Airport Light Generation Method Based on AIP Document and Image Data

Abstract

The present invention belongs to the technical field of airport lighting layout, and specifically relates to a method for generating airport lighting based on AIP documents and image data. It aims to solve the problem that the traditional lighting system has fixed settings and manual planning, which leads to layout and attribute configuration errors and makes it difficult to meet airport lighting needs. The present invention extracts runway / taxiway geometric attributes and spatial relationships by parsing AIP text, aligns satellite images and GIS data to obtain accurate image data; standardizes, features match, and fuses and superimposes the two types of data to generate runway and taxiway layout maps; based on the layout map and rule base, grid sampling is used in combination with adaptive processing to obtain airport lighting positions; cognitive rules and multi-objective models are constructed, the target attributes are solved using the MOPSO algorithm, and the actual attributes are controlled in conjunction with the enhanced Hi‑IDRL algorithm. The present invention meets the diverse lighting simulation needs of airports through dynamic optimization of layout and attributes.

Description

Technical Field

[0001] The present invention belongs to the technical field of airport lighting layout, and in particular relates to an airport lighting generation method based on AIP documents and image data. Background Art

[0002] Airport lighting is the cornerstone of aviation safety and the key to operational efficiency. It provides irreplaceable visual reference for aircraft takeoff and landing and ground movement in complex environments. At night or in low visibility, it provides pilots with critical visual guidance to ensure that the aircraft is accurately aligned with the runway, lands and takes off safely, and prevents runway incursions, runway excursions, or taxiway deviations. In addition, in the field of flight simulation training, the high-fidelity airport lighting system creates a zero-risk training environment for pilots through the simulator visual system, helping them to repeatedly hone their core operating skills, procedural application capabilities, and special situation handling capabilities in complex weather conditions and night flights.

[0003] However, the current airport lighting system still faces many challenges in modeling and simulating through simulator vision systems: on the one hand, during the operation of the simulated lighting system, the traditional simulated lighting system adopts fixed attribute parameter configuration, which makes it difficult to accurately reproduce the complex and changeable real environmental conditions and differentiated operation scenarios. For example, when simulating different visibility (such as heavy fog, heavy rain), day and night time transitions and flight traffic fluctuations, if the light color, flashing frequency and brightness are kept uniformly set, it will not only cause pilots to form visual judgment habits that are inconsistent with real operations in simulation training (such as the light brightness is not adapted and enhanced in low visibility, making it difficult to train the ability to rely on lights to judge taxiing trajectories), but also fail to simulate the actual lighting requirements of ground operations. On the other hand, in the initial modeling stage, the lighting layout modeling of existing simulation scenes mainly relies on manual input of massive light position parameters, and it is necessary to replicate the light position attributes of the real airport one by one (such as the difference in the distance between the light positions of the taxiway and the runway, and the light colors corresponding to different distances from the runway end). This not only consumes a lot of modeling time and cost, but also due to the complexity of the real airport lighting system, it is easy to have problems such as the simulated light position layout not matching the actual one and the attribute parameter configuration errors. As a result, the simulation scene cannot accurately reflect the lighting logic of the real airport, which weakens the authenticity and effectiveness of flight training and makes it difficult to meet the simulation requirements of the simulator for the lighting system.

[0004] Therefore, it is necessary to provide an airport lighting generation method based on AIP documents and image data, and dynamically adjust the light position layout and light position attributes to meet the needs of simulated airport lighting. Summary of the Invention

[0005] To address the aforementioned issues in the prior art, namely, that traditional lighting systems employ fixed attribute settings, and that lighting layouts rely primarily on manual planning and deployment, which are prone to improper layouts and attribute configuration errors, making it difficult to meet the requirements of simulated airport lighting. In a first aspect, the present invention proposes an airport lighting generation method based on AIP documents and image data, comprising the following steps:

[0006] Obtaining the AIP text and performing semantic parsing to extract the geometric attributes and spatial relationships of the runway / taxiway; the geometric attributes and spatial relationships of the runway / taxiway include the size and position information of the runway centerline, runway edgeline, taxiway centerline, and taxiway edgeline;

[0007] Obtaining satellite images and GIS data, extracting control points from each, calculating a covariance matrix of the control points, and using affine transformation and thin plate spline transformation based on the covariance matrix to achieve coordinate alignment of the satellite image and GIS data to obtain aligned image data;

[0008] Generate a runway / taxiway layout diagram by performing data format standardization, feature extraction and matching, data fusion, and overlay on the geometric attributes and spatial relationships of the runway / taxiway and the aligned image data;

[0009] Based on the layout diagram and the rules of the rule base, a spatiotemporal pyramid layering strategy is used to generate a multi-resolution candidate point set on both sides of the runway / taxiway centerline; the candidate point set is adaptively processed to obtain airport light positions; the attributes of the airport light positions include: light type, color, illumination range, and light number;

[0010] Cognitive rules and multi-objective models are constructed, and the MOPSO algorithm is used to solve the target attributes of the airport light positions. Based on the target attributes and actual attributes of the airport light positions, the Hi-IDRL algorithm is used to control the attributes of the airport light positions.

[0011] In some preferred embodiments, the candidate points are adaptively processed as follows:

[0012] Generate a feature vector based on the fusion of the geometric features of the layout diagram, the rules of the rule base, and the airport operation parameters; the airport operation parameters include three-dimensional laser point cloud elevation information, meteorological influencing factors, and flight flow parameters;

[0013] Input the feature vector into the GNN-Transformer hybrid model to calculate the invalid probability score of the candidate point;

[0014] Inputting the feature vector into a multi-dimensional risk assessment model to calculate the technical risk, safety risk, economic risk and total risk value of the candidate point;

[0015] The candidate point is determined to be an invalid point based on the invalid probability score and the total risk value of the candidate point, and a reinforcement learning algorithm is used to select a processing strategy to process the invalid point; the processing strategy includes deletion and position adjustment.

[0016] Beneficial effects of the present invention:

[0017] The geometric attributes and spatial relationships of runways / taxiways extracted from AIP text are combined with the image data after satellite imagery and GIS data alignment to generate accurate and reliable runway and taxiway layout maps;

[0018] Based on the layout diagram and rule base, a spatiotemporal pyramid layering strategy is used to generate a multi-resolution candidate point set on both sides of the runway / taxiway centerline. This candidate point set is then adaptively processed to obtain an optimized layout of airport lighting positions. This not only meets basic safety requirements, but also effectively avoids layout blind spots and redundant points caused by traditional methods, making the candidate point distribution more consistent with the actual operational needs of the airport.

[0019] The collaborative work of the MOPSO algorithm for solving target attributes and the Hi-IDRL algorithm for controlling airport light location attributes can achieve efficient, accurate, and intelligent optimization of simulated airport light location attributes. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0021] Figure 1 It is a flowchart of the steps of the airport lighting generation method based on AIP documents and image data of the present invention.

[0022] Figure 2 The present invention is a flowchart of the steps of obtaining aligned image data in the method for generating airport lights based on AIP documents and image data.

[0023] Figure 3 The present invention is a flowchart of the steps of generating a runway / taxiway layout diagram of an airport lighting generation method based on AIP documents and image data. DETAILED DESCRIPTION

[0024] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the relevant invention are shown in the accompanying drawings.

[0025] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0026] In order to more clearly explain the airport light generation method based on AIP document and image data of the present invention, the following is combined with Figure 1 Each step in the embodiment of the present invention is described in detail.

[0027] The present invention proposes an airport lighting generation method based on AIP documents and image data, see Figure 1 The method is deployed to the simulator to realize the simulator's simulation requirements for the lighting system, including the following steps:

[0028] Obtaining the AIP text and performing semantic parsing to extract the geometric attributes and spatial relationships of the runway / taxiway; the geometric attributes and spatial relationships of the runway / taxiway include the dimensions and location information (coordinate points) of the runway centerline, runway edgeline, taxiway centerline, and taxiway edgeline;

[0029] In this embodiment, the AIP text (Aeronautical Information Publication) is an aeronautical information publication specified by the International Civil Aviation Organization (ICAO), which contains standardized data on key facilities such as airport runways and taxiways;

[0030] Semantic parsing can use the existing BERT model to perform semantic parsing on AIP text;

[0031] Obtain satellite images and GIS data, extract control points from each, calculate the covariance matrix of the control points, and use the covariance matrix in combination with affine transformation and thin plate spline transformation to achieve coordinate alignment of the satellite image and GIS data, thereby obtaining aligned image data.

[0032] In this embodiment, the aligned image data is obtained by:

[0033] Step 1: Obtain satellite images and GIS data. Specifically, obtain satellite remote sensing images from a satellite data service provider platform. The satellite remote sensing images are selected based on geographic location, resolution, and shooting time parameters, and the resolution must be better than 0.5 meters. Obtain GIS data from a geographic information public service platform. The GIS data includes vector layers and terrain data, and its coordinate system must be consistent with the actual geographic coordinate system.

[0034] Step 2: Extract control points from the satellite image and GIS data, use SIFT or SURF algorithm to detect feature points on the satellite image, manually select feature points with clear geographical identification as control points, and mark their pixel coordinates ( xs , ys ); Select the geographic coordinate points corresponding to the satellite image control points in the GIS vector data ( Xg , Yg ), if the GIS data coordinate system is inconsistent, use the coordinate conversion tool (coordinate conversion function in the GDAL library) to convert it to the same coordinate system as the satellite image;

[0035] Step 3, calculate the covariance matrix of the control points, and form coordinate pairs with the pixel coordinates of the satellite image control points and the geographic coordinates of the GIS data control points, that is, each set of coordinate pairs ( xs , ys , Xg , Yg ) as the data matrix D A row of multiple coordinate pairs is arranged in order to form a data matrix D , the covariance matrix is ​​calculated according to the following formula:

[0036] ;

[0037] in, for D No. row vector, for D The mean vector of ;

[0038] Step 4: Based on the covariance matrix, affine transformation and thin plate spline transformation are used to achieve coordinate alignment of satellite images and GIS data. Specifically, the covariance matrix and the coordinate pairs of the control points are substituted into the least squares model. The least squares method finds the optimal function matching of the data by minimizing the sum of squares of the errors. In the affine transformation parameter solution, the pixel coordinates of the satellite image control points are used as input, and the geographic coordinates of the GIS data control points are used as the desired output to construct an error function:

[0039] ;

[0040] in, is the number of control points, For sum index, from 1 to , Respectively The actual coordinate values ​​of the horizontal and vertical coordinates of the control points, Respectively The original coordinate values ​​of the horizontal and vertical coordinates of the control points, are the scaling and rotation parameters, are translation parameters, respectively expressed in axis, The translation distance in the axis direction;

[0041] By respectively 、 Find the partial derivative and set it equal to 0 to get a linear equation system. Solve the equation system and get the affine transformation formula. 、 , the affine transformation formula is:

[0042] ;

[0043] Where, 、 are the horizontal and vertical coordinate values ​​of the transformed point respectively; 、 Respectively represent the horizontal and vertical coordinate values ​​of the original point; 1 is to satisfy the homogeneous coordinate representation;

[0044] Apply the obtained parameters to the satellite image, perform affine transformation calculation on the coordinates of each pixel in the image, and realize preliminary geometric correction and coordinate adjustment of the satellite image;

[0045] Selected Based on the control points, the interpolation function of the thin plate spline transformation is constructed. The thin plate spline transformation is based on the bending theory of the elastic thin plate, and its interpolation function form is:

[0046] ;

[0047] ;

[0048] ;

[0049] in, For the The pixel coordinates of the control points in the satellite image, is the weight coefficient, are the polynomial coefficients, For distance function, The coordinates on the plane are Point to control points distance;

[0050] Substitute the coordinates of the control points into the interpolation function, construct a linear equation system, and solve the weight coefficients and polynomial coefficients For any point in the satellite image, the interpolation function is used to calculate its corresponding coordinates in the GIS coordinate system. , perform nonlinear geometric distortion correction on satellite images, so that satellite images and GIS data can be accurately aligned in details;

[0051] After completing the affine transformation and thin plate spline transformation, the coordinates of the satellite image have changed. Bilinear interpolation or bicubic interpolation is used for image resampling. Taking bilinear interpolation as an example, for each pixel position in the new image after transformation, , first determine its corresponding position in the original satellite image , which is usually not an integer coordinate, find The four nearest integer coordinate pixels around , according to the grayscale value or spectral value of these four pixels, the ( x , y ) position, thereby obtaining the grayscale value or spectral value of the new image The pixel value of the position is obtained. All pixel positions of the new image are traversed according to this method to generate satellite image data that is precisely aligned with the GIS data coordinates. The entire coordinate alignment process is completed to generate image data that is aligned with the GIS data coordinates.

[0052] Generate a runway / taxiway layout diagram by performing data format standardization, feature extraction and matching, and data fusion and overlay on the geometric attributes and spatial relationships of the runway / taxiway and the aligned image data;

[0053] In this embodiment, data format standardization processing specifically includes: performing a structured conversion on the geometric attributes and spatial relationships of the runway / taxiway, i.e., text data; then, based on the structured coordinate information, using a GIS library, converting the line information (runway centerline, runway edgeline, taxiway centerline, taxiway edgeline) in the text data into geometric objects; and using a GIS coordinate conversion tool (such as the gdaltransform command in GDAL) to convert their coordinates into a target coordinate system to ensure consistency in spatial position between the two; converting the aligned image data from raster image data to vector data (for example, an image segmentation algorithm can be used to segment the runway and taxiway areas from the image, and then using a vectorization tool to convert the edge contours in the image into vector lines); and using the conversion parameters obtained during the geometric distortion correction and coordinate alignment process to generate image data aligned with the GIS data coordinates, converting its coordinates into the target coordinate system, thereby converting them into the same coordinate system. This ensures consistency in format and spatial position between the two types of data, reduces errors caused by differences in data format and coordinate system, and lays the foundation for accurate overlay.

[0054] Feature extraction and matching, specifically, extracting key features from structured and geometricized text data. This can be done using the aforementioned methods for extracting geometric attributes and spatial relationships. These features include type features (runway centerline, runway edgeline, taxiway centerline, taxiway edgeline), dimension information (calculated runway / taxiway length, width, curvature, and azimuth), location information (coordinate points, such as line endpoints, turning points, and edge feature coordinates), and functional attributes (whether it is used for precision approach).

[0055] Extract features from vectorized image data using computer vision algorithms (such as SIFT, SURF, and ORB) to extract the same types of features as those in text data.

[0056] Based on the extracted geometric feature point coordinates, a nearest neighbor matching algorithm is used to initially match the runway / taxiway features in the text data with the image data. Semantic similarity (e.g., cosine similarity) is calculated to further verify and adjust the matching relationship to ensure that the matched runways / taxiways are the same object.

[0057] The data fusion performs weighted fusion of geometric information according to the credibility of the data. The data with high credibility accounts for a larger proportion in the fusion, further eliminating data contradictions and redundancy, and improving the accuracy of the superimposed data. Specifically, the credibility is set for the geometric attributes and spatial relationship of the runway / taxiway and the aligned image data respectively, and the geometric information of the two is weighted fused. For example, for the runway width information, if the credibility of the geometric attributes and spatial relationship of the runway / taxiway is , the credibility of the aligned image data is , the width after fusion is: ,in, and They are the geometric attributes and spatial relationships of the runway / taxiway, and the runway width in the aligned image data. 、 The sum is 1;

[0058] Overlaying involves placing the two types of data in the same spatial coordinate system and accurately placing the line information at the corresponding actual scene location in the image data. This allows the geometric lines of the runway / taxiway to overlap with the actual runway / taxiway area in the image, forming a comprehensive data display that includes both geometric attribute information and actual scene information. For example, in GIS software, by overlaying a line feature layer of the runway centerline with a processed satellite image layer, the user can intuitively see the specific location of the line in the image and the alignment of the runway centerline represented by the line with the actual runway in the image.

[0059] During the overlay process, if there is a data conflict between the geometric attributes and spatial relationships of the runway / taxiway and the aligned image data (for example, the runway length in the geometric attributes and spatial relationships of the runway / taxiway is significantly different from the length measured in the image data), the geometric attributes and spatial relationships of the runway / taxiway will prevail, that is, set The fused width is obtained with a confidence level of 1. Local corrections are performed on the image data (e.g., through interpolation or deformation algorithms). The corrected image data is then superimposed with the geometric attributes and spatial relationships of the runway / taxiway. This optimizes the overlay effect, eliminates conflicts and redundancies between the data, highlights key information, and ultimately generates an accurate and reliable runway and taxiway layout.

[0060] Based on the layout diagram and the rules of the rule base, a spatiotemporal pyramid layering strategy is used to generate a multi-resolution candidate point set on both sides of the runway / taxiway centerline, and the candidate point set is adaptively processed to obtain airport light positions; the attributes of the airport light positions include: light type, color, illumination range, and light number;

[0061] In this embodiment, the rule base rules are lighting specifications extracted from ICAO Annex 14 and FAA AC 150 / 5345-43 standard documents, such as the runway edge light spacing Dj requirement and the taxiway edge light spacing requirement in the rule base;

[0062] Based on the coordinate information of the runway and taxiway centerlines in the layout diagram, their spatial range in the target coordinate system is determined. Based on this range, a spatiotemporal pyramid layering strategy is used to divide the airport area into L layers (for example, L = 3, respectively: high layer covering the global area, middle layer covering the taxiways, and bottom layer covering the key areas). The resolution of each layer is set and the number of particles is allocated. Each particle represents a candidate point, forming a candidate point set.

[0063] Adaptive processing is performed on the candidate points in the following manner:

[0064] A feature vector is generated by fusing (sequentially combining features) the geometric features of the layout diagram, the rules of the rule base, and the airport operation parameters; the airport operation parameters include three-dimensional laser point cloud elevation information (which can be obtained by scanning the airport area with radar), meteorological influencing factors (which can be obtained from the airport meteorological monitoring system, including wind speed, wind direction, and visibility), and flight flow parameters (which can be obtained from the airport flight scheduling system, including peak hours and busy runways).

[0065] The feature vector is input into the GNN-Transformer hybrid model to calculate the invalidity probability score of the candidate point. Specifically, the spatial topology of the airport area is modeled using GNN (graph neural network). The candidate point and its adjacent points are used as nodes of the graph, and the connection relationship between points is used as edges. The information of the neighboring nodes is aggregated through the message passing mechanism to extract the spatial features of the candidate point. Then, the spatial features extracted by GNN are input into the Transformer model together with the original feature vector. The Transformer's multi-head attention mechanism is used to capture the long-range dependencies between different features and learn the complex associations between features. Finally, the invalidity probability score of the candidate point is output through a fully connected layer and a Sigmoid activation function. The score ranges from 0 to 1. The higher the value, the greater the possibility that the candidate point is invalid.

[0066] The feature vector is input into the multi-dimensional risk assessment model to calculate the technical risk, safety risk, economic risk and total risk value of the candidate point. The formula is as follows:

[0067] ;

[0068] ;

[0069] ;

[0070] ;

[0071] Where, 、 、 、 They are technical risk, safety risk, economic risk and total risk value; is the distance (between the candidate point and the ideal light position), Score the interference degree of the candidate point to the navigation equipment, is the probability that the candidate point will cause flight conflict or accident, is the score of the candidate point’s degree of line of sight occlusion, 、 、 The construction cost, maintenance cost, and energy consumption cost of setting up lighting positions at candidate points are respectively, 、 、 、 、 、 All are weight coefficients;

[0072] Based on the invalid probability score and total risk value of the candidate point, the candidate point is determined to be an invalid point, and a reinforcement learning algorithm is used to select a processing strategy to process the invalid point; the processing strategy includes deletion and position adjustment; specifically, the feature vector, invalid probability score and total risk value of the candidate point are integrated into a state vector; multiple processing strategies for invalid points, such as deleting invalid points and adjusting the position of invalid points, are defined as action space; when the effect after invalid point processing is used as a reward function, the total risk value is reduced by a certain proportion, and a corresponding positive reward is given. Reinforcement learning uses the Q-learning algorithm according to the current state to select an action to be executed from the action space. During the training process, through continuous interaction, according to the reward feedback obtained after executing the action, its own policy network parameters are updated, and the optimal action selection strategy under different states is learned, such as adjusting the position of the point, and obtaining rewards based on the effect after adjustment, thereby optimizing subsequent adjustment decisions;

[0073] The position adjustment includes performing position adjustment using an interpolation method;

[0074] The position adjustment includes establishing a multi-objective optimization function and adopting a multi-objective particle swarm optimization algorithm to perform position adjustment;

[0075] The improved multi-objective particle swarm optimization algorithm (IM-MOPSO algorithm) is used for position adjustment. The method is as follows:

[0076] Step A1: Each invalid candidate point is corrected by the projection operator. The projection formula is: ,in, is the point obtained after projection transformation, is the boundary position, is the boundary correction coefficient, and the candidate points after correcting the invalid candidate points are used as the compliant candidate point set to ensure that all candidate points meet the ICAO specifications;

[0077] Step A2: Map the qualified candidate point set to the quantum bit space, calculate the measurement probability of each point to generate the initial quantum state parameters, output the layered particle swarm, quantum coding matrix and key area identification, dynamically adjust the sparse threshold according to the invalid probability score and environmental parameters, generate a sparse mask to screen valid candidate points, and form an optimized candidate point set;

[0078] Obtain weather (wind speed, visibility) and flight flow parameters in real time and dynamically adjust target weights , invalid probability score corrects the target weight coefficient, calculates the sparse threshold according to the environment change rate, generates a mask M based on the sparse threshold, and screens valid candidate points;

[0079] In step A3, the IM-MOPSO algorithm is used to optimize the candidate point set. A multi-objective optimization function is set, and the target priorities are adjusted using a dynamic weight formula. The candidate point positions are updated using an improved particle update formula, and a Pareto front is constructed through non-dominated sorting. The historical elite solution set is integrated and iteratively updated to ultimately determine the optimal light position layout. Based on the optimized light position layout, light attributes are automatically assigned according to the rules of the rule base.

[0080] The IM-MOPSO algorithm is used to optimize the candidate point set. Specifically, 1. Set the multi-objective optimization function:

[0081] ;

[0082] Where, For the accuracy of light position layout, To ensure flight safety, For energy consumption, For construction and maintenance costs, is the weight coefficient;

[0083] The improved particle update formula is:

[0084] ;

[0085] Where, is the velocity vector, is the position vector, is the number of iterations, is the inertia weight, is the individual learning factor, is the social learning factor, 、 are all random numbers, particle The best position in history, With particles The best location on the same floor; is a sparse mask vector;

[0086] 3. Calculate the non-dominated level and congestion distance of each solution under the multi-objective optimization function to construct the Pareto front solution set; integrate the current Pareto front with the historical elite solution set, update the elite solution set through non-dominated sorting, and determine the individual optimal and global optimal solutions; decide whether to terminate the optimization based on the degree of convergence of the elite solution set or the maximum number of iterations; if terminated, output the final light position layout;

[0087] In this way, through hierarchical calculation, quantum-inspired search, and dynamic weight adjustment, the computational complexity of traditional MOPSO is reduced, achieving efficient real-time optimization of airport lighting layout. Position adjustments are then used to optimize the layout of airport lighting positions. This adaptive processing not only meets basic safety requirements but also effectively avoids layout blind spots and redundant points caused by traditional methods, making the distribution of candidate points more closely aligned with the actual operational needs of the airport.

[0088] Construct cognitive rules and multi-objective models, and use the MOPSO algorithm to solve the target attributes of airport light positions. Based on the target attributes and actual attributes of the airport light positions, an enhanced Hi-IDRL algorithm is used to control the attributes of the airport light positions.

[0089] In this embodiment, the method for constructing the cognitive rules and multi-objective model is as follows:

[0090] The cognitive rules include constructing cognitive rules based on personnel cognitive needs, light attributes, and environmental conditions. The cognitive rules are constructed as follows: Constructing a causal relationship: A causal relationship is established between the environmental condition node of low visibility and severe weather (such as heavy fog and heavy rain) and the light attribute node of adjusting the approach light color to a warm tone (orange), indicating that the former is the trigger for the latter adjustment, that is, the severe environment causes the light color to be adjusted to enhance warning and safety. There is a causal relationship between the environmental condition node corresponding to the specific time of 2-4 a.m. and the light attribute node of reducing the flashing frequency of the runway edge lights, because personnel have high visual fatigue during this period, and the flashing frequency is adjusted to reduce fatigue.

[0091] Establishing influence relationships: Lights of different colors affect the distribution of visual attention and subjective perceptions of personnel. For example, an influence relationship is established between the white light node and the human cognitive node of pilots' ability to clearly identify runway boundaries at night. This indicates that white runway edge lights help pilots clearly identify runway boundaries, while lights with a fast flashing frequency have an impact on the human cognitive node of personnel's visual fatigue level. These nodes are connected through influence relationships.

[0092] Dependency building: The light location attribute node has dependencies on the environmental condition nodes of different airport areas (runways, taxiways, and aprons). The runway edge light node is dependent on the environmental condition node of the runway area. Its attribute adjustment needs to consider specific environmental factors of the runway area, such as runway length and slope.

[0093] The multi-objective model includes a personnel satisfaction objective model, a flight safety objective model, and an energy consumption objective model;

[0094] The personnel satisfaction target model is: ;

[0095] The flight safety target model is:

[0096] ;

[0097] The energy consumption target model is:

[0098] ;

[0099] Where, is the actual light color, For the optimal light color, is the actual light flashing frequency, For the optimal light flashing frequency, is the actual light brightness, For optimal light brightness, is the matching function, For visibility, is the visibility function, is the boot function, 、 、 、 are weight coefficients, is power, is the total number of airport lights; among them, 、 、 The airport operation experts, pilots and ground staff were asked to rate the relative importance of lighting attributes and environmental factors under the personnel satisfaction target;

[0100] The MOPSO algorithm is used to solve the target attributes of the airport light positions, and its method is as follows:

[0101] Initialize the particle swarm, where each particle represents a set of light attribute combinations C, F, and B;

[0102] Calculate the function value of each target model of each particle in the current state, and determine the fitness value of the particle according to the target function value;

[0103] Iteratively update the particle speed and position, calculate the fitness value of each particle after the update, compare this value with the historical optimal fitness value, update the individual optimal position of each particle, compare the fitness values ​​of all particles, update the global optimal position of the entire group, and continuously adjust the particle search direction until the stopping condition is met (for example, the maximum number of iterations is reached, the change in the global optimal position in 10 consecutive iterations is less than the set threshold of 0.01, and the average change rate of the objective function value is less than the set threshold of 0.001). The target attributes of the airport light position are obtained;

[0104] Based on the target attributes and actual attributes of the airport light positions, the enhanced Hi-IDRL algorithm is used to control the attributes of the airport light positions. The method is as follows:

[0105] Get the current light position target attributes, actual attributes, environmental parameters and operating mode to build the state space ;

[0106] Constructing the action space ;

[0107] in, Adjust according to the rate of change of regional environment;

[0108] Set up a multi-scale reward function (using a quantile regression network to predict multi-scale rewards to achieve a more reasonable reward trade-off), including three time-scale components: safety, energy consumption, and maintenance;

[0109] Based on the state space, the causal structure between environment states is learned by minimizing the causal reconstruction loss to generate a dynamic causal graph; the minimized causal reconstruction loss is: , where Causal reconstruction loss, measuring the actual environment state and the reconstructed environment state The degree of difference between is the weight coefficient, are the sparsity and acyclic constraints of the causal graph, impose constraints on the causal graph for functions, is the set of edges in the causal graph;

[0110] The dynamic causal diagram is: , where is the regional environmental change rate, For the moment t The set of edges in a causal graph;

[0111] Using the spatiotemporal attention mechanism, based on the causal relationship of the dynamic causal graph, we can capture the spatial dimension light position association and the dynamic change of the time dimension respectively. Specifically, we calculate the spatial attention through the multi-head attention mechanism. , capturing the dynamic changes of light attributes and environmental status from the time dimension, and calculating the time attention using the multi-head attention mechanism after transformation , and finally fused to get , can fully grasp the characteristics of light positions in time and space dimensions, and provide richer and more accurate information for strategy optimization;

[0112] Using quantum-inspired natural gradient optimization strategy, we construct quantum state representation parameter space and use Hamiltonian Calculate the nonlocal correlation between parameters, combined with the Fisher information matrix Forming a modified natural gradient update rule, ,in, is the quantum entanglement matrix, are policy network parameters, is the Fisher information matrix, is the learning rate, is the quantum correlation weight, Strategy optimization objective function, is the gradient operator of the policy network parameters;

[0113] This rule can update the algorithm's strategy parameters more efficiently and accurately, improving the algorithm's optimization efficiency and accuracy;

[0114] To solve the single-point bottleneck and communication delay problems of the central coordinator, we set up a distributed multi-agent collaborative optimization framework to generate action vectors through joint policy gradients. The method is as follows:

[0115] After obtaining the current state space, historical data, and communication topology, the airport is divided into N control areas, and the agent in each area receives the local state;

[0116] Each regional agent calculates the local gradient based on the policy network, adopts the gradient aggregation method enhanced by federated learning, and distributes the gradient through adaptive weights. ,in, is the number of regional samples, The temperature parameter is used, and global knowledge injection is combined to optimize the gradient exchange and integration process. At the same time, a decentralized communication protocol is introduced, and a rumor protocol is used to realize multi-agent gradient information diffusion. The Byzantine fault tolerance algorithm is combined to identify abnormal nodes.

[0117] Based on the local gradients, the central coordinator calculates the joint policy gradient and integrates the post-action incentive information of each region to generate the integrated action vector ;

[0118] Apply safety constraint projection to the integrated motion vector to generate the final motion vector ,in, is the projection function;

[0119] In order to ensure that the strategy update can meet safety constraints and have the flexibility to deal with complex scenarios, an airport operation knowledge graph is constructed to transform safety regulations and brightness standards into symbolic constraints. ; through differentiable symbolic-neural mapping functions , integrating the symbolic constraint into the reinforcement learning loss function, we get the total loss function: ,in, is the reinforcement learning loss, is the policy function, is the weight coefficient;

[0120] After executing the action, new state information is collected, and a quantile regression network is used to predict multi-scale rewards. The parameters of the state space, dynamic causal graph, and quantum state representation are updated to continuously improve the accuracy and adaptability of the control of airport light properties.

[0121] In this way, the MOPSO algorithm initializes a large number of particles to represent different combinations of light attributes, searches in parallel in the solution space, and utilizes information sharing and iterative update mechanisms between particles. It can quickly find the optimal or relatively optimal combination of light attributes that meets the multi-objective model (personnel satisfaction, flight safety, and energy consumption) on a global scale, providing a theoretically optimal solution for airport light attribute optimization. The enhanced Hi-IDRL algorithm is based on the actual operating status (actual attributes) of airport light locations. Through continuous interaction with the environment, it quickly adjusts actions (light attribute adjustments) according to the feedback of the reward function, responding to environmental changes in real time during actual operation. The collaborative work of the MOPSO algorithm to solve the target attributes and the enhanced Hi-IDRL algorithm to control the attributes can achieve efficient, accurate, and intelligent simulation of airport light attribute optimization.

[0122] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of the present invention.

[0123] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0124] The terms "first", "second", etc. are used to distinguish similar objects, rather than to describe or indicate a particular order or sequence.

[0125] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0126] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. The method for generating airport lights based on AIP documents and image data is characterized by: The method comprises the following steps: Obtaining the AIP text and performing semantic parsing to extract the geometric attributes and spatial relationships of the runway / taxiway; the geometric attributes and spatial relationships of the runway / taxiway include the size and position information of the runway centerline, runway edgeline, taxiway centerline, and taxiway edgeline; Obtaining satellite images and GIS data, extracting control points from each, calculating a covariance matrix of the control points, and using affine transformation and thin plate spline transformation based on the covariance matrix to achieve coordinate alignment of the satellite image and GIS data to obtain aligned image data; Generate a runway / taxiway layout diagram by performing data format standardization, feature extraction and matching, data fusion, and overlay on the geometric attributes and spatial relationships of the runway / taxiway and the aligned image data; Based on the layout diagram and the rules of the rule base, a spatiotemporal pyramid layering strategy is used to generate a multi-resolution candidate point set on both sides of the runway / taxiway centerline; the candidate point set is adaptively processed to obtain airport light positions; the attributes of the airport light positions include: light type, color, illumination range, and light number; A cognitive rule and multi-objective model are constructed, and the MOPSO algorithm is used to solve the target attributes of the airport light positions. Based on the target attributes and actual attributes of the airport light positions, the enhanced Hi-IDRL algorithm is used to control the attributes of the airport light positions. The method is as follows: Obtain the current light position target attributes, actual attributes, environmental parameters and operating mode to construct the state space; Based on the state space, the causal structure between environment states is learned by minimizing the causal reconstruction loss to generate a dynamic causal graph; Using the spatiotemporal attention mechanism, based on the causal relationship of the dynamic causal graph, the spatial dimension light position association and the dynamic changes in the temporal dimension are captured respectively; A quantum-inspired natural gradient optimization strategy is used to construct a quantum state representation parameter space. The Hamiltonian is used to calculate the non-local correlation between parameters, and the Fisher information matrix is ​​combined to form a modified natural gradient update rule. Based on the distributed multi-agent collaborative optimization framework, action vectors are generated through joint policy gradients; A knowledge graph of airport operations is constructed, and safety regulations and brightness standards are converted into symbolic constraints. The symbolic constraints are integrated into the reinforcement learning loss function through a differentiable symbolic-neural mapping function. After the action is executed, new state information is collected, and a quantile regression network is used to predict multi-scale rewards, and the parameters of the state space, dynamic causal graph, and quantum state representation are updated.

2. The method for generating airport lighting based on AIP documents and image data according to claim 1, characterized in that: The method for adaptively processing the candidate points is as follows: Generate a feature vector based on the fusion of the geometric features of the layout diagram, the rules of the rule base, and the airport operation parameters; the airport operation parameters include three-dimensional laser point cloud elevation information, meteorological influencing factors, and flight flow parameters; Input the feature vector into the GNN-Transformer hybrid model to calculate the invalid probability score of the candidate point; Inputting the feature vector into a multi-dimensional risk assessment model to calculate the technical risk, safety risk, economic risk and total risk value of the candidate point; The candidate point is determined to be an invalid point based on the invalid probability score and the total risk value of the candidate point, and a reinforcement learning algorithm is used to select a processing strategy to process the invalid point; the processing strategy includes deletion and position adjustment.

3. The method for generating airport lighting based on AIP documents and image data according to claim 2, characterized in that: The position adjustment includes performing position adjustment using an interpolation method; The position adjustment includes establishing a multi-objective optimization function and adopting an improved multi-objective particle swarm optimization algorithm to perform the position adjustment.

4. The method for generating airport lighting based on AIP documents and image data according to claim 3, characterized in that: The improved multi-objective particle swarm optimization algorithm is used for position adjustment, and the method is as follows: Step A1: Each invalid candidate point is corrected by a projection operator, and the candidate points after the invalid candidate points are corrected are used as a set of compliant candidate points; Step A2: Map the qualified candidate point set to the quantum bit space, calculate the measurement probability of each point to generate the initial quantum state parameters, output the layered particle swarm, quantum coding matrix and key area identification, dynamically adjust the sparse threshold according to the invalid probability score and environmental parameters, generate a sparse mask to screen valid candidate points, and form an optimized candidate point set; In step A3, the IM-MOPSO algorithm is used to optimize the candidate point set. A multi-objective optimization function is set, and the target priorities are adjusted using a dynamic weight formula. The candidate point positions are updated using an improved particle update formula, and a Pareto front is constructed through non-dominated sorting. The historical elite solution set is integrated and iteratively updated to ultimately determine the optimal light position layout. Based on the optimized light position layout, light attributes are automatically assigned according to the rules of the rule base. The improved particle update formula is: ； Where, is the velocity vector, is the position vector, is the number of iterations, is the inertia weight, is the individual learning factor, is the social learning factor, 、 are all random numbers, particle The best position in history, With particles The best location on the same floor; is the sparse mask vector; The multi-objective optimization function: ； Where, For the accuracy of light position layout, To ensure flight safety, For energy consumption, For construction and maintenance costs, is the weight coefficient.

5. The method for generating airport lighting based on AIP documents and image data according to claim 1, characterized in that: The method for constructing cognitive rules and multi-objective models is as follows: The cognitive rules include constructing cognitive rules based on personnel cognitive needs, lighting properties, and environmental conditions; The multi-objective model includes a personnel satisfaction objective model, a flight safety objective model, and an energy consumption objective model; The personnel satisfaction target model is: ； The flight safety target model is: ； Where, is the actual light color, For the optimal light color, is the actual light flashing frequency, For the optimal light flashing frequency, is the actual light brightness, For optimal light brightness, is the matching function, For visibility, is the visibility function, is the boot function, 、 、 、 are all weight coefficients.

6. The method for generating airport lighting based on AIP documents and image data according to claim 5, characterized in that: The MOPSO algorithm is used to solve the target attributes of the airport light positions, and its method is as follows: Initialize the particle swarm, where each particle represents a set of light attribute combinations C, F, and B; Calculate the function value of each target model of each particle in the current state, and determine the fitness value of the particle according to the target function value; Iteratively update the speed and position of the particles, calculate the fitness value of each particle after the update, compare this value with the historical optimal fitness value, update the individual optimal position of each particle, compare the fitness values ​​of all particles, update the global optimal position of the entire group, and continuously adjust the search direction of the particles until the stopping condition is met to obtain the target attributes of the airport light position.

7. The method for generating airport lighting based on AIP documents and image data according to claim 1, characterized in that: The distributed multi-agent collaborative optimization framework generates action vectors by combining policy gradients. The method is as follows: After obtaining the current state space, historical data, and communication topology, the airport is divided into N control areas, and each area agent receives the local state; Each regional agent calculates local gradients based on the policy network, using federated learning-enhanced gradient aggregation, adaptive weight allocation, and global knowledge injection to optimize gradient exchange and integration. A decentralized communication protocol is introduced, using a rumor protocol to achieve multi-agent gradient information diffusion, and a Byzantine fault-tolerant algorithm is used to identify abnormal nodes. Based on the local gradients, the central coordinator calculates the joint policy gradient and integrates the post-action incentive information of each region to generate an integrated action vector; Apply safety constraint projection to the integrated motion vector to generate the final motion vector.

8. The method for generating airport lighting based on AIP documents and image data according to claim 1, characterized in that: The data fusion is to set credibility for the geometric attributes and spatial relationship of the runway / taxiway and the aligned image data respectively, and perform weighted fusion on the geometric information of the two.

9. The method for generating airport lighting based on AIP documents and image data according to claim 8, characterized in that: During the overlay, if any data conflict occurs between the geometric attributes and spatial relationships of the runway / taxiway and the aligned image data, the image data is corrected based on the geometric attributes and spatial relationships of the runway / taxiway, and the corrected image data is overlaid with the geometric attributes and spatial relationships of the runway / taxiway.

Images