Response methods and systems for meteorological risks in air ticketing

Abstract

This invention discloses a method and system for responding to meteorological risks in airline ticketing, relating to the field of artificial intelligence technology. The method includes: acquiring multi-source meteorological disaster early warning data and flight schedule data, performing dynamic spatiotemporal alignment, and identifying whether the flight's take-off and landing points or key points along the flight route fall within the spatiotemporal impact range of the meteorological disaster early warning data; if so, extracting a cross-modal composite feature set based on the spatiotemporal alignment results, inputting the composite feature set into a target disaster level prediction model, using the target disaster level prediction model to predict the impact of the meteorological disaster on the flight, outputting the disaster impact level of the flight, and, based on the disaster impact level, calling a preset parameterized refund and change rule template to generate a refund and change strategy corresponding to the disaster impact level. This invention solves the technical problems in related technologies where airlines' refund and change rule processes are rigid and inefficient, increasing service costs and reducing passenger experience when responding to meteorological disasters.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology or other related fields, and more specifically, to a method and system for responding to weather risks in air ticketing. Background Technology

[0002] As a crucial pillar of the integrated transportation system and modern service industry, air transport's operational efficiency and passenger service quality are highly dependent on the stability of the meteorological environment. However, the frequent occurrence of sudden meteorological disasters such as typhoons, heavy rain, blizzards, dense fog, thunderstorms, severe convective weather, and sandstorms has become the primary external factor leading to flight delays, cancellations, diversions, and large-scale passenger congestion.

[0003] In related technologies, airlines' main mechanisms for responding to weather disasters still rely on the traditional model of "human perception - human judgment - human decision-making - human execution," which has significant shortcomings. First, weather warning information is disconnected from the ticketing system, resulting in severe information silos. While existing aviation operation control systems can access weather disaster warnings issued by meteorological bureaus, this data is typically in the form of text announcements, graphical warning maps, or raw grid data, lacking structured interfaces and unable to be automatically linked with ticketing, flight planning, and passenger order systems. Second, flight adjustment and refund / change decisions heavily rely on human experience and lack quantitative basis. After a weather event occurs, dispatchers and operations control personnel primarily rely on personal experience to determine whether to cancel flights, merge flight segments, or initiate diversion procedures. Such judgments lack unified standards, often leading to drastically different outcomes for the same weather event across different airlines and regions due to differences in experience. More seriously, initiating refund / change rules relies entirely on manual operation: customer service personnel must verify each passenger's order to see if it meets the "cancellation due to weather" criteria, requiring passengers to upload weather screenshots, flight cancellation notices, and other supporting documents, making the review process cumbersome. Thirdly, the rigid and inefficient refund and change process deteriorates the passenger experience and significantly increases service costs. Current mainstream ticketing systems use static rule engines, with fixed conditions for refunds and changes. At the same time, refund and change applications require manual review on a case-by-case basis. During peak periods, the daily processing volume exceeds 100,000 applications, greatly increasing the pressure on customer service. The average processing time exceeds 8 minutes, far exceeding the normal 30-second standard.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method and system for responding to meteorological risks in air ticketing, at least to solve the technical problems in related technologies where airlines' refund and change rules and processes are rigid and inefficient, leading to increased service costs and reduced passenger experience when dealing with meteorological disasters.

[0006] According to one aspect of the present invention, a method for responding to meteorological risks in air ticketing is provided, comprising: acquiring multi-source meteorological disaster early warning data and flight plan data, wherein the flight plan data includes flight departure and arrival points, route key points, and expected flight time windows; dynamically aligning the meteorological disaster early warning data and the flight plan data in a unified spatiotemporal coordinate system, and identifying whether the flight departure and arrival points or route key points fall within the spatiotemporal range of the early warning impact in the meteorological disaster early warning data; and, if the flight departure and arrival points or route key points fall within the spatiotemporal range of the early warning impact, extracting a cross-modal composite feature set based on the spatiotemporal alignment result, wherein the composite feature set includes at least one of the following: disaster type and level. The system includes the following features: dynamic distance between the warning center point and key points of the airport / airway; spatiotemporal proportion of the warning coverage segment; and time overlap coefficient between flight schedule time and disaster duration. These features are input into a target disaster level prediction model, which predicts the impact of meteorological disasters on flights and outputs the disaster impact level of the flights. The target disaster level prediction model is a model pre-trained using a deep learning neural network architecture based on a self-attention mechanism and a long short-term memory network. Based on the disaster impact level, a preset parameterized refund and rebooking rule template is used to generate a refund and rebooking strategy corresponding to the disaster impact level. This strategy includes passenger notification content and customer service approval policy.

[0007] According to another aspect of the present invention, a response system for meteorological risks in air ticketing is also provided, comprising: a warning data acquisition unit, configured to acquire multi-source meteorological disaster warning data and flight plan data, wherein the flight plan data includes flight take-off and landing points, route key points, and expected flight time windows; a spatiotemporal alignment unit, configured to dynamically align the meteorological disaster warning data and the flight plan data based on a unified spatiotemporal coordinate system, and identify whether the flight take-off and landing points or route key points fall within the warning impact spatiotemporal range in the meteorological disaster warning data; and a composite feature extraction unit, configured to extract a cross-modal composite feature set based on the spatiotemporal alignment result when the flight take-off and landing points or route key points fall within the warning impact spatiotemporal range, wherein the composite feature set includes at least one of the following: disaster type. The system includes: a disaster impact level output unit, a disaster impact level output unit, and a disaster impact level output unit. The composite feature set is input into a target disaster level prediction model, which predicts the impact of meteorological disasters on flights and outputs the disaster impact level of the flights. The target disaster level prediction model is a model pre-trained based on a deep learning neural network architecture with a self-attention mechanism and a long short-term memory network. A refund / change policy generation unit is used to generate a refund / change policy corresponding to the disaster impact level by calling a preset parameterized refund / change rule template. The refund / change policy includes passenger notification content and customer service approval strategy.

[0008] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform any of the above-described methods for responding to air ticketing weather risks.

[0009] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the air ticketing weather risk response method described above.

[0010] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the air ticketing weather risk response method described in any one of the above embodiments.

[0011] The aforementioned publicly available functionalities enable precise matching of real-time weather warning data with flight operation plans, automatic identification of affected flights and passenger orders, and subsequent output of quantified risk levels through a constructed dynamic impact level prediction model. Finally, based on the level of disaster impact, a preset parameterized refund and rebooking rule template is invoked to generate a refund and rebooking strategy corresponding to the level of disaster impact. This shifts refund and rebooking decisions from fixed rules to intelligent, tiered responses based on the actual degree of impact. Warning information can be pushed to passengers in real time, and customer service backends are simultaneously notified to approve strategies in advance, significantly shortening the response cycle, greatly reducing manual intervention, improving the efficiency of refund and rebooking processing, effectively reducing service costs, and transforming the passenger experience from passive acceptance to proactive awareness and convenient operation. This solves the technical problems of rigid and inefficient refund and rebooking rules and processes, increased service costs, and reduced passenger experience in airlines' response to weather disasters. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0013] Figure 1 This is a flowchart of an optional response method for weather risks in air ticketing according to an embodiment of the present invention;

[0014] Figure 2 This is a schematic diagram of an optional intelligent response system for air ticketing weather risks according to an embodiment of the present invention;

[0015] Figure 3 This is a schematic diagram of the workflow of an optional data integration module according to an embodiment of the present invention;

[0016] Figure 4 This is a schematic diagram of the workflow of an optional intelligent analysis module according to an embodiment of the present invention;

[0017] Figure 5 This is a schematic diagram of an optional data storage module according to an embodiment of the present invention;

[0018] Figure 6 This is a schematic diagram of an optional visualization module according to an embodiment of the present invention;

[0019] Figure 7 This is a schematic diagram of an optional air ticketing weather risk response device according to an embodiment of the present invention;

[0020] Figure 8 This is a hardware structure block diagram of an electronic device (or mobile device) for a weather risk response method for air ticketing according to an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] To facilitate understanding of the present invention by those skilled in the art, some terms or nouns involved in the various embodiments of the present invention are explained below:

[0024] Long Short-Term Memory (LSTM) networks are a type of recurrent neural network structure with time-series memory capabilities, used to process the temporal dependencies of historical weather and flight data.

[0025] The Transformer model, or Transformer for short, is a deep learning architecture based on a self-attention mechanism, used to model the global dependencies between weather warning data and flight schedule data in parallel.

[0026] Graph Neural Network (GNN) is a neural network model that learns and infers directly on graph-structured data. It is used to fuse meteorological, flight, and rule graph information to output flight cancellation probabilities and early warnings.

[0027] BERT, or Bidirectional Encoder Representations from Transformers, is a pre-trained language representation model based on a bidirectional Transformer encoder. It is used for deep parsing of refund and change policy texts and automatically extracting rule-based semantic relationships.

[0028] Knowledge Graph (KG) is a structured knowledge base that represents entities and their semantic relationships using a graph structure. It integrates heterogeneous data such as weather, flights, rules, and passengers to support intelligent reasoning and decision-making.

[0029] The Parameterized Template Engine (PTE) is an execution engine that can dynamically generate structured text or policies based on input variables. It is used to automatically generate tiered refund, change, and ticketing policy texts.

[0030] A multimodal classifier (MMC) is a machine learning model that integrates multiple types of input features for joint classification, used to comprehensively assess the impact of disasters on flights.

[0031] The Spatio-Temporal Alignment Engine (STAE) is a computational module that enables precise matching of meteorological early warning grid data with flight take-off and landing points, waypoints, and time windows.

[0032] It should be noted that the method and apparatus for responding to meteorological risks in air ticketing disclosed herein can be used in the field of artificial intelligence technology to achieve intelligent response to meteorological risks in air ticketing, and can also be used in any field other than the field of artificial intelligence technology to achieve intelligent response to meteorological risks in air ticketing. This disclosure does not limit the application field of the method and apparatus for responding to meteorological risks in air ticketing.

[0033] It should be noted that in this disclosure, customer information is collected and analyzed, and users are provided with corresponding operation entry points to choose whether to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0034] The following embodiments of the present invention can be applied to various systems / applications / devices for responding to meteorological risks in air ticketing. The present invention is applicable to intelligent air transport service systems, specifically to scenarios involving airline ticketing management, operational control, and passenger service collaboration. For example, when a sudden meteorological disaster occurs, the airline's operations control center automatically obtains weather warnings and assesses in real time the impact on the flight network; when faced with a large number of refund and rebooking requests due to weather conditions, the customer service system automatically pre-generates tiered refund and rebooking policies to assist in rapid manual approval and reduce service backlog; passengers proactively receive personalized travel warnings and rebooking guidance based on risk levels before flights may be affected, improving service proactivity; airlines intelligently schedule and pre-configure strategies for emergency ticketing resources during major holidays and periods of high seasonal disaster incidence, achieving proactive pre-disaster response.

[0035] This invention enables automatic semantic alignment and spatiotemporal correlation between meteorological disaster early warning information and ticketing rules, improving the accuracy of risk identification. Through a parameterized template engine, it automatically generates differentiated refund and change policies, shortening the effective time of refund and change policies from hours to minutes, achieving millisecond-level synchronization of early warnings, flight status, and policy changes, and improving customer service processing efficiency by more than 60%. While retaining the authority for manual intervention, the automation rate reaches more than 95%, significantly reducing manual review costs and passenger complaint rates, and improving overall service satisfaction.

[0036] The present invention will now be described in detail with reference to various embodiments.

[0037] Example 1

[0038] According to an embodiment of the present invention, an embodiment of a method for responding to weather risks in air ticketing is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0039] Figure 1 This is a flowchart of an optional response method for weather risks in air ticketing according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0040] Step S101: Obtain multi-source meteorological disaster early warning data and flight plan data. The flight plan data includes flight take-off and landing points, key points of the route, and expected flight time windows.

[0041] Optionally, step S101 includes: obtaining the latest meteorological data source for the current time period from designated meteorological agencies and satellite meteorological centers, and extracting early warning data containing multiple disaster types from the meteorological data source to obtain multi-source meteorological disaster early warning data. The multi-source meteorological disaster early warning data includes: disaster type, intensity, spatial range, radius of influence, and effective time window; obtaining flight operation map information from airlines, and extracting flight plan data from the flight operation map information; performing data cleaning and data standardization processing on the multi-source meteorological disaster early warning data and flight plan data to obtain preprocessed multi-source meteorological disaster early warning data and flight plan data.

[0042] It should be noted that in step S101, by connecting to designated authoritative data sources such as meteorological bureaus and satellite meteorological centers, real-time meteorological warning information streams can be periodically subscribed to. Structured warning records containing disaster types such as typhoons, rainstorms, blizzards, fog, thunderstorms, and sandstorms can be extracted. Each record includes a disaster type identifier, intensity level, geographical coordinate range, impact radius parameters, and effective start and end time windows, forming multi-source meteorological disaster warning data. Simultaneously, flight plan data is extracted from the airline's internal flight operation map system. This data includes the departure airport, landing airport, key route coordinate sequence, and estimated flight time windows for each segment of each flight, with time accuracy down to the minute. For the acquired raw meteorological and flight data, data cleaning operations are performed, including removing records with missing fields, abnormal coordinates, or out-of-order timestamps. The coordinate systems, time formats, and unit standards of different data sources are also uniformly converted. For example, WGS-84 coordinates (World Geodetic System 1984, a globally universal geodetic coordinate system that defines the Earth's ellipsoid parameters and geographic coordinate benchmarks for uniformly expressing latitude, longitude, and altitude) are uniformly converted to geographic latitude and longitude; UTC time (Coordinated Universal Time, an international standard time benchmark based on atomic clocks, which is not adjusted by region or season and serves as a reference standard for time synchronization in global meteorology, aviation, and communications) is uniformly converted to local time in the East Eighth Time Zone; and wind speed units are uniformly converted to meters per second. The final output is preprocessed multi-source meteorological disaster warning data and flight schedule data with a consistent format and standardized structure. This processing flow helps the subsequent spatiotemporal matching module to accurately align the meteorological impact range with flight trajectories, improving data compatibility and input quality.

[0043] Step S102: Based on a unified spatiotemporal coordinate system, dynamically align the meteorological disaster early warning data and flight plan data in spatiotemporal space, and identify whether the take-off and landing points or key points of the flight route fall within the spatiotemporal range of the early warning impact in the meteorological disaster early warning data.

[0044] It should be noted that this embodiment establishes a unified spatiotemporal coordinate system, mapping the preprocessed meteorological disaster warning data and flight schedule data to the same reference frame. The meteorological disaster warning data is described using a gridded spatial range, with each grid cell containing the center latitude and longitude, the radius of influence, and the effective time window. The flight schedule data is represented as a discrete waypoint sequence, with each waypoint associated with precise latitude and longitude coordinates and the expected flight time. This embodiment uses a spatiotemporal alignment engine to calculate the spatial distance between the coordinates of each flight's departure and arrival airports and key points along the route and the center point of each meteorological warning grid, determining whether the point is within the warning's influence radius. Simultaneously, it compares whether the expected flight time of the waypoint falls within the warning's effective time window. When both spatial location and time window meet the overlap condition, the waypoint is marked as a "point affected by the warning." The same judgment logic is applied to departure and arrival airports to identify whether the airport is within the warning coverage area.

[0045] Furthermore, this embodiment supports point-by-point scanning across multiple flight segments to identify whether a flight crosses a warning area at any point during its flight, thereby enabling fine-grained spatiotemporal coupling analysis of flight trajectories and the scope of meteorological impact. This process helps to establish a precise correlation between flights and disaster events, providing structured input for subsequent impact level prediction and improving the coverage completeness and spatial accuracy of risk identification.

[0046] Step S103: When the flight take-off and landing point or waypoint falls within the spatiotemporal range of the warning impact, extract a cross-modal composite feature set based on the spatiotemporal alignment result. The composite feature set includes at least one of the following: disaster type and level, dynamic distance between the warning center point and the airport / waypoint key point, spatiotemporal proportion of the warning coverage segment, and time overlap coefficient between flight schedule time and disaster duration.

[0047] Optionally, the step of extracting a cross-modal composite feature set based on the spatiotemporal alignment results includes: extracting disaster ontology features, spatiotemporal correlation features, and rule semantic features based on the spatiotemporal alignment results. The disaster ontology features include at least one of the following: disaster warning type, disaster level, and disaster duration. The spatiotemporal correlation features include at least one of the following: dynamic distance between the warning center point and key points of the airport / airway, spatiotemporal proportion of the warning-covered flight segment, and time overlap coefficient between flight schedule time and disaster duration. The rule semantic features include at least one of the following: rule features of the refund and change rules and related policy documents issued by the airline through clause semantic analysis. The spatiotemporal proportion of the warning-covered flight segment is obtained by calculating the proportion of the flight segment length within the warning radius in the total length of the flight segment and multiplying it by the proportion of the overlap time between the flight segment and the effective warning period. The cross-modal composite feature set is obtained by combining the disaster ontology features, spatiotemporal correlation features, and rule semantic features.

[0048] Among them, the disaster characteristics can include the disaster warning type (such as classification labels for typhoons, thunderstorms, and heavy fog) and the disaster level (such as structured level codes for yellow, orange, and red), as well as the duration of the disaster (the total duration calculated from the start and end time of the effective time window, in hours). The spatiotemporal correlation characteristics include the dynamic Euclidean distance (e.g., ground distance calculated based on WGS-84 coordinates, in kilometers) between the warning center point and each affected waypoint, used to measure the spatial proximity of flights to the disaster core area; the spatiotemporal proportion of the warning-covered flight segments is obtained through two-stage calculation: first, each flight segment is segmented, and the length of the segment within the warning influence radius is calculated as a percentage of the total flight segment length; second, the overlap between the time period covered by the segment and the warning effective time window is calculated as a percentage of the flight's planned flight time; finally, the spatial proportion and the temporal proportion are multiplied to obtain a comprehensive spatiotemporal coverage ratio, used to quantify the spatiotemporal overlap of flights affected by weather; the time overlap coefficient between the flight's planned time and the disaster duration is obtained by calculating the intersection time between the flight's expected flight time interval over the warning area and the disaster warning effective time window, divided by the total duration of the disaster, to obtain a normalized time overlap coefficient, reflecting the degree of matching between flight operations and the peak period of the disaster.

[0049] Furthermore, the rule semantic features are extracted by parsing the refund and rebooking rules documents issued by airline operators and using language models such as BERT to extract structured semantic relationships, such as condition-action rules like "free rebooking is available if the flight is canceled more than 3 hours ago" and "full refund is available if the destination is closed more than 2 hours ago." These are then transformed into quantifiable and coded rule triggering condition features. Combining the aforementioned disaster ontology features, spatiotemporal correlation features, and rule semantic features, a cross-modal composite feature set is formed, with each feature expressed in a standardized numerical vector form. This feature extraction process helps provide a unified representation of multi-source heterogeneous information for subsequent impact level prediction models, improving the model's ability to express complex weather-flight coupling relationships and enhancing the fine-grainedness and scenario adaptability of risk assessment.

[0050] Step S104: Input the composite feature set into the target disaster level prediction model, use the target disaster level prediction model to predict the impact of meteorological disasters on flights, and output the disaster impact level of flights. The target disaster level prediction model is a model pre-trained based on a deep learning neural network architecture with a self-attention mechanism and a long short-term memory network.

[0051] The target disaster level prediction model can adopt a hybrid structure that combines a deep learning neural network architecture based on self-attention mechanism with a long short-term memory network (LSTM). During the training phase, the model has been supervised learning through historical meteorological disaster events and corresponding flight operation data. The learning content includes disaster warning records, flight trajectories, actual delay / cancellation status, and manually labeled impact level labels. In the model structure, the LSTM layer is used to model the dynamic evolution of flight plan time series data and capture the temporal dependence of flight operation status at different time points with changes in meteorological conditions.

[0052] It should be noted that the target disaster level prediction model takes as input a standardized cross-modal composite feature set. Its dimensions can include the concatenation of numerical vectors of disaster ontology features, spatiotemporal correlation features, and rule semantic features. The model outputs three discrete level labels: low impact, medium impact, and high impact, corresponding to the mild, moderate, and severe degrees of meteorological interference to flight operations, respectively. Each level is generated by a probability distribution output by a Softmax layer and then thresholded. Furthermore, this target disaster level prediction model supports parallel inference for multiple flights, capable of processing composite feature sets of hundreds of flights at once. The output is an independent impact level prediction value for each flight. The prediction process is entirely based on feature input, without relying on manual intervention or hard-coded rules. The model's response latency during the inference phase is controlled at the millisecond level, supporting real-time risk assessment in high-concurrency scenarios. This helps to integrate the temporal dynamics and cross-dimensional correlations of structured features, improving the generalization ability to complex meteorological-flight coupled scenarios, making the impact level prediction results closer to the actual flight interruption trends, and providing a stable and consistent quantitative basis for subsequent refund and rebooking strategies.

[0053] Optionally, the process of training the target disaster level prediction model includes: constructing an initial disaster level prediction model based on a long short-term memory network and a deep learning neural network architecture based on a self-attention mechanism, wherein the long short-term memory network is used to model the temporal dependency of historical flight delay sequences, and the deep learning neural network architecture based on a self-attention mechanism is used to capture the global cross-modal correlation between multi-source meteorological features and flight features; obtaining model training samples and dividing the model training samples into multiple training subsets, wherein each training subset includes historical meteorological warning data, historical actual flight delay / cancellation records, and pre-labeled impact level labels; inputting each training subset into the initial disaster level prediction model, and using the initial disaster level prediction model to output impact level labels; after the initial disaster level prediction model has completed training on all training subsets, using a pre-acquired validation set to validate the model; and determining that the model training is complete when the accuracy of the impact level labels output by the initial disaster level prediction model reaches a predetermined accuracy requirement, thus obtaining the target disaster level prediction model.

[0054] It should be noted that in the process of training the target disaster level prediction model, this embodiment can build an initial disaster level prediction model based on a long short-term memory network (LSTM) and a deep learning neural network architecture based on a self-attention mechanism. The long short-term memory network is used to model the delay sequence change pattern of continuous time points in historical flight operations, and to capture the dynamic evolution trend of flight operation status before and after different weather events, such as the time series pattern of the on-time rate of flights gradually decreasing after three consecutive days of severe convective weather.

[0055] Furthermore, a deep learning neural network architecture based on a self-attention mechanism is constructed as a multi-head attention module to perform global correlation modeling on input multi-source meteorological features (such as disaster type, intensity, and radius of influence) and flight features (such as flight segment length, airport altitude, and historical cancellation rate). By calculating the correlation weights between each feature dimension, it automatically identifies high-risk combination patterns such as "typhoon red warning + plateau airport" or "dense fog + low visibility route", achieving nonlinear interactive expression of cross-modal features without the need for manual pre-setting rules.

[0056] It should be noted that the model training samples are derived from the historical aviation operation database, which covers complete records of all meteorological disaster events in the past three years. Each sample includes meteorological warning data (type, intensity, spatiotemporal range), actual flight operation records (take-off / landing time, whether it was canceled, delay duration), and three-level impact labels (low, medium, high) manually annotated by the operation control center. The total number of samples is no less than 100,000, covering typical scenarios of different seasons, regions, and route types.

[0057] Furthermore, in this embodiment, the training samples can be divided into multiple training subsets according to the time series. Each subset contains meteorological and flight data within an independent time period, ensuring that the training process does not introduce future information leakage while preserving temporal continuity. The training subsets are arranged sequentially in time to avoid data overlap and improve the model's generalization ability to temporal evolution trends. Subsequently, each training subset can be input into the initialized disaster level prediction model. The model outputs the predicted impact level label for the corresponding flight. The loss function adopts the form of cross-entropy to measure the difference between the predicted level and the manually labeled true label. The training process iteratively optimizes the model parameters through the backpropagation algorithm until the loss value tends to stabilize.

[0058] Step S105: Based on the level of disaster impact, call the preset parameterized refund and change rule template to generate a refund and change policy corresponding to the level of disaster impact. The refund and change policy includes passenger notification content and customer service approval policy.

[0059] Optionally, the step of generating a refund and rebooking strategy corresponding to the disaster impact level by calling a preset parameterized refund and rebooking rule template includes: calling the preset parameterized refund and rebooking rule template according to the disaster impact level, wherein the parameterized refund and rebooking rule template is constructed based on the rule semantic features of the cross-modal composite feature set, and the parameterized refund and rebooking rule template contains a mapping rule table of four-dimensional condition combinations of disaster type, airport closure status, delay duration, and cabin class and corresponding refund and rebooking actions; prioritizing passenger orders, wherein the priority classification is based on factors including ticket price, trip urgency, and historical service. Scoring; Based on priority classification, parameterized refund and rebooking rule templates are used to determine the passenger notification content and customer service approval strategy corresponding to the disaster impact level, and a refund and rebooking strategy is generated. The passenger notification content includes at least one of the following: notification triggering method, affected flight number, estimated delay duration, optional rebooking options and fee waiver period. The customer service approval strategy includes at least one of the following: the pre-approved refund and rebooking options corresponding to the first category of disaster impact level are confirmed and effective by customer service with one click; the flexible rebooking combination options corresponding to the second category of disaster impact level require manual review; and the suggested options corresponding to the third category of disaster impact level push a self-service rebooking link.

[0060] This embodiment uses the disaster impact level of the flight output by the target disaster level prediction model to call the preset parameterized refund and change rule template. This template is constructed by semantic parsing of historical refund and change policy documents and includes the combination logic of four-dimensional conditions: disaster type (such as typhoon, blizzard), airport closure status (whether it is completely closed), delay time range (e.g., 0-2 hours, 2-6 hours, >6 hours), and cabin class (economy class, business class, first class). Each combination corresponds to a clear refund and change action. For example, "typhoon red warning + airport closure + delay >6 hours + first class" triggers "full refund + transit accommodation compensation".

[0061] Furthermore, the parameterized refund and rescheduling rule template is stored in the form of a structured rule table. Each row corresponds to a combination of conditions and an action mapping, supporting dynamic addition, deletion, modification, and query. Its field design is entirely based on the rule semantic feature extraction results from the cross-modal composite feature set, ensuring that the template content is consistent with the actual policy terms at the semantic level and avoiding human translation bias.

[0062] It should be noted that before generating the refund and change policy, this embodiment performs priority classification on passenger orders. The classification criteria include ticket price (full price, discount, special price), trip urgency (such as same-day round trip, international transit, medical travel tags), and historical service ratings (past complaints and satisfaction records). The classification results can be represented by discrete rating levels, which serve as an auxiliary input dimension for policy generation to achieve differentiated service responses.

[0063] Furthermore, based on a joint judgment of the level of disaster impact and passenger priority classification, the system can retrieve matching rule entries from the parameterized refund and rebooking rule template to determine the corresponding passenger notification content and customer service approval strategy. The passenger notification content includes the notification triggering method (e.g., SMS, App push, voice call), affected flight number, expected delay duration range, list of optional rebooking segments, and fee waiver validity period. All content fields are dynamically filled from the template, supporting multilingual output and time localization adaptation.

[0064] It should be noted that the customer service approval strategy in this embodiment can be divided into three execution paths based on the impact level. For example: the first type is the pre-approval scheme corresponding to the high impact level, where the system automatically generates a refund / change instruction and marks it as "automatically effective," and the customer service only needs to view and confirm it without secondary review; the second type is the flexible change combination scheme corresponding to the medium impact level, such as "50% reduction in change fee + one free upgrade," which requires customer service personnel to log in to the backend to review the passenger's specific itinerary and manually activate it; the third type is the suggested scheme corresponding to the low impact level, where the system only pushes a self-service change link to the passenger with a risk warning, without triggering customer service intervention. This helps to achieve standardized, conditional, and hierarchical expression of refund / change strategies, which can support differentiated service responses to different passenger groups in disaster concurrency scenarios, improve the flexibility and configurability of strategy execution, reduce the frequency of manual intervention, and maintain the consistency and compliance of policy implementation.

[0065] Through the above steps, multi-source meteorological disaster early warning data and flight plan data can be obtained. The flight plan data includes flight departure and arrival points, key points along the route, and expected flight time windows. Based on a unified spatiotemporal coordinate system, the meteorological disaster early warning data and flight plan data are dynamically spatiotemporally aligned, and it is identified whether the flight departure and arrival points or key points along the route fall within the spatiotemporal impact range of the meteorological disaster early warning data. If the flight departure and arrival points or key points along the route fall within the spatiotemporal impact range, a cross-modal composite feature set is extracted based on the spatiotemporal alignment results. The composite feature set includes at least one of the following: disaster type and level, early warning center point, and airport / route key points. The system considers factors such as the dynamic distance of points, the spatiotemporal proportion of flight segments covered by early warnings, and the time overlap coefficient between flight schedules and the duration of disasters. This composite feature set is input into a target disaster level prediction model. The model then predicts the impact of meteorological disasters on flights and outputs the disaster impact level of the flights. The target disaster level prediction model is a pre-trained model based on a deep learning neural network architecture with a self-attention mechanism and a long short-term memory network. Based on the disaster impact level, a preset parameterized refund and rebooking rule template is used to generate a refund and rebooking strategy corresponding to the disaster impact level. This strategy includes passenger notification content and customer service approval procedures. In this embodiment, real-time weather warning data can be accurately matched with flight operation plans to automatically identify affected flights and passenger orders. Subsequently, a dynamic impact level prediction model is constructed to output a quantitative risk level. Finally, based on the level of disaster impact, a preset parameterized refund and change rule template is called to generate a refund and change policy corresponding to the level of disaster impact. This shifts the refund and change decision-making from fixed rules to intelligent graded response based on the actual degree of impact. Warning information can be pushed to passengers in real time, and the customer service backend is simultaneously notified to approve the policy in advance. This significantly shortens the response cycle, greatly reduces manual intervention, improves the efficiency of refund and change processing, effectively reduces service costs, and transforms the passenger experience from passive acceptance to proactive awareness and convenient operation. This solves the technical problems in related technologies where airlines' refund and change rules and processes are rigid and inefficient, increasing service costs and reducing passenger experience when dealing with weather disasters.

[0066] Optionally, after predicting the impact of meteorological disasters on flights using the target disaster level prediction model and outputting the disaster impact level of flights, the method further includes: using a graph neural network to perform multi-dimensional fusion based on the disaster impact level and dynamic graph. The multi-dimensional fusion includes: aggregating the spatial radiation range of disaster nodes, the take-off and landing coordinates of each flight node and historical delay patterns. The dynamic graph includes impact relationship nodes and applicable relationship nodes. Impact relationship nodes include disaster event ID, affected flight ID, impact intensity score, and impact start and end time. Applicable relationship nodes include flight ID, applicable refund and change rule ID, rule activation priority, and trigger condition logical expression. Based on the multi-dimensional fusion results, the method outputs the probability of flight cancellation / delay and outputs the affected flight number and the expected delay duration.

[0067] It should be noted that the dynamic graph in this embodiment may include impact relationship nodes and applicable relationship nodes. The impact relationship nodes may include disaster event ID, affected flight ID, impact intensity score (mapped from the level output by the prediction model to a continuous numerical value), and impact start and end time. The applicable relationship nodes include flight ID, applicable refund and change rule ID, rule effective priority (numerical encoding), and trigger condition logical expression (such as "delay ≥ 3h and cabin class = first class"). Furthermore, in this embodiment, the graph neural network can use the dynamic graph as the input structure, treating each impact relationship node and applicable relationship node as entities in the graph, and connecting them with edges to form a heterogeneous graph. The impact relationship node establishes an "impact" edge with the flight node, and the flight node establishes an "applicable" edge with the applicable rule node. The GNN aggregates the spatial radiation range of disaster nodes (with the warning center as the center and the impact radius as the scale), the take-off and landing coordinates of each flight node (used to calculate the geographical distribution density), and historical delay patterns (i.e., the mean and variance of past delays of the flight under similar weather conditions) through a multi-layer message passing mechanism, thereby achieving joint encoding of node semantics and topological structure.

[0068] It should be noted that the graph neural network uses a graph attention mechanism during the aggregation process to dynamically adjust the weights of different neighboring nodes. For example, when the historical delay rate of a flight is significantly higher than that of similar routes, its own historical pattern is given higher weight in the prediction. When multiple flights are concentrated within the same disaster impact radius, the spatial clustering effect is enhanced, thereby improving the ability to identify regional systemic delays.

[0069] Furthermore, based on the aforementioned multi-dimensional fusion results, the graph neural network outputs the predicted cancellation probability and delay duration for each affected flight. The cancellation probability is generated by using a sigmoid function to perform binary classification modeling on the fused features, while the delay duration is estimated by a regression head using continuous values ​​on the fused vector. The output results complement the disaster impact level, forming a refined impact description integrating "level + probability + duration." The flight cancellation probability is used to assist in prioritizing high-risk flights, and the predicted delay duration is used to populate key fields in passenger notifications. This facilitates the extension from static level prediction to dynamic behavior prediction, enhancing the system's perception of flight operation disturbance trends, making refund and rebooking strategies more closely aligned with actual passenger travel itinerary changes, and improving the foresight and accuracy of service responses.

[0070] Optionally, after generating a refund and rebooking policy corresponding to the disaster impact level by calling a preset parameterized refund and rebooking rule template, the policy also includes: triggering a reassessment instruction to reassess the disaster impact level of affected flights if the latest meteorological disaster warning data shows an upgrade or downgrade; and updating the pushed passenger notification content and customer service approval policy based on the latest disaster impact level obtained from the reassessment.

[0071] In this embodiment, a reassessment command is automatically triggered when the latest meteorological disaster warning data is upgraded or downgraded. This command is activated by the meteorological data subscription module after detecting a change in the warning level (e.g., from orange to red, or from yellow to blue). The system then re-performs spatiotemporal alignment and composite feature extraction on all flights marked as affected, and inputs the updated feature set into the target disaster level prediction model to recalculate the disaster impact level of each flight. The reassessment process does not rely on manual intervention. The system identifies the source of the change based on the timestamp and version number of the warning data, ensuring that recalculation is only initiated after an authoritative institution releases an official update. This avoids frequent strategy fluctuations due to temporary data volatility. The recalculation scope covers all flight records for which refunds and rebookings have not yet been completed, ensuring that the assessment results remain synchronized with the latest meteorological situation.

[0072] It should be noted that, based on the latest disaster impact level obtained from the reassessment, the system can automatically compare the currently generated refund and rebooking policies with the parameterized templates matching the new level. If the policy category changes (e.g., from medium impact to high impact), the content update process will be initiated immediately: fields such as the estimated delay duration, fee waiver period, and available rebooking options in the passenger notification will be dynamically replaced with the corresponding values ​​of the new policy. If the sent notification is not sent through an instant channel (e.g., email), the system will automatically use the updated version in subsequent pushes; if it is sent through an instant channel (e.g., SMS, App pop-up), the system will generate a supplementary notification explaining the changes to avoid misleading information.

[0073] Furthermore, the customer service approval strategy is updated synchronously. When the original strategy was "pre-approval effective" but the new level triggers "manual review required," the system automatically cancels the instruction marked as automatically effective, re-places the relevant refund / change request into the customer service pending review queue, and adds a note about the reason for the change. Conversely, when the level is lowered, causing the strategy to change from "manual review" to "self-service link push," the system disables the intervention permission of the customer service interface, freeing up human resources, and pushes self-service operation guidance to passengers. This helps maintain the consistency between refund / change policies and real-time weather conditions, reduces the risk of policy misuse or passenger complaints due to information lag, and improves the system's adaptability and service continuity in dynamic disaster evolution scenarios.

[0074] Optionally, it also includes: performing mandatory data integrity checks and data standardization transformations sequentially on the refund and rescheduling policies corresponding to the disaster impact level, wherein the mandatory data integrity checks include at least one of the following: ID validity checks, data type compliance checks, and timestamp continuity checks; and the data standardization transformations include at least one of the following: unified format specifications, time zone alignment, and character encoding conversion; calling the adapter through the heterogeneous database, performing database operations under the transaction control framework through the adapter, and using parameterized query defense SQL statements to encrypt and store sensitive fields in the refund and rescheduling policies after data standardization transformation.

[0075] The mandatory data integrity verification may include validating the ID validity of key fields in the refund and rebooking policy to ensure that the flight ID, disaster event ID, and rule ID all exist in the system's master data dictionary, avoiding invalid references or unregistered entities; at the same time, it performs data type compliance verification to verify that the delay duration is a non-negative value, the cabin class is a preset enumerated value, and the notification triggering method is a valid enumerated type; the timestamp continuity verification checks whether the policy's effective time is earlier than the warning release time, and whether the end time is not earlier than the start time, to prevent logically contradictory time intervals.

[0076] Furthermore, the data standardization and conversion process includes unified format specifications, such as unifying all dates and times to UTC+8 format while retaining millisecond precision to ensure unambiguous cross-system interaction; time zone alignment operations map and convert the local time of the passenger's location to the time zone of the airport where the flight departs or arrives, so that "expected delay until 18:00" in the notification accurately corresponds to the actual local time at the airport; character encoding conversion converts text from non-UTF-8 sources (such as old rules exported from historical systems) to Unicode encoding, ensuring the complete display of multilingual notification content (such as Chinese, English, and Pinyin).

[0077] It should be noted that the standardized refund and change policy data is accessed to the target storage system through a heterogeneous database call adapter. The adapter automatically selects the corresponding driver and connection protocol based on the target database type to achieve transparent access to heterogeneous data sources. The adapter encapsulates all database write operations under the transaction control framework to ensure that the storage behavior of each refund and change policy meets the atomicity requirements. A single transaction includes the writing of the policy master record, the associated rule ID, the passenger ID, and the encrypted fields. Failure in any step will trigger an overall rollback.

[0078] Optionally, it also includes: for the meteorological disaster early warning data after spatiotemporal alignment, rendering a meteorological heat map using a preset visualization icon library, wherein the meteorological heat map uses color gradients to distinguish meteorological intensity levels and marks the level of disaster impact; for the flight schedule data after spatiotemporal alignment, drawing a flight topology network diagram using a preset data visualization chart library, mapping the operational status of airport nodes to the flight topology network diagram, wherein the operational status of airport nodes includes at least: delay rate / cancellation rate; and pushing disaster early warning information and flight status update information to the user terminal.

[0079] It should be noted that for the meteorological disaster early warning data after spatiotemporal alignment, this embodiment can render a meteorological heat map through a preset visualization icon library. The heat map uses a geographic information system as the base map and uses color gradients to express the meteorological intensity level, such as blue representing weak impact, yellow representing moderate impact, and red representing strong impact. The color intensity is positively correlated with the early warning intensity score. At the same time, text labels are superimposed on the area covered by the heat map to clearly indicate the level of flight disaster impact (such as "high", "medium", "low") in the current area, realizing the superimposed display of two layers of information on meteorological intensity and aviation impact.

[0080] Furthermore, the rendering engine of the meteorological heat map supports a dynamic update mechanism. When new meteorological warning data enters the system and completes spatiotemporal alignment, the heat map is automatically redrawn. The color distribution smoothly transitions with the changes in the affected area, avoiding sudden frame jumps. The annotation information is displayed adaptively according to the user's zoom level, ensuring readability in both regional and city-level views.

[0081] It should be noted that for the flight schedule data after spatiotemporal alignment, this embodiment uses a preset data visualization chart library to draw a flight topology network diagram. For example, nodes represent airports and edges represent flight routes. The size of the nodes is proportional to the number of flights. The node colors change dynamically according to the operating status. The operating status includes at least the delay rate (e.g., 0-10% is green, 10-30% is orange, and >30% is red) and the cancellation rate (e.g., 0-5% is light gray, 5-15% is dark gray, and >15% is black). The two are presented with two-color overlay or layered ring markings, so that users can simultaneously perceive the delay pressure and the risk of flight interruption.

[0082] Furthermore, the flight topology network map and the meteorological heat map adopt a linked rendering mechanism. When a user selects a high-impact area in the heat map, the topology map automatically highlights all take-off and landing points or routes passing through that area, and displays the number of affected flights and the average delay time on the edge line, forming a two-way visual mapping of "meteorological impact - flight response".

[0083] It should be noted that the system pushes disaster warning information and flight status updates to users, with the content dynamically generated based on user roles: Operations control personnel receive a comprehensive situational view including heatmaps and topology maps; customer service personnel receive a list of affected flights and a summary of policy changes; and passengers receive personalized notifications via app or SMS, including flight numbers, estimated delay ranges, available rebooking options, and access points. All push information is synchronized with real-time backend data to ensure that every status update is the latest assessment result. This helps build an intuitive and interactive disaster response visualization interface, enhancing users' understanding of complex spatiotemporal situations, lowering the information interpretation threshold, improving decision-making response speed and operational accuracy, and achieving efficient transformation from data to understanding.

[0084] The following describes in detail another optional implementation method.

[0085] This invention provides an intelligent response system for meteorological risks in airline ticketing. By deeply integrating real-time, accurate meteorological disaster early warning information with airlines' core ticketing and customer service processes, it utilizes big data and artificial intelligence, combined with AI algorithms, to analyze the disaster impact level of flight routes and output the disaster impact level. The "disaster impact level of flight routes" is introduced as a weighting factor into the rule engine (e.g., high disaster impact level → SMS warning, generating relevant refund and change policies in advance), accelerating passenger refunds and changes, assisting customer service review, reducing operating costs, improving operational efficiency, and enhancing passenger experience and satisfaction.

[0086] Figure 2 This is a schematic diagram of an optional intelligent response system for air ticketing weather risks according to an embodiment of the present invention, such as... Figure 2 As shown, the intelligent response system may include: raw data (including multi-source meteorological disaster data and flight plan data), core processor (including data integration, intelligent analysis (interfacing with airline rule data and analysis data, as well as deep learning models and knowledge graphs), context data), persistence layer, and visualization module. Among these, the key components include: data integration module, intelligent analysis module, data storage module, and visualization module. The following is an illustrative explanation of the analysis of these modules.

[0087] Figure 3 This is a schematic diagram of the workflow of an optional data integration module according to an embodiment of the present invention, such as... Figure 3 As shown, it includes a data access layer (which can access meteorological bureau data, flight information, and airline operators' refund and change rules), a data cleaning and standardization section, a standard data pool, a data integration layer (including spatiotemporal matching and deep extraction of multi-dimensional features), a key semantic relationship construction section, a knowledge graph storage and modeling section (constructing knowledge graphs of entities, relationships, and attributes), a semantic reasoning and decision engine section, LSTM and Transformer models for processing spatiotemporal sequence features and structured features, a meteorological disaster flight impact level prediction model, and finally, data push.

[0088] The data integration module, as the core data hub of the system, is responsible for three main functions: data acquisition, data integration, and data push. Based on a unified spatial-temporal coordinate system, it utilizes advanced spatiotemporal matching technology to precisely and dynamically align meteorological warning grid data with flight plans / route networks (determining whether flight take-off and landing points, waypoints, and estimated flight times fall within the spatiotemporal range of warning impact), forming a cross-modal spatiotemporal alignment engine. Based on the alignment results, it deeply extracts and integrates multi-dimensional features: disaster ontology features (warning type, level, duration), spatiotemporal correlation features (dynamic distance between the warning center point and key airport / route points, spatiotemporal proportion of warning-covered flight segments, overlap coefficient between flight plans and disaster peak periods), and rule semantic features (understanding refund and change rules and related policy documents through clause semantic parsing). By integrating the aforementioned features and business rules, a multi-dimensional structured knowledge graph is constructed, containing key semantic relationships: "Impact" relationship: disaster events and potentially affected flights; "Applicability" relationship: flights and applicable refund and change rules. Deeply embedded intelligent risk assessment and impact evaluation functions (as the core embodiment of the knowledge graph's value) utilize advanced models such as LSTM and Transformer to process aligned spatiotemporal sequence features (historical weather, flight operations) and structured features (current warnings, flight schedules, extracted features), running a meteorological disaster-flight impact level prediction model. Based on real-time warning information and fused features, this model accurately predicts the scope of affected flights and their quantified impact levels (e.g., high, medium, low risk). The predicted "impact level" serves as a core dynamic attribute, providing real-time feedback and enhancing the knowledge graph, upgrading it into an intelligent decision-making foundation containing real-time risk information.

[0089] Figure 4 This is a schematic diagram of the workflow of an optional intelligent analysis module according to an embodiment of the present invention, such as... Figure 4As shown, it includes the following layers: First layer: Knowledge graph construction layer (input: integrating real-time disaster early warning data, flight spatiotemporal trajectory, dynamic risk rating and structured business rules, analyzing airline refund and rebooking rules and related policy documents through the model, generating disaster-airport mapping relationship, delay threshold-rule triggering mechanism, output structured indicator graph); Second layer: Graph neural network enhancement layer (dynamic graph multi-dimensional fusion, aggregating the spatial radiation range of disaster nodes, flight node take-off and landing coordinates and historical delay patterns, associating passenger itinerary nodes and rule nodes, generating personalized flight early warning and flight impact level); Third layer: Business decision layer (multimodal decomposer, parameterized template engine, generating flight impact information and passenger early warning prompts, finally outputting: personalized refund and rebooking solutions based on flight impact level).

[0090] The intelligent analysis module, based on the intelligent knowledge graph constructed by the data integration module (deeply integrating real-time disaster early warning, flight spatiotemporal trajectory, dynamic risk rating, and structured business rules), establishes a closed-loop system of disaster impact assessment and service optimization. The knowledge graph construction layer uses deep learning models such as BERT to analyze airline refund and rebooking rules and related policy documents, accurately solidifying the disaster-airport mapping relationship (quantifying the probability of operational disruption through the correlation matrix between disaster type / intensity and airport geographical coverage), delay threshold-rule triggering mechanism (constructing a decision tree model for delay duration tiers and refund / rebooking eligibility), and other core logic. The graph neural network (GNN) enhancement layer performs multi-dimensional fusion based on the dynamic graph—by aggregating the spatial radiation range of disaster nodes, flight node takeoff and landing coordinates, and historical delay patterns, it outputs the probability of flight cancellation / delay and expected duration; it associates passenger itinerary nodes with rule nodes to generate personalized early warnings (e.g., "Flights passing through typhoon areas, delays exceeding the threshold can be rebooked for free"); when the meteorological station upgrades the warning, it refreshes the risk rating and service terms in real time through re-aggregation of neighboring node features. The business decision-making level deploys a multimodal classifier to assess the flight impact level (high / medium / low), driving a parameterized template engine to generate tiered contingency plans: for high-impact flights, the "disaster weight × (priority rebooking channel)" strategy is activated, automatically matching transfer options; for medium-impact flights, flexible rebooking combinations are implemented (e.g., a 50% reduction in rebooking fees); and self-service rebooking links are pushed to low-impact flights. The reinforcement learning component, through a knowledge graph feedback mechanism, deposits validated and effective strategy relationships (e.g., "typhoon red alert + airport closure → automatic full refund and compensation") into new edges of the graph, ultimately forming a complete intelligent chain of "disaster perception → impact quantification → contingency plan generation → service push → effect feedback → strategy evolution," systematically improving passenger satisfaction and airline emergency response efficiency.

[0091] Figure 5 This is a schematic diagram of an optional data storage module according to an embodiment of the present invention, such as... Figure 5As shown, this module includes: core processing flow, database adapter, execution of database operations, and security and reliability assurance. In the event of storage failure / rollback, it outputs error information and rollback notification. In the event of successful storage, it outputs a data persistence completion notification.

[0092] The data storage module receives the refund and rescheduling rules from the upstream module. Through the core processing flow, it sequentially performs mandatory data integrity checks (including ID validity, data type compliance, and timestamp continuity verification), data standardization conversion (unified format specifications, time zone alignment, and character encoding conversion), and utilizes a heterogeneous database adaptation mechanism, with dynamic connection pool scheduling optimizing resource utilization. Subsequently, it executes database operations under a transaction control framework via adapter calls, employs parameterized queries to defend against structured query language injection attacks, encrypts sensitive fields, and uses end-to-end operation log auditing to ensure traceability. Simultaneously, it guarantees atomicity, consistency, isolation, and durability through a transaction management mechanism, automatically triggering rollback operations in case of anomalies. Finally, it establishes a bi-state feedback mechanism: outputting error messages and rollback notifications when storage fails, and outputting a "data persistence complete" flag when storage succeeds, thus constructing an end-to-end secure and reliable data storage system.

[0093] Figure 6 This is a schematic diagram of an optional visualization module according to an embodiment of the present invention, such as... Figure 6 As shown, this visualization module can connect to the data storage module, and then the Action can asynchronously retrieve data. Combined with Vuex state management, view data consistency, including Vue components in the view layer and the visualization rendering engine.

[0094] The visualization module integrates three core processes: weather warning, flight impact analysis, and refund / change policy management. Through modular design, it achieves dynamic response across the entire process: a three-layer visualization system connects the warning map, flight monitoring dashboard, and refund / change policy editing console. It dynamically renders weather heatmaps (typhoon paths / rainstorm intensity), draws flight topology network diagrams, dynamically maps airport nodes to operational status (delay rate / cancellation rate), and highlights threatened routes (e.g., red segments passing through thunderstorm areas) in real time based on risk ratings from the backend knowledge graph, forming a "weather-flight" dual-layer overlay analysis. When a disaster warning is triggered, a configurable rule engine generates operational disruption probabilities based on disaster type and airport distance, matches airline refund / change terms, and combines GNN-predicted flight delay probabilities to dynamically generate tiered solutions (e.g., "red warning + flight cancellation → full refund + priority for transfers"), automatically outputting change conditions and other terms. Management ensures full-domain synchronization of solutions, guaranteeing data consistency across views; users can define disaster areas on the warning map, select high-risk flights on the flight monitoring dashboard, and finally generate customized solutions in the refund / change policy editing console. At the same time, the mechanism of allowing airlines to adjust their permissions on refund and change rules is retained. By combining caching strategies to obtain meteorological warning data, flight dynamics and rule bases, the system can ultimately achieve the process from disaster identification and flight locking to the generation of refund and change policies and risk warnings.

[0095] In this embodiment of the invention, by deeply coupling multi-source meteorological early warning data with the airline ticketing system, a fully intelligent closed loop is achieved, from disaster perception to the automatic generation of refund and rebooking strategies. At the data processing level, through a spatiotemporal alignment engine and cross-modal feature extraction, a precise spatiotemporal correlation is established between discrete meteorological grid data and flight plan data, improving the accuracy of matching the meteorological impact range with flight operation status and significantly reducing the misjudgment rate of affected flight identification. At the intelligent analysis level, by combining the disaster impact level prediction capabilities of the Long Short-Term Memory (LSTM) network and the Transformer model, and the dynamic aggregation of historical flight delay patterns and spatial distributions by graph neural networks, the predicted results of flight cancellation probability and delay duration are closer to the actual operational disturbance trends, enhancing the system's forward-looking response to complex meteorological events.

[0096] In this embodiment of the invention, at the strategy generation level, a parameterized refund and change rule template based on four-dimensional condition combinations enables refund and change policies to automatically adapt to disaster level, cabin class, delay duration, and airport status. This achieves both standardization and differentiation in policy generation, reducing manual configuration workload and improving the consistency of rule implementation. At the service response level, passenger notification content and customer service approval strategies are automatically triggered in a tiered manner based on impact level and passenger priority. This supports three parallel paths: pre-approval, manual review, and self-service push, shortening the refund and change decision-making cycle, reducing the frequency of manual intervention at the customer service end, and improving service response efficiency.

[0097] At the interactive presentation level, the dual-layer linkage visualization of meteorological heat map and flight topology network map realizes an intuitive mapping between meteorological impact and flight operation status, helping operators to quickly identify high-risk areas and key routes, and improve the decision-making efficiency and spatial awareness of emergency command.

[0098] The following is a detailed description with reference to another embodiment.

[0099] Example 2

[0100] The air ticketing weather risk response device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in Embodiment 1 above.

[0101] Figure 7 This is a schematic diagram of an optional air ticketing weather risk response device according to an embodiment of the present invention, such as... Figure 7 As shown, the air ticketing weather risk response device may include: a warning data acquisition unit 71, a spatiotemporal alignment unit 72, a composite feature extraction unit 73, a disaster impact level output unit 74, and a refund and rescheduling strategy generation unit 75.

[0102] Among them, the early warning data acquisition unit 71 is used to acquire multi-source meteorological disaster early warning data and flight plan data. The flight plan data includes flight take-off and landing points, key points of the route, and expected flight time windows.

[0103] The spatiotemporal alignment unit 72 is used to dynamically align meteorological disaster early warning data and flight plan data based on a unified spatiotemporal coordinate system, and to identify whether the take-off and landing points or key points of the flight route fall within the spatiotemporal range of the early warning impact in the meteorological disaster early warning data.

[0104] The composite feature extraction unit 73 is used to extract a cross-modal composite feature set based on the spatiotemporal alignment result when the flight take-off and landing point or waypoint falls within the spatiotemporal range of the warning impact. The composite feature set includes at least one of the following: disaster type and level, dynamic distance between the warning center point and the airport / waypoint key point, spatiotemporal proportion of the warning coverage segment, and time overlap coefficient between flight schedule time and disaster duration.

[0105] The disaster impact level output unit 74 is used to input the composite feature set into the target disaster level prediction model, use the target disaster level prediction model to predict the impact of meteorological disasters on flights, and output the disaster impact level of flights. The target disaster level prediction model is a model pre-trained based on a deep learning neural network architecture with a self-attention mechanism and a long short-term memory network.

[0106] The refund and change policy generation unit 75 is used to generate a refund and change policy corresponding to the level of disaster impact by calling a preset parameterized refund and change rule template. The refund and change policy includes passenger notification content and customer service approval policy.

[0107] The aforementioned air ticketing weather risk response device can acquire multi-source meteorological disaster early warning data and flight plan data through the early warning data acquisition unit 71. The spatiotemporal alignment unit 72 dynamically aligns the meteorological disaster early warning data and flight plan data based on a unified spatiotemporal coordinate system, and identifies whether the flight's take-off and landing points or route key points fall within the spatiotemporal impact range of the meteorological disaster early warning data. The composite feature extraction unit 73 extracts a cross-modal composite feature set based on the spatiotemporal alignment result when the flight's take-off and landing points or route points fall within the spatiotemporal impact range of the early warning data. The composite feature set is then input into the target disaster level prediction model through the disaster impact level output unit 74. This model predicts the impact of meteorological disasters on flights and outputs the disaster impact level of the flights. The target disaster level prediction model is a model pre-trained based on a deep learning neural network architecture with a self-attention mechanism and a long short-term memory network. The refund and change policy generation unit 75 generates a refund and change policy corresponding to the disaster impact level by calling a preset parameterized refund and change rule template. The refund and change policy includes passenger notification content and customer service approval strategy. In this embodiment, real-time weather warning data can be accurately matched with flight operation plans to automatically identify affected flights and passenger orders. Subsequently, a dynamic impact level prediction model is constructed to output a quantitative risk level. Finally, based on the level of disaster impact, a preset parameterized refund and change rule template is called to generate a refund and change policy corresponding to the level of disaster impact. This shifts the refund and change decision-making from fixed rules to intelligent graded response based on the actual degree of impact. Warning information can be pushed to passengers in real time, and the customer service backend is simultaneously notified to approve the policy in advance. This significantly shortens the response cycle, greatly reduces manual intervention, improves the efficiency of refund and change processing, effectively reduces service costs, and transforms the passenger experience from passive acceptance to proactive awareness and convenient operation. This solves the technical problems in related technologies where airlines' refund and change rules and processes are rigid and inefficient, increasing service costs and reducing passenger experience when dealing with weather disasters.

[0108] Optionally, the composite feature extraction unit includes: a feature extraction module, used to extract disaster ontology features, spatiotemporal correlation features, and rule semantic features based on spatiotemporal alignment results. The disaster ontology features include at least one of the following: disaster warning type, disaster level, and disaster duration. The spatiotemporal correlation features include at least one of the following: dynamic distance between the warning center point and key points of the airport / airway, spatiotemporal proportion of the warning-covered flight segment, and time overlap coefficient between flight schedule time and disaster duration. The rule semantic features include at least one of the following: rule features of the refund and change rules and related policy documents issued by the airline through clause semantic analysis. The spatiotemporal proportion of the warning-covered flight segment is obtained by calculating the proportion of the flight segment length within the warning radius in the total length of the flight segment and multiplying it by the overlap time proportion between the flight segment and the effective warning period. The composite feature determination module is used to integrate the disaster ontology features, spatiotemporal correlation features, and rule semantic features to obtain a cross-modal composite feature set.

[0109] Optionally, during the training of the target disaster level prediction model, the air ticketing meteorological risk response system further includes: a disaster level prediction model initialization module, used to construct an initial disaster level prediction model based on a long short-term memory network and a deep learning neural network architecture based on a self-attention mechanism. The long short-term memory network is used to model the temporal dependencies of historical flight delay sequences, and the deep learning neural network architecture based on a self-attention mechanism is used to capture the global cross-modal correlation between multi-source meteorological features and flight features; acquiring model training samples and dividing them into multiple training subsets, each subset including historical meteorological warning data, historical actual flight delay / cancellation records, and pre-labeled impact level labels; a model data input module, used to input each training subset into the initial disaster level prediction model and output impact level labels using the initial disaster level prediction model; and a model validation module, used to validate the model using a pre-acquired validation set after the initial disaster level prediction model has completed training on all training subsets. If the accuracy of the impact level labels output by the initial disaster level prediction model reaches a predetermined accuracy requirement, the model training is considered complete, and the target disaster level prediction model is obtained.

[0110] Optionally, the refund and rescheduling strategy generation unit includes: a template invocation module, used to invoke a preset parameterized refund and rescheduling rule template according to the level of disaster impact, wherein the parameterized refund and rescheduling rule template is constructed based on the rule semantic features of a cross-modal composite feature set, and the parameterized refund and rescheduling rule template contains a mapping rule table of four-dimensional condition combinations of disaster type, airport closure status, delay duration, and cabin class and corresponding refund and rescheduling actions; an order priority classification module, used to prioritize passenger orders, wherein the priority classification is based on factors including ticket price, trip urgency, and historical service rating; and a refund and rescheduling strategy generation module, used for Based on priority classification, parameterized refund and rebooking rule templates are used to determine the passenger notification content and customer service approval strategy corresponding to the disaster impact level, generating a refund and rebooking strategy. The passenger notification content includes at least one of the following: notification triggering method, affected flight number, estimated delay duration, optional rebooking options and fee waiver period. The customer service approval strategy includes at least one of the following: the pre-approved refund and rebooking options corresponding to the first category of disaster impact level are confirmed and effective by customer service with one click; the flexible rebooking combination options corresponding to the second category of disaster impact level require manual review; and the suggested options corresponding to the third category of disaster impact level push a self-service rebooking link.

[0111] Optionally, the air ticketing weather risk response system further includes: a multi-dimensional fusion unit, used to predict the impact of weather disasters on flights using a target disaster level prediction model and output the disaster impact level of flights, and then use a graph neural network to perform multi-dimensional fusion based on the disaster impact level and dynamic graph. The multi-dimensional fusion includes: aggregating the spatial radiation range of disaster nodes, the take-off and landing coordinates of each flight node and historical delay patterns, and the dynamic graph includes impact relationship nodes and applicable relationship nodes. The impact relationship nodes include disaster event ID, affected flight ID, impact intensity score, and impact start and end time. The applicable relationship nodes include flight ID, applicable refund and change rule ID, rule effective priority, and trigger condition logical expression; and an affected flight number output module, used to output the probability of flight cancellation / delay based on the multi-dimensional fusion results, and output the affected flight number and the expected delay duration.

[0112] Optionally, the air ticketing weather risk response system also includes: a reassessment instruction triggering unit, used to trigger a reassessment instruction to reassess the disaster impact level of affected flights if the latest weather disaster warning data shows an upgrade or downgrade, after generating a refund and change policy corresponding to the disaster impact level by calling a preset parameterized refund and change rule template according to the disaster impact level; and a passenger notification update unit, used to update the pushed passenger notification content and customer service approval policy based on the latest disaster impact level obtained from the reassessment.

[0113] Optionally, the air ticketing weather risk response system further includes: a refund and change policy verification unit, used to sequentially perform mandatory data integrity verification and data standardization transformation on the refund and change policies corresponding to the disaster impact level, wherein the mandatory data integrity verification includes at least one of the following: ID validity verification, data type compliance verification and timestamp continuity verification; the data standardization transformation includes at least one of the following: unified format specification, time zone alignment and character encoding conversion; and an adaptation unit, used to call the adapter through the heterogeneous database, perform database operations under the transaction control framework through the adapter, and use parameterized query defense SQL statements to encrypt and store sensitive fields in the refund and change policy after data standardization transformation.

[0114] Optionally, the air ticketing weather risk response system also includes: a heat map rendering unit, used to render a weather heat map from the spatiotemporally aligned weather disaster warning data using a preset visualization icon library, wherein the weather heat map uses color gradients to distinguish weather intensity levels and marks the level of disaster impact; a topology network diagram drawing unit, used to draw a flight topology network diagram from the spatiotemporally aligned flight schedule data using a preset data visualization icon library, mapping the operational status of airport nodes to the flight topology network diagram, wherein the operational status of airport nodes includes at least: delay rate / cancellation rate; and pushing disaster warning information and flight status update information to the user terminal.

[0115] Optionally, the early warning data acquisition unit includes: an early warning data acquisition module, used to acquire the latest meteorological data sources for the latest time period from designated meteorological agencies and satellite meteorological centers, and extract early warning data containing multiple disaster types from the meteorological data sources to obtain multi-source meteorological disaster early warning data, wherein the multi-source meteorological disaster early warning data includes: disaster type, intensity, spatial range, radius of influence, and effective time window; a flight operation map acquisition module, used to acquire flight operation map information from airlines and extract flight plan data from the flight operation map information; and a preprocessing module, used to perform data cleaning and data standardization processing on the multi-source meteorological disaster early warning data and flight plan data to obtain preprocessed multi-source meteorological disaster early warning data and flight plan data.

[0116] The aforementioned air ticketing weather risk response device may also include a processor and a memory. The aforementioned early warning data acquisition unit 71, spatiotemporal alignment unit 72, composite feature extraction unit 73, disaster impact level output unit 74, and refund / rescheduling strategy generation unit 75 are all stored in the memory as program units. The processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0117] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and intelligent response to weather risks in airline ticketing can be achieved by adjusting kernel parameters.

[0118] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0119] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the air ticketing weather risk response method of any one of the above embodiments.

[0120] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the air ticketing weather risk response method of any one of the above embodiments.

[0121] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the air ticketing weather risk response method described in various embodiments of this application.

[0122] This application also provides a computer program product, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the air ticketing weather risk response method described in various embodiments of this application.

[0123] Figure 8 This is a hardware structure block diagram of an electronic device (or mobile device) for a weather risk response method in air ticketing according to an embodiment of the present invention. Figure 8 As shown, an electronic device may include one or more ( Figure 8 The processor 802 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 804 for storing data may also be included. In addition, it may include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a keyboard, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 8The structure shown is for illustrative purposes only and does not limit the structure of the electronic device described above. For example, the electronic device may also include components that are more... Figure 8 The more or fewer components shown, or having the same Figure 8 The different configurations shown.

[0124] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0125] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0126] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0127] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0128] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0129] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for responding to weather risks in air ticketing, characterized in that, include: Acquire multi-source meteorological disaster early warning data and flight plan data, wherein the flight plan data includes flight take-off and landing points, key points of the route, and expected flight time windows; Based on a unified spatiotemporal coordinate system, the meteorological disaster early warning data and the flight plan data are dynamically spatiotemporally aligned, and it is identified whether the flight take-off and landing points or key points of the route fall within the spatiotemporal range of the early warning impact in the meteorological disaster early warning data. When the flight take-off and landing point or waypoint falls within the spatiotemporal range of the warning impact, a cross-modal composite feature set is extracted based on the spatiotemporal alignment result. The composite feature set includes at least one of the following: disaster type and level, dynamic distance between the warning center point and key airport / waypoint, spatiotemporal proportion of the warning coverage segment, and time overlap coefficient between flight schedule time and disaster duration. The composite feature set is input into the target disaster level prediction model, and the target disaster level prediction model is used to predict the impact of meteorological disasters on flights, and the disaster impact level of flights is output. The target disaster level prediction model is a model pre-trained based on a deep learning neural network architecture with a self-attention mechanism and a long short-term memory network. Based on the level of disaster impact, a preset parameterized refund and rebooking rule template is invoked to generate a refund and rebooking strategy corresponding to the level of disaster impact. The refund and rebooking strategy includes passenger notification content and customer service approval strategy.

2. The method for responding to meteorological risks in air ticketing according to claim 1, characterized in that, The steps for extracting cross-modal composite feature sets based on spatiotemporal alignment results include: Based on the spatiotemporal alignment results, disaster ontology features, spatiotemporal correlation features, and rule semantic features are extracted. The disaster ontology features include at least one of the following: disaster warning type, disaster level, and disaster duration. The spatiotemporal correlation features include at least one of the following: dynamic distance between the warning center point and key points of the airport / airway, spatiotemporal proportion of the warning coverage segment, and time overlap coefficient between flight schedule time and disaster duration. The rule semantic features include at least one of the following: rule features of the refund and rebooking rules and related policy documents issued by the airline through clause semantic parsing. The spatiotemporal proportion of the warning coverage segment is obtained by calculating the proportion of the flight segment length within the warning radius in the total length of the flight segment and multiplying it by the proportion of the overlap time between the flight segment and the effective warning period. The cross-modal composite feature set is obtained by combining the disaster ontology features, the spatiotemporal correlation features, and the rule semantic features.

3. The method for responding to meteorological risks in air ticketing according to claim 1, characterized in that, The process of training the target disaster level prediction model includes: An initial disaster level prediction model is constructed based on a long short-term memory network and a deep learning neural network architecture based on a self-attention mechanism. The long short-term memory network is used to model the temporal dependency of historical flight delay sequences, and the deep learning neural network architecture based on a self-attention mechanism is used to capture the global cross-modal correlation between multi-source meteorological features and flight features. Obtain model training samples and divide the model training samples into multiple training subsets, wherein each training subset includes historical weather warning data, historical actual flight delay / cancellation records, and pre-labeled impact level labels; Each of the training subsets is input into the initialized disaster level prediction model, and the impact level label is output using the initialized disaster level prediction model; After the initialized disaster level prediction model has completed training on all the training subsets of data, the model is validated using a pre-acquired validation set. If the accuracy of the impact level label output by the initialized disaster level prediction model reaches the predetermined accuracy requirement, the model training is considered complete, and the target disaster level prediction model is obtained.

4. The method for responding to meteorological risks in air ticketing according to claim 1, characterized in that, The step of generating a refund / rescheduling policy corresponding to the disaster impact level by calling a preset parameterized refund / rescheduling rule template includes: The preset parameterized refund and rebooking rule template is invoked according to the disaster impact level. The parameterized refund and rebooking rule template is constructed based on the rule semantic features in the cross-modal composite feature set. The parameterized refund and rebooking rule template includes a mapping rule table of four-dimensional conditions, namely disaster type, airport closure status, delay duration and cabin class, and corresponding refund and rebooking actions. Passenger orders are prioritized, and the priority classification is based on factors such as ticket price, trip urgency, and historical service rating. Based on the priority classification, the parameterized refund and rebooking rule template is used to determine the passenger notification content and customer service approval strategy corresponding to the disaster impact level, and the refund and rebooking strategy is generated. The passenger notification content includes at least one of the following: notification triggering method, affected flight number, estimated delay duration, optional rebooking options and fee waiver period. The customer service approval strategy includes at least one of the following: the pre-approved refund and rebooking options corresponding to the first type of disaster impact level are confirmed and effective by customer service with one click; the flexible rebooking combination options corresponding to the second type of disaster impact level require manual review; and the suggested options corresponding to the third type of disaster impact level push a self-service rebooking link.

5. The method for responding to meteorological risks in air ticketing according to claim 1, characterized in that, After using the target disaster level prediction model to predict the impact of meteorological disasters on flights and outputting the disaster impact level of flights, the process also includes: Using a graph neural network based on the disaster impact level and dynamic graph, multi-dimensional fusion is performed. The multi-dimensional fusion includes: aggregating the spatial radiation range of disaster nodes, the take-off and landing coordinates of each flight node and historical delay patterns. The dynamic graph includes impact relationship nodes and applicable relationship nodes. The impact relationship nodes include disaster event ID, affected flight ID, impact intensity score, and impact start and end time. The applicable relationship nodes include flight ID, applicable refund and change rule ID, rule effective priority, and trigger condition logical expression. Based on the multi-dimensional fusion results, the probability of flight cancellation / delay is output, along with the affected flight numbers and the estimated delay duration.

6. The method for responding to meteorological risks in air ticketing according to claim 1, characterized in that, After generating a refund / rescheduling policy corresponding to the disaster impact level by calling a preset parameterized refund / rescheduling rule template, the process further includes: If the latest meteorological disaster warning data shows an upgrade or downgrade, a reassessment instruction will be triggered to reassess the disaster impact level of the affected flights; Based on the latest disaster impact level obtained from the reassessment, the content of the passenger notifications and customer service approval strategies have been updated.

7. The method for responding to meteorological risks in air ticketing according to claim 1, characterized in that, Also includes: For the refund and rescheduling policies corresponding to the disaster impact level, data integrity mandatory verification and data standardization transformation are performed sequentially. The data integrity mandatory verification includes at least one of the following: ID validity verification, data type compliance verification and timestamp continuity verification. The data standardization transformation includes at least one of the following: unified format specification, time zone alignment and character encoding conversion. The adapter is invoked through a heterogeneous database, and database operations are performed through the adapter under the transaction control framework. Parameterized query defense SQL statements are used to encrypt and store sensitive fields in the refund and reschedule policy after data standardization transformation.

8. The method for responding to meteorological risks in air ticketing according to claim 1, characterized in that, Also includes: For the meteorological disaster early warning data after spatiotemporal alignment, a meteorological heat map is rendered using a preset visualization icon library. The meteorological heat map uses color gradients to distinguish meteorological intensity levels and marks the level of disaster impact. For the flight schedule data after spatiotemporal alignment, a flight topology network diagram is drawn using a preset data visualization chart library, and the operating status of airport nodes is mapped to the flight topology network diagram. The operating status of the airport nodes includes at least: delay rate / cancellation rate. Disaster warning information and flight status updates are pushed to users.

9. The method for responding to meteorological risks in air ticketing according to claim 1, characterized in that, The steps for obtaining multi-source meteorological disaster early warning data and flight schedule data include: The latest meteorological data for the specified time period is obtained from designated meteorological agencies and satellite meteorological centers, and warning data for multiple disaster types contained in the meteorological data sources are extracted to obtain the multi-source meteorological disaster warning data. The multi-source meteorological disaster warning data includes: disaster type, intensity, spatial range, radius of influence, and effective time window. Obtain flight operation map information from the airline, and extract the flight plan data from the flight operation map information; The multi-source meteorological disaster early warning data and the flight plan data are cleaned and standardized to obtain pre-processed multi-source meteorological disaster early warning data and flight plan data.

10. A response system for meteorological risks in air ticketing, characterized in that, include: The early warning data acquisition unit is used to acquire multi-source meteorological disaster early warning data and flight plan data, wherein the flight plan data includes flight take-off and landing points, key points of the route, and expected flight time windows; The spatiotemporal alignment unit is used to dynamically align the meteorological disaster early warning data with the flight plan data based on a unified spatiotemporal coordinate system, and to identify whether the flight take-off and landing points or key points of the route fall within the spatiotemporal range of the early warning impact in the meteorological disaster early warning data. The composite feature extraction unit is used to extract a cross-modal composite feature set based on the spatiotemporal alignment result when the flight take-off and landing point or waypoint falls within the spatiotemporal range of the warning impact. The composite feature set includes at least one of the following: disaster type and level, dynamic distance between the warning center point and the airport / waypoint key point, spatiotemporal proportion of the warning coverage segment, and time overlap coefficient between flight schedule time and disaster duration. The disaster impact level output unit is used to input the composite feature set into the target disaster level prediction model, use the target disaster level prediction model to predict the impact of meteorological disasters on flights, and output the disaster impact level of flights. The target disaster level prediction model is a model pre-trained based on a deep learning neural network architecture with a self-attention mechanism and a long short-term memory network. The refund and rebooking strategy generation unit is used to generate a refund and rebooking strategy corresponding to the disaster impact level by calling a preset parameterized refund and rebooking rule template. The refund and rebooking strategy includes passenger notification content and customer service approval strategy.

Images