A dynamic airspace gridding management and service system for urban low-altitude environment

Abstract

The application discloses a kind of dynamic airspace gridding management and service system for urban low-altitude environment, it is related to urban low-altitude airspace management technical field, including grid subdivision module, semantic layer module, space-time body generation module.Grid subdivision module is based on the three-dimensional reconstruction of the improved GeoSOT coding system to the city three hundred meters below low-altitude airspace, building dense area uses 5m×5m×2m fine grid subdivision, open airspace is merged into cluster grid by octree algorithm, forms non-uniform resolution airspace grid.Semantic layer module superimposes landing field buffer zone, emergency landing area, high-voltage line corridor isolation zone logic label, realizes spatial entity and management attribute decoupling mapping.Space-time body generation module introduces time component for grid unit and constructs four-dimensional space-time body, as trajectory reservation and conflict detection minimum unit.The system can adapt to complex scene of urban low-altitude, realizes airspace fine dynamic control.

Description

Technical Field

[0001] This invention belongs to the field of urban low-altitude airspace management technology, specifically a dynamic airspace grid management and service system for urban low-altitude environments. Background Technology

[0002] Currently, urban low-altitude airspace management below 300 meters mostly uses uniform-sized grids for airspace division, failing to differentiate between densely built-up areas and open airspace. 3D airspace reconstruction does not rely on the improved GeoSOT coding system, nor does it utilize the octree algorithm for grid merging. Existing airspace management merely overlays basic geographic identifiers onto the spatial grid, without assigning dedicated logical labels to takeoff and landing buffer zones, emergency landing zones, and high-voltage power line corridor isolation zones. Geographical entities and management attributes remain coupled, failing to form an independent logical semantic airspace grid layer.

[0003] Existing airspace grid division methods suffer from insufficient spatial accuracy in densely built-up areas, while open airspace contains a large number of redundant grids. Furthermore, the 3D airspace partitioning is poorly adapted to the actual spatial morphology of urban low-altitude areas. Relying solely on spatial geometric attributes for airspace division fails to differentiate the management attributes of different functional zones, resulting in a chaotic binding relationship between airspace labels and spatial grids, hindering refined airspace attribute management.

[0004] Existing airspace management uses only spatial grids as the basic management unit, without introducing a time dimension component to the grid unit. It lacks a four-dimensional spatiotemporal structure that includes spatial coordinates and time slices, which cannot meet the spatiotemporal control requirements of dynamic flight of aircraft, nor can it provide the smallest physical unit that can be adapted for aircraft trajectory reservation and conflict detection. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art;

[0006] Therefore, this invention proposes a dynamic airspace grid management and service system for urban low-altitude environments, comprising:

[0007] The mesh partitioning module is used to reconstruct the three-dimensional space of the low-altitude airspace below 300 meters in the city based on the improved GeoSOT coding system. For densely built areas, a fine basic mesh of 5 meters by 5 meters by 2 meters is used for partitioning. For open airspace, the octree algorithm is used to merge them into cluster meshes to generate a basic airspace mesh set with non-uniform resolution.

[0008] The semantic layer module is used to overlay logical labels of take-off and landing field buffer zones, emergency landing zones, and high-voltage line corridor isolation zones on the geometric attributes of the basic airspace grid set, establish a decoupled mapping relationship between geospatial entities and management attributes, and form an airspace grid layer with logical semantics.

[0009] The spacetime volume generation module is used to introduce a time dimension component into each independent grid cell in the airspace grid layer with logical semantics, construct a four-dimensional spacetime volume containing spatial coordinates and time slices, and establish the four-dimensional spacetime volume as the smallest physical unit for subsequent aircraft trajectory reservation and conflict detection.

[0010] Furthermore, after constructing the four-dimensional spacetime volume containing spatial coordinates and time slices, the process also includes:

[0011] The dynamic weight calculation module is used to calculate the dynamic weight attributes of the grid cells within the four-dimensional spacetime body.

[0012] A multidimensional weight vector model is constructed to define a real-time weight vector for each grid cell in the four-dimensional spatiotemporal body. The real-time weight vector includes the ground population dynamic density index obtained from base station signaling data parsing, the obstacle occupancy rate index obtained from lidar point cloud updates, the wind speed and visibility risk value index obtained from microscale meteorological simulation, the aircraft traffic saturation index of the currently reserved grid cell, and the electromagnetic environment interference level index.

[0013] The fuzzy comprehensive evaluation method is used to calculate each index in the real-time weight vector to generate a comprehensive airworthiness index that characterizes the airworthiness status of the grid cell;

[0014] The system compares the comprehensive airworthiness index with a preset warning threshold. When the comprehensive airworthiness index exceeds the warning threshold, the system automatically switches the state of the grid cell from open to restricted or closed, and writes the state change information into the message queue of the global scheduling bus.

[0015] Furthermore, the construction method of the multidimensional weight vector model includes:

[0016] The raw data associated with each grid cell in the four-dimensional spatiotemporal body is collected. The raw data includes ground population density data from base station signaling, obstacle point cloud and image data from lidar and visual sensors, wind speed and visibility data from micro-weather stations and numerical simulation models, aircraft reservation flow data recorded from the global scheduling bus, and electromagnetic signal strength data from spectrum monitoring equipment.

[0017] A standardized processing function is defined for each type of raw data to process the raw data into index values ​​with consistent dimensions and value ranges. The index values ​​include ground population dynamic density index, obstacle occupancy rate index, wind speed and visibility risk value index, aircraft flow saturation index, and electromagnetic environment interference level index.

[0018] Each index value obtained from the processing is used as a dimension component in a vector and combined according to a predetermined dimensional order to form a real-time weight vector that characterizes the comprehensive environmental state of the grid cell under a specific time slice.

[0019] Furthermore, before calculating the various indicators in the real-time weight vector using the fuzzy comprehensive evaluation method, the following steps are also included:

[0020] The data fusion module is used to fuse heterogeneous sensing data.

[0021] Establish a heterogeneous data access channel to comprehensively receive sensing echo data, broadcast automatic correlation surveillance data, remote identification data, and visual data from 5G or 6G mobile communication technology base stations, and construct a raw observation dataset of low-altitude targets.

[0022] Feature extraction and identification are performed on the observation records belonging to non-cooperative targets in the original observation dataset. Multi-source data cross-validation is used to fill the blind spots of single sensor detection, forming a highly reliable situational awareness base map covering the entire airspace.

[0023] The high-reliability situational awareness base map is spatially registered with the four-dimensional spatiotemporal volume, so that each grid cell can be associated with a real-time environmental state description.

[0024] Furthermore, after forming a highly reliable situational awareness base map covering the entire airspace, the method further includes:

[0025] The airspace occupancy prediction module is used to predict future airspace occupancy.

[0026] Retrieve the historical state vector of the UAV recorded in the high-reliability situational awareness base map for the previous several time steps. The historical state vector includes position, velocity, acceleration and attitude information.

[0027] The extracted historical state vector is input into a pre-trained long short-term memory network model, and the model inference outputs the predicted trajectory coordinates of the UAV in the next few time steps.

[0028] The predicted trajectory coordinates are discretized and mapped to a series of continuous grid occupancy sequences within the four-dimensional spacetime, thereby realizing the serialized expression of the flight situation in digital space. The grid occupancy sequence contains the airspace occupancy status at future moments.

[0029] Furthermore, after realizing the serialized representation of the flight status in digital space, it also includes:

[0030] The reservation verification module is used to perform four-dimensional trajectory reservation verification based on the grid occupancy sequence.

[0031] Receive the expected flight path data submitted by the aircraft and parse the expected flight path data into a target grid sequence within the four-dimensional spacetime;

[0032] Call the real-time weight vector and status flag corresponding to each grid cell in the target grid sequence, perform grid deduplication operation, and determine whether the target grid sequence has been locked by other tasks or whether its comprehensive airworthiness index exceeds the standard within a specified time period;

[0033] If the verification result indicates the presence of conflicting or high-risk grids, the conflict information is fed back to the scheduling logic interface to trigger the subsequent cooperative avoidance algorithm; otherwise, the target grid sequence is marked as reserved.

[0034] Furthermore, the triggering of subsequent cooperative avoidance algorithms, which are executed through improved heuristic search logic, specifically includes:

[0035] The real-time weight vectors of each grid cell in the four-dimensional spacetime volume are used as cost factors for path search to construct a cost function. The cost function is composed of the actual cost of reaching the current grid from the starting point and the heuristic cost of the estimated destination.

[0036] A time-dimensional penalty term is introduced into the cost function. By adjusting the hovering time of the UAV in a specific grid cell or the fine-tuning coefficient of the flight speed, the cost difference brought about by different avoidance strategies is quantified.

[0037] Search for an optimal connectivity path through the target grid sequence within the four-dimensional spacetime, avoiding grid cells that are restricted or closed during the search process, and generate a preliminary conflict-free flight plan.

[0038] Furthermore, after generating the preliminary conflict-free flight plan, the process also includes:

[0039] The real-time correction module is used to perform asynchronous real-time corrections for unexpected situations during operation.

[0040] Real-time monitoring of the aircraft's status during operation; when a yaw event caused by a sudden gust of wind or equipment failure is detected, the edge computing node is activated to analyze the local airspace region where the yaw event occurred.

[0041] In the four-dimensional spacetime volume, a local grid region affected by the yaw event is delineated, and the real-time weight vector and occupancy status of the existing grid cells in the local grid region are re-evaluated.

[0042] Based on the priority attributes of the tasks, dynamic priority rearrangement is performed on all aircraft within the local grid area, and heading correction commands are issued to aircraft corresponding to high-priority tasks. Waiting or detour logic is executed on aircraft corresponding to low-priority tasks to resolve local spatiotemporal conflicts.

[0043] Furthermore, the comprehensive airworthiness index is compared with a preset warning threshold. When the comprehensive airworthiness index exceeds the warning threshold, the method further includes:

[0044] The state assessment module is used to perform cascading state assessments on surrounding grid cells.

[0045] Centered on a grid cell marked as restricted or closed, retrieve several neighboring grid cells in adjacent layers;

[0046] Extract the current real-time weight vector of the neighboring grid cells, and focus on analyzing the changing trends of the obstacle occupancy rate index and the aircraft traffic saturation index;

[0047] If the trend indicates that the risk is spreading to the surrounding area, the comprehensive airworthiness index of the neighboring grid cell is automatically increased, and its status flag is updated synchronously according to the new index value, so as to realize the gradient transmission and early warning of the risk status.

[0048] Furthermore, inputting the extracted historical state vector into the pre-trained long short-term memory network model also includes quantifying the uncertainty of the predicted trajectory:

[0049] Monte Carlo dropout technique is used to perform multiple forward propagations on the long short-term memory network model during the inference phase to generate multiple predicted trajectory branches;

[0050] The multiple predicted trajectory branches are mapped to different grid occupancy sequences, and the probability of a high-density grid cluster appearing in each grid occupancy sequence is calculated.

[0051] The probability information is packaged together with the original grid occupancy sequence and passed as a confidence parameter to the four-dimensional trajectory reservation verification module for risk selection when there are multiple candidate routes.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] Based on the improved GeoSOT coding system, a three-dimensional spatial reconstruction of the urban low-altitude airspace below 300 meters is performed. Densely built areas are divided using a fine basic grid of 5 meters by 5 meters by 2 meters. Open airspace is merged into cluster grids using an octree algorithm, which can generate a basic airspace grid set with non-uniform resolution. The differentiated grid division form fits the spatial characteristics of different airspaces. The fine grid matches the complex three-dimensional spatial morphology of densely built areas. The cluster grid simplifies the grid composition of open airspace. The three-dimensional spatial reconstruction fits the three-dimensional spatial attributes of the urban low-altitude airspace. The non-uniform resolution grid set reduces invalid grid divisions, so that the airspace division form is consistent with the actual airspace environment.

[0054] Logical labels for takeoff and landing buffer zones, emergency landing zones, and high-voltage power line corridor isolation zones are overlaid on the geometric attributes of the basic airspace grid set. This establishes a decoupled mapping relationship between geospatial entities and management attributes, forming an airspace grid layer with logical semantics. This allows for the separation of spatial geometry and management attribute identifiers, with logical labels for different functional areas independently corresponding to grid cells, avoiding confusion in the binding of spatial entities and management attributes. A time dimension component is introduced into each independent grid cell with logical semantics, constructing a four-dimensional spatiotemporal volume containing spatial coordinates and time slices. This four-dimensional spatiotemporal volume serves as the smallest physical unit for aircraft trajectory reservation and conflict detection. The time dimension component and spatial grid cells combine to form a spatiotemporal control unit adapted to dynamic flight, conforming to the spatiotemporal variation characteristics of aircraft flight and matching the control requirements of trajectory reservation and conflict detection. Attached Figure Description

[0055] Figure 1 This is a sequence diagram of a dynamic airspace grid management and service system for urban low-altitude environments as described in this invention.

[0056] Figure 2 A flowchart illustrating the construction method of a multidimensional weight vector model;

[0057] Figure 3 A flowchart for heterogeneous sensing data fusion in the data fusion module;

[0058] Figure 4 A diagram illustrating the cost function analysis of the collaborative avoidance algorithm;

[0059] Figure 5 Radar map of environmental risks in urban low-altitude airspace. Detailed Implementation

[0060] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] See Figure 1 The system employs a grid partitioning module and an improved GeoSOT coding system to reconstruct the three-dimensional space of low-altitude airspace below 300 meters in urban areas. In densely built-up areas, a refined base grid of 5m x 5m x 2m is used for partitioning. In open airspace, an octree algorithm is used to merge the base grid into larger cluster grids, thus generating a base airspace grid set with non-uniform spatial resolution. The semantic layer module overlays logical labels for management areas, such as takeoff and landing buffer zones, emergency landing zones, and high-voltage power line corridor isolation zones, onto the geometric attributes of this base airspace grid set. This establishes a decoupled mapping relationship between the geospatial grid and management attributes, forming an airspace grid layer with logical semantics. The spatiotemporal volume generation module introduces a time dimension component into each independent grid cell in the logically semantic airspace grid layer, constructing a four-dimensional spatiotemporal volume containing spatial coordinates and time slices. This four-dimensional spatiotemporal volume is established as the smallest physical unit for subsequent aircraft trajectory reservation and conflict detection.

[0062] In one embodiment of the present invention, a dynamic airspace grid management and service system for urban low-altitude environments performs dynamic weight attribute calculations. Each grid cell within a four-dimensional spatiotemporal volume defines its real-time weight vector through a multi-dimensional weight vector model. (See [reference]). Figure 2 The construction of the real-time weight vector begins with the collection of associated raw data. This raw data includes ground population density data from base station signaling, obstacle point cloud and image data from lidar and visual sensors, wind speed and visibility data from micro-weather stations and numerical simulation models, aircraft reservation flow data recorded by the global scheduling bus, and electromagnetic signal strength data from spectrum monitoring equipment. In a specific implementation, taking a grid cell located above a city's commercial area as an example, base station signaling data analysis shows that the ground population density in the projected area of ​​this grid cell reaches 15,000 people per square kilometer during peak hours. LiDAR point cloud update data indicates that the obstacle occupancy rate within the grid cell has increased to 12% due to temporary construction work involving tower cranes. Microscale weather simulation outputs a wind speed of 5 meters per second and a visibility of 800 meters. Aircraft reservation flow data shows that three drones are currently scheduled to occupy this grid cell. Spectrum monitoring data indicates that the electromagnetic interference level is moderate. This raw data forms the input basis for the multi-dimensional weight vector model.

[0063] In specific implementation, a standardized processing function is defined for each type of raw data. Ground population density data is normalized by dividing by the maximum carrying capacity of the area. Obstacle point cloud and image data are converted into an obstacle occupancy rate index by calculating the ratio of occupied volume to the total grid volume. Wind speed and visibility data are mapped to wind speed and visibility risk value indicators by combining wind speed risk curves and visibility threshold functions. Aircraft reservation flow data is used to obtain an aircraft flow saturation index by the ratio of the current reservation number to the maximum grid capacity. Electromagnetic signal strength data is quantified into an electromagnetic environment interference level index according to a preset interference level table. The processed index values ​​have consistent dimensions and value ranges. In some embodiments, the specific form of the standardized processing function can be adjusted for different urban environments. For example, for the population density index, the standardized processing function can be designed as linear normalization, so that the index value falls between 0 and 1, expressed by the formula:

[0064]

[0065] Among them: I pop D represents the standardized dynamic population density index on the ground. current D represents the raw data of ground population density currently collected. max This represents the preset maximum carrying capacity of the area. In practice, each processed index value is used as a dimension component of a vector, and they are combined according to a predetermined dimensional order to form a real-time weight vector representing the comprehensive environmental state of the grid cell in a specific time slice. The real-time weight vector can be expressed as V=[I pop ,I obs ,I weather ,I traffic ,I em ], where I obs I represents the obstacle occupancy rate index. weather I represents the risk value index for wind speed and visibility. traffic I represents the flow saturation index of an aircraft. em This indicates the level of electromagnetic interference.

[0066] In some embodiments, the fuzzy comprehensive evaluation method is used to calculate the various indicators in the real-time weight vector. The fuzzy comprehensive evaluation method first defines a membership function for each indicator, maps the indicator value to fuzzy linguistic variables such as "low risk", "medium risk", and "high risk", and then, based on the pre-set weight allocation matrix, weighted synthesis of the membership degrees of each indicator is performed to generate a comprehensive airworthiness index that represents the airworthiness status of the grid cell. In practical implementation, taking the aforementioned commercial area grid unit as an example, after the membership function is calculated for each indicator of the real-time weight vector, the ground population dynamic density indicator belongs to the "high risk" level with a degree of 0.8, the obstacle occupancy rate indicator belongs to the "medium risk" level with a degree of 0.6, the wind speed and visibility risk value indicator belongs to the "low risk" level with a degree of 0.9, the aircraft flow saturation indicator belongs to the "high risk" level with a degree of 0.7, and the electromagnetic environment interference level indicator belongs to the "medium risk" level with a degree of 0.5. The weight allocation matrix is ​​set to [0.3, 0.2, 0.2, 0.2, 0.1] according to the airspace management strategy. The comprehensive airworthiness index is 0.68 obtained by weighted synthesis. In practice, the comprehensive airworthiness index is compared with a preset warning threshold. When the comprehensive airworthiness index exceeds the warning threshold, the system automatically switches the grid cell's status from open to restricted or closed, and writes the status change information to the message queue of the global scheduling bus. For example, if the warning threshold is set to 0.65, the aforementioned grid cell will be marked as restricted because its comprehensive airworthiness index of 0.68 exceeds the threshold. Optionally, the warning threshold can be dynamically adjusted according to airspace category or time period. For example, during nighttime hours, the warning threshold can be appropriately relaxed to allow for a higher comprehensive airworthiness index.

[0067] In one embodiment of the present invention, the dynamic airspace grid management and service system for urban low-altitude environments performs fusion processing of heterogeneous sensing data before calculating the various indicators in the real-time weight vector using the fuzzy comprehensive evaluation method. (See [reference]). Figure 3 The system establishes a heterogeneous data access channel, which comprehensively receives sensing echo data, automatic correlation surveillance (ALS) data, remote identification (RAD) data, and visual data from optoelectronic payloads from 5G or 6G base stations to construct a raw observation dataset for low-altitude targets. In an example scenario over a city's core area, the heterogeneous data access channel simultaneously receives signal reflection point clouds from multiple 6G base stations, ALS messages from a logistics drone, a RAD sequence from an unregistered model aircraft, and video frames of moving objects captured by a fixed surveillance camera. After timestamp alignment, these data collectively constitute a raw observation dataset for a specific airspace.

[0068] In practical implementation, the system extracts features and identifies observation records belonging to non-cooperative targets in the original observation dataset, and uses multi-source data cross-validation to fill the blind spots of single-sensor detection. Taking the aforementioned unregistered model aircraft as an example, the automatic correlation surveillance (ALS) data does not contain its information, forming a detection blind spot. However, the sensing echo from the 6G mobile communication technology (SCR) base station provides its approximate azimuth and radial velocity. The visual data from the optoelectronic payload extracts its contour, color, and motion features through image recognition algorithms. Although the remote identification data is unregistered, it provides its physical address code. The system associates these features with the identifiers and generates a temporary fused trajectory identifier for the target. In some embodiments, multi-source data cross-validation is achieved by calculating the consistency of state estimates of the same target from different data sources. For example, for the position estimation of a target at the same time, the spatial correlation between the position estimated from the sensing echo from the 6G base station, visual angle measurement, and remote identification data can be compared. When the correlation is lower than a threshold, the data point is marked as low confidence and triggers a more in-depth correlation analysis. In practical implementation, by continuously correlating and verifying multiple independent observations from heterogeneous data access channels, the system can gradually fill in visual sensor blind spots caused by building obstruction, or compensate for monitoring gaps in areas without broadcast automatic correlation surveillance (ALS) signal coverage. Ultimately, this results in a highly reliable situational awareness map covering the entire airspace. This map includes not only the precise four-dimensional trajectories of cooperative targets but also the fused trajectories of non-cooperative targets and their confidence assessments. Optionally, for high-value or high-risk areas, the data sampling frequency of the heterogeneous data access channels can be increased to improve the update rate of the highly reliable situational awareness map.

[0069] In some embodiments, the multi-source data cross-validation process can be quantified by a comprehensive confidence function, expressed as follows:

[0070]

[0071] Where: W represents the overall confidence level of the judgment on a certain target or spatial state, n represents the number of independent data sources participating in the fusion, w i This represents the preset weight assigned to the i-th data source, where weight w i Based on sensor type, historical accuracy, and current signal-to-noise ratio, c is dynamically adjusted. i Let c represent the confidence score of the observation data provided by the i-th data source at the current time. iThe quality of the data source itself (such as signal strength and image clarity) determines this. In practice, the system spatially registers the high-reliability situational awareness base map with the four-dimensional spatiotemporal volume. The spatial registration process uses a unified geographic coordinate reference system and time base to map the real-time position and state attributes of each target or environmental feature in the high-reliability situational awareness base map to the corresponding grid cell in the four-dimensional spatiotemporal volume, ensuring that each grid cell is associated with a real-time environmental state description. For example, the fused real-time position, speed, and identification information of the aforementioned unregistered model aircraft are written into the attributes of the grid cell in its current four-dimensional spatiotemporal volume.

[0072] In one embodiment of the present invention, the dynamic airspace grid management and service system for urban low-altitude environments, after forming a highly reliable situational awareness base map covering the entire airspace, performs predictions of future airspace occupancy. The system retrieves the historical state vectors of UAVs recorded in the highly reliable situational awareness base map for the previous several time steps. Taking a UAV performing an inspection mission in an industrial park as an example, the highly reliable situational awareness base map continuously records the historical state of the UAV for 100 time steps at 0.1-second intervals over the past 10 seconds. Each historical state vector contains the UAV's three-dimensional position coordinates, three-axis velocity components, three-axis acceleration components, and pitch, roll, and yaw attitude angle information. These historical state vectors constitute the input sequence of the Long Short-Term Memory network model. In some embodiments, the length of the previous several time steps and the sampling interval can be dynamically configured according to the task type. For high-speed UAVs, the sampling interval may be shortened to 0.05 seconds, and the number of time steps may be increased to 200 steps to ensure that the input sequence can fully capture its motion pattern. In practice, the system inputs the extracted historical state vector into a pre-trained long short-term memory network model. The long short-term memory network model has learned a large number of normal and abnormal flight trajectory patterns during the training phase. Through the gating mechanism inside the model, it infers the input sequence and outputs the predicted trajectory coordinates of the UAV in the next 20 time steps (i.e. the next 2 seconds). The predicted trajectory coordinates are a series of three-dimensional position points arranged in chronological order.

[0073] In some embodiments, the system utilizes Monte Carlo dropout technology to perform multiple forward propagations of the Long Short-Term Memory (LSTM) network model during the inference phase to quantify the uncertainty of prediction. During each forward propagation, the dropout layer in the LSM model randomly "drops" a portion of neuron connections according to the probability retained during training. Due to this randomness, each forward propagation generates slightly different predicted trajectory coordinates, thus forming multiple predicted trajectory branches. In a specific implementation, for the aforementioned industrial inspection drone, the system performs 100 Monte Carlo dropout forward propagations, generating 100 slightly different predicted trajectory branches. Each predicted trajectory branch represents a possible future flight path. The system discretizes each of these 100 predicted trajectory branches according to the spatial grid partitioning rules of a four-dimensional spacetime volume, mapping it to a corresponding future grid occupancy sequence. A grid occupancy sequence consists of a series of grid cells encoded as to be occupied in future time slices. In a specific implementation, the system calculates the probability of a high-density grid cluster appearing in each grid occupancy sequence. A high-density grid cluster refers to the phenomenon where the number of predicted occupied grid cells exceeds a set threshold within a local spatial range. Optionally, the formula for calculating the probability of high-density grid clusters is expressed as follows:

[0074]

[0075] Where: P hd This represents the probability of a high-density grid cluster appearing in the grid occupancy sequence for the predicted trajectory of the drone, where M represents the total number of forward propagations in Monte Carlo dropout, and C represents the probability of such a cluster appearing. k This represents the number of occupied grid cells belonging to a predefined high-risk local spatial region in the grid occupancy sequence obtained in the k-th forward propagation. N represents the total number of grid cells contained in this high-risk local spatial region, θ represents the preset density threshold, and I(·) is an index function, which takes a value of 1 when the condition in parentheses is true, and 0 otherwise. For example, if the total number of grid cells N in the high-risk region is 50 and the density threshold θ is set to 0.3, then in 100 simulations, there will be 35 instances of C. k If the value is greater than 15, then the probability P of a high-density grid cluster is... hd It is 0.35.

[0076] In its implementation, the system discretizes the predicted trajectory coordinates generated from the original single deterministic prediction (i.e., the model output without Monte Carlo dropout) and maps them to a series of continuous grid occupancy sequences within a four-dimensional spacetime. This grid occupancy sequence is a serialized representation of the flight situation in digital space, containing the most probable airspace occupancy state at future moments. It can be understood that the system calculates the probability information P... hdThis most likely original grid occupancy sequence is packaged together to form a prediction result package with a confidence attribute. This prediction result package is passed as a confidence parameter to the subsequent four-dimensional trajectory reservation verification module. Optionally, the confidence parameter may include the high-density grid cluster probability P. hd In addition, it can also include other uncertainty measures such as the variance or entropy of the predicted trajectory. It can be understood that by combining the sequence prediction capability of the Long Short-Term Memory network model with the uncertainty quantification of Monte Carlo dropout, the system not only provides a sequential representation of future airspace occupancy but also provides a probabilistic basis for risk assessment.

[0077] In one embodiment of the present invention, the dynamic airspace grid management and service system for urban low-altitude environments, after realizing the serialized representation of flight status in digital space, initiates a four-dimensional trajectory reservation verification process based on the grid occupancy sequence. The system receives the intended flight path data submitted by the aircraft, such as the flight path of a logistics drone from an urban distribution center to the top of an office building. The intended flight path data contains a series of four-dimensional waypoints ordered by time. The system parses these four-dimensional waypoints into the corresponding target grid sequence within the four-dimensional spacetime. The target grid sequence consists of a series of uniquely encoded grid units with timestamps. In specific implementation, the system calls the real-time weight vector and status flag corresponding to each grid unit in the target grid sequence to perform a grid deduplication operation. The grid deduplication operation determines whether the target grid sequence has been locked by other tasks within a specified time period or whether its comprehensive airworthiness index exceeds a preset safety threshold. Referring to Table 1, taking a simple target grid sequence involving five grid units (G1 to G5) as an example, its verification process can be compared with the following data.

[0078] Table 1: Example Table of Target Grid Sequence Reservation Verification Status

[0079]

[0080] In practical implementation, based on the results of the grid deduplication operation, if the verification result indicates the presence of conflicting or high-risk grids, the system feeds back the conflict information to the scheduling logic interface, triggering the subsequent cooperative avoidance algorithm; otherwise, the target grid sequence is marked as reserved. Taking the table above as an example, since grid cell G3 is in a "restricted" state at time slice T3 and is locked by task B, the system determines that the current reservation request has a spatiotemporal conflict at grid cell G3. Therefore, it sends the conflict information, including the grid cell G3 code, time T3, and the conflicting task B identifier, to the scheduling logic interface. In some embodiments, the grid deduplication operation is executed concurrently, allowing for the simultaneous verification of the status of multiple grid cells to improve the response speed for complex long-haul route verification.

[0081] In some embodiments, after the subsequent cooperative avoidance algorithm is triggered, the cooperative avoidance algorithm performs path replanning through improved heuristic search logic. The cooperative avoidance algorithm uses the real-time weight vector of each grid cell in the four-dimensional spacetime volume as the cost factor for path search and constructs a cost function. The cost function F(n) is composed of the actual cost g(n) from the starting point to the current grid and the heuristic cost h(n) estimated to the destination. The formula is expressed as F(n) = g(n) + h(n).

[0082] In practical implementation, the actual cost g(n) from the starting point to the current grid is a weighted accumulation of relevant indicators in the real-time weight vector of the traversed grid cells along the path. For example, the actual cost g(n) can mainly accumulate the aircraft traffic saturation index and obstacle occupancy rate index. The heuristic cost h(n) is usually estimated using the Manhattan distance or Euclidean distance from the current grid to the target grid. The cooperative avoidance algorithm introduces a time-dimensional penalty term P(t) into the cost function. By adjusting the hovering time of the UAV in a specific grid cell or the fine-tuning coefficient of the flight speed, the cost difference brought about by different avoidance strategies is quantified. After introducing the time penalty term, the cost function is updated to F(n) = g(n) + h(n) + P(t). For example, if the avoidance strategy requires the drone to hover and wait for 5 seconds in a certain grid cell, the time-dimensional penalty term P(t) will be calculated based on a function positively correlated with time cost, resulting in a positive penalty value, which is added to the overall cost F(n). If the avoidance strategy involves slightly accelerating to pass through the conflict grid ahead of time, the time-dimensional penalty term P(t) may be a small value or even a negative value to reflect the time savings. In specific implementation, the cooperative avoidance algorithm searches for a starting and ending point of the target grid sequence within the four-dimensional spacetime, avoiding the optimal connectivity path of grid cells in a restricted or closed state, generating a preliminary conflict-free flight plan. The conflict-free flight plan is also expressed in the form of a set of four-dimensional grid sequences. Optionally, the calculation of the heuristic cost h(n) can incorporate more factors, such as considering the direction of risk diffusion, making the search more inclined to stay away from high-risk areas. It can be understood that the cost function based on the real-time weight vector and the introduced time-dimensional penalty term jointly guide the search direction of the cooperative avoidance algorithm, enabling the generated conflict-free flight plan to avoid risks while also taking into account time efficiency. Optionally, when the search space is large, optimization strategies such as jump point search can be used to accelerate the process of finding the optimal connected path.

[0083] See Figure 4This is a cost function analysis diagram of the cooperative obstacle avoidance algorithm, showing the changing trends of different cost factors with path nodes in the cooperative obstacle avoidance path planning of low-altitude UAVs. The core is the quantification of the cost composition of path search. In the cost structure changes, nodes 1-8 are dominated by heuristic costs, with the total cost decreasing slowly, and the algorithm prioritizing convergence towards the endpoint. Nodes 9-15 are dominated by actual costs, with the total cost changing synchronously with the actual cost, and the algorithm focusing on avoiding high-risk grids. The time penalty term reaches its peak at node 8, corresponding to the highest time cost of hovering / circling in the avoidance strategy, and then gradually decreases as the conflict eases. The smooth total cost curve indicates that the introduction of the time penalty term effectively balances path safety and time efficiency. The total cost curve flattens out after node 12, indicating that the path is close to the optimal solution, and the cost increment of subsequent nodes mainly comes from the accumulation of actual costs for the remaining distance.

[0084] In one embodiment of the present invention, the dynamic airspace grid management and service system for urban low-altitude environments, after generating a preliminary conflict-free flight plan, initiates an asynchronous real-time correction module for unforeseen circumstances during operation. The system monitors the status of the aircraft in real time. When a yaw event caused by a sudden gust of wind or equipment failure is detected, edge computing nodes are activated to analyze the local airspace area where the yaw event occurred. In a specific example scenario, a drone performing a power line inspection mission encounters a sudden gust of wind while flying over a riverbank area, causing its actual position to deviate from the predetermined flight path by more than 15 meters. The monitoring system detects this yaw event by comparing its real-time position with the predetermined grid sequence in a four-dimensional spacetime volume and immediately activates the edge computing nodes deployed nearby. The edge computing nodes receive the identity, current coordinates, speed, and predetermined flight path information of the yawed drone. In practical implementation, the system delineates a local grid region affected by the yaw event within a four-dimensional spatiotemporal volume. Using the current position of the yawing UAV as the center and a preset safety radius as the boundary, a spherical spatial region is defined. All four-dimensional spatiotemporal volume grid cells covered by this spherical spatial region are defined as local grid regions. For example, if the safety radius is set to 50 meters, this local grid region may involve hundreds of basic grid cells. Edge computing nodes re-evaluate the real-time weight vectors and occupancy status of existing grid cells within the local grid region. The re-evaluation process includes instantly acquiring the latest sensor data within the region, updating indicators such as wind speed and visibility risk values, obstacle occupancy rates, and recalculating the affected aircraft traffic saturation index, forming the latest situational snapshot of the local grid region.

[0085] In practice, based on the priority attributes of the tasks, a dynamic priority reordering is performed on all aircraft within the local grid area. This dynamic priority reordering considers factors such as the aircraft's task type, remaining battery power, and the value of its carried items, calculating a dynamic priority score for each task. Optionally, the dynamic priority score S... prioIt can be calculated using the following formula:

[0086]

[0087] Wherein: S prio This indicates a dynamic priority score, where U represents the task urgency, which is predefined by the task attributes, and E... used This indicates that the aircraft has consumed energy, E total Let U represent the total energy of the aircraft, V represent the value coefficient of the carried items, and α, β, γ are the weighting factors for each item, with α + β + γ = 1. Assume there are three aircraft within a local grid area: a power line inspection drone A that deviates from its flight path, a drone B performing aerial photography, and a drone C spraying pesticides. Using the above formula, drone A receives a higher U value due to its mission attribute (power line inspection involves public safety), and has sufficient remaining power; its dynamic priority score S is... prio (A) Highest; Drone C has a dynamic priority score of S due to the high value of the pesticides it carries and the urgency of the mission. prio (C) Secondly; Drone B's aerial photography mission has a lower urgency and item value, resulting in a dynamic priority score of S. prio (B) Lowest. In some embodiments, dynamic priority reordering is performed periodically or event-triggered to ensure that priorities reflect real-time conditions during the duration of a conflict.

[0088] In practical implementation, based on the results of dynamic priority reordering, the system issues heading correction commands to the aircraft corresponding to high-priority tasks, while executing waiting or detour logic on the aircraft corresponding to low-priority tasks to resolve local spatiotemporal conflicts. Taking the above scenario as an example, the system issues a command to the highest-priority power line inspection drone A to fine-tune its heading to quickly pass through the local grid area, a command to the next-highest-priority spraying drone C to decelerate and briefly hover in its current grid cell, and a command to the lowest-priority aerial photography drone B to detour along a preset backup route. It can be understood that by combining the rapid local decision-making of edge computing with dynamic priority scheduling, the system can asynchronously and in real-time correct conflicts caused by unforeseen circumstances.

[0089] In practical implementation, when the comprehensive airworthiness index exceeds the warning threshold, causing a grid cell to be marked as restricted or closed, the system simultaneously activates the status assessment module to perform a chain reaction status assessment on surrounding grid cells. The status assessment module uses the grid cell marked as restricted or closed as the center and retrieves several layers of neighboring grid cells. For example, using the closed grid cell G0 as the center, it retrieves all its directly adjacent grid cells (first-layer neighbors) and the grid cells one layer away (second-layer neighbors). The system extracts the current real-time weight vectors of these neighboring grid cells, focusing on analyzing the changing trends of the obstacle occupancy rate index and the aircraft traffic saturation index. In some embodiments, the changing trends are obtained by comparing the current index value with the average value within a historical sliding window or the index value at the previous moment. For example, if the closed grid cell G0 is due to a sudden high wind speed risk, the real-time weight vector of its first-layer neighboring grid cell G1 shows that its aircraft traffic saturation index has increased from 0.2 to 0.5 in the last 3 seconds, and data from the lidar indicates a slight upward trend in the obstacle occupancy rate index at the edge of G1. In practice, if the trend indicates that the risk is spreading to the surrounding area, the system automatically increases the comprehensive airworthiness index of neighboring grid cells and updates their status flags synchronously according to the new index value, realizing the gradient transmission and early warning of risk status. Taking the aforementioned G1 cell as an example, the system adjusts its comprehensive airworthiness index according to its upward trend through a predefined diffusion function, for example, increasing the comprehensive airworthiness index from 0.48 to 0.58 by 0.1. If the new comprehensive airworthiness index exceeds the warning threshold, the status of grid cell G1 is updated from "open" to "restricted," and an early warning is issued to potentially affected aircraft.

[0090] See Figure 5 This is a radar map of urban low-altitude airspace environmental risks, used to compare the differences in five core environmental indicators across three airspace categories, providing a clear basis for comprehensive airworthiness index calculation and airspace status management. High-risk areas are characterized by all indicators being high, especially in ground population density and wind speed and visibility, which are close to their maximum values. These areas should be marked as restricted / closed, allowing only high-priority missions to pass under strict control. Normal areas are characterized by all indicators being at a moderate level, with manageable risk, suitable as regular flight corridors. Continuous monitoring of wind speed, visibility, and aircraft traffic fluctuations is necessary to prevent sudden spikes in indicators that trigger risk warnings. Low-risk areas are characterized by all indicators being low, representing ideal open airspace capable of handling higher aircraft traffic. They can serve as priority detour areas in conflict avoidance situations, reducing overall flight costs.

[0091] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A dynamic airspace grid management and service system for urban low-altitude environments, characterized in that, include: The mesh partitioning module is used to reconstruct the three-dimensional space of the low-altitude airspace below 300 meters in the city based on the improved GeoSOT coding system. For densely built areas, a fine basic mesh of 5 meters by 5 meters by 2 meters is used for partitioning. For open airspace, the octree algorithm is used to merge them into cluster meshes to generate a basic airspace mesh set with non-uniform resolution. The semantic layer module is used to overlay logical labels of take-off and landing field buffer zones, emergency landing zones, and high-voltage line corridor isolation zones on the geometric attributes of the basic airspace grid set, establish a decoupled mapping relationship between geospatial entities and management attributes, and form an airspace grid layer with logical semantics. The spacetime volume generation module is used to introduce a time dimension component into each independent grid cell in the airspace grid layer with logical semantics, construct a four-dimensional spacetime volume containing spatial coordinates and time slices, and establish the four-dimensional spacetime volume as the smallest physical unit for subsequent aircraft trajectory reservation and conflict detection.

2. The dynamic airspace grid management and service system for urban low-altitude environments according to claim 1, characterized in that, After constructing the four-dimensional spacetime volume containing spatial coordinates and time slices, the process also includes: The dynamic weight calculation module is used to calculate the dynamic weight attributes of the grid cells within the four-dimensional spacetime body. A multidimensional weight vector model is constructed to define a real-time weight vector for each grid cell in the four-dimensional spatiotemporal body. The real-time weight vector includes the ground population dynamic density index obtained from base station signaling data parsing, the obstacle occupancy rate index obtained from lidar point cloud updates, the wind speed and visibility risk value index obtained from microscale meteorological simulation, the aircraft traffic saturation index of the currently reserved grid cell, and the electromagnetic environment interference level index. The fuzzy comprehensive evaluation method is used to calculate each index in the real-time weight vector to generate a comprehensive airworthiness index that characterizes the airworthiness status of the grid cell; The system compares the comprehensive airworthiness index with a preset warning threshold. When the comprehensive airworthiness index exceeds the warning threshold, the system automatically switches the state of the grid cell from open to restricted or closed, and writes the state change information into the message queue of the global scheduling bus.

3. The dynamic airspace grid management and service system for urban low-altitude environments according to claim 2, characterized in that, The construction methods of the multidimensional weight vector model include: The raw data associated with each grid cell in the four-dimensional spacetime body is collected. The raw data includes ground population density data from base station signaling, obstacle point cloud and image data from lidar and visual sensors, wind speed and visibility data from micro-weather stations and numerical simulation models, aircraft reservation flow data recorded from the global scheduling bus, and electromagnetic signal strength data from spectrum monitoring equipment. A standardized processing function is defined for each type of raw data to process the raw data into index values ​​with consistent dimensions and value ranges. The index values ​​include ground population dynamic density index, obstacle occupancy rate index, wind speed and visibility risk value index, aircraft flow saturation index, and electromagnetic environment interference level index. Each index value obtained from the processing is used as a dimension component in a vector and combined according to a predetermined dimensional order to form a real-time weight vector that characterizes the comprehensive environmental state of the grid cell under a specific time slice.

4. A dynamic airspace grid management and service system for urban low-altitude environments according to claim 3, characterized in that, Before calculating the various indicators in the real-time weight vector using the fuzzy comprehensive evaluation method, the following steps are also included: The data fusion module is used to fuse heterogeneous sensing data. Establish a heterogeneous data access channel to comprehensively receive sensing echo data, broadcast automatic correlation surveillance data, remote identification data, and visual data from 5G or 6G mobile communication technology base stations, and construct a raw observation dataset of low-altitude targets. Feature extraction and identification are performed on the observation records belonging to non-cooperative targets in the original observation dataset. Multi-source data cross-validation is used to fill the blind spots of single sensor detection, forming a highly reliable situational awareness base map covering the entire airspace. The high-reliability situational awareness base map is spatially registered with the four-dimensional spatiotemporal volume, so that each grid cell can be associated with a real-time environmental state description.

5. A dynamic airspace grid management and service system for urban low-altitude environments according to claim 4, characterized in that, After forming a highly reliable situational awareness base map covering the entire airspace, the following is also included: The airspace occupancy prediction module is used to predict future airspace occupancy. Retrieve the historical state vector of the UAV recorded in the high-reliability situational awareness base map for the previous several time steps. The historical state vector includes position, velocity, acceleration and attitude information. The extracted historical state vector is input into a pre-trained long short-term memory network model, and the model inference outputs the predicted trajectory coordinates of the UAV in the next few time steps. The predicted trajectory coordinates are discretized and mapped to a series of continuous grid occupancy sequences within the four-dimensional spacetime, thereby realizing the serialized expression of the flight situation in digital space. The grid occupancy sequence contains the airspace occupancy status at future moments.

6. A dynamic airspace grid management and service system for urban low-altitude environments according to claim 5, characterized in that, After realizing the serialized representation of the flight status in digital space, it also includes: The reservation verification module is used to perform four-dimensional trajectory reservation verification based on the grid occupancy sequence. Receive the expected flight path data submitted by the aircraft and parse the expected flight path data into a target grid sequence within the four-dimensional spacetime; Call the real-time weight vector and status flag corresponding to each grid cell in the target grid sequence, perform grid deduplication operation, and determine whether the target grid sequence has been locked by other tasks or whether its comprehensive airworthiness index exceeds the standard within a specified time period; If the verification result indicates the presence of conflicting or high-risk grids, the conflict information is fed back to the scheduling logic interface to trigger the subsequent cooperative avoidance algorithm; otherwise, the target grid sequence is marked as reserved.

7. A dynamic airspace grid management and service system for urban low-altitude environments according to claim 6, characterized in that, The subsequent cooperative avoidance algorithm is triggered, and the cooperative avoidance algorithm is executed through an improved heuristic search logic, specifically including: The real-time weight vectors of each grid cell in the four-dimensional spacetime volume are used as cost factors for path search to construct a cost function. The cost function is composed of the actual cost of reaching the current grid from the starting point and the heuristic cost of the estimated destination. A time-dimensional penalty term is introduced into the cost function. By adjusting the hovering time of the UAV in a specific grid cell or the fine-tuning coefficient of the flight speed, the cost difference brought about by different avoidance strategies is quantified. Search for an optimal connectivity path through the target grid sequence within the four-dimensional spacetime, avoiding grid cells that are restricted or closed during the search process, and generate a preliminary conflict-free flight plan.

8. A dynamic airspace grid management and service system for urban low-altitude environments according to claim 7, characterized in that, After generating the preliminary conflict-free flight plan, the process also includes: The real-time correction module is used to perform asynchronous real-time corrections for unexpected situations during operation. Real-time monitoring of the aircraft's status during operation; when a yaw event caused by a sudden gust of wind or equipment failure is detected, the edge computing node is activated to analyze the local airspace region where the yaw event occurred. In the four-dimensional spacetime volume, a local grid region affected by the yaw event is delineated, and the real-time weight vector and occupancy status of the existing grid cells in the local grid region are re-evaluated. Based on the priority attributes of the tasks, dynamic priority rearrangement is performed on all aircraft within the local grid area, and heading correction commands are issued to aircraft corresponding to high-priority tasks. Waiting or detour logic is executed on aircraft corresponding to low-priority tasks to resolve local spatiotemporal conflicts.

9. A dynamic airspace grid management and service system for urban low-altitude environments according to claim 8, characterized in that, The comprehensive airworthiness index is compared with a preset warning threshold. When the comprehensive airworthiness index exceeds the warning threshold, the method further includes: The state assessment module is used to perform cascading state assessments on surrounding grid cells. Centered on a grid cell marked as restricted or closed, retrieve several neighboring grid cells in adjacent layers; Extract the current real-time weight vector of the neighboring grid cells, and focus on analyzing the changing trends of the obstacle occupancy rate index and the aircraft traffic saturation index; If the trend indicates that the risk is spreading to the surrounding area, the comprehensive airworthiness index of the neighboring grid cell is automatically increased, and its status flag is updated synchronously according to the new index value, so as to realize the gradient transmission and early warning of the risk status.

10. A dynamic airspace grid management and service system for urban low-altitude environments according to claim 9, characterized in that, The step of inputting the extracted historical state vector into a pre-trained long short-term memory network model also includes quantifying the uncertainty of the predicted trajectory: Monte Carlo dropout technique is used to perform multiple forward propagations on the long short-term memory network model during the inference phase to generate multiple predicted trajectory branches; The multiple predicted trajectory branches are mapped to different grid occupancy sequences, and the probability of a high-density grid cluster appearing in each grid occupancy sequence is calculated. The probability information is packaged together with the original grid occupancy sequence and passed as a confidence parameter to the four-dimensional trajectory reservation verification module for risk selection when there are multiple candidate routes.

Images