An emergency command method and system for low altitude flight

Abstract

The application provides an emergency command method and system for low-altitude flight, which comprises the following steps: acquiring real-time flight monitoring data of a target area; determining the type of an emergency event in the target area according to an abnormal data subset in the real-time flight monitoring data; calling a flight emergency prediction model corresponding to the type of the emergency event; the flight emergency prediction model is trained based on flight features and airspace features selected from historical emergency event data, and the types of input parameters and output parameters of each flight emergency prediction model are consistent in the training stage and the application stage; outputting a flight path adjustment strategy and an emergency resource allocation strategy according to the flight emergency prediction model to generate an emergency command instruction; executing the emergency command instruction and updating the parameters of the flight emergency prediction model according to real-time feedback data after execution. The application can overcome the problem of coordinated improvement of flight safety and emergency response efficiency in a complex airspace environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to an emergency command method and system for low-altitude flight. BACKGROUND

[0002] In recent years, with the wide application of low-altitude carriers such as unmanned aerial vehicles and navigation aircraft, low-altitude flight emergency command technology has become a core link to ensure flight safety. Existing technologies mostly generate instructions based on preset emergency rule libraries or single-dimensional flight emergency prediction models, such as triggering fixed obstacle avoidance paths through aircraft positioning data or starting emergency response programs according to meteorological warning thresholds. Although such methods can cope with routine scenarios, they are difficult to achieve rapid and accurate decision-making in complex working conditions such as sudden airspace conflicts and extreme weather disturbances.

[0003] Current low-altitude emergency command systems have significant technical bottlenecks: first, static rule libraries cannot dynamically adapt to the multi-factor coupling relationship between aircraft state mutations, temporary airspace regulations, and meteorological parameter fluctuations, resulting in mismatch between generated flight path adjustment instructions and real-time environment; second, single flight emergency prediction models lack cross-dimensional data collaboration mechanisms, and flight control instructions and resource scheduling strategies produce temporal and spatial logical conflicts due to data source fragmentation; third, structural deviation between model training data and application scene parameters causes decision feature drift, resulting in emergency resource deployment location deviation or obstacle avoidance path planning failure; fourth, there is a lack of optimization mechanism driven by instruction execution effect feedback, and the system response capability continuously deteriorates in the face of high-frequency emergencies.

[0004] Due to insufficient dynamic adaptability, lack of multi-source data collaboration, and weak closed-loop optimization capability, existing technologies face systematic defects such as instruction generation lag, frequent strategy mutual exclusivity, and repeated resource scheduling in low-altitude emergency command, necessitating the construction of an emergency command architecture that real-time integrates multi-dimensional state parameters, dynamically corrects decision logic, and has self-optimization capability to break through the bottleneck of coordinated improvement of flight safety and emergency response efficiency in complex airspace environments. SUMMARY

[0005] Therefore, the embodiments of the present application provide at least an emergency command method for low-altitude flight.

[0006] The technical solution of the embodiment of the present application is implemented in the following manner: on one hand, the embodiment of the present application provides an emergency command method for low-altitude flight, which comprises the following steps: acquiring real-time flight monitoring data of a target region; the real-time flight monitoring data comprises aircraft state parameters, meteorological parameters and airspace occupation parameters; determining an emergency event type in the target region according to an abnormal data subset in the real-time flight monitoring data; calling a flight emergency prediction model corresponding to the emergency event type; the flight emergency prediction model is trained based on flight features and airspace features filtered from historical emergency event data, and the types of input parameters and output parameters of each flight emergency prediction model are consistent in the training stage and the application stage; outputting flight path adjustment strategies and emergency resource allocation strategies according to the flight emergency prediction model to generate emergency command instructions; and executing the emergency command instructions and updating parameters of the flight emergency prediction model according to real-time feedback data after execution.

[0007] On the other hand, the embodiment of the present application provides a computer system comprising a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the processor implements the steps in the above method when executing the program.

[0008] The emergency command method for low-altitude flight provided by the present application can effectively utilize the correlation between real-time dynamic data and historical training data in the low-altitude flight scene, break through the limitations of single-dimensional decision-making in traditional emergency response, and significantly improve the real-time performance and reliability of emergency command in a complex airspace environment. By forcibly constraining the consistency of input and output parameter types of the flight emergency prediction model in the training stage and the application stage, the heading, height and speed parameters in the flight path adjustment strategy and the path, position and supply parameters in the resource allocation strategy are ensured to form a synergistic mapping relationship in the spatial dimension and the time dimension, thereby eliminating the risk of strategy conflict caused by model feature drift. Based on the feedback data after instruction execution, the model parameters are dynamically updated, thereby solving the interference problem of uncertain factors such as meteorological mutation and dynamic change of airspace occupation on emergency decision-making, and finally realizing the synchronous optimization of aircraft obstacle avoidance path planning and precise allocation of emergency resources, and comprehensively improving the efficiency and safety of low-altitude emergency event disposal. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 An implementation flowchart of the emergency command method for low-altitude flight provided by the embodiment of the present application is shown in the figure;

[0010] Figure 2A hardware entity schematic diagram of a computer system provided for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0011] An emergency command method for low-altitude flight is provided by an embodiment of the present application, which can be executed by a processor of a computer system. The computer system can refer to a server, a notebook computer, a tablet computer, a desktop computer, a mobile device, and the like, which have data processing capabilities.

[0012] Figure 1 An implementation flowchart of an emergency command method for low-altitude flight provided for an embodiment of the present application is shown in the figure. Figure 1 As shown in the figure, the method comprises the following steps: S100: acquiring real-time flight monitoring data of a target area; wherein the real-time flight monitoring data comprises aircraft state parameters, meteorological parameters, and airspace occupation parameters.

[0013] Optionally, the real-time flight monitoring data is a comprehensive data set collected in real time, for example, by a sensor network, a radar system, and an aviation communication device deployed in the target area, for comprehensively reflecting the dynamic state of the current low-altitude flight environment. The aircraft state parameters specifically comprise real-time flight speed, flight altitude, heading angle, attitude angle, acceleration, remaining fuel quantity, and navigation system state of the aircraft, and the like, which are core operation indexes, and these parameters are transmitted in real time to the command center through a data link between the aircraft onboard equipment and the ground control system; the meteorological parameters cover atmospheric environment indexes in the target area, including wind speed, wind direction, temperature, humidity, air pressure, visibility, precipitation intensity, thunderstorm activity area, and turbulence intensity, and such data is collected by multi-source fusion of meteorological satellites, ground meteorological stations, and meteorological detection equipment carried by unmanned aerial vehicles; the airspace occupation parameters are used to describe the distribution density of aircrafts in the current airspace, the boundary coordinates of temporary control areas, the geographic positions and height restrictions of obstacles (such as buildings, high-voltage lines, and mountains), and other flight plan conflict areas, and the like, which are airspace resource usage conditions, and are acquired by collaborative monitoring of an airspace management system and an air traffic control radar. Specifically, the collection frequency of the aircraft state parameters needs to meet at least one update per second to ensure accurate capture of the aircraft dynamics; the collection of the meteorological parameters needs to be combined with a short-term weather forecasting model and real-time observation data to cover the comprehensive influence of static environment and dynamic changes; the update of the airspace occupation parameters needs to be synchronized with the air traffic control instructions to ensure the timeliness of the airspace dynamic information. Through parallel collection and fusion of the above three types of parameters, the real-time flight monitoring data can construct a three-dimensional dynamic mapping model of the aircrafts, meteorological environment, and airspace resources in the target area, and provide a data basis for subsequent emergency event identification and decision-making.

[0014] S200: determining an emergency event type in the target area according to an abnormal data subset in the real-time flight monitoring data.

[0015] Exemplarily, the abnormal data subset is a set of parameters exceeding the preset safety range, extracted from the real-time flight monitoring data by a dynamic threshold screening mechanism, for characterizing the flight risks or emergencies possibly existing in the target region. Specifically, the dynamic threshold screening mechanism adopts the preset aircraft state threshold, the preset meteorological fluctuation threshold, and the preset airspace capacity threshold as the screening criteria: for the aircraft state parameters, when the flight speed deviates from the allowable deviation range of the route planning speed, the flight height is lower than the minimum safety height limit, or the heading angle mutation exceeds the angle tolerance, the abnormal aircraft data is triggered; for the meteorological parameters, if the mutation gradient of the wind speed in unit time exceeds the safe flight condition, the visibility decline rate reaches the critical value affecting the navigation accuracy, or the air pressure fluctuation intensity causes the aircraft attitude instability risk, the abnormal meteorological data is marked; for the airspace occupation parameters, when the aircraft density in the airspace exceeds the maximum capacity limit, the overlap degree of the temporary control region and the flight path reaches the conflict threshold, or the obstacle distribution dispersion causes the obstacle avoidance difficulty to rise, the airspace abnormal data is determined. After the abnormal data subset is verified by multi-dimensional parameter cross-validation, it is input into the event type association model. Based on the mapping relationship between the same abnormal characteristics and the event type in the historical emergency event case library, the model calculates the similarity between the current abnormal data and the historical cases by a hierarchical matching mechanism, and outputs the basic classification label of the emergency event type and its priority level. For example, if the abnormal data subset contains aircraft height drop and strong crosswind meteorological data, the model will match the composite event type of "aircraft loss of control" and "meteorological mutation", and generate the priority order according to the historical processing time limit and resource demand parameters. Finally, the event feature label will be associated with the subsequent model calling conditions to ensure the pertinence of the emergency response strategy.

[0016] As an implementation, in step S200, the emergency event type in the target region is determined according to the abnormal data subset in the real-time flight monitoring data, which can specifically include: in step S210, a first data subset corresponding to the aircraft state parameters, a second data subset corresponding to the meteorological parameters, and a third data subset corresponding to the airspace occupation parameters are extracted from the real-time flight monitoring data.

[0017] The real-time flight monitoring data as a set of multi-source heterogeneous data needs to be classified and extracted into three independent data subsets according to the parameter type to support refined analysis. The first data subset is a special set of aircraft state parameters, specifically including aircraft real-time speed, altitude, heading angle, attitude angle, acceleration, remaining fuel quantity, and navigation system alarm state, and other core indicators of aircraft operation. Such parameters are transmitted in real time through the data link of the aircraft onboard sensor and the ground control system, and are classified and stored according to the unique identifier of the aircraft with a timestamp as the index. The second data subset is a special set of meteorological parameters, including wind speed, wind direction, temperature, humidity, air pressure, visibility, precipitation intensity, thunderstorm activity area coordinates, and turbulence intensity level in the target area. Such data is collected by multiple nodes of meteorological satellites, ground weather radars, and unmanned aerial vehicles equipped with micro weather stations, and is aligned according to the grid airspace coordinates to form a time-space continuous weather field data. The third data subset is a special set of airspace occupation parameters, including the aircraft density distribution heat map in the airspace, the boundary geofence coordinates of the temporary control area, the three-dimensional space coordinates and height limit of obstacles such as high-rise buildings, high-voltage transmission towers, and mountains, and the time-space occupation state of other flight plan conflict areas. Such parameters are generated by fusing the dynamic topological map of the airspace management system and the scanning data of the air traffic control radar. Specifically, in the extraction process, the original data needs to be classified and filtered according to the parameter type label. For example, aircraft state parameters are separated by analyzing the protocol identifier in the data packet, the second data subset is extracted using special coding rules for meteorological data, and the third data subset is obtained based on the data interface protocol of the airspace management system, ensuring the integrity and independence of each subset data, and providing structured input for subsequent anomaly detection.

[0018] Step S220: dynamically threshold filtering the first data subset, the second data subset, and the third data subset respectively to obtain abnormal aircraft data in the first data subset that exceeds a preset aircraft state threshold, abnormal meteorological data in the second data subset that exceeds a preset meteorological fluctuation threshold, and abnormal airspace data in the third data subset that exceeds a preset airspace capacity threshold.

[0019] Exemplarily, the dynamic threshold screening mechanism compares each data subset against preset aircraft state safety thresholds, meteorological fluctuation tolerance thresholds, and airspace capacity limitation thresholds to identify abnormal data deviating from normal ranges. The preset aircraft state thresholds include maximum allowable deviation values of flight speed (e.g., ±15% of the route planning speed), minimum safety limits of flight height (e.g., 100 meters of vertical buffer distance from ground obstacles), sudden deviation tolerance of heading angle (e.g., angle change of more than 10 degrees within 3 seconds), and critical warning values of fuel reserves (e.g., flight time supported by remaining fuel less than 15 minutes); the preset meteorological fluctuation thresholds cover instantaneous mutation gradient of wind speed (e.g., wind speed increase of 5 meters / second within 10 seconds), sudden drop rate of visibility (e.g., 500 meters per minute), fluctuation intensity of air pressure (e.g., air pressure change of more than 2 hundred pascals within 1 minute), and dangerous level of turbulence intensity (e.g., medium or above turbulence duration of more than 30 seconds); and the preset airspace capacity thresholds include maximum aircraft density in a unit airspace grid (e.g., no more than 5 aircrafts per cubic kilometer), upper limit of overlap degree of temporary control area and flight path (e.g., path overlap ratio of more than 60%), and safety factor of dispersion degree of obstacle distribution (e.g., obstacle spacing less than 1.5 times of the minimum avoidance distance). In the screening process, if the flight speed in the first data subset exceeds the maximum allowable deviation value for 3 consecutive sampling periods, the abnormal aircraft data is marked; if the wind speed in the second data subset suddenly changes by more than the preset threshold between two adjacent collection time points, the abnormal meteorological data is determined; and if the aircraft density heat map in the third data subset shows that the number of aircrafts in a certain airspace grid exceeds the capacity limit, the abnormal airspace data is generated. Through layer-by-layer threshold comparison and multi-period data sliding window verification, the accuracy of abnormal data determination and the anti-noise interference ability are ensured.

[0020] Step S230: input the abnormal aircraft data, abnormal meteorological data, and abnormal airspace data into the event type association model to determine the priority and association parameters of the emergency event type.

[0021] Exemplarily, the event type association model is a multi-modal fusion classification model trained based on a historical emergency event case base, which outputs the emergency event type and its handling priority by analyzing the similarity of the combined features of abnormal data and historical event patterns. The model input includes abnormal aircraft data (such as flight altitude drop sequence), abnormal weather data (such as strong side wind speed mutation gradient), and abnormal airspace data (such as obstacle distribution dispersion exceeding standard). First, the three types of abnormal data are standardized: aircraft state parameters are normalized by maximum-minimum mapping to the [0, 1] interval, weather parameters are standardized by Z-score to eliminate dimensional differences, and airspace parameters are segmented linearly scaled to adapt to different airspace scales. Subsequently, the model converts the standardized abnormal data into feature vectors: the aircraft feature vector is composed of speed deviation, altitude offset, and heading angle offset rate; the weather feature vector includes wind speed mutation gradient, visibility decline rate, and air pressure fluctuation intensity; and the airspace feature vector integrates airspace occupation density, control region overlap, and obstacle dispersion. After generating a comprehensive abnormal feature matrix by multi-dimensional splicing, the model uses a hierarchical matching mechanism to calculate the similarity (such as cosine similarity or dynamic time warping algorithm) with the feature matrix in the historical case base, filters the highest matching historical case, and extracts its event type identifier (such as "aircraft loss of control-weather mutation compound event"), handling priority weight (such as resource emergency scheduling level), and resource association parameters (such as required rescue equipment type and quantity). Further, the model dynamically corrects the priority weight in combination with the dynamic change trend of real-time data (such as the increasing rate of continuous flight altitude decline), for example, upgrading the originally planned secondary event to a primary event and adding the fuel supply requirement in the associated parameters, thereby generating an emergency event classification result closely adapted to the current situation.

[0022] As an implementation, in step S230, the abnormal aircraft data, abnormal weather data, and abnormal airspace data are input into the event type association model to determine the priority and associated parameters of the emergency event type, which can specifically include: in step S231, the abnormal aircraft data, abnormal weather data, and abnormal airspace data are respectively standardized to generate a first standardized data set corresponding to the abnormal aircraft data, a second standardized data set corresponding to the abnormal weather data, and a third standardized data set corresponding to the airspace data.

[0023] The data standardization process aims to eliminate the dimensional differences and numerical range differences of different parameters, so that multi-source heterogeneous data can be fused and analyzed on a unified scale. For abnormal aircraft data, the standardization process first calculates the deviation of each parameter (i.e. the difference between the actual value and the standard value) according to the normal range of the preset aircraft state parameters (such as the standard range of flight speed is 200-300 knots, the safe range of flight altitude is 500-1500 meters, and the stable fluctuation range of heading angle is ±5 degrees), and maps the deviation to the [0, 1] interval through the max-min normalization method to generate a first standardized data set. For example, the flight speed deviation is the difference between the actual speed and the planned speed, if the actual speed is 320 knots and the planned speed is 280 knots, the deviation is +40 knots, which is converted to 0.8 after normalization (assuming the maximum allowed deviation is 50 knots); for abnormal weather data, the Z-score standardization method is used, based on the mean and standard deviation of historical weather data, the wind speed mutation gradient (such as an increase of 5 meters / second in 10 seconds), the visibility decline rate (such as a decrease of 500 meters per minute), and the air pressure fluctuation intensity (such as a change of 2 hundred pascals in 1 minute) are converted to standardized values conforming to the normal distribution to generate a second standardized data set; for abnormal airspace data, the airspace occupancy density (such as 8 aircrafts per square kilometer, exceeding the preset threshold of 5 aircrafts), the temporary control area overlap degree (such as a flight path overlapping with a control area by 70%), and the obstacle distribution dispersion degree (such as an obstacle spacing less than 1.2 times the safety distance) are adjusted to a unified scale according to the airspace size and geographic coordinate range through the piecewise linear scaling method to generate a third standardized data set. The standardized data set has comparability and model input compatibility, providing a basis for subsequent feature vector construction.

[0024] Step S232: According to the aircraft state parameter type in the first standardized data set, an aircraft feature vector containing flight speed deviation, altitude offset, and heading angle offset rate is generated; according to the weather parameter type in the second standardized data set, a weather feature vector containing wind speed mutation gradient, visibility decline rate, and air pressure fluctuation intensity is generated; according to the airspace parameter type in the third standardized data set, an airspace feature vector containing airspace occupancy density, temporary control area overlap degree, and obstacle distribution dispersion degree is generated.

[0025] The eigenvector construction is a key step to convert the normalized parameter sequence into a structured feature expression. The aircraft eigenvector is composed of flight speed deviation (such as 0.8 normalized), height offset (such as the difference between the current height and the safe height normalized to 0.6), and heading angle offset rate (such as the change rate of the heading angle in the next 3 seconds normalized to 0.7) in a fixed order, forming a vector with a dimension of 3; the weather eigenvector integrates wind speed mutation gradient (such as 1.2 normalized by Z-score), visibility decline rate (normalized to -0.5), and air pressure fluctuation intensity (normalized to 0.9), forming a vector with a dimension of 3; the airspace eigenvector contains airspace occupation density (such as 0.75 scaled), temporary control area overlap (0.88 scaled), and obstacle distribution dispersion (0.63 scaled), also forming a vector with a dimension of 3. Specifically, each element in the aircraft eigenvector represents the abnormality degree of a specific aircraft state parameter, the weather eigenvector reflects the influence intensity of weather mutation on flight safety, and the airspace eigenvector quantifies the severity of airspace resource conflict. For example, when the aircraft eigenvector is [0.8, 0.6, 0.7], the weather eigenvector is [1.2, -0.5, 0.9], and the airspace eigenvector is [0.75, 0.88, 0.63], it can represent a complex abnormal scenario of high aircraft speed, low height, unstable heading, accompanied by strong wind speed mutation and low visibility, and high airspace congestion and control area overlap.

[0026] Step S233: The aircraft eigenvector, weather eigenvector, and airspace eigenvector are multi-dimensionally spliced to generate a comprehensive abnormal feature matrix, and the comprehensive abnormal feature matrix is input into a pre-trained event type association model; wherein the event type association model is trained based on the mapping relationship between the same type of eigenvectors extracted from historical emergency event cases and event type labels.

[0027] The multi-dimensional concatenation constructs a comprehensive abnormal feature matrix with dimension 3x3 or 9x1 by stacking three feature vectors row by row or column by column. For example, if the aircraft feature vector is [0.8, 0.6, 0.7], the meteorological feature vector is [1.2, -0.5, 0.9], and the airspace feature vector is [0.75, 0.88, 0.63], the row concatenation generates a 3x3 matrix [[0.8, 0.6, 0.7], [1.2, -0.5, 0.9], [0.75, 0.88, 0.63]], or the column concatenation generates a 9x1 vector [0.8, 0.6, 0.7, 1.2, -0.5, 0.9, 0.75, 0.88, 0.63]. The matrix is input into the pre-trained event type association model, which, based on the mapping relationship between similar feature matrices and event type labels (such as "aircraft out of control-strong crosswind combined event") in the historical emergency event case library, performs pattern matching through a deep neural network or an ensemble learning algorithm. Each case in the historical case library contains a feature matrix, an event type identifier (such as a unique code E-1024), a processing priority weight (such as a first priority corresponding to resource emergency scheduling), and resource association parameters (such as a requirement for 2 fire-fighting drones). The model establishes a mapping rule from the feature matrix to the event type and parameters through supervised learning, ensuring that the current comprehensive abnormal feature matrix can trigger similar classification logic as the historical cases.

[0028] Step S234: Through the hierarchical matching mechanism in the event type association model, the comprehensive abnormal feature matrix is compared with the feature matrices in the historical emergency event cases for similarity, at least one historical case with the highest matching degree to the comprehensive abnormal feature matrix is determined, and the event type identifier, processing priority weight, and resource association parameters recorded in the historical case are extracted.

[0029] The hierarchical matching mechanism adopts a multi-stage similarity calculation strategy: first, the Euclidean distance or cosine similarity is used in the first layer to quickly screen out the top K (such as K=10) candidate historical cases; in the second layer, the dynamic time warping (DTW) algorithm or Manhattan distance is used to refine the comparison, considering the time series trend of the feature matrix (such as the continuous increase of flight speed deviation); finally, in the third layer, the context compatibility of the candidate cases is verified by the expert rule engine (such as whether the current airspace allows the same type of resource scheduling). For example, if the cosine similarity between the current comprehensive abnormal feature matrix and the feature matrix of historical case C-205 is 0.92 (the highest), and the similarity with C-198 is 0.85, the event type identifier "aircraft fuel warning-airspace congestion combined event", the processing priority weight "level one", and the resource association parameter "fuel supply unmanned aerial vehicle 2, temporary navigation beacon 1 group" of C-205 are extracted as priority. In the matching process, the model calculates the similarity confidence (such as 0.92 corresponds to a confidence of 95%) at the same time, and only when the confidence exceeds the preset threshold (such as 80%), is the matching effective.

[0030] Step S235: According to the event type identifier of the historical case with the highest matching degree, determine the basic classification label of the current emergency event type; according to the processing priority weight and the resource association parameter, and combining the real-time change trend of the aircraft feature vector in the comprehensive abnormal feature matrix, dynamically adjust the priority level and the associated parameter combination corresponding to the basic classification label.

[0031] The basic classification label is inherited from the event type identifier of the matching historical case, such as "aircraft out of control-strong crosswind combined event". The initial value of the priority level is the processing priority weight in the historical case (such as level two), but it needs to be dynamically adjusted according to the parameter change trend in the real-time aircraft feature vector: if the flight speed deviation increases continuously in the next 3 sampling periods (such as from 0.8 to 0.9), the priority is raised from level two to level one; if the "number of firefighting unmanned aerial vehicles" in the resource association parameter is 2 in the historical case, but the current obstacle distribution dispersion is higher (such as from 0.63 to 0.75), an additional unmanned aerial vehicle is added to cope with the complex airspace environment. Dynamic adjustment is achieved through a preset weight correction coefficient, for example, the priority promotion coefficient is 0.2 (the coefficient is accumulated every period the deviation increases by 0.1), and the resource quantity correction formula is "baseline number x (1+dispersion increment / 0.1)".

[0032] Step S236: Based on the adjusted priority level and associated parameter combination, generate a structured output result containing the emergency event type, priority order, and resource association rule, and dynamically bind the structured output result with the calling conditions of the flight emergency prediction model.

[0033] The structured output result is packaged in JSON or XML format, containing the fields "Event Type: Uncontrolled Aircraft - Strong Crosswind Compound Event", "Priority: Level 1", "Resource Requirement: 3 Firefighting Drones, 1 Set of Temporary Navigation Beacons", and "Processing Time Window: 5 Minutes". The dynamic binding process verifies whether the invocation conditions of the flight emergency prediction model meet the resource type (e.g., whether the model supports "firefighting drone dispatch"), the time window (e.g., whether the model can generate a strategy within 5 minutes), and the airspace constraints (e.g., whether the model is compatible with the current airspace regulation). For example, if the invocation conditions of model M-1024 include "resource type = firefighting drone" and "time window ≤ 5 minutes", the model is bound with the structured output result; if model M-1025 only supports "time window ≥ 10 minutes", the model is excluded. The binding result generates a model invocation queue, ensuring that only qualified prediction models are triggered in the subsequent steps.

[0034] Step S240: According to the priority and associated parameters, generate an event feature label containing the emergency event type, and match the event feature label with the invocation conditions of the flight emergency prediction model.

[0035] The event feature label is a structured data object used to encapsulate the core attributes of the emergency event type and its handling constraints. The priority field reflects the urgency of event handling, for example, a first-level priority indicates an aircraft out-of-control event that needs immediate response, and a second-level priority corresponds to a weather disturbance event that can be delayed for handling. The associated parameter field contains resource type requirements (such as fire-fighting drones, temporary navigation beacons), handling time windows (such as requiring path adjustment to be completed within 5 minutes), and airspace impact ranges (such as a 2-kilometer radius control area). When generating the label, first parse the resource type requirement parameter in the associated parameter, refine the resource identifier to a specific device model and deployment location (such as "Model A fire-fighting drone needs to be deployed to coordinates X, Y"), and sort the resource priority according to the emergency resource scheduling rules (such as preferentially scheduling the closest available device); at the same time, based on the time window limit in the handling time parameter (such as generating command instructions within 30 seconds), the time-sensitive identifier is expanded to time-sensitive classification (such as high sensitivity requiring real-time feedback) and time-sensitive superposition effect (such as cumulative delay control of multiple instruction parallel execution); further combined with the obstacle distribution density in the airspace impact range parameter, the impact range identifier is mapped to a spatial weight distribution matrix (such as higher avoidance weight needs to be allocated in high-density obstacle areas) and regional linkage relationship topology (such as the coordinated effective condition of adjacent airspace control instructions). Finally, through logical splicing, a composite event feature label containing resource allocation logic, time constraint logic, and spatial coverage logic is generated, for example, "first-level priority-aircraft fuel alarm event: 2 fuel supply drones need to be dispatched to coordinates P, Q within 3 minutes, and other aircraft are prohibited from entering the airspace of height layers H1 to H2". When the label is matched with the calling conditions of the flight emergency prediction model, it needs to verify the consistency of the model input parameter type with the resource type, time window, and spatial range in the label, for example, only call the prediction model that supports the "fuel supply" resource type and is compatible with the "3-minute response" time constraint, to ensure the precise adaptation of the model output strategy to the event handling requirements.

[0036] As an implementation, in step S240, an event feature label containing an emergency event type is generated according to the priority and the associated parameter, which can specifically include: in step S241, based on the level division result in the priority, the resource type requirement parameter, the handling time parameter, and the airspace impact range parameter in the associated parameter are parsed to generate an initial label component corresponding to the emergency event type. The initial label component contains a resource type identifier, a time constraint identifier, and an impact range identifier.

[0037] The construction of the initial label component is indexed by priority levels, and the resource type requirement parameters (such as "firefighting drones, fuel supply equipment") in the associated parameters, the processing time limit parameters (such as "the path adjustment needs to be completed within 5 minutes") and the airspace influence range parameters (such as "a control area with a radius of 2 kilometers") are parsed to generate basic label elements. The resource type identifier is a standardized coded string, for example, "RES-FIRE-DRONE-001" represents a firefighting drone with model number 001; the time limit constraint identifier adopts the combination form of time stamp interval and time sensitivity mark, for example, "T_WINDOW: [T0, T0+300 seconds], SENSITIVITY: HIGH" represents a 5-minute response window with high sensitivity; the influence range identifier defines the spatial range through the geographic fence coordinates and vertical height limit, for example, "GEO_ZONE: POLYGON((x1, y1), (x2, y2),...), ALTITUDE: [100 meters, 500 meters]" represents an airspace area with a polygon coordinate as the horizontal boundary and a vertical height limit of 100 meters to 500 meters. Specifically, in the parsing process, the device name and model number in the resource type requirement parameter need to be mapped to the pre-defined resource coding library to ensure the uniqueness of the identifier; the processing time limit parameter needs to be converted into an absolute timestamp sequence and synchronized with the system clock; the airspace influence range parameter is converted into a standardized spatial description format through a geographic information system (GIS) interface. The initial label component is stored in the form of a tuple or a key-value pair, for example, "{resource: RES-FIRE-DRONE-001, time limit: T_WINDOW, space: GEO_ZONE}", which provides structured input for subsequent refinement and expansion.

[0038] Step S242: According to the emergency resource scheduling rules in the resource type requirement parameters, the resource type identifier in the initial label component is refined to generate an updated resource type identifier containing resource priority sorting and resource conflict avoidance strategies.

[0039] The refinement process extends the deployment priority, scheduling path and conflict avoidance conditions of the resource type identifier according to the strategies in the emergency resource scheduling rule base. For example, the initial resource type identifier "RES-FIRE-DRONE-001" is refined into "RES-FIRE-DRONE-001-P1 (priority 1, deployed to coordinate P1, closest to the current airspace)" and "RES-FIRE-DRONE-001-P2 (priority 2, deployed to coordinate P2, backup node)" through the "distance priority" strategy in the rule base; at the same time, if the conflict equipment is included in the resource type requirement parameters (such as the need to call fire unmanned aerial vehicle and weather monitoring unmanned aerial vehicle at the same time), the conflict avoidance strategy "ALT_AVOID: H_LAYER ≠ 300 meters" is added to prohibit the operation of two types of equipment at the same height layer. The refined resource type identifier forms an executable resource scheduling instruction by adding priority markers, deployment coordinates and conflict rules, such as "resource type identifier updated to: {RES-FIRE-DRONE-001-P1@P1, RES-WEATHER-DRONE-005-P2@P2, AVOID_H_LAYER=300 meters}", ensuring the compatibility of resource scheduling logic and airspace control rules.

[0040] Step S243: Based on the time window limit condition in the processing timeliness parameter, the timeliness constraint identifier in the initial tag component is dynamically extended to generate an updated timeliness constraint identifier containing time sensitivity classification and timeliness superposition effect.

[0041] Dynamic extension upgrades the simple time window of the initial time constraint identifier to a multi-dimensional time logic description by introducing a time sensitivity grading mechanism and an age overlap effect model. For example, the initial identifier "T_WINDOW: [T0, T0+300 seconds]" is extended to "T_SENSITIVITY: HIGH (red alert, real-time monitoring), T_SUB_TASKS: {path planning: [T0, T0+60 seconds], resource scheduling: [T0+60 seconds, T0+200 seconds], instruction issuance: [T0+200 seconds, T0+300 seconds]}, T_OVERLAP_TOLERANCE: ≤10 seconds (allowing sub-task overlap delay)". The time sensitivity grading is automatically associated according to event priority, for example, a first priority event corresponds to "HIGH" sensitivity, requiring real-time feedback; the time overlap effect ensures that the overall response time meets the window constraints by calculating the upper limit of the cumulative delay when multiple tasks are executed in parallel (such as a total delay of no more than 10 seconds). The extended time constraint identifier is stored through a time tree structure or a hierarchical key-value pair, such as "{main window: T0-T0+300 seconds, sub-task: [planning, scheduling, issuance], overlap tolerance: 10 seconds}", providing precise control logic for time-driven instruction execution.

[0042] Step S244: Based on the area overlap data in the airspace influence range parameter and the obstacle distribution density, the influence range identifier in the initial label component is multi-dimensionally mapped to generate an updated influence range identifier containing space weight allocation and area linkage relationship.

[0043] Multi-dimensional mapping upgrades the static geographic description of the initial influence range identifier to a dynamic space strategy through a space weight allocation algorithm and a region linkage rule engine. For example, the initial identifier "GEO_ZONE: POLYGON((x1,y1),...), ALTITUDE: [100 meters, 500 meters]" is mapped to "SPACE_WEIGHT: ZONE_A=0.8 (high priority for avoidance), ZONE_B=0.3 (low priority)" in combination with obstacle distribution density data (such as an obstacle density of 0.8 per square kilometer in area A); at the same time, if there is an airspace control linkage relationship between area A and adjacent area C (such as area A being activated and area C entering a monitoring state), the linkage rule "LINKED_ZONES: ZONE_A→ZONE_C (monitoring mode: preliminary control)" is added. The updated influence range identifier forms a dynamic avoidance strategy and area coordination mechanism through a space weight matrix and a linkage relationship table, such as "space identifier updated to: {weight: ZONE_A=0.8, ZONE_B=0.3, linkage: ZONE_A→ZONE_C}", providing decision-making basis for adaptive allocation of airspace resources.

[0044] Step S245: The updated resource type identifier, the updated time constraint identifier and the updated impact range identifier are logically spliced to form a composite event feature label containing resource allocation logic, time constraint logic and spatial coverage logic.

[0045] The logical splicing integrates the refined parameters of the three types of identifiers into a unified event feature label through logical operators and conditional expressions. For example, the "RES-FIRE-DRONE-001-P1@P1" in the resource type identifier is combined with the conflict avoidance rule "AVOID_H_LAYER=300 meters", the "T_SENSITIVITY: HIGH" in the time constraint identifier and the "SPACE_WEIGHT: ZONE_A=0.8" in the spatial identifier through logical AND relationship to form a composite logical statement: "IF resource=RES-FIRE-DRONE-001-P1@P1 AND time sensitivity=HIGH AND spatial weight≥0.8 THEN execute avoidance strategy: ALTITUDE≠300 meters". The composite event feature label is stored in structured text or XML / JSON format, for example "{resource allocation logic: schedule RES-FIRE-DRONE-001-P1 to P1 and avoid 300 meters height layer, time constraint logic: complete within T0+300 seconds and subtask overlap delay≤10 seconds, spatial coverage logic: preferentially avoid weight≥0.8 area and coordinate monitoring area C}", ensuring the independence and resolvability of each logical unit.

[0046] Step S246: According to the compatibility verification result of each logical unit in the composite event feature label, adjust the conflict part of the internal parameters of the label, and inject the unique identification code of the emergency event type to generate an event feature label matching the call conditions of the flight emergency prediction model.

[0047] Compatibility verification detects conflicts in resource, time, and space logic through a rule engine. For example, if the resource scheduling path "RES-FIRE-DRONE-001-P1@P1" overlaps with the spatial weight area "ZONE_A=0.8" and the area prohibits the entry of the aircraft, a conflict alert is triggered, and the resource deployment coordinates are automatically adjusted to "P1'" (outside ZONE_A); at the same time, if there is a conflict between the time window "T0+300 seconds" and the resource arrival time "T0+250 seconds" (e.g., the resource does not arrive on time), the time window is dynamically extended to "T0+350 seconds". After the conflict is resolved, a unique identification code (e.g., "EVENT-ID-20231105001") is injected into the label, and the final event feature label is generated: "EVENT-ID-20231105001: resource=RES-FIRE-DRONE-001-P1@P1', time=T0+350 seconds, space=outside ZONE_A and weight≥0.5, compatibility status=verification passed". The unique identification code is created through a hash algorithm or a serial number generator, ensuring global uniqueness and traceability.

[0048] Step S247: Through the resource allocation logic, time constraint logic, and space coverage logic in the event feature label, a mapping relationship between the emergency event type and the target flight emergency prediction model input parameters is established, ensuring the parameter consistency of the event feature label in the model calling stage.

[0049] The mapping relationship is achieved through a parameter matching table, which one-to-one corresponds the logical units in the event feature label to the model input parameters. For example, "RES-FIRE-DRONE-001-P1@P1'" in the resource allocation logic is mapped to "resource_type=fire_drone, deploy_coord=P1'" in the model input parameters; "T0+350 seconds" in the time constraint logic is mapped to "time_limit=350"; and "ZONE_A outside and weight≥0.5" in the space coverage logic is converted to "avoid_zones=ZONE_A, priority_weight=0.5". If the input conditions of the target flight emergency prediction model M-1024 require "resource_type=fire_drone, time_limit≤400, avoid_zones≠NULL", the parameters of the event feature label completely match those of the model, triggering the model call. In the mapping process, the consistency of parameter types, value ranges, and logical constraints needs to be verified, such as whether the time limit is an integer value and whether the avoidance area coordinates are in a valid geographic format, to ensure the legality of the model input and the reliability of the prediction results.

[0050] Step S300: calling a flight emergency prediction model corresponding to the emergency event type; the flight emergency prediction model is trained based on the filtered flight features and airspace features in the historical emergency event data, and the types of input parameters and output parameters of each flight emergency prediction model are consistent in the training stage and the application stage.

[0051] Exemplarily, the flight emergency prediction model is a machine learning model customized for a specific emergency event type, and the training data of the model comes from the flight feature set and the airspace feature set filtered from historical emergency event cases. The flight feature set includes the speed adjustment records, height correction trajectories, heading angle recovery rates, and other parameters directly related to the flight path of the aircraft in historical events; the airspace feature set includes the dynamic change data of airspace capacity, obstacle avoidance path planning records, and temporary regulation area adjustment strategies during the handling of historical events. In the model training stage, redundant features with an association degree with the emergency handling results lower than a preset threshold (such as irrelevant environmental noise data) are removed through redundancy analysis, and the core flight features and the core airspace features are retained as input variables, while the flight path adjustment strategies and emergency resource allocation strategies verified effective in historical events are taken as output variables, to ensure the structural consistency of the model input and output in the training and application stages. For example, for the “low-altitude thunderstorm weather avoidance” event type, the model input is the thunderstorm area wind speed time series data, the current height of the aircraft, and the available avoidance path of the airspace, and the output is the climb height instruction, the fly-around path coordinates, and the weather radar scanning frequency adjustment parameters. When calling the model, the target model with the highest response weight and the smallest error range is selected from the candidate model pool according to the event feature label, and the real-time abnormal data is reorganized into an input data set in the standardized format in the training stage, to trigger the model to perform path simulation and resource demand prediction.

[0052] As an implementation, step S300, calling a flight emergency prediction model corresponding to the emergency event type, can specifically include: step S310: obtaining a plurality of candidate flight emergency prediction models associated with the event feature label; each candidate flight emergency prediction model corresponds to a different emergency event handling stage, and the type of input parameter used by each candidate flight emergency prediction model in the training stage is consistent with the type of parameter in the current real-time flight monitoring data.

[0053] The candidate flight emergency prediction model is generated by screening from a pre-defined model pool through resource allocation logic, time constraint logic and spatial coverage logic in the event feature label. For example, if the event feature label contains "resource type identifier = fire drone model A-001", "time constraint identifier = 5-minute response window" and "spatial coverage identifier = avoidance area ZONE_A", all models in the model pool that support fire drone scheduling, 5-minute time window and ZONE_A area avoidance rules are listed as candidates. Each candidate model corresponds to a specific emergency handling stage, for example, model M-1024 is dedicated to path planning in the flight path emergency avoidance stage, and the input parameters include aircraft speed, height, heading angle and obstacle coordinates; model M-1025 is dedicated to demand prediction in the resource scheduling stage, and the input parameters include resource type, deployment coordinates and time window. The input parameter type of the candidate model must be strictly consistent with the parameters in the current real-time flight monitoring data, for example, if the real-time data contains aircraft speed deviation, wind speed mutation gradient and air space occupation density, the training data of the candidate model must contain the same parameters to ensure the structural compatibility of input and output. Through the matching of event feature label and model metadata, for example, the comparison of "resource type identifier" in the label and "supported resource type" field in the model metadata, all candidate models that meet the conditions are selected to form an initial candidate queue.

[0054] Step S320: According to the historical verification results of the candidate flight emergency prediction model, the response weight and error range of each candidate flight emergency prediction model are determined.

[0055] The historical verification result is obtained by evaluating the performance of the model in historical emergency event cases, including response speed, prediction accuracy and resource scheduling success rate. The response weight reflects the priority of the model in similar events, for example, model M-1024 successfully generates an effective path strategy 95 times in 100 historical verifications, and its response weight is 0.95; the error range is calculated by the deviation between the prediction result and the actual disposal result, for example, the average absolute error of the predicted resource demand quantity and the actual use quantity of model M-1025 is 8%, and its error range is 8%. Specifically, the response weight is calculated by weighted average method, considering that the performance of the model in the recent period is more important (such as the verification results in the last 30 days account for 70% weight); the error range is quantified by root mean square error (RMSE) or mean absolute percentage error (MAPE). For example, the historical response weight of model M-1024 is 0.92 (9 times successful in the last 10 verifications), and the error range is 5%; the response weight of model M-1025 is 0.85, and the error range is 12%. The weight and error data are stored in the model performance database, and are obtained and updated to the candidate model queue through a real-time query interface.

[0056] Step S330: Select at least one target flight emergency prediction model from the candidate flight emergency prediction models, which has a response weight higher than a preset weight threshold and an error range lower than a preset error threshold.

[0057] The preset weight threshold and error threshold are dynamically adjusted according to the safety level of the emergency event type. For example, a first priority event requires a response weight ≥ 0.9 and an error range ≤ 10%, and a second priority event is relaxed to a weight ≥ 0.8 and an error ≤ 15%. In the screening process, the candidate models are sorted in descending order of response weight, and the models with error exceeding the limit are excluded. For example, the candidate queue contains models M-1024 (weight 0.92, error 5%), M-1025 (weight 0.85, error 12%), and M-1026 (weight 0.88, error 9%), and if the preset threshold is weight ≥ 0.85 and error ≤ 10%, M-1025 is excluded due to error 12%, and the final target model is M-1024 and M-1026. If multiple models meet the conditions, a single model or a combination of models is selected according to the scene requirements, for example, a high complexity event triggers multiple model collaborative prediction (such as M-1024 is responsible for path planning, and M-1026 is responsible for resource conflict detection), while a simple event only calls a single model. The selection result of the target model forms the final call list and is injected into the event processing pipeline.

[0058] Step S340: Recombine the abnormal aircraft data, abnormal weather data and abnormal airspace data according to the input format of the target flight emergency prediction model in the training phase to generate a standardized input data set.

[0059] Data reorganization must strictly follow the training data format of the target model, including parameter order, data dimension and standardization rules. For example, the input format of model M-1024 requires a four-dimensional vector of [flight speed deviation, height offset, heading angle deviation rate, obstacle distribution dispersion]. The flight speed deviation in the real-time abnormal data is 0.8, the height offset is 0.6, the heading angle deviation rate is 0.7, and the obstacle distribution dispersion is 0.75, so the vector [0.8, 0.6, 0.7, 0.75] is reorganized. If the model input contains time series data (such as wind speed mutation gradient of the past 5 sampling periods), the real-time abnormal weather data needs to be aligned according to the time window sliding to generate a time series matrix. For example, the wind speed mutation gradient sequence in the abnormal weather data is [0.5, 0.7, 0.9, 1.1, 1.3], which is reorganized into [[0.5, 0.7, 0.9], [0.7, 0.9, 1.1], [0.9, 1.1, 1.3]] according to the model's requirement of 3-order time window. The matching of parameter type and dimension needs to be verified during data reorganization, such as whether the airspace occupancy density parameter is a floating point number, whether the time series length meets the model requirements, to ensure the integrity and legality of the input data.

[0060] Step S350: input the standardized input dataset into the target flight emergency prediction model, trigger the target flight emergency prediction model to perform flight path simulation and resource demand prediction, obtain flight path adjustment strategy and emergency resource allocation strategy.

[0061] The target model performs prediction logic according to the standardized input dataset, for example, the model M-1024 generates a heading correction angle, a target height layer, and a speed control curve through neural network inference to form a flight path adjustment strategy "heading + 15 degrees, climb to 600 meters, speed down to 250 knots"; the model M-1026 predicts the number of fire unmanned aerial vehicles required, deployment coordinates, and supply cycle through regression analysis to generate an emergency resource allocation strategy "dispatch 2 unmanned aerial vehicles to coordinates (X1, Y1), supply every 30 minutes". During the prediction process, the conflict detection module built in the model checks the space-time compatibility of the path and the resource, for example, whether the heading correction angle conflicts with the unmanned aerial vehicle deployment area, if there is a conflict, trigger strategy iteration optimization, until a conflict-free solution is generated. Finally, the output strategy is packaged into a structured instruction set, for example, the flight path adjustment strategy includes a waypoint sequence [WPT1, WPT2, WPT3], a height instruction [HOLD 600 meters], and a speed instruction [MAINTAIN 250 knots]; the emergency resource allocation strategy includes a resource scheduling queue [DRONE-001→(X1, Y1), DRONE-002→(X2, Y2)] and a time trigger [T0+1800 seconds supply]. The strategy data is encoded through a standardized protocol (such as JSON) and transmitted to the instruction execution module to complete the closed loop from model prediction to actual control.

[0062] As an embodiment, the training process of the flight emergency prediction model includes the following steps: step S10: obtaining a historical emergency event dataset; the historical emergency event dataset includes a plurality of processed emergency event cases, each emergency event case contains aircraft parameters, meteorological parameters, airspace parameters at the time of emergency event occurrence, and path adjustment records and resource allocation records after emergency event processing.

[0063] The historical emergency event dataset is constructed by integrating the archived data of the air traffic control system, meteorological monitoring database, and airspace management platform. Each case covers complete data of the emergency event handling period. Aircraft parameters include real-time flight speed, height, heading angle, attitude angle, fuel remaining, and navigation warning status at the time of the event. For example, in a "aircraft fuel warning event" case, the aircraft parameters are recorded as "speed 280 knots, height 600 meters, heading angle deviation +12 degrees, and fuel remaining 10%". Meteorological parameters include wind speed, wind direction, visibility, air pressure, and turbulence intensity time series data in the event occurrence area. For example, "wind speed sudden gradient is 5 meters / second, and visibility decreases to 500 meters within 30 minutes". Airspace parameters record airspace occupancy density, obstacle distribution coordinates, and temporary control area boundaries. For example, "obstacle distribution dispersion is 0.8 (high-density area accounts for 80%), and control area overlap with flight path is 70%". Path adjustment records detailed description of the actual executed heading correction angle, climb / descent instruction, and speed control curve. For example, "heading +15 degree correction, climb to 800 meters, and speed decrease to 250 knots". Resource allocation records include dispatched equipment type (such as fire unmanned aerial vehicle model A-001), deployment coordinates, supply period, and execution effect score (such as resource arrival time deviation ≤30 seconds). The dataset is aligned with time stamp and airspace coordinates to ensure the spatio-temporal consistency of multi-source data, providing high-quality input for feature extraction and model training.

[0064] Step S20: Extract flight features related to flight speed, height deviation, and heading angle deviation from the aircraft parameters of each emergency event case, and extract airspace features related to airspace capacity, obstacle distribution, and temporary control area from the airspace parameters.

[0065] The extraction of flight feature set focuses on the key indicators of aircraft dynamic behavior: flight speed feature is represented by the deviation amount (such as +20 knots) between the actual speed at the time of the event and the planned speed; height deviation is the difference between the actual height and the safe height (such as -100 meters); heading angle deviation is quantified by the standard deviation or maximum instantaneous deviation rate (such as +10 degrees within 5 seconds) of the heading angle during the event duration. For example, in a certain case, the flight feature set is "speed deviation +20 knots, height deviation -100 meters, and heading angle deviation rate 0.8"; the airspace feature set is extracted from the airspace parameters, including airspace capacity (such as 6 aircrafts in a unit grid, exceeding the threshold of 5 aircrafts), obstacle distribution (such as the area with obstacle spacing less than 1.2 times the safe distance accounts for 60%), and the overlap degree of temporary control area and flight path (such as coordinate overlap ratio 75%). The feature extraction process removes noise data (such as sensor instantaneous outliers) through data cleaning rules, and generates time series features (such as average height deviation in the past 5 minutes) using sliding window statistical method, ensuring that the feature set fully reflects the event situation.

[0066] Step S30: Redundancy analysis is performed on the flight feature set and the airspace feature set to remove redundant features with a correlation degree lower than a preset correlation threshold with the emergency event handling result, to obtain an optimized target flight feature set and a target airspace feature set.

[0067] The redundancy analysis filters out core features by calculating the statistical correlation degree of the features and the emergency event handling result (such as path adjustment effect score, resource scheduling success rate). Pearson correlation coefficient is used to quantify linear correlation, for example, the correlation coefficient between flight speed deviation and path adjustment effect is 0.85 (strong positive correlation), and the correlation coefficient between fuel reserve and resource scheduling success rate is 0.1 (weak correlation); mutual information method is used to evaluate nonlinear correlation, for example, the mutual information value between obstacle distribution dispersion and avoidance path complexity is 0.7 (high correlation). The preset correlation threshold is 0.3 (features below this threshold are considered redundant), for example, the correlation coefficient between the airspace feature "historical usage frequency of temporary control area" and the handling result is 0.2, and it is removed. Further, through feature combination influence analysis, the synergistic effect of the remaining features is verified, for example, the joint contribution of the "height deviation + airspace capacity" combination to the path adjustment effect is 0.9, which is higher than the contribution of a single feature. Finally, the target flight feature set retains "speed deviation, height deviation, heading angle deviation rate", and the target airspace feature set retains "airspace capacity, obstacle distribution dispersion, temporary control area overlap", forming a simplified and high-explanation feature set.

[0068] As an implementation, in step S30, the redundancy analysis is performed on the flight feature set and the airspace feature set to remove redundant features with a correlation degree lower than a preset correlation threshold with the emergency event handling result, to obtain an optimized target flight feature set and a target airspace feature set, which can specifically include: step S31: calculating a first correlation coefficient between each flight feature in the flight feature set and the emergency event handling result, and a second correlation coefficient between each airspace feature in the airspace feature set and the emergency event handling result.

[0069] The first correlation coefficient and the second correlation coefficient quantify the influence degree of the flight features and the airspace features on the emergency event handling result through statistical analysis methods. The first correlation coefficient adopts a Pearson correlation coefficient or a Spearman rank correlation coefficient to measure the linear or nonlinear correlation between the flight features (such as flight speed deviation, height deviation, and heading angle deviation rate) and the handling result (such as path adjustment effect score and resource scheduling success rate). For example, the Pearson correlation coefficient of the flight speed deviation and the path adjustment effect is 0.85, indicating that there is a strong positive correlation between them; and the Spearman rank correlation coefficient of the fuel reserve and the resource scheduling success rate is 0.1, showing weak correlation or no significant correlation. The second correlation coefficient is also applied to the relationship between the airspace features (such as airspace capacity, obstacle distribution dispersion, and temporary control area overlap) and the handling result, for example, the correlation coefficient of the obstacle distribution dispersion and the avoidance path complexity is 0.7, indicating that a high dispersion area needs a more complex avoidance strategy. The correlation coefficient calculation is based on the feature values and the actual handling result records in the historical emergency event data set, and a complete correlation matrix is generated by traversing all feature-result pairs, providing a quantitative basis for subsequent feature screening.

[0070] Step S32: removing target flight features with a first correlation coefficient lower than a first correlation coefficient threshold from the flight feature set, and removing target airspace features with a second correlation coefficient lower than a second correlation coefficient threshold from the airspace feature set.

[0071] The preset first correlation coefficient threshold and second correlation coefficient threshold are set according to the safety fault tolerance requirements of the emergency event type, for example, the first correlation coefficient threshold is 0.3 (eliminating weakly correlated features with an absolute value <0.3), and the second correlation coefficient threshold is 0.25. The removal process traverses the correlation matrix, marks the features lower than the threshold as redundant, and performs a deletion operation. For example, “fuel reserve” in the flight feature set is eliminated because the first correlation coefficient is 0.1 (<0.3); “temporal control area historical use frequency” in the airspace feature set is removed because the second correlation coefficient is 0.2 (<0.25). The removal operation needs to verify the feature independence to avoid mistakenly deleting features with potential synergistic effects, for example, “heading angle deviation rate” and “height deviation” have low individual correlation coefficients, but their combination may have a significant impact on the handling result, at which time the deletion is temporarily suspended and enters the combination analysis stage.

[0072] Step S33: performing combination analysis on the remaining flight features and the remaining airspace features after removal to determine the influence weight of different feature combinations on the handling result.

[0073] The combined analysis uses a multiple regression model or a random forest algorithm to evaluate the combined contribution of the feature combination to the treatment result. For example, the combination of "flight speed deviation" and "height deviation" in the remaining flight features has a regression coefficient of 0.65 (0.45 and 0.40 respectively) for the path adjustment effect calculated by the multiple regression model, indicating that the combined feature has a higher impact weight; the combination of "airspace capacity" and "obstacle distribution dispersion" in the remaining airspace features has a feature importance score of 0.8 (0.5 and 0.6 respectively) output by the random forest algorithm, reflecting that the synergistic effect of the combined features is significant. The combined analysis generates a feature combination weight matrix, for example, the weight of "flight speed deviation + airspace capacity" is 0.75, and the weight of "height deviation + obstacle distribution dispersion" is 0.68, providing data support for screening core features.

[0074] Step S34: According to the impact weight, the top N core flight features are screened from the remaining flight features, and the top M core airspace features are screened from the remaining airspace features.

[0075] The weight ranking is based on the combined analysis results, and the top N and top M cutoff thresholds are set (e.g. N=3, M=2). For example, the "flight speed deviation" weight is 0.75, the "height deviation" weight is 0.68, and the "heading angle deviation rate" weight is 0.62 in the remaining flight features, ranking the top three; the "airspace capacity" weight is 0.80, and the "obstacle distribution dispersion" weight is 0.73 in the remaining airspace features, ranking the top two. The significance of the weight difference needs to be verified during the screening process, for example, using T-test to confirm whether there is a statistically significant difference (P<0.05) between the weight of the third-ranked "heading angle deviation rate" (0.62) and the fourth-ranked "remaining fuel alarm frequency" (0.55), if the difference is not significant, the cutoff range is expanded (e.g. N=4). The finally screened core flight features and core airspace features need to cover more than 80% of the variation of the treatment results in the historical cases to ensure the explanatory power of the feature set.

[0076] Step S35: Merge the core flight features and the core airspace features into the target flight feature set and the target airspace feature set.

[0077] The merging operation is achieved by aligning the feature names with the data structure, for example, the core flight features "flight speed deviation, height deviation, heading angle offset rate" are classified by parameter type with the core airspace features "airspace capacity, obstacle distribution dispersion", respectively forming the target flight feature set and the target airspace feature set. The merged feature set needs to meet the model input format requirements, for example, the time series features need to be uniformly sampled (e.g. once per second), and the spatial features need to be converted to a standard coordinate system (e.g. WGS-84). Verify the integrity of the merged feature set, for example, check if there are missing values or inconsistent dimensions (e.g. flight speed is knots, airspace capacity is per square kilometer), and eliminate dimension differences through standardization processing (e.g. Z-score). Finally, the target feature set is used for subsequent model training, for example, the target flight feature set is input into the path adjustment model, and the target airspace feature set is input into the resource scheduling model, to ensure the parameter consistency of the model training data and the real-time emergency event data.

[0078] Step S40: According to the target flight feature set and the target airspace feature set, a plurality of initial flight emergency prediction models are constructed, and each initial flight emergency prediction model is iteratively trained in a cross-validation manner.

[0079] The initial model construction is based on diversified algorithms to adapt to different prediction needs: the first type of model uses supervised learning algorithms (such as random forest, gradient boosting decision tree), takes the target flight feature set and the target airspace feature set as input, and takes the path adjustment strategy as output, for example, when inputting "speed deviation + 20 knots, airspace capacity 6", the output is "heading correction + 10 degrees"; the second type of model applies time series analysis (such as LSTM, ARIMA), takes the time series flight features (such as the height deviation sequence of the past 10 minutes) and weather parameters as input, and predicts the resource demand curve. Cross-validation uses K-fold strategy (such as 5-fold), divides the historical data set into training subset and validation subset, for example, the first 4 folds are used to train the model, and the 5th fold is used to verify the prediction accuracy. In each iteration, the model adjusts the parameters through back propagation or gradient descent method, for example, the random forest model optimizes the node splitting rule through the Gini coefficient, and the LSTM model updates the weight matrix through the error back propagation of the time step. Record the mean square error (MSE) or classification accuracy of each validation during the training process, and dynamically adjust the hyperparameters (such as learning rate, tree depth) to improve the generalization ability.

[0080] As an implementation, step S40, according to the target flight feature set and the target airspace feature set, a plurality of initial flight emergency prediction models are constructed, which can specifically include: step S41: divide the historical emergency event data set into training subset and validation subset; the training subset is used to generate the initial flight emergency prediction model, and the validation subset is used to evaluate the model performance.

[0081] The historical emergency event dataset is divided into a training subset and a validation subset by a preset ratio (e.g., 8:2 or 7:3), ensuring that the two types of datasets are balanced in terms of event type, time distribution, and coverage of spatial features. The training subset is used for fitting and optimizing model parameters, such as selecting 80% of historical case data (e.g., 80 out of 100 cases) covering various emergency event types such as "aircraft loss of control", "meteorological mutation", and "airspace congestion". The validation subset is reserved for the remaining 20% of case data to evaluate the model's generalization ability on unknown data. Time leakage should be avoided during the division process, such as the case timestamp of the validation subset should be later than the training subset to prevent the model from overfitting due to time correlation. Specifically, if the historical dataset is arranged in chronological order (from the earliest to the latest event), the first 80% of the time period data is used as the training subset, and the last 20% is used as the validation subset. After data division, the distribution consistency of the target flight feature set and the target airspace feature set in the two subsets should be verified, such as the mean of the flight speed deviation in the training subset is +15 knots, and the validation subset is +16 knots, indicating that the data distribution has no significant deviation, meeting the model training requirements.

[0082] Step S42: According to the target flight feature set and the target airspace feature set, extract the training data corresponding to the core flight feature and the core airspace feature from the training subset.

[0083] The training data extraction is based on the target flight feature set (such as flight speed deviation, height deviation, and heading angle deviation rate) and the target airspace feature set (such as airspace capacity and obstacle distribution dispersion), to filter the corresponding parameter columns from the training subset, and perform data cleaning and formatting. For example, the "flight speed deviation" in the target flight feature set corresponds to the difference between the real-time speed and the planned speed of each case record in the training subset, which is extracted through a numerical field; the "height deviation" is calculated from the actual height and safe height of the aircraft height parameter in the case. The "airspace capacity" in the airspace feature set is obtained by counting the number of aircrafts in the airspace grid, and the "obstacle distribution dispersion" is calculated based on the ratio of obstacle distance to safety distance. The extracted training data needs to be processed for missing values (such as linear interpolation filling) and standardization (such as Z-score standardization) to ensure that the input data meets the model training requirements. For example, the flight speed deviation of a certain case is +20 knots, which is converted to 0.75 after standardization (assuming the mean is +15 knots and the standard deviation is 6.7 knots), and the airspace capacity is 6 aircrafts per square kilometer, which is standardized to 1.2 (mean is 5, standard deviation is 0.83).

[0084] Step S43: Use a supervised learning algorithm to fit the training data for multiple rounds to generate a first flight emergency prediction model; the first flight emergency prediction model is used to predict flight path adjustment strategies.

[0085] The supervised learning algorithm trains the model to generate path planning instructions by establishing a mapping relationship between input features (target flight feature set and target airspace feature set) and output labels (path adjustment strategies in historical cases). For example, the random forest algorithm takes flight speed deviation, height deviation, and airspace capacity as input, and outputs heading correction angle, climb height, and speed adjustment amount. During training, the model optimizes node splitting rules and feature weights through multiple iterations. For example, in the first iteration, the model may prefer to split nodes based on flight speed deviation to generate preliminary path strategies. In subsequent iterations, the airspace capacity feature is introduced to optimize the complexity of the avoidance path. After each fitting, the model evaluates the prediction accuracy (such as the mean absolute error of the heading angle prediction value and the actual value) through a validation subset, and adjusts the hyperparameters (such as tree depth and learning rate) based on the error gradient. The final generated model can output path adjustment instructions highly consistent with historical handling strategies based on real-time flight features and airspace features, such as inputting a flight speed deviation of +20 knots and an airspace capacity of 6 aircraft, and the model outputting "heading correction +12 degrees, climb to 800 meters."

[0086] As an implementation, in step S43, a supervised learning algorithm is used to fit the training data for multiple rounds to generate a first type of flight emergency prediction model, which can specifically include: step S431: determining a set of labeled results related to flight path adjustment in the training data; the set of labeled results includes the path change instructions actually executed in historical cases and the corresponding execution effect scores.

[0087] The set of labeled results is extracted from the path adjustment records of historical cases, covering specific instructions such as heading correction angle, target height, speed adjustment amount, and execution effect score (such as conflict avoidance success rate and time deviation after instruction execution). For example, the labeled result of a certain case is "heading correction +15 degrees, climb to 800 meters, speed down to 250 knots", and the execution effect score is 0.92 (based on a comprehensive score of path deviation and time efficiency). During the labeling process, the instruction format needs to be unified, such as heading angle in degrees, height in meters, and speed in knots, to ensure the consistency of the model output. The execution effect score is calculated through quantitative indicators, such as time deviation, which is the difference between actual execution time and planned time (e.g., -30 seconds), and is converted into a standardized score in the range of 0-1 (e.g., 1- |deviation| / maximum allowed deviation).

[0088] Step S432: Take the core flight features and core airspace features as input variables, and take the path change instructions and execution effect scores as output variables to build a regression analysis model.

[0089] The regression analysis model takes the target flight characteristics (flight speed deviation, height deviation, heading angle deviation rate) and airspace characteristics (airspace capacity, obstacle distribution dispersion) as independent variables, and the path change instruction (such as the heading correction angle) and the execution effect score as dependent variables, to establish a multi-output regression model. For example, the gradient boosting regression tree (GBRT) model generates a comprehensive path strategy by jointly optimizing the prediction errors of the heading angle, height, and speed. During model training, the loss function is designed as a weighted mean square error (MSE), with the heading angle error weight being 0.6, the height error being 0.3, and the speed error being 0.1, to reflect the priority differences of different instructions. After standardizing the input data (such as standardizing the flight speed deviation to 0.75), the model updates the decision tree structure and leaf node weight through multiple iterations, gradually approaching the optimal fitting state.

[0090] Step S433: In each fitting process, the weight coefficient in the model is adjusted according to the deviation value between the predicted path instruction of the regression analysis model and the actual path instruction.

[0091] After each fitting, the model calculates the deviation value between the predicted instruction and the actual instruction (such as a heading angle deviation of +3 degrees and a height deviation of -20 meters), and updates the weight coefficient through the back propagation algorithm. For example, the gradient descent method adjusts the gain threshold of feature splitting in the regression tree according to the deviation gradient, and optimizes the high-weight error items (such as heading angle deviation) first. At the same time, a regularization term (such as L2 regularization) is introduced into the loss function to prevent the model from overfitting due to excessive complexity. The adjusted weight coefficient needs to be re-evaluated for the performance of the validation subset. If the validation error continues to decrease, continue iteration; if there is an upward trend, trigger the early stopping mechanism to prevent performance degradation.

[0092] Step S434: When the deviation value change of K consecutive fitting is less than the preset change threshold, it is determined that the regression analysis model has reached the convergence state, and the model parameters at this time are saved as the first type of flight emergency prediction model.

[0093] The convergence judgment condition is set to the deviation value change (such as the heading angle error fluctuation range) of K=5 consecutive training being less than the threshold Δ=0.01. For example, the heading angle error sequence of a certain model in iteration is 0.15→0.12→0.11→0.10→0.09, with a maximum change of 0.03 (0.15-0.12), which does not meet the threshold requirement; after further training, the error sequence is 0.08→0.07→0.06→0.055→0.053, with a change of 0.027 (0.08-0.053), which is still higher than the threshold; finally, when the sequence is 0.05→0.049→0.048→0.047→0.046, the change is 0.004, which is lower than the threshold, and convergence is determined. The model parameters (such as decision tree structure, leaf node weight, feature importance) are serialized and stored for real-time prediction.

[0094] Step S44: Trend fitting on training data is performed using a time series analysis algorithm to generate a second type of flight emergency prediction model; the second type of flight emergency prediction model is used to predict an emergency resource allocation strategy.

[0095] The time series analysis algorithm captures the timing rules of historical resource scheduling data to predict future resource demand and supply cycles. For example, the autoregressive integrated moving average (ARIMA) model takes the timing data of resource scheduling duration, resource consumption rate, and supply interval as input to construct a resource demand prediction curve. During training, the model first performs difference processing (such as first-order difference to eliminate trend items) on non-stationary timing data, and then determines the optimal lag order (such as AR term order p=2 and MA term order q=1) through autocorrelation function (ACF) and partial autocorrelation function (PACF). After model fitting, the matching degree (such as root mean square error RMSE) of the prediction result and the actual resource usage record is evaluated through the validation subset, and the parameters are adjusted to minimize the error. For example, in a certain case, after the ARIMA(2,1,1) model is fitted to the timing data of the resource consumption rate, a 30-minute consumption rate prediction curve is generated, and the RMSE of the actual record curve is 0.08, indicating that the model has high prediction accuracy. The final model can output a dynamic resource scheduling strategy, such as "deploy 2 fire unmanned aerial vehicles to coordinates (X, Y) every 20 minutes".

[0096] As an implementation, in step S44, a time series analysis algorithm is used to perform trend fitting on training data to generate a second type of flight emergency prediction model, which can specifically include: step S441: extracting time series parameters related to resource allocation in the training data; the time series parameters include resource scheduling duration, resource consumption rate, and resource supply interval.

[0097] The time series parameters are extracted from historical resource allocation records in time windows, such as resource scheduling duration sequence records the time from the issuance of the command to the arrival of the resource (such as 30 minutes, 28 minutes, and 32 minutes); resource consumption rate sequence statistics resource usage per unit time (such as fire foam consumption rate of 100 liters / minute); resource supply interval sequence records the time difference between two consecutive supply operations (such as 120 minutes, 115 minutes, and 125 minutes). During the extraction process, the timestamps need to be aligned to ensure the continuity of the sequence, such as sampling the resource consumption rate at 5-minute intervals to generate equally spaced timing data.

[0098] Step S442: Stationarity test is performed on the time series parameters, and difference processing is performed on non-stationary sequences until a stationary time series is obtained.

[0099] The stationarity test uses the Augmented Dickey-Fuller (ADF) test. If the ADF statistic is less than the critical value (e.g., p < 0.05), the sequence is considered stationary. For example, the ADF statistic for the resource scheduling duration sequence is -2.5 (p = 0.03), indicating that the sequence is stationary; the ADF statistic for the resource consumption rate sequence is -1.8 (p = 0.15), requiring first-order differencing. After differencing, the sequence is retested. If it is still not stationary, second-order differencing is performed until the ADF statistic meets the stationarity condition. For example, the original consumption rate sequence has an ADF statistic of -3.1 (p = 0.02) after first-order differencing, and is therefore considered stationary.

[0100] Step S443: Based on the stationary time series, construct an autoregressive integral moving average model and determine the optimal order and lag parameters of the model.

[0101] The optimal order of the Autoregressive Integral Moving Average (ARIMA) model is determined through analysis of the autocorrelation function (ACF) and partial autocorrelation function (PACF). For example, if the ACF of the stationary resource consumption rate series is truncated after lag 2 and the PACF is truncated after lag 1, then the model order is set to ARIMA(1,1,2). Parameter optimization employs a grid search method, traversing all combinations of p(0-3), d(1-2), and q(0-3) to select the parameter combination with the smallest AIC (Akaike Information Criterion) or BIC (Bayesian Information Criterion). For example, ARIMA(1,1,1) has an AIC value of 250, while ARIMA(2,1,1) has 245; the latter is ultimately chosen.

[0102] Step S444: Using the optimal order and lag parameter, backfit the historical resource allocation data to generate a resource demand prediction curve.

[0103] After model fitting, the prediction accuracy is validated using historical data. For example, the ARIMA(2,1,1) model takes a 120-minute historical resource consumption rate sequence as input and generates a predicted curve for the next 60 minutes. The root mean square error (RMSE) between the predicted curve and the actual recorded curve is 0.05, indicating that the prediction results are reliable. The model also outputs confidence intervals (e.g., 95% confidence level) to provide a risk boundary reference for resource scheduling.

[0104] Step S445: Based on the matching degree between the resource demand prediction curve and the actual resource usage curve, optimize the parameters of the autoregressive integral moving average model, and mark the optimized model as the second type of flight emergency prediction model.

[0105] The optimization process improves the matching accuracy by adjusting the order of the ARIMA model or introducing a seasonal parameter (SARIMA). For example, if the prediction curve shows a systematic deviation during periodic periods (such as daily peak hours), a seasonal order is added (e.g., SARIMA(2,1,1)(1,1,1,24)). The optimized model is evaluated using a validation subset. If the RMSE decreases to 0.03, it is marked as a Class II flight emergency prediction model, and its parameters are stored for real-time prediction. For example, if the model outputs "3 firefighting drones need to be dispatched within the next hour, with a resupply interval of 90 minutes," the deviation from the actual demand is ±1 drone, meeting the emergency response accuracy requirements.

[0106] In an optional derivative implementation, the method further includes a step of jointly verifying the first type of flight emergency prediction model and the second type of flight emergency prediction model, which may specifically include: Step S45: obtaining the heading change parameters, altitude correction parameters and speed control parameters in the flight path adjustment strategy output by the first type of flight emergency prediction model, and extracting the resource scheduling path parameters, resource delivery location parameters and resupply time interval parameters in the emergency resource allocation strategy output by the second type of flight emergency prediction model.

[0107] The heading change parameter indicates the correction amount of the aircraft's heading angle in the flight path adjustment strategy, such as a heading offset of +15 degrees to avoid obstacle areas; the altitude correction parameter is used to adjust the aircraft's vertical flight level, such as climbing to 800 meters to avoid low-altitude turbulence; the speed control parameter defines the adjustment curve of the aircraft's speed, such as reducing from 300 knots to 250 knots within 120 seconds to match the resource scheduling rhythm. The resource scheduling path parameter describes the movement trajectory coordinate sequence of emergency resources (such as firefighting drones), such as a straight path from the base coordinates (X1, Y1) to the target area coordinates (X2, Y2); the resource deployment location parameter specifies the specific geographical coordinates and altitude level of resource deployment, such as deploying a navigation beacon at coordinates (X3, Y3) and an altitude of 500 meters; the resupply time interval parameter specifies the time difference between two consecutive resource resupply operations, such as performing a refueling operation every 180 minutes. The above parameters are extracted from the output results of the two types of models through a standardized data interface and converted into a unified spatial coordinate system (such as WGS-84) and timestamp format to ensure the spatiotemporal consistency of subsequent collision detection.

[0108] Step S46: Based on the spatial coverage of the heading change parameters and the resource scheduling path parameters, determine the overlapping area of ​​their trajectories in the airspace coordinate system, and detect whether there are conflict segments in the overlapping area where the heading offset direction is opposite to the resource scheduling direction.

[0109] The spatial coverage area is determined by calculating the intersection region of the aircraft's adjusted trajectory (e.g., the coordinate sequence of the aircraft within the next 10 minutes corresponding to a heading of +15 degrees) and the resource scheduling path (e.g., the movement trajectory of a UAV from X1,Y1 to X2,Y2) within the airspace grid. For example, if the aircraft trajectory crosses grid G7 within the time window T0+300 seconds to T0+420 seconds, while the UAV passes through the same grid G7 within the time window T0+360 seconds to T0+480 seconds, then the two trajectories are determined to have spatiotemporal overlap in the G7 region. Further analysis of the movement direction within the overlapping section: if the aircraft's heading offset direction is northeast (+15 degrees), while the UAV's movement direction is southwest (from X1,Y1 to X2,Y2), then the two directions are opposite, posing a risk of opposing conflict. The collision detection algorithm marks the collision zone by calculating the angle between the vectors (e.g., the angle between the aircraft's heading vector and the UAV's movement vector exceeds 150 degrees), and records the collision time window (e.g., from T0+360 seconds to T0+420 seconds) and the spatial coordinate range (e.g., coordinates X=123.45, Y=45.67 to X=123.50, Y=45.70 within the G7 grid).

[0110] Step S47: Based on the aircraft climb or descent commands corresponding to the altitude correction parameters, analyze the vertical altitude constraints in the resource deployment location parameters and identify the compatibility status between the aircraft climb or descent commands and the resource deployment altitude threshold.

[0111] The aircraft climb commands (e.g., "climb to 800 meters") or descent commands (e.g., "descent to 500 meters") corresponding to the altitude correction parameters need to be verified for compatibility with the vertical altitude restrictions in the resource deployment location parameters. For example, if the resource deployment location parameters stipulate that navigation beacons must be deployed between 600 and 700 meters, and the target altitude of the aircraft climb command is 800 meters, the aircraft will cross this altitude layer without conflicting with the resource deployment altitude threshold. However, if the target altitude of the aircraft descent command is 550 meters, and the resource deployment altitude threshold is 500 to 600 meters, then the aircraft altitude of 550 meters is within the resource deployment layer, and further detection of spatial overlap is required. Compatibility status is divided into three categories: fully compatible (aircraft altitude and resource altitude do not intersect), partially compatible (there is a brief overlap but the time window is misaligned), and incompatible (altitude layers overlap and time windows conflict). For example, if an aircraft maintains an altitude of 550 meters between T0+300 seconds and T0+360 seconds, while the resource deployment time window is between T0+330 seconds and T0+390 seconds, it is considered to be in an incompatible state.

[0112] Step S48: Compare the expected flight speed change curve in the speed control parameters with the resource arrival time window in the resupply interval parameters to verify whether the peak period of the speed change curve exceeds the fault tolerance boundary of the resource arrival time window.

[0113] The expected flight speed change curve in the speed control parameters describes the trend of the aircraft's speed over time. For example, it linearly decreases from 300 knots to 250 knots between T0+0 and T0+120 seconds, and then remains stable. The resource arrival time window in the resupply interval parameter specifies that resources must arrive within a specific time range. For example, the first UAV must arrive at coordinates X2, Y2 before T0+600 seconds. During verification, the peak period of the speed change curve is extracted (e.g., the period with the largest speed gradient during the deceleration phase from T0+0 to T0+120 seconds is from T0+60 to T0+90 seconds), and it is checked whether this period conflicts with the resource arrival window. For example, if the estimated time for resource scheduling path is 540 seconds, the estimated arrival time of the UAV is T0+540 seconds, which does not overlap with the peak period of speed control (T0+60 seconds to T0+90 seconds), and is therefore considered compatible. However, if the resource scheduling time is extended to 660 seconds (arrival time T0+660 seconds) due to path adjustment, and the speed control curve requires the aircraft to accelerate to 280 knots within the period from T0+600 seconds to T0+660 seconds, then the peak period (acceleration phase) overlaps with the resource arrival window, and it is necessary to verify whether the acceleration command causes the aircraft to occupy conflict airspace during the resource arrival period.

[0114] Step S49: When a conflict zone, incompatible vertical height limit conditions, or peak period exceeding the fault tolerance boundary is detected, a parameter adjustment instruction set is generated. The parameter adjustment instruction set includes heading offset angle compensation value, resource deployment altitude offset, or smoothing correction coefficient of speed change curve.

[0115] The parameter adjustment instruction set is generated based on the conflict type: For heading conflict sections, a heading offset angle compensation value is calculated to eliminate the risk of opposing movement; for example, the original heading +15 degrees is adjusted to +18 degrees, causing the aircraft trajectory to deviate from the conflict grid G7. For altitude incompatibility states, the resource deployment altitude offset is adjusted; for example, the navigation beacon deployment altitude is changed from 600-700 meters to 650-750 meters to avoid overlap with the aircraft's 550-meter altitude layer. For conflicts during peak speed periods, a smoothing correction coefficient is introduced to adjust the gradient of the speed change curve; for example, the gradient during the deceleration phase is reduced from 5 knots / second to 3 knots / second, and the deceleration time window is extended to T0+180 seconds to stagger the resource arrival time. The adjustment instruction set is generated through optimization algorithms (such as linear programming or genetic algorithms) to ensure that the adjusted parameters minimize the impact on the original strategy while meeting emergency objectives.

[0116] Step S410: Synchronously feed back the parameter adjustment instruction set to the first type of flight emergency prediction model and the second type of flight emergency prediction model, triggering both to regenerate the flight path adjustment strategy and emergency resource allocation strategy based on the adjusted parameters, until all conflict sections, incompatible states and boundary overflow conditions are eliminated.

[0117] The feedback mechanism injects the heading offset angle compensation value into the input parameters of the first type of model, triggering it to recalculate the path trajectory; the resource deployment altitude offset is input into the second type of model to adjust the resource deployment logic; the speed smoothing correction coefficient is simultaneously updated to both types of models to coordinate the spatiotemporal rhythm of the aircraft and resources. For example, the first type of model regenerates the detour path based on a +18-degree heading to avoid grid G7; the second type of model modifies the navigation beacon delivery altitude to 700 meters and extends the resupply interval to 200 minutes. During the iteration process, the conflict detection module continuously verifies the compatibility of the new strategy. If residual conflicts still exist (such as the adjusted aircraft trajectory overlapping with another resource path), steps S45 to S49 are repeated until all conflicts are eliminated. The iteration terminates when there are no new conflicts in two consecutive verification results and all historical conflict markers are cleared.

[0118] Step S411: Record the logical association rules between the flight path adjustment strategy and the emergency resource allocation strategy after the final elimination of the conflict, and embed the logical association rules into the output parameter verification mechanism of the flight emergency prediction model.

[0119] Logical association rules encapsulate the constraints of coordination strategies through structured descriptions, such as "when the heading deviation angle is ≥ +15 degrees, the resource deployment altitude must be ≥ 600 meters" or "the peak velocity period must be at least 60 seconds earlier than the resource arrival window." These rules are encoded as executable verification conditions and embedded in the pre-verification module of the model output interface. For example, in subsequent predictions, when the first type of model generates a heading +15 degree instruction, the verification module automatically retrieves association rules, forcing the second type of model to apply the ≥ 600-meter restriction in the resource deployment altitude parameter. The rule base is dynamically expanded through case learning, such as adding a new conflict type, "joint avoidance rules when sudden weather changes lead to insufficient visibility," continuously enhancing the model's collaborative prediction capabilities and conflict prevention efficiency.

[0120] Step S50: During the iterative training process, the model parameters are dynamically adjusted according to the difference between the simulation results and the actual processing results of each initial flight emergency prediction model for historical cases. Training is stopped when the difference is lower than the preset convergence threshold, and the initial flight emergency prediction model that has been trained is marked as a flight emergency prediction model.

[0121] The discrepancy is calculated by comparing the deviation between the model's predictions and the actual handling records. For example, if the predicted heading angle for a path adjustment strategy is +12 degrees and the actual correction is +15 degrees, the discrepancy is 3 degrees; if the predicted resource demand is 3 aircraft and 4 aircraft are actually dispatched, the discrepancy is 1 aircraft. The discrepancy index is normalized to a 0-1 range (e.g., 3 degrees / maximum allowable deviation 15 degrees = 0.2) and weighted to form the total model discrepancy (e.g., path discrepancy weight 0.6, resource discrepancy weight 0.4). Dynamic parameter adjustment uses adaptive optimization algorithms, such as the Adam algorithm adjusting the learning rate based on the discrepancy gradient, and decision tree models reducing overfitting through pruning. The model is considered converged when the total discrepancy fluctuation range of N consecutive iterations (e.g., N=5) is less than a preset convergence threshold (e.g., 0.05). For example, if a model's difference sequence is 0.15→0.12→0.11→0.10→0.09, with a fluctuation range of 0.06 (0.15-0.09), which is higher than the threshold of 0.05, it needs to continue iterating. If the sequence is 0.08→0.07→0.06→0.055→0.053, with a fluctuation range of 0.027 (0.08-0.053), which is lower than the threshold, then training should stop. Finally, the validated models are labeled as flight emergency prediction models and injected into the model library for real-time use. For example, model M-1024 is labeled as "Flight Path Adjustment Model - Strong Wind Scenario," and model M-1025 is labeled as "Resource Scheduling Model - Airspace Congestion Scenario."

[0122] Step S400: Based on the flight emergency prediction model, output flight path adjustment strategy and emergency resource allocation strategy, and generate emergency command instructions.

[0123] The flight path adjustment strategy is a dynamic path planning scheme generated by the model based on real-time anomaly data and historical handling patterns. For example, it includes heading offset angle, target flight altitude layer, speed control curve, and waypoint sequence. The emergency resource allocation strategy covers the dispatch path of rescue equipment, the deployment coordinates of temporary navigation beacons, the activation sequence of emergency communication relay nodes, and the refueling cycle. When generating emergency command instructions, the spatial trajectory of the heading parameters and resource dispatch path in the path adjustment strategy is first analyzed to detect potential conflict points between them in the airspace coordinate system, such as the risk of spatiotemporal collision caused by the overlap between the heading correction direction and the resource transportation route. After eliminating the conflict by adjusting the heading offset angle or the curvature of the resource path, it is further verified whether the target flight altitude covers the vertical safety zone of the resource delivery area (e.g., the drone delivery altitude must avoid high-voltage lines). If there is an altitude mismatch, the flight altitude parameters are dynamically corrected. Simultaneously, the synchronization rate of the speed adjustment curve and the resource refueling time window is calibrated to ensure that the acceleration or deceleration phase matches the resource arrival time. Finally, the flight control command set is generated by integrating the correction parameters of heading, altitude, and speed, and the resource scheduling command set is encapsulated into a standardized signal queue. After verification by airspace control rules, executable emergency command commands are output.

[0124] As one implementation method, step S400 involves generating emergency command instructions based on the flight path adjustment strategy and emergency resource allocation strategy output by the flight emergency prediction model. Specifically, this may include: parsing the heading, altitude, and speed parameters in the flight path adjustment strategy to generate a path parameter set; parsing the resource path, location, and resupply parameters in the emergency resource allocation strategy to generate a resource parameter set; detecting the intersection points of the heading in the path parameter set and the movement trajectory in the resource parameter set within the airspace coordinate system, determining the conflict time window, and adjusting the heading or trajectory curvature to generate an updated path and resource parameter set; and verifying the path parameter set. If the target altitude of the combined flight path does not cover the vertical safety zone of the resource delivery location, the altitude parameters are dynamically adjusted to generate the final path parameters if not. The synchronization rate between the speed adjustment curve and the resource replenishment cycle is matched, and the time axis of the speed change phase is calibrated to generate optimized speed parameters. The corrected heading, altitude, and speed parameters are integrated to generate a flight control command set, and the adjusted resource path, position, and cycle parameters are integrated to generate a resource scheduling command set. The two command sets are converted into standardized signals and encapsulated into a command queue. The execution sequence is verified to comply with airspace control rules. If there are conflicts, the sequence is readjusted until the verification is passed, and then the final emergency command is output.

[0125] Specifically, the system analyzes the heading change parameters, altitude correction parameters, and speed control parameters in the flight path adjustment strategy to generate a path parameter set containing the heading offset angle, target flight altitude, and speed adjustment curves. Simultaneously, it analyzes the resource scheduling path parameters, resource delivery location parameters, and resupply time interval parameters in the emergency resource allocation strategy to generate a resource parameter set containing resource movement trajectories, delivery coordinate sequences, and resupply cycle sequences. Based on the heading offset angle in the path parameter set and the resource movement trajectory in the resource parameter set, it detects the intersection points of their trajectories in the airspace coordinate system and determines the conflict time window corresponding to the trajectory intersection point. During the conflict time window, adjust the heading offset angle or the path curvature of the resource movement trajectory to generate an updated path parameter set and an updated resource parameter set to eliminate spatiotemporal conflicts. Verify the vertical alignment between the target flight altitude in the updated path parameter set and the delivery coordinate sequence in the resource parameter set to identify whether the target flight altitude covers the vertical safety zone of the resource delivery location. If not, dynamically adjust the target flight altitude based on the upper and lower limits of the vertical safety zone to generate a highly compatible final path parameter set. Finally, compare the speed adjustment curve in the final path parameter set with the resupply information in the updated resource parameter set. The system performs a periodic sequence matching of the synchronization rate between speed change periods and resource arrival periods, and calibrates the acceleration or deceleration phases of the speed adjustment curve based on the synchronization rate to generate optimized speed control parameters with time synchronization. It integrates the heading offset angle, highly compatible target flight altitude, and optimized speed control parameters from the final path parameter set to generate a flight control command set containing heading, altitude, and speed commands. Simultaneously, it integrates the resource movement trajectory, delivery coordinate sequence, and resupply cycle sequence from the updated resource parameter set to generate a resource scheduling command set containing path navigation, coordinate positioning, and periodic trigger commands. The flight control command set and resource scheduling command set are converted according to preset command encoding rules to generate standardized flight control signals and standardized resource scheduling signals. These standardized flight control signals and standardized resource scheduling signals are logically encapsulated, and emergency event identifiers and time stamps are injected to form an executable emergency command command queue. The system verifies whether the execution timing of the flight control signals and resource scheduling signals in the emergency command command queue meets airspace control rules and resource deployment constraints. If conflicts exist, the system returns to the step of adjusting the conflict time window and regenerates the commands until all signals pass verification before outputting the final emergency command command.

[0126] Step S500: Execute emergency command instructions and update the parameters of the flight emergency prediction model based on real-time feedback data after execution.

[0127] For example, the execution of emergency command instructions is completed collaboratively by the air traffic control system, the aircraft's autopilot module, and the ground resource scheduling platform. Real-time feedback data includes indicators such as the aircraft's response accuracy to instructions (e.g., the deviation between the actual heading angle and the instruction value), the delay between the resource deployment time and the planned time, and the dynamic flight density after airspace changes. A deviation index is generated by comparing the expected response data with the actual feedback data. If the deviation exceeds the tolerance threshold (e.g., a heading angle deviation greater than 2 degrees or a resource delay exceeding 5 minutes), a model parameter update mechanism is triggered: an incremental learning algorithm is used to inject new feedback data into the model. Without changing the original model structure, the weight parameters are adjusted using gradient descent to reduce prediction errors in similar scenarios. For example, if the model underestimates the navigation beacon deployment time in a "low visibility emergency landing" event, incremental learning will correct the weights of the resource scheduling path planning module, optimizing the resource allocation strategy for subsequent similar events. The updated model needs to be jointly validated using historical data and new data to ensure its overall performance stability in complex scenarios.

[0128] As one implementation method, step S500 involves executing emergency command instructions and updating the parameters of the flight emergency prediction model based on real-time feedback data after execution. Specifically, this may include step S510: monitoring aircraft response data, resource scheduling status data, and airspace change data during the execution of emergency command instructions.

[0129] For example, during the execution of emergency command instructions, three types of key data are collected in real time through a multi-source sensor network and data link: Aircraft response data includes the actual execution parameters of the aircraft's heading correction, altitude adjustment, and speed control commands, such as the actual heading angle deviation of +14.5 degrees (expected +15 degrees), the actual altitude climb to 798 meters (expected 800 meters), and the actual speed drop to 248 knots (expected 250 knots); resource scheduling status data covers the deployment progress and operational status of emergency resources (such as firefighting drones and navigation beacons), for example, the drone's departure from its base... The data includes the deviation distance (e.g., 50 meters) between the actual movement trajectory from ground coordinates (X1, Y1) to the target area (X2, Y2) and the planned path; the actual activation time of the navigation beacon at coordinates (X3, Y3) (T0+620 seconds, expected T0+600 seconds); and the dynamic changes in the density of aircraft distribution within the airspace after the airspace change command is executed (e.g., the number of aircraft in airspace grid G7 decreases from 6 to 4), adjustments to the boundaries of temporary control areas (e.g., the radius of the control area increases from 2 kilometers to 3 kilometers), and updates to obstacle status (e.g., the coordinates of newly added high-voltage power line towers). The monitoring data is timestamped and aligned with the airspace coordinates to ensure spatiotemporal consistency, providing a basis for subsequent deviation analysis.

[0130] Step S520: Compare the aircraft response data with the expected response data in the flight path adjustment strategy to generate the first deviation index.

[0131] The expected response data consists of path adjustment commands output by the flight emergency prediction model, such as a heading correction of +15 degrees, climbing to 800 meters, and reducing speed to 250 knots. The deviation between the actual response data and the expected values ​​is calculated using quantitative indicators: heading angle deviation is |14.5-15|=0.5 degrees, altitude deviation is |798-800|=2 meters, and speed deviation is |248-250|=2 knots. The first deviation indicator uses a weighted average method to synthesize the deviations of each parameter. For example, with a heading angle weight of 0.6, an altitude weight of 0.3, and a speed weight of 0.1, the indicator value is (0.5×0.6 + 2×0.3 + 2×0.1) = 0.3 + 0.6 + 0.2 = 1.1. The indicator thresholds are set according to flight safety standards. For example, a heading angle tolerance of ±1 degree, an altitude tolerance of ±5 meters, and a speed tolerance of ±3 knots corresponds to a first deviation indicator threshold of 1.8 (weighted calculated value). Exceeding the threshold indicates a significant deviation in command execution.

[0132] Step S530: Compare the resource scheduling status data with the expected resource data in the emergency resource allocation strategy to generate a second deviation index.

[0133] Expected resource data includes resource deployment coordinates, arrival time, and quantity. For example, a firefighting drone should arrive at (X2, Y2) at T0+600 seconds, and a navigation beacon should be activated at (X3, Y3) at T0+600 seconds. In actual scheduling data, the drone arrival time is T0+620 seconds (delayed by 20 seconds), and the beacon activation time is T0+615 seconds (earlier by 15 seconds). The second deviation index is calculated by weighting the absolute value of the time deviation with the quantity deviation. For example, the time deviation weight is 0.7, and the quantity deviation weight is 0.3. Assuming there is no deviation in the number of drones, the index value is (20 × 0.7 + 0 × 0.3) = 14. The preset threshold is set according to the urgency of resource scheduling. For example, the time deviation threshold is ±30 seconds, and the quantity deviation threshold is ±1 drone. The corresponding second deviation index threshold is 21 (time deviation 30 × 0.7 + quantity deviation 1 × 0.3 = 21.3). The current index of 14 is not exceeded, but the trend needs to be recorded.

[0134] Step S540: When the first deviation index or the second deviation index exceeds the preset fault tolerance threshold, the model parameter update mechanism is triggered, and the adjustment amount of the model parameters is calculated based on the current deviation index.

[0135] For example, the preset fault tolerance thresholds can be dynamically adjusted according to event priority. For instance, the first deviation threshold for a Level 1 event is set to 1.5, and the second deviation threshold to 18; for Level 2 events, the thresholds are relaxed to 2.0 and 24. If the current first deviation index is 1.8 (exceeding the Level 1 threshold of 1.5) and the second deviation is 14 (not exceeding the threshold), then a parameter update for the flight path adjustment model is triggered. The adjustment amount is calculated using the error backpropagation algorithm. For example, the model weight gradient corresponding to a heading angle deviation of 0.5 degrees is Δθ = 0.5 × learning rate 0.01 = 0.005, and the gradient corresponding to an altitude deviation of 2 meters is Δh = 2 × 0.01 = 0.02. The adjustment injection mechanism determines the parameter correction sign (e.g., increasing the heading weight coefficient) based on the deviation direction (e.g., insufficient heading correction).

[0136] Step S550: Incremental learning algorithm is used to inject the adjustment into the flight emergency prediction model, and the prediction accuracy of the updated model is re-verified.

[0137] Incremental learning algorithms iteratively optimize parameters by adding small batches of new data without altering the model structure. For example, the weight of the heading correction module in the flight path adjustment model, originally 0.85, is updated to 0.855 after an adjustment of +0.005; the weight of the altitude correction module is adjusted from 0.75 to 0.77. The updated model is then jointly validated using historical datasets and newly added feedback data. For instance, using cross-validation, the average heading angle error of the updated model in 100 historical cases decreases from 1.2 degrees to 1.0 degrees, and the error in 20 newly added feedback cases is 1.1 degrees, meeting the preset performance threshold (e.g., average error ≤ 1.5 degrees).

[0138] As one implementation method, in step S550, an incremental learning algorithm is used to inject the adjustment amount into the flight emergency prediction model, which may specifically include: step S551: extracting a new dataset from the real-time feedback data that is consistent with the type of the model input parameters.

[0139] For example, the new dataset selects parameters from real-time feedback data that align with the model's input features. For instance, if the input features of a flight emergency prediction model include flight speed deviation, altitude deviation, and airspace capacity, the new data must contain the same fields: in a given feedback case, the speed deviation is +20 knots, the altitude deviation is -5 meters, and the airspace capacity is 5 aircraft / square kilometer. Data extraction must verify field integrity, such as removing invalid records with missing airspace capacity values, to ensure that the new dataset's feature dimensions are completely consistent with the training data.

[0140] Step S552: Perform feature alignment on the new dataset to ensure that the feature dimensions of the new dataset are exactly the same as those used in the training phase.

[0141] For example, feature alignment includes data normalization and dimensionality mapping. For instance, during the training phase, flight speed deviations are normalized using max-min normalization (the range [-50, +50] sections are mapped to [0,1]), and the +20 sections in the new data need to be converted to (20+50) / 100=0.7; airspace capacity is measured in aircraft / square kilometer during the training phase, and the 5 aircraft in the new data need to be directly retained. If the new data contains features not seen during the training phase (such as newly added obstacle type encoding), dimensionality consistency is maintained by filling in default values ​​or removing the feature.

[0142] Step S553: ​​Without changing the original model structure, input the new dataset into the flight emergency prediction model and calculate the loss function value between the output of the flight emergency prediction model and the actual feedback result.

[0143] For example, the model structure is fixed, such as the tree structure of a random forest or the number of layers in a neural network. When a new dataset is input, the model outputs a prediction instruction (e.g., heading correction +14 degrees), and calculates the loss function value (e.g., mean squared error MSE = (14 - 14.5) by comparing this with the actual heading correction value (+14.5 degrees). 2 =0.25). The loss function value reflects how well the current model fits the new data, and it is gradually reduced through iteration.

[0144] Step S554: Based on the loss function value, the weight parameters in the flight emergency prediction model are adjusted using the gradient descent method, so that the prediction error of the flight emergency prediction model for new data is gradually reduced.

[0145] Gradient descent updates model weights based on the gradient of the loss function. For example, in a neural network, the original weight matrix W of the heading correction layer is [0.85, 0.12]. The loss gradient is calculated as ∂L / ∂W = [0.02, -0.005], and the learning rate η = 0.01. After updating, W = W - η × ∂L / ∂W = [0.85 - 0.01 × 0.02, 0.12 - 0.01 × (-0.005)] = [0.8498, 0.12005]. Through multiple iterations (e.g., 10 iterations), the loss value decreases from 0.25 to 0.18, indicating that the model gradually adapts to the new data distribution.

[0146] Step S555: After each parameter adjustment, perform local verification on the flight emergency prediction model to ensure that the overall performance of the adjusted flight emergency prediction model on historical datasets and new datasets is not lower than the preset performance threshold.

[0147] For example, local validation can employ stratified sampling. For instance, 50 cases are randomly selected from the historical dataset and all 20 cases from the newly added dataset, and a comprehensive performance metric (such as Mean Absolute Error, MAE) is calculated. For example, if the historical data MAE increases from 1.2 degrees to 1.3 degrees (with an allowable threshold of 1.5), while the newly added data MAE decreases from 1.1 degrees to 1.0 degrees, the overall performance remains within acceptable limits. If adjustments cause the historical data MAE to exceed the limit (e.g., increase to 1.6 degrees), the parameters are rolled back and the learning rate is reduced and readjusted until the performance constraints of both datasets are met. After successful validation, the updated model parameters are persistently stored for subsequent emergency command and control.

[0148] Figure 2 A hardware entity diagram of a computer system provided as an embodiment of the present invention, such as... Figure 2 As shown, the hardware entity of the computer system 1000 includes a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can run on the processor 1001, and the processor 1001 executes the program to implement the steps in the method of any of the above embodiments.

Claims

1. An emergency command method for low-altitude flight, characterized in that, The method includes: acquiring real-time flight monitoring data of a target area; the real-time flight monitoring data includes aircraft status parameters, meteorological parameters, and airspace occupancy parameters; determining the type of emergency event within the target area based on a subset of abnormal data in the real-time flight monitoring data; invoking a flight emergency prediction model corresponding to the type of emergency event; the flight emergency prediction model is trained based on flight features and airspace features selected from historical emergency event data, and the types of input and output parameters for each flight emergency prediction model are consistent during the training and application phases; generating emergency command instructions based on the flight path adjustment strategy and emergency resource allocation strategy output by the flight emergency prediction model; executing the emergency command instructions, and determining the type of emergency event within the target area based on the results of the execution. The parameters of the flight emergency prediction model are updated based on real-time feedback data. The step of determining the type of emergency event within the target area based on abnormal data subsets in the real-time flight monitoring data includes: extracting a first data subset corresponding to aircraft status parameters, a second data subset corresponding to meteorological parameters, and a third data subset corresponding to airspace occupancy parameters from the real-time flight monitoring data; performing dynamic threshold filtering on the first, second, and third data subsets respectively to obtain abnormal aircraft data exceeding a preset aircraft status threshold in the first data subset, abnormal meteorological data exceeding a preset meteorological fluctuation threshold in the second data subset, and abnormal airspace data exceeding a preset airspace capacity threshold in the third data subset. The data includes: inputting the abnormal aircraft data, abnormal weather data, and abnormal airspace data into the event type association model to determine the priority and association parameters of the emergency event type; generating event feature tags containing the emergency event type based on the priority and association parameters, and matching the event feature tags with the calling conditions of the flight emergency prediction model; calling the flight emergency prediction model corresponding to the emergency event type includes: obtaining multiple candidate flight emergency prediction models associated with the event feature tags; each candidate flight emergency prediction model corresponds to a different emergency event handling stage, and the input parameter type used by each candidate flight emergency prediction model during the training stage is consistent with the parameter class in the current real-time flight monitoring data. The model is consistent with the target model; based on the historical verification results of the candidate flight emergency prediction models, the response weight and error range of each candidate flight emergency prediction model are determined; at least one target flight emergency prediction model is selected from the candidate flight emergency prediction models whose response weight is higher than a preset weight threshold and whose error range is lower than a preset error threshold; the abnormal aircraft data, abnormal meteorological data, and abnormal airspace data are reorganized according to the input format of the target flight emergency prediction model during the training phase to generate a standardized input dataset; the standardized input dataset is input into the target flight emergency prediction model to trigger the target flight emergency prediction model to perform flight path simulation and resource demand prediction, thereby obtaining flight path adjustment strategies and emergency resource allocation strategies.

2. The method according to claim 1, characterized in that, The training process of the flight emergency prediction model includes: acquiring a historical emergency event dataset; the historical emergency event dataset includes multiple processed emergency event cases, each emergency event case containing aircraft parameters, meteorological parameters, airspace parameters at the time of the emergency, and path adjustment records and resource allocation records after the emergency; extracting flight feature sets related to flight speed, altitude deviation, and heading angle offset from the aircraft parameters of each emergency event case, and extracting airspace feature sets related to airspace capacity, obstacle distribution, and temporary control areas from the airspace parameters; performing redundancy analysis on the flight feature sets and airspace feature sets, and removing... After removing redundant features whose correlation with the emergency response results is lower than a preset correlation threshold, an optimized target flight feature set and target airspace feature set are obtained. Based on the target flight feature set and target airspace feature set, multiple initial flight emergency prediction models are constructed, and each initial flight emergency prediction model is iteratively trained using cross-validation. During the iterative training process, the model parameters are dynamically adjusted according to the difference between the simulation results and actual processing results of each initial flight emergency prediction model for historical cases, until the difference is lower than a preset convergence threshold, at which point training stops, and the trained initial flight emergency prediction model is marked as the flight emergency prediction model.

3. The method according to claim 2, characterized in that, The redundancy analysis of the flight feature set and airspace feature set, removing redundant features whose correlation with the emergency event handling result is lower than a preset correlation threshold, to obtain an optimized target flight feature set and target airspace feature set includes: calculating a first correlation coefficient between each flight feature in the flight feature set and the emergency event handling result, and a second correlation coefficient between each airspace feature in the airspace feature set and the emergency event handling result; removing target flight features from the flight feature set whose first correlation coefficient is lower than the first correlation coefficient threshold, and removing target airspace features from the airspace feature set whose second correlation coefficient is lower than the second correlation coefficient threshold; performing a combination analysis on the remaining flight features and remaining airspace features after removal to determine the influence weight of different feature combinations on the handling result; selecting the top N core flight features by weight from the remaining flight features based on the influence weight, and selecting the top M core airspace features by weight from the remaining airspace features; and merging the core flight features and core airspace features into the target flight feature set and target airspace feature set.

4. The method according to claim 3, characterized in that, The step of constructing multiple initial flight emergency prediction models based on the target flight feature set and the target airspace feature set includes: dividing the historical emergency event dataset into a training subset and a validation subset; the training subset is used to generate initial flight emergency prediction models, and the validation subset is used to evaluate model performance; extracting training data corresponding to core flight features and core airspace features from the training subset based on the target flight feature set and the target airspace feature set; performing multiple rounds of fitting on the training data using a supervised learning algorithm to generate a first type of flight emergency prediction model; the first type of flight emergency prediction model is used to predict flight path adjustment strategies; and performing trend fitting on the training data using a time series analysis algorithm to generate a second type of flight emergency prediction model; the second type of flight emergency prediction model is used to predict emergency resource allocation strategies.

5. The method according to claim 4, characterized in that, The process of using a supervised learning algorithm to perform multiple rounds of fitting on the training data to generate a first-type flight emergency prediction model includes: determining a set of labeled results related to flight path adjustment in the training data; the labeled result set includes actual path change instructions executed in historical cases and corresponding execution effect scores; constructing a regression analysis model by using the core flight features and core airspace features as input variables and the path change instructions and execution effect scores as output variables; adjusting the weight coefficients in the model according to the deviation between the predicted path instructions and the actual path instructions in each round of fitting; determining that the regression analysis model has reached convergence when the change in deviation after K consecutive rounds of fitting is less than a preset change threshold, and saving the model parameters at this time as the first-type flight emergency prediction model; the process of using time series analysis... The algorithm performs trend fitting on the training data to generate a second type of flight emergency prediction model, including: extracting time series parameters related to resource allocation from the training data; the time series parameters include resource scheduling duration, resource consumption rate, and resource replenishment interval; performing stationarity tests on the time series parameters and differentiating non-stationary sequences until a stationary time series is obtained; constructing an autoregressive integral moving average model based on the stationary time series and determining the optimal order and lag parameters of the model; using the optimal order and lag parameters to backfit historical resource allocation data to generate a resource demand prediction curve; optimizing the parameters of the autoregressive integral moving average model based on the matching degree between the resource demand prediction curve and the actual resource usage curve, and labeling the optimized model as the second type of flight emergency prediction model.

6. The method according to claim 1, characterized in that, The execution of the emergency command instructions and the updating of the parameters of the flight emergency prediction model based on real-time feedback data after execution include: monitoring aircraft response data, resource scheduling status data, and airspace change data during the execution of the emergency command instructions; comparing the aircraft response data with the expected response data in the flight path adjustment strategy to generate a first deviation index; comparing the resource scheduling status data with the expected resource data in the emergency resource allocation strategy to generate a second deviation index; when the first deviation index or the second deviation index exceeds a preset fault tolerance threshold, triggering a model parameter update mechanism and calculating the adjustment amount of the model parameters based on the current deviation index; using an incremental learning algorithm to inject the adjustment amount into the flight emergency prediction model and re-verifying the prediction accuracy of the updated model.

7. The method according to claim 6, characterized in that, The step of injecting the adjustment amount into the flight emergency prediction model using an incremental learning algorithm includes: extracting a new dataset with the same type of input parameters as the model from the real-time feedback data; performing feature alignment processing on the new dataset to ensure that the feature dimensions of the new dataset are exactly the same as those used in the training phase; inputting the new dataset into the flight emergency prediction model without changing the original model structure, and calculating the loss function value between the output of the flight emergency prediction model and the actual feedback result; adjusting the weight parameters in the flight emergency prediction model using gradient descent based on the loss function value, so that the prediction error of the flight emergency prediction model for new data gradually decreases; and performing local verification on the flight emergency prediction model after each parameter adjustment to ensure that the overall performance of the adjusted flight emergency prediction model on historical datasets and new datasets is not lower than a preset performance threshold.

8. A computer system comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 7.

Images