Method and system for joint early warning of agricultural disasters based on meteorological data elements

Abstract

This invention discloses a joint early warning method and system for agricultural disasters based on meteorological data elements. The method includes constructing a multi-source meteorological data acquisition system to collect meteorological data from the target agricultural area; preprocessing the collected multi-source meteorological data; constructing a targeted agricultural disaster early warning indicator system based on the main types of agricultural disasters in the target area and the growth characteristics and maturity stages of different crops; constructing an agricultural disaster early warning model integrating machine learning algorithms and training and optimizing the model; classifying agricultural disasters into early warning levels according to preset early warning level thresholds and combining the types and maturity stages of crops in the target area, and generating early warning information; and disseminating the generated early warning information through preset channels. This invention integrates multi-source meteorological data to construct a multi-dimensional early warning indicator system and a targeted early warning model, achieving accurate and real-time early warning of agricultural disasters.

Description

Technical Field

[0001] This invention relates to the field of agricultural disaster early warning technology, specifically to a joint early warning method and system for agricultural disasters based on meteorological data elements. Background Technology

[0002] Agriculture is a fundamental industry of the national economy, but its production process is highly susceptible to weather conditions. Frequent meteorological disasters such as droughts, floods, frosts, hail, and heatstroke can not only lead to reduced crop yields or even crop failure, but also damage agricultural production facilities, causing huge economic losses and seriously threatening national food security and sustainable agricultural development. Existing agricultural disaster early warning technologies largely rely on traditional meteorological observation data and empirical models, which have several shortcomings: First, the data sources are limited, mostly relying on observation data from ground meteorological stations, resulting in limited coverage and difficulty in accurately reflecting spatial differences in meteorological conditions within a region. For areas with complex terrain and sparse meteorological stations, the early warning accuracy is even lower. Second, the early warning models lack specificity. Traditional models are mostly general-purpose models that do not fully consider the growth characteristics of different crops, the disaster sensitivity at different growth stages, and the agricultural production characteristics of different regions, leading to discrepancies between early warning results and actual agricultural disaster occurrences. Third, the data processing timeliness is poor. Meteorological data is characterized by its real-time nature and large volume, making it difficult for existing technologies to quickly clean, integrate, and analyze massive amounts of meteorological data. This results in delayed early warning information dissemination, failing to provide agricultural producers with sufficient response time. Fourth, the early warning indicator system is incomplete, failing to fully consider the synergistic effects of meteorological factors and relying solely on a single meteorological indicator for early warning, which easily leads to false alarms or missed warnings. Therefore, how to integrate multi-source meteorological data, construct accurate, efficient, and targeted agricultural disaster early warning models, and achieve early and accurate warnings of agricultural disasters has become an urgent technical problem to be solved in the field of agricultural disaster early warning technology. Summary of the Invention

[0003] The purpose of this invention is to provide a joint early warning method and system for agricultural disasters based on meteorological data elements, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The joint early warning method for agricultural disasters based on meteorological data elements includes the following steps: S1. Multi-source meteorological data acquisition: Construct a multi-source meteorological data acquisition system to collect meteorological data of the target agricultural area. The meteorological data includes ground meteorological observation data, satellite remote sensing meteorological data, radar meteorological data, numerical weather prediction data, agricultural microclimate station data, regional automatic station data, and AI real-scene monitoring data. S2. Data preprocessing: Preprocess the multi-source meteorological data collected in step S1, including data cleaning, data standardization and data fusion. S3. Construction of Agricultural Disaster Types and Early Warning Indicator System: Based on existing ground meteorological observation data and satellite inversion parameters, combined with digital elevation models, physical statistical models and machine learning methods are used to reconstruct the fine distribution field of microclimate elements such as temperature, humidity, light and wind speed in the three-dimensional space of agricultural areas. Based on the main types of agricultural disasters in the target area, combined with the growth characteristics and growth period of different crops, a targeted agricultural disaster early warning indicator system is constructed. S4. Early Warning Model Construction and Training: Based on the preprocessed meteorological dataset of the target area and the constructed early warning indicator system, an agricultural disaster early warning model integrating machine learning algorithms is constructed. An evaluation model of the impact of meteorological data on crop quality is established, and the model is trained and optimized. The formula for the evaluation model of the impact of meteorological data on crop quality is as follows: ; S5. Agricultural Disaster Early Warning and Classification: Input the pre-processed multi-source meteorological data into the optimized early warning model to obtain the probability of agricultural disasters; based on the preset early warning level threshold, combined with the crop type and growth period of the target area, classify the early warning level of agricultural disasters and generate early warning information. S6. Warning Information Release and Update: The generated warning information will be released through preset channels. At the same time, the latest meteorological data will be collected in real time to update the warning model and dynamically adjust the warning level and warning information.

[0005] S7. Disaster Management and Protection Effect Assessment: When drought occurs in relevant agricultural areas, control the operation of sprinkler irrigation systems in the relevant areas. By comparing humidity data before and after the operation, automatically generate a protection effect assessment report. It can also record data from the entire process of early warning-response-feedback for model adaptive optimization.

[0006] Furthermore, in step S1, the ground meteorological observation data includes temperature, precipitation, relative humidity, wind speed, wind direction, sunshine duration, and air pressure; satellite remote sensing meteorological data includes cloud top temperature, vegetation index, surface temperature, and spatial distribution data of precipitation intensity; radar meteorological data includes echo intensity, echo top height, and vertically integrated liquid water content; numerical weather prediction data includes forecast data for temperature, precipitation, wind speed, and wind direction for the next 1-7 days; agricultural microclimate station data is meteorological data within a small area of ​​the agricultural region; regional automatic station data is data from various electrical devices and sensors used in automated agricultural production applications within the agricultural region; and AI real-scene monitoring data is image data obtained from monitoring the agricultural region.

[0007] Furthermore, in step S2, data cleaning employs an outlier detection algorithm to remove outliers and a missing value imputation algorithm to fill in missing data. The outlier detection algorithm combines box plot and Z-score methods. The missing value imputation algorithm selects the appropriate imputation method based on the missing data rate: when the missing rate is ≤5%, nearest neighbor imputation is used; when 5% < missing rate ≤20%, random forest-based missing value imputation is used; and when the missing rate >20%, the corresponding data samples are removed and supplemented with collected data.

[0008] Furthermore, in step S2, data standardization uses the min-max standardization method to transform meteorological data of different dimensions and units into the [0,1] interval; data fusion is based on a weighted average fusion algorithm, and the weights are determined according to the accuracy and reliability of different data sources.

[0009] Furthermore, in step S3, the satellite inversion parameters include crop leaf area index (LAI) and land surface temperature; the machine learning method is one of random forest or neural network; the agricultural disaster types include drought, flood, frost, heat damage, and hail; and corresponding early warning indicators are constructed for different disaster types. Drought disaster early warning indicators include: number of consecutive days without effective precipitation, soil relative humidity, vegetation water supply index, and standardized precipitation evapotranspiration index; Flood disaster early warning indicators include: 24-hour cumulative precipitation, 72-hour cumulative precipitation, precipitation intensity, river water level, and soil saturation moisture content; Frost disaster warning indicators include: minimum temperature, daily temperature range, minimum surface temperature, and duration of frost. High-temperature heat hazard warning indicators include: daily maximum temperature, number of consecutive days of high temperature, intensity of high temperature, and relative humidity. Hail disaster warning indicators include radar echo intensity, echo top height, vertical integral liquid water content, and probability of hail.

[0010] Furthermore, in step S3, a predictive model that integrates high-precision weather forecasts, real-time monitoring information, and quality can be established, and an AI decision support model that generates the optimal crop collection index and drying environment index can be generated, providing a scientific basis for refined and standardized production.

[0011] Furthermore, in step S4, the early warning model adopts a fusion algorithm combining gradient boosting decision tree algorithm and long short-term memory network; the model training adopts an adaptive moment estimation optimizer to optimize model parameters, with cross-entropy loss function as loss function; and a grid search algorithm is used to optimize model hyperparameters.

[0012] Furthermore, in step S5, the warning levels are divided into four levels: Level IV (General Warning), Level III (Severe Warning), Level II (Severe Warning), and Level I (Extremely Severe Warning). The specific judgment rules are as follows: when the probability of disaster occurrence is ∈ [0.2, 0.4), it is judged as a Level IV warning; when the probability of disaster occurrence is ∈ [0.4, 0.6), it is judged as a Level III warning; when the probability of disaster occurrence is ∈ [0.6, 0.8), it is judged as a Level II warning; and when the probability of disaster occurrence is ∈ [0.8, 1.0], it is judged as a Level I warning.

[0013] Furthermore, in step S6, the publishing channels include mobile terminal APP, SMS, WeChat official account, agricultural information platform, and rural broadcasting; it also includes a customized early warning information function, which pushes personalized early warning information and response measures suggestions based on the user's crop type, planting area, and geographical location.

[0014] To achieve the above objectives, the present invention also provides the following technical solution: A joint early warning system for agricultural disasters based on meteorological data elements includes: The data acquisition module is used to construct a multi-source meteorological data acquisition system to collect meteorological data of the target agricultural area. The meteorological data includes ground meteorological observation data, satellite remote sensing meteorological data, radar meteorological data, and numerical weather prediction data. The preprocessing module is used to preprocess the collected multi-source meteorological data, including data cleaning, data standardization, and data fusion. The module is used to construct a targeted agricultural disaster early warning indicator system based on the main types of agricultural disasters in the target area and the growth characteristics and growth period of different crops. The construction and training module is used to build an agricultural disaster early warning model that integrates machine learning algorithms based on the preprocessed meteorological dataset of the target area and the constructed early warning indicator system, and to train and optimize the model. The classification module is used to input real-time preprocessed multi-source meteorological data into the optimized early warning model to obtain the probability of agricultural disasters; based on the preset early warning level threshold, combined with the crop type and growth period of the target area, the early warning level of agricultural disasters is classified and early warning information is generated. The publishing and updating module is used to publish the generated early warning information through preset publishing channels, while collecting the latest meteorological data in real time, updating the early warning model in real time, and dynamically adjusting the early warning level and early warning information.

[0015] The control module is used to acquire drought data of relevant agricultural areas divided by the segmentation module. When the drought in a relevant agricultural area exceeds the threshold, it controls the sprinkler system in that area to work. The sprinkler system opens the solenoid valve group corresponding to the drought risk area, while the valves in other areas remain closed, so as to achieve precise delivery of energy and water.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. Comprehensive data sources and wide coverage: This invention integrates meteorological data from multiple sources, including ground meteorological observation, satellite remote sensing, radar, and numerical weather prediction, and can also expand to agricultural Internet of Things data. This effectively makes up for the shortcomings of limited coverage and insufficient accuracy of a single data source, and can comprehensively and accurately reflect the meteorological conditions of the target area. 2. Highly efficient and reliable data processing: By employing a data cleaning method that combines outlier detection and missing value imputation, a min-max standardization method, and a weighted average fusion algorithm, efficient preprocessing of multi-source meteorological data was achieved, eliminating data noise and dimensional differences, improving data quality and fusion accuracy, and providing reliable data support for the subsequent construction of early warning models; 3. The early warning model has high accuracy and strong generalization ability: The present invention uses a fusion algorithm combining GBDT and LSTM to construct an early warning model, which can effectively extract nonlinear features in meteorological data and capture the time series dependence of meteorological data. At the same time, the model can be further optimized through ensemble learning, which improves the generalization ability and early warning accuracy of the model and reduces the false alarm and missed alarm rates. 4. Highly targeted and practical early warning system: This invention constructs a targeted early warning indicator system for different types of agricultural disasters, different crop growth characteristics and growth stages. At the same time, it can realize personalized early warning information push, which can provide accurate early warning information and response suggestions for different users, thus improving the practicality of the early warning method. 5. Strong real-time early warning and rapid response: This invention can collect and process meteorological data in real time, dynamically update the early warning model, realize the real-time release and dynamic adjustment of early warning information, provide agricultural producers and operators with sufficient response time, and effectively reduce disaster losses. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the joint early warning method for agricultural disasters based on meteorological data elements according to the present invention.

[0018] Figure 2 This is a schematic diagram of the module structure of the agricultural disaster joint early warning device based on meteorological data elements according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figures 1 to 2 The present invention provides a technical solution: To address the shortcomings of existing agricultural disaster early warning technologies, such as single data sources, insufficient targeting of early warning models, poor timeliness of data processing, and incomplete early warning indicator systems, this invention provides a joint early warning method for agricultural disasters based on meteorological data elements. By integrating multi-source meteorological data, a multi-dimensional early warning indicator system and a targeted early warning model are constructed to achieve accurate and real-time early warning of agricultural disasters.

[0021] Includes the following steps: S1. Multi-source meteorological data acquisition: Construct a multi-source meteorological data acquisition system to collect meteorological data from the target agricultural area. This meteorological data includes surface meteorological observation data, satellite remote sensing meteorological data, radar meteorological data, numerical weather prediction data, agricultural microclimate station data, regional automatic weather station data, and AI-powered real-scene monitoring data. Specifically, surface meteorological observation data includes temperature, precipitation, relative humidity, wind speed, wind direction, sunshine duration, and air pressure; satellite remote sensing meteorological data includes cloud top temperature, vegetation index, surface temperature, and spatial distribution data of precipitation intensity; radar meteorological data includes echo intensity, echo top height, and vertically integrated liquid water content; numerical weather prediction data includes forecasts of temperature, precipitation, wind speed, and wind direction for the next 1-7 days; agricultural microclimate station data is meteorological data within a small area of ​​the agricultural region; regional automatic weather station data includes data from various electrical devices and sensors used in automated agricultural production within the agricultural region; and AI-powered real-scene monitoring data is image data obtained from monitoring the agricultural region.

[0022] S2. Data preprocessing: Preprocess the multi-source meteorological data collected in step S1, including data cleaning, data standardization and data fusion. The data cleaning process involves: outlier detection algorithms to remove outliers from meteorological data, and missing value imputation algorithms to fill in missing data. The outlier detection algorithm combines box plotting and Z-score methods. First, box plotting is used to initially select a candidate set of outliers, then Z-score is used to further validate the data in the candidate set to determine the final outliers. The missing value imputation algorithm selects the appropriate imputation method based on the missing data rate. When the missing rate is ≤5%, nearest neighbor imputation is used; when 5% < missing rate ≤20%, random forest-based missing value imputation is used; when the missing rate >20%, the corresponding data sample is removed and supplemented with meteorological data of the same time period and type. Data standardization: The min-max standardization method was used to transform meteorological data of different dimensions and units to the [0,1] interval, eliminating the impact of dimensional differences on subsequent analysis; Data fusion: The preprocessed multi-source meteorological data are fused based on a weighted average fusion algorithm. The weights are determined according to the accuracy and reliability of different data sources. The higher the accuracy and reliability, the greater the weight, resulting in a unified meteorological dataset for the target area. S3. Construction of Agricultural Disaster Types and Early Warning Indicator System: Based on existing ground meteorological observation data and satellite inversion parameters, combined with digital elevation models, physical statistical models and machine learning methods are used to reconstruct the fine distribution field of microclimate elements such as temperature, humidity, light and wind speed in the three-dimensional space of agricultural areas. Based on the main types of agricultural disasters in the target area, combined with the growth characteristics and growth period of different crops, a targeted agricultural disaster early warning indicator system is constructed. Among them, the satellite inversion parameters include crop leaf area index (LAI) and land surface temperature; the machine learning method is one of random forest and neural network; the agricultural disaster types include drought disaster, flood disaster, frost disaster, high temperature heat damage, and hail disaster; Develop corresponding early warning indicators for different types of disasters: Drought disaster early warning indicators include: number of consecutive days without effective precipitation, soil relative humidity, vegetation water supply index, and standardized precipitation evapotranspiration index; Flood disaster early warning indicators include: 24-hour cumulative precipitation, 72-hour cumulative precipitation, precipitation intensity, river water level, and soil saturation moisture content; Frost disaster warning indicators include: minimum temperature, daily temperature range, minimum surface temperature, and duration of frost. High-temperature heat hazard warning indicators include: daily maximum temperature, number of consecutive days of high temperature, intensity of high temperature, and relative humidity. Hail disaster warning indicators include radar echo intensity, echo top height, vertical integral liquid water content, and probability of hail.

[0023] This step also enables the establishment of a predictive model that integrates high-precision weather forecasts, real-time monitoring information, and quality data, and generates an AI decision support model that provides the optimal crop collection index and drying environment index, thus providing a scientific basis for refined and standardized production.

[0024] S4. Early Warning Model Construction and Training: Based on the preprocessed meteorological dataset of the target area and the constructed early warning indicator system, an agricultural disaster early warning model integrating machine learning algorithms is constructed. An evaluation model of the impact of meteorological data on crop quality is established, and the model is trained and optimized. The formula for the evaluation model of the impact of meteorological data on crop quality is as follows: In the formula, Tcqi is the crop climate quality index, a1 and b1 are the combined weights of climate suitability indicators and meteorological disaster indicators affecting crop quality (initially set a1=0.8 and b1=0.2), xi is the weight of each climate suitability evaluation indicator affecting crop quality, Mi is the evaluation level of the climate suitability evaluation indicator; yj is the weight of each meteorological disaster indicator affecting quality, Nj is the evaluation level of the meteorological disaster evaluation indicator; n and m are the sample sizes of the two summation terms.

[0025] The specific steps are as follows: S41. Dataset partitioning: Divide the meteorological dataset of the target area into a training set and a test set in a 7:3 ratio. The training set is used for model training, and the test set is used for model performance verification. S42. Early Warning Model Construction: An early warning model is constructed using a fusion algorithm combining the Gradient Boosting Decision Tree (GBDT) algorithm and the Long Short-Term Memory (LSTM) network. The GBDT algorithm is used to extract nonlinear and high-dimensional features from meteorological data, while the LSTM network is used to capture the time-series dependencies of meteorological data. The features extracted by GBDT are input into the LSTM network for time-series modeling, and the probability of agricultural disasters is output. S43. Model Training: The fusion early warning model is trained using the training set. The model parameters are optimized using the adaptive moment estimation (Adam) optimizer. The cross-entropy loss function is used as the loss function. The loss function is minimized through iterative training to obtain the initial early warning model. S44. Model Optimization: The hyperparameters of the initial early warning model are optimized using a grid search algorithm to determine the optimal combination of hyperparameters and obtain the optimized agricultural disaster early warning model. The hyperparameters include the number of decision trees, learning rate, and maximum tree depth of the GBDT, as well as the number of hidden layer neurons, number of iterations, and batch size of the LSTM network. S5. Agricultural Disaster Early Warning and Classification: Input the pre-processed multi-source meteorological data into the optimized early warning model to obtain the probability of agricultural disasters; based on the preset early warning level threshold, combined with the crop type and growth period of the target area, classify the early warning level of agricultural disasters and generate early warning information. The warning levels are divided into four levels: Level IV (general warning), Level III (relatively severe warning), Level II (serious warning), and Level I (extremely severe warning). The warning level thresholds for different disaster types are determined based on historical disaster data and actual agricultural production losses. Specific rules for determining the warning level: When the probability of disaster occurrence is ∈ [0.2, 0.4), it is determined as a Level IV warning; when the probability of disaster occurrence is ∈ [0.4, 0.6), it is determined as a Level III warning; when the probability of disaster occurrence is ∈ [0.6, 0.8), it is determined as a Level II warning; when the probability of disaster occurrence is ∈ [0.8, 1.0], it is determined as a Level I warning. S6. Warning Information Release and Update: The generated warning information will be released through preset channels, while the latest meteorological data will be collected in real time to update the warning model and dynamically adjust the warning level and warning information. The release channels include mobile terminal APP, SMS, WeChat official account, agricultural information platform, and rural broadcast.

[0026] S7. Disaster Management and Protection Effect Assessment: When drought occurs in relevant agricultural areas, control the operation of sprinkler irrigation systems in the relevant areas. By comparing humidity data before and after the operation, automatically generate a protection effect assessment report. It can also record data from the entire process of early warning-response-feedback for model adaptive optimization.

[0027] To further improve the accuracy and practicality of early warning, the present invention may also adopt the following optimization schemes: 1. In step S1, agricultural IoT data collection is added. The agricultural IoT data includes soil temperature and humidity, crop canopy temperature, and soil nutrient content data. These data are then analyzed in conjunction with multi-source meteorological data to improve the targeting of the early warning model. 2. In step S3, the Analytic Hierarchy Process (AHP) is introduced to assign weights to different early warning indicators. The weights are determined based on the degree of influence of the indicators on the occurrence of disasters, thereby improving the scientific nature of the early warning indicator system. 3. In step S4, an ensemble learning algorithm is used to integrate GBDT, LSTM, and support vector machine (SVM) algorithms. The final probability of disaster occurrence is determined through a voting mechanism, which further improves the generalization ability and early warning accuracy of the model. 4. In step S6, a customized early warning information function is added, which pushes personalized early warning information and response suggestions based on the user's crop type, planting area, geographical location and other information.

[0028] The present invention will be further described in detail below with reference to specific embodiments. Example 1: Taking drought disaster early warning in a rice-growing area as an example, the agricultural disaster early warning method based on meteorological data of the present invention will be described in detail. Step S1: Multi-source meteorological data acquisition: Determine the target agricultural area as a major rice-producing region and collect multi-source meteorological data for this area, including: Surface meteorological observation data: Daily average temperature, daily precipitation, relative humidity, wind speed, and sunshine duration data for the past 5 years were collected from 10 surface meteorological stations in the region, with a sampling frequency of 1 time per hour; Satellite remote sensing meteorological data: Spatial distribution data of vegetation index (NDVI), land surface temperature (LST), and precipitation intensity in this region were acquired through MODIS satellites, with a spatial resolution of 1 km × 1 km and a temporal resolution of 1 day; Radar meteorological data: Echo intensity and echo top height data are collected by two Doppler weather radars in the region, with a sampling frequency of 6 minutes / time; Numerical weather forecast data: The daily average temperature and daily precipitation forecast data for the region for the next 7 days are obtained from the National Meteorological Center. Step S2, Data Preprocessing: Data cleaning: Box plot method was used to initially screen out the candidate set of outliers in precipitation and temperature data, and then the Z-score method (with a threshold of ±3) was used to perform secondary verification on the data in the candidate set to remove the final outliers; for missing values ​​in soil relative humidity data, since the missing value rate was 8%, a missing value imputation method based on random forest was used for imputation. Data standardization: The min-max standardization method was used to convert data of different dimensions such as temperature (unit: °C), precipitation (unit: mm), and relative humidity (unit: %) to the [0,1] interval; Data fusion: The preprocessed multi-source data are fused based on a weighted average fusion algorithm, with the weight of ground meteorological observation data set to 0.4, satellite remote sensing data set to 0.3, numerical weather prediction data set to 0.2, and radar data set to 0.1, to obtain a unified meteorological dataset for the target area. Step S3: Construction of Agricultural Disaster Types and Early Warning Indicator System: The main agricultural disaster in the target area is rice drought. Combining the growth characteristics of rice during the booting stage, an early warning indicator system for drought disaster is constructed, including: number of consecutive days without effective precipitation, soil relative humidity, vegetation water supply index, and standardized precipitation evapotranspiration index. The weights of each indicator are determined to be 0.3, 0.3, 0.2, and 0.2, respectively, using the analytic hierarchy process (AHP).

[0029] Step S4, Early Warning Model Construction and Training: S41. Dataset partitioning: The meteorological dataset of the target area is divided into a training set and a test set in a 7:3 ratio. The training set contains data from the past 3.5 years, and the test set contains data from the past 1.5 years. S42. Early warning model construction: The early warning model is constructed using a fusion algorithm combining GBDT and LSTM. The GBDT algorithm is set with 100 decision trees, a learning rate of 0.1, and a maximum tree depth of 6. The LSTM network is set with 64 hidden layer neurons, 100 iterations, and a batch size of 32. S43. Model Training: The model is trained using the training set. The Adam optimizer is used to optimize the model parameters. The cross-entropy loss function is used as the loss function. The model is iteratively trained until the loss function converges to obtain the initial warning model. S44. Model Optimization: The grid search algorithm was used to optimize the hyperparameters of the initial model. The optimal combination of hyperparameters was determined to be: GBDT decision tree number 120, learning rate 0.08, maximum tree depth 5; LSTM hidden layer neurons number 80, iteration count 120, batch size 64, to obtain the optimized drought disaster early warning model. Step S5, Agricultural Disaster Early Warning and Classification: Input the real-time collected and preprocessed meteorological data into the optimized early warning model to obtain the probability of rice drought disaster occurrence; based on the historical drought disaster data and rice yield reduction in the region, set the early warning level thresholds: Level IV early warning (general early warning) probability threshold is [0.2, 0.4), Level III early warning (relatively severe early warning) is [0.4, 0.6), Level II early warning (severe early warning) is [0.6, 0.8), and Level I early warning (extremely severe early warning) is [0.8, 1.0]; if the model outputs a drought disaster occurrence probability of 0.65, it is determined to be a Level II early warning. Step S6, Issuance and Update of Early Warning Information: The Level II drought early warning information will be issued through three channels: mobile terminal APP, SMS, and rural broadcast. At the same time, suggestions for drought response measures for rice (such as timely irrigation and spraying of drought-resistant agents) will be pushed. The latest meteorological data will be collected every 6 hours to update the early warning model and dynamically adjust the early warning level. Please see Figure 2 The present invention further provides a joint early warning system for agricultural disasters based on meteorological data elements. This system is used to implement the aforementioned joint early warning method for agricultural disasters based on meteorological data elements, specifically including: The data acquisition module is used to construct a multi-source meteorological data acquisition system to collect meteorological data of the target agricultural area. The meteorological data includes ground meteorological observation data, satellite remote sensing meteorological data, radar meteorological data, and numerical weather prediction data. The preprocessing module is used to preprocess the collected multi-source meteorological data, including data cleaning, data standardization, and data fusion. The module is used to construct a targeted agricultural disaster early warning indicator system based on the main types of agricultural disasters in the target area and the growth characteristics and growth period of different crops. The construction and training module is used to build an agricultural disaster early warning model that integrates machine learning algorithms based on the preprocessed meteorological dataset of the target area and the constructed early warning indicator system, and to train and optimize the model. The classification module is used to input real-time preprocessed multi-source meteorological data into the optimized early warning model to obtain the probability of agricultural disasters; based on the preset early warning level threshold, combined with the crop type and growth period of the target area, the early warning level of agricultural disasters is classified and early warning information is generated. The publishing and updating module is used to publish the generated early warning information through preset publishing channels, while collecting the latest meteorological data in real time, updating the early warning model in real time, and dynamically adjusting the early warning level and early warning information.

[0030] The control module is used to acquire drought data of relevant agricultural areas divided by the segmentation module. When the drought in a relevant agricultural area exceeds the threshold, it controls the sprinkler system in that area to work. The sprinkler system opens the solenoid valve group corresponding to the drought risk area, while the valves in other areas remain closed, so as to achieve precise delivery of energy and water.

[0031] Tests showed that the early warning model in this embodiment achieved an accuracy rate of 92%, a false alarm rate of 5%, and a missed alarm rate of 3%. Compared with traditional early warning methods, the accuracy rate was improved by more than 15%, enabling it to provide accurate and timely drought disaster early warning services for the rice-growing area.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A joint early warning method for agricultural disasters based on meteorological data elements, characterized in that, Includes the following steps: S1. Multi-source meteorological data acquisition: Construct a multi-source meteorological data acquisition system to collect meteorological data of the target agricultural area. The meteorological data includes ground meteorological observation data, satellite remote sensing meteorological data, radar meteorological data, numerical weather prediction data, agricultural microclimate station data, regional automatic station data, and AI real-scene monitoring data. S2. Data preprocessing: Preprocess the multi-source meteorological data collected in step S1, including data cleaning, data standardization and data fusion. S3. Construction of Agricultural Disaster Types and Early Warning Indicator System: Based on existing ground meteorological observation data and satellite inversion parameters, combined with digital elevation models, physical statistical models and machine learning methods are used to reconstruct the fine distribution field of microclimate elements such as temperature, humidity, light and wind speed in the three-dimensional space of agricultural areas. Based on the main types of agricultural disasters in the target area, combined with the growth characteristics and growth period of different crops, a targeted agricultural disaster early warning indicator system is constructed. S4. Early Warning Model Construction and Training: Based on the preprocessed meteorological dataset of the target area and the constructed early warning indicator system, an agricultural disaster early warning model integrating machine learning algorithms is constructed. An evaluation model of the impact of meteorological data on crop quality is established, and the model is trained and optimized. The formula for the evaluation model of the impact of meteorological data on crop quality is as follows: ; S5. Agricultural disaster early warning and classification: Input the pre-processed multi-source meteorological data into the optimized early warning model to obtain the probability of agricultural disasters. Based on the preset warning level thresholds, combined with the crop type and growth stage of the target area, the warning level of agricultural disasters is classified and warning information is generated. S6. Warning Information Issuance and Update: The generated warning information will be issued through preset channels, while the latest meteorological data will be collected in real time to update the warning model and dynamically adjust the warning level and warning information. S7. Disaster Management and Protection Effect Assessment: When drought occurs in relevant agricultural areas, control the operation of sprinkler irrigation systems in the relevant areas. By comparing humidity data before and after the operation, automatically generate a protection effect assessment report. It can also record data from the entire process of early warning-response-feedback for model adaptive optimization.

2. The joint early warning method for agricultural disasters based on meteorological data elements as described in claim 1, characterized in that, In step S1, the ground meteorological observation data includes temperature, precipitation, relative humidity, wind speed, wind direction, sunshine duration, and air pressure; satellite remote sensing meteorological data includes cloud top temperature, vegetation index, surface temperature, and spatial distribution data of precipitation intensity; radar meteorological data includes echo intensity, echo top height, and vertically integrated liquid water content; numerical weather prediction data includes forecast data for temperature, precipitation, wind speed, and wind direction for the next 1-7 days; agricultural microclimate station data is meteorological data within a small area of ​​the agricultural region; regional automatic station data is data from various electrical devices and sensors used in automated agricultural production in the agricultural region; and AI real-scene monitoring data is image data obtained from monitoring the agricultural region.

3. The joint early warning method for agricultural disasters based on meteorological data elements as described in claim 1, characterized in that, In step S2, data cleaning employs an outlier detection algorithm to remove outliers and a missing value imputation algorithm to fill in missing data. The outlier detection algorithm combines box plot and Z-score methods. The missing value imputation algorithm selects the appropriate imputation method based on the missing data rate: when the missing rate is ≤5%, nearest neighbor imputation is used; when 5% < missing rate ≤20%, random forest-based missing value imputation is used; and when the missing rate >20%, the corresponding data samples are removed and supplemented with collected data.

4. The joint early warning method for agricultural disasters based on meteorological data elements as described in claim 1, characterized in that, In step S2, data standardization uses the min-max standardization method to transform meteorological data of different dimensions and units into the [0,1] interval; data fusion is based on the weighted average fusion algorithm, and the weights are determined according to the accuracy and reliability of different data sources.

5. The joint early warning method for agricultural disasters based on meteorological data elements as described in claim 1, characterized in that, In step S3, the satellite inversion parameters include crop leaf area index (LAI) and land surface temperature; the machine learning method is one of random forest or neural network; the agricultural disaster types include drought, flood, frost, heat damage, and hail; and corresponding early warning indicators are constructed for different disaster types. Drought disaster early warning indicators include: number of consecutive days without effective precipitation, soil relative humidity, vegetation water supply index, and standardized precipitation evapotranspiration index; Flood disaster early warning indicators include: 24-hour cumulative precipitation, 72-hour cumulative precipitation, precipitation intensity, river water level, and soil saturation moisture content; Frost disaster warning indicators include: minimum temperature, daily temperature range, minimum surface temperature, and duration of frost. High-temperature heat hazard warning indicators include: daily maximum temperature, number of consecutive days of high temperature, intensity of high temperature, and relative humidity. Hail disaster warning indicators include radar echo intensity, echo top height, vertical integral liquid water content, and probability of hail.

6. The joint early warning method for agricultural disasters based on meteorological data elements as described in claim 1, characterized in that, In step S3, a predictive model that integrates high-precision weather forecasts, real-time monitoring information, and quality can be established, and an AI decision support model that generates the optimal crop collection index and drying environment index can be generated, providing a scientific basis for refined and standardized production.

7. The joint early warning method for agricultural disasters based on meteorological data elements as described in claim 1, characterized in that, In step S4, the early warning model adopts a fusion algorithm combining gradient boosting decision tree algorithm and long short-term memory network; the model training adopts adaptive moment estimation optimizer to optimize model parameters, and cross-entropy loss function is used as loss function; grid search algorithm is used to optimize model hyperparameters.

8. The joint early warning method for agricultural disasters based on meteorological data elements as described in claim 1, characterized in that, In step S5, the warning levels are divided into four levels: Level IV (general warning), Level III (relatively severe warning), Level II (serious warning), and Level I (extremely severe warning); the specific judgment rule is: when the probability of disaster occurrence ∈ [0.2, 0.4), it is judged as a Level IV warning; When the probability of a disaster occurring is ∈ [0.4, 0.6), it is classified as a Level III warning; when the probability of a disaster occurring is ∈ [0.6, 0.8), it is classified as a Level II warning. When the probability of a disaster occurring is ∈ [0.8, 1.0], it is determined to be a Level I warning.

9. The joint early warning method for agricultural disasters based on meteorological data elements as described in claim 1, characterized in that, In step S6, the publishing channels include mobile terminal APP, SMS, WeChat official account, agricultural information platform, and rural broadcasting; it also includes a customized early warning information function, which pushes personalized early warning information and response measures suggestions based on the user's crop type, planting area, and geographical location.

10. A joint early warning system for agricultural disasters based on meteorological data elements, characterized in that: include: The data acquisition module is used to construct a multi-source meteorological data acquisition system to collect meteorological data of the target agricultural area. The meteorological data includes ground meteorological observation data, satellite remote sensing meteorological data, radar meteorological data, and numerical weather prediction data. The preprocessing module is used to preprocess the collected multi-source meteorological data, including data cleaning, data standardization, and data fusion. The module is used to construct a targeted agricultural disaster early warning indicator system based on the main types of agricultural disasters in the target area and the growth characteristics and growth period of different crops. The construction and training module is used to build an agricultural disaster early warning model that integrates machine learning algorithms based on the preprocessed meteorological dataset of the target area and the constructed early warning indicator system, and to train and optimize the model. The partitioning module is used to input the real-time preprocessed multi-source meteorological data into the optimized early warning model to obtain the probability of agricultural disasters. Based on the preset warning level thresholds, combined with the crop type and growth stage of the target area, the warning level of agricultural disasters is classified and warning information is generated. The publishing and updating module is used to publish the generated early warning information through preset publishing channels, while collecting the latest meteorological data in real time, updating the early warning model in real time, and dynamically adjusting the early warning level and early warning information. The control module is used to acquire drought data of relevant agricultural areas divided by the segmentation module. When the drought in a relevant agricultural area exceeds the threshold, it controls the sprinkler system in that area to work. The sprinkler system opens the solenoid valve group corresponding to the drought risk area, while the valves in other areas remain closed, so as to achieve precise delivery of energy and water.

Images