Unmanned aerial vehicle navigation method and system based on spatial three-dimensional grid coding

Abstract

The application provides a UAV navigation method and system based on airspace three-dimensional grid coding, and relates to the technical field of path planning. The method comprises the following steps: reading the UAV request flight information of a target airspace in a preset time zone; performing three-dimensional grid size adaptation analysis on a plurality of UAV attribute data sets respectively to obtain a plurality of adapted grid sizes; performing simultaneous zone collision risk prediction on the plurality of UAV attribute data sets to output a predicted collision risk coefficient, and compensating the predicted collision risk coefficient; performing grid size optimization based on the plurality of adapted grid sizes and the compensated collision risk coefficient to obtain an optimal grid size; and performing airspace three-dimensional grid coding in the preset time zone according to the optimal grid size to perform UAV navigation control. The technical problem that the uniform three-dimensional grid size of the airspace in the prior art is difficult to adapt to diversified operation requirements and is prone to increase collision risk or reduce operation efficiency is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of path planning, in particular to a UAV navigation method and system based on airspace three-dimensional grid coding. BACKGROUND

[0002] The aviation control agency can set a unified three-dimensional grid size for a specific airspace, and require all UAVs entering the area to follow the same set of grid parameters for path planning and airspace interaction. The main purpose of this approach is to ensure consistency in airspace management, simplify flight coordination, and improve flight safety.

[0003] However, with the diversification of UAV application scenarios, the number of multi-type UAVs working together in the same airspace has increased significantly. The differences in size, speed, and task type of different UAVs are significant, and a unified grid size is difficult to adapt to diversified operation requirements, which can easily increase the risk of collision or reduce operational efficiency.

[0004] Therefore, there is an urgent need for a UAV navigation method that can dynamically adjust the three-dimensional grid size according to the UAV attributes and airspace environment to achieve a dynamic balance between airspace utilization efficiency and flight safety. SUMMARY

[0005] The present application provides a UAV navigation method and system based on airspace three-dimensional grid coding to solve the technical problems of the prior art that a unified three-dimensional grid size for airspace is difficult to adapt to diversified operation requirements and can easily increase the risk of collision or reduce operational efficiency.

[0006] The technical solution of the present application to solve the above technical problems is as follows:

[0007] In a first aspect, the present application provides a UAV navigation method based on airspace three-dimensional grid coding, comprising:

[0008] Reading the UAV request flight information of the target airspace in a preset time zone, wherein the UAV request flight information includes a plurality of UAV attribute data sets;

[0009] Performing three-dimensional grid size adaptation analysis according to the plurality of UAV attribute data sets to obtain a plurality of adapted grid sizes;

[0010] Performing simultaneous zone collision risk prediction according to the plurality of UAV attribute data sets, outputting a predicted collision risk coefficient, and compensating the predicted collision risk coefficient based on the predicted dynamic obstacle information and predicted weather conditions in the preset time zone to obtain a compensated collision risk coefficient;

[0011] Performing grid size optimization based on the plurality of adapted grid sizes and the compensated collision risk coefficient to obtain an optimal grid size;

[0012] Perform the airspace three-dimensional grid coding in the preset time zone according to the optimal grid size, and perform the unmanned aerial vehicle navigation control.

[0013] In the second aspect, the application provides an unmanned aerial vehicle navigation system based on airspace three-dimensional grid coding, comprising:

[0014] An information reading module is configured to read unmanned aerial vehicle request flight information of a target airspace in a preset time zone, wherein the unmanned aerial vehicle request flight information comprises a plurality of unmanned aerial vehicle attribute data sets;

[0015] A grid size analysis module is configured to perform three-dimensional grid size adaptive analysis according to the plurality of unmanned aerial vehicle attribute data sets respectively, and obtain a plurality of adaptive grid sizes;

[0016] A risk prediction module is configured to perform collision risk prediction in the same time zone according to the plurality of unmanned aerial vehicle attribute data sets, output a predicted collision risk coefficient, and compensate the predicted collision risk coefficient based on predicted dynamic obstacle information and predicted meteorological conditions in the preset time zone to obtain a compensated collision risk coefficient;

[0017] A parameter optimization module is configured to perform grid size optimization based on the plurality of adaptive grid sizes and the compensated collision risk coefficient, and obtain an optimal grid size;

[0018] An output execution module is configured to perform airspace three-dimensional grid coding in the preset time zone according to the optimal grid size, and perform the unmanned aerial vehicle navigation control.

[0019] The application has the following beneficial effects:

[0020] Compared with the prior art, firstly, the unmanned aerial vehicle request flight information in the preset time zone of the target airspace is read, which provides a reliable data basis for subsequent grid size adaptation analysis, collision risk prediction, grid optimization and other steps. Secondly, three-dimensional grid size adaptation analysis is performed according to a plurality of unmanned aerial vehicle attribute data sets to obtain a plurality of adaptive grid sizes, and a grid size adaptation analyzer is trained through historical flight data to match personalized adaptive grid sizes for a plurality of unmanned aerial vehicle attribute data sets, thereby providing a reference for subsequent global grid optimization. Thirdly, the collision risk prediction in the same zone is performed according to a plurality of unmanned aerial vehicle attribute data sets, and the predicted collision risk coefficient is output. The predicted collision risk coefficient is compensated based on the predicted dynamic obstacle information and the predicted meteorological conditions in the preset time zone to obtain a compensated collision risk coefficient. The compensated collision risk coefficient takes into account both the characteristics of the unmanned aerial vehicle itself and the external environmental impact, thereby providing a key basis for the safety optimization of the three-dimensional grid size. Further, grid size optimization is performed based on the plurality of adaptive grid sizes and the compensated collision risk coefficient to obtain an optimal grid size. The optimal grid size is obtained through iterative optimization and dynamic adjustment of the size of the compensation collision risk coefficient. The optimal grid size can maximize the personalized requirements of different unmanned aerial vehicles, and can reduce the amount of calculation by maintaining a larger grid size in a high-risk scenario to ensure the real-time obstacle avoidance, and can appropriately reduce the grid size in a low-risk scenario to improve the spatial resolution and enhance the navigation accuracy, thereby achieving a dynamic balance between safety and efficiency. Finally, the three-dimensional grid coding of the airspace in the preset time zone is performed according to the optimal grid size, and the unmanned aerial vehicle navigation control is performed, thereby converting the airspace into a coded digital grid to realize the safe and efficient flight of the unmanned aerial vehicle in the preset time period.

[0021] Through the above technical solution, the unmanned aerial vehicle request flight information in the preset time zone of the target airspace is read, the adaptive grid size is determined according to the unmanned aerial vehicle attribute data set, the collision risk in the same zone is predicted, the predicted collision risk coefficient is output, the predicted collision risk coefficient is compensated based on the predicted dynamic obstacle information and the predicted meteorological conditions, the compensated collision risk coefficient which is more suitable for the actual scenario is obtained, and the optimal grid size is obtained through iterative optimization. Finally, the three-dimensional grid coding of the airspace is performed and the navigation control is performed accordingly. In this way, the matching accuracy of the grid size and the unmanned aerial vehicle is improved through personalized adaptation analysis, the collision risk control ability is strengthened, and finally the precise balance between airspace safety and operation efficiency is realized. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A flowchart of the unmanned aerial vehicle navigation method based on the three-dimensional grid coding of the airspace provided by the present application is shown.

[0023] Figure 2 A structural diagram of the unmanned aerial vehicle navigation system based on the three-dimensional grid coding of the airspace provided by the present application is shown.

[0024] In the drawings, the components represented by the respective reference numerals are as follows:

[0025] The information reading module 11, the grid size analysis module 12, the risk prediction module 13, the parameter optimization module 14, and the output execution module 15. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative work fall within the protection scope of the present application.

[0027] In the description of the present application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0028] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that a person skilled in the art can realize the present application without using these specific details. In other examples, well-known structures and processes will not be described in detail in order to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the shown embodiments, but is consistent with the broadest scope in accordance with the principles and characteristics disclosed.

[0029] Embodiment one, as shown in the present application, provides a UAV navigation method based on spatial three-dimensional grid coding. Figure 1

[0030] S10: reading the UAV request flight information of the target airspace in a preset time zone, wherein the UAV request flight information includes a plurality of UAV attribute data sets.

[0031] ​In a preset time zone, there are multiple types of unmanned aerial vehicles in a target airspace, and their sizes, flight speeds, flight altitudes and executed tasks are significantly different, which will cause the flight requirements (such as precision and efficiency) and safety constraints (such as collision avoidance distance) of the unmanned aerial vehicles in the airspace to be difficult to meet through a unified fixed grid size, and the traditional airspace division method is easy to cause resource waste or collision risk.

[0032] To solve the above problems, the unmanned aerial vehicle request flight information of the target airspace in the preset time zone is read, wherein the unmanned aerial vehicle request flight information includes a plurality of unmanned aerial vehicle attribute data sets.

[0033] Specifically, step S10 in the method includes:

[0034] The unmanned aerial vehicle request flight information includes a plurality of unmanned aerial vehicle attribute data sets, wherein the unmanned aerial vehicle attribute data includes a task type, an aircraft type, overall size data, a flight altitude, a flight speed, a flight area and a flight time period.

[0035] In the embodiment of the application, the unmanned aerial vehicle request flight information of the target airspace in the preset time zone is read, wherein the unmanned aerial vehicle request flight information includes a plurality of unmanned aerial vehicle attribute data sets, and the unmanned aerial vehicle attribute data includes a task type, an aircraft type, overall size data, a flight altitude, a flight speed, a flight area and a flight time period. Exemplarily, the unmanned aerial vehicle attribute data can be read from multiple channels: for static attribute data such as the aircraft type, the overall size data, the flight area and the flight time period, the data is obtained from the device registration information and the airspace application platform, and the factory parameters are matched according to the unique identifier; for dynamic attribute data such as the real-time flight altitude and the flight speed, the data can be collected by an on-board sensor, and the time stamp and the area marker are combined to ensure the space-time matching, and then the information is integrated through a multi-source data fusion interface. The static data is regularly synchronized and cached, and the dynamic data is real-time verified and filtered, to generate a plurality of unmanned aerial vehicle attribute data sets, to provide complete and reliable basic inputs for three-dimensional grid adaptation analysis and collision risk prediction.

[0036] The task type refers to the specific task executed by the unmanned aerial vehicle, such as geographic mapping, logistics transportation, emergency rescue and aerial photography entertainment, and different tasks have significant differences in the demand for airspace precision: precise tasks (such as mapping) require smaller grid units to ensure detail capture, and wide-area tasks (such as forest monitoring) can use larger grids to improve efficiency, and the task type is the core basis for subsequent three-dimensional grid size adaptation analysis, and determines the basic demand for the grid size.

[0037] Among them, the aircraft type includes multi-rotor unmanned aerial vehicle, fixed-wing unmanned aerial vehicle, vertical take-off and landing unmanned aerial vehicle and the like, and the flight characteristics such as the maneuverability, turning radius and wind resistance of different types of unmanned aerial vehicles are different, for example, the fixed-wing unmanned aerial vehicle has high speed and low turning flexibility, and the multi-rotor unmanned aerial vehicle has strong maneuverability but slow speed, which are key features in collision risk prediction, for example, high-speed fixed-wing unmanned aerial vehicles need a longer early warning distance.

[0038] Among them, the overall size data refers to the overall maximum size of the unmanned aerial vehicle and the object performing the task, for example, the overall size data of the transport unmanned aerial vehicle is the overall maximum size of the unmanned aerial vehicle + delivery object, the larger the overall size, the larger the grid that needs to be adapted, and it is also the basis for evaluating the collision possibility, for example, the larger the size, the higher the probability of physical conflict with other objects.

[0039] Among them, the flight height refers to the altitude or relative height at which the unmanned aerial vehicle plans to fly, and the air space occupation and meteorological conditions (such as wind speed) of different height layers may be different, which is the core basis for dividing the height dimension of the three-dimensional grid, for example, there may be more dynamic obstacles such as civil unmanned aerial vehicles and kites in low altitude, and there may be surveying or inspection unmanned aerial vehicles in high altitude, which can avoid vertical collision through height layering.

[0040] Among them, the flight speed refers to the cruising speed or maximum speed of the unmanned aerial vehicle, and high-speed unmanned aerial vehicles cross a larger air space range in unit time and have a shorter reaction time for collision warning, so they have a higher weight in collision risk prediction, and also affect the dynamic adjustment of grid size, for example, high-speed scenarios may need larger grids to reduce the computational burden of path planning.

[0041] Among them, the flight area refers to the geographical range where the unmanned aerial vehicle plans to operate, such as a certain area of a city or a certain industrial park, which is the basis for dividing the longitude-latitude dimension of the air space and determines the spatial range of the three-dimensional grid code.

[0042] Among them, the flight period refers to the specific time interval during which the unmanned aerial vehicle plans to fly, such as 9:00-11:00 on a certain day, and the unmanned aerial vehicle density, dynamic obstacles (such as birds, kites, etc.) and meteorological conditions in the same area at different time periods may differ significantly, which is a key basis for collision risk prediction in the same area and a basis for dynamic adjustment of grid size over time.

[0043] In summary, compared with the prior art, the present application reads the unmanned aerial vehicle flight request information in the preset time zone of the target air space, wherein the unmanned aerial vehicle flight request information includes a plurality of unmanned aerial vehicle attribute data sets. In this way, the unmanned aerial vehicle attribute data completely describes the flight characteristics of the unmanned aerial vehicle from three dimensions of task demand, aircraft characteristics and space-time parameters, and these data are the data basis for subsequent grid size adaptation analysis, collision risk prediction, grid optimization and the like.

[0044] S20: Perform three-dimensional grid size adaptation analysis according to the plurality of unmanned aerial vehicle attribute data sets respectively, and obtain a plurality of adapted grid sizes.

[0045] The unmanned aerial vehicle attribute data directly affects the reasonable division of the three-dimensional grid size: if the grid division is too large, it may not meet the precision requirements of precision tasks (such as geographic mapping), and may increase the collision risk due to insufficient safety buffer space; if the division is too small, it will greatly increase the computational burden of path planning and reduce the execution efficiency of wide-area tasks (such as large-area inspection). Therefore, based on the historical flight data that has been verified to be effective, the correlation between attribute data and grid size is learned through a model, and the grid size is matched and adapted for different unmanned aerial vehicle attribute data.

[0046] To solve the above problems, the application performs three-dimensional grid size adaptation analysis according to the plurality of unmanned aerial vehicle attribute data sets respectively, and obtains a plurality of adapted grid sizes.

[0047] Specifically, step S20 in the method comprises:

[0048] Based on the historical unmanned aerial vehicle flight detection records of the target airspace, a plurality of sample unmanned aerial vehicle attribute data sets are collected, and the historical high-frequency grid size corresponding to different sample unmanned aerial vehicle attribute data sets is set as a sample grid size, thereby obtaining a plurality of sample grid sizes, wherein the historical high-frequency grid size is the mode of the historical grid size under the pre-set flight task index;

[0049] The plurality of sample unmanned aerial vehicle attribute data sets are used as input, and the plurality of sample grid sizes are used as supervision to train a deep learning model to convergence, thereby obtaining a grid size adaptation analyzer;

[0050] The grid size adaptation analyzer is used to perform three-dimensional grid size adaptation analysis according to the plurality of unmanned aerial vehicle attribute data sets respectively, and output a plurality of adapted grid sizes.

[0051] In the embodiments of the present application, sample data is first collected. Specifically, based on historical unmanned aerial vehicle flight detection records of a target airspace, attribute data of multiple unmanned aerial vehicles is collected as multiple sample unmanned aerial vehicle attribute data sets, and historical high-frequency grid sizes corresponding to different sample unmanned aerial vehicle attribute data sets are collected and set as sample grid sizes, to obtain multiple sample grid sizes. The historical high-frequency grid size is the mode of historical grid sizes under preset flight task indicators, and the preset flight task indicators refer to preset indicators such as no collision, timely completion of tasks, and path planning efficiency meeting standards. Illustratively, attribute data of a large number of unmanned aerial vehicles is extracted from historical flight detection records of a target airspace to form multiple sample unmanned aerial vehicle attribute data sets. These samples cover flight tasks in different scenarios in the past and are representative. Then, for each sample unmanned aerial vehicle attribute data set, historical grid sizes meeting preset flight task indicators (such as no collision, timely completion of tasks, and path planning efficiency meeting standards) are screened out, and the mode is taken as the historical high-frequency grid size. For example, based on historical unmanned aerial vehicle flight detection records of a target airspace, attribute data of a certain type of small surveying and mapping unmanned aerial vehicle is collected, and 0.02°×0.02°×50m grid size is used for this type of unmanned aerial vehicle. When 80% of the preset flight task indicators are met, and this size has the highest frequency, it is set as the historical high-frequency grid size corresponding to this sample unmanned aerial vehicle attribute data set. In this way, multiple sample unmanned aerial vehicle attribute data sets and corresponding multiple sample grid sizes are finally obtained according to the same method, to form input-output sample pairs required for model training.

[0052] Secondly, a plurality of sample unmanned aerial vehicle attribute data sets are used as input, a plurality of sample grid sizes are used as supervision, a deep learning model is trained to convergence, and a grid size adaptation analyzer is obtained. Illustratively, the grid size adaptation analyzer can be trained through the following technical path: 1. Data preparation: first, the plurality of sample unmanned aerial vehicle attribute data sets and the corresponding plurality of sample grid sizes are preprocessed, the category type attributes (such as task type, aircraft type) are converted into numerical features through one-hot encoding or embedding layer, the numerical type attributes (such as overall size data, flight height, flight speed) are standardized, such as normalized to the [0, 1] interval, to ensure the consistency of the input features, and then randomly divided into a training set (used for model parameter learning), a validation set (used for monitoring overfitting and adjusting hyperparameters), and a test set (used for final evaluation of the model generalization ability) according to the ratio of 7:1.5:1.5. 2. Model construction: a hybrid neural network architecture is used, mainly composed of a feature encoding layer, a deep feature fusion layer, and an output regression layer, wherein the feature encoding layer designs branch networks for different types of attributes, and the category features (such as task type, aircraft type, etc.) are mapped to low-dimensional dense vectors using embedding layers, and the numerical features (such as overall size data, flight height, flight speed, etc.) are preliminarily nonlinearly transformed using fully connected layers, and then integrated into a unified feature vector through concatenation operation; the deep feature fusion layer includes 3-5 fully connected hidden layers, with the number of neurons decreasing in turn, such as 512→256→128, each layer is connected to a ReLU activation function to introduce nonlinearity, and a Dropout layer (dropout rate=0.2) and a batch normalization layer are added to suppress overfitting and enhance the model's fitting ability for complex attribute combinations; the output regression layer uses a fully connected layer containing 3 neurons corresponding to the longitude step, latitude step, and height step, and the output layer does not use an activation function, but directly outputs continuous numerical grid size prediction results. 3. Model training: the plurality of sample unmanned aerial vehicle attribute data in the training set are used as input, and the plurality of sample grid sizes are used as supervision labels, the Adam optimizer is used with an initial learning rate of 0.001, which is dynamically decayed during training, the mean square error (MSE) is used as the loss function, and the model parameters are updated through backpropagation iteration. After each round of training, the loss is evaluated using the validation set. When the validation set loss does not decrease for 10 consecutive rounds and the test set loss tends to be stable (such as a fluctuation amplitude of less than 0.001), the model is considered to have converged, and the training is stopped, obtaining a grid size adaptation analyzer with good generalization ability.

[0053] Finally, the grid size adaptation analyzer is used to perform three-dimensional grid size adaptation analysis on the basis of a plurality of unmanned aerial vehicle attribute data sets, and a plurality of adaptive grid sizes are output. Illustratively, the pre-trained grid size adaptation analyzer is called, and a plurality of unmanned aerial vehicle attribute data sets are input. The grid size adaptation analyzer predicts and outputs corresponding adaptive grid sizes for each unmanned aerial vehicle attribute data set on the basis of learned historical rules. For example, taking a certain logistics unmanned aerial vehicle as an example, the unmanned aerial vehicle attribute data set {task type: logistics transportation, aircraft type: fixed-wing unmanned aerial vehicle, overall size: 0.5m x 0.4m x 0.4m, flight height: 50m, flight speed: 20m / min, flight area: a certain industrial park, and flight period: 9:00-12:00} is input. The grid size adaptation analyzer outputs adaptive grid sizes of longitude 55m, latitude 55m, and height 10m on the basis of rules in historical data. In this way, the individual characteristics of the unmanned aerial vehicle are accurately matched, and the effective grid sizes verified in historical flight are reused, thereby providing individual benchmark parameters for subsequent global grid size optimization.

[0054] To sum up, compared with the prior art, the present application performs three-dimensional grid size adaptation analysis on the basis of a plurality of unmanned aerial vehicle attribute data sets, and obtains a plurality of adaptive grid sizes. In this way, the grid size adaptation analyzer is trained by using historical flight data, individual adaptive grid sizes are matched for a plurality of unmanned aerial vehicle attribute data sets, and individual benchmarks are provided for subsequent global grid optimization.

[0055] S30: Simultaneous zone collision risk prediction is performed on the basis of the plurality of unmanned aerial vehicle attribute data sets, a predicted collision risk coefficient is output, the predicted collision risk coefficient is compensated on the basis of predicted dynamic obstacle information and predicted meteorological conditions in the preset time zone, and a compensated collision risk coefficient is obtained.

[0056] Unmanned aerial vehicle attributes are basic factors affecting collision risk. For example, a high-speed unmanned aerial vehicle crosses a wider airspace range in a unit of time, and has a higher possibility of crossing paths with other unmanned aerial vehicles in the same region, so the collision risk is relatively higher. Further, dynamic obstacles and meteorological conditions as external environmental factors further affect the collision risk. For example, the high-frequency appearance of dynamic obstacles such as birds and kites increases the probability of sudden collision of the unmanned aerial vehicle with non-aircraft, and strong winds can cause the unmanned aerial vehicle to deviate from the preset flight route, expand the trajectory deviation range with other unmanned aerial vehicles, and indirectly increase the collision risk.

[0057] To solve the above problems, the present application performs simultaneous zone collision risk prediction on the basis of the plurality of unmanned aerial vehicle attribute data sets, outputs a predicted collision risk coefficient, and compensates the predicted collision risk coefficient on the basis of predicted dynamic obstacle information and predicted meteorological conditions in the preset time zone, thereby obtaining a compensated collision risk coefficient.

[0058] Specifically, step S30 in the method comprises:

[0059] The plurality of unmanned aerial vehicle attribute data sets are clustered, and unmanned aerial vehicle attribute data sets in the same flight area and in the same flight period are added to the same attribute data set space to obtain a plurality of attribute data set spaces, wherein each attribute data set space has an identification of a flight area and an identification of a flight period;

[0060] The unmanned aerial vehicle collision risk prediction channel is pre-trained, wherein the unmanned aerial vehicle collision risk prediction channel comprises P collision risk prediction units, the collision risk prediction units are constructed based on a deep learning model, and P is an integer greater than or equal to 5 and less than or equal to 30;

[0061] The unmanned aerial vehicle collision risk prediction channel is used to perform simultaneous area collision risk prediction according to the plurality of attribute data set spaces respectively, a plurality of predicted collision risk probabilities are output, and the maximum predicted collision risk probability is taken as the predicted collision risk coefficient.

[0062] In the embodiments of the present application, firstly, a plurality of unmanned aerial vehicle attribute data sets are clustered, and unmanned aerial vehicle attribute data sets in the same flight area and in the same flight period are added to the same attribute data set space to obtain a plurality of attribute data set spaces, wherein each attribute data set space has an identification of a flight area and an identification of a flight period. Exemplarily, the premise of collision risk is that unmanned aerial vehicles are active in the same flight area and in the same flight period, therefore, according to the two core dimensions of flight area and flight period, the plurality of collected unmanned aerial vehicle attribute data sets are clustered: all unmanned aerial vehicle attribute data sets in the same flight area (such as overlapping latitude and longitude ranges) and in the same flight period (such as 9:00-10:00) are classified into one category to obtain a plurality of attribute data set spaces, each attribute data set space has an identification of a flight area (such as east longitude 116.3°-116.4°, north latitude 39.9°-40.0°) and an identification of a flight period (such as 2024.08.07 09:00-10:00), as a unique marker, representing all possible unmanned aerial vehicle sets that may collide in this space-time range.

[0063] Secondly, a pre-trained unmanned aerial vehicle collision risk prediction channel, wherein the unmanned aerial vehicle collision risk prediction channel comprises P collision risk prediction units, each of which is constructed based on a deep learning model, and P is an integer satisfying 5≤P≤30. The multiple collision risk prediction units are constructed to offset the inherent bias of a single model through model diversity. Different collision risk prediction units can adopt different network structures, training data subsets, etc., to form complementary risk prediction models, thereby improving the robustness of the overall prediction. Further, the value range of P (5≤P≤30) is determined based on the historical flight data characteristics of the target airspace and engineering practical experience: the lower limit 5 ensures the basic threshold of model diversity, avoiding the convergence of prediction and the difficulty in covering complex risk patterns due to too few units; the upper limit 30 balances the prediction accuracy and the calculation cost, preventing redundant calculation and affecting the efficiency of real-time risk assessment due to too many units.

[0064] Exemplarily, the collision risk prediction unit can be trained by the following technical path: 1. Data preparation: a large number of attribute data set spaces are collected from a historical flight database of a target airspace, a sample attribute data set space set is formed, a proportion of occurrence of a collision event corresponding to a statistical sample attribute data set space set is obtained, a collision risk probability of each sample attribute data set space is labeled, ranging from 0 to 1, and the closer to 1 represents the greater the collision risk, such as 0.8 representing high risk and 0.2 representing low risk, a corresponding sample collision risk probability set is obtained, then the sample attribute data set space set and the sample collision risk probability set are cross-validated by P, P parts of training data are obtained, and each part of the training data is divided into a training set (used for parameter learning), a validation set (used for hyperparameter tuning), and a test set (used for generalization ability evaluation) according to a ratio of 7:1.5:1.5. 2. Model construction: a hybrid convolutional neural network architecture of feature fusion-local correlation extraction-risk prediction can be used, mainly composed of an attribute embedding layer, a spatio-temporal feature convolution layer, and a risk prediction layer. The attribute embedding layer is designed for multiple unmanned aerial vehicle attributes in the sample space, and a multi-head embedding branch is designed. The category attributes are mapped to a dense vector through an 8-dimensional embedding layer, the numerical attributes are nonlinearly transformed through a 16-dimensional fully connected layer, and then the correlation weight between unmanned aerial vehicles is calculated through an attention mechanism. The single unmanned aerial vehicle feature vector of 32 dimensions is weighted and aggregated, and finally the feature vectors of all unmanned aerial vehicles in the space are stacked into an N×32 feature matrix (N is the number of unmanned aerial vehicles in the space). The spatio-temporal feature convolution layer includes two consecutive convolution blocks. The first convolution block uses 16 3×3 convolution kernels to perform local correlation extraction (capture the adjacent correlation of unmanned aerial vehicle speed and height) on the N×32 feature matrix, and accesses a ReLU activation function and a 2×2 max pooling layer (dimension reduction to N / 2×16). The second convolution block uses 32 2×2 convolution kernels to further extract high-order features (such as the risk pattern of multiple high-speed unmanned aerial vehicles at the same height layer), and outputs an N / 4×32 feature matrix, which is compressed into a 32-dimensional global feature vector through a global average pooling layer. The risk prediction layer is composed of two fully connected layers. The first layer (64 neurons) performs nonlinear fusion on the global features, accesses a Dropout layer (dropout rate=0.3) to suppress overfitting, and the second layer (1 neuron) outputs a collision risk probability in the range of 0-1 through a Sigmoid activation function, realizing the mapping from attribute features to risk probability.3. Model training: independently train P models with P training data, each model takes the sample attribute data set space of the training set as input, takes the corresponding sample collision risk probability as the supervision label, uses the Adam optimizer, the initial learning rate is 0.0005, and is attenuated to 0.5 of the previous value every 50 rounds, uses the binary cross entropy loss function to optimize the model parameters, and evaluates the loss and prediction accuracy (such as the proportion of samples with a probability error of ≤0.1) with the validation set at the end of each round during training. When the validation set loss does not decrease for 15 consecutive rounds and the test set AUC (area under the curve) is ≥0.92, the model is considered to have converged, and the training is stopped. Finally, P collision risk prediction units with strong generalization ability are obtained.

[0065] Finally, the unmanned aerial vehicle collision risk prediction channel is used to predict the collision risk in the same zone according to the multiple attribute data set spaces, output multiple predicted collision risk probabilities, and the maximum predicted collision risk probability is taken as the predicted collision risk coefficient. Illustratively, the pre-trained unmanned aerial vehicle collision risk prediction channel is called, multiple attribute data set spaces are input, and multiple collision risk prediction units respectively output a predicted collision risk probability. After mean calculation, the predicted collision risk probability is obtained, added to the multiple predicted collision risk probabilities, and the maximum predicted collision risk probability is taken as the predicted collision risk coefficient. The maximum predicted collision risk probability is selected for safety and conservatism. Even if most collision risk prediction units consider the risk to be low, as long as one collision risk prediction unit predicts a high risk, the high risk is taken as the benchmark to avoid underestimating the risk due to the prediction bias of the collision risk prediction unit, resulting in a collision accident.

[0066] Specifically, the "using the unmanned aerial vehicle collision risk prediction channel to predict the collision risk in the same zone according to the multiple attribute data set spaces" includes:

[0067] Randomly select a first attribute data set space in the multiple attribute data set spaces, and obtain the number of first attribute data sets in the first attribute data set space as the first flight complexity;

[0068] Multiply the first flight complexity by the ratio of the historical maximum flight complexity in the historical time range to obtain K by P integer;

[0069] Randomly select K collision risk prediction units from the P collision risk prediction units of the unmanned aerial vehicle collision risk prediction channel, predict K collision risk probabilities according to the first attribute data set space, and obtain a first predicted collision risk probability after mean calculation, and add it to the multiple predicted collision risk probabilities.

[0070] In the embodiments of the present application, first, a first attribute data set space is randomly selected in a plurality of attribute data set spaces, and the number of first attribute data sets in the first attribute data set space is obtained as a first flight complexity. For example, one attribute data set space is randomly selected from a plurality of attribute data set spaces as a first attribute data set space, and the number of first attribute data sets in the first attribute data set space, such as 10, is obtained as a first flight complexity. The flight complexity directly reflects the degree of airspace congestion. The higher the flight complexity, the higher the possibility of crossing paths between unmanned aerial vehicles, the higher the collision risk, and the more collision risk prediction units needed.

[0071] Secondly, the number K of collision risk prediction units to be called is dynamically determined according to the flight complexity. Specifically, the ratio of the first flight complexity to the historical maximum flight complexity in the historical time range is multiplied by P to obtain K, wherein K = ⌈(first flight complexity / historical maximum flight complexity in historical time range)*P⌉, and ⌈⌉ is the ceiling. For example, the historical maximum flight complexity in the historical time range is obtained, such as 20. If the first flight complexity is 10 and P is 15, then K = ⌈(10 / 20)*15⌉ = 8, that is, 8 collision risk prediction units need to be randomly called in the unmanned aerial vehicle collision risk prediction channel. In this way, the higher the flight complexity, the higher the collision risk of unmanned aerial vehicles, and the more collision risk prediction units are needed to participate to reduce accidental errors. The lower the flight complexity, the less collision risk prediction units are needed to meet the accuracy requirements, reducing the calculation cost.

[0072] Finally, K collision risk prediction units are randomly selected from P collision risk prediction units in the unmanned aerial vehicle collision risk prediction channel, K collision risk probabilities are predicted according to the first attribute data set space, and a first predicted collision risk probability is obtained after mean calculation and added to a plurality of predicted collision risk probabilities. Randomly selecting K collision risk prediction units avoids systematic bias caused by fixed unit combination, mean calculation smoothens the prediction fluctuation of a single collision risk prediction unit, and obtains more stable results.

[0073] Further, the "compensating the predicted collision risk coefficient based on the predicted dynamic obstacle information and the predicted weather condition in the preset time zone to obtain a compensated collision risk coefficient" comprises:

[0074] The attribute data set space corresponding to the predicted collision risk coefficient is taken as a marked attribute data set space, and an identification flight area and an identification flight period of the marked attribute data set space are obtained;

[0075] The predicted dynamic obstacle information and the predicted weather condition of the identification flight area in the identification flight period are obtained, wherein the predicted dynamic obstacle information is a predicted dynamic obstacle occurrence frequency, and the predicted weather condition includes a predicted wind speed.

[0076] Based on the historical unmanned aerial vehicle flight detection records of the target airspace, taking the attribute data set space as a conditional constraint, the unmanned aerial vehicle collision risk growth rate is analyzed according to the predicted dynamic obstacle occurrence frequency and the predicted wind speed, and a predicted risk growth amplitude is output;

[0077] The sum of 1 and the predicted risk growth amplitude is taken as a predicted risk growth coefficient, and the product of the predicted risk growth coefficient and the predicted collision risk coefficient is taken as a compensated collision risk coefficient.

[0078] In the embodiments of the present application, first, the attribute data set space corresponding to the predicted collision risk coefficient is taken as a marked attribute data set space, and an identified flight area and an identified flight period of the marked attribute data set space are obtained. For example, the attribute data set space corresponding to the predicted collision risk coefficient is the attribute data set space with the maximum predicted collision risk probability, and the influence of environmental factors needs to be considered first. Therefore, it is taken as the marked attribute data set space, and the corresponding identified flight area and identified flight period are obtained, for example, a certain scenic spot, Saturday 14:00-16:00.

[0079] Secondly, obtain the predicted dynamic obstacle information and the predicted meteorological condition of the identified flight area in the identified flight period. The dynamic obstacle refers to a dynamic object such as a bird, a kite, or a balloon. The predicted dynamic obstacle information is the predicted dynamic obstacle occurrence frequency. The predicted meteorological condition includes a predicted wind speed. The dynamic obstacle can physically collide with the UAV, directly increasing the collision risk. A too large wind speed can cause the UAV to lose control and deviate from the predetermined flight route, indirectly increasing the collision risk. Illustratively, the predicted wind speed can be obtained through a meteorological forecasting platform. For example, wind speed prediction data of the identified flight area in the identified flight period is collected according to the national meteorological data center, the regional meteorological monitoring network, and the like. Illustratively, the predicted dynamic obstacle information can be obtained based on a dynamic obstacle prediction model constructed by machine learning. For example, from the historical flight monitoring records of the target airspace, a large number of flight area-flight period combinations are extracted as sample input features, such as a certain park area + weekday 9:00-11:00, a certain scenic area + weekend 14:00-16:00, and the like. At the same time, the occurrence number of dynamic obstacles (such as birds, kites, balloons, and the like) in a unit time corresponding to each sample input feature is counted, such as 3 times per hour, as sample dynamic obstacle information. A time series model (such as LSTM) is used to train the sample input features as independent variables and the sample dynamic obstacle information as dependent variables until the model prediction error is > 90%, obtaining the trained dynamic obstacle prediction model. Then, the identified flight area and the identified flight period are input into the pre-trained dynamic obstacle prediction model, and the predicted dynamic obstacle information is output, such as 5 times per hour. In this way, the environment influence data matched with the target flight area and the flight period can be accurately obtained, providing a reliable basis for collision risk compensation.

[0080] Again, based on the historical UAV flight detection records of the target airspace, the attribute data set space is constrained as a conditional constraint, and the UAV collision risk growth rate is analyzed according to the predicted dynamic obstacle appearance frequency and the predicted wind speed, and the predicted risk growth amplitude is output. Illustratively, based on the historical UAV flight detection records of the target airspace, the attribute data set space is constrained as a conditional constraint, the historical flight cases with a dynamic obstacle appearance frequency of 1 time / hour and a wind speed of 1 m / s are first screened out, and the collision probability (such as 0.05) is calculated as the baseline risk value. Then, under the same attribute data set space constraint, the actual collision probability corresponding to different dynamic obstacle appearance frequencies and wind speed combinations is statistically stratified, for example, when the dynamic obstacle appearance frequency is 2 times / hour (increased by 100% compared with the baseline value) and the wind speed is 4 m / s (increased by 3 m / s compared with the baseline), the corresponding collision probability is 0.08, when the dynamic obstacle appearance frequency is 3 times / hour (increased by 200% compared with the baseline value) and the wind speed is 5 m / s (increased by 4 m / s compared with the baseline), the corresponding collision probability is 0.11. Based on these historical data, through multivariate regression analysis or piecewise fitting, a mapping rule of dynamic obstacle appearance frequency, wind speed and collision risk growth rate is constructed, for example, the fitting obtains that the collision risk growth rate is 5% when the dynamic obstacle appearance frequency increases by 50%, the collision risk growth rate is 4% when the wind speed increases by 1 m / s, and when both increase, the growth rate has a superposition effect, that is, the total growth is the sum of the respective growth rates, for example, when the dynamic obstacle appearance frequency increases by 50% and the wind speed increases by 1 m / s, the total collision risk growth rate is 5%+4%=9%, finally, the predicted dynamic obstacle appearance frequency (such as 2.5 times / hour, increased by 150% compared with the baseline) and the predicted wind speed (such as 5 m / s, increased by 4 m / s compared with the baseline) are substituted into the above mapping rule, and the total risk growth amplitude = 5% x (150% / 50%) + 4 x 4% = 15% + 16% = 31% is calculated as the predicted risk growth amplitude. In this way, the comparability of historical data and predicted scenarios is ensured through the same attribute data set space constraint, the quantitative superposition rule accurately reflects the comprehensive influence of environmental factors on collision risk, and the risk growth analysis is more in line with the actual flight scenario.

[0081] Finally, the sum of 1 and the predicted risk growth rate is taken as a predicted risk growth coefficient, and the product of the predicted risk growth coefficient and the predicted collision risk coefficient is taken as a compensated collision risk coefficient, wherein the predicted risk growth coefficient = 1 + predicted risk growth rate, and the compensated collision risk coefficient = predicted risk growth coefficient x predicted collision risk coefficient. For example, if the predicted risk growth rate is 31%, and the predicted collision risk coefficient is 0.2, then the predicted risk growth coefficient = 1 + 31% = 1.31, and the compensated collision risk coefficient = 0.2 x 1.31 = 0.262. Since the predicted collision risk coefficient is only based on the prediction of the unmanned aerial vehicle itself, and the compensated collision risk coefficient takes into account the influence of environmental factors (predicted dynamic obstacle appearance frequency, predicted wind speed), it can more truly reflect the risk level of the actual flight scene, and provide more reliable safety constraints for subsequent grid size optimization.

[0082] In summary, compared with the prior art, the present application simultaneously predicts the collision risk in the time zone according to the plurality of unmanned aerial vehicle attribute data sets, outputs a predicted collision risk coefficient, and compensates the predicted collision risk coefficient based on the predicted dynamic obstacle information and the predicted meteorological conditions in the preset time zone to obtain a compensated collision risk coefficient. In this way, the compensated collision risk coefficient takes into account not only the characteristics of the unmanned aerial vehicle itself, but also the influence of the external environment, thereby providing a key basis for the safety optimization of the three-dimensional grid size.

[0083] S40: performing grid size optimization based on the plurality of adaptive grid sizes and the compensated collision risk coefficient to obtain an optimal grid size.

[0084] The foregoing steps calculate the adaptive grid size (reflecting the baseline grid requirement of different unmanned aerial vehicle individuals) and the compensated collision risk coefficient (comprehensive safety constraints integrating environmental factors), based on which grid size optimization can be performed.

[0085] To solve the above problems, the present application performs grid size optimization based on the plurality of adaptive grid sizes and the compensated collision risk coefficient to obtain an optimal grid size.

[0086] Specifically, step S40 in the method comprises:

[0087] performing importance evaluation according to a plurality of task types of the plurality of unmanned aerial vehicle attribute data sets, and configuring a plurality of attribute weights according to the importance evaluation results;

[0088] obtaining a grid size adjustment threshold, wherein the grid size adjustment threshold comprises a longitude step interval, a latitude step interval, and a height step interval;

[0089] randomly selecting any parameter in the longitude step interval, the latitude step interval, and the height step interval to obtain a first grid size;

[0090] respectively, and a first overall size difference is obtained by weighted summation according to the plurality of attribute weights;

[0091] The random selection of the grid size and the iterative calculation of the overall size difference are continued until a preset convergence number is reached, and the grid size corresponding to the minimum overall size difference is output as the initial optimal grid size;

[0092] The ratio of the compensation collision risk coefficient to a preset maximum compensation collision risk coefficient is set as a grid size adjustment coefficient, and the product of the grid size adjustment coefficient and the initial optimal grid size is taken as the optimal grid size.

[0093] In the embodiments of the present application, first, the importance of a plurality of task types of a plurality of unmanned aerial vehicle attribute data sets is evaluated, and a plurality of attribute weights are configured according to the importance evaluation result. For example, the importance of a plurality of task types can be evaluated according to task urgency, social value, etc., for example, emergency rescue > medical material transportation > mapping > recreational aerial photography, etc., and then attribute weights are configured for each unmanned aerial vehicle according to the importance evaluation result. The higher the importance, the greater the attribute weight configured, for example, the attribute weight of emergency rescue is 0.9, and the attribute weight of recreational aerial photography is 0.3. In this way, in the subsequent calculation of grid size deviation, the adaptation requirement of high-weight unmanned aerial vehicles has a greater impact on the final result.

[0094] Secondly, a grid size adjustment threshold is obtained, wherein the grid size adjustment threshold includes a longitude step interval, a latitude step interval and a height step interval. For example, in order to avoid excessively large grid size (insufficient precision) or excessively small grid size (excessive calculation load), the step boundaries of the grid size can be preset in combination with the geographical features of the target airspace (such as smaller step size in densely populated urban areas, larger step size in suburban areas), the type of unmanned aerial vehicle (smaller step size for small unmanned aerial vehicles, larger step size for large unmanned aerial vehicles) and the historical optimal size range, for example, the longitude step interval is 10-100 meters, i.e. the minimum / maximum interval of the grid in the east-west direction, the latitude step interval is 10-100 meters, i.e. the minimum / maximum interval of the grid in the north-south direction, and the height step interval is 100-200 meters, i.e. the minimum / maximum interval of the grid in the vertical direction. In this way, it is ensured that the optimization result is within a realistic feasible range.

[0095] Thirdly, any parameter is randomly selected within the longitude step interval, the latitude step interval and the height step interval to form a first grid size. For example, a value combination is randomly selected within the longitude, latitude and height step intervals, such as longitude 45 meters, latitude 45 meters and height 15 meters, to form a first grid size.

[0096] Further, the deviation of each of the plurality of adaptive grid sizes is calculated based on the first grid size, and the first overall size difference is obtained by weighting and summing the deviations according to the plurality of attribute weights. For example, if the first grid size is longitude 45 meters, latitude 45 meters, and height 15 meters, and the plurality of adaptive grid sizes are (longitude 55 meters, latitude 55 meters, height 10 meters) and (longitude 58 meters, latitude 58 meters, height 18 meters), the deviations are calculated as follows: longitude deviation = |45-55| = 10, latitude deviation = |45-55| = 10, height deviation = |15-10| = 5, longitude deviation = |45-58| = 13, latitude deviation = |45-58| = 13, height deviation = |18-10| = 8. Then, the first overall size difference is obtained by weighting and summing the deviations according to the corresponding attribute weights, for example, if the attribute weights are 0.4 and 0.3, the first overall size difference = 0.4 x (10+10+5) + 0.3 x (13+13+8) = 20.2. The smaller the first overall size difference, the smaller the weighted and combined deviation of the first grid size from the individual adaptive grid sizes of the plurality of UAVs, that is, the closer the first grid size is to the global optimal solution under the task priority constraint.

[0097] Further, the random selection and iterative calculation of the grid size and the overall size difference are continued using the above-mentioned random selection and deviation calculation method. Each time a new grid size is generated and the corresponding overall size difference is calculated until the preset convergence number is reached. The grid size corresponding to the minimum overall size difference is output as the initial optimal grid size, wherein the preset convergence number is a pre-set maximum iteration limit, such as 500 times. The value of the preset convergence number needs to balance the optimization accuracy and the calculation cost. Too few preset convergence numbers may miss better solutions, and too many preset convergence numbers will increase unnecessary calculation load. Those skilled in the art can dynamically set the value according to actual needs. For example, after 500 iterations, the grid size corresponding to the minimum overall size difference is longitude 50 meters, latitude 50 meters, and height 20 meters, which is the initial optimal grid size. In this way, through a large number of random explorations, more potential grid size combinations can be covered, effectively avoiding the limitations of local optimal solutions, and ensuring that the initial optimal grid size can best meet the individual adaptive requirements of all UAVs and also prioritize the size requirements of high-weight task UAVs, laying a reasonable foundation for subsequent risk adjustment.

[0098] Finally, the ratio of the compensation collision risk coefficient and a preset maximum compensation collision risk coefficient is set as a grid size adjustment coefficient, and the product of the grid size adjustment coefficient and the initial optimal grid size is taken as the optimal grid size, wherein the preset maximum compensation collision risk coefficient is the highest risk threshold acceptable in the airspace, which can be dynamically set according to the airspace type (such as urban core area, suburban open area, etc.), control level (such as temporary flight restricted area, regular operation area, etc.), for example, the general operation airspace can be set to 0.8. Illustratively, if the compensation collision risk coefficient is 0.262 and the preset maximum compensation collision risk coefficient is 0.8, then the grid size adjustment coefficient = 0.262 / 0.8 = 0.33. The grid size adjustment coefficient can reflect the proportional relationship between the current collision risk and the maximum acceptable risk. The closer the grid size adjustment coefficient is to 1, the closer the current collision risk is to the upper limit of the airspace, and a larger grid size needs to be maintained to ensure real-time response capability. Conversely, the smaller the grid size adjustment coefficient, the greater the risk margin, and the grid size can be appropriately reduced to improve control accuracy. Illustratively, if the initial optimal grid size is longitude 50 meters, latitude 50 meters, and height 20 meters, the optimal grid size is calculated to be longitude 50 x 0.33 = 16.5, latitude 50 x 0.33 = 16.5, and height 20 meters x 0.33 = 6.6 meters. This is because the larger the compensation collision risk coefficient, the higher the current collision risk, the closer the grid size adjustment coefficient to 1, and the optimal grid size remains relatively large to reduce the amount of calculation and improve the real-time performance of path planning. The smaller the compensation collision risk coefficient, the lower the current collision risk, the closer the grid size adjustment coefficient to 0, and the optimal grid size is relatively small to improve the spatial resolution and enhance the navigation trajectory accuracy. In this way, by dynamically adjusting the grid size, a larger grid size is maintained when the collision risk is high to reduce the amount of calculation and improve the real-time performance of path planning, and the grid size is refined when the collision risk is low to improve the spatial resolution and enhance the navigation trajectory accuracy, thereby achieving an optimal balance between safety and efficiency under limited computing power.

[0099] In summary, compared with the prior art, the present application performs grid size optimization based on the several adaptive grid sizes and the compensation collision risk coefficient to obtain the optimal grid size. In this way, by highlighting the adaptive requirements of important tasks through weights, and by dynamically adjusting the size through iterative optimization and the compensation collision risk coefficient, the optimal grid size is ultimately obtained, which can maximize the satisfaction of individualized needs of different unmanned aerial vehicles, and can maintain a larger grid size in high-risk scenarios to reduce the amount of calculation and ensure the real-time performance of obstacle avoidance, and can appropriately reduce the grid size in low-risk scenarios to improve the spatial resolution and enhance the navigation accuracy, thereby achieving a dynamic balance between safety and efficiency.

[0100] S50: performing airspace three-dimensional grid encoding in the preset time zone according to the optimal grid size, and performing unmanned aerial vehicle navigation control.

[0101] In the embodiments of the present application, according to the optimal grid size, such as longitude 16.5, latitude 16.5, and height 6.6 meters, the target airspace in the preset time zone is divided into uniform three-dimensional grid units, each unit is assigned a unique geographic code (such as three-dimensional Geohash), when the unmanned aerial vehicle navigates, the space-time A* algorithm can be used to plan the path in the grid space, the generated path is decomposed into grid sequence instructions, such as target longitude and latitude, height, and arrival time millisecond stamp, etc., which are issued to the unmanned aerial vehicle flight control system through low delay communication for navigation control. In this way, the airspace is converted into a digital grid that can be coded, realizing safe and efficient flight of the unmanned aerial vehicle in the preset time period.

[0102] In summary, the embodiments of the present application have at least the following technical effects:

[0103] Compared with the prior art, the present application first reads the unmanned aerial vehicle request flight information of the target airspace in the preset time zone, wherein the unmanned aerial vehicle request flight information includes a plurality of unmanned aerial vehicle attribute data sets. In this way, the unmanned aerial vehicle attribute data completely describes the flight characteristics of the unmanned aerial vehicle from three dimensions of task demand, aircraft characteristics, and space-time parameters, and these data are the data basis for subsequent grid size adaptation analysis, collision risk prediction, and grid optimization.

[0104] Secondly, the present application performs three-dimensional grid size adaptation analysis according to the plurality of unmanned aerial vehicle attribute data sets to obtain a plurality of adapted grid sizes. In this way, the grid size adaptation analyzer is trained by historical flight data, and individualized adapted grid sizes are matched for the plurality of unmanned aerial vehicle attribute data sets, providing individual benchmarks for subsequent global grid optimization.

[0105] Thirdly, the present application performs collision risk prediction in the preset time zone according to the plurality of unmanned aerial vehicle attribute data sets, outputs a predicted collision risk coefficient, and compensates the predicted collision risk coefficient based on the predicted dynamic obstacle information and the predicted weather conditions in the preset time zone to obtain a compensated collision risk coefficient. In this way, the compensated collision risk coefficient takes into account both the characteristics of the unmanned aerial vehicle and the external environmental impact, providing a key basis for the safety optimization of the three-dimensional grid size.

[0106] Further, the present application performs grid size optimization based on the plurality of adapted grid sizes and the compensated collision risk coefficient to obtain an optimal grid size. In this way, by highlighting the adaptation requirements of important tasks through weights, and dynamically adjusting the size through iterative optimization and the compensated collision risk coefficient, the optimal grid size is finally obtained, which can maximize the individualized needs of different unmanned aerial vehicles, and in high-risk scenarios, by maintaining a larger grid size, the calculation amount is reduced to ensure the real-time obstacle avoidance, and in low-risk scenarios, the grid size is appropriately reduced to improve the spatial resolution to enhance the navigation accuracy, realizing the dynamic balance of safety and efficiency.

[0107] Finally, the application performs the three-dimensional grid coding of the airspace in the preset time zone according to the optimal grid size, and controls the navigation of the unmanned aerial vehicle. In this way, the airspace is converted into a digital grid that can be coded, and the safe and efficient flight of the unmanned aerial vehicle in the preset time period is realized.

[0108] Through the above technical solution, the application reads the unmanned aerial vehicle request flight information in the preset time zone of the target airspace, determines the adaptive grid size according to the unmanned aerial vehicle attribute data set, then performs the same time zone collision risk prediction, outputs the predicted collision risk coefficient, and compensates the predicted collision risk coefficient based on the predicted dynamic obstacle information and the predicted meteorological conditions, to obtain a more realistic scenario. The compensated collision risk coefficient is obtained, and the optimal grid size is obtained by iterative optimization. Finally, the three-dimensional grid coding of the airspace is performed according to the optimal grid size, and the navigation control is performed. In this way, the matching accuracy of the grid size and the unmanned aerial vehicle is improved through personalized adaptive analysis, the collision risk control ability is strengthened, and finally the precise balance between airspace safety and operation efficiency is realized.

[0109] Embodiment two, as shown in Figure 2 based on the same inventive concept of the unmanned aerial vehicle navigation method based on the three-dimensional grid coding of the airspace provided in embodiment one, the application embodiment also provides an unmanned aerial vehicle navigation system based on the three-dimensional grid coding of the airspace, comprising:

[0110] The information reading module 11 is configured to read the unmanned aerial vehicle request flight information in the preset time zone of the target airspace, wherein the unmanned aerial vehicle request flight information comprises a plurality of unmanned aerial vehicle attribute data sets;

[0111] The grid size analysis module 12 is configured to perform three-dimensional grid size adaptive analysis according to the plurality of unmanned aerial vehicle attribute data sets, and obtain a plurality of adaptive grid sizes;

[0112] The risk prediction module 13 is configured to perform the same time zone collision risk prediction according to the plurality of unmanned aerial vehicle attribute data sets, output the predicted collision risk coefficient, and compensate the predicted collision risk coefficient based on the predicted dynamic obstacle information and the predicted meteorological conditions in the preset time zone, to obtain the compensated collision risk coefficient;

[0113] The parameter optimization module 14 is configured to perform grid size optimization based on the plurality of adaptive grid sizes and the compensated collision risk coefficient, and obtain the optimal grid size;

[0114] The output execution module 15 is configured to perform the three-dimensional grid coding of the airspace in the preset time zone according to the optimal grid size, and control the navigation of the unmanned aerial vehicle.

[0115] The information reading module 11 is configured to:

[0116] The flight information request of the unmanned aerial vehicle includes a plurality of unmanned aerial vehicle attribute data sets, wherein the unmanned aerial vehicle attribute data includes a task type, an aircraft type, overall size data, a flight height, a flight speed, a flight area, and a flight period.

[0117] The grid size analysis module 12 is specifically configured to:

[0118] Based on the historical unmanned aerial vehicle flight detection records of the target airspace, a plurality of sample unmanned aerial vehicle attribute data sets are collected, and the historical high-frequency grid size corresponding to different sample unmanned aerial vehicle attribute data sets is set as a sample grid size, thereby obtaining a plurality of sample grid sizes, wherein the historical high-frequency grid size is the mode of the historical grid size under the preset flight task index;

[0119] The plurality of sample unmanned aerial vehicle attribute data sets are used as input, and the plurality of sample grid sizes are used as supervision to train a deep learning model to convergence, thereby obtaining a grid size adaptation analyzer;

[0120] The grid size adaptation analyzer is used to perform three-dimensional grid size adaptation analysis on the plurality of unmanned aerial vehicle attribute data sets respectively, and a plurality of adapted grid sizes are output. The risk prediction module 13 is specifically configured to:

[0121] The plurality of unmanned aerial vehicle attribute data sets are clustered, and unmanned aerial vehicle attribute data sets in the same flight area and in the same flight period are added to the same attribute data set space, thereby obtaining a plurality of attribute data set spaces, wherein each attribute data set space has an identifier of a flight area and an identifier of a flight period;

[0122] A pre-trained unmanned aerial vehicle collision risk prediction channel is used, wherein the unmanned aerial vehicle collision risk prediction channel includes P collision risk prediction units, the collision risk prediction units are constructed based on a deep learning model, and P is an integer greater than or equal to 5 and less than or equal to 30;

[0123] The unmanned aerial vehicle collision risk prediction channel is used to perform simultaneous zone collision risk prediction on the plurality of attribute data set spaces respectively, a plurality of predicted collision risk probabilities are output, and the maximum predicted collision risk probability is used as the predicted collision risk coefficient.

[0124] Specifically, the "unmanned aerial vehicle collision risk prediction channel is used to perform simultaneous zone collision risk prediction on the plurality of attribute data set spaces respectively" includes:

[0125] A first attribute data set space is randomly selected in the plurality of attribute data set spaces, and the number of first attribute data sets in the first attribute data set space is obtained as a first flight complexity;

[0126] multiplying the ratio of the first flight complexity and the historical maximum flight complexity in a historical time range by P and taking an integer to obtain K;

[0127] randomly selecting K collision risk prediction units in the P collision risk prediction units of the unmanned aerial vehicle collision risk prediction channel, and obtaining K collision risk probabilities according to the first attribute data set space prediction, and obtaining a first predicted collision risk probability through mean calculation and adding the first predicted collision risk probability to the plurality of predicted collision risk probabilities.

[0128] Further, the "compensating the predicted collision risk coefficient based on the predicted dynamic obstacle information and the predicted meteorological condition in the preset time zone to obtain a compensated collision risk coefficient" comprises:

[0129] taking the attribute data set space corresponding to the predicted collision risk coefficient as a marked attribute data set space, and obtaining an identified flight area and an identified flight period of the marked attribute data set space;

[0130] obtaining the predicted dynamic obstacle information and the predicted meteorological condition of the identified flight area in the identified flight period, wherein the predicted dynamic obstacle information is a predicted dynamic obstacle occurrence frequency, and the predicted meteorological condition comprises a predicted wind speed;

[0131] based on the historical unmanned aerial vehicle flight detection record of the target airspace, taking the attribute data set space as a conditional constraint, performing unmanned aerial vehicle collision risk growth ratio analysis according to the predicted dynamic obstacle occurrence frequency and the predicted wind speed, and outputting a predicted risk growth amplitude;

[0132] taking the sum of 1 and the predicted risk growth amplitude as a predicted risk growth coefficient, and taking the product of the predicted risk growth coefficient and the predicted collision risk coefficient as a compensated collision risk coefficient.

[0133] The parameter optimization module 14 is specifically configured to:

[0134] performing importance evaluation according to the plurality of task types of the plurality of unmanned aerial vehicle attribute data sets, and configuring a plurality of attribute weights according to the importance evaluation result;

[0135] obtaining a grid size adjustment threshold, wherein the grid size adjustment threshold comprises a longitude step interval, a latitude step interval and a height step interval;

[0136] randomly selecting any parameter in the longitude step interval, the latitude step interval and the height step interval for combination to obtain a first grid size;

[0137] taking the first grid size as a reference, respectively performing deviation calculation on the plurality of adaptive grid sizes, and performing weighted summation according to the plurality of attribute weights to obtain a first overall size difference.

[0138] Continue to randomly select the grid size and iteratively calculate the overall size difference until a preset convergence number is reached, and output the grid size corresponding to the minimum overall size difference as the initial optimal grid size;

[0139] Set the ratio of the compensation collision risk coefficient to the preset maximum compensation collision risk coefficient as the grid size adjustment coefficient, and set the product of the grid size adjustment coefficient and the initial optimal grid size as the optimal grid size.

[0140] The output execution module 15 is specifically configured to:

[0141] According to the optimal grid size, perform airspace three-dimensional grid encoding in the preset time zone, and perform unmanned aerial vehicle navigation control.

[0142] In summary, the embodiments of the present application have at least the following technical effects:

[0143] Compared with the prior art, firstly, the information reading module is used to read the unmanned aerial vehicle flight request information in the preset time zone, thereby providing a reliable data basis for subsequent grid size adaptation analysis, collision risk prediction, grid optimization, and the like. Secondly, the grid size analysis module is used to perform three-dimensional grid size adaptation analysis on the basis of a plurality of unmanned aerial vehicle attribute data sets, to obtain a plurality of adapted grid sizes, to train the grid size adaptation analyzer by using historical flight data, and to match personalized adapted grid sizes for the plurality of unmanned aerial vehicle attribute data sets, thereby providing a benchmark for subsequent global grid optimization. Thirdly, the risk prediction module is used to predict the collision risk in the same zone on the basis of the plurality of unmanned aerial vehicle attribute data sets, to output a predicted collision risk coefficient, and to compensate the predicted collision risk coefficient on the basis of the predicted dynamic obstacle information and the predicted weather conditions in the preset time zone, thereby obtaining a compensated collision risk coefficient. The compensated collision risk coefficient takes into account both the unmanned aerial vehicle characteristics and the external environmental influences, thereby providing a key basis for the safe optimization of the three-dimensional grid size. Further, the parameter optimization module is used to optimize the grid size on the basis of the plurality of adapted grid sizes and the compensated collision risk coefficient, to obtain an optimal grid size, to highlight the adaptation requirements of important tasks by using weights, to dynamically adjust the size by using iterative optimization and the compensated collision risk coefficient, and to finally obtain the optimal grid size. The optimal grid size can maximize the satisfaction of the personalized requirements of different unmanned aerial vehicles, can reduce the calculation amount by maintaining a larger grid size in a high-risk scenario to ensure the real-time obstacle avoidance, and can appropriately reduce the grid size in a low-risk scenario to improve the spatial resolution and enhance the navigation accuracy, thereby achieving the dynamic balance between safety and efficiency. Finally, the output execution module is used to perform the three-dimensional grid coding of the airspace in the preset time zone on the basis of the optimal grid size, and to perform the unmanned aerial vehicle navigation control, thereby converting the airspace into a codable digital grid and realizing the safe and efficient flight of the unmanned aerial vehicle in the preset time period. In this way, the personalized adaptation analysis improves the matching accuracy of the grid size and the unmanned aerial vehicle, strengthens the collision risk control capability, and finally achieves the precise balance between airspace safety and operation efficiency.

[0144] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0145] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0147] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0148] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0149] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments described and shown, and it is therefore intended that the application cover any and all variations that fall within the scope of the present application. Accordingly, the application is not limited by the foregoing description, but is only limited by the scope of the appended claims.

Claims

1. A UAV navigation method based on spatial three-dimensional mesh coding, characterized in that, The methods include: Read the drone flight request information within a preset time zone in the target airspace, wherein the drone flight request information includes several drone attribute datasets; Based on the aforementioned UAV attribute datasets, 3D mesh size adaptation analysis is performed to obtain several adapted mesh sizes, including: Based on historical UAV flight detection records of the target airspace, multiple sample UAV attribute datasets are collected, and the historical high-frequency grid size corresponding to different sample UAV attribute datasets is set as the sample grid size to obtain multiple sample grid sizes. Among them, the historical high-frequency grid size is the mode of the historical grid size that meets the preset flight mission indicators. Using the multiple sample drone attribute datasets as input and the multiple sample grid sizes as supervision, a deep learning model is trained until convergence to obtain a grid size adaptor analyzer. Using the mesh size adaptation analyzer, three-dimensional mesh size adaptation analysis is performed on the several UAV attribute datasets respectively, and several adapted mesh sizes are output; Based on the aforementioned several UAV attribute datasets, simultaneous collision risk prediction is performed, a predicted collision risk coefficient is output, and the predicted collision risk coefficient is compensated based on the predicted dynamic obstacle information and predicted weather conditions within the preset time zone to obtain a compensated collision risk coefficient. Based on the aforementioned adaptive mesh sizes and collision risk compensation coefficients, mesh size optimization is performed to obtain the optimal mesh size, including: The importance of several task types in the aforementioned several UAV attribute datasets is evaluated, and the weights of several attributes are configured based on the importance evaluation results. Obtain the grid size adjustment threshold, wherein the grid size adjustment threshold includes longitude step interval, latitude step interval and altitude step interval; The first grid size is obtained by randomly selecting any parameter within the longitude step interval, latitude step interval, and altitude step interval and combining them. Based on the first grid size, the deviations of the plurality of adapted grid sizes are calculated respectively, and the first overall size difference is obtained by weighted summation according to the plurality of attribute weights; Continue to randomly select the mesh size and iteratively calculate the overall size difference until the preset number of convergences is reached, and output the mesh size corresponding to the minimum overall size difference as the initial optimal mesh size; The ratio of the compensated collision risk coefficient to the preset maximum compensated collision risk coefficient is set as the mesh size adjustment coefficient, and the product of the mesh size adjustment coefficient and the initial optimal mesh size is taken as the optimal mesh size; Perform spatial three-dimensional mesh encoding within the preset time zone based on the optimal mesh size, and then perform UAV navigation control.

2. The UAV navigation method based on spatial three-dimensional mesh coding according to claim 1, characterized in that, The drone request flight information includes several drone attribute datasets, wherein the drone attribute data includes mission type, aircraft type, overall size data, flight altitude, flight speed, flight area and flight time period.

3. The UAV navigation method based on spatial three-dimensional mesh coding according to claim 2, characterized in that, Based on the aforementioned datasets of drone attributes, simultaneous collision risk prediction is performed, and the predicted collision risk coefficient is output, including: Cluster the aforementioned drone attribute datasets, and add drone attribute datasets that are in the same flight area and the same flight time period to the same attribute dataset space to obtain multiple attribute dataset spaces, wherein each attribute dataset space has an identifier for the flight area and an identifier for the flight time period; A pre-trained drone collision risk prediction channel, wherein the drone collision risk prediction channel includes P collision risk prediction units, the collision risk prediction units are constructed based on a deep learning model, and P is an integer greater than or equal to 5 and less than or equal to 30; Using the aforementioned UAV collision risk prediction channel, collision risk prediction is performed simultaneously in the space of the multiple attribute datasets, outputting multiple predicted collision risk probabilities, and the maximum predicted collision risk probability is used as the predicted collision risk coefficient.

4. The UAV navigation method based on spatial three-dimensional mesh coding according to claim 3, characterized in that, Using the aforementioned UAV collision risk prediction channel, simultaneous zone collision risk prediction is performed based on the spatial distribution of the multiple attribute datasets, including: Randomly select a first attribute dataset space within the plurality of attribute dataset spaces, and obtain the number of first attribute datasets in the first attribute dataset space as the first flight complexity; The ratio of the first flight complexity to the historical maximum flight complexity within the historical time range is multiplied by P and rounded to obtain K; K collision risk prediction units are randomly selected from the P collision risk prediction units in the UAV collision risk prediction channel. K collision risk probabilities are spatially predicted based on the first attribute dataset. The average value is used to calculate the first predicted collision risk probability, which is then added to the plurality of predicted collision risk probabilities.

5. The UAV navigation method based on spatial three-dimensional mesh coding according to claim 1, characterized in that, The predicted collision risk coefficient is compensated based on the predicted dynamic obstacle information and predicted weather conditions within the preset time zone to obtain the compensated collision risk coefficient, including: The attribute dataset space corresponding to the predicted collision risk coefficient is used as the labeled attribute dataset space to obtain the identified flight area and labeled flight time period of the labeled attribute dataset space; Obtain predicted dynamic obstacle information and predicted weather conditions for the identified flight area during the identified flight period, wherein the predicted dynamic obstacle information is the predicted frequency of occurrence of dynamic obstacles, and the predicted weather conditions include predicted wind speed; Based on historical UAV flight detection records of the target airspace, and with the attribute dataset space as a condition constraint, the UAV collision risk growth ratio is analyzed according to the predicted dynamic obstacle occurrence frequency and predicted wind speed, and the predicted risk growth rate is output. The sum of 1 and the predicted risk growth rate is used as the predicted risk growth coefficient, and the product of the predicted risk growth coefficient and the predicted collision risk coefficient is used as the compensation collision risk coefficient.

6. A UAV navigation system based on spatial three-dimensional mesh coding, characterized in that, For performing the method according to any one of claims 1-5, comprising: The information reading module is used to read the drone flight request information in the target airspace within a preset time zone, wherein the drone flight request information includes several drone attribute datasets; The mesh size analysis module is used to perform three-dimensional mesh size adaptation analysis based on the several UAV attribute datasets to obtain several adapted mesh sizes; The risk prediction module is used to predict the collision risk in the same time zone based on the aforementioned several UAV attribute datasets, output the predicted collision risk coefficient, and compensate the predicted collision risk coefficient based on the predicted dynamic obstacle information and predicted weather conditions in the preset time zone to obtain the compensated collision risk coefficient. The parameter optimization module is used to optimize the mesh size based on the several adaptive mesh sizes and the compensation collision risk coefficient to obtain the optimal mesh size; the output execution module is used to perform spatial three-dimensional mesh encoding in the preset time zone according to the optimal mesh size and to perform UAV navigation control.

Images