CN-121477193-B - Multi-source data fusion type airplane wake vortex real-time identification and alarm method and system

CN121477193BCN 121477193 BCN121477193 BCN 121477193BCN-121477193-B

Abstract

The invention relates to the technical field of civil aviation operation safety and intelligent air traffic control, and discloses a multi-source data fusion method and system for identifying and alarming wake vortexes of an airplane in real time. The method comprises the steps of carrying out multi-source data acquisition through various radars, carrying out uniform time stamp alignment on each source data to obtain a wake vortex data set, constructing a parallel space-time deep learning wake vortex identification model, extracting time domain features through a gating circulation unit, extracting space domain features through a convolutional neural network, fusing the time domain features and the space domain features, finally predicting the identification result and the existence probability of wake vortices, carrying out judgment by combining dynamic probability thresholds, restraining false alarm and false alarm, finally generating a wake vortex identification result and a judgment result, pushing and outputting the wake vortex identification result and the judgment result to each terminal to prompt and record logs, and sending the wake vortex identification result and the judgment result to a tower and a unit. The invention realizes real-time and reliable identification of wake vortexes under complex weather and multi-machine conditions, improves the identification accuracy and robustness, and reduces false alarm and missing alarm.

Inventors

WANG XUAN
LI SHANGJUN
Dai Qianqiao
WANG QIANHUI
PAN WEIJUN
ZUO QINGHAI

Assignees

中国民用航空飞行学院

Dates

Publication Date: 20260508
Application Date: 20260107

Claims (8)

1. The method for identifying and alarming the wake vortexes of the airplane by multi-source data fusion in real time is characterized by comprising the following steps of: step 1, multi-source data acquisition is carried out by using an airport foundation LiDAR/radar, a runway edge wind sensor, two Doppler coherent laser radars A and B and a profile radar; Step 2, space-time alignment and fusion, namely performing uniform time stamp alignment on all source data of airport foundation LiDAR/radar, doppler laser coherent radar and wind profile radar to obtain a wake vortex data set; Step 3, wake vortex identification, namely constructing a parallel space-time deep learning wake vortex identification model comprising a gating circulation unit and a convolutional neural network, preprocessing a wake vortex data set, and dividing the wake vortex data set into two input parts, namely inputting sequences Sum block input Inputting the sequence The input gating circulation unit extracts time domain characteristics and inputs blocks The method comprises the steps of inputting the characteristics into a convolutional neural network, extracting spatial domain characteristics, fusing the temporal domain characteristics and the spatial domain characteristics, and finally predicting the recognition result and the existence probability of wake vortexes; Step 4, a probability and dynamic threshold strategy is that the dynamic probability threshold is combined to judge, and false alarm and missing report are restrained; Step 5, alarm generation, pushing and recording, namely generating a wake vortex identification result and a judgment result, pushing and outputting the wake vortex identification result and the judgment result to a tower and a unit, prompting to each terminal, and recording a log; the probability and dynamic threshold strategy is that the model is directly adopted to output the existence probability of wake vortex at the moment t As a judging basis, combining the on-site atmospheric stability, the background wind speed/wind direction, the airplane model and the operation stage given by the runway edge wind sensor and the wind profile radar to generate an adaptive threshold value at the moment t Setting feature vectors And (3) collecting factors including on-site atmospheric stability, background wind speed/wind direction, airplane type and operation stage, and writing a threshold value as follows: ; Wherein, the Is a basic threshold bias; generating a learnable weight vector of the model for the threshold; And Respectively a minimum value and a maximum value of the threshold value, and is used for limiting the threshold value range; clipping function for section for limiting input value to [ ]; A and B are continuously and simultaneously satisfied in a plurality of frames on the recognition result of the Doppler coherent laser radar Triggering alarm, merging and limiting current for repeated event, and prompting according to threshold amplitude and duration.
2. The method for identifying and alarming the wake vortexes of the multi-source data fusion aircraft according to claim 1 is characterized in that in the step 1, the vertical interval distance between the A Doppler coherent laser radar and the B Doppler coherent laser radar is not more than 10m, the scanning period is separated by half a period, in the detection process, a range-altitude display scanning mode is adopted to scan a section along the direction perpendicular to the flight direction of the aircraft to form a real-time wind field data set, radar measurement values comprise time stamps and position parameters, the position parameters comprise the range, the azimuth angle and the elevation angle, wind profile radars synchronously acquire meteorological elements near an airport and the radars, including the wind speed, the wind direction, the temperature, the humidity, the boundary layer stability and the turbulence intensity, and are used for acquiring background wind field and atmospheric stability data, and three-dimensional track information of an aircraft entering the departure field is acquired through an airport foundation radars and a monitoring system, including the longitude and latitude/distance, the altitude and the heading/the ground speed.
3. The method for identifying and alarming wake vortexes of an aircraft based on multi-source data fusion according to claim 1, wherein the method in step 1 further comprises the steps of taking wind field information acquired by each of the two Doppler coherent laser radars A and B as independent input channels, automatically learning feature fusion by a deep learning network to form complete wind field data, namely wind field data acquired by the radars ; The step 2 specifically comprises the following steps: Converting an aircraft track acquired by an airport foundation LiDAR/radar from a geographic coordinate system to a runway coordinate system taking a runway center line as a reference, resampling and interpolating on a uniform time step and a space grid, converging to form a data block with consistent space and time, and respectively carrying out normalization processing on input data of two Doppler coherent laser radar wind fields A and B by considering the influence of a background wind field so as to facilitate training and identification of a parallel deep learning network, wherein the calculation expression is as follows: ; in the formula, For wind field data acquired by the radar, And sigma are the mean and standard deviation calculated in a given time window or spatial neighborhood respectively, Is a standardized result.
4. The method for identifying and alarming wake vortexes of an aircraft based on multi-source data fusion according to claim 3, wherein in said step 3, a sequence is input The input gating circulation unit extracts time domain features as follows: Step 311, the gate control circulation unit inputs at t time With the last time hidden state For input, the hidden state at the moment t is output through forgetting and introducing control information of the update gate and the reset gate ; Two-way gating circulation units are adopted to carry out two-way coding and learning representation on the time domain information of the Doppler coherent laser radar A and the Doppler coherent laser radar B, so that complete past and future context information of each point in an input sequence is concerned; step 312, using The radial velocity wind field is processed by a layer bidirectional GRU network, and comprises a forward sequence and a backward sequence with a plurality of independent hidden layers, and then the feed forward is fused to the same output layer, hidden layer vectors of forward and backward input characteristic processes are respectively obtained through learning And : , ; In the formula, Mapping operators for the GRUs; and (3) with Respectively represent standardized results Forward and backward inputs as sequence data; Step 313, obtaining fusion characteristics by splicing hidden layer vector output of the forward and backward input characteristic process It is calculated as: ; in the formula, Is a time domain fusion feature.
5. The method for identifying and alarming wake vortexes of an aircraft based on multi-source data fusion according to claim 4, wherein in said step 3, the blocks are input The spatial domain features are extracted by the input convolutional neural network, and the spatial domain features are specifically as follows: Step 321, dividing wind field data into a plurality of patches, and then applying a convolutional neural network to the obtained Patch set, wherein the convolutional process is expressed as follows: ; in the formula, To represent a feature map of the output of the convolutional layer, A set of convolution kernels representing output channels c, Non-linear activation function for element by element; step 322, activating after each convolution layer to generate a spatial domain feature map, then spatial domain features The method comprises the following steps: ; in the formula, To extract a function based on characteristics of a convolutional neural network, the function comprises a plurality of convolutional layers, an activating layer and an optional normalizing layer, which are used for extracting characteristic diagrams from input Extracting spatial domain features; step 323, time-space feature fusion, namely adopting a gating weighting fusion module to fuse the time domain into the features at first And spatial domain features Mapping to consistent dimensions: ; in the formula, And (3) with In order for the parameters to be able to be learned, 、 Respectively is 、 The unified dimension characteristics after linear mapping; step 324, inputting the two paths of the unified dimension characteristics into a gating unit after splicing to obtain a gating vector : ; In the formula, And (3) with In order for the parameters to be able to be learned, Representing feature vector stitching; Step 325, weighting and fusing the two paths of the unified dimension features based on the gating vector to obtain a fused feature F: ; in the formula, Representing element-by-element multiplication; step 326, inputting the fusion characteristic F into a fully-connected classification network to obtain the wake vortex existence probability: ; in the formula, For the probability of the presence of wake vortexes, The network is classified for full connectivity.
6. The method for identifying and alarming wake vortexes in real time by multi-source data fusion according to claim 5, wherein when training the deep learning network, multiplicative speckle noise enhancement is adopted for the training data set before preprocessing, which is defined as: ; Wherein, the And for standard deviation sigma from interval Randomly sampling to form a multi-intensity noise distribution; Wind field data values for the original training samples; is the wind field data value after noise enhancement.
7. The method for identifying and alarming wake vortexes of multi-source data fusion aircraft according to claim 6, wherein in step 5, the system fuses aircraft track data acquired by airport foundation LiDAR/radar, associates wake vortexes identification result with specific flight/model, runway and azimuth and space position/altitude slice elements, and characterizes relation between wake vortexes and aircraft operation, when wake vortexes exist probability Crossing adaptive thresholds And generating events when the duration and dual radar consistency constraints are satisfied, the events including numbers, time, spatial locations, probabilities/thresholds, levels, associated flights/models, and recommended treatments; The output channel comprises a tower display and acousto-optic reminding, an operation system interface, a unit end FMS/EFB message, merging/deduplication, frequency limiting and life cycle management of events, pushing a structured message and a visual prompt to the unit end through a system bus, wherein the message content comprises a time stamp, a runway and azimuth, a flight/model, a existence probability and threshold value, an alarm level, uncertainty and spatial position/altitude slicing and suggested disposal actions, all events and user interactions are reserved in a log form, and the log records identification results, threshold parameters, alarm life cycle, personnel confirmation and disposal results and is used for post-hoc traceability, statistical evaluation and model calibration.
8. The utility model provides a multisource data fusion's aircraft wake vortex real-time identification and warning system which characterized in that includes: The acquisition module is used for carrying out multi-source data acquisition through an airport foundation LiDAR/radar, a runway edge wind sensor, two Doppler coherent laser radars A and B and a profile radar; the alignment module is used for executing uniform time stamp alignment on all source data of the airport foundation LiDAR/radar, the Doppler laser coherent radar and the wind profile radar to obtain a wake vortex data set; the recognition module is used for constructing a parallel space-time deep learning wake vortex recognition model, comprising a gating circulation unit and a convolutional neural network, preprocessing a wake vortex data set and dividing the wake vortex data set into two input parts, namely, inputting sequences Sum block input Inputting the sequence The input gating circulation unit extracts time domain characteristics and inputs blocks The method comprises the steps of inputting the characteristics into a convolutional neural network, extracting spatial domain characteristics, fusing the temporal domain characteristics and the spatial domain characteristics, and finally predicting the recognition result and the existence probability of wake vortexes; the alarm module is used for judging by combining with the dynamic probability threshold value and inhibiting false alarm and missing alarm; the output module is used for generating, pushing and recording alarms, namely generating wake vortex identification results and judgment results, pushing and outputting the wake vortex identification results and judgment results to the tower and the unit to each terminal to prompt and record logs; the method for judging by combining the dynamic probability threshold value is used for inhibiting false alarm and false omission, and specifically comprises the steps of directly adopting a model to output the existence probability of wake vortexes at the moment t As a judging basis, combining the on-site atmospheric stability, the background wind speed/wind direction, the airplane model and the operation stage given by the runway edge wind sensor and the wind profile radar to generate an adaptive threshold value at the moment t Setting feature vectors And (3) collecting factors including on-site atmospheric stability, background wind speed/wind direction, airplane type and operation stage, and writing a threshold value as follows: ; Wherein, the Is a basic threshold bias; generating a learnable weight vector of the model for the threshold; And Respectively a minimum value and a maximum value of the threshold value, and is used for limiting the threshold value range; clipping function for section for limiting input value to [ ]; A and B are continuously and simultaneously satisfied in a plurality of frames on the recognition result of the Doppler coherent laser radar Triggering alarm, merging and limiting current for repeated event, and prompting according to threshold amplitude and duration.

Description

Multi-source data fusion type airplane wake vortex real-time identification and alarm method and system Technical Field The invention relates to the technical field of civil aviation operation safety and intelligent air traffic control, in particular to a method and a system for identifying and alarming wake vortexes of an airplane by multi-source data fusion in real time. Background The wake vortex hazard is derived from wingtip vortex, and forms rolling moment and pneumatic disturbance on a subsequent aircraft, so that an interval is required to be strictly managed in a terminal area. Airport sides typically operate according to visual and weather estimates or according to fixed interval rules of the model class (Heavy/Medium/Light). The Doppler coherent laser radar can measure radial wind vectors through a coherent detection principle, the RHI mode carries out section scanning along the direction perpendicular to a flight path, wing tip vortex can be detected, the wind profile radar provides background wind field and atmosphere stability information, and the foundation LiDAR/radar and monitoring system can provide three-dimensional flight paths (longitude, latitude, distance, altitude, heading, ground speed and the like) of an aircraft. At present, a wake vortex detection and identification method based on single radar data is difficult to fuse with multi-source information such as tracks, weather and the like, and cannot form a complete and accurate space-time background wind field and a wake vortex field. The real-time wake vortex strength and dissipation rate cannot be reflected only by depending on the model type and the operation rule of a fixed interval, so that safety redundancy or efficiency is insufficient, and the time sequence dependence and the space structure characteristic are difficult to capture simultaneously based on a single-channel time sequence or a single-frame image recognition network. The time and space coordinates of multi-source data (foundation radar, bicoherence Doppler laser radar A/B, wind profile radar) are difficult to align uniformly, so that the data are difficult to splice and model directly. Wake vortex observation has strong time sequence and spatial structure, and the traditional method is difficult to fully model, so that the joint characterization of the time sequence and the spatial structure by the model is insufficient, and the robustness and the generalization capability are limited. The probability uncertainty of the identification result under different meteorological and model working conditions is high, the threshold setting lacks self-adaptability, the dynamic threshold based on working conditions (stability, background wind, model and operation stage) is lacking, and false alarm are difficult to be compatible. Disclosure of Invention Aiming at the problems, the invention aims to provide a method and a system for identifying and alarming the wake vortexes of an airplane in real time by multi-source data fusion, which are used for constructing a set of engineering landing multi-source data fusion and on-line identification scheme, realizing the real-time and reliable identification of the wake vortexes under complex weather and multi-machine conditions, improving the identification accuracy and robustness and reducing false alarm and missing alarm. The technical proposal is as follows: A multi-source data fusion airplane wake vortex real-time identification and alarm method comprises the following steps: step 1, multi-source data acquisition is carried out by using an airport foundation LiDAR/radar, a runway edge wind sensor, two Doppler coherent laser radars A and B and a profile radar; Step 2, space-time alignment and fusion, namely performing uniform time stamp alignment on all source data of airport foundation LiDAR/radar, doppler laser coherent radar and wind profile radar to obtain a wake vortex data set; Step 3, wake vortex identification, namely constructing a parallel space-time deep learning wake vortex identification model comprising a gating circulation unit and a convolutional neural network, preprocessing a wake vortex data set, and dividing the wake vortex data set into two input parts, namely inputting sequences Sum block inputInputting the sequenceThe input gating circulation unit extracts time domain characteristics and inputs blocksThe method comprises the steps of inputting the characteristics into a convolutional neural network, extracting spatial domain characteristics, fusing the temporal domain characteristics and the spatial domain characteristics, and finally predicting the recognition result and the existence probability of wake vortexes; Step 4, a probability and dynamic threshold strategy is that the dynamic probability threshold is combined to judge, and false alarm and missing report are restrained; and 5, generating, pushing and recording the alarm, namely generating a wake vortex identification result and a judgment result,