A Smart Method for Low-Altitude Flight Track Misalignment Repair by Integrating Multi-Source Sensor Data

Abstract

This invention relates to the field of intelligent flight control technology, specifically to an intelligent method for repairing low-altitude flight track deviations by fusing multi-source sensor data. The method includes: correcting jitter in the visual image stream using a motion compensation module; resampling the ADS-B positioning signal using Lagrange interpolation; mapping the radar point cloud to the image coordinate system; generating a spatiotemporally synchronized fused trajectory representation vector; extracting a set of curvature abrupt change points, environmental occlusion factors, and airspace visibility gradients based on this vector; constructing a multidimensional offset feature tensor by combining it with aircraft dynamic constraints; inputting this tensor into a pre-trained spatiotemporal graph neural network; outputting a risk-level repair decision including repair angle, climb rate, and velocity compensation; injecting the repair decision into a flight control simulator for verification; and outputting a final repair command when the trajectory deviation is less than a safety threshold. This method possesses high-precision and closed-loop-executable track repair capabilities and is suitable for flight control systems in complex low-altitude environments.

Description

Technical Field

[0001] This invention relates to the field of intelligent flight control technology, and in particular to an intelligent method for repairing low-altitude flight track deviations by integrating multi-source sensor data. Background Technology

[0002] With the gradual opening of urban low-altitude airspace and the rapid development of drone swarm operations, low-altitude aircraft are widely used in logistics delivery, urban inspection, and environmental monitoring. However, due to complex factors such as building obstruction, electromagnetic interference, and sudden weather changes in the low-altitude flight environment, the actual flight paths of aircraft often deviate to varying degrees, easily leading to path drift, heading errors, and even safety accidents. Therefore, real-time flight path repair, trajectory deviation correction, and risk avoidance have become key research directions in current low-altitude flight path management. With the maturity of multi-sensor fusion technology and graph neural network algorithms, intelligent repair of flight path deviations through multi-source data fusion analysis is gradually becoming a core means to improve flight stability and safety.

[0003] However, existing technologies still have several prominent problems in practical applications: First, some methods rely on only a single data source (such as GPS or image sequences), which has poor robustness under occlusion scenarios or signal failure conditions and makes it difficult to obtain complete flight track status; Second, multi-source data fusion often lacks a unified spatiotemporal alignment mechanism, leading to information redundancy or conflict between data, affecting the accuracy of repair judgment; Third, existing repair strategies mostly rely on static path planning or rule matching, which is difficult to respond to time-varying risk factors such as dynamic obstacles and sudden weather changes, and lacks adaptability to complex environments; Fourth, repair strategies lack a closed-loop verification mechanism before execution, making it impossible to dynamically evaluate their executability and stability in the flight control system. Summary of the Invention

[0004] This invention provides an intelligent method for repairing low-altitude flight track deviations by integrating multi-source sensor data. This method integrates multi-source sensor data, has high robustness, and can realize dynamic reasoning and closed-loop verification, so as to effectively solve the bottleneck problems of current low-altitude track control technology.

[0005] A method for intelligent repair of low-altitude flight track deviation by integrating multi-source sensor data includes the following steps:

[0006] S1: Receives radar point cloud sequence, ADS-B positioning signal, and visual image stream. The motion compensation module performs jitter correction on the visual image stream. The Lagrange interpolation method is used to resample the ADS-B signal in time. The radar point cloud sequence is mapped to a synchronized image coordinate system based on the timestamp. The spatiotemporally synchronized fusion trajectory representation vector is output.

[0007] S2: Based on the fused trajectory representation vector, extract the curvature abrupt change point set, environmental occlusion factor, and airspace visibility gradient, and construct a multidimensional offset feature tensor by combining the aircraft dynamics constraints;

[0008] S3: Input the multidimensional offset feature tensor into a pre-trained spatiotemporal graph neural network, associate the dynamic obstacle motion vector with the spatial topology grid, and output a risk-level repair decision including repair angle, climb rate, and speed compensation amount.

[0009] S4: Inject the risk classification repair decision into the flight control simulator for backpropagation verification. When the trajectory deviation value is less than the safety threshold, generate the final repair command; otherwise, return to S3 to adjust the decision parameters.

[0010] Optionally, S1 includes:

[0011] S11: Receives radar point cloud sequence, ADS-B positioning signal and visual image stream, inputs visual image stream into motion compensation module, calculates inter-frame motion vectors using optical flow method, performs jitter correction on visual image stream, and outputs jitter-corrected visual image stream;

[0012] S12: Perform non-uniform time series analysis on the ADS-B positioning signal, and use the Lagrange interpolation method to resample the position coordinates at equally spaced time points to generate the resampled ADS-B positioning signal.

[0013] S13: Extract the timestamps from the jitter-corrected visual image stream, map the radar point cloud sequence to the image coordinate system based on the same timestamps, and convert the radar three-dimensional point cloud coordinates into two-dimensional image plane coordinates through the perspective transformation matrix to generate timestamp-aligned radar point cloud mapping data.

[0014] Optionally, after generating the timestamp-aligned radar point cloud mapping data, the resampled ADS-B positioning signal and the timestamp-aligned radar point cloud mapping data are spatially registered in the image coordinate system. The position data is fused by Kalman filtering, and the feature point trajectory of the jitter-corrected visual image stream is combined to generate a spatiotemporally synchronized fused trajectory representation vector.

[0015] Optionally, S2 includes:

[0016] S21: Based on the spatiotemporally synchronized fused trajectory representation vector, the trajectory curvature derivative is calculated using a differential geometry algorithm, and trajectory points with curvature change rates exceeding a preset threshold are extracted to generate a set of curvature abrupt change points;

[0017] S22: Perform semantic segmentation on the jitter-corrected visual image stream, extract building and terrain edge features, and simultaneously analyze the reflection intensity distribution of the timestamp-aligned radar point cloud mapping data. Combine edge features and reflection intensity distribution to calculate occlusion coverage and generate environmental occlusion factor.

[0018] S23: Combining the meteorological code in the resampled ADS-B positioning signal with the attenuation coefficient of the radar point cloud sequence, construct a spatial visibility gradient model and output a dynamically updated spatial visibility gradient.

[0019] S24: Stack the curvature mutation point set, environmental occlusion factor, airspace visibility gradient and aircraft dynamics constraint parameters by channel to construct a multidimensional offset feature tensor that includes spatial location, environmental risk and ontological constraints.

[0020] Optionally, S23 includes:

[0021] S231: Analyze the meteorological code in the resampled ADS-B positioning signal and extract visibility interference parameters, including cloud height, precipitation intensity and turbulence index;

[0022] S232: Calculate the attenuation coefficient of the radar point cloud sequence and quantify the signal attenuation by the deviation between the radar echo power and the theoretical propagation model.

[0023] S233: Integrate visibility interference parameters and attenuation coefficients to construct an airspace visibility gradient model. The model is defined as a three-dimensional function with geographic coordinates and timestamps as inputs.

[0024] S234: Utilize a time sliding window mechanism to dynamically update the parameter weights of the spatial visibility gradient model and output the dynamically updated spatial visibility gradient.

[0025] Optionally, the pre-trained spatiotemporal graph neural network includes an input encoding layer, a graph convolutional layer, a temporal convolutional layer, a cross-modal fusion layer, and a decision output layer.

[0026] Optionally, S3 includes:

[0027] S31: Input the multidimensional offset feature tensor into the pre-trained spatiotemporal graph neural network, extract the static structural features of the spatial topological grid through the graph convolutional layer, and capture the temporal evolution pattern of the motion vector of dynamic obstacles through the temporal convolutional layer.

[0028] S32: Based on the temporal evolution pattern of static structural features and dynamic obstacle motion vectors of the spatial topological grid, cross-modal attention fusion is performed to generate an obstacle threat field that incorporates environmental constraints;

[0029] S33: Calculate the trajectory repair path in an obstacle threat field that incorporates environmental constraints, and output the original repair decision vector containing the repair angle, climb rate, and velocity compensation amount;

[0030] S34: Based on the collision probability of the dynamic obstacle motion vector and the risk level mapping table of the spatial topology grid, the original repair decision vector is labeled with risk level and a risk-level repair decision is generated.

[0031] Optionally, S4 includes:

[0032] S41: Inject risk-level repair decisions into the flight control simulator, perform trajectory repair simulation through a six-degree-of-freedom flight dynamics model, and output simulated flight trajectory data;

[0033] S42: The trajectory deviation is calculated based on the Euclidean distance between the simulated flight trajectory data and the reference trajectory, and the flight control parameters are iteratively optimized using a backpropagation verification mechanism;

[0034] S43: Determine whether the trajectory deviation value is less than the safety threshold. If the trajectory deviation value is less than the safety threshold, generate the final repair instruction. Otherwise, return to S3 to adjust the repair angle, climb rate and speed compensation parameters in the risk classification repair decision.

[0035] S44: When the trajectory deviation value is less than the safety threshold for three consecutive iterations, the final repair command is locked and transmitted to the flight control system for execution.

[0036] Optionally, S42 further includes:

[0037] S421: When the trajectory deviation value is less than the safety threshold for the first time, the safety margin compensation module is activated to calculate the dynamic safety margin compensation amount based on historical trajectory deviation data.

[0038] S422: The dynamic safety margin compensation amount is superimposed on the repair angle, climb rate and speed compensation amount parameters in the risk classification repair decision to generate an enhanced repair decision;

[0039] S423: Reinject the enhanced repair decision into the flight control simulator to perform backpropagation verification and update the trajectory deviation value.

[0040] The beneficial effects of this invention are:

[0041] This invention achieves unified processing and spatial registration of three types of sensor information with different sources, sampling frequencies, and accuracy levels by introducing jitter-corrected visual image streams, ADS-B positioning signals resampled using Lagrange interpolation, and radar point cloud mapping data aligned with timestamps. In particular, by using Kalman filtering fusion and combining visual feature point trajectories to supplement positional information, a fused trajectory representation vector with high robustness and high spatiotemporal consistency is generated. This provides a high-quality basic input for subsequent feature extraction and offset repair, effectively avoiding flight path judgment errors and false triggering risks caused by a single data source.

[0042] This invention, through the construction of a multidimensional offset feature tensor, integrates a set of curvature abrupt change points, environmental occlusion factors, airspace visibility gradients, and aircraft dynamics constraint parameters, reflecting the aircraft's current trajectory status, environmental threats, and controllable boundaries in a multidimensional and multi-channel manner. In S3, the pre-trained spatiotemporal graph neural network possesses powerful structural modeling and temporal awareness capabilities. It can extract deep features from the airspace topological grid and dynamic obstacle motion vectors through graph convolutional layers and temporal convolutional layers, respectively. Combined with a cross-modal attention fusion mechanism, it constructs an obstacle threat field, thereby outputting risk-leveled repair decisions. Compared to traditional rule-based judgment or two-dimensional path planning methods, this mechanism can more effectively adapt to complex changes in the flight environment, achieving high-precision repair strategy reasoning for offset trajectories.

[0043] This invention introduces a flight control simulator and a six-degree-of-freedom flight dynamics model to simulate the execution of risk-level remediation decisions. Deviation values ​​are calculated using the Euclidean distance between the trajectory and the baseline track, and control parameters are iteratively optimized based on a backpropagation verification mechanism to ensure the remediation strategy can be safely executed in a real flight control system. A safety margin compensation module is also designed. When the deviation value first falls below a safety threshold, the compensation mechanism is automatically activated. The dynamic safety margin compensation amount is calculated by combining historical flight deviation trends, and the compensation amounts for the remediation angle, climb rate, and velocity are weighted and adjusted to form a reinforced remediation decision. Finally, through multiple rounds of verification and the "three consecutive threshold satisfactions" criterion, the optimal control command is locked and written into the flight control system, significantly improving flight safety and the adaptive robustness of trajectory control. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the S3 process in an embodiment of the present invention. Detailed Implementation

[0047] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0048] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0049] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0050] like Figures 1-2 As shown, a method for intelligent repair of low-altitude flight track deviation by fusing multi-source sensor data includes the following steps:

[0051] S1: Receives radar point cloud sequence, ADS-B positioning signal, and visual image stream. The motion compensation module performs jitter correction on the visual image stream. Lagrange interpolation is used to resample the ADS-B signal. The radar point cloud sequence is mapped to a synchronized image coordinate system based on a timestamp. A spatiotemporally synchronized fusion trajectory representation vector is output, specifically:

[0052] S11, Visual image stream jitter correction: First, receive radar point cloud sequences, ADS-B positioning signals, and visual image streams from the airborne multi-sensor system, where:

[0053] The radar point cloud sequence is acquired by airborne lidar or millimeter-wave radar and contains the three-dimensional spatial coordinates and reflection intensity information of the target reflection point;

[0054] ADS-B positioning signals are acquired in real time through the airborne ADS-B system, including flight parameters such as latitude and longitude, weather codes, altitude, and heading angle.

[0055] The visual image stream is captured by an embedded airborne camera system at a fixed frame rate, resulting in noticeable inter-frame jitter.

[0056] The aforementioned visual image stream is input into the motion compensation module, where optical flow methods (such as the Lucas-Kanade or Farneback methods) are used to calculate the motion vectors between adjacent frames in the image sequence. This allows for frame-by-frame motion compensation processing, achieving dynamic jitter correction. The output jitter-corrected visual image stream will be used for subsequent timestamp synchronization and spatial registration operations.

[0057] S12, ADS-B positioning signal Lagrange resampling: Non-uniform time-series analysis is performed on the received ADS-B positioning signal because its sampling time intervals are unequal, making direct time matching with the image stream difficult. Lagrange interpolation is used to reconstruct the original ADS-B trajectory coordinate sequence, selecting the timestamp of each target. Using the timestamps of visual image frames as interpolation points, second- or third-order interpolation is performed based on the latitude and longitude coordinates of neighboring ADS-B points to generate a set of equally spaced resampled ADS-B positioning signals. This interpolation sequence has a stable time step, which facilitates subsequent joint processing with data from other sensor sources.

[0058] S13, Radar point cloud timestamp mapping and perspective transformation: Extract the timestamp of each frame of the jitter-corrected visual image stream, filter the radar point cloud sequence at the corresponding time point, perform timestamp-based point cloud filtering, and then construct the perspective transformation matrix. This matrix is ​​generated based on the intrinsic and extrinsic parameter calibration results of the airborne radar, completing the radar's three-dimensional coordinate points ( (to the image's two-dimensional pixel coordinate system) Conversion of )

[0059] ;

[0060] in, The image intrinsic parameter matrix, This represents the rotation and translation matrix from radar to camera. The transformation result is radar point cloud mapping data for each frame of image, where each point contains its original 3D position and corresponding image position coordinates. And retain attributes such as the reflection intensity value of the radar point.

[0061] After image jitter correction and radar point cloud mapping are completed, the resampled ADS-B positioning signal and the timestamp-aligned radar point cloud mapping data are registered in a unified image coordinate system. A Kalman filter is used for multi-source position estimation fusion, where the predicted state comes from the ADS-B positioning signal, and the observed state is constructed from the radar point cloud data. The filter iteratively updates the position state.

[0062] ;

[0063] in, For the current estimated position, For observations (radar mapping points), For Kalman gain, This is the observation matrix.

[0064] Meanwhile, sparse feature point trajectories (such as corner points and edge points) are extracted from the jitter-corrected visual image stream to further assist in localization information fusion and trajectory smoothing.

[0065] Finally, the fused ADS-B trajectory points, radar reflection points, and image feature trajectories are arranged according to a unified timestamp and spatial coordinate system to construct a spatiotemporally synchronized fused trajectory representation vector with the following structure:

[0066] ;

[0067] in, This indicates the location of the filtered ADS-B points. This represents the reprojection position of the point cloud in image coordinates. It is a set of image feature point trajectories.

[0068] S2: Based on the fused trajectory representation vector, extract the curvature abrupt change point set, environmental occlusion factor, and airspace visibility gradient. Combine this with aircraft dynamic constraints to construct a multidimensional offset feature tensor, specifically:

[0069] S21, Curvature mutation point extraction: First, the trajectory sequence corresponding to the ADS-B positioning signal filtering result in the spatiotemporally synchronized fused trajectory representation vector is differentially processed. Based on the differential geometry algorithm, the curvature derivative between continuous trajectory points is calculated to capture the local change trend of the track in three-dimensional space.

[0070] Specifically, let the resampled ADS-B trajectory sequence be: ;

[0071] Calculate the spatial curvature at each point: ;

[0072] in, For velocity vectors, It is the acceleration vector;

[0073] Taking the derivative further, we obtain the curvature derivative: ;

[0074] when Exceeding the set threshold for the rate of change of curvature (e.g., 0.5m) If a point is found to be a curvature inflection point, then that point is determined to be a curvature inflection point. This ultimately forms a set of curvature inflection points used to identify key turning points where flight deviations may occur.

[0075] S22, Environmental occlusion factor calculation: Perform semantic segmentation algorithms, such as DeepLabV3+ or SegNet, on the jitter-corrected visual image stream to identify building outlines, terrain undulation edges, and obstacle boundaries in the image, and obtain an image-level occlusion region annotation map.

[0076] Simultaneously, the reflection intensity values ​​of each point are extracted from the timestamp-aligned radar point cloud mapping data to construct a three-dimensional reflectivity distribution map. This intensity value is typically related to the object's material, density, and degree of occlusion.

[0077] Image semantic edge features are fused with radar reflectivity maps, and a pixel-level joint masking calculation method is used to calculate the occlusion coverage rate in each frame of the image, i.e., the proportion of the occluded area to the entire image. This coverage rate is expressed numerically as the environmental occlusion factor, ranging from 0 to 1, with the closer to 1 indicating poorer visibility and more severe occlusion.

[0078] S23, Airspace Visibility Gradient Construction: Combining the meteorological code in the resampled ADS-B positioning signal with the attenuation coefficient of the radar point cloud sequence, an airspace visibility gradient model is constructed, outputting a dynamically updated airspace visibility gradient. The processing flow includes:

[0079] S231: Extract meteorological coding information from the resampled ADS-B positioning signal and parse out key visibility interference parameters, including cloud height, precipitation intensity, and turbulence index.

[0080] S232: Calculate the attenuation coefficient for each point in the radar point cloud sequence, according to the following formula:

[0081] ;

[0082] in, For radar power under ideal propagation, To measure the echo power, This refers to the degree of signal attenuation.

[0083] S233: By integrating the above meteorological parameters and attenuation coefficients, a three-dimensional function-defined airspace visibility gradient model is constructed as follows:

[0084] Cloud height, precipitation intensity, turbulence index, attenuation coefficient;

[0085] Among them, the function It is a linear weighted model, and its parameter weights are obtained through training with historical flight big data, reflecting the actual impact weight of each factor on visibility.

[0086] S234: Introducing a time sliding window mechanism (such as updating every 10 seconds) to dynamically adjust the model parameter weights, so that the visibility gradient model can be updated in real time with changes in flight path and weather, and finally outputs a dynamically updated airspace visibility gradient.

[0087] S24, Construction of Multidimensional Offset Feature Tensor: The various features extracted in the above sub-steps are stacked and integrated into a unified structure. The construction method is as follows:

[0088] Channel 1: Binary label matrix of the curvature mutation point set (1 represents mutation point, 0 represents non-mutation);

[0089] Channel 2: Shading heatmap formed by mapping environmental shading factors over time;

[0090] Channel 3: Spatial visibility gradient according to ( The visibility distribution tensor formed by sampling;

[0091] Channel 4: Aircraft dynamics constraint parameters from the flight control system, such as maximum rate of climb, maximum turning radius, and maximum speed, will be statically filled and encoded. The features from each channel will be aligned along the time axis and stacked along the depth dimension (channel dimension) to generate the final multidimensional offset feature tensor.

[0092] ;

[0093] in, For the trajectory length, For spatial dimension resolution, Given the number of channels (≥4), this tensor will serve as the core input to the pre-trained spatiotemporal graph neural network in S3.

[0094] S3: Input the multi-dimensional offset feature tensor into a pre-trained spatiotemporal graph neural network, associate the dynamic obstacle motion vector with the spatial topology grid, and output a risk-level repair decision including repair angle, climb rate, and velocity compensation amount. Specifically:

[0095] S31, Graph Neural Network Input and Temporal Feature Extraction: The multidimensional offset feature tensor output from S2 is used as input to load a pre-trained spatiotemporal graph neural network model that has completed offline training. This tensor includes four channels (curvature abrupt change point set, environmental occlusion factor, airspace visibility gradient, and aircraft dynamics constraint parameters), where:

[0096] The set of curvature abrupt change points is used to describe changes in the trajectory curve;

[0097] Environmental shading factors and airspace visibility gradient characterize environmental risks;

[0098] Aircraft dynamics constraint parameters are used to physically limit repair actions.

[0099] The input data stream first enters the model's input encoding layer, where the tensor is decoupled into two sub-streams:

[0100] One is a spatial topological grid, whose structure originates from the spatial connectivity formed by curvature abrupt changes and environmental factors;

[0101] The other is the dynamic obstacle motion vector, which represents the relative position and velocity of various nearby flying objects or risk targets appearing within the trajectory range.

[0102] Subsequently, the spatial topology grid data stream enters the graph convolutional layer, where the local structural features (static connections) of each node are extracted using its adjacency matrix; the dynamic obstacle motion vector data stream enters the temporal convolutional layer, where the node attributes are extracted in the temporal dimension through one-dimensional convolution operations, and the corresponding temporal evolution pattern is output.

[0103] The pre-trained spatiotemporal graph neural network architecture includes the following processing layers:

[0104] Input encoding layer: Decouples the multidimensional offset feature tensor into two sub-streams: dynamic obstacle motion vector and spatial topological grid.

[0105] Graph convolutional layer: Utilizes the adjacency matrix of the spatial topological grid to extract the structural relationships at each location based on the GCN (Graph Convolutional Network) structure.

[0106] Temporal convolutional layer: Based on one-dimensional convolution operations, it extracts the temporal evolution features of the motion vectors of dynamic obstacles and processes the motion trend in time-series trajectories.

[0107] Cross-modal fusion layer: Uses attention gating mechanism to fuse spatial structure and temporal features, outputting a semantically enhanced obstacle threat field.

[0108] The decision output layer uses a graph path search algorithm and a risk classification model in the obstacle threat field to output the final risk classification and remediation decision.

[0109] S32, Cross-modal Attention Fusion: The feature tensors output by the graph convolutional layer and the temporal convolutional layer are input into the cross-modal fusion layer, and the information between different modalities is weighted and integrated through an attention gating mechanism. The core calculation formula of this mechanism is as follows:

[0110] ;

[0111] in, For query vectors (such as static structural features), As the key vector, For value vectors (such as dynamic motion features), This is the dimension normalization factor.

[0112] This method yields a combined pattern with the highest semantic coupling between spatial structure and temporal evolution, outputting an obstacle threat field that incorporates environmental information constraints.

[0113] The obstacle threat field is a three-dimensional tensor that provides a set of scores at each trajectory point location, including spatial risk level and directional correction reference.

[0114] S33, Generation of Original Repair Decision Vector: Based on the obstacle threat field, the decision output layer inside the model calls the reinforcement graph search algorithm to evaluate the optimal path combination for trajectory repair at each high-risk point, calculate the optimal trajectory repair action for each node, and generate the original repair decision vector consisting of the following three quantities:

[0115] Repair angle: The angle that should be adjusted relative to the current heading;

[0116] Rate of ascent: The required rate of ascent / descent in the vertical direction;

[0117] Speed ​​compensation amount: The speed correction amount required to accelerate or decelerate during the repair process.

[0118] The repair decision vector is output in the form of a time series, providing a quantitative basis for subsequent risk grading.

[0119] S34, Risk-based remediation decision generation: The original remediation decision vector needs to be labeled with a safety level in conjunction with an external risk assessment mechanism. The specific method is as follows:

[0120] Extract the dynamic obstacle motion vector that coincides with the current trajectory from the multidimensional offset feature tensor;

[0121] Collision probability calculations are performed on it based on a prediction model, taking into account its speed, direction, and proximity rate;

[0122] The location of the trajectory points in the airspace topology grid is compared with a predefined risk level mapping table to generate a static risk level (e.g., low, medium, high).

[0123] By combining the two risk factors mentioned above, and based on the comprehensive threat coefficient... definition:

[0124] ;

[0125] in, Let be the collision probability. It is a static level. The fusion weights are obtained by tuning the parameters of the historical model.

[0126] Finally, the system outputs risk-based remediation decisions with risk labels (such as "low-risk remediation", "medium-risk remediation", and "high-risk avoidance").

[0127] S4: Inject the risk-level repair decision into the flight control simulator for backpropagation verification. When the trajectory deviation is less than the safety threshold, generate the final repair command; otherwise, return to S3 to adjust the decision parameters. Specifically:

[0128] S41, Simulation of Repair Decision Execution: The risk-level repair decision output from S3 is input into the flight control simulation module. This module is built on a high-fidelity flight control simulator and integrates a six-degree-of-freedom flight dynamics model conforming to aerospace dynamics standards. The model considers the following six state variables:

[0129] Spatial location coordinates;

[0130] Attitude angles (pitch, roll, yaw);

[0131] Velocity vector (linear velocity and angular velocity);

[0132] Dynamic parameters (rate of climb and speed compensation).

[0133] Control input (correction of heading adjustment values ​​such as angle);

[0134] External disturbances (environmental disturbances such as wind shear and airflow);

[0135] The simulator dynamically solves and runs the flight path repair process, outputting complete simulated flight trajectory data, which includes flight path point sequences, velocity change curves, attitude evolution sequences, and other content.

[0136] S42, Trajectory Deviation Calculation and Backpropagation Verification: Based on the comparison between simulated flight trajectory data and a preset baseline trajectory, the deviation is calculated using point-by-point Euclidean distance difference.

[0137] ;

[0138] in, The location point in the simulated trajectory. As the reference track position point, For a moment The trajectory deviation value. Accumulate the trajectory deviation values ​​across all time steps to define the overall deviation index:

[0139] ;

[0140] like If the deviation exceeds the set safety threshold (e.g., 2.5m), it means that the current risk-level repair decision cannot effectively control the deviation range in the dynamic model. In this case, the backpropagation verification mechanism is executed according to the error gradient to update the flight control parameters in the simulator (e.g., speed compensation ratio, angular velocity limit, pitch adjustment parameters, etc.) to enhance the dynamic adaptability to the original repair action.

[0141] Following step S42, after the safety margin compensation mechanism triggers and reinforces the repair decision generation, a key dynamic safety enhancement mechanism is also included to further reduce the cumulative trajectory risk value, as detailed below:

[0142] Activate the safety margin compensation module: When the detected trajectory deviation value is first less than the safety threshold (i.e. Threshold, or trigger safety margin compensation module, is used to enhance the robustness of the current repair action. This module extracts deviation trend curves from similar scenarios in the historical database, constructs a statistical model, and calculates the dynamic safety margin compensation amount for the target trajectory point. Its definition is:

[0143] ;

[0144] in, For the first time in history The deviation value under similar conditions This is an empirical adjustment coefficient.

[0145] S422, Generate Enhanced Repair Decision: Add the above dynamic safety margin compensation amounts to the current repair parameters respectively:

[0146] Repair angle ;

[0147] Climb rate ;

[0148] Speed ​​compensation amount ;

[0149] A set of enhanced repair decision parameter vectors is obtained to ensure increased margin space within the control boundary and reduced fault tolerance risk.

[0150] S423, Re-injecting the flight control simulator for verification: The enhanced repair decision is re-inputted into the flight control simulator, a complete trajectory simulation and deviation calculation are performed to form enhanced repaired simulated flight trajectory data, and the backpropagation optimization in step S42 is performed again until the final trajectory deviation control requirements are met.

[0151] S43, Trajectory Deviation Threshold Judgment and Feedback Loop: After each simulation calculation, it is determined whether the current trajectory deviation value is less than the safety threshold. If it is satisfied, the current repair action is retained and the next judgment cycle is entered; if it is not satisfied, the error is fed back to S3, the key parameters (repair angle, climb rate, speed compensation amount) in the risk classification repair decision are adjusted, the repair strategy is regenerated and a new round of simulation verification is entered, forming a closed-loop control process.

[0152] This feedback loop executes at most three times to ensure response efficiency and command stability;

[0153] S44, Final Repair Command Locking and Control Writing: When the trajectory deviation value is less than the safety threshold for three consecutive iterations, it is determined that the current repair action has flight control feasibility and safety margin, and the current repair parameters are immediately locked to generate the final repair command;

[0154] ;

[0155] in, For the final repair angle, For the final climb rate, This is the final speed compensation amount.

[0156] The final repair command is written into the flight control system (such as the autopilot / APU system) via the control bus to ensure that the track repair action is executed accurately according to the preset parameters during the actual flight phase.

[0157] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0158] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent repair of low-altitude flight track deviation by integrating multi-source sensor data, characterized in that, Includes the following steps: S1: Receives radar point cloud sequence, ADS-B positioning signal, and visual image stream. The motion compensation module performs jitter correction on the visual image stream. The Lagrange interpolation method is used to resample the ADS-B signal in time. The radar point cloud sequence is mapped to a synchronized image coordinate system based on the timestamp. The spatiotemporally synchronized fusion trajectory representation vector is output. S2: Based on the fused trajectory representation vector, extract the curvature abrupt change point set, environmental occlusion factor, and airspace visibility gradient, and construct a multidimensional offset feature tensor by combining the aircraft dynamics constraints; Specifically, it includes: S21: Based on the spatiotemporally synchronized fused trajectory representation vector, the trajectory curvature derivative is calculated using a differential geometry algorithm, and trajectory points with curvature change rates exceeding a preset threshold are extracted to generate a set of curvature abrupt change points; S22: Perform semantic segmentation on the jitter-corrected visual image stream, extract building and terrain edge features, and simultaneously analyze the reflection intensity distribution of the timestamp-aligned radar point cloud mapping data. Combine edge features and reflection intensity distribution to calculate occlusion coverage and generate environmental occlusion factor. S23: Combining the meteorological code in the resampled ADS-B positioning signal with the attenuation coefficient of the radar point cloud sequence, construct a spatial visibility gradient model and output a dynamically updated spatial visibility gradient. S24: Stack the curvature mutation point set, environmental occlusion factor, airspace visibility gradient and aircraft dynamics constraint parameters by channel to construct a multidimensional offset feature tensor that includes spatial location, environmental risk and ontological constraints. S3: Input the multidimensional offset feature tensor into a pre-trained spatiotemporal graph neural network, associate the dynamic obstacle motion vector with the spatial topology grid, and output a risk-level repair decision including repair angle, climb rate, and speed compensation amount. S4: Inject the risk classification repair decision into the flight control simulator for backpropagation verification. When the trajectory deviation value is less than the safety threshold, generate the final repair command; otherwise, return to S3 to adjust the decision parameters.

2. The intelligent repair method for low-altitude flight track deviation fusion based on multi-source sensor data according to claim 1, characterized in that, S1 includes: S11: Receives radar point cloud sequence, ADS-B positioning signal and visual image stream, inputs visual image stream into motion compensation module, calculates inter-frame motion vectors using optical flow method, performs jitter correction on visual image stream, and outputs jitter-corrected visual image stream; S12: Perform non-uniform time series analysis on the ADS-B positioning signal, and use the Lagrange interpolation method to resample the position coordinates at equally spaced time points to generate the resampled ADS-B positioning signal. S13: Extract the timestamps from the jitter-corrected visual image stream, map the radar point cloud sequence to the image coordinate system based on the same timestamps, and convert the radar three-dimensional point cloud coordinates into two-dimensional image plane coordinates through the perspective transformation matrix to generate timestamp-aligned radar point cloud mapping data.

3. The intelligent repair method for low-altitude flight track deviation fusion based on multi-source sensor data according to claim 2, characterized in that, After generating the timestamp-aligned radar point cloud mapping data, the resampled ADS-B positioning signal and the timestamp-aligned radar point cloud mapping data are spatially registered in the image coordinate system. The position data is fused by Kalman filtering, and the feature point trajectory of the jitter-corrected visual image stream is combined to generate a spatiotemporally synchronized fused trajectory representation vector.

4. The intelligent repair method for low-altitude flight track deviation fusion based on multi-source sensor data according to claim 1, characterized in that, S23 includes: S231: Analyze the meteorological code in the resampled ADS-B positioning signal and extract visibility interference parameters, including cloud height, precipitation intensity and turbulence index; S232: Calculate the attenuation coefficient of the radar point cloud sequence and quantify the signal attenuation by the deviation between the radar echo power and the theoretical propagation model. S233: Integrate visibility interference parameters and attenuation coefficients to construct an airspace visibility gradient model. The model is defined as a three-dimensional function with geographic coordinates and timestamps as inputs. S234: Utilize a time sliding window mechanism to dynamically update the parameter weights of the spatial visibility gradient model and output the dynamically updated spatial visibility gradient.

5. The intelligent method for repairing low-altitude flight track deviation by fusing multi-source sensor data according to claim 4, characterized in that, The pre-trained spatiotemporal graph neural network includes an input encoding layer, a graph convolutional layer, a temporal convolutional layer, a cross-modal fusion layer, and a decision output layer.

6. The intelligent repair method for low-altitude flight track deviation fusion based on multi-source sensor data according to claim 5, characterized in that, S3 includes: S31: Input the multidimensional offset feature tensor into the pre-trained spatiotemporal graph neural network, extract the static structural features of the spatial topological grid through the graph convolutional layer, and capture the temporal evolution pattern of the motion vector of dynamic obstacles through the temporal convolutional layer. S32: Based on the temporal evolution pattern of static structural features and dynamic obstacle motion vectors of the spatial topological grid, cross-modal attention fusion is performed to generate an obstacle threat field that incorporates environmental constraints; S33: Calculate the trajectory repair path in an obstacle threat field that incorporates environmental constraints, and output the original repair decision vector containing the repair angle, climb rate, and velocity compensation amount; S34: Based on the collision probability of the dynamic obstacle motion vector and the risk level mapping table of the spatial topology grid, the original repair decision vector is labeled with risk level and a risk-level repair decision is generated.

7. The intelligent repair method for low-altitude flight track deviation fusion based on multi-source sensor data according to claim 6, characterized in that, S4 includes: S41: Inject risk-level repair decisions into the flight control simulator, perform trajectory repair simulation through a six-degree-of-freedom flight dynamics model, and output simulated flight trajectory data; S42: The trajectory deviation is calculated based on the Euclidean distance between the simulated flight trajectory data and the reference trajectory, and the flight control parameters are iteratively optimized using a backpropagation verification mechanism; S43: Determine whether the trajectory deviation value is less than the safety threshold. If the trajectory deviation value is less than the safety threshold, generate the final repair instruction. Otherwise, return to S3 to adjust the repair angle, climb rate and speed compensation parameters in the risk classification repair decision. S44: When the trajectory deviation value is less than the safety threshold for three consecutive iterations, the final repair command is locked and transmitted to the flight control system for execution.

8. The intelligent method for repairing low-altitude flight track deviation by fusing multi-source sensor data according to claim 7, characterized in that, S42 further includes: S421: When the trajectory deviation value is less than the safety threshold for the first time, the safety margin compensation module is activated to calculate the dynamic safety margin compensation amount based on historical trajectory deviation data. S422: The dynamic safety margin compensation amount is superimposed on the repair angle, climb rate and speed compensation amount parameters in the risk classification repair decision to generate an enhanced repair decision; S423: Reinject the enhanced repair decision into the flight control simulator to perform backpropagation verification and update the trajectory deviation value.

Images