal model for multi-disaster linkage; The multi-disaster linkage spatiotemporal model is subjected to spectral clustering and a preset gradient boosting tree fusion process to obtain a disaster impact prediction model; The disaster impact prediction model is subjected to adaptive filtering and graded judgment processing to obtain the disaster impact level of multiple disaster types on each crop in all preset areas; An adversarial network is generated based on the disaster impact levels and historical warning response strategies to determine response strategies corresponding to all disaster impact levels for each crop.

[0006] Secondly, this application also provides a multi-disaster-linked grain and cotton crop meteorological risk classification early warning system, including: an acquisition unit, configured to acquire first information, wherein the first information includes soil salinity change data, frost meteorological data, drought meteorological data, wind disaster meteorological data, and crop monitoring data within a preset time period; an analysis unit configured to analyze the spatiotemporal relationship of disaster types on the first information, and to generate a multi-disaster linkage spatiotemporal model based on the spatiotemporal relationship obtained by the analysis; A processing unit, configured to perform spectral clustering and a preset gradient boosting tree fusion process on the multi-disaster linkage spatiotemporal model to obtain a disaster impact prediction model; A judgment unit, configured to perform adaptive filtering and hierarchical judgment processing on the disaster impact prediction model to obtain the disaster impact level of multiple disaster types on each crop in all preset areas; The response unit is used to generate an adversarial network based on the disaster impact level and historical warning response strategies, and determine the response strategies corresponding to all disaster impact levels of each crop.

[0007] The beneficial effects of the present invention are: By accurately analyzing the correlation between single disasters and multiple disasters, the present invention combines algorithms such as spatiotemporal tensor decomposition, Granger causality test, and graph convolutional network to generate a spatiotemporal model of multi-disaster linkage, which can capture the spatiotemporal correlation and compound effects between multiple disasters. In particular, the present invention adopts a customized disaster risk grading method based on the differences in the distribution areas of grain and cotton crops and the differences in crop planting types, and conducts targeted risk assessment and disaster grading according to the climatic conditions, soil types, and crop planting characteristics of different regions. This innovative combination of regional differences, crop types, and analysis of the linkage of multiple disasters provides more accurate and personalized decision-making support for agricultural meteorological risk management, thereby effectively improving the accuracy and real-time performance of disaster warnings and helping agricultural producers to respond promptly in complex situations of multiple disaster linkages.

[0008] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the embodiments of the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0010] Figure 1 A schematic flow chart of a method for grading meteorological risk warning for grain and cotton crops for multi-disaster linkage according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the multi-disaster linkage grain and cotton crop meteorological risk classification early warning system described in an embodiment of the present invention.

[0011] In the figure: 701, acquisition unit; 702, analysis unit; 703, processing unit; 704, judgment unit; 705, response unit. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0013] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are used only to distinguish the description and should not be understood as indicating or implying relative importance.

[0014] Example 1:

[0015] This embodiment provides a multi-disaster linkage method for grain and cotton crop meteorological risk classification and early warning.

[0016] See also Figure 1 , the figure shows that the method includes step S1, step S2, step S3, step S4 and step S5.

[0017] Step S1: Acquire first information, where the first information includes soil salinity change data, frost weather data, drought weather data, wind disaster weather data, and crop monitoring data within a preset time period; It can be understood that the first information in this step is a collection of various disaster-related data, including soil salinity changes, frost, drought, wind disaster meteorological data and crop monitoring data. Soil salinity change data reflects the salinization of the soil within a predetermined time period, which is particularly important for evaluating the crop growth environment, especially under climatic conditions such as drought and high temperature. The intensification of soil salinization may lead to the inhibition of crop growth. Frost meteorological data focuses on the spatiotemporal characteristics of low temperature occurrence. Frost has a greater impact on crop growth, especially crops such as cotton are sensitive to low temperatures. Predicting the timing of frost can effectively avoid losses. Drought meteorological data is directly related to changes in precipitation and soil moisture, affecting the water supply of crops, while wind disaster meteorological data mainly includes information such as wind speed, wind direction and duration, which is crucial for evaluating the physical destructive effects of wind disasters on crops.

[0018] Crop monitoring data provides information on the physiological and health status of crops at different growth stages, helping to assess their adaptability to environmental changes. By integrating these different types of data, we can comprehensively and accurately depict the external environmental conditions for crop growth and their changing trends under different climatic conditions. This plays a fundamental role in predicting the impact of subsequent disasters and formulating response strategies.

[0019] The unique benefit of acquiring this data is that it enables the formation of a comprehensive, cross-sectoral disaster impact assessment framework. Unlike traditional methods that rely solely on meteorological data or a single disaster type, this step combines data from multiple disasters to more precisely capture the spatial and temporal interactions between disasters and their combined impacts on crops.

[0020] Step S2: Analyze the spatiotemporal relationship of disaster types on the first information, and build a model based on the spatiotemporal relationship obtained by the analysis to generate a multi-disaster linkage spatiotemporal model; It can be understood that this step, through spatiotemporal relationship analysis and graph convolutional network modeling, effectively reveals the spatiotemporal linkage effects between disasters. By integrating and modeling multidimensional spatiotemporal data, it ensures the multidimensionality and accuracy of disaster prediction. This enables the processing of complex interactions between disasters, providing a powerful and flexible theoretical framework for subsequent disaster risk assessment and response strategy development, suitable for scenarios involving multiple disasters with concurrent or compound effects. In this step, step S2 includes steps S21, S22, S23, and S24.

[0021] Step S21: extracting the spatial and temporal variation features of the first information into multiple feature tensors based on a spatiotemporal tensor decomposition method to obtain a spatiotemporal feature data set; It's understandable that this step first converts the acquired primary information—including soil salinity change data, frost weather data, drought weather data, windstorm weather data, and crop monitoring data—into a multidimensional spatiotemporal data structure. This data involves multiple variables, such as time (by hour, day, month, etc.), space (by geographic location or region), and different disaster types (frost, drought, etc.). To effectively integrate this multidimensional data, each type of disaster data is first converted into a matrix, where rows represent time, columns represent space or region, and the matrix values ​​represent the disaster intensity or impact.

[0022] Using the spatiotemporal tensor decomposition method, these data matrices are concatenated into tensors, forming a multidimensional tensor structure. This method performs high-dimensional analysis on disaster data, using time, space, and disaster type as tensor dimensions. This approach allows us to extract the characteristics of disasters across different temporal and spatial dimensions, ensuring that the spatiotemporal distribution and impact of each disaster are effectively captured.

[0023] In this process, space-time tensor decomposition, by breaking down the raw data into multiple latent factors, can reveal the temporal and spatial evolution patterns of different disasters and help identify potential relationships between them. For example, the occurrence of frost disasters may fluctuate cyclically in winter, while drought disasters may exhibit seasonal variations, and these changes are reflected in the space-time tensor. Through this decomposition method, we can extract the principal components related to disaster changes, further providing a data foundation with temporal and spatial characteristics for subsequent modeling.

[0024] The resulting spatiotemporal data sets provide accurate input for subsequent disaster linkage analysis, spatiotemporal modeling, and disaster risk assessment. Spatiotemporal tensor decomposition efficiently structures complex spatiotemporal disaster data, helping to identify potential relationships between disasters across different temporal and spatial scales, thereby providing deep insights for subsequent disaster prediction and risk assessment.

[0025] Step S22: performing a relationship analysis on the data in the spatiotemporal feature data set based on the Granger causality test method to obtain a causal relationship between disasters; As you can understand, this step first targets a collection of spatiotemporal feature data, which exhibits complex dependencies across time and space. For example, frost may affect changes in soil salinity, while drought may exacerbate the impact of windstorms on crops. Here, we treat these spatiotemporal feature data as time series data and apply Granger causality tests to each pair of disasters (e.g., frost and soil salinity changes, drought and windstorms, etc.).

[0026] This step constructs a Granger causality test model by using the time series data of the preset lag period. Assuming that we select k lag periods, we can build the following linear regression model: ,in, Y t is the time series to be predicted (dependent variable), α ο is the constant term (intercept), is the length of the lag period, indicating how far back in time we consider the data to predict the current value, β i is the regression coefficient of the lagged term of the time series being predicted, Y t-i is the time series predicted at time ti, is the regression coefficient of the lagged term of the predicted time series (independent variable), X t-j is the time series predicted at time tj, is the error term, which represents the unpredictable random component.

[0027] Granger causality tests reveal the strength of the causal relationship between each pair of disasters, further revealing the spatiotemporal dependencies between them. In particular, in scenarios involving multiple disasters, this method can effectively reveal potential interactions and transmission mechanisms between disasters. For example, droughts can exacerbate frost by affecting soil moisture, and frost can affect crop growth, ultimately having profound impacts on the growth of grain and cotton crops.

[0028] Granger causality testing can precisely determine the causal relationships between different disasters, revealing their transmission pathways and spatiotemporal interactions. This step not only helps understand how each disaster impacts others over time but also provides a clear causal network structure for subsequent spatiotemporal modeling. Furthermore, Granger causality testing can effectively identify significant causal relationships between disasters, providing data support for disaster linkage modeling and risk assessment.

[0029] Step S23: constructing a spatiotemporal correlation graph between disasters based on the causal relationship between the disasters, wherein the disaster types are regarded as nodes in the graph, the interactions between the disasters are regarded as edges in the graph, and a graph convolution operation is performed to extract the spatial and temporal characteristics of the linkage between the disasters; It's understandable that in this step, disaster types (such as frost, drought, and windstorms) are treated as nodes in a spatiotemporal correlation graph. Each node represents a disaster type, and its connections in the graph (i.e., edges) represent the interactions between disasters. Through the aforementioned Granger causality test, we have already established causal relationships between disasters. Therefore, the edges of the graph can be defined based on these causal relationships. If one disaster can historically predict the occurrence of another, there will be an edge connecting the two disaster nodes in the graph, indicating a spatiotemporal dependency between them. The edges of this graph not only connect disaster types but also contain weights regarding the causal strength and spatiotemporal relationships between the disasters.

[0030] Next, a graph convolutional network (GCN) is used to process the spatiotemporal correlation graph between disasters. A GCN is a deep learning method for learning on graph data, capable of propagating and fusing information through edges between nodes. In a GCN, each node (i.e., a disaster type) aggregates information from its neighboring nodes (i.e., other disasters with which it has a causal relationship), thereby updating its feature representation. Through multiple layers of graph convolution operations, the model gradually extracts complex interaction patterns between disasters and maps these patterns to spatiotemporal features. The final representation of each node contains information about the spatial and temporal relationships between that disaster and other disasters, accurately reflecting the interaction effects between disasters.

[0031] In this way, the spatiotemporal correlation graph not only reveals the direct causal relationships between disasters but also captures the complex indirect dependencies between multiple disaster types. Through this "neighborhood propagation" mechanism, graph convolutional networks can efficiently process spatial and temporal patterns in disaster data, avoiding the limitations of traditional methods that rely on single-hazard modeling. In particular, in scenarios involving multiple disaster types, GCNs are able to handle nonlinear linkages between disasters, ensuring that their multidimensional dependencies are fully captured.

[0032] Step S24: Modeling is performed based on the deep generative adversarial network and the spatiotemporal correlation graph between the disasters, and a multi-disaster linkage spatiotemporal model is constructed by generating linkage patterns and spatiotemporal evolution laws between different disasters.

[0033] As you can understand, this step builds on the spatiotemporal correlation graph between disasters generated in the previous step. We already know the causal relationships and spatiotemporal dependencies between disaster types. The nodes in the graph represent disaster types, while the edges represent the causal connections and spatiotemporal dependencies between disasters. Based on this foundation, a deep generative adversarial network (GAN) is used to further model the linkage patterns between disasters. A GAN consists of two components: a generator and a discriminator.

[0034] The generator's task is to generate possible disaster linkage patterns based on the spatiotemporal correlation map and the characteristics of disaster types. The generator learns from input data (such as the spatiotemporal relationships of disasters and historical data) to generate disaster linkage scenarios that conform to spatiotemporal dependencies. For example, based on the spatiotemporal relationship between frost and soil salinity, the generator can generate spatiotemporal evolution patterns of alternating or linked occurrences of frost and salinized soils.

[0035] The discriminator's task is to evaluate whether the linkage patterns generated by the generator conform to the real-world patterns of disaster linkage and whether the generated disaster linkage scenarios are consistent with actual observed historical data. By comparing the generated patterns with the spatiotemporal relationships in historical data, the discriminator provides feedback to guide the generator in adjusting its output, making the generated disaster linkage patterns more realistic and consistent with actual conditions.

[0036] Through adversarial training of the generator and discriminator, the generative adversarial network gradually learns the complex interaction patterns and spatiotemporal evolution of disasters. Ultimately, the generator is able to produce a spatiotemporal model of multi-hazard interaction. This model not only captures the direct and indirect interaction effects between disasters, but also simulates the compound impacts of disasters at different temporal and spatial scales. This model provides deep insights into multi-hazard interaction scenarios and can provide dynamic predictions for disaster risk assessment and early warning.

[0037] Step S3: The multi-disaster linkage spatiotemporal model is subjected to spectral clustering and a preset gradient boosting tree fusion process to obtain a disaster impact prediction model; It can be understood that this step, through the fusion of spectral clustering and gradient boosting, can effectively extract key disaster features from complex spatiotemporal data and make high-precision predictions of disaster impacts. Spectral clustering helps reveal the spatiotemporal relationships and similarities between disasters, while gradient boosting accurately models the nonlinear characteristics of disaster impacts by gradually optimizing the loss function. Ultimately, this fusion method enables the disaster prediction model to not only handle single disasters but also adapt to complex scenarios involving multiple disaster types, providing more accurate and comprehensive support for subsequent disaster risk assessments and response strategies. In this step, step S3 includes steps S31, S32, S33, and S34.

[0038] Step S31: performing local weighting processing on the multi-disaster linkage spatiotemporal model based on a local weighted regression algorithm to obtain weighted disaster spatiotemporal characteristic data; This step first requires using the spatiotemporal characteristic data of disasters from the multi-hazard linkage spatiotemporal model. This data includes information on the variations of various disasters (such as frost, drought, and windstorms) across time and space. Each disaster dataset is typically represented as a time series, where each time point corresponds to a disaster impact value (such as temperature, precipitation, wind speed, etc.). This step then requires determining the target prediction point. This target data point is typically the disaster impact value at a specific time point and spatial region. For example, predicting the intensity of drought in a specific region over a certain period of time, or predicting the impact of frost on the growth of a specific crop.

[0039] Once the target data point is determined, the next key step is to calculate the weights of the surrounding data points. Local weighted regression assigns weights to each data point using a distance function. This step uses a Gaussian kernel function to measure the similarity between the data point and the target point. This process calculates weights based on the spatial distance (and temporal distance) between each disaster data point and the target point, assigning higher weights to data points closer to the target point and lower weights to data points farther away. These weights play a crucial role in the subsequent regression process, controlling the impact of each data point on the prediction results.

[0040] After calculating the weights, we then use a weighted regression model to fit the model. Specifically, local weighted regression uses weighted least squares to calculate the regression coefficients.

[0041] As you can understand, through local weighted regression, we ultimately obtain weighted regression coefficients for each disaster in the area surrounding the target data point. These regression coefficients represent the degree of impact of each disaster on the target point. By combining the regression coefficients with the corresponding disaster spatiotemporal characteristic data, we can generate weighted disaster spatiotemporal characteristic data.

[0042] Step S32: converting the weighted disaster spatiotemporal characteristic data into a graph structure and performing spectral clustering analysis to obtain disaster spatiotemporal characteristic clustering data; It is understood that the nodes in this step represent different disaster types and regions. For example, a node can be "the impact of frost disaster in a certain area" or "the intensity of drought disaster in a certain period of time."

[0043] Edges represent the spatiotemporal connections between different nodes (hazards). For example, frost may affect soil salinization, and since these two are spatiotemporally connected, an edge is added to the graph to represent their interaction. Edge weights are set based on factors such as the strength of the causal relationship between the hazards and their spatiotemporal proximity.

[0044] Based on the spatiotemporal dependencies between disasters, a similarity matrix is ​​constructed. Each element in the matrix represents the similarity between one disaster and another. This similarity is the similarity calculated by the Gaussian kernel function in the previous step. Then, by calculating the eigenvalues ​​and eigenvectors of the Laplacian matrix (the degree matrix minus the similarity matrix), which is a diagonal matrix where each element is the degree of a node (i.e., the sum of the weights of the edges connecting to the node), the eigenvectors corresponding to the smallest eigenvalues ​​are selected. These eigenvectors provide the embedding space of the nodes (i.e., mapping the nodes to a low-dimensional space). The eigenvectors reflect the similarity between nodes, so clustering can be performed using these eigenvectors as representations of the nodes.

[0045] By calculating the eigenvalues ​​and eigenvectors of the Laplacian matrix and selecting the eigenvectors corresponding to the smallest eigenvalues, these eigenvectors provide the node embedding space (i.e., mapping the nodes to a low-dimensional space). The eigenvectors reflect the similarity between nodes. Finally, the K-means clustering algorithm is used to cluster the eigenvectors, generating clustered data on the spatiotemporal characteristics of disasters. Each cluster represents a group of spatiotemporally similar disasters with similar impact patterns and linkage effects. Through clustering, disasters can be grouped according to their spatiotemporal characteristics, providing valuable information for subsequent disaster prediction and risk assessment.

[0046] After spectral clustering, the resulting clustered data of disaster spatiotemporal characteristics can be used to represent the spatiotemporal relationship patterns between different disasters. This clustered data not only assigns a cluster label to each disaster but also reveals spatial and temporal similarities and dependencies between disasters. For example, in some regions, frost and drought may co-occur within a given season and exacerbate each other, while in other regions, they may occur completely independently. The clustering results clearly indicate which disasters share similar spatiotemporal characteristics and which have more similar patterns of impact on crops.

[0047] Step S33: clustering the spatiotemporal characteristics of disasters as a Bayesian network node, where the node represents the disaster risk of different regions and crops under different disaster scenarios. The conditional probability of different disaster combinations occurring is calculated through the Bayesian network, and the weight of each disaster type when a compound disaster occurs is estimated; It can be understood that this step converts the spatiotemporal clustering data into nodes in a Bayesian network. Each cluster represents a group of similar disaster characteristics across time and space. The nodes of the Bayesian network represent different disaster types (such as frost, drought, and windstorms) as well as the disaster risks for specific regions or crops. The value of each node represents the risk level or intensity of the disaster under specific spatiotemporal conditions.

[0048] In a Bayesian network, directed edges connect nodes to represent causal relationships between disasters. We leverage the causal relationship data between disasters obtained previously through Granger causality testing to construct directed edges in the network. For example, if frost can predict changes in soil salinity in historical data, or if droughts can lead to exacerbated windstorms, we add directed edges between these disaster nodes in the Bayesian network to represent the causal relationship. The direction of each edge indicates the direction of the causal relationship, and the strength of the edge indicates the strength of the causal influence.

[0049] The steps of constructing a Bayesian network include determining the network structure and defining a conditional probability table (CPT). The network structure is determined based on the causal relationship between disasters (such as the causal path obtained by the Granger causality test), and appropriate parent nodes and child nodes are added to each disaster node in the network to form a directed acyclic graph (DAG).

[0050] For each node, a conditional probability table (CPT) needs to be defined. This is the probability distribution of the node's state given the state of its parent node. For example, the probability of frost occurring given a drought, or the probability of crop loss given a windstorm.

[0051] Once the structure of the Bayesian network and the conditional probability table are defined, we can use Bayesian reasoning to calculate the conditional probabilities of different disaster combinations. The Bayesian network uses probabilistic reasoning to calculate the probability distribution of each disaster node under different conditions. This involves two steps: forward propagation and backward reasoning. Forward propagation: Starting from known conditions (such as the occurrence of certain disasters), forward propagation calculates the conditional probabilities of other disasters in the Bayesian network. This helps us understand the probability of each disaster type when a specific disaster combination occurs. Backward reasoning: If the probability of certain disasters is known, we can infer the conditional probabilities of other disasters. Through this reasoning method, we can estimate the probability of different disaster combinations.

[0052] For example, if the probability of frost and drought occurring in a certain period is high, the Bayesian network can calculate the conditional probability of wind disasters occurring when these disasters occur, as well as the impact of each disaster on crops.

[0053] The conditional probabilities derived through Bayesian inference can be further used to estimate the weight of each disaster type when a complex disaster occurs. A complex disaster refers to multiple disasters occurring simultaneously or alternately, resulting in a combined impact on crops. Bayesian networks can calculate the probability of each disaster type in a complex disaster scenario and estimate the "weight" of each disaster—the extent of its impact on crops or a region—based on the conditional probabilities and causal relationships between the different disasters.

[0054] Step S34: using the disaster spatiotemporal characteristic clustering data and the weight of each disaster type when a compound disaster occurs as input features, combining the first information with a weighted logistic regression algorithm for training and modeling to obtain a disaster impact prediction model.

[0055] It can be understood that the weighted logistic regression algorithm in this step is a classic regression method, which is suitable for predicting binary or multi-classification problems. In this step, we use weighted logistic regression to train the disaster impact prediction model. The goal of the logistic regression model is to predict whether the impact of disasters on crops reaches a certain threshold based on the input feature data, or to evaluate the risk level of crops suffering disasters. The core of weighted logistic regression is to better adapt to the imbalance or regional differences in the data by assigning different weights to different samples. For example, certain types of disasters may have a greater impact on crops in specific areas. At this time, higher weights can be given to data samples in these areas to ensure that the model has better prediction results in these key areas. This method takes into account the weight of the samples, so that high-weight samples (such as disaster data in high-risk areas) are given more attention. Specifically, the loss function of logistic regression is as follows: ,in, L The loss weight for each sample is, W i is the weight of the i-th sample, which is related to factors such as the importance of the sample, regional risk or sensitivity, is the true label of the i-th sample, is the predicted probability of the i-th sample.

[0056] Through the weighted logistic regression training process, we can obtain regression coefficients and use these coefficients to model the relationship between input features and output outcomes. During the training process, the algorithm optimizes the loss function, enabling the model to accurately predict the impact of disasters based on the input spatial and temporal characteristics of disasters, disaster weights, and other environmental information.

[0057] After training is completed, the model can predict whether crops will be affected by disasters and the extent of the impact of disasters in a certain period of time in the future based on the new disaster spatiotemporal characteristic data.

[0058] Step S4: performing adaptive filtering and graded judgment processing on the disaster impact prediction model to obtain the disaster impact level of multiple disaster types on each crop in all preset areas; It can be understood that this step, through adaptive filtering and hierarchical judgment processing, effectively reduces the impact of external noise and uncertainty on disaster prediction results, thereby improving the accuracy and reliability of disaster predictions. Adaptive filtering can dynamically adjust the prediction process to adapt to disaster data under different meteorological and environmental conditions, ensuring the stability of prediction results. Hierarchical judgment processing generates accurate disaster risk assessment results based on the actual disaster impact level, making disaster warning and response strategies more specific and personalized. In this step, step S4 includes steps S41, S42, and S43.

[0059] Step S41: predicting impact data based on the disaster impact prediction model, and denoising and preprocessing the predicted disaster impact data based on an integrated variational fuzzy adaptive filtering method to obtain filtered disaster impact data; It's understood that this step first uses the previously constructed disaster impact prediction model to predict disaster impact based on input temporal and spatial disaster data (such as meteorological data, crop growth monitoring data, and historical disaster data). The disaster impact prediction model generates a series of disaster impact predictions for different regions and crops. These predictions reflect the potential impact of different disaster types (such as frost, drought, and windstorms) on crops, typically expressed as loss percentages or risk levels.

[0060] For example, through the prediction model, the system may calculate that frost will have a 20% impact on crops in one area, while drought will have a 30% impact on crops in another area. These prediction results provide data support for subsequent disaster warning and management decisions.

[0061] Then, by identifying the underlying patterns of change in the data within the time series and inferring the distribution of noise, the stability of the prediction is improved. The integrated variational fuzzy adaptive filtering method in this step combines the advantages of variational filtering and fuzzy logic. First, this step uses variational filtering to smooth and remove noise by minimizing the model's energy function. This method improves the stability of the prediction by identifying the underlying patterns of change in the data within the time series and inferring the distribution of noise. In disaster impact prediction, variational filtering can model the uncertainty in time series data and effectively remove interference caused by noise or sudden events.

[0062] Secondly, fuzzy logic is used to address data uncertainty. In the disaster forecasting process, fuzzy logic can help the system understand and process incomplete or ambiguous input data (such as fuzzy descriptions of crop growth status) and transform it into more accurate output results. Fuzzy adaptive filtering adapts to dynamic changes in forecast data by adjusting fuzzy rules and membership functions, thereby enhancing the system's adaptability and robustness.

[0063] During the integrated variational fuzzy adaptive filtering process, the filter parameters are dynamically adjusted based on the characteristics of each predicted value. Through this adaptive mechanism, the system adjusts the weight of each data point in real time, assigning higher weight to predicted values ​​near the target point and lower weight to data farther away. Adaptive filtering can reduce the impact of noise from areas far from the target on prediction results based on the spatiotemporal characteristics of disaster data.

[0064] Step S42: classify the disaster impact areas of the preset areas based on the principal component analysis method to obtain the disaster impact level of each preset area; It can be understood that this step first standardizes the disaster impact data and then calculates the covariance matrix of the standardized data, which is used to describe the linear relationships between different features. Next, the covariance matrix undergoes eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors. The eigenvalue represents the importance of each principal component (i.e., the proportion of data variance explained), while the eigenvector represents the direction of the principal component. Finally, based on the size of the eigenvalue, the first few principal components with the largest eigenvalues ​​are selected. These principal components retain the majority of the variation in the data. By selecting an appropriate number of principal components, high-dimensional data can be mapped to a low-dimensional space.

[0065] After principal component analysis (PCA), we project the disaster impact data onto the newly generated principal component space. Each principal component represents a major source of variation in the original disaster data. For example, the first principal component might represent the combined effects of frost and drought, while the second principal component might represent the relationship between wind damage and soil moisture.

[0066] Based on the principal component scores (i.e., the projection value of each sample on each principal component), we can aggregate the disaster impacts for each region. For example, for a specific pre-defined region, we weight the impact values ​​of different disasters by the principal component scores to obtain the comprehensive disaster impact level for that region.

[0067] The PCA results allow us to assign a comprehensive impact score to each disaster-affected area, representing the overall risk level of the area under multiple disaster scenarios. We then use these comprehensive impact scores to rank the disaster impacts of the pre-defined areas.

[0068] For example, regions can be divided into multiple levels based on the range of comprehensive impact scores, such as: Low risk (0-10%): The disaster has a minor impact and causes minor damage to crops; Medium risk (10%-30%): The disaster has a moderate impact and some preventive measures may be needed; High risk (30%-50%): The disaster has a significant impact, causing serious damage to crops and requiring immediate response measures; Very high risk (>50%): The impact of the disaster is extremely severe, with heavy crop losses and requiring urgent response.

[0069] The above classification process ultimately results in a disaster impact rating for each pre-defined region. These ratings provide decision-makers with a clear assessment of disaster risk and can help formulate regional response strategies, such as allocating water resources, implementing disaster prevention and mitigation measures, or adjusting crop planting plans.

[0070] Step S43: Based on the distribution of grain and cotton crops in each preset area, weighted analysis and reclassification of the disaster-affected areas in the preset areas are performed to obtain the disaster impact level of multiple disasters on each crop in all preset areas.

[0071] It will be appreciated that this step first obtains grain and cotton crop distribution data within a predetermined area. This data typically includes information such as the planting area, planting density, and growth stage of each crop (e.g., grain crops, cotton, etc.) at different geographic locations. Crop distribution data can be collected using remote sensing technology, agricultural survey data, or farmland monitoring networks.

[0072] Different crops respond differently to disasters. For example, cotton may be more sensitive to frost, while grain crops may be more vulnerable to drought. Crop distribution data can be used to more accurately analyze the impact of crop disasters within each region.

[0073] After obtaining the crop distribution data, it is necessary to conduct a weighted analysis of the disaster impact area in each pre-set region. The core of the weighted analysis is to adjust the disaster impact value based on the sensitivity and distribution of different crops. Among them, the weighted analysis considers the following factors: Disaster sensitivity by crop type: Different crops have varying susceptibility to disasters. For example, cotton is highly sensitive to frost, while grain crops are more tolerant to drought. For each crop, a sensitivity weight corresponding to the corresponding disaster type needs to be defined. These weights reflect the crop's responsiveness to a specific disaster.

[0074] Crop density: The density and regional distribution of crops directly influences the overall impact of a disaster on crops. If a region has a larger area planted with a particular crop, the total cost of disaster losses will be higher. Therefore, crop density plays a significant role in the weighting process.

[0075] The weighted analysis is calculated using the following formula: in, For the preset areaR k Internal crops C j The weighted disaster impact value of For crops C j Disaster D i The sensitivity weight of For disasters D i For crops C j The impact value of For crops C j In the preset area R k The distribution density within is n, and n is the total number of disaster types.

[0076] After weighted analysis, the disaster impact value we obtain for each region and crop will represent its comprehensive risk level. Next, we need to reclassify based on these weighted disaster impact values.

[0077] The purpose of reclassification is to set different risk levels based on the impact of disasters on different crops and regions. For example, we can classify the disaster impact levels as follows: Low risk level: The disaster impact value is low, crop losses are small, and the risk is low.

[0078] Medium risk level: The disaster impact is moderate, crops are severely affected, and preventive measures need to be taken.

[0079] High risk level: The disaster impact value is high, crops are seriously threatened, and emergency response measures are urgently needed.

[0080] Very high risk level: The disaster impact value is extremely high, crop losses are serious, and disaster prevention measures need to be taken immediately.

[0081] The disaster impact level for each region and crop is compared with a set threshold by weighted impact value. If the disaster impact value exceeds a certain threshold, the corresponding crop and region will be assigned a high risk level; otherwise, it will be assigned a low risk level.

[0082] For example, if the frost impact in a region is 40%, and grain crops are less sensitive to frost, the disaster impact level in the region may be assessed as medium risk; while for cotton crops, the impact of frost may be more severe, and the region may be assessed as high risk.

[0083] Step S5: generating an adversarial network based on the disaster impact levels and historical warning response strategies to determine the response strategies corresponding to all disaster impact levels for each crop.

[0084] It can be understood that this step uses a generative adversarial network (GAN) to generate crop disaster response strategies. This step can dynamically generate the most appropriate response measures based on the disaster impact level and historical response strategies for each crop. Compared with traditional methods, GAN can automatically generate personalized and flexible response strategies for complex and changing disaster scenarios. The generated response strategies can more accurately reflect the actual impact of different disasters on crops, ensuring that crops are properly protected. In this step, step S5 includes steps S51, S52, and S53.

[0085] Step S51: Generate an adversarial network based on the disaster impact level and historical warning response strategies, wherein the generative adversarial network includes a generator and a discriminator. The generator generates response strategies for each disaster level based on the crop disaster impact level data, and the discriminator evaluates whether the generated response strategies for each disaster level meet preset conditions based on the historical warning response strategies. It's understood that the disaster impact level in this step is derived from the analysis results in the previous step and represents the disaster risk level for each crop in different regions and disaster scenarios. The disaster impact level is quantified as a numerical value, typically categorized as low, medium, or high. Historical early warning response strategies include past disaster response measures and strategies, such as irrigation measures to combat drought or protective measures against frost. This historical data helps the discriminator determine whether the generated response strategy is effective.

[0086] The generator's task is to generate response strategies based on input data on crop disaster impact levels. For each disaster level, the generator attempts to generate appropriate response measures for each crop by learning the characteristics and patterns of historical early warning response strategies. The generator inputs the crop disaster impact level under different disaster scenarios and maps this input data into a space of response strategies.

[0087] The discriminator's task is to evaluate whether each response strategy output by the generator meets the standards of historical warning response strategies. Based on the effectiveness of historical response strategies, the discriminator determines whether the generated response strategy is reasonable and feasible.

[0088] After adversarial training, the generator will output a response strategy corresponding to each disaster level. This process will adjust and optimize the response strategy based on the severity of the disaster and regional characteristics. Each predefined region and crop will be tailored to its disaster impact level, providing personalized, targeted response measures. Adversarial training can be achieved by optimizing the following objective function: in, is the loss function of the Generative Adversarial Network (GAN), which represents the objective function of the generator and discriminator in adversarial training. To represent the sample Distribution from real data The expected value of For the discriminator For real data samples The output, The input latent variable z comes from the distribution P of the latent space z The expected value of For the discriminator Evaluation of coping strategies on the generator output, represents the “realism” score of the discriminator on the generated strategy.

[0089] For example, low-risk crops may only require some minor protection against frost damage, while high-risk crops may require more complex responses such as greenhouse heating.

[0090] For drought disasters, medium-risk areas may generate coping strategies such as increasing irrigation or reducing crop water demand.

[0091] Step S52: performing adversarial training and prediction based on the historical warning response strategy and the generative adversarial network to obtain a predicted response strategy output by the trained generative adversarial network; It's understandable that through adversarial training and prediction, this step can generate the most appropriate crop response strategies for different disaster scenarios, ensuring personalized and precise strategies. By learning from historical warning and response strategies, the generative adversarial network can generate high-quality responses in complex environments with multiple disasters. The discriminator's feedback mechanism ensures that the generated strategies meet actual needs and the standards of historically effective response strategies. Ultimately, this process enhances the intelligence of disaster warning and management systems, helping agricultural managers make more informed decisions, reducing disaster losses and improving emergency response efficiency.

[0092] Step S53: Optimize and select the predicted response strategies based on a particle swarm optimization algorithm to obtain the optimal response strategies corresponding to each disaster impact level of each crop.

[0093] It's easy to understand that this step first initializes a particle swarm, where each particle represents a candidate solution for a response strategy. Each particle's position represents a potential response strategy (e.g., a specific response to a disaster), while the particle's velocity determines the particle's search direction and step size in space.

[0094] Next, the particle swarm calculates its fitness function. The goal is to find a response strategy that optimizes multiple objectives. In disaster response strategy optimization, the fitness function typically consists of multiple objectives, such as: Minimizing disaster losses: How effectively each strategy minimizes losses under different disaster scenarios; Minimizing costs: The implementation cost of each strategy; Maximizing response efficiency: The speed and timeliness of the response strategy's execution; Sustainability: Ensuring the strategy's continued execution to avoid excessive consumption of the environment or resources. Each particle calculates its fitness value based on these objectives. A higher fitness value indicates a more optimal response strategy.

[0095] The fitness function is as follows: Among them, Fitness is the fitness value, which indicates the overall quality of the response strategy; (Disaster Loss Reduction) indicates the ability of the response strategy to reduce crop or resource losses; (Cost) indicates the cost of implementing the response strategy, including the cost of materials, labor, and other related resources; (Efficiency) is the emergency response efficiency, which indicates the response speed and processing efficiency of the response strategy when a disaster occurs; (Sustainability) is sustainability, which indicates the long-term impact of the response strategy on resources and the environment, ensuring that the response strategy will not lead to excessive consumption of resources or damage to the environment; α, β, γ, and δ are the weights of the objectives, indicating the relative importance of each objective to the final optimization result.

[0096] Through multiple iterations, the particle swarm optimization algorithm continuously searches for and refines response strategies until it finds the one that best meets all objectives. Ultimately, the optimized particle swarm outputs the optimal response strategy for each crop under each disaster impact level. These strategies are the result of comprehensive optimization, balancing multiple factors such as disaster losses, costs, efficiency, and sustainability.

[0097] Example 2:

[0098] like Figure 2 As shown, this embodiment provides a multi-disaster linkage-oriented grain and cotton crop meteorological risk classification early warning system, see Figure 2 The system shown includes an acquisition unit 701 , an acquisition unit 702 , an acquisition unit 703 , an acquisition unit 704 , and an acquisition unit 705 .

[0099] An acquisition unit 701 is configured to acquire first information, wherein the first information includes soil salinity change data, frost weather data, drought weather data, wind disaster weather data, and crop monitoring data within a preset time period; An analysis unit 702 is configured to analyze the spatiotemporal relationship of disaster types on the first information, and to generate a multi-disaster linkage spatiotemporal model based on the spatiotemporal relationship obtained by the analysis; The processing unit 703 is configured to perform spectral clustering and a preset gradient boosting tree fusion process on the multi-disaster linkage spatiotemporal model to obtain a disaster impact prediction model; The judgment unit 704 is configured to perform adaptive filtering and graded judgment processing on the disaster impact prediction model to obtain the disaster impact level of multiple disaster types on each crop in all preset areas; The response unit 705 is configured to generate an adversarial network based on the disaster impact levels and historical warning response strategies, and determine response strategies corresponding to all disaster impact levels for each crop.

[0100] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated on here.

[0101] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

[0102] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A multi-disaster linkage-oriented grain and cotton crop meteorological risk classification early warning method, characterized in that: include: Acquiring first information, the first information including soil salinity change data, frost meteorological data, drought meteorological data, wind disaster meteorological data, and crop monitoring data within a preset time period; Analyzing the spatiotemporal relationship of disaster types using the first information, and building a model based on the spatiotemporal relationship obtained from the analysis to generate a spatiotemporal model for multi-disaster linkage; The multi-disaster linkage spatiotemporal model is subjected to spectral clustering and a preset gradient boosting tree fusion process to obtain a disaster impact prediction model; The disaster impact prediction model is subjected to adaptive filtering and graded judgment processing to obtain the disaster impact level of multiple disaster types on each crop in all preset areas; An adversarial network is generated based on the disaster impact levels and historical warning response strategies to determine response strategies corresponding to all disaster impact levels for each crop.

2. The method for grading and warning meteorological risks of grain and cotton crops for multi-disaster linkage according to claim 1 is characterized in that , analyzing the spatiotemporal relationship of disaster types on the first information, and modeling based on the spatiotemporal relationship obtained by the analysis, including: Extracting the spatial and temporal variation features of the first information into multiple feature tensors based on a spatiotemporal tensor decomposition method to obtain a spatiotemporal feature data set; Performing a relationship analysis on the data in the spatiotemporal feature data set based on the Granger causality test method to obtain a causal relationship between disasters; Based on the causal relationship between the disasters, a spatiotemporal correlation graph between disasters is constructed. In this graph, the types of disasters are considered as nodes, the interactions between disasters are considered as edges, and graph convolution operations are used to extract the spatial and temporal characteristics of the linkage between disasters. Based on the deep generative adversarial network and the spatiotemporal correlation graph between the disasters, modeling is carried out. By generating the linkage patterns and spatiotemporal evolution laws between different disasters, a multi-disaster linkage spatiotemporal model is constructed.

3. The method for grading and warning meteorological risks of grain and cotton crops for multi-disaster linkage according to claim 1 is characterized in that , the multi-disaster linkage spatiotemporal model is subjected to spectral clustering and a preset gradient boosting tree fusion process, including: Performing local weighting processing on the multi-hazard linkage spatiotemporal model based on a local weighted regression algorithm to obtain weighted disaster spatiotemporal characteristic data; Converting the weighted disaster spatiotemporal characteristic data into a graph structure and performing spectral clustering analysis to obtain disaster spatiotemporal characteristic clustering data; The disaster spatiotemporal characteristic clustering data is used as a Bayesian network node, where the node represents the disaster risk of different regions and crops under different disaster scenarios. The conditional probability of different disaster combinations occurring is calculated through the Bayesian network, and the weight of each disaster type when a compound disaster occurs is estimated; The disaster spatiotemporal characteristic clustering data and the weight of each disaster type when a compound disaster occurs are used as input features, and training modeling is performed in combination with the first information and a weighted logistic regression algorithm to obtain a disaster impact prediction model.

4. The method for grading and warning meteorological risks of grain and cotton crops for multi-disaster linkage according to claim 1 is characterized in that , the disaster impact prediction model is subjected to adaptive filtering and hierarchical judgment processing, including: The impact data is predicted based on the disaster impact prediction model, and the predicted disaster impact data is denoised and preprocessed based on the integrated variational fuzzy adaptive filtering method to obtain filtered disaster impact data; Based on the principal component analysis method, the disaster impact areas of the preset areas are classified to obtain the disaster impact level of each preset area; Based on the distribution of grain and cotton crops in each preset area, the disaster-affected areas in the preset areas are weighted analyzed and reclassified to obtain the disaster impact level of multiple disasters on each crop in all preset areas.

5. The method for grading and warning meteorological risks of grain and cotton crops for multi-disaster linkage according to claim 1 is characterized in that Based on the disaster impact level and historical warning response strategies, an adversarial network is generated to determine the response strategies corresponding to all disaster impact levels of each crop, including: A generative adversarial network is generated based on the disaster impact level and historical warning response strategies, wherein the generative adversarial network includes a generator and a discriminator. The generator generates response strategies for each disaster level based on crop disaster impact level data, and the discriminator evaluates whether the generated response strategies for each disaster level meet preset conditions based on historical warning response strategies. Conducting adversarial training and prediction based on the historical warning response strategy and the generative adversarial network to obtain a prediction response strategy output by the trained generative adversarial network; The prediction response strategy is optimized and selected based on the particle swarm optimization algorithm to obtain the optimal response strategy corresponding to each disaster impact level of each crop.

6. A multi-disaster linkage-oriented grain and cotton crop meteorological risk classification early warning system, characterized by: include: an acquisition unit, configured to acquire first information, wherein the first information includes soil salinity change data, frost meteorological data, drought meteorological data, wind disaster meteorological data, and crop monitoring data within a preset time period; an analysis unit configured to analyze the spatiotemporal relationship of disaster types on the first information, and to generate a multi-disaster linkage spatiotemporal model based on the spatiotemporal relationship obtained by the analysis; A processing unit, configured to perform spectral clustering and a preset gradient boosting tree fusion process on the multi-disaster linkage spatiotemporal model to obtain a disaster impact prediction model; A judgment unit, configured to perform adaptive filtering and hierarchical judgment processing on the disaster impact prediction model to obtain the disaster impact level of multiple disaster types on each crop in all preset areas; The response unit is used to generate an adversarial network based on the disaster impact level and historical warning response strategies, and determine the response strategies corresponding to all disaster impact levels of each crop.

7. The multi-disaster linkage-oriented grain and cotton crop meteorological risk grading early warning system according to claim 6 is characterized in that: The analysis unit comprises: A first analysis subunit is configured to extract the spatial and temporal variation features of the first information into a plurality of feature tensors based on a spatiotemporal tensor decomposition method to obtain a spatiotemporal feature data set; A second analysis subunit is configured to perform a relationship analysis on the data in the spatiotemporal feature data set based on the Granger causality test method to obtain a causal relationship between disasters; The third analysis subunit is configured to construct a spatiotemporal correlation graph between disasters based on the causal relationship between the disasters, wherein the spatial and temporal characteristics of the linkage between disasters are extracted by graph convolution operation by treating the disaster types as nodes in the graph and the interactions between the disasters as edges in the graph; The fourth analysis subunit is used to build a model based on the deep generative adversarial network and the spatiotemporal correlation graph between the disasters, and to construct a multi-disaster linkage spatiotemporal model by generating linkage patterns and spatiotemporal evolution laws between different disasters.

8. The multi-disaster linkage-oriented grain and cotton crop meteorological risk grading early warning system according to claim 6 is characterized in that: The processing unit includes: A first processing subunit is configured to perform local weighted processing on the multi-hazard linkage spatiotemporal model based on a local weighted regression algorithm to obtain weighted disaster spatiotemporal characteristic data; A second processing subunit is configured to convert the weighted disaster spatiotemporal characteristic data into a graph structure and perform spectral clustering analysis to obtain disaster spatiotemporal characteristic clustering data; A third processing subunit is configured to use the disaster spatiotemporal characteristic clustering data as a Bayesian network node, wherein the node represents the disaster risk of different regions and crops under different disaster scenarios, calculate the conditional probability of different disaster combinations through the Bayesian network, and estimate the weight of each disaster type when a compound disaster occurs; The fourth processing subunit is used to use the disaster spatiotemporal feature clustering data and the weight of each disaster type when a compound disaster occurs as input features, combine the first information and the weighted logistic regression algorithm for training and modeling, and obtain a disaster impact prediction model.

9. The multi-disaster linkage-oriented grain and cotton crop meteorological risk grading early warning system according to claim 6 is characterized in that: The judging unit includes: The first judgment subunit is used to predict the impact data according to the disaster impact prediction model, and to denoise and preprocess the predicted disaster impact data based on the integrated variational fuzzy adaptive filtering method to obtain filtered disaster impact data; The second judgment subunit is used to classify the disaster impact areas of the preset areas based on the principal component analysis method to obtain the disaster impact level of each preset area; The third judgment subunit is used to perform weighted analysis and reclassification of disaster-affected areas in the preset areas based on the distribution of grain and cotton crops in each preset area, and obtain the disaster impact level of multiple disasters on each crop in all preset areas.

10. The multi-disaster linkage-oriented grain and cotton crop meteorological risk grading early warning system according to claim 6 is characterized in that: The response unit includes: A first response subunit is configured to generate an adversarial network based on the disaster impact level and historical warning response strategies, wherein the generative adversarial network includes a generator and a discriminator. The generator generates response strategies for each disaster level based on the crop disaster impact level data, and the discriminator evaluates whether the generated response strategies for each disaster level meet preset conditions based on the historical warning response strategies. A second response subunit is configured to perform adversarial training and prediction based on the historical warning response strategy and the generative adversarial network, and obtain a predicted response strategy output by the trained generative adversarial network; The third response subunit is used to optimize and select the predicted response strategy based on the particle swarm optimization algorithm to obtain the optimal response strategy corresponding to each disaster impact level of each crop.

Images