Search

CN-122024533-A - AI-based low-altitude airspace information center cooperative control method

CN122024533ACN 122024533 ACN122024533 ACN 122024533ACN-122024533-A

Abstract

The invention relates to the technical field of airspace control, and discloses a low-altitude airspace information center collaborative control method based on AI, the method comprises the step of constructing a closed-loop intelligent decision mechanism through six core steps of multisource perception fusion, abnormal behavior discrimination, dynamic risk assessment, environment constraint re-planning, communication collaborative optimization and collaborative execution decision. The method comprises the steps of achieving microsecond time synchronization through a Beidou or GPS satellite time service system, improving target identification accuracy through combining a weighted evidence fusion theory, combining an R-tree and space-time bounding box dual rule engine to accurately detect abnormal flight, quantifying collision risks among aircrafts through a three-layer stacked multi-head graph attention network, achieving track re-planning under environmental constraint through a multi-dimensional environment voxel model, and carrying out communication collaborative optimization through a ground and satellite mixed networking architecture, so that management and control communication is enabled to avoid terrain shielding and electromagnetic interference, global coverage is achieved, accurate adjustment is achieved, and accuracy, instantaneity and safety of management and control are guaranteed.

Inventors

  • WANG FEI
  • WANG XUESONG
  • ZENG XUEFENG
  • DU QINGFENG

Assignees

  • 北京迅奥科技有限公司

Dates

Publication Date
20260512
Application Date
20260210

Claims (10)

  1. 1. The AI-based low-altitude airspace information center collaborative management and control method is characterized by comprising the following steps of: the method comprises the steps of acquiring initial perception data of a low-altitude airspace in a multi-mode manner to obtain a perception data set; Performing space-time calibration and multi-source feature fusion on the perception data set to obtain a structured airspace situation data set; comparing identity association with rules based on the airspace situation data set, and if the airspace situation data set of the current airspace position falls into a no-fly zone or is in a track deviation situation and the deviation value exceeds a preset deviation threshold value, adding a corresponding aircraft ID into an identification list to generate an abnormal aircraft identification list; performing dynamic risk assessment according to the abnormal aircraft identification list, the dynamic aircraft spatial distribution diagram and the risk propagation model based on the graph attention network to obtain a high risk interaction pair set; based on the related aircrafts and the influence ranges in the high-risk interaction pair set, combining a multidimensional environment element model and the current position of the aircrafts, predicting conflict tracks through a track prediction network and carrying out environment constraint track re-planning so as to generate a compliance track adjustment instruction; according to the compliance track adjustment instruction, carrying out communication channel quality assessment through a directional antenna array and a convolution self-encoder, and adaptively adjusting communication parameters to generate a communication parameter optimization instruction; And the compliance track adjustment instruction and the communication parameter optimization instruction are coded and classified in a combined mode to generate a normal aircraft state list and an abnormal aircraft state list, and multi-agent game scheduling and multi-department collaborative abnormal early warning response are respectively executed to obtain a low-altitude airspace collaborative management and control execution strategy.
  2. 2. The AI-based low-altitude airspace information center collaborative management and control method according to claim 1, wherein the step of comparing identity association with rules based on airspace situation data sets, and generating an abnormal aircraft identification list if an offset value exceeds a preset offset threshold value comprises the steps of: Loading a no-fly area rule base stored in GeoJSON format, constructing an R-tree space index structure, loading a registered flight plan database containing a take-off time window, a route, an altitude layer and an operation main body field, and constructing a flight plan space-time bounding box index; Extracting a target ID, longitude and latitude, altitude and aircraft type in the airspace situation data set, performing R-tree range query on longitude and latitude coordinates of each aircraft, and judging whether the aircraft falls into a geometrical boundary of a no-fly area; If the current actual position does not fall into the no-fly zone, extracting a reference track point sequence in the current aircraft flight plan, and acquiring the Euclidean distance between the current actual position and the nearest reference track point in the airspace situation data set; if the Euclidean distance exceeds a preset offset threshold range in the horizontal direction, determining that the track deviation exceeds the limit; writing the original perceived data snapshot in the aircraft ID, event type, occurrence time and airspace situation data set which meet the invasion of the no-fly zone or the deviation of the track from the overrun condition into an abnormal aircraft identification list; And pushing the abnormal aircraft identification list to a log system for persistent storage and synchronously triggering a driving-off instruction generator, verifying whether the abnormal aircraft has a legal communication link and an identity authentication state, and marking the abnormal aircraft identification list as a high-risk target if the abnormal aircraft has no legal communication link and the identity authentication state so as to generate the abnormal aircraft identification list containing the aircraft ID, the deviation amount and the risk level.
  3. 3. The AI-based low-altitude spatial information center collaborative management method according to claim 1, wherein performing a dynamic risk assessment according to the abnormal aircraft identification list, a dynamic aircraft spatial distribution map, and a risk propagation model based on a graph attention network to obtain a set of high-risk interaction pairs comprises: Acquiring three-dimensional boundary grid data in a CityGML format from airspace management geographic information, wherein the three-dimensional boundary grid data comprises a building LOD2 model, a terrain elevation and an airspace layered structure; Extracting the positions and the speed vectors of all the aircrafts in the airspace situation data set, projecting the positions and the speed vectors to the three-dimensional boundary grid, and constructing a dynamic aircraft space distribution diagram with 10 frames updated per second by combining the aircraft IDs and the risk grades in the abnormal aircraft identification list; Traversing all aircraft pairs in the dynamic aircraft space distribution diagram, calculating Euclidean distance and relative speed vectors, and generating an N multiplied by N dimensional aircraft space matrix, wherein N is the total number of aircraft in a airspace; The aircraft space matrix is encoded together with the aircraft type, physical size and maximum maneuvering overload capacity attribute in the airspace situation data set into graph node characteristics and edge weights, and the graph node characteristics and edge weights are input into a graph attention network model formed by three layers of stacked multi-head graph attention layers; The drawing and annotating force network model carries out supervision training based on a plurality of real or simulated close-range events in a historical conflict case library, and outputs collision risk scores in the interval of 0 to 1 of each pair of aircrafts; and marking the aircraft pairs with scores larger than a preset threshold value as high-risk interaction pairs, and combining risk grades in the abnormal aircraft identification list to obtain a high-risk interaction pair set containing aircraft ID pairs, collision risk scores and associated abnormal identifications.
  4. 4. The AI-based low-altitude spatial information center collaborative management and control method according to claim 1, wherein based on the associated aircraft and the range of influence in the set of high-risk interaction pairs, in combination with a multidimensional environment voxel model and a current position of the aircraft, predicting a collision track through a track prediction network and performing environment constraint track re-planning to generate a compliance track adjustment instruction, comprising: Receiving echo intensity data from a weather radar, interference power spectral density data of an electromagnetic spectrum scanner and three-dimensional static obstacle grid data; Uniformly mapping echo intensity data, interference power spectrum density data and three-dimensional static obstacle grid data to a three-dimensional voxel space with 5 multiplied by 5 resolution by a voxelization engine, respectively generating a meteorological disturbance layer, an electromagnetic disturbance layer and an obstacle distribution layer, and recording a voxel data set of horizontal wind speed, vertical wind speed, turbulence intensity, signal attenuation index, co-frequency interference source position and static obstacle voxel occupation state in each layer; Carrying out space-time interpolation on the voxel data set by adopting a sliding time window, and deducing the environment state evolution in the future preset time by utilizing a linear prediction or LSTM time sequence model to obtain a multidimensional environment voxel model; Extracting a wind speed gradient threshold value, a signal attenuation region coordinate and a static obstacle envelope surface from the multidimensional environment element model as safety constraint parameters; Extracting the current position and speed in the airspace situation data set according to the ID and the influence range of the associated aircraft in the high-risk interaction pair set, and inputting the current position and speed in the airspace situation data set into a track prediction network in combination with safety constraint parameters, wherein a track prediction network encoder receives historical track points in the past preset time, and a decoder predicts a plurality of track points in the future preset time to form conflict track prediction; performing point-by-point inquiry on the track points predicted in the conflict track prediction to obtain an inquiry result; and inquiring environment parameters of corresponding voxels in the multidimensional environment voxel model point by point, judging whether the situation of violating wind speed gradient, signal intensity or obstacle occupation constraint exists, if so, calling a trajectory re-planner to conduct environment constraint trajectory re-planning under the collision risk score of the high risk interaction pair set and the influence range constraint, and generating a new waypoint sequence conforming to the environment constraint so as to generate an aircraft compliance trajectory adjustment instruction comprising an avoidance strategy, a time window and three-dimensional path points.
  5. 5. The AI-based low-altitude-spatial-domain information center collaborative management method according to claim 4, characterized by performing communication channel quality assessment with a convolutional self-encoder through a directional antenna array according to the compliance track adjustment instruction, and adaptively adjusting communication parameters to generate a communication parameter optimization instruction, comprising: extracting three-dimensional path points and time windows in the aircraft compliance track adjustment instruction, calculating the sight paths of each waypoint and the nearest low-altitude private network base station by combining a three-dimensional city model, and determining the optimal communication echo sampling time and the ground station receiving azimuth angle; Activating a base station antenna through a directional antenna array at the sampling moment according to the compliance track adjustment instruction to enable a main lobe of the base station antenna to be aligned to the azimuth angle, receiving a downlink communication signal of an aircraft, and generating an original echo signal stream with microsecond time-frequency marks; inputting the original echo signal stream to a four-layer 1D convolution self-encoder through the convolution self-encoder, extracting multipath effect, co-channel interference and atmospheric noise characteristics through reconstruction errors, and performing communication channel quality assessment to obtain a communication channel quality index vector containing CINR, time delay expansion and Doppler frequency shift; According to the communication channel quality index vector, communication parameters such as the aircraft transmitting power, the modulation mode, the retransmission strategy and the like are adaptively adjusted according to a preset mapping table, and the avoidance strategy and the time window in the compliance track adjustment instruction are combined to generate a communication parameter optimization instruction synchronous with the track adjustment.
  6. 6. The AI-based low-altitude airspace information center collaborative management and control method according to claim 5, wherein the co-ordination track adjustment instruction and the communication parameter optimization instruction are jointly encoded and classified to generate a normal aircraft state list and an abnormal aircraft state list, and the multi-agent game scheduling and the multi-department collaborative anomaly early warning response are respectively executed to obtain a low-altitude airspace collaborative management and control execution strategy, which comprises: Extracting a waypoint sequence, a speed profile and a obstacle avoidance margin in the aircraft compliance track adjustment instruction, and transmitting power, a modulation mode and a retransmission strategy in a communication parameter optimization instruction, and jointly encoding the compliance track adjustment instruction and the communication parameter optimization instruction into 128-dimensional running state feature vectors, wherein the first 64 dimensions represent track adjustment features, and the later 64 dimensions represent communication parameter features; the operation state feature vector is input into an integrated classifier formed by three basic models of a training gradient lifting decision tree, a random forest and a lightweight gradient elevator and classified, and the integrated classifier is trained based on historical normal and abnormal flight records and outputs a binary classification result of the operation state of the aircraft through a weighted voting mechanism; according to the binary classification result, combining collision risk scores in the high risk interaction pair set and risk grades in the abnormal aircraft identification list to generate a normal aircraft state list and an abnormal aircraft state list; Extracting an aircraft ID, an operation state feature vector and a risk level from each item in the abnormal aircraft state list, establishing a triplet relation of an aircraft-event-instruction through a Neo4j graph database, matching a preset emergency response mode library, respectively executing multi-department collaborative abnormal early warning response, and generating an emergency response scheme comprising multi-department linkage actions of public security, civil aviation and emergency management, a resource scheduling list and strict time window constraint; Inputting the aircraft ID and the running state feature vector in the normal aircraft state list into a multi-agent game scheduling optimizer, respectively executing multi-agent game scheduling, and solving globally optimal take-off time slots, route assignment and altitude layer configuration by Nash equilibrium to obtain a flight scheduling scheme; And loading the emergency response scheme and the flight scheduling scheme to a distributed task scheduling engine, decomposing the emergency response scheme and the flight scheduling scheme into atomic operations, and distributing the atomic operations to standard interfaces of public security, civil aviation and emergency management units through a message queue to obtain a low-altitude airspace cooperative management and control execution strategy.
  7. 7. The AI-based low-altitude spatial information center collaborative management and control method according to claim 6, wherein inputting the operational state feature vector into an integrated classifier composed of three base models of a training gradient lifting decision tree, a random forest and a lightweight gradient lifting machine and classifying the integrated classifier comprises: extracting a training sample set containing normal flight records and abnormal flight records from a historical flight record database; extracting 128-dimensional characteristics of a waypoint sequence, a speed profile, an obstacle avoidance margin, a transmitting power, a modulation mode and a retransmission strategy from each flight record in the training sample set, and labeling class labels to obtain a labeled training data set; Dividing the labeling training data set into a training set and a verification set according to the proportion of 7:3, respectively training three base models of a gradient lifting decision tree, a random forest and a lightweight gradient lifting machine by using the training set, and performing super-parameter tuning on the training set to obtain three optimized base models; Evaluating the accuracy, recall, F1 score and AUC value of the three base models on a verification set, and determining weighted voting weights of 0.4, 0.3 and 0.3 of the three base models according to the performance of the verification set; And respectively inputting the running state feature vectors into the three base models to obtain probability prediction values of the base models, carrying out weighted summation on the probability prediction values according to the weighted voting weights to obtain final classification probability, judging that the running state feature vectors are abnormal running classes if the final classification probability is larger than a preset threshold value, otherwise judging that the running state feature vectors are normal running classes, and outputting binary classification results of the running states of the aircraft.
  8. 8. The AI-based low-altitude spatial information center collaborative management and control method according to claim 6, wherein inputting aircraft IDs and operational state feature vectors in the normal aircraft state inventory to a multi-agent gaming schedule optimizer respectively performs multi-agent gaming scheduling to solve globally optimal departure time slots, route assignments, and altitude layer configurations with nash equalization, comprising: Extracting an aircraft ID, expected take-off time, a destination, a voyage, aircraft performance parameters and operation state feature vectors of each aircraft from the normal aircraft state list, constructing a multi-agent game model, regarding each aircraft as an independent agent and defining a strategy space thereof, and obtaining a strategy space comprising a take-off time slot set, a voyage set and a height layer set; The utility function comprehensively considers the departure time deviation cost, the route length cost, the altitude layer preference cost and the conflict risk cost, and sets constraint conditions based on the current airspace capacity dynamic quota to obtain a game model containing the utility function and the constraint conditions; Solving Nash equilibrium for the game model by adopting an iterative optimal response algorithm, initializing each agent random selection strategy combination, selecting an optimal strategy which enables a self utility function to be maximum under the condition that other agent strategies are fixed in each iteration, and carrying out iterative updating to obtain Nash equilibrium strategies; And verifying whether the Nash equilibrium strategy meets all constraint conditions, if so, outputting the globally optimal take-off time slot, the route assignment and the altitude layer configuration of each aircraft, and combining the globally optimal take-off time slot, the route assignment and the altitude layer configuration into a flight scheduling scheme to obtain a detailed flight plan comprising take-off time, a route point sequence, cruising altitude and predicted arrival time.
  9. 9. The AI-based low-altitude airspace information center collaborative management method according to claim 6, wherein for each item in the abnormal aircraft state list, extracting an aircraft ID, an operation state feature vector and a risk level, establishing a triplet relation of an aircraft-event-instruction through a Neo4j graph database, matching a preset emergency response mode library, respectively executing multi-department collaborative anomaly early warning response, and generating an emergency response scheme including multi-department linkage actions, resource scheduling lists and strict time window constraints of public security, civil aviation and emergency management, comprising: Extracting the aircraft ID, the running state feature vector, the risk level, the abnormal event type and the occurrence time of each item from the abnormal aircraft state list; The aircraft ID is used as an aircraft node, an abnormal event type is used as an event node, a compliance track adjustment instruction and a communication parameter optimization instruction are used as instruction nodes, a triplet relation of the aircraft-event-instruction is established in a Neo4j graph database, wherein the aircraft node and the event node are connected through an occurrence relation, and the event node and the instruction node are connected through a triggering relation; according to the risk level and the abnormal event type, matching a corresponding emergency response mode from a preset emergency response mode library, wherein the emergency response mode library comprises response flows classified according to the risk level and the event type, and comprises a level I response mode, a level II response mode and a level III response mode; For the I-level response mode, generating a mild response flow comprising warning notification sent by an empty control department and autonomous avoidance performed by the aircraft, for the II-level response mode, generating a moderate response flow comprising forced instruction sent by the empty control department, coordination airspace resource of civil aviation department and forced avoidance performed by the aircraft; respectively executing multi-department collaborative anomaly early warning response according to the matched emergency response mode, and generating a multi-department linkage action list; Generating a resource scheduling list, determining the number of personnel to be scheduled, equipment types, vehicle configuration and communication channels of each department, and determining a resource deployment position according to the current position and a predicted track of an abnormal aircraft; Setting strict time window constraint according to emergency degree of abnormal event to generate emergency response scheme including multi-department linkage action, resource scheduling list and time window constraint.
  10. 10. A computer readable storage medium for storing computer readable instructions which when read by a computer are capable of operating an AI-based low altitude spatial information center collaborative management method according to any of claims 1-9.

Description

AI-based low-altitude airspace information center cooperative control method Technical Field The invention relates to the technical field of airspace control, in particular to a low-altitude airspace information center collaborative control method based on AI. Background With the deep fusion of the general aviation and unmanned aerial vehicle technology and urban air traffic, the low-altitude airspace management and control provides higher requirements for the precise identification, dynamic risk assessment and collaborative communication optimization capability of the aircraft. Currently, low-altitude airspace management is gradually evolving from static management based on fixed rules to intelligent collaborative decision-making based on AI. Although the traditional low-altitude control method has a certain progress in the aspects of multi-mode sensor data acquisition, basic track monitoring and simple abnormal alarming, the traditional low-altitude control method still has obvious defects in the aspects of sensing data, risk assessment, track re-planning and communication optimization and finally obtaining a closed-loop decision mechanism of a collaborative control execution strategy, and cannot effectively support real-time collaboration and intelligent control of all elements of a low-altitude airspace. The existing scheme is based on a preset no-fly zone rule base or a fixed airspace classification structure, lacks a dynamic identification and self-adaptive adjustment mechanism for high-risk aircraft interaction relation, complex environment constraint and communication channel quality change, and meanwhile, rarely uses actual execution feedback for reversely optimizing multimode perception fusion weight, risk assessment model parameters or a track re-planning strategy after an abnormal event triggers an emergency response. The method ensures that when the method is used for coping with abnormal situations such as complex meteorological condition change, electromagnetic interference burst, aircraft short-distance conflict or illegal invasion of a low-altitude airspace, the response speed and decision accuracy have hysteresis, and the intelligent depth and the safety guarantee breadth of the cooperative control of the low-altitude airspace are limited. Therefore, how to construct an AI-driven low-altitude airspace information center collaborative management and control method with the capability of obtaining cross-department collaborative execution strategies from multi-mode perception data to anomaly identification and risk assessment to trajectory re-planning and communication optimization finally, so as to continuously mine the interactive risk, environment constraint change and communication channel quality evolution law of the aircraft from real-time operation data, dynamically optimize multi-source perception fusion, intelligent risk prediction, environment constraint trajectory re-planning, communication parameter self-adaptive adjustment and multi-department emergency response strategies according to the method, and the method becomes a technical problem to be solved in the current low-altitude airspace management field. Disclosure of Invention The invention provides an AI-based low-altitude airspace information center collaborative control method, which solves the technical problems that a closed-loop intelligent decision mechanism which is cooperatively executed from multi-mode perception is lacking in the prior art, high-risk aircraft interaction and complex environment constraint cannot be dynamically and adaptively identified, and response speed and decision accuracy are lagged. The invention provides an AI-based low-altitude airspace information center collaborative control method, which comprises the following steps: In a first aspect, an AI-based low-altitude spatial information center collaborative control method includes: Acquiring initial perception data of a low-altitude airspace through a radar, an optical sensor, an ADS-B receiver, an infrared detector and other multi-mode sensors to obtain a perception data set containing a time stamp, a space coordinate, a speed vector and signal characteristics; Performing space-time calibration by using the timestamp information in the perception data set, and performing multi-source feature fusion on space coordinates, speed vectors and signal features to obtain a space domain situation data set containing aircraft position, track and identity features; Extracting identity characteristics and track information in the airspace situation data set to perform identity association and rule comparison, and adding a corresponding aircraft ID into an identification list to generate an abnormal aircraft identification list if the airspace situation data set of the current airspace position falls into a no-fly zone or is in a track deviation situation and the deviation value exceeds a preset deviation threshold (the horizontal deviation is 50 meters or