Dangerous weather intelligent early warning system based on data analysis

Abstract

The invention relates to the technical field of meteorology, in particular to a dangerous weather intelligent early warning system based on data analysis, which comprises a multi-source weather data acquisition module for acquiring dangerous weather data such as meteorology, geology and hydrology; the multi-source weather data transmission module transmits data to the cloud through a communication interruption coping mechanism and sets access authority; the multi-source weather data processing module is used for cleaning and proofreading the data and extracting key weather characteristics; the avalanche judgment model building module is used for building an avalanche judgment model based on an avalanche occurrence probability and a confidence interval by using GAN enhanced feature data; the early warning generation module simulates an avalanche path and an influence area through an avalanche physics model and a shallow water model, evaluates a danger level and issues an early warning; and the model updating module dynamically optimizes the avalanche judgment model based on the latest data, so that the accuracy and timeliness of the early warning capability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorology, and in particular to an intelligent early warning system for dangerous weather based on data analysis. Background Art

[0002] The plateau area has a high altitude and complex and changeable terrain, and the transportation conditions are relatively inconvenient, resulting in a low density of automatic weather stations and radar sites. Remote sensing observations are often restricted by factors such as terrain shielding. Blizzards in plateau areas are often accompanied by severe weather such as low temperatures, strong winds and blowing snow, and sometimes even trigger avalanches, posing a serious threat to transportation, electricity, agriculture and animal husbandry, and personal safety. Therefore, timely and accurate avalanche prediction is of vital importance to effectively reduce accidents and economic losses.

[0003] Some existing solutions can deploy automatic weather stations and ground-based radars in plateau areas, and realize the preliminary feedback of observation data through low-power wide area networks or satellite communications. Combined with weather forecasts and satellite data, these solutions can make macro judgments on large-scale avalanches. However, these early warning systems usually lack refined analysis methods when determining the affected area, and fail to accurately identify the specific location, propagation path and impact range of the avalanche, which makes the early warning information less targeted and difficult to more accurately predict the impact range of plateau blizzards. In addition, under the condition of limited communication conditions in plateau areas, these early warning systems often lack effective emergency communications and information transmission mechanisms, resulting in low efficiency in warning issuance and emergency response, and cannot meet the needs of emergency handling.

[0004] In order to improve the accuracy and timeliness of early warning capabilities, a dangerous weather intelligent early warning system based on data analysis is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent early warning system for dangerous weather based on data analysis, which obtains meteorological, geological, hydrological and other dangerous weather data through a multi-source weather data acquisition module; a multi-source weather data transmission module transmits data to the cloud through a communication interruption response mechanism and sets access rights; a multi-source weather data processing module cleans and proofreads the data, and extracts key weather features; an avalanche determination model establishment module uses GAN to enhance feature data and establish an avalanche determination model based on avalanche occurrence probability and confidence interval; an early warning generation module simulates avalanche paths and impact areas through avalanche physics models and shallow water models, evaluates the danger level and issues early warnings; and a model update module dynamically optimizes the avalanche determination model based on the latest data, thereby improving the accuracy and timeliness of the early warning capability. To achieve the above purpose, the present invention provides the following technical solutions:

[0006] A dangerous weather intelligent early warning system based on data analysis, comprising:

[0007] Multi-source weather data collection module, used to collect dangerous weather data including meteorological data, geological data and hydrological data;

[0008] A multi-source weather data transmission module, used to transmit the dangerous weather data to the cloud server using a communication interruption response mechanism, and to set access permissions for the dangerous weather data, and to perform hierarchical permission management on the dangerous weather data according to role identities;

[0009] A multi-source weather data processing module is used to perform data cleaning and quality proofreading on the dangerous weather data to obtain a first data set; and to filter the first data set by using a Pearson correlation coefficient to obtain weather characteristic data;

[0010] An avalanche determination model establishment module is used to perform data enhancement on the weather characteristic data using a GAN model to obtain a second data set; establish an avalanche determination model, identify the second data set, and output an avalanche occurrence probability and an avalanche confidence interval;

[0011] The warning generation module calculates the initial flow velocity using an avalanche physics model according to the dangerous weather data if the avalanche occurrence probability and the avalanche confidence interval meet the first condition, and performs path simulation using a shallow water model according to the initial flow velocity to generate an avalanche impact area; generates an avalanche hazard level according to the avalanche impact area and issues warning information;

[0012] The dangerous weather prediction model updating module is used to obtain the latest dangerous weather data, and automatically perform data analysis and update the avalanche determination model.

[0013] Furthermore, the dangerous weather data includes:

[0014] The meteorological data include temperature, humidity, air pressure, wind speed and direction, and precipitation;

[0015] The geological data include snow depth, terrain slope, rock and soil layer distribution, and historical events of avalanche disasters;

[0016] The hydrological data include snow melt rates, river flow and soil moisture.

[0017] Furthermore, the communication interruption response mechanism includes:

[0018] If the network is interrupted, the hazardous weather data is stored in a local storage device;

[0019] If the network is restored, the local storage device synchronizes the hazardous weather data using an incremental upload method.

[0020] Furthermore, the access permission setting process includes:

[0021] Encrypt the hazardous weather data during transmission using an encryption protocol to generate encrypted weather data;

[0022] The role identities include meteorologist, emergency personnel and administrator, and the access rights to the weather encrypted data are set according to different role identities, including:

[0023] The meteorologist has the authority to view and modify the multi-source weather data acquisition module, the multi-source weather data transmission module, and the multi-source weather data processing module;

[0024] The emergency personnel have the authority to view the multi-source weather data acquisition module and the warning generation module;

[0025] The administrator has the authority to view and modify the multi-source weather data acquisition module, the multi-source weather data transmission module, the multi-source weather data processing module, the avalanche determination model establishment module, the warning generation module and the dangerous weather prediction model update module.

[0026] Furthermore, the data cleaning and the quality proofreading specifically include: using time series interpolation method and spatial interpolation method to complete the dangerous weather data, and cross-validating the dangerous weather data from different data sources to obtain the first data set.

[0027] Furthermore, the implementation process of the Pearson correlation coefficient includes:

[0028] Calculating the Pearson correlation coefficient between each of the dangerous weather data and the avalanche occurrence event;

[0029] A characteristic absolute value threshold is set, and if the Pearson correlation coefficient is greater than the characteristic absolute value threshold, the dangerous weather data corresponding to the Pearson correlation coefficient is filtered as the weather characteristic data.

[0030] Furthermore, the implementation process of the avalanche determination model includes:

[0031] Constructing the avalanche determination model, including an input layer, a hidden layer, an output layer and a Bayesian layer; the input layer is used to receive the second data set; the hidden layer is used to extract data features of the second data set; the output layer is used to predict the avalanche occurrence probability and the avalanche confidence interval; the Bayesian layer introduces uncertainty analysis to obtain the confidence of the prediction result;

[0032] The loss function is set to a weighted combination of negative log-likelihood and KL divergence; the negative log-likelihood is used to evaluate the loss value of the avalanche occurrence probability, and the KL divergence is used to evaluate the loss value of the avalanche confidence interval;

[0033] Splitting the second data set into a training set and a test set, and inputting the sets into the avalanche determination model; the training set is used to train the avalanche determination model, and the test set is used to evaluate the avalanche determination model;

[0034] The avalanche determination model after the test is saved.

[0035] Furthermore, the first condition includes: the minimum value of the avalanche confidence interval is greater than a preset confidence threshold, the width of the avalanche confidence interval is less than a preset interval width, and the probability of avalanche occurrence is greater than a preset probability threshold.

[0036] Furthermore, the process of generating the avalanche hazard level includes:

[0037] Obtain each geographic grid, if the avalanche flow velocity corresponding to the geographic grid exceeds a flow velocity threshold, and the avalanche depth exceeds a depth threshold, then the geographic grid is the avalanche impact area;

[0038] For the avalanche affected area, the avalanche hazard level is set according to the range in which the avalanche flow velocity exceeds the flow velocity threshold, and the range in which the avalanche depth exceeds the depth threshold;

[0039] The avalanche risk levels include a low risk level, a medium risk level and a high risk level.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. The present invention adopts a communication interruption response mechanism. In the case of network interruption, key dangerous weather data can be safely stored in local storage devices. After the network is restored, the data is synchronized using the incremental upload method, which enhances the reliability and security of data transmission and avoids data loss or delay caused by communication interruption. At the same time, the dangerous weather data in the transmission process is encrypted through an encryption protocol, and hierarchical authority management is set based on role identity, which not only ensures that different role identities can accurately access the required modules, but also improves the accuracy and timeliness of avalanche risk warnings.

[0042] 2. The present invention uses the GAN model to enhance the weather feature data, improves the robustness of the model, and effectively solves the problem of insufficient historical avalanche data. On this basis, an avalanche judgment model including an input layer, a hidden layer, an output layer, and a Bayesian layer is constructed. The model is finely optimized through the loss function of the weighted combination of negative log-likelihood and KL divergence, which improves the prediction accuracy of the probability of avalanche and the calculation reliability of the confidence interval, providing an accurate basis for risk assessment, thereby improving the accuracy and timeliness of avalanche risk warning.

[0043] 3. The present invention calculates the initial flow velocity through the avalanche physics model, and simulates the path in combination with the shallow water model to generate a detailed map of the avalanche impact area, realizing the accurate division of the avalanche impact area and risk level assessment, thereby making the warning information more targeted and timely. In addition, based on the gridded area, the avalanche flow velocity and depth thresholds are set to divide the three risk levels of low, medium and high, and the corresponding warning information is issued, which can lock the high-risk areas in advance, effectively improving the accuracy and timeliness of the avalanche risk warning, thereby preparing for the adoption of effective preventive measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A schematic diagram of the structure of a dangerous weather intelligent early warning system based on data analysis of the present invention;

[0045] Figure 2 It is a flow chart of the communication interruption coping mechanism of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] See also Figure 1 to Figure 2 The present invention provides a dangerous weather intelligent early warning system based on data analysis, and the technical solution is as follows:

[0048] Area A is located in a high-altitude mountainous area with complex terrain. It is one of the areas prone to avalanches. The climate conditions in the area are changeable, the snow is thick in winter, and snowmelt is prone to occur after the temperature rises in spring, triggering avalanches. In addition, the population density in Area A is low, but the main transportation arteries and some villages may be located in the potential avalanche impact area. Once an avalanche occurs, it will have a significant impact on transportation and residents' lives. In response to the avalanche monitoring and early warning needs in Area A, Example 1 is as follows:

[0049] like Figure 1 As shown, a dangerous weather intelligent early warning system based on data analysis includes:

[0050] refer to Figure 1 The multi-source weather data collection module in is used to collect dangerous weather data including meteorological data, geological data and hydrological data.

[0051] Furthermore, the dangerous weather data includes:

[0052] The meteorological data include temperature, humidity, air pressure, wind speed and direction, and precipitation;

[0053] The geological data include snow depth, terrain slope, rock and soil layer distribution, and historical events of avalanche disasters;

[0054] The hydrological data include snow melt rates, river flow and soil moisture.

[0055] Specifically, according to the existing automatic weather stations in the plateau area, or the automatic weather stations deployed in the avalanche-prone areas, equipped with temperature, humidity, air pressure, wind speed and direction, and precipitation sensors, data are automatically recorded every 10 minutes. Then, the meteorological satellite images and meteorological radar data are integrated to obtain geological related information. Among them, the snow depth is identified by satellite imaging technology, the snow accumulation area and depth, the terrain slope is calculated using the digital elevation model (DEM), the rock and soil layer distribution is obtained by geological radar scanning and analysis, and the historical events of avalanche disasters are integrated according to the time, location and impact range of past avalanche events. Satellite images and remote sensing technology are used to calculate NDSI changes and identify the melting rate of snow. River flow and soil moisture data are integrated through ground hydrological monitoring stations, remote sensing technology and hydrological numerical models.

[0056] Through the multi-source weather data collection module, the system can integrate dangerous weather data sets from different sources to provide comprehensive data support for subsequent modules, thereby effectively improving the accuracy and timeliness of avalanche risk warnings.

[0057] refer to Figure 1 The multi-source weather data transmission module is used to transmit the dangerous weather data to the cloud server using the communication interruption response mechanism, and to set access permissions for the dangerous weather data, so as to perform hierarchical permission management on the dangerous weather data according to role identity.

[0058] Furthermore, the communication interruption response mechanism includes:

[0059] If the network is interrupted, the hazardous weather data is stored in a local storage device;

[0060] If the network is restored, the local storage device synchronizes the hazardous weather data using an incremental upload method.

[0061] Specifically, in this embodiment, Figure 2 FIG. 1 is a flow chart of the communication interruption response mechanism of the present invention. Figure 2As shown in the figure, the local storage device uses the storage device of the automatic weather station. First, check the network status to determine whether the network is normal. If normal, use the normal data transmission mode and transmit it to the cloud server. If the network is interrupted or unstable, the system automatically switches to the local storage mode and stores the real-time collected dangerous weather data (such as temperature, humidity, snow depth and soil moisture, etc.) in chronological order to ensure the integrity and continuity of the data. Among them, the circular overwrite storage strategy is used to ensure that the latest data is saved first, avoid data loss due to insufficient storage space, and improve the utilization rate of local storage space. When the network connection of the monitoring point is restored to stability, the system will start the incremental upload mechanism to synchronize the newly added data stored locally during the interruption to the cloud server, reducing the amount of data transmitted repeatedly and improving the transmission efficiency, thereby ensuring the stable operation of the system in harsh environments and improving the accuracy and timeliness of avalanche risk warnings.

[0062] Furthermore, the access permission setting process includes:

[0063] Encrypt the hazardous weather data during transmission using an encryption protocol to generate encrypted weather data;

[0064] The role identities include meteorologist, emergency personnel and administrator, and the access rights to the weather encrypted data are set according to different role identities, including:

[0065] The meteorologist has the authority to view and modify the multi-source weather data acquisition module, the multi-source weather data transmission module, and the multi-source weather data processing module;

[0066] The emergency personnel have the authority to view the multi-source weather data acquisition module and the warning generation module;

[0067] The administrator has the authority to view and modify the multi-source weather data acquisition module, the multi-source weather data transmission module, the multi-source weather data processing module, the avalanche determination model establishment module, the warning generation module and the dangerous weather prediction model update module.

[0068] Specifically, in this embodiment, the encryption protocol uses the TLS protocol to ensure that the data is not tampered with during transmission. Among the three types of user roles set up in the system, meteorologists are mainly responsible for the analysis and processing of weather data, emergency personnel are mainly concerned with the monitoring and early warning information of disasters in order to respond quickly to disasters, and administrators comprehensively manage the system. The corresponding subdivided permission allocation table is shown in Table 1, which gives the three types of users the permission to use the system.

[0069] Table 1 Permission allocation table

[0070] Module Name Meteorologist's authority Emergency Personnel Authority Administrator privileges Multi-source weather data acquisition module View, Modify Check View, Modify Multi-source weather data transmission module View, Modify none View, Modify Multi-source weather data processing module View, Modify none View, Modify Avalanche determination model building module none none View, Modify Warning generation module none Check View, Modify Dangerous weather forecast model update module none none View, Modify

[0071] Through permission settings and data encryption measures, the system not only ensures data security, but also clarifies the division of responsibilities, improves management efficiency, and supports rapid emergency response to warning information when avalanches are predicted. This not only ensures the efficient operation of the system, but also provides reliable technical support for relevant departments, improving the accuracy and timeliness of avalanche risk warnings.

[0072] refer to Figure 1 The multi-source weather data processing module is used to clean and quality check the dangerous weather data to obtain a first data set; the first data set is screened by the Pearson correlation coefficient to obtain weather characteristic data.

[0073] Furthermore, the data cleaning and the quality proofreading specifically include: using time series interpolation method and spatial interpolation method to complete the dangerous weather data, and cross-validating the dangerous weather data from different data sources to obtain the first data set.

[0074] Specifically, in this embodiment, since there may be missing data in meteorological data (such as temperature and humidity) and hydrological data (such as snow melt rate), linear interpolation is used to fill in the missing time point data. For the spatial distribution in geological data (such as terrain slope and rock and soil layer distribution) and the spatial distribution in hydrological data (such as river flow and soil moisture), the inverse distance weighted method is used to fill in.

[0075] In order to ensure the consistency and accuracy of data between different data sources, it is necessary to cross-validate the hazardous weather data from different data sources. For example, compare whether the precipitation of the automatic weather station and the meteorological satellite in the overlapping area and time period is consistent. After aligning the time index, use the Spearman rank correlation coefficient method or the Pearson correlation coefficient method to calculate the correlation coefficient of the two data. If the correlation coefficient is high (such as greater than 0.8), it means that the data consistency of the two data sources is good. Otherwise, it is necessary to further check the data source and processing process, and the data may need to be re-cleaned. After time series interpolation, spatial interpolation and cross-data source cross-validation, all data are integrated to generate the first data set for subsequent model training. Through the above data cleaning and quality proofreading, the data from different data sources are synchronized, which effectively improves the reliability of the data and provides a reliable data foundation for subsequent modules, thereby improving the accuracy of avalanche risk warning.

[0076] Furthermore, the implementation process of the Pearson correlation coefficient includes:

[0077] Calculating the Pearson correlation coefficient between each of the dangerous weather data and the avalanche occurrence event;

[0078] A characteristic absolute value threshold is set, and if the Pearson correlation coefficient is greater than the characteristic absolute value threshold, the dangerous weather data corresponding to the Pearson correlation coefficient is filtered as the weather characteristic data.

[0079] Specifically, in the present embodiment, the feature absolute value threshold is set to 0.7. For each type of data (such as temperature, precipitation, etc.), it is aligned with the sequence of avalanche events, and the Pearson correlation coefficient between them and avalanche events is calculated. If the Pearson correlation coefficient is greater than 0.7, this data is selected into the weather characteristic data. As shown in Table 2, the correlation coefficient of each data type and whether it is screened as weather characteristic data are shown. By selecting feature data that is highly correlated with avalanche risk (such as temperature, wind speed, precipitation, snow depth, etc.), irrelevant or redundant data can be effectively eliminated, reducing the computational burden of the avalanche determination model. This not only improves the training efficiency of the model, but also enhances its prediction accuracy, so that avalanche risk warning can be performed more accurately.

[0080] Table 2 Screening results

[0081] Data Types Pearson correlation coefficient Feature absolute value threshold (0.7) Filter results temperature 0.82 Greater than 0.7 Filter to weather feature data Precipitation 0.76 Greater than 0.7 Filter to weather feature data Wind speed 0.65 Less than 0.7 No filter humidity 0.73 Greater than 0.7 Filter to weather feature data Snow depth 0.88 Greater than 0.7 Filter to weather feature data Terrain slope 0.72 Greater than 0.7 Filter to weather feature data Soil moisture 0.78 Greater than 0.7 Filter to weather feature data Snow Melt Rate 0.75 Greater than 0.7 Filter to weather feature data River flow 0.55 Less than 0.7 No filter Distribution of rock and soil layers 0.68 Less than 0.7 No filter

[0082] refer to Figure 1 The avalanche determination model establishment module is used to use the GAN model to enhance the weather characteristic data to obtain a second data set; establish an avalanche determination model, identify the second data set, and output the avalanche occurrence probability and avalanche confidence interval.

[0083] Among them, in the data collection process of avalanche risk warning, since avalanches are extremely low-probability events, it is difficult to obtain a large amount of accurately labeled avalanche event data. GAN can use the collected weather feature data for training to generate synthetic data that is highly similar to the real data distribution. This provides more training resources for the model, effectively improving the generalization ability of the avalanche judgment model, so that it can more accurately judge and predict when facing actual avalanche events.

[0084] Furthermore, the implementation process of the avalanche determination model includes:

[0085] Constructing the avalanche determination model, including an input layer, a hidden layer, an output layer and a Bayesian layer; the input layer is used to receive the second data set; the hidden layer is used to extract data features of the second data set; the output layer is used to predict the avalanche occurrence probability and the avalanche confidence interval; the Bayesian layer introduces uncertainty analysis to obtain the confidence of the prediction result;

[0086] The loss function is set to a weighted combination of negative log-likelihood and KL divergence; the negative log-likelihood is used to evaluate the loss value of the avalanche occurrence probability, and the KL divergence is used to evaluate the loss value of the avalanche confidence interval;

[0087] Splitting the second data set into a training set and a test set, and inputting the sets into the avalanche determination model; the training set is used to train the avalanche determination model, and the test set is used to evaluate the avalanche determination model;

[0088] The avalanche determination model after the test is saved.

[0089] In this embodiment, the avalanche determination model is constructed based on a Bayesian neural network (BNN), including an input layer, a hidden layer, an output layer and a Bayesian layer. In order to optimize the model performance, the loss function is expressed as:

[0090] L = NLL + α × KL;

[0091] Among them, L is the loss value, NLL is the negative log-likelihood value, KL is the KL divergence value, and α is the weight parameter, which can be set to 0.8.

[0092] Next, the second data set was split into a training set and a test set in a ratio of 8 to 2. The training set was input into the avalanche determination model, and the Adam optimizer and the learning rate decay strategy were used to perform multiple iterations of training until the loss function converged. The test set was used to evaluate the avalanche determination model, using mean square error (MSE) and prediction interval coverage (PICP) as evaluation indicators. The results of BNN after the test were MSE of 7.4% and PICP of 91%. It can not only predict the probability of avalanche occurrence, but also evaluate the uncertainty of the prediction results, output confidence, and provide more accurate avalanche risk warnings, thereby providing more valuable reference information for evacuation decisions for avalanches.

[0093] Table 3 Judgment results

[0094]

[0095] refer to Figure 1 In the warning generation module, if the avalanche occurrence probability and the avalanche confidence interval meet the first condition, the initial flow velocity is calculated using the avalanche physics model according to the dangerous weather data, and the path simulation is performed using the shallow water model according to the initial flow velocity to generate the avalanche impact area; the avalanche hazard level is generated according to the avalanche impact area and warning information is issued.

[0096] Furthermore, the first condition includes: the minimum value of the avalanche confidence interval is greater than a preset confidence threshold, the width of the avalanche confidence interval is less than a preset interval width, and the probability of avalanche occurrence is greater than a preset probability threshold.

[0097] Specifically, in this embodiment, the preset confidence threshold is 50%, the preset interval width is 10%, and the preset occurrence probability threshold is 50%. The judgment result is shown in Table 3. If the above three conditions are met at the same time, it is judged as "yes", indicating that the next step of warning generation can be performed; otherwise, it is judged as "no". By setting the first condition, avalanche events that meet the warning conditions can be effectively screened out to avoid the occurrence of false alarms and missed alarms. In addition, combined with the comprehensive judgment of avalanche occurrence probability and confidence interval, the reliability of model prediction and scientific decision-making can be improved, thereby improving the accuracy of avalanche risk warning.

[0098] Furthermore, the initial flow velocity is calculated according to the physical characteristics of the avalanche to simulate the initial motion state of the avalanche, which is expressed as:

[0099]

[0100] Among them, v0 is the initial flow velocity, g is the gravitational acceleration, H is the vertical drop of the avalanche starting point, and μ is the friction coefficient.

[0101] The process of obtaining H is to select steep areas (such as slope greater than 30°) as possible avalanche starting points based on the terrain slope, then obtain the lowest point height below the avalanche path, and subtract these two values ​​to obtain H. The friction coefficient is 0.1 to 0.5, and the friction coefficient can be calculated by reverse calculation based on the avalanche path and sliding speed in historical avalanche data.

[0102] Specifically, area A is divided into regular geographic grids, each of which contains key parameters related to avalanche propagation, such as terrain slope, geological roughness, and snow depth, which can accurately depict local terrain features. At the starting point of the avalanche, the initial flow velocity and initial snow depth are assigned as initial conditions for the simulation based on model calculations and monitoring data. The avalanche propagation process gradually expands downstream of the grid according to physical laws, and the explicit Euler method is used to numerically solve the shallow water equation. The shallow water equation describes the conservation of mass and momentum in avalanche flow. Through step-by-step iterative calculations, the depth changes and velocity distribution during avalanche flow are dynamically simulated. After the simulation is completed, the system outputs the avalanche depth and velocity distribution corresponding to each geographic grid, showing the propagation characteristics of avalanches in area A, and can clearly identify potential high-risk areas, thereby providing data support for disaster warning and emergency response.

[0103] Furthermore, the process of generating the avalanche hazard level includes:

[0104] Obtain each geographic grid, if the avalanche flow velocity corresponding to the geographic grid exceeds a flow velocity threshold, and the avalanche depth exceeds a depth threshold, then the geographic grid is the avalanche impact area;

[0105] Among them, the geographic grid is a grid structure used to divide alpine geographical areas. The characteristics of each area (avalanche flow velocity, avalanche depth) can be analyzed separately, thereby accurately capturing regional changes.

[0106] For the avalanche affected area, the avalanche hazard level is set according to the range in which the avalanche flow velocity exceeds the flow velocity threshold, and the range in which the avalanche depth exceeds the depth threshold;

[0107] The avalanche risk levels include a low risk level, a medium risk level and a high risk level.

[0108] Among them, in this embodiment, the velocity threshold is set to 1.5m / s, and the depth threshold is set to 0.5m. The low risk level is a yellow warning, indicating that the possibility of avalanche impact is low, the velocity does not exceed the velocity threshold, and the depth does not exceed the depth threshold; the medium risk level is an orange warning, indicating that the impact possibility is medium, the velocity exceeds the velocity threshold within 1.5m / s, and the depth does not exceed the depth threshold within 0.5m; the high risk level is a red warning, indicating that the impact possibility is large, the velocity exceeds the velocity threshold by more than 1.5m / s, and the depth exceeds the depth threshold by more than 0.5m, and emergency measures must be initiated. As shown in Table 4, the risk levels corresponding to different geographic grids are given. By setting the velocity and depth thresholds and dividing them according to the grid area, it is possible to effectively optimize resource allocation, thereby achieving an accurate assessment of the avalanche impact area. In this mode, the specific location of each grid can quickly determine its risk level, thereby improving the accuracy of disaster warning. At the same time, combined with dynamic monitoring of flow velocity and depth, real-time capture of changes can enable the system to flexibly adapt to real-time changes in avalanche conditions, improve the real-time and accuracy of early warnings, and provide stronger protection for responding to avalanche disasters.

[0109] Table 4 Avalanche hazard level results

[0110] Geographic Grid Exceeding flow rate threshold (m / s) Exceeding the depth threshold (m) Risk Level Risk Color (0，0) 0 0 Low risk Yellow Warning (0，1) 0.8 0.2 Medium risk Orange warning (0，2) 1.8 0.8 High risk Red Alert (1，0) 1.2 0.2 High risk Red Alert

[0111] refer to Figure 1 The dangerous weather prediction model update module in the system is used to obtain the latest dangerous weather data, automatically perform data analysis and update the avalanche determination model. The model that has been continuously updated and optimized can not only be gradually optimized and applicable to the current real-time situation of avalanches, but also provide reference and reference for other weather-related disaster predictions.

[0112] To further verify the actual application effect of the dangerous weather intelligent early warning system based on data analysis in area A based on Example 1, Example 2 is as follows:

[0113] A dangerous weather intelligent early warning system based on data analysis, comprising:

[0114] Multi-source weather data collection module, used to collect dangerous weather data including meteorological data, geological data and hydrological data;

[0115] A multi-source weather data transmission module, used to transmit the dangerous weather data to the cloud server using a communication interruption response mechanism, and to set access permissions for the dangerous weather data, and to perform hierarchical permission management on the dangerous weather data according to role identities;

[0116] A multi-source weather data processing module is used to perform data cleaning and quality proofreading on the dangerous weather data to obtain a first data set; and to filter the first data set by using a Pearson correlation coefficient to obtain weather characteristic data;

[0117] An avalanche determination model establishment module is used to perform data enhancement on the weather characteristic data using a GAN model to obtain a second data set; establish an avalanche determination model, identify the second data set, and output an avalanche occurrence probability and an avalanche confidence interval;

[0118] The warning generation module calculates the initial flow velocity using an avalanche physics model according to the dangerous weather data if the avalanche occurrence probability and the avalanche confidence interval meet the first condition, and performs path simulation using a shallow water model according to the initial flow velocity to generate an avalanche impact area; generates an avalanche hazard level according to the avalanche impact area and issues warning information;

[0119] The dangerous weather prediction model updating module is used to obtain the latest dangerous weather data, and automatically perform data analysis and update the avalanche determination model.

[0120] Specifically, when the present invention and the original system in area A are used, the specific comparative data of various key indicators are shown in Table 5. It can be seen that the present invention significantly enhances the avalanche prediction capability, especially in reducing missed reports and false alarms, and effectively improves the accuracy of the system. In addition, the transmission success rate is also improved to a certain extent. The data transmission mechanism ensures the transmission success rate in extreme cases, thereby improving the timeliness of warning information.

[0121] Table 5 Comparison data of key indicators

[0122] Key Metrics Prediction accuracy False negative rate False Positive Rate Average time from data collection to warning release Data transmission success rate Original system 78.5％ 15.6％ 12.3％ 32 minutes 86.0％ The present invention 84.5％ 7.6％ 9.3％ 24 minutes 98.5％

[0123] Furthermore, if the avalanche hazard level of a geographic grid in region A is a high risk level, the improved Dijkstra algorithm can be used to generate a minimum risk path plan for the traffic route.

[0124] Specifically, the nodes in the high-risk area are marked as obstacle nodes, so that the Dijkstra algorithm will prioritize avoiding these nodes for path planning. According to different geographical grids, the traffic information of the corresponding area is extracted, and the data are quantitatively described and weighted summed in combination with factors such as the probability of avalanche occurrence, the weight of the avalanche-affected area, and the traffic capacity to calculate the path weights of different grids. At the same time, through the real-time collection of weather, geological and traffic data, the value of the risk factor is dynamically updated to ensure that the path planning can adapt to environmental changes in a timely manner to improve the accuracy and safety of the planning.

[0125] In summary, through the multi-source weather data acquisition module and the multi-source weather data transmission module, the system can obtain and integrate dangerous weather data, and ensure data integrity and real-time performance with the help of communication interruption response mechanism and authority management, providing a reliable basis for analysis and decision-making. Then, the multi-source weather data processing module uses data cleaning and proofreading to eliminate interference factors, improves data quality through cleaning and proofreading, and uses the Pearson correlation coefficient to screen key features, thereby improving the efficiency and accuracy of the model. Then, through the avalanche determination model establishment module, the system introduces the GAN model for data enhancement, which improves the generalization ability of the model for extreme situations. The avalanche determination model outputs the probability and confidence interval of avalanche occurrence, which improves the prediction accuracy of the model. The warning generation module accurately divides the affected area and risk level, helps resource allocation, and reduces the losses caused by disasters. Finally, the dangerous weather prediction model update module can continuously collect and analyze the latest data, dynamically optimize the avalanche determination model, adapt it to the changing environmental conditions, ensure its robustness in long-term operation, and further improve the timeliness and accuracy of the warning.

[0126] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dangerous weather intelligent early warning system based on data analysis, characterized in that: include: Multi-source weather data collection module, used to collect dangerous weather data including meteorological data, geological data and hydrological data; A multi-source weather data transmission module, used to transmit the dangerous weather data to the cloud server using a communication interruption response mechanism, and to set access permissions for the dangerous weather data, and to perform hierarchical permission management on the dangerous weather data according to role identities; A multi-source weather data processing module is used to perform data cleaning and quality proofreading on the dangerous weather data to obtain a first data set; and to filter the first data set by using a Pearson correlation coefficient to obtain weather characteristic data; An avalanche determination model establishment module is used to perform data enhancement on the weather characteristic data using a GAN model to obtain a second data set; establish an avalanche determination model, identify the second data set, and output an avalanche occurrence probability and an avalanche confidence interval; The warning generation module calculates the initial flow velocity using an avalanche physics model according to the dangerous weather data if the avalanche occurrence probability and the avalanche confidence interval meet the first condition, and performs path simulation using a shallow water model according to the initial flow velocity to generate an avalanche impact area; generates an avalanche hazard level according to the avalanche impact area and issues warning information; The dangerous weather prediction model updating module is used to obtain the latest dangerous weather data, and automatically perform data analysis and update the avalanche determination model.

2. The dangerous weather intelligent early warning system based on data analysis according to claim 1 is characterized in that: The hazardous weather data include: The meteorological data include temperature, humidity, air pressure, wind speed and direction, and precipitation; The geological data include snow depth, terrain slope, rock and soil layer distribution, and historical events of avalanche disasters; The hydrological data include snow melt rates, river flows and soil moisture.

3. The dangerous weather intelligent early warning system based on data analysis according to claim 1 is characterized in that: The communication interruption response mechanism includes: If the network is interrupted, the hazardous weather data is stored in a local storage device; If the network is restored, the local storage device synchronizes the hazardous weather data using an incremental upload method.

4. The dangerous weather intelligent early warning system based on data analysis according to claim 1 is characterized in that: The access permission setting process includes: Encrypt the hazardous weather data during transmission using an encryption protocol to generate encrypted weather data; The role identities include meteorologist, emergency personnel and administrator, and the access rights to the weather encrypted data are set according to different role identities, including: The meteorologist has the authority to view and modify the multi-source weather data acquisition module, the multi-source weather data transmission module, and the multi-source weather data processing module; The emergency personnel have the authority to view the multi-source weather data acquisition module and the warning generation module; The administrator has the authority to view and modify the multi-source weather data acquisition module, the multi-source weather data transmission module, the multi-source weather data processing module, the avalanche determination model establishment module, the warning generation module and the dangerous weather prediction model update module.

5. The dangerous weather intelligent early warning system based on data analysis according to claim 1 is characterized in that: The data cleaning and the quality checking specifically include: using time series interpolation method and spatial interpolation method to complete the dangerous weather data, and cross-validating the dangerous weather data from different data sources to obtain the first data set.

6. The dangerous weather intelligent early warning system based on data analysis according to claim 1 is characterized in that: The implementation process of the Pearson correlation coefficient includes: Calculating the Pearson correlation coefficient between each of the dangerous weather data and avalanche occurrence events; A characteristic absolute value threshold is set, and if the Pearson correlation coefficient is greater than the characteristic absolute value threshold, the dangerous weather data corresponding to the Pearson correlation coefficient is filtered as the weather characteristic data.

7. The dangerous weather intelligent early warning system based on data analysis according to claim 1 is characterized in that: The implementation process of the avalanche determination model includes: Constructing the avalanche determination model, including an input layer, a hidden layer, an output layer and a Bayesian layer; the input layer is used to receive the second data set; the hidden layer is used to extract data features of the second data set; the output layer is used to predict the avalanche occurrence probability and the avalanche confidence interval; the Bayesian layer introduces uncertainty analysis to obtain the confidence of the prediction result; The loss function is set to a weighted combination of negative log-likelihood and KL divergence; the negative log-likelihood is used to evaluate the loss value of the avalanche occurrence probability, and the KL divergence is used to evaluate the loss value of the avalanche confidence interval; Splitting the second data set into a training set and a test set, and inputting the sets into the avalanche determination model; the training set is used to train the avalanche determination model, and the test set is used to evaluate the avalanche determination model; The avalanche determination model after the test is saved.

8. The dangerous weather intelligent early warning system based on data analysis according to claim 1 is characterized in that: The first condition includes: the minimum value of the avalanche confidence interval is greater than a preset confidence threshold, the width of the avalanche confidence interval is less than a preset interval width, and the avalanche occurrence probability is greater than a preset probability threshold.

9. The dangerous weather intelligent early warning system based on data analysis according to claim 1 is characterized in that: The generation process of the avalanche hazard level includes: Obtain each geographic grid, if the avalanche flow velocity corresponding to the geographic grid exceeds a flow velocity threshold, and the avalanche depth exceeds a depth threshold, then the geographic grid is the avalanche impact area; For the avalanche affected area, the avalanche hazard level is set according to the range in which the avalanche flow velocity exceeds the flow velocity threshold, and the range in which the avalanche depth exceeds the depth threshold; The avalanche risk levels include a low risk level, a medium risk level and a high risk level.

Images