High-resolution low-altitude meteorological risk early warning method

Abstract

The invention discloses a high-resolution low-altitude meteorological risk early warning method, and relates to the technical field of disaster early warning. The method comprises the following components: S1, a data acquisition and preprocessing step, S2, a multi-disaster association model construction and training step, S3, a high-resolution gridding risk calculation step, S4, a risk assessment and grading step, and S5, a linkage early warning issuing and dynamic updating step. By combining historical disaster data and real-time data, a correlation model of low-altitude meteorological risks, geological disasters and flood disasters is constructed, linkage early warning of multiple disasters is achieved, meanwhile, by continuously collecting data in real time, the correlation model of multiple disasters is dynamically updated and optimized, and the early warning efficiency is improved. The risk value and the early warning level of the grid unit are adjusted in time according to the new calculation result, the latest early warning information is issued, the timeliness and accuracy of the early warning information are ensured through the dynamic updating mechanism, disaster changes can be coped with in time, and disaster losses are reduced.

Description

Technical Field

[0001] This invention relates to the field of disaster early warning technology, specifically a high-resolution low-altitude meteorological risk early warning method. Background Technology

[0002] With the acceleration of global climate change and urbanization, extreme weather events and geological disasters occur frequently, posing a serious threat to human society and the natural environment. As an important factor affecting surface meteorology and the occurrence of disasters, the accurate monitoring and early warning of low-altitude meteorological conditions are of great significance for disaster prevention and mitigation.

[0003] Traditional meteorological and geological disaster early warning technologies have several shortcomings: First, traditional methods often rely on single-hazard models for early warning, neglecting the correlation and mutual influence between multiple disasters, resulting in incomplete and inaccurate early warning results. Second, due to low spatial resolution, traditional early warning methods struggle to accurately reflect risk differences across different regions, making early warning information less targeted and actionable. Furthermore, traditional methods lag behind in data processing and model updates, failing to reflect changes in disaster conditions in real time, thus reducing the timeliness and effectiveness of early warnings. Specifically, traditional technologies may rely on fixed data collection points and limited monitoring methods, leading to incomplete data acquisition; simultaneously, model training and optimization processes may lack flexibility, making it difficult to adapt to rapidly changing disaster environments.

[0004] In view of the shortcomings of traditional meteorological and geological disaster early warning technologies, the proposed high-resolution low-altitude meteorological risk early warning method in this invention is particularly important. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a high-resolution low-altitude meteorological risk early warning method. It can significantly improve the accuracy, timeliness and spatial resolution of early warning by integrating multi-source data acquisition and preprocessing technologies, constructing multi-hazard correlation models, implementing high-resolution gridded risk calculation, and implementing a coordinated early warning release and dynamic update strategy.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a high-resolution low-altitude meteorological risk early warning method, the specific steps of which are as follows: S1. Data Acquisition and Preprocessing Steps: Using meteorological satellites, meteorological radars, ground meteorological stations, geological disaster monitoring sensors, and hydrological monitoring equipment, low-altitude meteorological data, geological data, and hydrological data are collected in real time. After collection, the raw data is preprocessed by cleaning, denoising, and format conversion to remove abnormal and duplicate data, unify the format, and use interpolation and regression analysis to complete missing data. S2. Steps for constructing and training a multi-hazard correlation model: Combining historical disaster data and pre-processed real-time data, construct a correlation model between low-altitude meteorological risk and geological disasters, floods, and other multi-hazards. By analyzing the correlation between low-altitude meteorological elements and the occurrence of other disasters, determine the weight of each influencing factor, establish the mathematical relationship between changes in meteorological conditions and the probability of disaster occurrence, and train and optimize the model. S3. High-resolution gridded risk calculation steps: Divide the warning area into multiple high-resolution grid units. Based on the collected data and the correlation model, calculate the low-altitude meteorological risk value, geological disaster risk value and flood disaster risk value in each grid unit. When calculating, comprehensively consider the impact of low-altitude wind speed and precipitation intensity on the disaster-bearing body, as well as the effects of multiple factors such as mountain slope and river flow. S4. Risk Assessment and Classification Steps: Based on the risk value of each grid unit and in conjunction with the preset risk assessment standards, assess the risks of low-altitude meteorological disasters, geological disasters, and flood disasters respectively, and classify them into four levels of early warning: general, relatively severe, severe, and extremely severe. At the same time, take into account the risk situation of multiple disasters and determine the comprehensive risk early warning level of each grid unit. S5. Linked Early Warning Issuance and Dynamic Update Steps: Through multiple channels such as SMS, radio, television, internet platforms, and outdoor displays, timely release early warning information on low-altitude meteorological, geological disaster, and flood disaster risks, as well as comprehensive risk early warning information for each grid unit. Continuously collect data in real time, dynamically update and optimize the multi-hazard correlation model, adjust the grid unit risk value and early warning level in a timely manner based on the new calculation results, and issue the latest early warning.

[0007] Furthermore, the low-altitude meteorological data acquisition in S1 also includes using a low-altitude UAV equipped with a high-precision sensor to conduct low-altitude vertical layered data acquisition in specific complex terrain areas. The acquisition altitude range is 0-500 meters above the ground, with layered acquisition at 10-meter intervals to obtain meteorological data at different altitudes. For geological data acquisition, distributed fiber optic sensing technology is used, with fiber optic cables laid inside the mountain to monitor stress changes within the mountain in real time. Combined with ground-based ground-penetrating radar data on the rock and soil structure, a three-dimensional geological data acquisition system is formed. In addition to conventional river flow and water level monitoring, hydrological data acquisition also utilizes satellite remote sensing technology to obtain dynamic changes in river width. Combined with underwater topographic sonar data, a complete hydrological data acquisition scheme is constructed. During the data acquisition process, a data quality evaluation index system is established to monitor the accuracy and stability of the acquisition equipment in real time, ensuring the accuracy and reliability of the acquired data. This index system includes equipment calibration and data fluctuation coefficient parameters. By analyzing and evaluating these parameters, potential problems during the acquisition process can be identified and addressed in a timely manner, ensuring that the acquired data meets the requirements for subsequent processing and analysis.

[0008] Furthermore, in step S1, a data cleaning algorithm based on improved median filtering and adaptive weighted fusion is employed. For outlier handling in low-altitude meteorological data, the size of the sliding window centered on the data point is first determined. Calculate the median of the data within the window. By comparing the difference between a data point and the median, outliers are determined. For geological and hydrological data, a dynamic threshold range is set based on the physical meaning of the data and its historical distribution. Data exceeding this range are considered outliers. An adaptive weighting factor is introduced during the denoising process. The calculation formula is as follows: ;

[0009] in For the first Data points, The mean of the data. To prevent extremely small positive numbers with a denominator of zero, the data is fused using this weighting factor to effectively remove noise interference. In terms of data completion, for missing values ​​in low-altitude meteorological data, a spatiotemporal correlation prediction model is used, considering the correlation between adjacent time points and spatial locations, and a completion algorithm based on Lagrange interpolation is constructed. For geological and hydrological data, the seasonal variation patterns of historical data are utilized, combined with a grey prediction model to complete missing values, thereby improving the completeness and usability of the data.

[0010] Furthermore, in S2, a spatiotemporal fusion multi-hazard correlation model based on a single formula is adopted, using preprocessed and normalized encoded low-altitude meteorological data, geological data, and hydrological data as input vectors. The probability of geological disasters and floods triggered by low-altitude meteorological conditions can be calculated using the following formula. : ;

[0011] in The length of the time series. For the number of spatial locations, Indicates time Spatial location The feature vector at that location, and These are the trainable weight matrix and bias vector at corresponding positions, used to extract spatiotemporal features and calculate their importance weights. and It involves fusing the weight matrix and bias vector, then fusing and mapping the weighted spatiotemporal features. , The weight matrix and bias vector are used to assist in the calculation. Using the sigmoid activation function, the calculation results are mapped to a probability range of 0 to 1. The mean squared error loss function is minimized by applying the stochastic gradient descent algorithm to historical disaster data. ;

[0012] in For the sample size, For the actual occurrence of the disaster, Predict probabilities for the model and iteratively optimize. and Parameters are used to establish the correlation between meteorological conditions and the probability of geological disasters and floods, thus completing the construction of a multi-hazard correlation model.

[0013] Furthermore, the grid cell risk value calculation in S3 employs a multi-factor coupled risk assessment model for low-altitude meteorological risk values. The calculation formula is: ;

[0014] in For the first The weights of each low-altitude meteorological factor were determined through statistical analysis of historical low-altitude meteorological disaster events and expert scoring. The number of low-altitude meteorological factors, For the first The standardized values ​​of each low-altitude meteorological factor are calculated using the following formula: ;

[0015] in For the first The original values ​​of the low-altitude meteorological factors, and These are the minimum and maximum values ​​of the factor, respectively. Geological disaster risk value The calculation formula is: ;

[0016] in , , The weights for mountain slope, soil and rock type, and groundwater level were determined through logistic regression analysis of historical geological disaster data. , , These are the standardized values ​​of the corresponding factors; Flood disaster risk value The calculation formula is: ; in , , The weights for river flow, water level, and channel morphology were determined through principal component analysis of historical flood data. , , These are the standardized values ​​of the corresponding factors. This multi-factor coupled risk assessment model comprehensively considers multiple factors and accurately calculates various risk values ​​for each grid cell.

[0017] Furthermore, in S4, a risk assessment method based on dynamic threshold adjustment is adopted. The warning level thresholds for low-altitude meteorological risk, geological disaster risk, and flood disaster risk are not fixed, but dynamically adjusted according to the distribution characteristics of historical disaster data in different seasons and regions. First, the historical data is classified according to season and region, and the mean value of each type of risk value under each category is calculated. and standard deviation Then, based on the risk level classification requirements, threshold ranges are set for different levels. Generally, the risk level threshold range is... The threshold range for a more severe risk level is: The threshold range for severe risk level is: The threshold range for particularly severe risk levels is: When determining the comprehensive risk warning level, the analytic hierarchy process (AHP) is used to construct a judgment matrix for low-altitude meteorological risk, geological disaster risk, and flood disaster risk. By calculating the eigenvectors and eigenvalues ​​of the judgment matrix, the weight of each type of disaster risk in the comprehensive risk assessment is determined. Then, based on the single-disaster risk level and weight of each grid unit, the comprehensive risk warning level is calculated, making the risk assessment and warning level classification more in line with the actual situation and improving the accuracy and effectiveness of the warning.

[0018] Furthermore, S5 employs a precise early warning push strategy based on user profiles. First, it collects users' basic information, historical disaster response behavior data, and social attribute data to construct user profiles. Then, based on the characteristics of different user profiles, it analyzes users' attention to and receiving preferences for different types of disaster early warning information. For users living in mountainous areas, it focuses on pushing geological disaster and low-altitude meteorological disaster early warning information. For users engaged in agricultural production, in addition to regular disaster early warnings, it also pushes meteorological disaster early warning information related to agricultural production. In terms of push channel selection, for younger user groups, it prioritizes social media platforms and mobile applications, while for older user groups, it uses SMS and broadcasts. Through this precise early warning push strategy, it improves the reach rate of early warning information and users' attention to early warning information, enabling relevant parties to take disaster prevention and mitigation measures more promptly and effectively based on the early warning information.

[0019] Furthermore, the S5 real-time dynamic update step employs a model update algorithm based on incremental learning. When new data is collected, it is first preprocessed, then merged with historical data to form an incremental dataset. During the model update process, the entire model is not retrained; instead, a local update strategy is used. For the disaster-related deep neural network model based on a spatiotemporal attention mechanism, the impact of new data on the parameters of each layer of the model is calculated. Based on the impact, the range of parameters that need to be updated is determined. For parameters with significant impact, a stochastic gradient descent algorithm is used for updating. The update formula is: ;

[0020] in For model parameters, For learning rate, For loss function For parameters For parameters with minor impact, a weighted average method is used for updating the gradient to maintain the stability and convergence of the model. At the same time, a model performance evaluation index system is established. After each model update, the prediction results of historical data and new data are evaluated to determine whether the model update is effective. If the model performance deteriorates after the update, the update strategy is adjusted and the model is updated again to ensure that the multi-hazard correlation model always maintains high prediction accuracy and adaptability.

[0021] Furthermore, a data security system is established throughout the entire early warning process. For the collected data, encrypted transmission technology is used to prevent theft or tampering during transmission. In terms of data storage, a distributed storage approach is adopted, distributing data across multiple server nodes and performing data backups to ensure data security and reliability. For users and systems accessing the data, a strict identity authentication and access control mechanism is established, allowing only authorized users and systems to access the corresponding data. During model training and operation, data involving user privacy and sensitive information is anonymized to prevent data leakage. Through this data security system, the security of the entire high-resolution low-altitude meteorological risk early warning method is ensured during data collection, processing, storage, and use, protecting the information security of users and related institutions.

[0022] Compared with existing technologies, this high-resolution low-altitude meteorological risk early warning method has the following advantages: I. This invention constructs a correlation model between low-altitude meteorological risk and multiple disasters, including geological disasters and floods, by combining historical disaster data and real-time data. This enables coordinated early warning for multiple disasters. Simultaneously, by continuously collecting data in real time, the multi-hazard correlation model is dynamically updated and optimized. Based on the new calculation results, the risk values ​​and warning levels of grid units are adjusted in a timely manner, and the latest warning information is released. This dynamic update mechanism ensures the timeliness and accuracy of the warning information, which helps to respond to disaster changes in a timely manner and reduce disaster losses.

[0023] Second, this invention constructs a high-resolution gridded risk calculation system, subdividing the early warning area into multiple high-resolution grid units. It comprehensively considers the effects of multiple factors such as low-altitude wind speed, precipitation intensity, mountain slope, and river flow, accurately calculating the low-altitude meteorological risk value, geological disaster risk value, and flood disaster risk value within each grid unit. This refined risk calculation method significantly improves the accuracy and spatial resolution of early warnings, making early warning information more specific and targeted. It helps relevant departments and the public to understand the risk situation more accurately and take effective preventive measures.

[0024] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0026] Figure 1 A flowchart illustrating a high-resolution low-altitude meteorological risk early warning method. Figure 2 This is a schematic diagram of the data flow for a high-resolution low-altitude meteorological risk early warning method. Detailed Implementation

[0027] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0028] Example 1: During the plum rain season in a hilly mountainous area in southern China, meteorological satellites captured large-scale low-altitude meteorological cloud images and precipitation distribution. Weather radar tracked the movement path and intensity of rain belts in real time. Ground meteorological stations collected basic meteorological data such as temperature, humidity, and air pressure. For complex terrain such as mountain valleys and steep slopes, multiple low-altitude drones equipped with high-precision sensors were deployed to collect low-altitude vertical layered data, acquiring wind speed, wind direction, and raindrop spectrum data at different altitudes from the ground to hundreds of meters below the ground. For geological data collection, optical fibers were laid at different depths inside the mountains, utilizing distributed optical fiber sensing technology for real-time monitoring. Stress changes within the mountain caused by rainwater infiltration, combined with ground-based ground-penetrating radar data on the rock and soil structure, form a three-dimensional geological data system. Hydrological data collection covers the flow and water level monitoring of major rivers and reservoirs in the region. Satellite remote sensing continuously acquires dynamic changes in river width, and underwater topographic sonar is used to detect river depth and siltation. After data collection, all raw data are processed using a data cleaning algorithm based on improved median filtering and adaptive weighted fusion. For outlier handling in low-altitude meteorological data, the sliding window size centered on the data point is first determined. Calculate the median of the data within the window. By comparing the difference between a data point and the median, outliers are determined. For geological and hydrological data, a dynamic threshold range is set based on the physical meaning of the data and its historical distribution. Data exceeding this range are considered outliers. An adaptive weighting factor is introduced during the denoising process. The calculation formula is as follows: ;

[0029] in For the first Data points, The mean of the data. To prevent extremely small positive numbers with a denominator of zero, in terms of data completion, for missing values ​​in low-altitude meteorological data, a spatiotemporal correlation prediction model is adopted, which considers the correlation between data at adjacent time points and spatial locations, and a completion algorithm based on Lagrange interpolation is constructed. For geological and hydrological data, the seasonal variation patterns of historical data are used, combined with a grey prediction model to complete missing values. The accuracy and stability of the equipment are monitored in real time through a data quality assessment index system.

[0030] By integrating historical disaster data such as landslides, debris flows, and flash floods that occurred in the region during the plum rain season over the past 30 years, and combining this data with preprocessed real-time meteorological, geological, and hydrological data, a multi-hazard correlation model between low-altitude meteorological risk and geological and flood disasters was constructed. The preprocessed and normalized coded low-altitude meteorological, geological, and hydrological data were used as input vectors. The probability of geological disasters and floods triggered by low-altitude meteorological conditions can be calculated using the following formula. : ;

[0031] in The length of the time series. For the number of spatial locations, Indicates time Spatial location The feature vector at that location, and These are the trainable weight matrix and bias vector at corresponding positions, used to extract spatiotemporal features and calculate their importance weights. and It involves fusing the weight matrix and bias vector, then fusing and mapping the weighted spatiotemporal features. , The weight matrix and bias vector are used to assist in the calculation. Using the sigmoid activation function, the correlation between meteorological elements such as the duration of continuous rainfall, hourly rainfall intensity, and low-level wind speed during the plum rain season and the probability of landslides and the rise in river flood levels is analyzed to determine the weights of each influencing factor. A mathematical relationship between changes in meteorological conditions and the probability of disasters is established. The stochastic gradient descent algorithm is employed, and the model is trained using historical disaster data. The model parameters are optimized by minimizing the mean squared error loss function. ;

[0032] in For the sample size, For the actual occurrence of the disaster, Predict probabilities for the model and iteratively optimize. and Parameters are used to establish the correlation between meteorological conditions and the probability of geological disasters and floods, complete the construction of a multi-hazard correlation model, and improve the accuracy of the model in predicting the probability of disasters.

[0033] The hilly area was divided into multiple high-resolution grid cells with a precision of 100 meters × 100 meters. Based on collected real-time rainfall data, mountain stress change data, river flow data, etc., and combined with a trained multi-hazard correlation model, a multi-factor coupled risk assessment model was used to calculate the risk value of each grid cell. For low-altitude meteorological risk values... The calculation formula is: ;

[0034] in For the first The weights of low-altitude meteorological factors, The number of low-altitude meteorological factors, For the first Standardized values ​​of low-altitude meteorological factors; Geological disaster risk value The calculation formula is: ;

[0035] in , , The weights are respectively for mountain slope, soil and rock type, and groundwater level. , , These are the standardized values ​​of the corresponding factors; Flood disaster risk value The calculation formula is: ;

[0036] in , , The weights for river flow, water level, and river channel morphology are respectively. , , These are the standardized values ​​of the corresponding factors.

[0037] Based on the risk values ​​of each grid unit, a risk assessment method based on dynamic threshold adjustment is adopted. Combined with the historical data distribution characteristics of the plum rain season in this region, the risks of low-altitude meteorological disasters, geological disasters, and floods are assessed separately. First, the historical data is classified according to different stages of the plum rain season, and the mean and standard deviation of each type of risk value are calculated to determine the threshold range of the four-level warning levels, and the warning levels are divided into general, relatively severe, serious, and extremely severe. At the same time, the analytic hierarchy process is used to construct a judgment matrix to determine the weight of the three types of disaster risks in the comprehensive assessment, and then the comprehensive risk warning level of each grid unit is calculated.

[0038] Early warning information is disseminated through multiple channels. For villagers living in mountainous areas, village broadcasts and outdoor displays are used to promptly broadcast geological disaster and low-altitude meteorological risk warnings for their respective grids. For residents in townships, comprehensive risk warning information is pushed through SMS, television, and local government affairs apps. Data is continuously collected in real time, and a model update algorithm based on incremental learning is used to dynamically optimize the multi-hazard correlation model. When new rainfall data is received, only the parameters in the model that are significantly affected by the new data are locally updated without retraining the entire model. Based on the calculation results of the updated model, the risk values ​​and warning levels of each grid unit are adjusted in a timely manner. The latest warning information is disseminated through existing channels. For example, when the hourly rainfall intensity of a certain grid unit exceeds the threshold, its geological disaster warning level is upgraded from "relatively severe" to "severe" and immediately pushed out.

[0039] Example 2: When a plain city in northern China experiences severe convective weather in summer, meteorological satellites monitor the formation and movement of rainstorm clouds, weather radar accurately determines the location and intensity of the rainstorm, and ground meteorological stations are densely distributed throughout the city to record temperature, humidity, and wind speed at a height of 10 meters in real time. For the "tunnel effect" areas between high-rise buildings, low-altitude drones are deployed to collect low-altitude vertical layered data, acquiring wind speed and direction changes at different altitudes and capturing potential short-term strong gusts. Geological data collection focuses on river embankments and artificial fill areas around the city, using distributed fiber optic sensing technology to monitor the impact of rainwater on the river embankments. The stress changes caused by soaking, combined with soil compaction data detected by ground-based ground-penetrating radar, form a geological data system. Hydrological data collection includes the flow rate of the main urban drainage network, the operation data of rainwater pumping stations, and the monitoring of water level and flow velocity in urban rivers. Satellite remote sensing is used to obtain changes in the width of the river surface, and underwater topographic sonar is used to understand the drainage capacity of the river channel. The collected data is processed using a data cleaning algorithm based on improved median filtering and adaptive weighted fusion to remove outliers and duplicate data, unify the format, and complete missing data through interpolation and regression analysis. At the same time, a data quality assessment index system is used to ensure stable equipment operation.

[0040] Historical disaster data from the past 20 years, including urban flooding and billboard collapses caused by summer rainstorms, were collected and combined with pre-processed real-time data to construct a multi-hazard correlation model between low-altitude meteorological risk (focusing on low-altitude strong winds) and flooding. The correlation between meteorological elements such as hourly rainfall intensity, duration, and low-altitude gust speed during summer rainstorms and the depth of water accumulation in low-lying areas, drainage network pressure, and probability of damage to outdoor facilities was analyzed. The weights of each factor were determined, and a mathematical relationship between meteorological conditions and the probability of disaster occurrence was established. Using the stochastic gradient descent algorithm and historical disaster data as training samples, the model parameters were optimized by minimizing the mean squared error loss function to improve the model's prediction accuracy for urban flooding and low-altitude strong wind disasters.

[0041] The urban administrative area is divided into high-resolution grid units with a precision of 50 meters × 50 meters. Based on real-time data such as rainfall intensity, low-altitude wind speed, drainage network flow, and riverbank stress, and combined with a multi-hazard correlation model, a multi-factor coupled risk assessment model is used to calculate the risk value of each grid unit. When calculating the low-altitude meteorological risk value, the impact of low-altitude gust wind speed on outdoor billboards and trees is given special consideration. When calculating the flood disaster risk value, the effects of river flow, drainage network water level, and urban terrain slope (such as low-lying areas) are taken into account to comprehensively assess the disaster risk of each grid unit.

[0042] Based on the risk values ​​of each grid unit, a risk assessment method based on dynamic threshold adjustment is adopted. Combined with the historical data characteristics of summer rainstorms in the city, four levels of early warning are divided for low-altitude meteorological risk and flood disaster risk. First, the historical data are classified according to the rainstorm characteristics of different months in summer. The mean and standard deviation of each type of risk value are calculated to determine the threshold range of general, relatively severe, serious, and extremely severe. At the same time, the analytic hierarchy process is used to determine the weight of the two disaster risks in the comprehensive assessment and calculate the comprehensive risk warning level of each grid unit. For example, for grids in low-lying areas of the city with dense high-rise buildings around them, the superimposed risks of urban flooding and low-altitude strong winds need to be considered simultaneously.

[0043] A precise push strategy is adopted for different user groups. For young office workers, early warning information for their grid is pushed through social media platforms and city service apps; for elderly residents, early warnings are issued through SMS and community broadcasts; for outdoor workers (such as construction workers and deliverymen), low-altitude strong wind and urban flooding risk warnings are pushed through industry-specific apps. Data is continuously collected in real time, and the model is dynamically optimized using an incremental learning-based model update algorithm. When new rainstorm center data is received, only the parameters most affected in the model are updated. The model's timeliness is ensured through local adjustments. Based on the updated calculation results, the risk value and early warning level of the grid unit are adjusted in a timely manner. For example, when the water depth of a grid unit reaches the threshold, its flood disaster warning level is upgraded from "general" to "severe" and updated information is issued immediately.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A high-resolution low-altitude meteorological risk early warning method, characterized in that, The specific steps of this method are as follows: S1. Data Acquisition and Preprocessing Steps: Using meteorological satellites, meteorological radars, ground meteorological stations, geological disaster monitoring sensors, and hydrological monitoring equipment, low-altitude meteorological data, geological data, and hydrological data are collected in real time. After collection, the raw data is preprocessed by cleaning, denoising, and format conversion to remove abnormal and duplicate data, unify the format, and use interpolation and regression analysis to complete missing data. S2. Steps for constructing and training a multi-hazard correlation model: Combining historical disaster data and pre-processed real-time data, construct a correlation model between low-altitude meteorological risk and geological disasters, floods, and other multi-hazards. By analyzing the correlation between low-altitude meteorological elements and the occurrence of other disasters, determine the weight of each influencing factor, establish the mathematical relationship between changes in meteorological conditions and the probability of disaster occurrence, and train and optimize the model. S3. High-resolution gridded risk calculation steps: Divide the warning area into multiple high-resolution grid units. Based on the collected data and the correlation model, calculate the low-altitude meteorological risk value, geological disaster risk value and flood disaster risk value in each grid unit. When calculating, comprehensively consider the impact of low-altitude wind speed and precipitation intensity on the disaster-bearing body, as well as the effects of multiple factors such as mountain slope and river flow. S4. Risk Assessment and Classification Steps: Based on the risk value of each grid unit and in conjunction with the preset risk assessment standards, assess the risks of low-altitude meteorological disasters, geological disasters, and flood disasters respectively, and classify them into four levels of early warning: general, relatively severe, severe, and extremely severe. At the same time, take into account the risk situation of multiple disasters and determine the comprehensive risk early warning level of each grid unit. S5. Linked Early Warning Issuance and Dynamic Update Steps: Through multiple channels such as SMS, radio, television, internet platforms, and outdoor displays, timely release early warning information on low-altitude meteorological, geological disaster, and flood disaster risks, as well as comprehensive risk early warning information for each grid unit. Continuously collect data in real time, dynamically update and optimize the multi-hazard correlation model, adjust the grid unit risk value and early warning level in a timely manner based on the new calculation results, and issue the latest early warning.

2. The high-resolution low-altitude meteorological risk early warning method according to claim 1, characterized in that, The low-altitude meteorological data acquisition in S1 also includes using low-altitude UAVs equipped with high-precision sensors to conduct low-altitude vertical layered data acquisition in specific complex terrain areas, obtaining meteorological data at different altitudes. For geological data acquisition, distributed fiber optic sensing technology is used, laying optical fibers inside the mountain to monitor stress changes inside the mountain in real time. Combined with ground-based ground-penetrating radar data on the rock and soil structure, a three-dimensional geological data acquisition system is formed. In addition to conventional river flow and water level monitoring, hydrological data acquisition also utilizes satellite remote sensing technology to obtain dynamic changes in river width, combined with underwater topographic sonar data, to construct a complete hydrological data acquisition scheme. During the data acquisition process, a data quality assessment index system is established to monitor the accuracy and stability of the acquisition equipment in real time. This index system includes equipment calibration degree and data fluctuation coefficient parameters.

3. The high-resolution low-altitude meteorological risk early warning method according to claim 1, characterized in that, In step S1, a data cleaning algorithm based on improved median filtering and adaptive weighted fusion is used. For outlier handling in low-altitude meteorological data, the size of the sliding window centered on the data point is first determined. Calculate the median of the data within the window. By comparing the difference between a data point and the median, outliers are determined. For geological and hydrological data, a dynamic threshold range is set based on the physical meaning of the data and its historical distribution. Data exceeding this range are considered outliers. An adaptive weighting factor is introduced during the denoising process. The calculation formula is as follows: in For the first Data points, The mean of the data. To prevent extremely small positive numbers with a denominator of zero, in terms of data completion, for missing values ​​in low-altitude meteorological data, a spatiotemporal correlation prediction model is adopted, which considers the correlation between data at adjacent time points and spatial locations, and a completion algorithm based on Lagrange interpolation is constructed. For geological and hydrological data, the seasonal variation patterns of historical data are used, combined with a grey prediction model to complete missing values.

4. The high-resolution low-altitude meteorological risk early warning method according to claim 1, characterized in that, In S2, a spatiotemporal fusion multi-hazard correlation model based on a single formula is adopted, using preprocessed and normalized encoded low-altitude meteorological data, geological data, and hydrological data as input vectors. The probability of geological disasters and floods triggered by low-altitude meteorological conditions can be calculated using the following formula. : in The length of the time series. For the number of spatial locations, Indicates time Spatial location The feature vector at that location, and These are the trainable weight matrix and bias vector at corresponding positions, used to extract spatiotemporal features and calculate their importance weights. and It involves fusing the weight matrix and bias vector, then fusing and mapping the weighted spatiotemporal features. , The weight matrix and bias vector are used to assist in the calculation. Using the sigmoid activation function, the calculation results are mapped to a probability range of 0 to 1. The mean squared error loss function is minimized by applying the stochastic gradient descent algorithm to historical disaster data. ； in For the sample size, For the actual occurrence of the disaster, Predict probabilities for the model and iteratively optimize. and Parameters are used to establish the correlation between meteorological conditions and the probability of geological disasters and floods, thus completing the construction of a multi-hazard correlation model.

5. The high-resolution low-altitude meteorological risk early warning method according to claim 1, characterized in that, The grid cell risk value calculation in S3 adopts a multi-factor coupled risk assessment model for low-altitude meteorological risk values. The calculation formula is: ； in For the first The weights of low-altitude meteorological factors, The number of low-altitude meteorological factors, For the first Standardized values ​​of low-altitude meteorological factors; Geological disaster risk value The calculation formula is: ； in , , The weights are respectively for mountain slope, soil and rock type, and groundwater level. , , These are the standardized values ​​of the corresponding factors; Flood disaster risk value The calculation formula is: ； in , , The weights for river flow, water level, and river channel morphology are respectively. , , These are the standardized values ​​of the corresponding factors.

6. The high-resolution low-altitude meteorological risk early warning method according to claim 1, characterized in that, The risk assessment method based on dynamic threshold adjustment used in S4 is not fixed for the warning level thresholds of low-altitude meteorological risk, geological disaster risk, and flood disaster risk. Instead, it is dynamically adjusted according to the distribution characteristics of historical disaster data in different seasons and regions. First, the historical data is classified according to season and region, and the mean value of each type of risk value under each category is calculated. and standard deviation Then, based on the risk level classification requirements, threshold ranges are set for different levels. Generally, the risk level threshold range is... The threshold range for a more severe risk level is: The threshold range for severe risk level is: The threshold range for particularly severe risk levels is: When determining the comprehensive risk warning level, the analytic hierarchy process (AHP) is used to construct a judgment matrix for low-altitude meteorological risk, geological disaster risk, and flood disaster risk. By calculating the eigenvectors and eigenvalues ​​of the judgment matrix, the weight of each type of disaster risk in the comprehensive risk assessment is determined. Then, based on the single-disaster risk level and weight of each grid unit, the comprehensive risk warning level is calculated.

7. The high-resolution low-altitude meteorological risk early warning method according to claim 1, characterized in that, The S5 employs a precise early warning push strategy based on user profiles. First, it collects users' basic information, historical disaster response behavior data, and social attribute data to construct user profiles. Then, based on the characteristics of different user profiles, it analyzes users' attention to and receiving preferences for different types of disaster early warning information. For users living in mountainous areas, it focuses on pushing geological disaster and low-altitude meteorological disaster early warning information. For users engaged in agricultural production, in addition to regular disaster early warnings, it also pushes meteorological disaster early warning information related to agricultural production. In terms of push channel selection, for younger user groups, it prioritizes social media platforms and mobile applications, while for older user groups, it uses SMS and broadcast methods.

8. The high-resolution low-altitude meteorological risk early warning method according to claim 1, characterized in that, The S5 real-time dynamic update step employs an incremental learning-based model update algorithm. When new data is collected, it is first preprocessed, then merged with historical data to form an incremental dataset. During the model update process, the entire model is not retrained; instead, a local update strategy is used. For the disaster-related deep neural network model based on spatiotemporal attention, the impact of new data on the parameters of each layer of the model is calculated. Based on the impact, the range of parameters that need to be updated is determined. For parameters with significant impact, a stochastic gradient descent algorithm is used for updating. The update formula is as follows: ； in For model parameters, For learning rate, For loss function For parameters For parameters with minor impact, a weighted average method is used for updating the gradient. At the same time, a model performance evaluation index system is established. After each model update, the prediction results of historical data and new data are evaluated to determine whether the model update is effective. If the model performance deteriorates after the update, the update strategy is adjusted and the model is updated again.

9. The high-resolution low-altitude meteorological risk early warning method according to claim 1, characterized in that, Throughout the entire early warning process, a data security system is established. For the collected data, encryption transmission technology is used to prevent the data from being stolen or tampered with during transmission. In terms of data storage, a distributed storage method is adopted, distributing the data across multiple server nodes and performing data backup. For users and systems accessing the data, a strict identity authentication and access control mechanism is established, allowing only authorized users and systems to access the corresponding data. During model training and operation, data involving user privacy and sensitive information is anonymized to prevent data leakage.

Images