Crop rainstorm disaster quantitative risk assessment method based on CNN-LSTM hybrid neural network

Abstract

The invention discloses a crop rainstorm disaster quantitative risk assessment method based on a CNN-LSTM hybrid neural network, and the method comprises the steps: obtaining multi-source data, carrying out the standardization processing, and carrying out the time-space superposition to construct a three-dimensional feature matrix; based on a CNN-LSTM hybrid neural network model, respectively extracting a spatial feature vector and a time sequence feature vector of the disaster situation data, completing fusion to obtain a fused feature vector, and outputting a risk probability and loss intensity; and continuing to comprehensively consider the crop loss caused by the risk probability and the loss intensity, completing the construction of a rainstorm disaster-crop coupling loss function, and finally generating the crop rainstorm disaster risk probability, the main crop loss intensity and the rainstorm disaster-crop coupling total loss based on the intelligent grid rainfall and wind speed forecast data. According to the method, the quantitative loss of crops is estimated by integrating rainstorm disaster-crop coupling dynamic response, and effective transformation from static disaster assessment to dynamic multi-source fusion response assessment is realized.

Description

Technical Field

[0001] The present invention belongs to the field of meteorological science and technology, and in particular relates to a quantitative risk assessment method for crop rainstorm disasters based on a CNN-LSTM hybrid neural network. Background Art

[0002] The frequency and destructiveness of rainstorm disasters have significantly increased, posing an increasingly severe threat to agricultural production. Traditional risk assessment methods, which often rely on historical disaster statistics or static meteorological threshold models, struggle to effectively integrate multi-source heterogeneous data (such as high-resolution remote sensing imagery and soil moisture time series data). These methods also fail to capture the complex nonlinear relationships between weather, environment, and crop response within the rainstorm disaster chain. For example, traditional models based on indicator weighting often overlook the dynamics of surface water bodies and differences in crop sensitivity during the growth period, resulting in insufficient spatial resolution and relatively poor timeliness in risk assessment results.

[0003] Deep learning technology has provided a breakthrough tool for rainstorm disaster risk assessment. Research has shown that machine learning algorithms such as random forests (RF) and support vector machines (SVM) significantly outperform traditional statistical methods in assessing rainstorm disaster risk, particularly when dealing with high-dimensional meteorological data and nonlinear relationships. However, existing research has tended to focus on single hazard factors (such as rainfall intensity) or static risk assessments, and has yet to fully integrate spatiotemporal coupling characteristics and crop physiological response mechanisms. For example, rice is three to five times more sensitive to waterlogging at the heading stage than at the seedling stage, yet traditional models often overlook these critical agronomic parameters. Summary of the Invention

[0004] Purpose of the invention: In view of the fact that the existing technology fails to better meet the needs of refined weather forecast services, the present invention provides a spatiotemporal verification method for precipitation forecast based on target object diagnosis.

[0005] Technical solution: To achieve the above-mentioned purpose, the present invention adopts the following technical solution: a quantitative risk assessment method for crop rainstorm disasters based on a CNN-LSTM hybrid neural network, comprising the following steps: S1, data preparation and preprocessing, acquiring multi-source data and standardizing the multi-source data to obtain multi-source grid data, the multi-source data including meteorological data, soil moisture data, crop growth monitoring data, geographical environment data and historical disaster data; the meteorological data includes actual and forecast precipitation, actual and forecast wind speed; S2, spatiotemporal superposition is performed on the multi-source grid data to construct a three-dimensional feature matrix X, X=[T, Lat, Lon, C], wherein T is the number of time steps, Lat is the latitude, Lon is the longitude, and C is the number of channels; the channels include real-time hourly precipitation, real-time daily precipitation, real-time process cumulative precipitation, forecast hourly precipitation, forecast daily precipitation, forecast process cumulative precipitation, real-time hourly maximum wind speed, forecast hourly maximum wind speed, soil moisture content 10 cm below the ground, soil moisture content 20 cm below the ground, soil moisture content 40 cm below the ground, NDVI index, crop growth period code, slope, TDI index, crop disaster area ratio, crop disaster area ratio, crop disaster area ratio, crop yield reduction rate, and agricultural economic loss; S3, a CNN-LSTM hybrid neural network model is constructed, and a spatial feature vector of meteorological data, geographical environment data, crop growth data and disaster data is extracted based on a CNN neural network model ; a time sequence feature vector of precipitation and wind speed data sequence, soil moisture content, crop growth data and disaster data is extracted based on an LSTM model containing an attention mechanism ; multi-source features of the spatial features and the time sequence features extracted by the CNN neural network model and the LSTM model are fused to complete the model construction to obtain a fusion feature vector , output a risk probability and a loss intensity data; S4, a loss function of a storm disaster-crop coupling is constructed by comprehensively considering the crop loss caused by the risk probability and the loss intensity ; S5, according to the model of S3 and the loss function of S4, a crop storm disaster risk probability, a main crop loss intensity and a storm disaster-crop coupling total loss in a future storm process are generated based on intelligent grid precipitation and wind speed prediction data.

[0006] Further, the data preparation and preprocessing of step S1 first performs the following processing on the multi-source data: (1) normalization of meteorological data: first, the meteorological data is down-scaled to obtain 1km×1km meteorological grid data, and the meteorological grid data is normalized respectively; the meteorological data includes real-time hourly precipitation, real-time daily precipitation, real-time process cumulative precipitation, real-time hourly maximum wind speed, forecast hourly precipitation, forecast daily precipitation, forecast process cumulative precipitation, and forecast hourly maximum wind speed; (2) The three-layer soil moisture data were downscaled to obtain three layers of soil relative humidity with a spatial resolution of 1 km × 1 km. The soil relative humidity is the ratio of the current water content to the field water holding capacity. The three layers of soil relative humidity include the soil relative humidity at 10 cm below the surface, 20 cm below the surface, and 40 cm below the surface. (3) The NDVI index obtained by satellite is processed into the daily NDVI index, and the crop growth period information is converted into categorical variable data according to the crop growth period to form the growth period stage coding data; (4) The slope is obtained by the following formula: and microtopography drainage capacity index TDI, , Where, represents the standard deviation of elevation, Indicates the horizontal distance, slope indicates the slope degree obtained according to the slope analysis in Gis. is the total runoff within the catchment area, It represents the flow rate per unit area; Then, the three layers of soil relative moisture, daily NDVI index, slope tan (slope), and microtopography drainage capacity index TDI of each grid were standardized respectively; Finally, based on the crop planting area and planting distribution, the county-scale disaster data were processed into standardized grid data with a spatial resolution of 1km*1km.

[0007] Furthermore, in the three-dimensional feature matrix X in step S2, Lat=3, Lon=3; the time step T=13, including 5 days before the rainstorm disaster to 7 days after the rainstorm disaster; and the number of channels C=20.

[0008] Furthermore, step S3 constructs the CNN-LSTM hybrid neural network model, which specifically includes the following steps: The CNN neural network model described in S3.1 obtains the spatial feature vectors of meteorological data, geographical environment data, crop growth data and disaster data, as well as the spatial feature vector sequences of the actual hourly precipitation, actual daily precipitation, actual process cumulative precipitation, actual hourly maximum wind speed, three-layer soil relative humidity, and daily NDVI index at each time step, and obtains a 128-dimensional spatial feature vector ; The LSTM model with attention mechanism described in S3.2 includes 13 time steps, and the attention mechanism is applied separately. The weight of each attention mechanism is calculated by the following formula to complete the construction of the LSTM model with attention mechanism. , Where, is the weight coefficient of the output value of the t-th time step, represents the weight function at the t-th time step, represents the weight function at the i-th time step, T represents the number of time steps, and its value is 13; is the hyperbolic tangent activation function, is the weight matrix, is the bias term, is the input vector at this time step The transpose of Represents the output vector at the tth time step;

[0009] Then the time series feature vector is extracted through the LSTM model including the attention mechanism , , Where, and are the weight coefficient and output vector at the t-th time step, is a 256-dimensional time series feature vector; S3.3 extracts the 128-dimensional spatial feature vectors from steps S3.1 and S3.2 and 256-dimensional time series feature vector The fusion is then weighted by the attention mechanism to obtain the final multi-source fusion feature vector , S3.4 The multi-source fusion feature vector obtained from steps S3.1-S3.3 , the model finally outputs the risk probability and loss intensity Two variables, among which: risk probability is obtained by binary classification method, with a value range of 0 to 1; loss intensity From the following formula, , Where, is the basic reduction rate, ranging from 0 to 1; is the growth period sensitivity coefficient of the crop at the current growth stage, It is the final multi-source fusion feature vector.

[0010] Furthermore, step S4 includes the following steps:

[0011] First, based on the risk probability, the binary cross entropy BCE method is used to calculate the loss L of the risk probability prob， , Where, L prob represents the risk probability of loss, The model predicts the probability of disaster occurrence for the sample. is the risk probability; Then, the loss intensity loss is calculated by the root mean square error (MSE) of the crop growth period sensitivity coefficient. , calculated by the following formula, , Where, and are the true loss intensity of the sample and the predicted loss intensity of the sample, respectively, and N represents the number of samples; Finally, the total loss of rainstorm disaster-crop coupling is established , , Where, It is a hyperparameter used to balance the two factor terms and its value is 0.7.

[0012] Furthermore, in step S5, the risk probability of crop rainstorm disasters, the loss intensity of major crops, and the total loss of rainstorm disaster-crop coupling are generated through the intelligent grid precipitation and wind speed forecast data, and a distribution map of risk probability, loss intensity, and rainstorm disaster-crop coupling total loss is formed with a spatial resolution of 1km×1km.

[0013] Beneficial effects: Compared with the existing technology, it has the following advantages: (1) By adopting the method of the present invention, multi-source data such as meteorological, soil, crop, and geographic information are used in model construction to estimate the quantitative loss of crops by integrating the dynamic response of rainstorm disaster-crop coupling, while dynamically capturing the precipitation-soil moisture hysteresis effect, thus achieving an effective transition from static disaster assessment to dynamic multi-source fusion response assessment; (2) The method of the present invention constructs a CNN-LSTM hybrid neural network model and uses deep learning methods to fully consider the effects of various factors in the disaster chain from rainstorm disasters to crop yield reduction disasters. The final risk assessment product greatly enriches the existing agricultural meteorological service products. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the overall technical process of the present invention.

[0015] Figure 2 This is a schematic diagram of the CNN-LSTM hybrid network architecture described in the present invention. DETAILED DESCRIPTION

[0016] The present invention is further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, modifications of various equivalent forms of the present invention made by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0017] To address the aforementioned challenges of existing technologies, this paper proposes a quantitative risk assessment method for rainstorm disasters affecting major crops based on a CNN-LSTM hybrid neural network. By integrating observational data such as precipitation and high winds, soil moisture observations, and crop growth monitoring data (NDVI index), a trinity input feature matrix ("air-land-agriculture") is constructed. This multi-source data fusion architecture achieves a spatial resolution of up to 1 kilometer, overcoming the limitations of traditional grid-scale assessments. The CNN branch extracts spatial characteristics of farmland microtopography (such as slope and drainage capacity), while the LSTM branch captures the temporal patterns of rainfall processes and soil moisture migration. An attention mechanism dynamically weights key nodes in the disaster chain, significantly enhancing the model's responsiveness to short-term, heavy rainfall. By introducing crop growth period sensitivity coefficients and a disaster loss resilience model, the traditional regression problem is transformed into a dual-output prediction task: "risk probability-loss intensity." The innovative use of a disaster-crop coupling loss function integrates vulnerability theory from agricultural disasters into a deep learning framework, addressing the inadequate representation of crop physiological characteristics in existing models.

[0018] This study reveals the interplay between rainstorm disaster chains and crop systems, promoting cross-innovation in agricultural disaster science and artificial intelligence. The developed "disaster chain dynamics model" simulates the multi-step progression of rainstorms, waterlogging, soil erosion, and crop yield reduction, filling a theoretical gap in existing risk assessment systems. It also provides an interpretable decision-making support tool for accurate agricultural insurance pricing and optimized pre-disaster resource scheduling (such as drainage equipment deployment). By integrating cutting-edge deep learning technologies with agricultural disaster theory, this study not only promotes a paradigm shift in risk assessment from a "static threshold" approach to a "dynamic process," but also provides innovative solutions for global food security and climate-resilient agriculture.

[0019] like Figure 1 and 2 As shown, the quantitative risk assessment method for rainstorm disasters of major crops based on the CNN-LSTM hybrid neural network of the present invention includes the following steps: S1: Preprocessing of multi-source data such as meteorological, soil, crop, and geographic information to provide a data foundation for subsequent model construction, including: The meteorological data are downscaled to a 1km×1km grid, including the actual hourly precipitation, actual daily precipitation, actual cumulative precipitation during the rainstorm process, and actual daily maximum wind speed of each observation station. The eight variables of each grid point, including the 3-hourly forecast precipitation, daily forecast precipitation, forecast cumulative precipitation during the rainstorm process, and forecast daily maximum wind speed, are normalized separately.

[0020] The soil moisture data were processed into grid data with a spatial resolution of 1 km × 1 km, and the daily soil relative humidity (current moisture content / field capacity) of each grid was calculated.

[0021] The NDVI index obtained by the MODIS satellite is temporally interpolated to obtain the daily continuous NDVI index, and the crop growth stage information is converted into categorical variable data (such as jointing stage, heading stage, flowering stage, etc.) to obtain the growth stage coding data.

[0022] The digital elevation model (DEM) is used to calculate the slope (degrees) and microtopography drainage index (TDI) of each grid.

[0023] Slope calculation formula: , Calculation method of microtopography drainage capacity index TDI: , in, The total runoff within the catchment area can be obtained by processing DEM data. The surface water flow path is determined using the ArcGIS hydrological analysis module. The runoff accumulation for each grid is calculated based on the water direction data. The runoff accumulation is converted to runoff per unit area (SCA) using the raster calculation tool: , The slope and micro-topography drainage capacity index of each grid are then processed into 1km×1km grid data.

[0024] The daily soil relative moisture, daily NDVI index, slope and microtopographic drainage index (TDI) of each grid point were standardized.

[0025] Based on the crop planting area and planting distribution, the county-scale historical disaster data were processed into grid data with a spatial resolution of 1km×1km.

[0026] S2: Construct a three-dimensional "air-land-agriculture" feature matrix for each 1 km × 1 km grid point. For each 1 km × 1 km grid point, within the time window T (e.g., 5 days before to 7 days after a rainstorm disaster, a total of 13 days), construct a three-dimensional feature matrix X, X = [T, Lat, Lon, C], where T is the number of time steps, Lat is latitude, Lon is longitude, and C is the number of channels. The channels include actual hourly precipitation, actual daily precipitation, actual cumulative precipitation, forecast hourly precipitation, forecast daily precipitation, forecast cumulative precipitation, actual hourly maximum wind speed, forecast hourly maximum wind speed, soil moisture 10 cm below the surface, soil moisture 20 cm below the surface, soil moisture 40 cm below the surface, NDVI index, crop growth period code, slope, TDI index, crop disaster area ratio, crop disaster area ratio, crop complete failure area ratio, crop yield reduction rate, and agricultural economic losses. Since the data used is 1km×1km grid data, it is a point in space. However, in order to preserve the spatial characteristics, we take a 3×3 neighborhood around the grid, that is, each grid uses a 3×3 neighborhood to represent the local spatial characteristics, Lat=3, Lon=3; time step T=13, feature channel C=20, including: precipitation, wind speed, soil moisture content, NDVI index, growth period code, slope, TDI index, etc.

[0027] S3: Construction of CNN-LSTM hybrid neural network model.

[0028] The CNN-LSTM hybrid neural network model structure consists of an input layer, a CNN layer, an LSTM layer, and an output layer. By combining CNN and LSTM with an attention mechanism, it can fully explore and utilize the intrinsic correlation between meteorological data, crop growth monitoring data, geographic environment data, soil moisture data, and disaster loss time series data, thereby obtaining quantitative loss assessment results.

[0029] S3.1: Use convolutional neural networks (CNNs) to extract spatial features. Specifically, a 2D CNN model is used to obtain the spatial feature vectors of static data such as geographic environmental data (slope and microtopography drainage capacity index), disaster data (affected area, disaster-stricken area, complete crop failure area, yield reduction data, and agricultural economic losses), as well as spatial feature vector sequences of data such as actual hourly precipitation, actual daily precipitation, actual cumulative precipitation, forecast hourly precipitation, forecast daily precipitation, forecast cumulative precipitation, actual hourly maximum wind speed, forecast hourly maximum wind speed, three-layer soil relative humidity, and daily NDVI index at each time step. The 2D CNN model consists of an input layer, a convolutional layer, a pooling layer, an activation function, a fully connected layer, and an output layer. In the convolutional layer, a 3×3 convolution kernel with 64 kernels is used to extract local features. The ReLU (Rectified Linear Unit) activation function is used to improve the problem of small gradients. In the pooling layer, a 2×2 maximum pooling method is used to reduce the data dimension. A fully connected layer is then used to obtain 128-dimensional spatial feature vectors of the slope and TDI index, as well as a 128-dimensional spatial feature vector sequence containing meteorological data, soil moisture data, and crop growth data for each time step. This is a 13×128 vector matrix for the entire time series (13 days).

[0030] S3.2: Extract time series features using a long short-term memory (LSTM) neural network with an attention mechanism. The LSTM model consists of three components: an input gate, a forget gate, and an output gate. It uses a two-LSTM layer structure, with 50 LSTM hidden units in the first layer and 32 hidden units in the second layer. A dropout layer is added to reduce the risk of overfitting between LSTM layers. The attention mechanism is applied to each of the 13 time steps, and the weight of each time step is calculated. The attention mechanism weight calculation method is as follows: , , in, is the input vector at this time step The transpose of is the output vector at the tth time step, is the hyperbolic tangent activation function, is the weight matrix, is the bias term, is the weight function at the t-th time step, is the weight coefficient of the output value of the t-th time step. Finally, the LSTM model including the attention mechanism outputs the weighted time series feature vector : , in, and are the weight coefficient and output at the t-th time step, is a 256-dimensional time series feature vector.

[0031] S3.3: The spatial features extracted by CNN and the temporal features extracted by LSTM are then fused to obtain the final multi-source fusion feature distribution. The feature vector of the last time step extracted by CNN is (128-dimensional) feature vector with LSTM (256 dimensions) are spliced ​​to obtain (384 dimensions), and then the fused feature vector is obtained through weighting by the attention mechanism .

[0032] After the fusion of multi-source features, the model outputs two variables: risk probability and loss intensity. The risk probability is output through a binary classification method (disaster occurred or not). (range 0 to 1), and the loss intensity was obtained by combining the regression model with the crop growth period sensitivity coefficient. (Range 0 to 1). During the loss strength calculation, the fused feature vector is passed through a fully connected layer (without activation) to obtain the base loss rate (range 0 to 1, representing 0% to 100%), and then combine the fused feature vector and the crop growth period code of the current grid to calculate the growth period sensitivity coefficient (The vector preset according to agricultural knowledge, such as 1.2 for tillering stage, 1.5 for jointing stage, 2.0 for heading stage, 1.8 for flowering stage, and 1.0 for maturity stage) is multiplied by the dot product of the fusion feature vector of the growth stage to obtain the sensitivity coefficient ; Then the basic production reduction rate Multiply again , and the loss strength is obtained ,Right now: , , in is the fused feature vector, is the base reduction rate, Determined by the crop growth period.

[0033] S4: Establish a rainstorm disaster-crop coupling loss function. The total loss function consists of two parts, including the loss of risk probability and the loss of loss intensity. The process is as follows:

[0034] S4.1: Calculate the loss of risk probability using the binary cross entropy (BCE) method : , Where, The model predicts the probability of disaster occurrence for the sample. is the risk probability.

[0035] The loss intensity is calculated by considering the root mean square error (MSE) of the sensitivity coefficient of the crop growth period , this invention only calculates the loss intensity loss of samples where disasters actually occurred (i.e. For samples without disasters, the theoretical loss intensity of the samples is 0, so this part of the loss is not calculated (or 0 loss is used).

[0036] , Where, and are the true loss intensity of the sample and the predicted loss intensity of the sample, respectively.

[0037] S4.2: Calculation of total losses from rainstorm disaster-crop coupling, , Where, It is a hyperparameter used to balance the two factors. The general empirical value is 0.7.

[0038] S5: According to the optimized model, the risk probability of crop rainstorm disasters, the loss intensity of major crops and the total loss of rainstorm disasters-crops coupled are generated based on the intelligent grid precipitation forecast data, and a distribution map of risk probability, loss intensity and rainstorm disasters-crops coupled total loss is drawn with a spatial resolution of 1km×1km.

[0039] The present invention can also be applied to scenarios including forming the total loss amount at the county level, and generating an insurance loss index based on the county-level predicted loss as a reference. The specific method is as follows:

[0040] Combined with the crop planting area distribution data, the total loss at the county level can be calculated. : in, is the crop area of ​​the unit grid, The average yield per unit area in the past five years. To predict the production reduction rate. In addition, the insurance loss index can be generated based on the county-level predicted losses. For reference.

[0041] in, It is the historical average output value of a crop in the county.

[0042] The present application can effectively improve the solution ability of the pain points such as low precision, poor timeliness and weak mechanism in agricultural disaster assessment through three innovations of multi-source data fusion, mechanism model embedding and double output architecture.

[0043] The above is the ideal embodiment according to the present application, and the related personnel can make various changes and modifications without deviating from the technical idea of the present application. The technical scope of the present application is not limited to the content of the specification, and the technical scope must be determined according to the scope of claims.

Claims

1. A quantitative risk assessment method for crop rainstorm disasters based on CNN-LSTM hybrid neural network, characterized by The following steps are involved: S1, data preparation and preprocessing, acquiring multi-source data and standardizing the multi-source data to obtain multi-source grid data, the multi-source data including meteorological data, soil moisture data, crop growth monitoring data, geographical environment data and historical disaster data; the meteorological data includes actual and forecast precipitation, actual and forecast wind speed; S2. Performing spatiotemporal superposition on the multi-source grid data to construct a three-dimensional feature matrix X, where X=[T, Lat, Lon, C], where T is the number of time steps, Lat is the latitude, Lon is the longitude, and C is the number of channels; the channels include actual hourly precipitation, actual daily precipitation, actual cumulative precipitation, forecast hourly precipitation, forecast daily precipitation, forecast cumulative precipitation, actual hourly maximum wind speed, forecast hourly maximum wind speed, soil moisture content 10 cm below the surface, soil moisture content 20 cm below the surface, soil moisture content 40 cm below the surface, NDVI index, crop growth period code, slope, TDI index, crop disaster area ratio, crop disaster area ratio, crop complete failure area ratio, crop yield reduction rate, and agricultural economic losses; S3: Build a CNN-LSTM hybrid neural network model and extract spatial feature vectors of meteorological data, geographical environment data, crop growth data, and disaster data based on the CNN neural network model. Extracting time series feature vectors of precipitation and wind speed data, soil moisture, crop growth data, and disaster data based on an LSTM model with an attention mechanism ; The multi-source features of spatial features and temporal features extracted by CNN neural network model and LSTM model are fused to complete the model construction and obtain the fused feature vector , output risk probability and loss intensity data; S4, comprehensively consider the crop losses caused by risk probability and loss intensity, and construct the loss function of rainstorm disaster-crop coupling ; S5, based on the model of S3 and the loss function of S4, generates the risk probability of crop rainstorm disasters, the loss intensity of major crops and the total loss of rainstorm disaster-crop coupling during future rainstorms based on the precipitation and wind speed forecast data of the smart grid.

2. The quantitative risk assessment method for crop rainstorm disasters based on the CNN-LSTM hybrid neural network according to claim 1 is characterized in that: In step S1, the data preparation and preprocessing are performed by first performing the following processing on the multi-source data: (1) Normalization of meteorological data: First, the meteorological data is downscaled to obtain 1 km × 1 km meteorological grid data, and the meteorological grid data is normalized separately; the meteorological data includes actual hourly precipitation, actual daily precipitation, actual process cumulative precipitation, actual hourly maximum wind speed, forecast hourly precipitation, forecast daily precipitation, forecast process cumulative precipitation, and forecast hourly maximum wind speed; (2) The three-layer soil moisture data were downscaled to obtain three layers of soil relative humidity with a spatial resolution of 1 km × 1 km. The soil relative humidity is the ratio of the current water content to the field water holding capacity. The three layers of soil relative humidity include the soil relative humidity at 10 cm below the surface, 20 cm below the surface, and 40 cm below the surface. (3) The NDVI index obtained by satellite is processed into the daily NDVI index, and the crop growth period information is converted into categorical variable data according to the crop growth period to form the growth period stage coding data; (4) The slope is obtained by the following formula: and microtopography drainage capacity index TDI, ， Where, represents the standard deviation of elevation, Indicates the horizontal distance, slope indicates the slope degree obtained according to the slope analysis in Gis. is the total runoff within the catchment area, It represents the flow rate per unit area; Then, the three layers of soil relative moisture, daily NDVI index, slope tan (slope), and microtopography drainage capacity index TDI of each grid were standardized respectively; Finally, based on the crop planting area and planting distribution, the county-scale disaster data were processed into standardized grid data with a spatial resolution of 1km*1km.

3. The quantitative risk assessment method for crop rainstorm disasters based on the CNN-LSTM hybrid neural network according to claim 1 is characterized in that: In the three-dimensional feature matrix X in step S2, Lat=3, Lon=3; the time step T=13, including 5 days before the rainstorm disaster to 7 days after the rainstorm disaster; the number of channels C=20.

4. The quantitative risk assessment method for crop rainstorm disasters based on the CNN-LSTM hybrid neural network according to claim 1 is characterized in that: Step S3 constructs a CNN-LSTM hybrid neural network model, which specifically includes the following steps: The CNN neural network model described in S3.1 obtains the spatial feature vectors of meteorological data, geographical environment data, crop growth data and disaster data, as well as the spatial feature vector sequences of the actual hourly precipitation, actual daily precipitation, actual process cumulative precipitation, actual hourly maximum wind speed, three-layer soil relative humidity, and daily NDVI index at each time step, and obtains a 128-dimensional spatial feature vector ; The LSTM model with attention mechanism described in S3.2 includes 13 time steps, and the attention mechanism is applied separately. The weight of each attention mechanism is calculated by the following formula to complete the construction of the LSTM model with attention mechanism. ， Where, is the weight coefficient of the output value of the t-th time step, represents the weight function at the t-th time step, represents the weight function at the i-th time step, T represents the number of time steps, and its value is 13; is the hyperbolic tangent activation function, is the weight matrix, is the bias term, is the input vector at this time step The transpose of Represents the output vector at the tth time step; Then the time series feature vector is extracted through the LSTM model including the attention mechanism , ， Where, and are the weight coefficient and output vector at the t-th time step, It is a 256-dimensional time series feature vector; S3.3 extracts the 128-dimensional spatial feature vectors from steps S3.1 and S3.2 and 256-dimensional time series feature vector The fusion is then weighted by the attention mechanism to obtain the final multi-source fusion feature vector , S3.4 The multi-source fusion feature vector obtained from steps S3.1-S3.3 , the model finally outputs the risk probability and loss intensity Two variables, among which: risk probability is obtained by binary classification method, with a value range of 0 to 1; loss intensity From the following formula, ， Where, is the basic reduction rate, ranging from 0 to 1; is the growth period sensitivity coefficient of the crop at the current growth stage, It is the final multi-source fusion feature vector.

5. The quantitative risk assessment method for crop rainstorm disasters based on the CNN-LSTM hybrid neural network according to claim 1 is characterized in that: Step S4 includes the following steps: First, based on the risk probability, the binary cross entropy BCE method is used to calculate the loss L of the risk probability prob， ， Where, L prob represents the risk probability of loss, The model predicts the probability of disaster occurrence for the sample. is the risk probability; Then, the loss intensity loss is calculated by the root mean square error (MSE) of the crop growth period sensitivity coefficient. , calculated by the following formula, ， Where, and are the true loss intensity of the sample and the predicted loss intensity of the sample, respectively, and N represents the number of samples; Finally, the total loss of rainstorm disaster-crop coupling is established , ， Where, It is a hyperparameter used to balance the two factor terms and its value is 0.

7.

6. The quantitative risk assessment method for crop rainstorm disasters based on the CNN-LSTM hybrid neural network according to claim 1 is characterized in that: In step S5, the risk probability of rainstorm disasters for crops, the loss intensity of major crops, and the total loss of rainstorm disasters-crops coupled are generated by using the precipitation and wind speed forecast data of the smart grid, and a distribution map of the risk probability, loss intensity, and total loss of rainstorm disasters-crops coupled is formed with a spatial resolution of 1km×1km.

Images