Meteorological disaster early warning response method and system based on meteorological early warning aircraft

Abstract

The invention discloses a meteorological disaster early warning response method and system based on a meteorological early warning aircraft, and the system comprises a data collection module which is used for constructing a database containing the real-time meteorological data, the terrain data, the historical disaster data and the early warning aircraft operation data; the multi-disaster terrain adaptation analysis module is used for analyzing waterlogging and gale comprehensive risks by using a synchronous superposition model in plains and analyzing landslide and thunder-gale chain risks by using a time sequence chain model in mountainous areas through a double-branch fusion neural network; the early warning response strategy generation module is used for generating differentiation strategies according to regional risks and terrains; and the terminal interaction module pushes information and receives feedback. The system dynamically generates protection guidance for different geographic features such as a plain high-risk area and a mountainous area high-risk area through multi-source meteorological data integration and analysis, and improves the timeliness and coverage of early warning information.

Description

Technical Field

[0001] The present invention relates to the technical field of early warning devices, and in particular to a meteorological disaster early warning response method and system based on a meteorological early warning aircraft. Background Art

[0002] Existing meteorological disaster warning and response technology primarily relies on a preset fixed threshold trigger mechanism. This mechanism sets critical values ​​for rainfall, wind speed, and other factors for a single disaster type (such as heavy rainfall or lightning). When the monitored data reaches the threshold, a warning message is issued via a meteorological early warning system. This system relies on dedicated meteorological networks, the internet, and other networks, connecting to provincial warning information dissemination systems. Upon receiving standardized warning signals, these signals are displayed on terminals through text, audio, and visual alarms. Furthermore, a hierarchical management model is employed: the provincial platform uniformly issues warning rules, while municipal and county-level platforms are responsible for push and review of notifications to terminals within their jurisdiction. Local users then provide feedback on the status of these notifications via their terminals.

[0003] Existing meteorological disaster warning and response technologies suffer from several significant flaws. In terms of risk assessment, relying solely on fixed threshold triggers for a single hazard fails to capture the compound risks of multiple overlapping hazards. For example, when heavy rainfall and strong winds occur simultaneously in plain areas, the amplifying effect of their synergistic impact is not considered, leading to an underestimation of the actual damage. In mountainous areas, ignoring the chain-like evolution of disasters makes it difficult to predict the temporal correlation risk—for example, early rainfall triggering later landslides, or lightning foreshadowing subsequent strong winds—resulting in delayed warnings of secondary disasters.

[0004] Therefore, it is necessary to improve the meteorological disaster warning response method and system based on meteorological early warning aircraft in the prior art to solve the above problems. Summary of the Invention

[0005] The present invention overcomes the shortcomings of the existing technology and provides a meteorological disaster warning response method and system based on a meteorological early warning aircraft, aiming to solve the problem that the existing technology uses a single fixed threshold trigger mechanism and a unified warning strategy, which cannot adapt to the chain risks of multiple disasters.

[0006] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a meteorological disaster early warning and response system based on a meteorological early warning aircraft, comprising: The data acquisition module is used to acquire and pre-process real-time meteorological monitoring data, terrain data, historical disaster data, and early warning aircraft operation data to build a call response system database; A multi-hazard terrain adaptation analysis module, connected to the data acquisition module, is used to extract meteorological and terrain characteristics based on preprocessed data, divide the region into plains and mountainous areas based on terrain type, perform risk analysis using a dual-branch fusion neural network model, and output risk levels for each region and hazard type; The dual-branch fusion neural network includes: a synchronous superposition sub-network in the plains and a temporal chain sub-network in the mountainous areas; The synchronous superposition sub-network in the plains integrates the waterlogging and high wind disaster factors to output the comprehensive waterlogging-high wind risk level; the temporal chain sub-network in the mountainous area outputs the landslide risk level and the lightning-high wind chain risk level based on the time attenuation factor and chain conduction probability; An early warning response strategy generation module, connected to the multi-hazard terrain adaptation analysis module, is used to generate differentiated targeted early warning response strategies based on the risk levels and terrain characteristics of plains and mountainous areas; The terminal interaction module is connected to the warning response strategy generation module and is used to push warning information to the target object and receive feedback.

[0007] In a preferred embodiment of the present invention, the plain synchronous overlay sub-network quantifies the combined risk level of waterlogging and high winds through the following steps: Input a feature vector containing meteorological features, terrain features, and historical correlation features; The feature extraction layer extracts waterlogging features and strong wind features respectively through parallel convolution modules; The synchronous association layer introduces a self-attention mechanism to calculate the synchronous association weights of the two sets of features and generate the synchronous occurrence coefficient; The risk output layer uses a fully connected network to fuse features, combines real-time features with the mean similarity of similar case sets to modify the basic risk index, and obtains a simultaneous superposition risk index for multiple disasters in the plains. The simultaneous superposition risk index of multiple disasters in the plain is quantified into a comprehensive risk level of waterlogging and strong winds ranging from 0 to 5 through mapping rules.

[0008] In a preferred embodiment of the present invention, the mountainous area temporal chain sub-network quantifies the landslide risk level and the lightning-high wind chain risk level through the following steps: Input feature vectors containing landslide-related features, lightning and gale-related features, and time series features; The time series feature extraction layer uses the LSTM network to process precipitation time series data and introduces the time decay coefficient to output the basic landslide risk index in mountainous areas; The temporal feature extraction layer uses the GRU network to learn the temporal association between lightning and strong winds and outputs the chain conduction probability and time delay factor; The risk output layer integrates real-time features and the mean similarity with similar case sets to calculate the mountain landslide time series chain risk index and the mountain lightning and gale chain risk index respectively. The landslide temporal chain risk index and the lightning and gale chain risk index in mountainous areas are quantified into landslide risk levels and lightning and gale chain risk levels of 0-5 respectively through mapping rules.

[0009] In a preferred embodiment of the present invention, the differentiated targeted early warning response strategy of the early warning response strategy generation module includes: Plain areas are divided into core, key, and general layers according to risk levels and spatial attributes, with differentiated warning content and push frequencies. In mountainous areas, warning targets are divided according to the stage of disaster evolution, and warning content is updated according to different stages.

[0010] In a preferred embodiment of the present invention, the call response system database includes: Real-time meteorological data table, storing meteorological monitoring data; The terrain spatial table stores spatial data on terrain type, slope, and geological disaster risk points. It uses a grid index structure to store terrain data and performs differentiated associative storage for plains and mountainous areas. A historical disaster case table stores the following information: the start and end time of the disaster, the grid ID set of the affected area, the real-time monitoring sequence of the meteorological elements that caused the disaster, terrain characteristic parameters, and a description of the disaster losses; cases are stored by disaster type and terrain type; The early warning aircraft operation status table stores the terminal online status and user operation records.

[0011] In a preferred embodiment of the present invention, the extraction of terrain features includes: performing spatial overlay analysis in plain areas through the GDAL library to extract the proportion of low-lying areas, distance from rivers, and river network density; using buffer zone analysis in mountainous areas to generate slope raster data to extract slope, density of geological disaster risk points, and altitude difference.

[0012] In a preferred embodiment of the present invention, the steps of constructing a similar case set include: Construct a feature-disaster association database classified by disaster type to store the mapping relationship between feature vectors of historical disaster cases and disaster results; The combination of meteorological and topographic features extracted in real time is converted into a multidimensional feature vector, and the Euclidean distance algorithm is used to calculate its similarity with the feature vectors of historical disaster cases; Filter the collection of historical disaster cases whose similarity exceeds the threshold and calculate the probability of disaster occurrence; Assign higher similarity calculation weights to recent cases, and strengthen the reflection of new disaster patterns through distance calculation coefficient adjustment; The mean similarity between real-time features and similar case sets is input into the synchronous overlay subnetwork in the plains and the temporal chain subnetwork in the mountainous areas to revise the comprehensive risk index and chain risk index.

[0013] The present invention provides a meteorological disaster early warning response method based on a meteorological early warning aircraft, comprising the steps of: S1. Collect and pre-process raw data related to meteorological disasters and build a response system database; the raw data includes: real-time meteorological monitoring data, terrain data, historical disaster data, and early warning aircraft operation data; S2. Extract meteorological and topographic features from preprocessed data, integrate historical disaster data, and construct a dual-branch fusion neural network model to conduct multi-hazard risk linkage analysis. This model assesses risks in plains and mountainous areas separately, and outputs risk levels for each region and each hazard type. S3. Generate targeted early warning response strategies based on the risk levels and terrain characteristics of different areas; S4. According to the characteristics and risk level of the warning object, push warning response information and obtain feedback.

[0014] In a preferred embodiment of the present invention, the preprocessing in step S1 includes data cleaning, data standardization and spatiotemporal alignment processing; Data cleaning filters outliers based on physical thresholds and statistical rules, and uses linear interpolation or neighborhood mean method to fill in short-term missing data; Data standardization unifies data formats and units, and performs coordinate conversion on spatial data; Spatiotemporal alignment matches real-time meteorological monitoring data and terrain data to spatial grids, assigns a unique ID to each spatiotemporal grid, and unifies the time granularity.

[0015] In a preferred embodiment of the present invention, the training steps of the dual-branch fusion neural network model are: The training set is constructed using historical data from the call response system database. The training set contains feature vector-risk level pairs for synchronous disaster cases in plains and chain disaster cases in mountainous areas. Using the mean square error loss function, the losses are calculated for the outputs of the synchronous superposition sub-network in the plains and the temporal chain sub-network in the mountainous areas, and then the weighted summation is performed. The Adam optimizer is used to optimize the model parameters.

[0016] The present invention solves the defects existing in the background technology and has the following beneficial effects: (1) The multi-hazard terrain adaptation analysis module of the present invention adopts a dual-branch fusion neural network architecture, and designs a synchronous superposition model and a time series chain model respectively according to the terrain differences between plains and mountainous areas. The plain model integrates the disaster-causing factors of waterlogging and strong winds to analyze the real-time comprehensive risk, and the mountain model analyzes the disaster chain evolution risk based on the time attenuation factor and chain conduction probability. It can accurately capture the risk characteristics of multiple disasters under different terrains. The synchronous superposition of the plains takes into account the immediate synergistic effect of disasters, and the time series chain of the mountainous areas focuses on the evolution law of disasters, which directly makes the output regional and disaster risk levels more in line with the actual disaster scenario.

[0017] (2) This invention adopts a terrain-driven differentiated targeting strategy. Plain areas are divided into core layers, key layers, and general layers according to risk levels and spatial attributes, and different warning contents and push frequencies are matched. Mountainous areas are pushed monitoring prompts, evacuation instructions, and equipment protection instructions according to the disaster evolution stage. At the same time, backup channels are activated for signal blind spots. This strategy fully combines the risk characteristics and terrain features of different regions. The layered push in the plains ensures that high-risk areas receive more intensive and critical information. The phased push in the mountainous areas fits the rhythm of disaster evolution. The backup channels make up for the problem of insufficient signal coverage, thus achieving accurate push and comprehensive coverage of warning information.

[0018] (3) The present invention introduces a dynamic training set construction mechanism and assigns higher weights to recent cases, so that the model can continuously absorb the characteristics of new disaster cases and strengthen its adaptability to the laws of recent disasters, thereby enhancing the robustness of the system in the context of climate change. Compared with the limitations of traditional static models that rely on training with full historical data, the present invention greatly improves the sustainable guarantee capability of early warning accuracy and further reduces the risk of model failure caused by the evolution of disaster patterns.

[0019] (4) This invention addresses the chain-like evolution of mountain disasters, constructs a temporal chain risk model, introduces a time decay factor to quantify the continued impact of previous precipitation on landslides, and designs a chain conduction probability correlation between lightning and strong winds. This effectively addresses the problem of delayed response to secondary disasters in existing technologies. This solution can predict in advance the landslide risk that may be caused by accumulated precipitation and the strong wind risk indicated by lightning activity. This advances the mountain disaster warning window by several hours, buying critical time for evacuating personnel from high-risk areas and protecting key facilities, further reducing losses caused by the chain conduction of disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive efforts. Figure 1 This is a module diagram of a meteorological disaster warning and response system according to a preferred embodiment of the present invention; Figure 2 This is a flow chart of a meteorological disaster warning response method according to a preferred embodiment of the present invention; Figure 3 This is a diagram of the risk linkage analysis model architecture of a preferred embodiment of the present invention; Figure 4 is an early warning strategy diagram of a preferred embodiment of the present invention; Figure 5It is a simplified structural diagram of a dual-branch fusion neural network model according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0023] This application targets areas with both plains and mountainous terrains, focusing on the core scenario of multi-disaster linkage in meteorological disaster prevention. In the early stage of the promotion of meteorological early warning aircraft, the existing technology adopted a single fixed threshold trigger mechanism and a unified early warning strategy, which was difficult to adapt to the chain risks formed by the superposition of multiple disasters. For example, heavy rainfall in mountainous areas may cause landslides and form compound disasters with short-term strong winds. At the same time, the early warning strategy of the existing technology does not take into account the terrain differences between plains and mountainous areas, and it is difficult to meet the differentiated needs of large-scale rapid response in plains and delayed defense against secondary disasters in mountainous areas, resulting in insufficient early warning accuracy and timeliness.

[0024] The concept of this application stems from a deep integration of meteorological disaster patterns and terrain characteristics: Based on historical data accumulated from the operation of meteorological early warning aircraft, it is found that multiple disaster types often occur simultaneously in plain areas, and risks have an immediate superposition; disasters in mountainous areas exhibit temporal chain characteristics. Based on this, we break away from the conventional thinking of a single threshold and a unified strategy and propose a differentiated response mechanism for different terrains: a risk superposition model is used in plain areas to quantify the real-time comprehensive risk of multiple disaster types, and a dynamic time series model is used in mountainous areas to predict the evolution trend of disaster chains, while optimizing the scenario adaptability of early warning information push. By integrating real-time meteorological monitoring data, historical disaster cases, and terrain characteristics, we can achieve precise triggering and hierarchical push of multiple disaster types, thereby improving the disaster response capabilities of grassroots units.

[0025] like Figure 1 As shown, a meteorological disaster warning and response system based on a meteorological early warning aircraft includes the following modules: The data acquisition module is used to obtain and pre-process raw data related to meteorological disasters. The raw data is stored in the call response system database. The raw data includes: real-time meteorological monitoring data, terrain data, historical disaster data and early warning aircraft operation data; A multi-hazard terrain adaptation analysis module, connected to the data acquisition module, is used to extract meteorological and terrain characteristics based on preprocessed data, divide the region into plains and mountainous areas based on terrain type, perform risk analysis using a dual-branch fusion neural network model, and output risk levels for each region and hazard type; The plain branch uses a synchronous superposition model to integrate waterlogging and high wind disaster factors to analyze the real-time comprehensive risk level of multiple disasters; the mountain branch uses a time series chain model to analyze the disaster chain evolution risk level based on the time attenuation factor and chain transmission probability. The early warning and response strategy generation module is connected to the multi-hazard terrain adaptation analysis module to generate differentiated targeted early warning and response strategies based on the risk level and terrain characteristics of plains and mountainous areas; The terminal interaction module is connected to the early warning response strategy generation module and is used to push early warning information to the target object and receive feedback.

[0026] In the data acquisition module, a call response system database is created and its storage structure and data specifications are defined. The call response system database includes: real-time meteorological data table, terrain space table, historical disaster case table, and early warning aircraft operation status table; The response system database is a relational database used to centrally store and manage meteorological disaster warning response data. It is built on a database management system and supports high-concurrency reading and writing as well as spatial data storage. Real-time meteorological data table: a structured table used to store minute-level / hourly dynamic meteorological monitoring data, including fields such as monitoring point ID, timestamp, rainfall, wind speed, wind direction, humidity, temperature, radar echo intensity, and satellite cloud image data; Topographic spatial table: uses a table structure designed by spatial database extension to store terrain data containing geometric information, such as plain / mountainous area division, boundary information, terrain slope, altitude, geological disaster risk points, land use type, micro-topographic characteristics, etc. Historical disaster case table: This table stores details of meteorological disaster events over the past 10 years, including the disaster ID, occurrence time, affected area, disaster type, extreme values ​​of meteorological factors that caused the disaster, loss statistics, and associated past call records and equipment failure records. Early warning aircraft operation status table: a table that records the real-time status of the meteorological early warning aircraft terminal (township version / building version / enterprise version), including terminal ID, timestamp, online / offline status, signal strength, battery power, and grassroots user operation records.

[0027] For example, the following table is the table structure design of the real-time weather data table:

[0028] The other tables also have similar table structure design, which will not be elaborated here.

[0029] The multi-hazard terrain adaptation analysis module is built based on a distributed computing framework. Through the collaborative operation of feature extraction engine, association rule miner and risk calculation model, it realizes the quantitative output of multi-hazard risk levels.

[0030] Meteorological feature extraction uses a sliding window algorithm, OpenCV library parsing, and DBSCAN clustering algorithm to obtain time series features such as hourly rainfall intensity, radar echo-related features, and ground lightning density. Terrain feature extraction uses the GDAL library, spatial overlay analysis, and buffer zone analysis to obtain raster data on mountain grid slope, the proportion of low-lying areas in plain grids, and distance to rivers. Feature vector construction associates meteorological and terrain features according to grid UUIDs to form a 128-dimensional feature vector, which is normalized and compressed to the [0,1] interval and then stored in the Milvus feature vector database.

[0031] The multi-hazard terrain adaptation analysis module requires the construction of a feature-disaster association database. This database is a structured database table that stores the corresponding relationships between meteorological and terrain feature combinations and historical disaster events. It is used to mine the implicit laws of “feature combination → disaster occurrence”. A feature combination refers to a collection of multiple meteorological and topographic features. Meteorological features include rainfall (magnitude / duration), wind speed, humidity, and radar echo characteristics. Topographic features include terrain type (plain / mountainous), slope, altitude, geological attributes, and land use type. The association database table structure is designed to construct a feature combination table and a disaster event association table. The feature combination table stores the unique identifier of a feature combination (the combination ID), as well as the meteorological and terrain features that make up the combination, and associates it with the corresponding disaster event through the combination ID. The disaster event association table stores the combination ID, the associated disaster type, a qualitative description of the disaster occurrence, and the terrain range where the association rule is effective.

[0032] In terms of feature combination classification and mapping rule establishment, sub-association libraries are constructed according to disaster type classification. Sub-libraries are established for different disaster types such as heavy rainfall, waterlogging, landslides, and lightning. Each sub-library only stores feature combinations and mapping relationships related to the corresponding disaster type. At the same time, based on historical data statistical mapping relationships, by searching historical disaster cases, the frequency of occurrence of corresponding disasters after the appearance of specific feature combinations is counted to form mapping rules. The rule content covers the composition of feature combinations, associated disaster types, description of the degree of association, and applicable terrain range.

[0033] In the dynamic maintenance mechanism of the association database, regular updates are implemented. Every quarter, the mapping relationship between feature combinations and disasters is supplemented based on new disaster cases. If the degree of correlation between a feature combination and disaster in a new case is inconsistent with the records in the database, the association rules will be updated. In addition, professional teams in the fields of meteorology and geology evaluate the mapping relationships in the association database every year, eliminate rules that have become invalid due to environmental changes such as terrain transformation and changes in land use types, and add new association rules that conform to current disaster laws.

[0034] The early warning response strategy generation module is built based on a rule engine and spatial analysis services. It dynamically generates targeted early warning strategies based on risk levels and terrain characteristics, including specific configurations of warning objects, content, channels, and frequency.

[0035] Plain areas: Based on risk levels and spatial queries, the grid sets of the core layer and key layer are screened. The population data within the grids are spatially linked with important POIs to generate a list of warning targets, including the name, address, and contact information of the person in charge. Mountainous areas: Divide warning areas by disaster stage, use buffer zone analysis to generate warning polygons within a 500-meter radius around landslide risk points, and filter key objects such as forest management stations and communication base stations through attribute queries, and store them in the warning object relationship table.

[0036] The push strategy configurator automatically selects push channels based on the signal coverage status of the warning target: 4G / 5G signal coverage areas enable early warning aircraft terminals, mobile phone text messages, and WeChat group robots; signal blind areas enable emergency broadcasts and the gong team dispatch system.

[0037] The terminal interaction module realizes two-way communication between the meteorological early warning terminal and the system platform through hardware interface adaptation and software protocol encapsulation, and supports early warning information push, status monitoring and user feedback collection.

[0038] Exemplary methods: like Figure 2 As shown, a meteorological disaster warning response method based on a meteorological early warning aircraft includes the following steps: S1. Collect and pre-process raw data related to meteorological disasters and build a response system database; the raw data includes: real-time meteorological monitoring data, terrain data, historical disaster data, and early warning aircraft operation data; S2. Extract meteorological and topographic features from preprocessed data, integrate historical disaster data, and construct a dual-branch fusion neural network model to conduct multi-hazard risk linkage analysis. This model assesses risks in plains and mountainous areas separately, and outputs risk levels for each region and each hazard type. S3. Generate targeted early warning response strategies based on the risk levels and terrain characteristics of different areas; S4. Push warning response information and obtain feedback based on the characteristics and risk level of the warning object; In step S1, real-time meteorological monitoring data, terrain data, historical disaster data and early warning aircraft operation data are obtained through multiple channels, and after pre-processing operations such as cleaning and standardization, they are stored in the database to form a full-dimensional basic data resource covering meteorological elements, geographic space, historical laws and equipment status, providing unified and reliable data input for the meteorological disaster early warning response system.

[0039] Among them, raw data refers to various types of initial information related to meteorological disaster warning analysis, including real-time meteorological monitoring data, which is used to reflect the current meteorological conditions; terrain data, which is used to characterize geographic spatial characteristics; historical disaster data, which is used to record the patterns of past disasters; early warning aircraft operation data, which is used to reflect the status of early warning equipment; the call response system database refers to the structured, secure and controllable data set formed by storing the pre-processed data according to specifications, which provides reliable data support for subsequent meteorological disaster risk analysis.

[0040] Raw data collection includes: Real-time meteorological monitoring data is collected through automatic weather stations in towns and villages to collect ground meteorological elements, including: rainfall, wind speed, wind direction, humidity, and temperature; radar echo intensity is obtained through radar stations; and satellite cloud map remote sensing data is received through meteorological satellites.

[0041] Topographic data is obtained by accessing the regional GIS map database and extracting spatial attribute data of plains and mountainous areas, including: boundary information, terrain slope, altitude, and geological disaster risk points; on this basis, high-resolution remote sensing image data is integrated to dynamically update micro-topographic features such as valleys and cliffs.

[0042] GIS map database is a database that stores geographic spatial information, including spatial data such as topography, administrative divisions, and geological disaster risk points, providing a basis for obtaining terrain data.

[0043] Historical disaster data is retrieved from the TianQing platform, including meteorological disaster cases in the past 10 years, including: disaster type, occurrence time, impact range, and degree of loss; at the same time, the historical operation logs of the early warning aircraft are retrieved, including: past early warning response records and equipment failure records.

[0044] The TianQing platform is a meteorological big data cloud platform that centrally stores and manages the province's meteorological observation, forecast, warning, and historical disaster case data, and is the core source of historical disaster data.

[0045] The operation data of the early warning aircraft is uploaded in real time by the township version, building version, and enterprise version early warning aircraft terminals, including equipment status information such as online / offline status, signal strength, battery power, and operation records of grassroots users.

[0046] Data preprocessing includes: data cleaning, data standardization and spatiotemporal alignment; Data cleaning removes outliers by filtering unreasonable data based on physical thresholds and statistical rules, including negative values ​​in rainfall monitoring and data with wind speeds exceeding physical extremes. Short-term missing data is supplemented by linear interpolation or neighborhood mean method to ensure data continuity. Image data such as radar echoes and satellite cloud images are denoised. Long-term missing data in radar blind spots in mountainous areas are spatially interpolated. Data standardization unifies data formats and units, performs coordinate conversion on spatial data, and achieves spatial matching between meteorological data and terrain data; and performs structured coding on text feedback data; Spatiotemporal alignment matches real-time meteorological monitoring data with terrain data to a spatial grid ranging from 5km×5km to 10km×10km. The data from discrete monitoring points are mapped to the grid center using the nearest neighbor method. Terrain units across grids are grid-cut to ensure unique terrain attributes within each grid. The real-time data and historical data are aligned to a unified time granularity (minutes) to ensure consistency in the spatiotemporal attributes of the data in subsequent analysis, avoiding analytical errors caused by spatiotemporal misalignment. A unique ID is assigned to each spatiotemporal grid to enable association of meteorological data, terrain data, and historical disaster cases within the same grid. The pre-processed data is stored in the call response system database, forming a multi-table associated, secure and controllable database.

[0047] like Figure 3 As shown, in step S2, meteorological and topographical features are extracted from the preprocessed data, historical disaster data are integrated, a multi-hazard risk linkage analysis model is constructed, and the risk level is output; Among them, meteorological characteristics refer to the core indicators extracted from pre-processed real-time meteorological monitoring data, which are closely related to the occurrence of meteorological disasters; terrain characteristics refer to the geographical attributes extracted from terrain data, which have a significant impact on the formation and development of disasters; historical disaster data fusion refers to the correlation analysis of historical disaster cases with real-time extracted characteristics, and the exploration of the inherent laws of characteristics and disaster occurrence; multi-disaster risk linkage analysis model refers to a mathematical model constructed based on the extracted characteristics and historical laws, which can quantify the risk of superposition or chain occurrence of multiple disasters; risk level is the output of the model, which is a grading result used to characterize the possibility of disaster occurrence and the degree of impact, and includes temporal and spatial attributes.

[0048] Multi-dimensional feature extraction is used to extract core indicators related to disasters; Based on different disaster types and terrain differences, key features are accurately extracted to provide input variables for model construction: Meteorological feature extraction: Heavy rainfall related: Extract hourly rainfall intensity, 3-hour accumulated rainfall, 6-hour accumulated rainfall, radar echo strength, and radar echo top height. A radar echo strength of ≥45dBZ indicates heavy rainfall, and the radar echo top height reflects the height of convection development. Lightning related: Extract ground-to-ground lightning frequency, ground-to-ground lightning density, and thunderstorm movement speed; Strong wind related: extract 10-minute average wind speed, maximum wind speed, and wind speed duration; Feature processing: Perform time sliding window calculation on the extracted meteorological features, calculate the average rainfall intensity in the past hour, and enhance the timeliness of the features.

[0049] The key to extracting meteorological characteristics related to heavy rainfall, lightning, and strong winds is to accurately capture the unique disaster-causing factors of different disasters and achieve refined early warning of multi-disaster superposition or chain risks; heavy rainfall is the core cause of disasters such as waterlogging and landslides. Its hourly rainfall intensity and accumulated rainfall directly reflect the intensity and sustainability of precipitation, and radar echo characteristics can predict precipitation development trends in advance; lightning ground flash data can quantify the risk of lightning strikes and avoid fires or equipment damage caused by lightning strikes; the wind speed characteristics of strong winds can assess their potential to damage buildings and outdoor facilities; heavy rainfall is often accompanied by lightning and strong winds to form complex disasters, and extracting only one type of feature will lead to a one-sided risk assessment.

[0050] Terrain feature extraction: Plain areas: Extract the proportion of low-lying areas, river network density, and distance to the river. The proportion of low-lying areas is the ratio of the area of ​​low-lying areas within a grid to the total area of ​​the grid. The river network density reflects the drainage capacity. The closer the distance to the river, the higher the risk of flooding. Mountainous areas: Extract slope, slope aspect, density of geological hazard points, and altitude difference; among them, a slope ≥ 25° indicates a high risk of landslide, sunny / shady slopes affect soil moisture content, and altitude difference reflects the degree of terrain undulation; Common features: Extract population density and the number of important places within the grid to assess the potential impact of disasters.

[0051] Historical disaster data fusion and pattern mining are used to establish association rules between features and disasters. By matching the similarity between the feature combinations extracted in real time and the feature combinations of historical disaster cases, the risk probability is directly mapped, providing an objective association basis for the model. Specifically, Construct a characteristic disaster association database, integrate historical disaster data and mine patterns. By constructing a characteristic disaster association database classified by disaster type, containing complete feature combinations and disaster results, the real-time feature combinations are converted into vectors and matched with historical disaster case vectors for similarity. Similar cases are screened to statistically analyze the probability of disaster occurrence to determine the current risk and refer to its impact. At the same time, the case database is regularly updated, the weights of recent cases are adjusted, and controversial cases are handled to establish the association rules between features and disasters.

[0052] Classification by disaster type: Historical disaster cases are divided into types such as heavy rainfall, waterlogging, landslides, and lightning disasters. Each disaster type is associated with its specific meteorological and topographical characteristics, achieving a precise binding between characteristics and disaster types. Quantify the strength of the association and statistically analyze the probability of disasters occurring under specific combinations of meteorological and topographical features in historical data, so as to clarify the impact of different feature combinations on disaster occurrence.

[0053] The characteristic values ​​of historical disaster cases were normalized, and the hourly rainfall intensity was converted into a percentage relative to the historical maximum value to eliminate the calculation deviation caused by different units and ensure the comparability of real-time characteristics with historical disaster cases.

[0054] The combination of meteorological and topographic features extracted in real time is converted into a multidimensional feature vector, and historical disaster cases are converted into feature vectors of the same dimension. The Euclidean distance algorithm is used to calculate the similarity between the real-time feature vector and the feature vector of the historical disaster case. The smaller the distance, the higher the similarity. A similarity threshold or a selection number is set to filter out a collection of cases similar to the current scenario. The probability of disaster occurrence in a collection of similar cases is calculated. If the corresponding disaster occurs in more than 70% of the cases, the risk of the disaster occurring under the current feature combination is determined to be high. The degree of disaster impact of similar cases is also recorded as a reference for the current scenario risk level. The degree of disaster impact includes the depth and duration of waterlogging. In addition, new disaster cases will be added to the case library every quarter. The cases must contain complete feature combinations and disaster results to ensure the timeliness of the case library. Assigning a higher similarity calculation weight to recent cases and multiplying a certain coefficient in the distance calculation so that emerging disaster patterns can be reflected in the matching results more quickly. Among them, emerging disaster patterns include changes in disaster-causing characteristics caused by climate change; Cases with similar feature combinations but large differences in disaster outcomes are automatically marked as “disputed cases” and manually calibrated with expert experience to avoid outliers interfering with matching accuracy.

[0055] Constructing a multi-hazard risk linkage analysis model: Differentiated models were designed for plain and mountainous areas. Based on the differences in disaster characteristics in plain and mountainous areas, a dual-branch fusion neural network model was designed to construct risk models for each region. Through the differentiated structures of the plain and mountain branches, end-to-end quantification of multi-hazard complex risks was achieved. The model inputs were meteorological, topographic, and historical feature vectors, and output was a risk level (0-5) for each region and hazard type.

[0056] The multi-hazard risk linkage analysis utilizes a dual-track analysis mechanism based on topographic differences and disaster characteristics between plains and mountainous areas. Through the differentiated design of a synchronous overlay model for plains and a time-series chain model for mountainous areas, this approach accurately quantifies the combined risks of multiple hazards. The model inputs preprocessed meteorological, topographic, and historical data, and outputs risk levels by region and hazard type, providing a basis for targeted early warning strategies.

[0057] like Figure 5 As shown in the figure, the overall architecture logic of the dual-branch fusion neural network model is: The input features are automatically routed to the corresponding branch network after the terrain classifier determines the area type: The plain regional characteristics are input into the plain synchronous superposition sub-network, and the output is the comprehensive risk level of waterlogging and strong winds; The mountainous area characteristics are input into the mountainous area time series chain sub-network, and the output is the landslide risk level and the lightning-strong wind chain risk level.

[0058] Among them, the plain areas focus on the simultaneous superposition risks of urban flooding and strong winds, and reflect the synergistic amplification effect of multiple disasters occurring simultaneously through the immediate superposition coefficient; the mountainous areas focus on the time-series chain risks of landslides and lightning, and introduce time attenuation factors to quantify the transmission impact of previous disasters on subsequent disasters, so that the model calculation logic is deeply adapted to the disaster evolution law dominated by terrain.

[0059] The steps to build the plain synchronization overlay subnetwork include: A multi-hazard simultaneous superposition risk model is adopted in plain areas. For the coordinated risk of urban flooding and strong winds, the real-time comprehensive risk of simultaneous occurrence of urban flooding and strong winds in plain areas is quantified, and the synergistic effect of the superposition of multiple disasters is considered. Urban flooding and strong winds in plain areas are often accompanied by severe convective weather, and the risks are instantly superimposed. The comprehensive risk level is quantified by integrating the disaster-causing factors and simultaneous occurrence probability of urban flooding and strong winds.

[0060] Input layer: 64-dimensional feature vector, including: Meteorological characteristics: hourly precipitation, 3-hour / 6-hour accumulated rainfall, 10-minute average wind speed, maximum wind speed, wind speed duration, radar echo intensity (feature weight is enhanced when ≥50dBZ); Topographic characteristics: proportion of low-lying areas, distance from rivers, river network density, and population density; Historical correlation features: the mean similarity between real-time features and similar case sets, i.e., the matching degree of plain synchronous disasters ; Feature extraction layer: The parallel convolution module uses two 1D convolution kernels to extract waterlogging characteristics (integrating hourly rainfall intensity, 3-hour / 6-hour cumulative rainfall, proportion of low-lying areas, distance to the river, and river network density) and high wind characteristics (10-minute average wind speed, maximum wind speed, wind speed duration, and population density). It outputs two sets of 32-dimensional feature maps, corresponding to the implicit features of the waterlogging risk index and the high wind risk index. Synchronous association layer: introduces the self-attention mechanism to calculate the synchronous association weights of two sets of features and simulate the synchronous occurrence coefficient ; Calculate the proportion of simultaneous occurrence of waterlogging and strong winds in historical data to generate a synchronous occurrence coefficient; Risk output layer: The fully connected network integrates the above features, maps them to a level of 0-5 through the Sigmoid activation function combined with linear scaling, and outputs the comprehensive risk level of the plain: Retrieve cases from the historical disaster case table that are similar to the current real-time meteorological and topographical features, calculate the mean similarity between the real-time feature vector and the similar case set, and use this as a weight to modify the basic risk index; Finally, the comprehensive risk index is obtained through the following formula: ,in, The index is a continuous value index for the simultaneous superposition of multiple disasters in plains. The value range is 0-5. The larger the value, the higher the combined risk of simultaneous occurrence of waterlogging and strong winds. They are the plain waterlogging risk index and the plain gale risk index; the plain waterlogging risk index is calculated based on hourly precipitation, 3-hour cumulative rainfall, 6-hour cumulative rainfall, proportion of low-lying areas, and distance from the river; the plain gale risk index is calculated based on 10-minute average wind speed, maximum wind speed, wind speed duration, and population density. The value range of the synchronous occurrence coefficient is 1.0-1.5, which is calculated based on the probability of synchronous occurrence of waterlogging and strong winds under real-time meteorological conditions; is the matching degree of synchronous disasters in plains, is the mean similarity between the real-time feature vector and the set of similar cases, and the higher the similarity, the closer it is to 1; Furthermore, the risk index is quantified into an integer risk level of 0-5 through mapping rules; Level 0 (No Risk): <0.5; Level 1 (low risk): 0.5 ≤ <1.5; Level 2 (medium-low risk): 1.5 ≤ <2.5; Level 3 (medium risk): 2.5 ≤ <3.5; Level 4 (high risk): 3.5 ≤ <4.5; Level 5 (extremely high risk): ≥4.5.

[0061] The final output is the comprehensive risk level of urban flooding and strong winds.

[0062] Synchronous superposition calculation is triggered when any of the following conditions are met: The radar echo intensity is ≥50dBZ (strong convective signal) and the maximum wind speed detected is ≥10m / s; In the past hour, the waterlogging risk index is greater than or equal to 2 and the gale risk index is greater than or equal to 1.

[0063] Among the conditions for triggering synchronous overlay calculation, when the matching degree between real-time features and historical synchronization cases is ≥0.7, overlay calculation is still triggered even if the radar echo intensity standard is not met.

[0064] Furthermore, the synchronous superposition characteristics of the plain are adapted through the new enhanced convective synchronous monitoring index, and the potential for the synchronous occurrence of waterlogging and strong winds is judged by the combination of radar echo top height and vertical wind shear, which is the synchronous occurrence coefficient. Provide real-time calculation basis, and automatically improve the grid with synchronous events accounting for ≥30% in the historical disaster case database. benchmark value, thereby ensuring that the superposition effect of high-risk areas is not underestimated.

[0065] The steps for constructing the mountainous temporal chain subnetwork include: A multi-hazard temporal chain risk model is used in mountainous areas. Targeted at the disaster chain evolution risk, the model predicts the chain evolution risks of "precipitation-landslide" and "lightning-strong wind" in mountainous areas, taking into account the temporal correlation and cumulative effects of disasters: mountain disasters exhibit temporal chain characteristics, with early precipitation easily triggering later landslides, and lightning activity often foreshadowing subsequent strong winds. The model introduces a time attenuation factor and a chain conduction probability, and uses the matching degree of historical chain cases as a correction term for the time attenuation factor and the conduction probability to quantify the evolution risk of the disaster chain.

[0066] Input layer: 64-dimensional feature vector, including: Landslide-related characteristics: current hourly rainfall intensity, cumulative rainfall in the past 6 / 12 / 24 hours, slope, density of geological hazard points, radar echo intensity, and the mean similarity between real-time precipitation characteristics and a collection of similar cases; Lightning-wind related characteristics: ground-to-ground lightning frequency, ground-to-ground lightning density, current wind speed, altitude difference, and the mean similarity between real-time lightning characteristics and a collection of similar cases; Temporal feature extraction layer: For landslide risk paths: LSTM networks are used to process precipitation time series data, including 6 / 12 / 24-hour accumulated rainfall. A time decay coefficient is introduced through a gating mechanism to simulate the cumulative effect of precipitation and output a basic landslide risk index in mountainous areas. The time decay coefficient is set based on the time interval between precipitation and landslides in historical landslide cases to quantify the ongoing impact of precipitation. For the lightning-high wind risk path: the GRU network is used to learn the temporal association between lightning and high winds, and the temporal attention mechanism is used to capture the pattern of high winds one hour after lightning, and output the chain conduction probability and time delay factor.

[0067] Based on the extracted time series features, two sets of fully connected layers are used to output the landslide risk level and the lightning-high wind chain risk level. The calculation logic includes: Time series chain risk index of landslides in mountainous areas: ,in, is the matching degree of mountain landslide chain, is the mean similarity between the real-time precipitation characteristics and the historical early precipitation-later landslide chain cases, It is a mountain landslide time series chain risk index with a value range of 0-5 and is a continuous value, reflecting the cumulative impact of previous precipitation on the current landslide risk; It is the basic risk index of landslide in mountainous areas, which is calculated based on the slope, density of geological hazard points, and radar echo intensity parameters; is the time attenuation coefficient, which represents the decay rate of the impact of previous precipitation over time; is the quantitative value of the radar echo intensity within the period t, where t is the time period. The three key periods of the first 6 hours, 12 hours, and 24 hours are selected to reflect the cumulative effect of precipitation of different durations.

[0068] Mountain area lightning-strong wind chain risk index: The frequency of strong winds occurring within one hour after lightning in historical disaster cases was counted as the chain transmission probability. A time delay factor was set (based on the temporal pattern of lightning and strong winds) to calculate the lightning-strong wind chain risk index in mountainous areas: ,in, The lightning-high wind chain matching degree in mountainous areas is calculated based on the similarity between real-time lightning characteristics and historical lightning-high wind chain cases; The lightning-wind chain risk index in mountainous areas has a value range of 0-5 and is a continuous value; The lightning risk index in mountainous areas is calculated based on parameters such as ground-to-ground lightning frequency and ground-to-ground lightning density; is the chain conduction probability, which is determined by the frequency of strong winds occurring within 1 hour after lightning in historical data; It is the time delay factor, which takes the value of 1.0 for 0-1 hour after the lightning occurs, 0.5 for 1-3 hours, and 0 after 3 hours, reflecting the attenuation law of temporal correlation.

[0069] The output mountain landslide time series chain risk index and mountain lightning-high wind chain risk index are both continuous values, which are mapped to 0-5 through the same mapping rules to form the landslide risk level and lightning-high wind chain risk level, respectively.

[0070] Dynamic adaptation mechanism: Furthermore, the temporal chain characteristics of mountainous areas are adapted by establishing a previous impact factor database to store the time interval distribution of mountain disaster chain cases in the past 10 years, which is the time attenuation coefficient and time delay factor The data support is provided for the dynamic adjustment of the soil moisture content. When the soil moisture content is monitored to be ≥80% (saturated state), It is automatically increased to 0.8 to shorten the decay period and strengthen the chain effect of previous precipitation.

[0071] For the dual-branch fusion neural network model, the historical data of the call response system database, including cases of synchronous disasters in plains and chain disasters in mountainous areas in the past 10 years, was used to construct a training set based on feature vector-risk level pairs; The loss function uses the mean square error (MSE) loss. The losses are calculated for the outputs of the two branches and then weighted summed. The weights of the plain and mountainous samples are allocated according to the frequency of regional disasters. The optimizer uses the Adam optimizer, assigning 1.2 times the weight to cases in the past two years to enhance the model's adaptability to recent disaster patterns. Model output: The output format is: Plain area: single value (0-5), representing the combined risk level of waterlogging and strong winds; Mountainous areas: double values ​​(0-5), representing the landslide risk level and lightning-strong wind chain risk level respectively.

[0072] In this embodiment, through the differentiated design of the synchronous superposition sub-network in the plains and the temporal chain sub-network in the mountainous areas, the risk calculation logic is precisely matched with the disaster characteristics dominated by the terrain: the plains capture the synergistic effect of the instant superposition of multiple disaster types, and the mountainous areas quantify the temporal correlation of disasters through the time attenuation factor and the chain conduction probability. Therefore, the risk level output by the dual-branch fusion neural network model not only includes the intensity of a single disaster type, but also reflects the evolution law of multiple disaster types driven by the terrain, providing a more practical quantitative basis for targeted early warning and response strategies.

[0073] like Figure 4 As shown, in step S3, a targeted early warning response strategy is generated based on the risk level and terrain characteristics. The core is to match differentiated early warning objects, content and frequency according to the differences and evolution laws of disaster risks in plains and mountainous areas, to ensure that early warning information accurately reaches high-risk areas and people, and improve response efficiency.

[0074] Targeted early warning and response plan for plain areas: Based on the simultaneous superposition risk index of multiple disasters in the plains, population density, and distribution of important places, the warning objects are divided into core layer, key layer, and general layer. Differentiated warning contents such as immediate risk avoidance instructions, prevention measures guidance, and risk prompts are designed for different levels, and the warning frequency is dynamically adjusted according to the risk level.

[0075] The core layer is ≥3.5 (high risk) and low-lying areas account for ≥50%, including low-lying residential areas, underground parking lots, and urban waterlogging points; key layers are 2.5≤ <3.5 (medium-high risk) and the distance from the river is ≤500 meters, including farmland along the river and industrial parks; the general layer is Areas with a risk of <2.5 (medium to low risk) are mainly ordinary residential areas.

[0076] The core layer warning content must include immediate evacuation instructions, clearly define the evacuation location, evacuation routes and prohibited items; the key layer warning content focuses on guidance on preventive measures, including: reinforcing outdoor billboards, suspending high-altitude operations, and checking drainage facilities; the general layer warning content is mainly risk warnings, reminding people to pay attention to subsequent warning information and close doors and windows.

[0077] when When the value is ≥3.5 and the strong convection synchronous monitoring indicator is continuously triggered, an early warning will be pushed every 15 minutes; when the value is ≤2.5 <3.5, push once every 30 minutes; when If the risk level is less than 2.5, notifications will be sent every hour. If the risk level decreases by 30% or more, the frequency will be automatically reduced to the next level.

[0078] Targeted early warning response plan for mountainous areas: According to the mountain landslide time series chain risk index and lightning-high wind chain risk index , combined with the density of geological hazard points, divide the warning objects according to the disaster evolution stage, combine the density of geological hazard points and divide the warning objects according to the disaster evolution stage, update the warning content according to the stage time sequence, and use multi-channel collaborative push to ensure information coverage in view of the difference in signal coverage.

[0079] Landslide risk dominant stage ≥3.5, the priority for early warning targets is villages within 500 meters of the landslide potential point and mountain road construction teams; lightning-strong wind chain stage ≥3.5, the warning targets are expanded to forest management stations and high-altitude communication base stations.

[0080] In the early stage of landslide risk (preliminary precipitation accumulation stage), the warning content is mainly based on monitoring prompts, and grassroots monitors are required to intensify inspections of valleys and steep cliffs; in the mid-term of landslide risk If the temperature is ≥3.5 and the soil moisture content is ≥80%, evacuation preparation instructions will be issued, and the evacuation assembly point and the list of items to be carried will be clarified; in the lightning-strong wind chain stage, the warning content will focus on equipment protection guidelines, including shutting down the power supply in the forest area, suspending outdoor operations, and reinforcing base station antennas.

[0081] In view of the differences in signal coverage in mountainous areas, multi-channel coordinated push is adopted: for areas with 4G / 5G signals, graphic and text warnings are pushed through early warning terminals, mobile phone text messages, and village-level WeChat groups; for areas with weak signals, township emergency broadcasts, gong teams, and grid workers’ home notifications are linked to ensure the “last mile” coverage of early warning information.

[0082] In step S4, pushing warning response information and obtaining feedback is a key link in realizing closed-loop management of warnings. It ensures that information reaches through multi-channel push and monitors the response effect through real-time feedback, providing a basis for dynamically adjusting the warning strategy.

[0083] Based on the regional characteristics and risk levels of the warning targets, differentiated push channels are dispatched: plain high-risk areas ≥3.5 will be prioritized through township version early warning terminal (sound and light alarm + text display), building version early warning machine (elevator screen scrolling play), mobile phone pop-up push, and synchronous linkage city emergency broadcast system; high-risk areas in mountainous areas ≥3.5 The focus is on activating village-level early warning aircraft, emergency broadcast vehicles, and drone aerial broadcasts, and additional paper early warning sheets will be delivered by grid workers.

[0084] High-risk warning information must be pushed within 5 minutes of policy generation, with a secondary push within 10 minutes for coverage areas (for those who haven't confirmed receipt). Medium- and low-risk warning information must be pushed within 15 minutes. During the push process, the channel delivery rate is monitored in real time. If the delivery rate of a particular channel falls below 70%, it will automatically switch to a backup channel (for example, if the AWACS terminal goes offline, SMS resending will be initiated immediately).

[0085] The above description is based on the ideal embodiment of the present invention. Based on the above description, relevant personnel can make various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the content of the specification and must be determined according to the scope of the claims.

Claims

1. A meteorological disaster warning and response system based on a meteorological early warning aircraft, characterized in that: Included modules: The data acquisition module is used to acquire and pre-process real-time meteorological monitoring data, terrain data, historical disaster data, and early warning aircraft operation data to build a call response system database; A multi-hazard terrain adaptation analysis module, connected to the data acquisition module, is used to extract meteorological and terrain characteristics based on preprocessed data, divide the region into plains and mountainous areas based on terrain type, perform risk analysis using a dual-branch fusion neural network model, and output risk levels for each region and hazard type; The dual-branch fusion neural network includes: a synchronous superposition sub-network in the plains and a temporal chain sub-network in the mountainous areas; The synchronous superposition sub-network in the plains integrates the waterlogging and high wind disaster factors to output the comprehensive waterlogging-high wind risk level; the temporal chain sub-network in the mountainous area outputs the landslide risk level and the lightning-high wind chain risk level based on the time attenuation factor and chain conduction probability; An early warning response strategy generation module, connected to the multi-hazard terrain adaptation analysis module, is used to generate differentiated targeted early warning response strategies based on the risk levels and terrain characteristics of plains and mountainous areas; The terminal interaction module is connected to the warning response strategy generation module and is used to push warning information to the target object and receive feedback.

2. The meteorological disaster warning and response system based on a meteorological early warning aircraft according to claim 1, characterized in that: The plain synchronous overlay subnetwork quantifies the combined risk level of waterlogging and high winds through the following steps: Input a feature vector containing meteorological features, terrain features, and historical correlation features; The feature extraction layer extracts waterlogging features and strong wind features respectively through parallel convolution modules; The synchronous association layer introduces a self-attention mechanism to calculate the synchronous association weights of the two sets of features and generate the synchronous occurrence coefficient; The risk output layer uses a fully connected network to fuse features, combines real-time features with the mean similarity of similar case sets to modify the basic risk index, and obtains a simultaneous superposition risk index for multiple disasters in the plains. The simultaneous superposition risk index of multiple disasters in the plain is quantified into a comprehensive risk level of waterlogging and strong winds ranging from 0 to 5 through mapping rules.

3. The meteorological disaster warning and response system based on a meteorological early warning aircraft according to claim 1, characterized in that: The mountainous area temporal chain subnetwork quantifies the landslide risk level and the lightning-high wind chain risk level through the following steps: Input feature vectors containing landslide-related features, lightning and gale-related features, and time series features; The time series feature extraction layer uses the LSTM network to process precipitation time series data and introduces the time decay coefficient to output the basic landslide risk index in mountainous areas; The temporal feature extraction layer uses the GRU network to learn the temporal association between lightning and strong winds and outputs the chain conduction probability and time delay factor; The risk output layer fuses the real-time features with the mean similarity of similar case sets to calculate the mountain landslide time series chain risk index and the mountain lightning and gale chain risk index respectively. The landslide temporal chain risk index and the lightning and gale chain risk index in mountainous areas are quantified into landslide risk levels and lightning and gale chain risk levels of 0-5 respectively through mapping rules.

4. The meteorological disaster warning and response system based on a meteorological early warning aircraft according to claim 1, characterized in that: The differentiated targeted early warning response strategies of the early warning response strategy generation module include: Plain areas are divided into core, key, and general layers according to risk levels and spatial attributes, with differentiated warning content and push frequencies. In mountainous areas, warning targets are divided according to the stage of disaster evolution, and warning content is updated according to different stages.

5. The meteorological disaster warning and response system based on a meteorological early warning aircraft according to claim 1, characterized in that: The call response system database includes: Real-time meteorological data table, storing meteorological monitoring data; The terrain spatial table stores spatial data on terrain type, slope, and geological disaster risk points. It uses a grid index structure to store terrain data and performs differentiated associative storage for plains and mountainous areas. A historical disaster case table stores the following information: the start and end time of the disaster, the grid ID set of the affected area, the real-time monitoring sequence of the meteorological elements that caused the disaster, terrain characteristic parameters, and a description of the disaster losses; cases are stored by disaster type and terrain type; The early warning aircraft operation status table stores the terminal online status and user operation records.

6. The meteorological disaster warning and response system based on a meteorological early warning aircraft according to claim 1, characterized in that: The extraction of terrain features includes: spatial overlay analysis in plain areas through the GDAL library to extract the proportion of low-lying areas, distance from rivers, and river network density; buffer zone analysis is used in mountainous areas to generate slope raster data to extract slope, density of geological disaster risk points, and altitude difference.

7. The meteorological disaster warning and response system based on a meteorological early warning aircraft according to claim 2, characterized in that: The steps for constructing a similar case set include: Construct a feature-disaster association database classified by disaster type to store the mapping relationship between feature vectors of historical disaster cases and disaster results; The combination of meteorological and topographic features extracted in real time is converted into a multidimensional feature vector, and the Euclidean distance algorithm is used to calculate its similarity with the feature vectors of historical disaster cases; Filter the collection of historical disaster cases whose similarity exceeds the threshold and calculate the probability of disaster occurrence; Assign higher similarity calculation weights to recent cases, and strengthen the reflection of new disaster patterns through distance calculation coefficient adjustment; The mean similarity between real-time features and similar case sets is input into the synchronous overlay subnetwork in the plains and the temporal chain subnetwork in the mountainous areas to revise the comprehensive risk index and chain risk index.

8. A meteorological disaster warning and response method based on a meteorological early warning aircraft, based on a meteorological disaster warning and response system based on a meteorological early warning aircraft according to any one of claims 1 to 7, characterized in that: Including steps: S1. Collect and pre-process raw data related to meteorological disasters and build a response system database; the raw data includes: real-time meteorological monitoring data, terrain data, historical disaster data, and early warning aircraft operation data; S2. Extract meteorological and topographic features from preprocessed data, integrate historical disaster data, and construct a dual-branch fusion neural network model to conduct multi-hazard risk linkage analysis. This model assesses risks in plains and mountainous areas separately, and outputs risk levels for each region and each hazard type. S3. Generate targeted early warning response strategies based on the risk levels and terrain characteristics of different areas; S4. According to the characteristics and risk level of the warning object, push warning response information and obtain feedback.

9. The meteorological disaster warning response method based on a meteorological early warning aircraft according to claim 8, characterized in that: The preprocessing in step S1 includes data cleaning, data standardization and spatiotemporal alignment; Data cleaning filters outliers based on physical thresholds and statistical rules, and uses linear interpolation or neighborhood mean method to fill in short-term missing data; Data standardization unifies data formats and units, and performs coordinate conversion on spatial data; Spatiotemporal alignment matches real-time meteorological monitoring data and terrain data to spatial grids, assigns a unique ID to each spatiotemporal grid, and unifies the time granularity.

10. The meteorological disaster warning response method based on a meteorological early warning aircraft according to claim 8, characterized in that: The training steps of the dual-branch fusion neural network model are: The training set is constructed using historical data from the call response system database. The training set contains feature vector-risk level pairs for synchronous disaster cases in plains and chain disaster cases in mountainous areas. Using the mean square error loss function, the losses are calculated for the outputs of the synchronous superposition sub-network in the plains and the temporal chain sub-network in the mountainous areas, and then the weighted summation is performed. The Adam optimizer is used to optimize the model parameters.

Images