A low-altitude airspace dynamic grid coding method and device, a terminal and a medium

Abstract

The application discloses a low-altitude airspace dynamic grid coding method and device, a terminal and a medium, and relates to the field of airspace management.The method acquires meteorological data in real time, wherein the meteorological data includes one or both of global meteorological data collected based on a micro weather station array and local meteorological data collected based on a UAV; the meteorological data is spatio-temporally aligned to determine to-be-predicted meteorological data; an improved long short-term memory neural network is used to predict the to-be-predicted meteorological data at intervals of a preset time to determine meteorological prediction values; the airspace is multi-level grid coded to determine local dynamic grids; and the local dynamic grids corresponding to the meteorological prediction values are dynamically adjusted according to the meteorological prediction values to determine target airspace grids. Therefore, the problem that the existing technology has limited processing capacity for nonlinear environmental factors and has coarse adjustment granularity and is difficult to meet high-precision flight control requirements can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of airspace management, and in particular to a low-altitude airspace dynamic grid coding method, device, terminal and medium. Background Art

[0002] With the increasing popularity of unmanned aerial vehicles (UAVs) in the low-altitude flight sector, the complexity and safety of airspace management are becoming increasingly prominent. Traditional airspace grid management technologies typically use static grid divisions (such as regular cube grids or latitude and longitude partitions). While computationally simple, these methods suffer from poor environmental adaptability, inefficient resource allocation, and delayed conflict prediction.

[0003] Dynamic grid coding offers advantages in improving airspace management efficiency, ensuring flight safety, and enhancing environmental adaptability. However, current dynamic grid coding, which often uses linear interpolation or threshold-triggered strategies, has limited ability to handle nonlinear environmental factors (such as rainfall and sudden visibility drops), and its adjustment granularity is coarse, making it difficult to meet the needs of high-precision flight control.

[0004] Therefore, the existing technology still needs to be improved and developed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a low-altitude airspace dynamic grid coding method, device, terminal and medium in response to the above-mentioned defects of the existing technology, aiming to solve the problem that the existing technology has limited processing ability for nonlinear environmental factors and the adjustment granularity is coarse, which is difficult to meet the needs of high-precision flight control.

[0006] The technical solutions adopted by the present invention to solve the problem are as follows:

[0007] In a first aspect, an embodiment of the present invention provides a low-altitude airspace dynamic grid coding method, wherein the method includes:

[0008] Acquiring meteorological data in real time, wherein the meteorological data includes one or both of global meteorological data collected by an array of micro-meteorological stations and local meteorological data collected by drones;

[0009] Performing spatiotemporal alignment on the meteorological data to determine meteorological data to be predicted;

[0010] At preset intervals, an improved long short-term memory neural network is used to make predictions based on the meteorological data to be predicted, and determine each meteorological prediction value;

[0011] Carry out multi-level grid coding on the airspace and determine each local dynamic grid;

[0012] The local dynamic grid corresponding to each of the meteorological forecast values ​​is dynamically adjusted according to each of the meteorological forecast values ​​to determine a target airspace grid.

[0013] In one implementation method, performing spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted includes:

[0014] Performing spatiotemporal alignment on the meteorological data using a Kriging interpolation method and a cubic spline interpolation method to determine spatiotemporal aligned meteorological data;

[0015] performing data extraction and calculation on the spatiotemporally aligned meteorological data based on a sliding window to determine instantaneous observation values ​​and statistics;

[0016] The meteorological data to be predicted is determined according to the instantaneous observation value and the statistical quantity.

[0017] In one implementation method, the improved long short-term memory neural network includes: a convolutional long short-term memory two-dimensional layer, a spatiotemporal attention layer, and a hierarchical long short-term memory neural network layer. The improved long short-term memory neural network is used to perform predictions based on the meteorological data to be predicted, and determine each meteorological prediction value, including:

[0018] Extracting features of the meteorological data to be predicted through the convolutional long short-term memory two-dimensional layer to obtain various spatiotemporal features;

[0019] Weighting each of the spatiotemporal features through the spatiotemporal attention layer;

[0020] The weighted spatiotemporal features are predicted through the hierarchical long short-term memory neural network layer to determine each of the meteorological forecast values.

[0021] In one implementation method, a multi-level grid encoding is performed on the airspace to determine each local dynamic grid, including:

[0022] Use the earth gridding technology to divide the airspace into grids and determine the reference grids for each area;

[0023] The local dynamic grid technology is combined with the quadtree and the octree to locally subdivide each of the regional reference grids, and the local dynamic grids corresponding to each of the regional reference grids are determined.

[0024] In one implementation method, dynamically adjusting the local dynamic grid corresponding to each of the weather forecast values ​​according to each of the weather forecast values ​​to determine the target airspace grid includes:

[0025] Obtaining a critical threshold corresponding to each of the meteorological forecast values, and determining a meteorological risk comprehensive index corresponding to each of the local dynamic grids based on each of the critical thresholds and each of the meteorological forecast values;

[0026] Inputting each of the meteorological risk comprehensive indices into an exponential grid size function to determine a target adjustment size corresponding to each of the local dynamic grids;

[0027] adjusting the corresponding local dynamic grid according to each target adjustment size, to determine the target airspace grid.

[0028] In an implementation method, the exponential grid size function is expressed as:

[0029]

[0030] wherein, is the target adjustment size, is the meteorological risk comprehensive index, is the maximum adjustable size, is the minimum adjustable size, is the adjustment sensitivity.

[0031] In an implementation method, the method further comprises:

[0032] acquiring actual meteorological values corresponding to each meteorological prediction value;

[0033] updating the improved long short-term memory neural network according to each meteorological prediction value and each actual meteorological value.

[0034] In a second aspect, the embodiments of the present application also provide a low-altitude airspace dynamic grid coding device, wherein the low-altitude airspace dynamic grid coding device comprises:

[0035] a meteorological data acquisition module, configured to acquire meteorological data in real time, wherein the meteorological data comprises one or both of global meteorological data collected based on a micro meteorological station array and local meteorological data collected based on a UAV;

[0036] a meteorological data alignment module, configured to perform space-time alignment on the meteorological data to determine to-be-predicted meteorological data;

[0037] a meteorological data prediction module, configured to use an improved long short-term memory neural network to predict each meteorological prediction value at intervals of a preset time based on the to-be-predicted meteorological data;

[0038] an airspace grid coding module, configured to perform multi-level grid coding on an airspace to determine each local dynamic grid;

[0039] a grid dynamic adjustment module, configured to dynamically adjust each local dynamic grid corresponding to each meteorological prediction value according to each meteorological prediction value, to determine a target airspace grid.

[0040] ​In a third aspect, an embodiment of the present invention further provides a terminal comprising a memory and one or more processors; the memory stores one or more programs; the program comprises instructions for executing any of the low-altitude airspace dynamic grid coding methods described above; and the processor is used to execute the program.

[0041] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium on which a plurality of instructions are stored, wherein the instructions are suitable for being loaded and executed by a processor to implement any of the above-mentioned low-altitude airspace dynamic grid coding methods.

[0042] Beneficial effects of the present invention: The embodiments of the present invention acquire meteorological data in real time, wherein the meteorological data includes one or both of global meteorological data collected by an array of micro-meteorological stations and local meteorological data collected by drones; perform spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted; use an improved long-short-term memory neural network to predict the meteorological data at preset intervals to determine each meteorological forecast value; perform multi-level grid encoding on the airspace to determine each local dynamic grid; and dynamically adjust the local dynamic grid corresponding to each meteorological forecast value based on each meteorological forecast value to determine the target airspace grid. Therefore, the present invention can effectively address the problem that existing technologies have limited processing capabilities for nonlinear environmental factors and coarse adjustment granularity, making it difficult to meet the requirements of high-precision flight control. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 It is a flow chart of the low-altitude airspace dynamic grid coding method provided by an embodiment of the present invention.

[0045] Figure 2 It is a flowchart of a specific embodiment of the low-altitude airspace dynamic grid coding method provided by an embodiment of the present invention.

[0046] Figure 3 This is a diagram of the improved long short-term memory neural network series structure provided by an embodiment of the present invention.

[0047] Figure 4 It is a schematic diagram of the internal modules of the low-altitude airspace dynamic grid coding device provided by an embodiment of the present invention.

[0048] Figure 5 This is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention discloses a method, device, terminal, and medium for low-altitude airspace dynamic grid coding. To make the objectives, technical solutions, and effects of the present invention more clear and explicit, the present invention is 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 present invention and are not intended to limit the present invention.

[0050] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the description of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any units and all combinations of one or more associated listed items.

[0051] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0052] With the increasing popularity of unmanned aerial vehicles (UAVs) in the low-altitude flight sector, the complexity and safety of airspace management are becoming increasingly prominent. Traditional airspace grid management technologies typically use static grid divisions (such as regular cube grids or latitude and longitude partitions). While computationally simple, these methods suffer from poor environmental adaptability, inefficient resource allocation, and delayed conflict prediction.

[0053] Dynamic grid coding offers advantages in improving airspace management efficiency, ensuring flight safety, and enhancing environmental adaptability. However, current dynamic grid coding, which often uses linear interpolation or threshold-triggered strategies, has limited ability to handle nonlinear environmental factors (such as rainfall and sudden visibility drops), and its adjustment granularity is coarse, making it difficult to meet the needs of high-precision flight control.

[0054] To address the aforementioned shortcomings of the prior art, the present invention provides a method for dynamic grid coding of low-altitude airspace. The method acquires meteorological data in real time, including one or both of global meteorological data collected by an array of micro-meteorological stations and local meteorological data collected by drones; performs spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted; uses an improved long-short-term memory neural network to predict the meteorological data at preset intervals to determine each meteorological forecast value; performs multi-level grid coding on the airspace to determine each local dynamic grid; and dynamically adjusts the local dynamic grid corresponding to each meteorological forecast value based on each meteorological forecast value to determine the target airspace grid. This method effectively addresses the existing problems of limited processing capabilities for nonlinear environmental factors and coarse adjustment granularity, making it difficult to meet the requirements of high-precision flight control.

[0055] Exemplary methods:

[0056] like Figure 1 As shown, the method includes:

[0057] Step S100: Acquire meteorological data in real time, wherein the meteorological data includes one or both of global meteorological data collected based on a micro-meteorological station array and local meteorological data collected based on a drone.

[0058] Specifically, meteorological data acquisition methods include micro-weather station arrays and drones. Micro-weather station arrays offer highly integrated and high-precision sensing capabilities, enabling the acquisition of highly accurate meteorological data. These arrays utilize military-grade sensors (such as the Vaisala WXT530) to simultaneously measure temperature, humidity, wind speed, and visibility. Temperature monitoring accuracy reaches ±0.3°C, and humidity error does not exceed ±2%RH (relative humidity). Combined with the second-level data transmission enabled by 5G IoT, these arrays establish a three-dimensional observation network with temporal and spatial resolution far exceeding that of traditional weather stations (one node per 5 square kilometers).

[0059] The micro-weather station array can provide the following: Full-factor weather monitoring: By integrating simultaneous measurement capabilities for multiple parameters, including temperature, humidity, wind speed, and visibility, the micro-weather station array establishes a comprehensive low-altitude environmental monitoring system. Its temperature measurement accuracy of ±0.3°C enables precise sensing of thermodynamic environmental changes, while its humidity tolerance of ±2%RH ensures accurate reflection of atmospheric water vapor conditions. Its wind speed range of 0-60 m / s covers observations from calm to strong winds, while its visibility monitoring range of 10-50 km effectively captures low-altitude visibility changes. High-temporal and spatial resolution networking: Its high-temporal and spatial resolution enables the micro-weather station array to be deployed at intervals of one station every five square kilometers, forming a cellular observation network. Leveraging 5G IoT technology, data transmission is achieved within seconds, with latency under 800 ms, and can analyze microclimate characteristics in areas as small as 100 m x 100 m. System reliability assurance: System reliability assurance measures include multi-sensor redundant design, data self-checking mechanisms, and lightning and corrosion-resistant structures, ensuring stable operation in extreme environments ranging from -40°C to +70°C. Airspace management support: As the core support for airspace management, the micro-weather station array not only provides high-quality training data for the improved long-short-term memory neural network, but also drives the dynamic adjustment of the airspace grid.

[0060] The drone-based collection method mainly uses drone sensors to collect local meteorological data during the movement of the drone, performs preliminary filtering processing through edge computing nodes, and provides real-time feedback of meteorological data in the area near the drone.

[0061] This embodiment collects meteorological data in real time, so that the meteorological values ​​of future time nodes can be predicted in real time based on various meteorological data, so as to make real-time dynamic adjustments to the airspace grid according to the predicted meteorological values, thereby improving the ability of the airspace grid to respond to environmental changes or emergencies.

[0062] Step S200: performing spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted.

[0063] Simply put, meteorological data is collected based on a micro-weather station array and drone sensors. The micro-weather station array includes a variety of different sensors. The meteorological data collected by different sensors are discrete and asynchronous. It is necessary to align the meteorological data in time and space to solve the problem of inconsistency in time and space dimensions of multi-source heterogeneous data (such as ground-based meteorological station data, drone mobile observation data, etc.). It provides highly consistent and high-resolution input data for the subsequent improved long-short-term memory neural network, so as to enhance the improved long-short-term memory neural network's ability to capture local meteorological mutations (such as low-level wind shear and sudden drop in visibility), while reducing the prediction error caused by data time and space deviation.

[0064] In one implementation, performing spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted includes:

[0065] Step S201: performing spatiotemporal alignment on the meteorological data using the Kriging interpolation method and the cubic spline interpolation method to determine spatiotemporal aligned meteorological data;

[0066] Step S202: extracting and calculating the spatiotemporally aligned meteorological data based on a sliding window to determine instantaneous observation values ​​and statistics;

[0067] Step S203: Determine the meteorological data to be predicted based on the instantaneous observation value and the statistics.

[0068] Specifically, the meteorological data is temporally and spatially aligned to obtain the temporally and spatially aligned meteorological data, which solves the temporal and spatial fragmentation problem of traditional meteorological data. For the two types of heterogeneous meteorological data, namely micro-weather station array data and drone sensor data, due to their differences in temporal and spatial resolution, the Kriging interpolation method is used to establish a continuous spatial field. The formula is as follows:

[0069] ;

[0070] in, is the grid point position to be interpolated The location of the weather station, is the number of weather stations, For weather stations Meteorological data (measured values ​​of meteorological elements), is the Kriging weight, The meteorological data after spatial alignment.

[0071] The time alignment is performed by using cubic spline interpolation, and the formula is as follows:

[0072] ;

[0073] is the target time point (aligned timestamp, such as 5-minute interval), It is The original observation time point, , , , are cubic spline coefficients respectively, and are solved by boundary conditions. The meteorological data are time-aligned.

[0074] Construct a sliding window and extract and calculate data based on the spatiotemporal aligned meteorological data of the sliding window, instantaneous observations and statistics: slide the window on the spatiotemporal aligned meteorological data in the form of a spatiotemporal cube, with the time span being the first preset time (for example, 1 hour, 12 time steps) and the spatial range being the first preset range (for example, 1 km×1 km, 10×10 grids). Each window contains instantaneous observations (meteorological elements such as temperature, humidity, wind speed, visibility, etc.), statistics (statistics calculate the mean and change trend of each meteorological data, including moving averages, etc.), spatial gradients (reflecting the position of the meteorological front to facilitate spatial alignment of meteorological data), etc.

[0075] After data preprocessing, the space-time cube will output a 5-dimensional (5D) tensor as the input of the improved long short-term memory neural network. The instantaneous observation value is , statistics include moving average and changing trend, among which the moving average for:

[0076] ,

[0077] in, represents the number of valid data points in the sliding window (the amount of valid spatiotemporally aligned meteorological data), Represents the current time step index, such as represents the 15th 5-minute interval, which is 75 minutes. represents the starting time step of the sliding window, if , which means going back 5 time steps from the current moment, express The instantaneous observation value at time .

[0078] Changing trends for:

[0079] ,

[0080] in, Indicates the rate of change under the time unit step (the rate of change of the instantaneous observation value), , the time unit is 5 minutes, then it can be said that the temperature is The rate of change in time step units is 0.5.

[0081] spatial gradient for:

[0082] ,

[0083] in, Indicates the differential change of meteorological elements (differential variable of instantaneous observation value), a spatial interval representing an east-west (longitude) direction, a spatial interval representing a north-south (latitude) direction.

[0084] The sliding window for spatio-temporal alignment can provide high-consistency and low-uncertainty input data for the improved long short-term memory neural network, and significantly improve the prediction reliability.

[0085] For example, meteorological data is input into the spatio-temporal alignment sliding window module, the discrete observation points are unified to a standard grid of 100m x 100m by using the Kriging interpolation method, and are aligned to a time resolution of 5 minutes by using the cubic spline interpolation, to generate spatio-temporally consistent meteorological data to be predicted.

[0086] In step S300, the improved long short-term memory neural network is used to predict according to the meteorological data to be predicted, and each meteorological prediction value is determined.

[0087] In short, considering the change speed of meteorological data and the calculation resource loss of network model prediction and grid adjustment, a preset time (such as 0.5 hours) is set, and the improved long short-term memory neural network is used to predict according to the meteorological data to be predicted every interval of the preset time to obtain each meteorological prediction value. In one implementation, the length of the preset time can be adjusted according to the last meteorological prediction value. For example, if the meteorological prediction value shows that the weather in the future time period is poor, the preset time is shortened, the number of adjustments of the airspace grid is increased, and the sudden situation caused by the bad weather is dealt with. If the meteorological prediction value shows that the weather in the future time period is good, the preset time is extended, the number of predictions and the number of adjustments of the airspace grid are reduced, and the loss of calculation resources is reduced.

[0088] In one implementation, the improved long short-term memory neural network includes a convolutional long short-term memory two-dimensional layer, a spatio-temporal attention layer, and a hierarchical long short-term memory neural network layer. The improved long short-term memory neural network is used to predict according to the meteorological data to be predicted, and each meteorological prediction value is determined, including:

[0089] In step S301, the convolutional long short-term memory two-dimensional layer is used to extract features from the meteorological data to be predicted, and each spatio-temporal feature is obtained.

[0090] In step S302, the spatio-temporal attention layer is used to weight each spatio-temporal feature.

[0091] In step S303, the hierarchical long short-term memory neural network layer is used to predict the weighted spatio-temporal features, and each meteorological prediction value is determined.

[0092] Specifically, the improved long short-term memory neural network receives the meteorological data to be predicted after spatiotemporal alignment (5-dimensional tensor: number of samples, time step, grid rows, grid columns, number of features), where the feature data include instantaneous values ​​and statistics of temperature, humidity, wind speed, and visibility.

[0093] Since the raw meteorological data in the forecasted meteorological data have orders of magnitude differences (such as temperature range -10~40℃, humidity 0~100%, wind speed 0~60m / s), the forecasted meteorological data is first normalized before feature extraction:

[0094] By standardizing features to unify dimensions, we can eliminate feature scale differences and prevent features with large values ​​(such as wind speed) from dominating model training. The feature standardization formula is as follows:

[0095] ,

[0096] in, For the meteorological data to be predicted, is the mean of the meteorological data to be predicted in the training set, is the standard deviation of the meteorological data to be predicted in the training set, The meteorological data to be predicted is standardized, which not only ensures that all input channels of the improved long short-term memory neural network have equal importance, but also improves numerical stability and prevents the problem of gradient explosion.

[0097] Based on the improved long short-term memory neural network, the forecast is made according to the meteorological data to be predicted, and the forecast values ​​of each meteorological data are determined, including: Figure 3 As shown in the figure, the local spatiotemporal feature extraction and temporal dynamic modeling are first performed through the convolutional long short-term memory 2D layer (ConvLSTM2D layer). Specifically, spatial feature extraction is performed by scanning the 50×50 grid data with a 3×3 convolution kernel to capture local meteorological patterns (such as cyclone center and temperature front); temporal dynamic modeling is performed through a gating mechanism (forget gate , input gate , output gate ) controls the flow of information and learns the short-term evolution of meteorological elements in the forecasted data. By extracting local spatiotemporal features, the raw meteorological data can be converted into high-level features with spatiotemporal correlations, providing a foundation for the subsequent attention mechanism.

[0098] Next, we construct a spatiotemporal attention layer. This layer is a modular component that enhances traditional long-short-term memory neural networks through a cascaded approach. Its core purpose is to weight and focus the input data (i.e., various spatiotemporal features) to achieve the goal of focusing on key areas and moments. The spatial attention formula for the spatiotemporal attention layer is as follows:

[0099] ,

[0100] in, For spatial attention, is the spatial attention weight matrix, Hidden With the spatiotemporal characteristics of the current input The splicing, The time step is used to highlight the areas with great influence on the weather forecast value (such as the area with sudden drop in visibility near the airport runway).

[0101] The temporal attention formula is as follows:

[0102] ,

[0103] in, For time attention, is the temporal attention weight vector, is a gated logic unit used to weight important historical moments (such as the humidity accumulation stage before rainfall begins), Represents the matrix composed of all hidden states from the beginning to the current moment, Indicates the ending time step.

[0104] The spatiotemporal attention layer weights data from key regions and time periods, reducing noise from irrelevant regions and moments and improving model efficiency. During the forecasting process, the spatiotemporal attention layer automatically focuses on high-risk areas (such as areas with plummeting visibility around airports) and key time points (such as one hour before severe convective weather).

[0105] A hierarchical long-short-term memory (LSTM) neural network layer is constructed to model long-term dependencies and update state. This layer consists of the following: a first layer (128 LSTM units, each containing a memory cell (storing long-term state) and three gating mechanisms: a forget gate, an input gate, and an output gate). This layer captures short- to medium-term changes (30 minutes to an hour) and outputs a complete time series, preserving detailed information. A second layer (64 LSTM units) models long-term trends (1 to 2 hours), extracts global features, and outputs the final weather forecast.

[0106] The state update mechanism of the hierarchical LSTM neural network layer is as follows:

[0107] ,

[0108] in, is the state transformation weight matrix, For the Forgotten Gate, is the input gate, represents element-wise multiplication, is the time step, for hidden output of the time step (size 128), for The input features corresponding to the time step (64 dimensions, output from the spatiotemporal attention layer), are obtained through the cell state Selectively memorize key meteorological evolution patterns. By separating spatial convolution (convolutional long short-term memory 2D layer) and temporal modeling (hierarchical long short-term memory neural network layer), this approach reduces computational overhead by 70% compared to traditional 3D convolutional neural networks (3D-CNNs) while maintaining spatiotemporal modeling capabilities. This also resolves the conflict between short-term fluctuations and long-term trends in weather forecasts.

[0109] Step S400: Perform multi-level grid coding on the airspace to determine each local dynamic grid.

[0110] Simply put, the airspace is coded with multi-level grids to build a hybrid grid system of "global coarse granularity + local fine granularity" to achieve full coverage of airspace management from macro-planning to micro-adjustment.

[0111] In one implementation, multi-level grid coding is performed on the airspace to determine each local dynamic grid, including:

[0112] Step S401: Using the earth gridding technology to divide the airspace into grids and determine the reference grids for each area;

[0113] Step S402: using a local dynamic grid technology combined with a quadtree and an octree to locally subdivide each of the regional reference grids, and determining each of the local dynamic grids corresponding to each of the regional reference grids.

[0114] Specifically, a double-layer grid system of earth subdivision grid and local dynamic grid that complies with the earth space grid coding rules is constructed.

[0115] At the global level, to accommodate the spatial extent of urban airspace, a 200m × 200m × 100m regional reference grid is established using Earth gridding technology as the basic coordinate system for airspace management. Globally unique identification is achieved through a variant encoding of Geohash (a system for encoding geographic coordinates as short strings). Each regional reference grid cell carries macroscopic airspace attributes, such as airspace type and basic meteorological conditions, providing a global path planning benchmark for aircraft. The regional reference network encoding formula is as follows:

[0116] ,

[0117] is the coordinates (latitude and longitude) of the origin of the southwest corner of the region, is the altitude, is the encoding connector, is the regional reference grid coordinate.

[0118] At the local level, dynamic mesh refinement is achieved based on the quadtree / octree data structure combined with local dynamic mesh technology to obtain a local dynamic mesh. The local dynamic mesh can support a minimum resolution of 5m×5m×5m, and the mesh is adaptively adjusted through real-time environmental perception. The local dynamic mesh formula is as follows:

[0119] ;

[0120] The coding structure is (parent area reference grid code ( ) Quadtree Path ( ):Octtree level( )),in, It is a hierarchical inheritance identifier in the local dynamic grid encoding, indicating the parent-child relationship between the local dynamic grid and the regional reference grid it describes. Its structural significance lies in separating the subdivision path of the parent grid of the earth subdivision grid and the local refined grid, forming a "parent→child" topological expression.

[0121] Step S500: Dynamically adjust the local dynamic grid corresponding to each of the meteorological forecast values ​​according to each of the meteorological forecast values ​​to determine a target airspace grid.

[0122] In simple terms, the improved long short-term memory neural network outputs high-precision weather forecast values ​​(including key indicators such as visibility, wind speed, and precipitation probability) for the next two hours at a preset interval (every 5 minutes). These weather forecast values ​​directly drive the dynamic adjustment decisions of the corresponding local dynamic grid. For example, Figure 2 As shown, visibility warning: When the improved long-short-term memory (LSTM) neural network predicts that visibility in a certain area will fall below 800 meters in the next 30 minutes, the local dynamic mesh in that area is refined from the standard 100-meter resolution to 30 meters. Furthermore, if the visibility forecast drops further below 500 meters, the mesh is upgraded to an ultra-high-precision mesh of less than 30 meters, and a 200-meter buffer zone is extended along the wind direction. The buffer zone density increases with visibility deterioration. Wind speed warning: The low-altitude mesh layer (0-120 meters) is disabled in the path of strong winds. Rainfall warning: If rainfall exceeds 20 mm / h, the drainage buffer zone within the airspace mesh is activated. Based on the above adjustments to the local dynamic mesh, a detour path is generated for the aircraft. The adjusted local dynamic mesh is fed back to the improved long-short-term memory (LSTM) neural network in real time, and the weight parameters are updated through online learning.

[0123] In one implementation, dynamically adjusting the local dynamic grid corresponding to each of the weather forecast values ​​according to each of the weather forecast values ​​to determine the target airspace grid includes:

[0124] Step S501: Acquire a critical threshold corresponding to each of the meteorological forecast values, and determine a meteorological risk comprehensive index corresponding to each of the local dynamic grids according to each of the critical thresholds and each of the meteorological forecast values;

[0125] Step S502: inputting each of the meteorological risk comprehensive indices into an exponential grid size function to determine a target adjustment size corresponding to each of the local dynamic grids;

[0126] Step S503: Adjust the local dynamic grid corresponding to each target adjustment size to determine the target spatial grid.

[0127] Specifically, in order to adjust the local dynamic grid more flexibly, a joint decision-making adjustment strategy based on multiple meteorological factors is adopted, and a comprehensive meteorological risk index is proposed. (Meteorological Risk Index) drives local dynamic mesh adjustments:

[0128] ;

[0129] in, is the visibility forecast value (m), is the critical visibility threshold (e.g. 500m), is the predicted wind speed value (m / s), is the critical wind speed threshold (e.g. 10m / s), is the predicted rainfall intensity (mm / h), is the critical threshold of rainfall intensity (e.g. 20 mm / h), 、 、 is the weight coefficient (the default values ​​are 0.5, 0.3, and 0.2, and the optimal solution can be determined through a large number of trainings). The larger the value of the meteorological risk comprehensive index, the higher the risk, which means that the airspace grid needs to be divided more accurately.

[0130] The comprehensive meteorological risk index will affect the size of the local dynamic grid. In order to avoid a large increase or decrease in the grid size, an exponential grid size function is proposed:

[0131] ,

[0132] in, Adjust the size for the target, is the comprehensive meteorological risk index, is the maximum adjustable size (such as 100m), The minimum adjustable size (such as 5m), To adjust the sensitivity.

[0133] The meteorological risk comprehensive index corresponding to each local dynamic grid is calculated according to each critical threshold and the meteorological forecast value corresponding to each local dynamic grid. The meteorological risk comprehensive index corresponding to each local dynamic grid is input into the exponential grid size function to obtain the target adjustment size corresponding to each local dynamic grid. Based on each target adjustment size, the local dynamic grid corresponding to each target adjustment size is adjusted to obtain the target airspace grid.

[0134] For example, suppose that an airspace management system needs to dynamically adjust the grid size to cope with meteorological risks, and the parameters are set as follows: (maximum grid size without risk); (minimum grid size at extreme risk); (Adjust sensitivity).

[0135] Assuming a low risk scenario ( ):

[0136] ,

[0137] When the risk is low, the grid size is close to the maximum value (coarse grain) to reduce computing resource usage.

[0138] Medium risk situations ( ):

[0139] ,

[0140] As the risk increases, the grid size is reduced to increase the resolution and capture local meteorological changes.

[0141] High-risk situations ( ):

[0142] ,

[0143] When the risk is extremely high, the grid size approaches the minimum value (fine granularity) to ensure accurate monitoring.

[0144] In this embodiment, the adjustment of the local dynamic grid is not a simple threshold trigger, but a multi-dimensional decision-making process based on the prediction results of the improved long short-term memory neural network.

[0145] In one implementation, the probabilistic predictions output by the improved LSTM neural network (e.g., "there is a 75% probability of visibility <500m") are fed into a risk assessment model to calculate the expected benefits of different grid adjustment strategies. For example, in areas densely populated with logistics drones, even a 30% probability of rain could trigger preventative grid refinement. Secondly, the spatiotemporal features extracted by the improved LSTM neural network are analyzed. If the forecast indicates a linear trend of severe convective weather, dynamic adjustment corridors are automatically generated along the movement path, enabling precise deployment of grid resources.

[0146] In one implementation, the method further includes:

[0147] Step H10: Obtaining actual meteorological values ​​corresponding to the meteorological forecast values;

[0148] Step H20: updating the improved long short-term memory neural network according to the weather forecast values ​​and the actual weather values.

[0149] Simply put, the actual meteorological values ​​of the adjusted local dynamic grid (the meteorological values ​​generated by the meteorological data collected by each sensor) are obtained, and the actual meteorological values ​​and meteorological forecast values ​​(such as the actual visibility change and the forecast deviation) are stored in the training library. The improved long-short-term memory neural network is re-output as a feedback signal, and the prediction accuracy is continuously optimized through online learning, forming an enhanced learning cycle of "observation-prediction-adjustment-verification".

[0150] Based on the above embodiment, the present invention also provides a low-altitude airspace dynamic grid coding device, such as Figure 4 As shown, the device includes:

[0151] Meteorological data acquisition module 01 is used to acquire meteorological data in real time, wherein the meteorological data includes one or both of global meteorological data collected based on a micro-meteorological station array and local meteorological data collected based on drones;

[0152] The meteorological data alignment module 02 is used to perform spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted;

[0153] The meteorological data prediction module 03 is used to predict the meteorological data to be predicted using an improved long short-term memory neural network at preset intervals to determine each meteorological prediction value;

[0154] The spatial domain grid coding module 04 is used to perform multi-level grid coding on the spatial domain and determine each local dynamic grid;

[0155] The grid dynamic adjustment module 05 is used to dynamically adjust the local dynamic grid corresponding to each meteorological forecast value according to each meteorological forecast value to determine the target airspace grid.

[0156] Based on the above embodiments, the application further provides a terminal, a principle block diagram of which can be shown as follows. Figure 5 The terminal includes a processor, a memory, a network interface, a display screen connected through a system bus. The processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the terminal is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement the low-altitude airspace dynamic grid encoding method. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0157] Those skilled in the art can understand that, Figure 5 The principle block diagram shown in the above embodiments is only a block diagram of part of the structure related to the application scheme, and does not constitute a limitation on the terminal to which the application scheme is applied. The specific terminal can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0158] In an implementation manner, the memory of the terminal stores more than one program, and is configured to execute the more than one program by more than one processor, which contains instructions for performing the low-altitude airspace dynamic grid encoding method.

[0159] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media used in the various embodiments provided herein may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0160] In summary, the present invention discloses a method, device, terminal, and medium for dynamic grid coding of low-altitude airspace. The method acquires meteorological data in real time, wherein the meteorological data includes one or both of global meteorological data collected by an array of micro-meteorological stations and local meteorological data collected by drones; performs spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted; uses an improved long-short-term memory neural network to predict the meteorological data at preset intervals to determine each meteorological forecast value; performs multi-level grid coding on the airspace to determine each local dynamic grid; and dynamically adjusts the local dynamic grid corresponding to each meteorological forecast value based on each meteorological forecast value to determine the target airspace grid. Therefore, the method can effectively solve the problem that the existing technology has limited processing capabilities for nonlinear environmental factors and coarse adjustment granularity, making it difficult to meet the needs of high-precision flight control.

[0161] It should be understood that the application of the present invention is not limited to the above examples. For those skilled in the art, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A low-altitude airspace dynamic grid coding method, characterized in that: The method comprises: Acquiring meteorological data in real time, wherein the meteorological data includes one or both of global meteorological data collected by an array of micro-meteorological stations and local meteorological data collected by drones; Performing spatiotemporal alignment on the meteorological data to determine meteorological data to be predicted; At preset intervals, an improved long short-term memory neural network is used to make predictions based on the meteorological data to be predicted, and determine each meteorological prediction value; Carry out multi-level grid coding on the airspace and determine each local dynamic grid; Dynamically adjusting the local dynamic grid corresponding to each of the meteorological forecast values ​​according to each of the meteorological forecast values ​​to determine a target airspace grid; Dynamically adjusting the local dynamic grid corresponding to each of the meteorological forecast values ​​according to each of the meteorological forecast values ​​to determine a target airspace grid includes: Obtaining a critical threshold corresponding to each of the meteorological forecast values, and determining a meteorological risk comprehensive index corresponding to each of the local dynamic grids based on each of the critical thresholds and each of the meteorological forecast values; Inputting each of the meteorological risk comprehensive indices into an exponential grid size function to determine a target adjustment size corresponding to each of the local dynamic grids; Adjust the local dynamic grid corresponding to each target adjustment size to determine the target spatial grid; The exponential grid size function is expressed as: ， in, Adjust the size for the target, is the comprehensive meteorological risk index, is the maximum adjustable size, is the minimum adjustable size, To adjust the sensitivity.

2. The low-altitude airspace dynamic grid coding method according to claim 1, characterized in that: Performing spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted includes: Performing spatiotemporal alignment on the meteorological data using a Kriging interpolation method and a cubic spline interpolation method to determine spatiotemporal aligned meteorological data; performing data extraction and calculation on the spatiotemporally aligned meteorological data based on a sliding window to determine instantaneous observation values ​​and statistics; The meteorological data to be predicted is determined according to the instantaneous observation value and the statistical quantity.

3. The low-altitude airspace dynamic grid coding method according to claim 1, characterized in that: The improved long short-term memory neural network includes: a convolutional long short-term memory two-dimensional layer, a spatiotemporal attention layer, and a hierarchical long short-term memory neural network layer. The improved long short-term memory neural network is used to perform predictions based on the meteorological data to be predicted, and determine each meteorological prediction value, including: Extracting features of the meteorological data to be predicted through the convolutional long short-term memory two-dimensional layer to obtain various spatiotemporal features; Weighting each of the spatiotemporal features through the spatiotemporal attention layer; The weighted spatiotemporal features are predicted through the hierarchical long short-term memory neural network layer to determine each of the meteorological forecast values.

4. The low-altitude airspace dynamic grid coding method according to claim 1, characterized in that: Perform multi-level grid coding on the airspace and determine each local dynamic grid, including: Use the earth gridding technology to divide the airspace into grids and determine the reference grids for each area; The local dynamic grid technology is combined with the quadtree and the octree to locally subdivide each of the regional reference grids, and the local dynamic grids corresponding to each of the regional reference grids are determined.

5. The low-altitude airspace dynamic grid coding method according to claim 1, characterized in that: The method further comprises: Obtaining actual meteorological values ​​corresponding to each of the meteorological forecast values; The improved long short-term memory neural network is updated according to each of the meteorological forecast values ​​and each of the actual meteorological values.

6. A low-altitude airspace dynamic grid coding device, characterized in that: The device comprises: A meteorological data acquisition module, configured to acquire meteorological data in real time, wherein the meteorological data includes one or both of global meteorological data collected by an array of micro-meteorological stations and local meteorological data collected by drones; A meteorological data alignment module is used to perform spatiotemporal alignment on the meteorological data to determine the meteorological data to be predicted; A meteorological data prediction module is used to predict the meteorological data to be predicted using an improved long short-term memory neural network at preset intervals to determine each meteorological prediction value; The spatial grid coding module is used to perform multi-level grid coding on the spatial domain and determine each local dynamic grid; A grid dynamic adjustment module is used to dynamically adjust the local dynamic grid corresponding to each meteorological forecast value according to each meteorological forecast value to determine a target airspace grid; Dynamically adjusting the local dynamic grid corresponding to each of the meteorological forecast values ​​according to each of the meteorological forecast values ​​to determine a target airspace grid includes: Obtaining a critical threshold corresponding to each of the meteorological forecast values, and determining a meteorological risk comprehensive index corresponding to each of the local dynamic grids based on each of the critical thresholds and each of the meteorological forecast values; Inputting each of the meteorological risk comprehensive indices into an exponential grid size function to determine a target adjustment size corresponding to each of the local dynamic grids; Adjust the local dynamic grid corresponding to each target adjustment size to determine the target spatial grid; The exponential grid size function is expressed as: ， in, Adjust the size for the target, is the comprehensive meteorological risk index, is the maximum adjustable size, is the minimum adjustable size, To adjust the sensitivity.

7. A terminal, characterized in that: The terminal includes a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the low-altitude airspace dynamic grid coding method as described in any one of claims 1-5; and the processor is used to execute the program.

8. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are loaded and executed by the processor to implement the steps of the low-altitude airspace dynamic grid coding method described in any one of claims 1-5 above.

Images