Global view generator for low-altitude economic unmanned aerial vehicle monitoring

Abstract

A global view generator for low-altitude economic unmanned aerial vehicle monitoring comprises a grouping strategy setting module, an unmanned aerial vehicle clustering and grouping module, a local view generation module and a view aggregation module which are connected in sequence, flexible definition and real-time adjustment of grouping logic are achieved through modular strategy design and dynamic expansion interfaces, and the overall view generator is high in practicability and easy to popularize. And the scene adaptation capability of the system is obviously improved. The unmanned aerial vehicle clustering and grouping module adopts a dynamic graph clustering model and an event triggering optimization mechanism, only performs local adjustment on related parts of incremental data, and ensures grouping efficiency and real-time response capability. And the local view generation module generates grouping boundaries by using a dynamic three-dimensional alpha-shape algorithm, constructs a local view by combining state feature mapping, and provides real-time data support for grouping and monitoring. The view aggregation module effectively solves the problem of boundary conflict between groups through a global optimization technology, ensures the consistency of global views in spatial relationship, grouping range and state characteristics, and provides reliable support for unmanned aerial vehicle monitoring and task coordination in a complex task scene.

Description

Technical Field

[0001] The present invention relates to a technology in the field of unmanned aerial vehicle (UAV) control, and specifically to a global view generator for low-altitude economic UAV monitoring. Background Art

[0002] Currently, as an important direction of the deep integration of aviation technology and intelligent management technology, the low-altitude economy is showing extensive application potential in multiple industrial fields. With the gradual deepening of the application of UAVs in scenarios such as logistics distribution, aerial patrol, emergency rescue, and exploration surveying and mapping, how to achieve the efficient management of UAV groups has become the core task to promote the development of the low-altitude economy. Most of the existing UAV grouping logics are based on preset rules, with fixed grouping strategies and lack of flexibility. It is difficult for users to dynamically adjust the grouping rules according to the changes in the task scenarios, which limits the system's adaptability to diverse requirements and results in poor scalability of the grouping model. In the clustering grouping process, the existing technologies usually rely on static clustering algorithms, such as K-means, and have poor support for incremental data. When the state or distribution of UAVs changes, it often requires global recalculation, leading to low computational efficiency, lagging update of the grouping state, and difficulty in meeting the actual requirements of real-time performance and stability. In addition, the generation of grouping boundaries is usually based on static geometric forms, which are difficult to be updated in real time with the dynamic changes in the distribution and state of UAVs, and the grouping information is prone to inconsistency, weakening the integrity and reliability of the grouping model. In terms of global view generation, the existing technologies mostly adopt simple static superposition methods, lacking optimization mechanisms and being unable to effectively handle the problem of grouping boundary conflicts, which affects the overall consistency and dynamic adaptability of the global view. Summary of the Invention

[0003] Aiming at the above deficiencies of the existing technology, the present invention proposes a global view generator for low-altitude economic UAV monitoring. Through modular strategy design and dynamic expansion interfaces, it realizes the flexible definition and real-time adjustment of grouping logic, significantly improving the system's scenario adaptability. The UAV clustering grouping module adopts a dynamic graph clustering model and an event-triggered optimization mechanism, and only makes local adjustments to the parts related to incremental data, ensuring grouping efficiency and real-time response ability. The local view generation module uses the dynamic three-dimensional α-shape algorithm to generate grouping boundaries and constructs a local view in combination with state feature mapping, providing real-time data support for grouping and monitoring. The view aggregation module effectively solves the problem of boundary conflicts between groups through global optimization technology, ensuring the consistency of the global view in terms of spatial relationships, grouping scopes, and state features, and providing reliable support for UAV monitoring and task coordination in complex task scenarios.

[0004] The present invention is realized through the following technical solutions:

[0005] The present invention relates to a global view generator for low-altitude economic drone monitoring, comprising: a grouping strategy setting module, a drone clustering and grouping module, a local view generation module, and a view aggregation module that are connected in sequence, where: the grouping strategy setting module calculates the comprehensive difference degree function between drones based on the input grouping logic and evaluation criteria, generates a grouping evaluation function, weight parameters, preliminary grouping rules, and grouping optimization objectives; the drone clustering and grouping module performs clustering processing on real-time incremental data through a dynamic graph clustering model according to the comprehensive difference degree function, grouping evaluation function, and preliminary grouping rules, and dynamically maintains grouping information and the neighborhood relationship between groups; the local view generation module generates the spatial boundary and local view of each group through a dynamic three-dimensional alpha shape algorithm according to the grouping information; the view aggregation module integrates each spatial boundary and local view through a global optimization algorithm to obtain a global view for drone state monitoring and task coordination in a dynamic environment.

[0006] The grouping logic is determined by spatial distance, remaining battery power, task type, and flight speed. By setting the weight w of each factor i , the comprehensive difference between drones is calculated through the comprehensive difference degree function: where: d k (x ik ,x jk ) is the distance metric function of the kth factor, w k is the weight parameter, satisfying For example, the spatial distance uses the Euclidean distance where: p i =(x i ,y i ,z i ) is the three-dimensional coordinate of drone i, and discrete variables such as task type use the symbolic distance metric The module further defines the grouping evaluation function E, with the intra-group compactness and inter-group separation as the core indicators to quantify the grouping quality: E = α·IntraClusterCohesion - β·InterClusterSeparation, where: α and β are the weight parameters of the evaluation function, respectively controlling the balance within and between groups. To support the scalability of the module, the system allows users to dynamically add new grouping factors through the policy engine. For example, when the user adds a new grouping factor x m , the comprehensive difference degree formula can be automatically updated to: where: w m is the weight of the new factor x m , satisfying The system dynamically adjusts the proportions of various factors through a weight normalization method, enabling new factors to seamlessly integrate into the grouping logic. The evaluation function E also adapts to the changes in the new grouping factors, and the grouping quality can be recalculated after the update. The module rationally distributes the impact of new factors on the within-group compactness or between-group separation by dynamically adjusting the weights of α and β. Users can optimize the weight w of the new factor according to actual needs m to quickly adapt to the new scenario.

[0007] The clustering process mentioned above specifically includes:

[0008] Step 1: Construct a dynamic neighborhood and event triggering mechanism. Through the dynamic neighborhood construction mechanism, map the spatial coordinates p i and state attributes s i of the UAV into a high-dimensional feature space to construct the neighborhood structure of the initial grouping. When incremental data is triggered, the module identifies the affected neighborhood set, and the affected neighborhoods trigger update events, thus avoiding the overhead of global recomputation.

[0009] The neighborhood structure mentioned above is N(p i ) = p j |D(p i , p j ) ≤ ∈, where: D(p i , p j ) is the distance metric in the feature space, and ∈ is the neighborhood radius.

[0010] The affected neighborhood set mentioned above is

[0011] Step 2: Use a dynamic graph clustering model to achieve incremental grouping optimization. For each affected grouping C k , update its center point μ k and feature distribution:

[0012] The affected groupings mentioned above include:

[0013] a. New data point addition: When the distance D(p new , μ k ) from the new point p new to any existing grouping center μ k satisfies D ≤ ∈, add it to the grouping and update the grouping features; otherwise, create a new grouping C new .

[0014] b. Change in the state of existing points: Recalculate the center point μ k and neighborhood structure of the affected groupings, and adjust the membership relationship of the points.

[0015] c. Group merging and splitting: When the center distance D(μ k and C l ) of two groups C k satisfies D ≤ ∈, group merging is performed. For groups with low density, splitting operations are performed according to the condition that |C l | is less than the threshold δ. k | is less than the threshold δ.

[0016] Step 3. Dynamically optimize the grouping results based on the multi-objective optimization method: Seek the optimal grouping configuration through the intra-group compactness and inter-group separation, and through the multi-objective optimization algorithm NSGA-II, specifically including:

[0017] 3.1. Define the objective function: Maximize the intra-group compactness that measures the similarity between samples within the group and minimize the inter-group separation that measures the difference between groups where: K is the number of groups, C k is the k-th group, and μ k is the group center.

[0018] 3.2. Pareto front search and population generation: Use the NSGA-II algorithm to generate the Pareto optimal solution set, and each solution represents a grouping configuration. Initialize a population of N individuals, each individual corresponding to a grouping configuration, and the initial parameters include the number of groups K and the feature center μ k .

[0019] 3.3. Genetic operations: Generate a new grouping configuration for the parameters of two individuals through cross processing of linear combination: C = αA + (1 - α)B, or perform mutation processing by adding random perturbations to the group center μ k : Δμ k = γ·u, where: α ∈ [0, 1] is the cross coefficient; for mutation, γ is the mutation amplitude, and u is a random vector of the standard normal distribution.

[0020] 3.4. For each generation of the population, recalculate IntraClusterCohesion and InterClusterSeparation, and screen the individuals of the next generation through non-dominated sorting.

[0021] 3.5. Stop iterating and obtain the optimal grouping configuration when the maximum number of iterations T max is reached or the Pareto solution set is stable.

[0022] Step 4. After each event is triggered, perform local optimization on the affected groups and ensure the rationality of the overall grouping logic through the global consistency mechanism, specifically including:

[0023] 4.1. Local optimization: Among them: is Group C k Gradient on the evaluation function E, and η is the learning rate.

[0024] 4.2. Evaluate the global consistency through the grouping stability index S: Among them: Var is the variance of the points within the group. The higher the S value, the more stable the grouping.

[0025] Step 5. Output the updated grouping relationships C1, C2, …, C K To the local view generation module for further boundary generation and view construction.

[0026] The described dynamic three-dimensional α-shape algorithm specifically includes:

[0027] Step i. Data preprocessing and spatial mapping: Receive the three-dimensional spatial coordinates p i =(x i , y i , z i ) and the state feature vector s i , and after ensuring the consistency of different feature dimensions through normalization processing, map the data to the logical nodes in the three-dimensional space to form a state feature map, where: s i includes battery power and task type.

[0028] The described normalization processing means: Among them: μ s and σ s are the feature mean and standard deviation respectively.

[0029] Step ii. Neighborhood construction and three-dimensional Delaunay triangulation: On the basis of the normalized data, construct the neighborhood relationship of the point set. Use the three-dimensional Delaunay triangulation algorithm to divide the point set into a group of non-overlapping tetrahedral elements satisfying: Among them: circumsphere(T j ) is the circumscribed sphere of the tetrahedron T j . The triangulation ensures the effectiveness of the space division and provides a geometric basis for the construction of the α-shape.

[0030] Step iii. Retain the tetrahedrons that meet the conditions according to the α value set by the user: Among them: radius(T j ) is the circumscribed sphere radius of the tetrahedron T j . By adjusting the α value, the shape of the boundary can be controlled. A smaller α value generates a more detailed boundary, while a larger α value generates a more extensive boundary.

[0031] Step iv. Optimize the generated α - shape boundary: By introducing the state feature weight w s , re - define the node weights on the boundary: where: λ k is the weight of the k - th feature, satisfying s ik is the eigenvalue. Adjust the positions of the boundary nodes based on the weighted weights to ensure that the boundary can accurately reflect the state feature distribution of the UAV.

[0032] Step v. When a UAV is added or removed, trigger the real - time boundary update mechanism: For the newly added node p new , calculate its distance D min to the nearest neighbor node of the existing boundary: If D min ≤α, then incorporate the node p new into the α - shape boundary; otherwise, create a new boundary cell.

[0033] Step vi. On the basis of boundary generation, map the UAV state features to the local view, specifically: where: is the three - dimensional coordinate set, is the boundary cell set, is the state feature set.

[0034] The global optimization algorithm mentioned above specifically includes:

[0035] Step a. Parsing and parameterization of the local view: Receive the local view output by the local view generation module, specifically where: is the node set of group k, is the group boundary set, is the group state feature set. Through a unified parameterization method, extract the geometric description of the group boundary , such as the boundary surface equation or the discrete boundary point set.

[0036] Step b. Detect boundary conflicts between groups using the Euclidean distance and spatial overlap degree indicators: Assume that the boundary surface equations of groups k and l are f k (x,y,z)=0 and f l (x,y,z)=0 respectively, then the conflict area is specifically: where: ∈ is the tolerance threshold for boundary conflicts.

[0037] Step c, construct the global optimization problem: By defining the optimization objective function to minimize the overlapping area between groups and maintain the stability of the grouping range, the optimization objective function E total = α·E overlap + β·E stability , where: is the overlapping volume between groups, is the change amount of the grouping node set, and α and β are weight parameters, satisfying α + β = 1.

[0038] Step d, use the gradient descent method to optimize the boundary surface equation to minimize the objective function E total : Let the gradient of the boundary surface f k (x, y, z) be The optimization update formula is: where: η is the learning rate.

[0039] Step e, after adjusting the grouping boundary, project the state feature into the global view: Let the state feature set of the global view be The fusion formula is: where: w k is the feature weight of group k, which is dynamically adjusted based on the importance of the group.

[0040] Step f, when the change amount ΔE total of the objective function E total < δ, or reaches the maximum number of iterations T max , terminate the optimization and output the global view including the global node set, boundary set, and state feature set where: δ is a preset threshold.

[0041] The grouping information described above includes: UAV information containing the 3D coordinates and state data of the UAVs, grouping boundary data, grouping state data, and grouping neighborhood relationships. Among them: The UAV information records the 3D coordinates and state data of each UAV within the group, which is used to reflect the spatial position and dynamic state of the UAVs in real time (such as battery level, flight speed, etc.), providing fine-grained support for UAV management and monitoring; The grouping boundary data is used to describe the geometric range of the group in 3D space; The grouping state data records the real-time attributes of the group (such as the number of UAVs, distribution uniformity, etc.), providing a basis for the dynamic optimization and state evaluation of the group; The neighborhood relationship between groups is used to describe the spatial adjacency and topological structure between groups, providing support for group collaborative operations and boundary conflict detection.

[0042] The present invention relates to a global view generation method for low-altitude economy UAV monitoring based on the above generator, including:

[0043] Step 1: Preprocess the input UAV feature data according to the preset grouping logic and evaluation criteria: standardize the three-dimensional spatial coordinates, remaining battery power, mission type, and flight speed of the UAVs, calculate the similarity between each UAV through the comprehensive difference function, and then use the grouping strategy to generate the preliminary grouping evaluation function, weight parameters, and optimization objectives.

[0044] Step 2: Perform local updates on the real-time incremental data: Through the dynamic neighborhood construction and event-triggering mechanism, real-time identify the local neighborhood affected by the UAV state change or newly added data, and perform local clustering reconstruction on this basis: update the center and feature distribution of each affected group, and at the same time classify new data points or merge and split the original groups according to the data increment situation. Set the neighborhood radius and density threshold throughout the process to balance the calculation efficiency and grouping accuracy, and avoid the delay problem caused by global recalculation.

[0045] Step 3: Use the dynamic three-dimensional α-shape algorithm to generate the spatial boundary and map the state features of each group: Standardize and spatially map the three-dimensional coordinates and state features of the UAVs. After constructing the logical nodes and state feature map, construct the neighborhood relationship through three-dimensional Delaunay triangulation, and filter the tetrahedral units according to the α value set by the user, so as to generate a detailed and adaptive grouping boundary. At the same time, optimize and adjust the boundary nodes through the state feature weight and set a real-time update mechanism.

[0046] Step 4: Integrate all local views through the global optimization algorithm to generate a unified and logical global view: Analyze the local views of each group, detect possible boundary conflicts between groups, construct the global objective function, and use the multi-objective evolutionary algorithm to iteratively optimize the boundary, dynamically adjust the fusion weight of the state features of each group until the preset termination condition is met. Technical effects

[0001] Through a dynamic adjustment mechanism, the present invention optimizes the grouping logic and scheme according to real-time task requirements and environmental changes, significantly improving the system's adaptability and efficiency. The present invention realizes local adjustment of incremental data through a dynamic graph clustering model, without global recalculation, greatly improving the efficiency and response speed of the grouping process. Combining multi-objective optimization methods ensures the quality and stability of the grouping results under dynamic changes. Using the dynamic three-dimensional α-shape algorithm, the present invention constructs grouping boundaries and local views that support real-time updates, providing high-precision geometric boundaries and state feature mappings for generating the global view, and enhancing the monitoring ability in complex task scenarios. In addition, the view aggregation module integrates grouping information through global optimization techniques to solve the problem of grouping boundary conflicts and ensure the consistency of the global view in terms of spatial relationships and state features. Compared with the simple overlay display method of the prior art, the present invention can dynamically adjust the view layout and state mapping to provide a more intuitive and accurate global monitoring view. Overall, the present invention significantly improves the efficiency, real-time performance, and adaptability of the UAV grouping and monitoring system, provides an innovative solution for UAV task management and resource optimization in the low-altitude economy, and can be widely applied to multi-task scenarios such as logistics distribution, aerial patrol, emergency rescue, and exploration and mapping. Brief Description of the Drawings

[0002] Figure 1 It is a schematic diagram of the system of the present invention;

[0003] Figure 2 It is a schematic diagram of the application of the embodiment;

[0004] Figure 3 It is a data structure diagram of the UAV global view of the present invention. Detailed Embodiment

[0005] As Figure 1 shown, this embodiment relates to a global view generator for UAV monitoring in the low-altitude economy, including: a grouping strategy setting module, a UAV clustering and grouping module, a local view generation module, and a view aggregation module connected in sequence, where: the grouping strategy setting module implements the grouping logic and evaluation criteria as extensible code components in an object-oriented manner; the UAV clustering and grouping module performs high-dimensional vectorization of incremental data in the feature space with NumPy for calculation; the local view generation module standardizes state features with NumPy and maps the UAV spatial coordinates and state information into unified three-dimensional feature node data; the view aggregation module performs unified processing and data integration on the local views through NumPy. The module receives the local views output by the local view generation module and programmatically loads the node coordinate data, boundary parameters, and state feature vectors of each group into a NumPy array in memory for unified vectorized operations.

[0006] The described grouping strategy setting module reads the weight parameters w of each grouping factor from the configuration file during system initialization i and relevant factor information, and loads the feature information (such as spatial position, remaining battery power, task type, flight speed, etc.) of the UAVs into the memory in the form of key-value pairs or arrays. To calculate the comprehensive difference degree function D(x i ,x j ), the module defines the corresponding calculation function in Python, and implements the distance metric function d k (x ik ,x jk ) in the form of conditional judgment or mathematical operations. For continuous factors such as Euclidean distance, basic arithmetic operations are used, and for discrete factors such as task type, 0 or 1 is returned through simple logical conditions. After the comprehensive difference degree is calculated, the calculation methods of the within-group compactness and between-group separation are defined according to the grouped results, and the evaluation function E is calculated in the form of a function. The loaded weight parameters α and β are used for real-time weighting in the code to obtain the corresponding grouping quality score. It is implemented with a policy engine for dynamic expansion. When the user adds a new grouping factor x m through the configuration file or graphical interface, the system automatically updates the weight vector at runtime, adds the weight w m of the new factor and performs normalization processing. At the same time, the calculation logic of the comprehensive difference degree and the evaluation function automatically adapts to the existence and weight changes of the new factor without modifying the core code. Through this modular and parameterized programming implementation method, the definition and expansion of the grouping strategy can maintain a high degree of flexibility at the code level, enabling users to quickly adapt to and configure new task scenarios. The grouping strategy setting module includes: a grouping factor reading unit, a comprehensive difference degree calculation unit, a grouping quality scoring unit, and a policy engine expansion unit, where: the grouping factor reading unit loads the feature information (such as spatial position, remaining battery power, task type, flight speed, etc.) of the UAVs according to the weight parameters w i of the grouping factors read from the configuration file and relevant factor information, and loads it into the memory in the form of key-value pairs or arrays; the comprehensive difference degree calculation unit calculates the comprehensive difference degree function D(x i ,x j) and adopt different calculation methods according to different types of factors (such as continuous factors like Euclidean distance or discrete factors like task type); the grouping quality scoring unit defines the calculation methods of within-group compactness and between-group separation based on the calculated comprehensive difference degree, and calculates the evaluation function E in the form of a function, and adjusts the grouping quality score in real time through the weighted parameters α and β; the policy engine extension unit dynamically extends the grouping policy according to the settings of the configuration file or graphical interface, automatically updates the weight vector and performs normalization processing to ensure that the calculation of the comprehensive difference degree and evaluation function adapts to the existence and weight changes of new factors, and at the same time, without modifying the core code, to ensure the flexibility and scalability of the system.

[0007] The described UAV clustering and grouping module uses NumPy to perform high-dimensional vectorization on the incremental data in the feature space for computing. During the system operation, it monitors the input of incremental data in real time, locally updates the affected neighborhood structure through an event-triggered mechanism, and uses NumPy to quickly calculate the distance metric between data points. When a new UAV joins or the attributes of an existing UAV change, it performs incremental calculation and iterative update on the grouping center points according to the dynamic neighborhood relationship, and in necessary, uses the vectorization operation implemented by NumPy to perform real-time processing on the merging or splitting of groups. The support for multi-objective optimization is completed by docking with the NSGA-II-based optimization algorithm in the DEAP library. In each generation of optimization, NumPy is used to quickly calculate and statistically analyze the Euclidean distance between the points within the group and the difference between the center points of different groups, and IntraClusterCohesion and InterClusterSeparation are used as the input quantities of the objective function and passed into the DEAP framework. The new group configurations generated after crossover and mutation operations are immediately evaluated and selected in the Python runtime environment, gradually approaching the Pareto front solution set. The local optimization steps triggered by incremental data are achieved through iterative update of the objective function gradient. Its update rule is called in a functional manner in the Python code. NumPy is used to perform differential operations on the group feature statistics, and the group center points are evolved in the direction of improving group stability and optimizing the objective function value according to the learning rate. Finally, the updated group configuration is output in real time, and the group relationship is passed to the subsequent local view generation for further spatial layout and logical modeling. This UAV clustering and grouping module includes: an incremental data processing unit, a local update unit, a multi-objective optimization unit, and a group configuration output unit, where: the incremental data processing unit uses NumPy to perform high-dimensional vectorization on the incremental data in the feature space, monitors the input of incremental data in real time, locally updates the affected neighborhood structure through an event-triggered mechanism, and quickly calculates the distance metric between data points; the local update unit dynamically calculates the group center points and performs incremental updates according to the joining of new UAVs or changes in the attributes of existing UAVs, and in necessary, performs group merging or splitting operations through NumPy; the multi-objective optimization unit docks with the NSGA-II-based optimization algorithm in the DEAP library. In each generation of optimization, it quickly calculates and statistically analyzes the Euclidean distance within the group and the difference between the center points of different groups, uses IntraClusterCohesion and InterClusterSeparation as the input of the objective function, and optimizes the group configuration through crossover and mutation operations, gradually approaching the Pareto front solution set; the group configuration output unit performs iterative update on the objective function gradient through the local optimization steps triggered by incremental data, adjusts the group center points according to the learning rate to improve group stability and optimize the objective function value, and finally outputs the updated group configuration in real time and passes the group relationship to the subsequent local view generation.

[0008] The described local view generation module standardizes the state features using NumPy, mapping the spatial coordinates and state information of the drones into unified three-dimensional feature node data. After obtaining the normalized data, it calls the Delaunay function of SciPy to perform three-dimensional Delaunay triangulation on the point set, generating a set of tetrahedral elements for filtering. Subsequently, a filtering function written in Python is used to calculate the radii of these triangulated elements, and tetrahedrons that do not meet the conditions are removed according to the given α value, thereby forming an α-shaped boundary adapted to the scene features. For state feature embedding and boundary optimization, the state feature vectors of the nodes are weighted using NumPy, dynamically adjusting the coordinate positions of the boundary nodes to more accurately reflect the feature distribution of the drone swarm. When a new drone joins or an existing drone leaves, the update function inside Python is called in an event-triggered manner to quickly locate and recalculate the affected boundary elements without repeating the processing of global data, thus achieving real-time update of the spatial boundary. This local view generation module includes: a feature standardization unit, a three-dimensional triangulation unit, a boundary filtering unit, and a boundary update unit, where: the feature standardization unit standardizes the spatial coordinates and state information of the drones using NumPy and maps them into unified three-dimensional feature node data; the three-dimensional triangulation unit calls the Delaunay function of SciPy to perform three-dimensional Delaunay triangulation on the standardized node set, generating a set of tetrahedral elements and providing them for subsequent filtering operations; the boundary filtering unit calculates the radii of these triangulated elements through the written filtering function, removes tetrahedral elements that do not meet the conditions according to the given α value, and finally forms an α-shaped boundary adapted to the scene features; the boundary update unit triggers an event to call the update function when a new drone joins or an existing drone leaves, quickly locates and recalculates the affected boundary elements, and does not need to repeat the processing of global data, thus achieving real-time update of the spatial boundary.

[0009] The described view aggregation module unifies and integrates the local views using NumPy. It receives the local views output from the local view generation, and programmatically loads the node coordinate data, boundary parameters, and state feature vectors of each group into NumPy arrays in memory for unified vectorized operations. For the problem of boundary conflict detection and optimization, numerical calculation functions based on NumPy in Python are used to quickly iteratively update the boundary surface equations and feature data between groups, and the detection efficiency is improved through the vectorization of the Euclidean distance and overlap measure functions. During the global optimization process, based on the NumPy array for the objective function E totalPerform gradient approximation calculations, and perform gradient descent iterations through a custom Python function to gradually reduce the boundary overlap and unstable factors between groups. For the fusion and mapping of state features, the weighted sum method is used to superimpose the state feature sets of each group, and the adaptive allocation of feature importance is achieved by dynamically adjusting the weights in the Python code. When the iteration reaches the convergence condition or exceeds the preset number of times, the optimized global view results (including nodes, boundaries, and state features) are output as a unified data structure and passed to the subsequent display and monitoring links to provide a consistent, optimized, and logically coherent global view for the system in a dynamic environment. This view aggregation module includes: a local view integration unit, a boundary conflict detection unit, a global optimization unit, and a view output unit, where: the local view integration unit loads the node coordinate data, boundary parameters, and state feature vectors from the local view generation module into NumPy arrays in memory through NumPy and performs unified vectorized operations; the boundary conflict detection unit uses NumPy for numerical calculations, quickly iteratively updates the boundary surface equations and feature data between groups, and improves the efficiency of conflict detection through vectorized operations of the Euclidean distance and overlap measure functions; the global optimization unit is based on the NumPy array for the objective function E total Perform gradient approximation calculations, and gradually reduce the boundary overlap and unstable factors between groups through a custom gradient descent iteration function; the view output unit outputs the optimized global view (including nodes, boundaries, and state features) as a unified data structure when the iteration reaches the convergence condition or exceeds the preset number of times, and passes it to the subsequent display and monitoring links to provide a consistent, optimized, and logically coherent global view for the system.

[0010] After specific actual experiments, in low-altitude economic scenarios such as UAV swarm management, status monitoring, and task coordination, construct a control system containing the above-mentioned generator, such as Figure 2 shown, including: a Web application layer, a business processing layer, and a data layer, where: the Web application layer is for the interaction between users and the system, providing grouping strategy definition, grouping status viewing, and global view result feedback; the business processing layer comprehensively processes grouping strategies, real-time UAV data, and grouping logic. The business processing layer completes the dynamic adjustment of grouping relationships, the generation of grouping boundaries, and the optimized output of the global view; the data layer uses InfluxDB to store the spatio-temporal data stream of UAVs, providing high throughput and low latency support for real-time monitoring and querying; MongoDB is used to store the global view to achieve flexible expansion and fast retrieval of data structures in a document-based storage method.

[0011] Users input the grouping logic settings and grouping evaluation criteria of UAVs through the Web interface and view the global view generated by the system in real time.

[0012] The Web - side application layer is built based on HTML5, CSS3, and Vue.js to construct a dynamic and responsive user interface, supporting users to upload data files in JSON format and YAML format. The system realizes communication with the business processing layer through Axios, submits the data input by users to the back - end service layer via RESTful API, and supports real - time processing and feedback. The user interface uses the Element Plus component library to build input forms and interactive panels to ensure the simplicity and intuitiveness of user operations. The visualization module uses D3.js and Three.js to achieve dynamic three - dimensional display of grouping boundaries and global views, and real - time displays grouping status, boundary changes, and the operating parameters of drones. The grouping status viewing page provides real - time updated grouping information, including core indicators such as the number of drones within a group, the in - group compactness, and the inter - group separation degree, presented in the form of charts, and uses ECharts to provide intuitive statistical data display. The system also supports the export function of grouping results. Users can choose to save the final grouping information and global view as CSV or PNG files for subsequent analysis and sharing.

[0013] Experimental data shows that for the dynamic clustering model and related algorithms of the present invention, compared with the traditional global recalculation method, the average calculation time for real - time data processing is reduced by about 40% to 60%, the grouping quality is significantly improved, and the overall grouping stability is increased by about 15% to 25%. Even in complex scenarios with the introduction of new grouping factors or frequent changes in drone status, rapid adaptive adjustment can be achieved. In the simulation environment, tests for up to 500 drones show that through the dynamic update mechanism, the average response time for boundary updates is reduced from the traditional 120 milliseconds to about 60 milliseconds, the processing speed is increased by about 50%, and the optimization of state feature embedding improves the boundary accuracy by about 20%. In addition, in the monitoring scenario with 100 drones and 30 groups constructed in the laboratory, the global optimization algorithm reduces the average overlapping area of grouping boundaries by about 60%, increases the grouping stability by about 25%, reduces the average number of optimization iterations from 80 times to 45 times, shortens the total response time from 200 milliseconds to 95 milliseconds, the decline rate of the objective function exceeds 80%, and further improves the accuracy of the grouping status information in the global view by about 18%.

[0014] Table 1 Comparison of Technical Characteristics

[0015] Through the collaborative application of multiple core technical features, the present invention is significantly superior to the prior art in terms of real-time performance, integrity, stability, and scalability. In terms of real-time performance, the present invention adopts a dynamic graph clustering model and a dynamic three-dimensional alpha shape algorithm, effectively solving the problems of incremental data processing and environmental change adaptation in dynamic scenarios. Different from the prior art that requires global recalculation, the present invention only performs local updates on the affected groups or view areas, and its average response time is reduced by about 40% to 60%, greatly improving the real-time response ability of the system and enabling rapid processing of UAV state changes and environmental disturbances. In terms of integrity, the present invention realizes the real-time update of group state features and global views, ensuring data consistency and comprehensiveness; the dynamic three-dimensional alpha shape algorithm adopted by the local view generation module, combined with state feature mapping, makes the generation of group boundaries fine and accurate, avoiding boundary distortion and state data loss. In terms of stability, through the event-triggered optimization mechanism and multi-objective evaluation function of the dynamic graph clustering model, the present invention has shown an increase in group stability of about 15% to 25% in laboratory tests, and the global optimization has reduced the average overlapping area of group boundaries by about 60%, and the overall response time has been reduced from 200 milliseconds to about 95 milliseconds, ensuring long-term stable operation. In terms of scalability, the modular design and standardized interfaces enable the system to be flexibly extended to meet diverse application requirements, and at the same time, user-defined grouping strategies are realized through dynamic extension interfaces, demonstrating excellent flexibility and scalability. In summary, the technical solution of the present invention is not only applicable to the grouping and monitoring of UAVs in the low-altitude economy scenario, but also provides efficient and reliable support for complex task execution and resource optimization, with broad application prospects and significant technical advantages.

[0016] The above specific embodiments can be locally adjusted by those skilled in the art in different ways without departing from the principles and purposes of the present invention. The protection scope of the present invention is subject to the claims and is not limited by the above specific embodiments, and all implementation solutions within its scope are subject to the present invention.

Claims

1. A global view generator for low-altitude economical UAV monitoring, characterized in that: include: The grouping strategy setting module, the UAV clustering grouping module, the local view generation module and the view aggregation module are connected in sequence, wherein: the grouping strategy setting module calculates the comprehensive difference function between UAVs based on the input grouping logic and evaluation criteria, generates the grouping evaluation function, weight parameters, preliminary grouping rules and grouping optimization targets; the UAV clustering grouping module clusters the real-time incremental data through the dynamic graph clustering model according to the comprehensive difference function, the grouping evaluation function and the preliminary grouping rules, and dynamically maintains the grouping information and the neighborhood relationship between the groups; the local view generation module generates the spatial boundary and local view of each group through the dynamic three-dimensional α shape algorithm according to the grouping information; the view aggregation module integrates the various spatial boundaries and local views through the global optimization algorithm to obtain the global view, which is used for UAV status monitoring and task coordination in a dynamic environment; The grouping information includes: drone information including drone three-dimensional coordinates and drone status data, group boundary data, group status data and group neighborhood relationship, wherein: drone information records the three-dimensional coordinates and status data of each drone in the group, which is used to reflect the spatial position and dynamic status of the drone in real time, and provide fine-grained support for drone management and monitoring; group boundary data is used to describe the geometric range of the group in three-dimensional space; group status data records the real-time properties of the group, and provides a basis for dynamic optimization and status evaluation of the group; the neighborhood relationship between groups is used to describe the spatial adjacency and topological structure between groups, and provide support for group collaborative operation and boundary conflict detection.

2. The global view generator for low-altitude economical UAV monitoring according to claim 1 is characterized in that: The grouping logic is determined by the spatial distance, remaining power, mission type and flight speed, and the weight w of each factor is set. i , the comprehensive difference between drones is calculated through the comprehensive difference function: Where: d k (x ik ,x jk ) is the distance metric function of the kth factor, w k is the weight parameter, satisfying For example, spatial distance is Euclidean distance Where: p i =(x i ,y i ,z i ) is the three-dimensional coordinate of UAV i, and discrete variables such as mission type are measured using signed distance The module further defines the evaluation function E of grouping, which takes the intra-group cohesion and inter-group separation as the core indicators to quantify the grouping quality: E = α·IntraClusterCohesion-β·InterClusterSeparation, where α and β are the weight parameters of the evaluation function, which control the balance within and between groups respectively. To support the scalability of the module, the system allows users to dynamically add new grouping factors through the policy engine. For example, when a user adds a new grouping factor x m When , the comprehensive difference formula can be automatically updated to: Where: w m Add the new factor x m The weight of The system dynamically adjusts the proportion of each factor through the weight normalization method, so that the newly added factors can be seamlessly integrated into the grouping logic. The evaluation function E also adapts to the changes of the newly added grouping factors. The grouping quality can be recalculated after updating. The module dynamically adjusts the weights of α and β to reasonably distribute the influence of the newly added factors on the tightness within the group or the separation between groups. The user can adjust the weight w of the newly added factors according to actual needs. m Perform optimization to quickly adapt to new scenarios.

3. The global view generator for low-altitude economical UAV monitoring according to claim 1 is characterized in that: The clustering process specifically includes: Step 1: Construct dynamic neighborhood and event trigger mechanism: Through the dynamic neighborhood construction mechanism, the spatial coordinates of the drone p i , state attributes i Mapped into high-dimensional feature space, constructing the neighborhood structure of the initial grouping, when the incremental data is triggered, the module identifies the affected neighborhood set, and the affected neighborhood triggers an update event, thus avoiding the overhead of global recalculation; The neighborhood structure is N(p i )=p j ∣D(p i ,p j )≤∈, where: D(p i ,p j ) is the distance metric in the feature space, ∈ is the neighborhood radius; The affected neighborhood set is in: Step 2: Use dynamic graph clustering model to implement incremental grouping optimization: For each affected group C k , about its center point μ k And the feature distribution is updated: Step 3: Dynamically optimize the grouping results based on the multi-objective optimization method: through the intra-group closeness and inter-group separation, the multi-objective optimization algorithm NSGA-II is used to find the optimal grouping configuration; Step 4: After each event is triggered, local optimization is performed on the affected groups, and the rationality of the overall grouping logic is ensured through a global consistency mechanism; Step 5: Output the updated grouping relationship C1, C2, ..., C K To the local view generation module for further boundary generation and view construction.

4. The global view generator for low-altitude economical UAV monitoring according to claim 3 is characterized in that: The affected groups include: a. New data points are added: When the new point p new To any existing group center μ k The distance D(p new ,μ k ) satisfies D≤∈, then add it to the group and update the group feature; otherwise, create a new group C new ; b. Change of existing point status: recalculate the center point μ of the affected group k and neighborhood structure, adjusting the affiliation of points; c. Group merging and splitting: When two groups C k and C l The center distance D(μ k ,μ l ) satisfies D≤∈, then grouping is performed. For groups with lower density, |C k |The split operation is performed when the condition is less than the threshold δ.

5. The global view generator for low-altitude economical UAV monitoring according to claim 3 is characterized in that: The step 3 specifically includes: 3.

1. Define the objective function: Maximize the intra-group closeness to measure the similarity between samples within the group and the minimum separation between groups, which measures the differences between groups Where: K is the number of groups, C k is the kth group, μ k For grouping centers; 3.

2. Pareto frontier search and population generation: Use the NSGA-II algorithm to generate the Pareto optimal solution set. Each solution represents a grouping configuration. Initialize the population of N individuals. Each individual corresponds to a grouping configuration. The initial parameters include the number of groups K and the feature center μ. k ; 3.

3. Genetic operation: The parameters of two individuals are transformed into a new group configuration through the crossover process of linear combination: C = αA + (1-α) B, or the group center μ k Add random perturbations to achieve mutation processing: Δμ k =γ·u, where: α∈[0,1] is the crossover coefficient; variation, γ is the variation amplitude, and u is a standard normal distribution random vector; 3.

4. For each generation of population, recalculate IntraClusterCohesion and InterClusterSeparation, and select the next generation of individuals through non-dominated sorting; 3.

5. When the maximum number of iterations T is reached max Or when the Pareto solution set is stable, the iteration is stopped and the optimal grouping configuration is obtained.

6. The global view generator for low-altitude economical UAV monitoring according to claim 3 is characterized in that: The step 4 specifically includes: 4.

1. Local optimization: in: For group C k The gradient on the evaluation function E, η is the learning rate; 4.

2. Evaluate global consistency through group stability index S: Where: Var is the variance of the points within the group.

7. The global view generator for low-altitude economical UAV monitoring according to claim 1 is characterized in that: The dynamic three-dimensional alpha shape algorithm specifically includes: Step i: Data preprocessing and spatial mapping: Receive the three-dimensional spatial coordinates p of the drone i =(x i ,y i ,z i ) and the state feature vector s i After ensuring the consistency of different feature dimensions through standardization, the data is mapped to the logical nodes in the three-dimensional space to form a state feature graph, where: i Including power, task type; Step ii: Neighborhood construction and 3D Delaunay triangulation: Based on the standardized data, the neighborhood relationship of the point set is constructed, and the point set is transformed into a 3D Delaunay triangulation algorithm. Split into a set of non-intersecting tetrahedral elements satisfy: Where: circumsphere(T j ) is a tetrahedron T j The circumscribed sphere of the,triangulation ensures the validity of the space division and provides a geometric,basis for the construction of the α shape; Step iii: retain the tetrahedrons that meet the conditions according to the α value set by the user: Where: radius(T j ) is a tetrahedron T j The radius of the circumscribed sphere is used to control the shape of the boundary by adjusting the α value. A smaller α value produces a more detailed boundary, while a larger α value produces a wider boundary. Step iv: Optimize the generated α shape boundary by introducing the state feature weight w s , redefine the node weights on the boundary: Where: k is the weight of the kth feature, satisfying s ik The position of the boundary nodes is adjusted based on the weighted weights to ensure that the boundary can accurately reflect the state characteristic distribution of the UAV; Step v: When adding or removing a drone, trigger the boundary real-time update mechanism: for the newly added node p new , calculate the distance D between it and the nearest neighbor node of the existing boundary min : If D min ≤α, then the node p new Include the α-shaped boundary; otherwise, create a new boundary cell; Step vi: Based on the boundary generation, the drone state features are mapped to the local view, specifically: in: is a set of three-dimensional coordinates, is the boundary cell set, is a set of state characteristics.

8. The global view generator for low-altitude economical UAV monitoring according to claim 1 is characterized in that: The global optimization algorithm specifically includes: Step a: parsing and parameterizing the local view: receiving the local view output by the local view generation module, specifically: in: is the set of nodes in group k, is the group boundary set, It is a set of grouping state features, and the grouping boundaries are extracted through a unified parameterization method. A geometric description of , such as boundary surface equations or a set of discrete boundary points; Step b: Use the Euclidean distance and spatial overlap index to detect the boundary conflicts between groups: Assume that the boundary surface equations of groups k and l are f k (x,y,z)=0 and f l (x,y,z)=0, then the conflict area Specifically: Where: ∈ is the tolerance threshold of boundary conflict; Step c: Construct a global optimization problem: By defining the optimization objective function to minimize the overlapping area between groups and maintain the stability of the group range, optimize the objective function E total =α·E overlap +β·E stability ,in: is the overlapping volume between groups, is the change in the grouped node set, α and β are weight parameters, satisfying α+β=1; Step d: Use the gradient descent method to optimize the boundary surface equation so that the objective function E total Minimize: Let the boundary surface f k The gradient of (x,y,z) is The optimization update formula is: Where: η is the learning rate; Step e: After adjusting the grouping boundaries, the state features Projection into the global view: Let the state feature set of the global view be The fusion formula is: Where: w k The feature weight of group k is dynamically adjusted based on the importance of the group; Step f: When the objective function E is satisfied total The change in ΔE total <δ, or the maximum number of iterations T is reached max The optimization is terminated and a global view containing the global node set, boundary set, and state feature set is output. Where: δ is the preset threshold.

9. A global view generation method for low-altitude economical UAV monitoring based on the generator described in any one of claims 1-8, characterized in that: include: Step 1: Preprocess the input drone feature data according to the preset grouping logic and evaluation criteria: standardize the drone's three-dimensional spatial coordinates, remaining power, mission type, and flight speed, and calculate the similarity between drones through a comprehensive difference function, and then use the grouping strategy to generate a preliminary grouping evaluation function, weight parameters, and optimization objectives; Step 2: Locally update the real-time incremental data: Through dynamic neighborhood construction and event triggering mechanism, the local neighborhood affected by the change of drone status or the newly added data is identified in real time, and local clustering reconstruction is performed on this basis: the center and feature distribution of each affected group are updated, and new data points are classified or the original groups are merged and split according to the data increment. In the whole process, the neighborhood radius and density threshold are set to balance the calculation efficiency and grouping accuracy, avoiding the delay problem caused by global recalculation; Step 3: Generate spatial boundaries and map state characteristics of each group using a dynamic three-dimensional α shape algorithm: standardize and spatially map the three-dimensional coordinates of the drone and its state characteristics, construct logical nodes and state characteristic graphs, construct neighborhood relationships through three-dimensional Delaunay triangulation, and filter tetrahedral units according to the α value set by the user, thereby generating detailed and adaptive group boundaries. At the same time, the boundary nodes are optimized and adjusted through the state characteristic weights and a real-time update mechanism is provided; Step 4: Integrate all local views through a global optimization algorithm to generate a unified and logical global view: After parsing the local views of each group and detecting possible boundary conflicts between groups, construct a global objective function, and use a multi-objective evolutionary algorithm to iteratively optimize the boundaries, dynamically adjust the fusion weights of each group state features, until the preset termination conditions are met.

Images