CN-121982944-A - Low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction

CN121982944ACN 121982944 ACN121982944 ACN 121982944ACN-121982944-A

Abstract

The application discloses a low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction, and belongs to the technical field of low-altitude flight safety. The method comprises the steps of obtaining non-cooperative target track data, carrying out feature extraction and vectorization, inputting a track feature vector sequence into a pre-trained multi-hidden-layer long-short-term memory cyclic neural network LSTM, outputting track prediction vectors and probability distribution, constructing a three-dimensional probability reachable set representing a flight influence range based on the prediction probability distribution, carrying out space geometric intersection analysis on the three-dimensional probability reachable set and a three-dimensional flight protection area of a cooperative aircraft, judging flight conflict risks, carrying out real-time rolling update on the track data through a sliding time window, and carrying out track prediction and conflict judgment in an iterative mode to realize rolling early warning. The method improves track prediction precision and conflict early warning accuracy, reduces false alarm rate of missing alarm, ensures sufficient early warning advance, and is suitable for low-altitude traffic management of complex highland terrains.

Inventors

ZHANG YUN
ZHANG KAIQIN
ZHANG XIAO
FU SHUAI
ZHENG NAN
YANG WENCHEN
ZHAN YANG
HE JIN
WU XIAONAN
CAO QIAN
HAN XIANGZHONG

Assignees

云南省交通规划设计研究院股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. A low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction is characterized by comprising the following steps: Acquiring real-time track data of a non-cooperative target, and carrying out feature extraction and vectorization processing on the real-time track data to obtain a track feature vector sequence; Inputting the track characteristic vector sequence into a pre-trained multi-hidden-layer long-short-term memory cyclic neural network LSTM, and outputting a track prediction vector sequence of a non-cooperative target in a future preset period and a corresponding track state prediction probability distribution through the multi-hidden-layer long-short-term memory cyclic neural network LSTM; based on the track state prediction probability distribution, combining a flight dynamics model of the non-cooperative target, and constructing a three-dimensional probability reachable set for representing the flight influence range of the non-cooperative target; Acquiring a preset flight route and a three-dimensional flight protection area of the cooperative aircraft, performing space geometry intersection analysis on the three-dimensional probability reachable set and the three-dimensional flight protection area, and judging that flight conflict risks exist if the intersection probability is larger than a preset safety probability threshold value; And carrying out real-time rolling update on track data of the non-cooperative target through a preset sliding time window, and carrying out track prediction and conflict judgment based on the updated track data in an iterative manner to realize rolling conflict early warning on the non-cooperative target.
2. The low-altitude non-cooperative target flight conflict pre-warning method based on motion trail prediction according to claim 1, wherein each LSTM unit in the multi-hidden-layer long-short-term memory cyclic neural network LSTM comprises a forgetting gate, an input gate, an output gate three-level gate control structure and a cell state with a dynamic weight self-cyclic mechanism, and the internal operation logic of the LSTM unit is as follows: forgetting the door: Controlling the cell state information retention ratio at the last moment; An input door: Generating candidate cell states Controlling the new information blending proportion; Cell status update: The track information is linearly transferred across time; Output door: Hidden layer output Screening effective information output of cell states; Wherein, the For the trajectory feature vector input at the current time, For the hidden layer state output at the previous moment, For the hidden layer state output at the current time, In order to be the state of the cell at the previous time, For the updated cell state at the current time, Is the candidate cell state at the current time, To forget the gate activation value of the gate, To input the gate activation value of the gate, To output the gate activation value of the gate, For the weight matrix corresponding to the forget gate, For the weight matrix corresponding to the input gate, For the weights corresponding to the candidate cell states, To output the weight matrix corresponding to the gate, For forgetting the bias term corresponding to the gate, For the bias term corresponding to the input gate, Is a bias term corresponding to the candidate cell state, In order to output the corresponding bias term of the gate, Activating a function for Sigmoid for mapping gating values to In the interval of the time period, Is a hyperbolic tangent activation function used to generate candidate cell states and hidden layer outputs, For the multiplication on an element-by-element basis, In order to splice the hidden layer output at the previous moment with the input vector at the current moment.
3. The low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction according to claim 2 is characterized in that a time back propagation BPTT algorithm combining gradient clipping and gradient smoothing mechanisms is adopted for parameter updating in a training process of the multi-hidden-layer long-short-term memory cyclic neural network LSTM, gradient clipping is used for scaling gradients exceeding a preset threshold in equal proportion, gradient smoothing mechanisms are used for relieving gradient attenuation by carrying out weighted average on gradients of adjacent moments, wherein the moment closest to the current moment is recorded as a first moment and the moment farthest from the current moment is recorded as a second moment, and gradient weight coefficients of the first moment are larger than those of the second moment.
4. The low-altitude non-cooperative target flight conflict pre-warning method based on motion trail prediction according to claim 3, wherein the preset threshold of gradient clipping is 1.0, the weight coefficient of gradient smoothing is attenuated according to time, the weight at the first moment is 0.8, and the weight at the second moment is 0.2.
5. The low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction according to claim 1 is characterized in that the three-dimensional probability reachable set is constructed based on a zirono polyhedron, and specifically comprises the steps of setting a dynamic constraint range under a plateau environment for a control variable of a non-cooperative target according to a flight dynamics model of the non-cooperative target, deriving all possible flight state sets of the non-cooperative target in a future period according to trail state prediction probability distribution, and converting the possible flight state sets into the three-dimensional probability reachable set characterized by the zirono polyhedron.
6. The low-altitude non-cooperative target flight conflict pre-warning method based on motion trail prediction according to claim 5, wherein the space geometry intersection analysis comprises the steps of calculating an intersection region of a three-dimensional probability reachable set and a three-dimensional flight protection region of a cooperative aircraft by adopting a three-dimensional space geometry intersection algorithm, and determining conflict probability of the intersection region according to probability distribution of the three-dimensional probability reachable set.
7. The low-altitude non-cooperative target flight conflict pre-warning method based on motion trail prediction according to claim 1 is characterized in that the preset safety probability threshold is dynamically adjusted, and is set according to the current geographic scene of the non-cooperative target, wherein a first threshold is adopted in an open area, and a second threshold lower than the first threshold is adopted in high-risk areas such as mountain areas or canyons.
8. The low-altitude non-cooperative target flight conflict pre-warning method based on motion trail prediction according to claim 7, wherein the first threshold is 8% and the second threshold is 3%.
9. The low-altitude non-cooperative target flight conflict pre-warning method based on motion trail prediction according to claim 1 is characterized in that the window length of the sliding time window is equal to a preset time period in the future, the sliding step length is smaller than the window length, when the sliding time window is used for carrying out rolling update on track data, track data at the earliest moment in the window are removed, the latest acquired track data are integrated, and the latest motion state of a non-cooperative target is always reflected in the data in the window.
10. The low-altitude non-cooperative target flight conflict pre-warning method based on motion trail prediction according to claim 1, wherein the future preset time period is 30 seconds, the window length of a sliding time window is 30 seconds, the sliding step length is 1 second, and the updating frequency of rolling conflict pre-warning is once per second.

Description

Low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction Technical Field The invention relates to the technical field of low-altitude flight safety, in particular to a low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction. Background With the rapid development of low-altitude economy, the number of low-altitude aircraft represented by unmanned aerial vehicles has increased dramatically, wherein non-cooperative targets (such as unauthorized "black-flying" unmanned aerial vehicles) pose a serious security threat to air traffic, known as "airborne mobile reefs". Particularly under complex terrain conditions such as a plateau, the low-altitude environment has the characteristics of serious terrain shielding, changeable meteorological conditions, complex electromagnetic environment and the like, so that the flight conflict early warning of non-cooperative targets faces a great challenge. In the existing low-altitude flight conflict early warning technology, track prediction of non-cooperative targets is a core link. Currently, the mainstream method mostly adopts a Recurrent Neural Network (RNN) to process trajectory time series data of non-cooperative targets. However, RNNs and variants thereof have the inherent disadvantage that when dealing with longer time sequences, the network is very prone to gradient extinction or gradient explosion problems during back propagation, resulting in a network that loses the ability to learn the long-range historical track information, cannot fully mine and exploit the track's context information, and thus results in a poor track prediction accuracy. The insufficient precision of the track prediction serving as a basis for early warning directly influences the reliability of subsequent conflict judgment. Further, the non-cooperative targets have the characteristics of strong maneuverability, random movement intention and high track uncertainty. Traditional deterministic trajectory prediction methods (i.e., outputting only one most probable predicted trajectory) have difficulty accurately characterizing their true flight impact range. If the conflict judgment is performed only according to a single and deterministic track prediction result, the conflict judgment is very easy to cause the missing report of the potential flight conflict (the real risk cannot be identified) or the false alarm (the non-existing risk is reported by mistake), and the requirement of actual safety monitoring cannot be met. In addition, existing pre-warning methods often lack dynamic iterative prediction mechanisms for highly mobile non-cooperative targets. This means that the system cannot dynamically correct and update the track prediction value according to the motion state updated by the target in real time, which results in insufficient early warning advance, and it is difficult to reserve sufficient collision avoidance decision time for air traffic controllers or surrounding cooperative aircrafts. Particularly under the complex topography of the plateau, the problem is more remarkable, and the urgent requirement on the low-altitude traffic safety prevention and control under the scene is difficult to meet. At present, although an attempt of combining track prediction and conflict judgment is made, a complete method which is specially aimed at a high-mobility non-cooperative target, can adapt to a complex low-altitude environment of a plateau and integrates track prediction and conflict early warning is not formed. In the prior art, obvious technical short plates exist in the aspects of track prediction precision, collision early warning accuracy, early warning timeliness and the like, and a more efficient and accurate low-altitude flight collision early warning method is needed. Disclosure of Invention Aiming at the problems existing at present, the invention provides a low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction, which realizes accurate prediction of a non-cooperative target trail through an improved long-short-term memory cyclic neural network, completes scientific judgment of flight conflict by combining a three-dimensional probability reachable set, realizes iteration trail prediction and rolling conflict early warning of a high-mobility non-cooperative target, improves accuracy and advance of low-altitude non-cooperative target flight conflict early warning under complex topography of a plateau, and provides reliable technical support for low-altitude traffic safety prevention and control. The technical scheme of the invention is as follows: A low-altitude non-cooperative target flight conflict early warning method based on motion trail prediction comprises the following steps: Acquiring real-time track data of a non-cooperative target, and carrying out feature extraction and vectorization processing on the real-time track data to o