Multi-scene oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method and system

Abstract

The application relates to the technical field of data processing, and discloses a multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method and system. The method comprises the following steps: generating multi-scene task constraint data through semantic analysis; constructing a dynamic potential field matrix based on a Gaussian function; establishing a continuous motion velocity field containing pressure gradient and viscosity diffusion; fusing a vortex velocity component to form an adaptive control velocity field; constructing a task allocation vector field to generate an operation area configuration scheme and convert the operation area configuration scheme into flight control instructions. The application solves the core problems of insufficient adaptability, difficulty in balancing real-time performance and optimality in the prior art, improves the intelligent scheduling capability and cooperative operation efficiency of a group of unmanned aerial vehicles in a complex and changeable environment through an innovative method combining semantic understanding and physical modeling.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a multi-scene oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method and system. BACKGROUND

[0002] The existing unmanned aerial vehicle operation resource configuration and scheduling technology mainly adopts a rule-based heuristic algorithm and a centralized optimization method, and realizes task allocation and path planning of a UAV group through a pre-defined task template and fixed parameter configuration. The traditional method is usually specially designed for a single application scene, such as agricultural plant protection, logistics distribution or environmental monitoring, and each scene has an independent algorithm architecture and parameter setting, and relies on manually preset task priority and space division strategy to guide the operation behavior of the unmanned aerial vehicle.

[0003] The existing technology has significant deficiencies, mainly in poor scene adaptability, difficulty in balancing real-time performance and optimality, limited environmental uncertainty processing capability, etc. When facing new application scenes or complex tasks mixed with multiple scenes, the existing method needs to redesign the algorithm and adjust the parameters, and lacks cross-scene generality and migratability. At the same time, the traditional discrete path planning and centralized scheduling mechanism has a sharp increase in computational complexity when processing large-scale UAV clusters and dynamic environmental changes, and it is difficult to obtain an optimal solution within a limited time, and has insufficient robustness to sudden situations such as communication interruption and equipment failure. SUMMARY

[0004] The present application provides a multi-scene oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method and system, which is used to solve the core problems of insufficient multi-scene adaptability and difficulty in balancing real-time performance and optimality in the prior art, and improves the intelligent scheduling capability and collaborative operation efficiency of the UAV group in a complex and variable environment through an innovative method combining semantic understanding and physical modeling.

[0005] In a first aspect, the present application provides a multi-scene oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method, which comprises:

[0006] Step S1: performing semantic analysis and constraint formalization processing on a natural language instruction input by a user to generate multi-scene task constraint data, the multi-scene task constraint data comprising spatial coordinates, task weights and scene types;

[0007] Step S2: constructing a dynamic potential field matrix reflecting multi-scene task priority distribution through Gaussian function combination based on the spatial coordinates and task weights in the multi-scene task constraint data;

[0008] Step S3: taking the dynamic potential field matrix as a driving source, establishing a continuous motion velocity field of the UAV group containing a pressure gradient term and a viscosity diffusion term;

[0009] Step S4: calculating a vortex influence coefficient according to a UAV capability evaluation value and a current position potential field intensity, generating a local vortex velocity component, and fusing the local vortex velocity component with the continuous motion velocity field to obtain a multi-scenario adaptive control velocity field;

[0010] Step S5: constructing a task allocation vector field based on the multi-scenario adaptive control velocity field and a scene type feature, generating a UAV operation area configuration scheme through dynamic probability calculation, and converting the UAV operation area configuration scheme into real-time flight control instructions to complete closed-loop resource scheduling.

[0011] In a second aspect, the present application provides a multi-scenario-oriented UAV operation resource dynamic configuration and scheduling system, which comprises:

[0012] A generation module is configured to perform semantic analysis and constraint formalization processing on a natural language instruction input by a user, and generate multi-scenario task constraint data, wherein the multi-scenario task constraint data comprises spatial coordinates, task weights, and scene types.

[0013] A construction module is configured to construct a dynamic potential field matrix reflecting multi-scenario task priority distribution by Gaussian function combination based on the spatial coordinates and task weights in the multi-scenario task constraint data.

[0014] An analysis module is configured to take the dynamic potential field matrix as a driving source, and establish a continuous motion velocity field of a UAV group containing a pressure gradient term and a viscosity diffusion term.

[0015] A fusion module is configured to calculate a vortex influence coefficient according to a UAV capability evaluation value and a current position potential field intensity, generate a local vortex velocity component, and fuse the local vortex velocity component with the continuous motion velocity field to obtain a multi-scenario adaptive control velocity field.

[0016] A conversion module is configured to construct a task allocation vector field based on the multi-scenario adaptive control velocity field and a scene type feature, generate a UAV operation area configuration scheme through dynamic probability calculation, and convert the UAV operation area configuration scheme into real-time flight control instructions to complete closed-loop resource scheduling.

[0017] In a third aspect, a multi-scenario-oriented UAV operation resource dynamic configuration and scheduling device is provided, which comprises a memory and at least one processor, wherein the memory stores instructions; and the at least one processor invokes the instructions in the memory, so that the multi-scenario-oriented UAV operation resource dynamic configuration and scheduling device performs the multi-scenario-oriented UAV operation resource dynamic configuration and scheduling method described above.

[0018] In a fourth aspect, a computer readable storage medium is provided, and the computer readable storage medium stores instructions which, when executed on a computer, cause the computer to perform the multi-scene oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method described above.

[0019] In the technical solutions provided in the present application, the multi-scene task constraint data is generated by performing semantic analysis and constraint formalization processing on the natural language instruction, which breaks through the limitation of the traditional unmanned aerial vehicle scheduling method relying on preset templates and fixed parameters, so that a single technical architecture can simultaneously process the operation requirements of multiple heterogeneous scenes such as agricultural plant protection, logistics distribution, emergency rescue, etc., and fundamentally solves the problem of insufficient scene adaptability in the prior art, and the standardized constraint data format ensures seamless conversion and unified processing of task information between different scenes. The technical features of constructing a dynamic potential field matrix based on spatial coordinates and task weights through Gaussian functions, innovatively introduce continuous mathematical functions into discrete task allocation problems, so that complex multi-task priority relationships can be quantitatively described and spatially mapped through unified mathematical expressions. Compared with the traditional grid division and weight allocation method, the dynamic potential field matrix can naturally handle the time-varying characteristics and spatial continuity of task priority, providing a smooth and differentiable driving source for subsequent motion planning. The dynamic potential field matrix is used as a driving source to establish a continuous motion velocity field containing a pressure gradient term and a viscosity diffusion term, and the principle of Navier-Stokes equation in fluid mechanics is used to convert the discrete motion problem of the unmanned aerial vehicle group into a continuous field theory problem. The pressure gradient term ensures the uniform distribution of the unmanned aerial vehicle group in space to avoid excessive aggregation, and the viscosity diffusion term ensures the continuity and coordination of motion in adjacent regions. This continuous processing method fundamentally changes the traditional point-to-point path planning idea, so that the entire unmanned aerial vehicle group presents a coordinated motion characteristic similar to fluid.

[0020] The technical features of calculating the vortex influence coefficient and generating the local vortex velocity component according to the UAV capability evaluation value and the current position potential field intensity, ingeniously introduce the vortex dynamics mechanism to solve the conflict avoidance problem between UAVs, the calculation of the vortex influence coefficient considers the individual capability difference of the UAV and the importance of the UAV in the task space, so that the UAV with strong capability and located in the key position obtains greater local influence, and the local vortex velocity component realizes elegant detour avoidance by generating a rotating effect, compared with the traditional hard constraint and re-planning mechanism, the vortex mechanism provides a more natural and efficient conflict solution. The technical features of constructing the task allocation vector field based on the multi-scene adaptive control velocity field and the scene type characteristics and generating the work area configuration scheme through dynamic probability calculation convert the abstract task allocation problem into a specific probability calculation problem, the task allocation vector field comprehensively considers the global motion trend and the local scene characteristics, the dynamic probability calculation ensures the quantitative comparability and decision transparency of the task allocation result, and finally converts into real-time flight control instructions to complete the closed-loop resource scheduling, forming a complete technical link from high-level semantic understanding to underlying hardware control. In the specific multi-scene UAV work application field, the introduction of the semantic analysis algorithm enables non-professional users to directly describe complex work requirements through natural language, the Gaussian function combination algorithm ensures the reasonable distribution and smooth transition of tasks with different priorities in space, the continuous motion velocity field algorithm realizes the coordinated control of large-scale UAV groups by referring to the theory of fluid mechanics, the vortex avoidance algorithm solves the conflict problem in the multi-UAV system through a physical heuristic method, and the probability allocation algorithm guarantees the scientificity and interpretability of task allocation. Organic combination of these algorithm characteristics provides a theoretical basis and implementation path for dynamic configuration and scheduling of UAV work resources in complex multi-scene. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings based on these drawings without any creative effort.

[0022] Figure 1 is an embodiment schematic diagram of the multi-scene UAV work resource dynamic configuration and scheduling method in the embodiment of the present application;

[0023] Figure 2 is an embodiment schematic diagram of the multi-scene UAV work resource dynamic configuration and scheduling system in the embodiment of the present application;

[0024] Figure 3 is a structural schematic diagram of the multi-scene UAV work resource dynamic configuration and scheduling device in the embodiment of the present application. DETAILED DESCRIPTION

[0025] The embodiment of the present application provides a multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method and system. The terms first, second, third, fourth and the like (if any) in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term includes or has and any variation thereof is intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Please refer to Figure 1 One embodiment of the multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method in the embodiment of the present application comprises the following steps.

[0027] Step S1: performing semantic analysis and constraint formalization processing on a natural language instruction input by a user, multi-scene task constraint data, the multi-scene task constraint data comprising spatial coordinates, task weights and scene types;

[0028] Step S2: based on the spatial coordinates and the task weights in the multi-scene task constraint data, constructing a dynamic potential field matrix reflecting multi-scene task priority distribution through Gaussian function combination;

[0029] Step S3: taking the dynamic potential field matrix as a driving source, establishing a continuous motion velocity field of a group of unmanned aerial vehicles containing a pressure gradient term and a viscosity diffusion term;

[0030] Step S4: calculating a vortex influence coefficient according to an unmanned aerial vehicle capability evaluation value and a current position potential field intensity, generating a local vortex velocity component and fusing the local vortex velocity component with the continuous motion velocity field to generate a multi-scene adaptive control velocity field;

[0031] Step S5: constructing a task allocation vector field based on the multi-scene adaptive control velocity field and scene type characteristics, generating an unmanned aerial vehicle operation region configuration scheme through dynamic probability calculation, and converting the unmanned aerial vehicle operation region configuration scheme into real-time flight control instructions to complete closed-loop resource scheduling.

[0032] It can be understood that the execution subject of the present application can be a multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling system, and can also be a terminal or a server, and the specific execution subject is not limited here. The embodiment of the present application takes a server as an execution subject for example.

[0033] Specifically, semantic parsing and constraint formalization process first decomposes natural language instructions into lexical units through lexical analysis, and then performs syntactic analysis to identify grammatical structures and semantic relationships, thereby extracting scenario identifiers, task type descriptors, and constraint condition descriptors to obtain a set of semantic elements. The pre-trained domain knowledge base contains standardized parameter templates for various scenarios such as agricultural plant protection, logistics distribution, emergency rescue, etc. Through matching and mapping, semantic elements are mapped to specific numerical parameters. Spatial coordinate extraction obtains latitude and longitude values through geographic information analysis, task weight is calculated by priority evaluation algorithm to obtain numerical importance coefficient, and scenario type is determined by classification algorithm of scenario identifier to determine specific operation mode. Multi-scenario task constraint data is structured as output, containing the basic data required for all subsequent calculation steps.

[0034] Dynamic potential field matrix construction is based on the spatial distribution characteristics of Gaussian function. Single point potential field distribution is obtained by calculating the Euclidean distance between task point and each point in space, combined with the exponential decay characteristics of Gaussian function to obtain the spatial influence range. Task priority potential field multiplies the task weight as amplitude coefficient with single point potential field distribution to obtain weighted spatial intensity distribution. Multi-scenario task priority distribution is obtained by numerically superimposing all task priority potential fields in the same spatial coordinate system, reflecting the competition and cooperation relationship of different tasks in space. The influence radius parameter is automatically set according to the scenario type, agricultural plant protection scenario is set to a larger radius to cover the continuous operation area, and logistics distribution scenario is set to a smaller radius to accurately locate the delivery point. Matrix processing discretizes continuous spatial distribution into a calculable numerical matrix format.

[0035] Unmanned aerial vehicle group continuous motion velocity field is established based on fluid mechanics principle. Task driving force vector is obtained by gradient operation on dynamic potential field matrix. Gradient operation calculates the potential field difference between adjacent grid points and divides by spatial distance to obtain force vector pointing to high priority area. Pressure gradient term is obtained by counting the number density of unmanned aerial vehicles in local area. When the density exceeds the threshold, an outward repulsive force is generated to prevent unmanned aerial vehicles from gathering excessively in the same area. Viscosity diffusion term performs Laplacian operator operation on velocity field to eliminate discontinuity and ensure smooth transition of velocity in adjacent areas. Dynamics equation solving vector synthesizes three force components according to physical laws to obtain velocity vector of each spatial point, obtaining continuous velocity field describing the coordinated motion trend of unmanned aerial vehicle group in the whole operation space.

[0036] The multi-scene adaptive control velocity field generation is realized by a fusion algorithm of local vortex mechanism and global velocity field. The UAV capability evaluation value comprehensively considers multi-dimensional indexes such as residual power, load capacity, sensor accuracy, and the like, and obtains a comprehensive capability value through weighted summation. The current position potential field strength is directly obtained by table lookup from the dynamic potential field matrix according to the real-time coordinates of the UAV. The vortex influence coefficient is obtained by multiplying the capability evaluation value and the potential field strength, and then normalized by dividing by the total capability of all UAVs. In the local vortex field construction, the distance weight factor is normalized by dividing the radial distance by the maximum action distance, and the spatial attenuation coefficient adopts an exponential function form to ensure rapid attenuation of the influence at a long distance. The vortex intensity value is distributed around the UAV to obtain a rotationally symmetric force field distribution, and the tangential velocity component is calculated by multiplying the vortex intensity value by a unit tangential vector to obtain a comprehensive control field with global guidance and local avoidance functions through vector superposition with the continuous motion velocity field.

[0037] The task allocation vector field construction is based on the feature vector extraction algorithm of the multi-scene adaptive control velocity field and the scene type feature. The scene type feature is encoded by a predefined feature template, and the agricultural plant protection scene is encoded as area coverage priority, and the logistics distribution scene is encoded as point-to-point accuracy priority. The task type feature vector is multiplied point by point with the velocity field value to obtain a spatial distribution reflecting the task execution suitability. The allocation probability value calculation inputs the current position of the UAV into the task allocation vector field for numerical query, and combines the capability evaluation value for weighted processing, and performs probability normalization through the softmax function. The region attribution determination determines the target work region of the UAV based on the maximum probability principle, the control instruction generation converts the region coordinates into a flight point sequence, and determines the flight height, speed parameters and work parameters according to the scene type to obtain a standardized flight control instruction format.

[0038] In a specific embodiment, step S1 comprises:

[0039] The natural language instruction is subjected to morphological analysis and syntactic parsing to obtain a semantic element set containing a scene identifier, a task type descriptor and a constraint condition descriptor;

[0040] The semantic element set is input into a pre-trained domain knowledge base for matching and mapping to obtain a standardized task parameter combination;

[0041] Based on the standardized task parameter combination, spatial coordinates and task weights are extracted to obtain coordinate values and weight values;

[0042] According to the scene identifier, a scene classification process is performed to obtain a scene type, and the scene type is combined with the spatial coordinates and the task weights to obtain multi-scene task constraint data.

[0043] Specifically, the data processing mechanism of morphological analysis and syntactic parsing is based on a hierarchical architecture of natural language processing. Morphological analysis first splits the input continuous character stream according to the lexical boundary, identifies independent lexical units through the maximum matching algorithm and dictionary lookup, and assigns each word a part-of-speech tag such as noun, verb, adjective, etc. Syntactic parsing builds a dependency relation tree based on the results of morphological analysis, identifies the grammatical function and semantic role of each component in the sentence through subject-predicate-object relationship identification and modification relationship analysis. Scene identifier extraction uses a keyword matching algorithm to maintain a dictionary of keywords for various job scene features. When identifying keywords in the fields of agriculture, logistics, and rescue, record their position and context information in the sentence. Task type descriptor extraction focuses on identifying verb phrases that express specific job behaviors, such as monitoring, spraying, and distribution. Through verb semantic role labeling, determine its attributes as the core action of the task. Constraint condition descriptor extraction identifies time adverbs, spatial prepositions, and quantity adjectives to obtain limiting descriptions such as immediate, full coverage, and high precision. The semantic element set organizes and stores these three types of information in a structured format.

[0044] The matching mapping process of the pre-trained domain knowledge base uses a combination of semantic similarity calculation and template matching. Semantic similarity calculation converts each element in the semantic element set into a word vector representation, then calculates the cosine distance with the pre-stored standard term vector in the knowledge base. The smaller the distance value, the higher the semantic similarity. Template matching uses a conditional rule engine to find predefined parameter configuration templates based on the combination of scene identifiers and task type descriptors. Each template contains standard parameter settings and constraint conditions for the scene. When multiple matching results are obtained, a confidence weighted fusion algorithm is used to weight and average the parameters of each matching item according to their matching scores to obtain the final standardized task parameter combination. The standardized task parameter combination contains structured data such as spatial range parameters, priority coefficients, job mode identifiers, and resource demand quantities. Each parameter is labeled with its data type, value range, and unit information.

[0045] The spatial coordinate extraction and the task weight extraction are based on parameter analysis and quantitative evaluation algorithms. The spatial coordinate extraction identifies the geographic location related parameters from the standardized task parameter combination. When the parameters contain specific coordinate values, they are directly extracted. When the parameters contain geographic entity names, they are converted into latitude and longitude coordinates through a geocoding service. When the parameters contain relative position descriptions, they are deduced into absolute coordinates through a spatial reasoning algorithm combined with known location information. The task weight extraction is obtained by a multidimensional evaluation algorithm, which comprehensively analyzes the time urgency, spatial importance and resource consumption of the task. The time urgency is quantified by the difference between the task deadline and the current time. The spatial importance is evaluated by the economic value and ecological value of the work area. The resource consumption is calculated by the estimated working time and energy consumption of the unmanned aerial vehicle. The scores of the three dimensions are summed by weighting to obtain the final weight value. The weight value is normalized to ensure that it is within the range of zero to one.

[0046] The scene classification processing adopts a decision tree classification algorithm based on feature matching. The construction of the decision tree is based on the feature combination of the scene identifier and the preset classification rules. Agricultural scene classification is determined by detecting the presence and combination mode of agricultural feature words such as farmland, crops and plant protection. Logistics scene classification is identified by detecting logistics feature words such as distribution, transportation and warehousing. Emergency scene classification is confirmed by detecting emergency feature words such as rescue, disaster and emergency. When the input contains multiple scene features, the dominant scene type is determined by a feature weight voting mechanism. Each feature word is assigned a different voting weight according to its importance in the scene. The scene type with the highest final voting score is selected as the main scene. The scene type is represented in the form of an enumeration constant, including predefined types such as agricultural work, logistics distribution, emergency rescue and environmental monitoring. Each type is associated with specific work parameters and constraints. The multi-scene task constraint data is obtained by data encapsulation operation. The spatial coordinates, task weights and scene types are combined according to the predefined data structure to obtain a composite data object containing complete constraint information.

[0047] When receiving the natural language instruction of performing pest monitoring operation on the tea garden, the lexical analysis identifies four lexical units of tea garden as a noun, pest as a noun, monitoring as a verb, and operation as a noun, the syntactic analysis determines the tea garden as the operation object, the pest monitoring as the composite task type, and the operation as the execution instruction. The semantic element set extraction obtains the scene identifier as the tea garden agriculture, the task type descriptor as the pest monitoring, and the constraint condition descriptor as the regular patrol. In the pre-training domain knowledge base matching process, the tea garden agriculture has the highest similarity with the agricultural operation template, the pest monitoring is matched to the parameter configuration of the monitoring task, and the regular patrol is matched to the time periodicity constraint. The standardized task parameter combination generates detailed configurations including the boundary coordinates of the tea garden plot, the monitoring frequency requirement, and the flight path parameters. The spatial coordinate extraction converts the text description of the tea garden into a specific set of geographical boundary coordinates, and the task weight is obtained by evaluating the importance of the pest monitoring on the crop yield to a medium priority weight value. The scene classification processing classifies the tea garden agriculture into the agricultural scene type and associates it to the parameters such as the standard flight height and the monitoring sensor configuration of the agricultural operation. The finally generated multi-scene task constraint data integrates the accurate geographical range of the tea garden, the priority weight of the pest monitoring, and the operation mode identifier of the agricultural scene, obtaining the standardized input data format for subsequent potential field construction and unmanned aerial vehicle scheduling algorithm.

[0048] In an embodiment, step S2 comprises:

[0049] Based on the spatial coordinates in the multi-scene task constraint data, a Gaussian potential field contribution function of each task point is constructed to obtain a single-point potential field distribution;

[0050] According to the task weight, a priority weighting process is performed on the single-point potential field distribution of each task point to obtain a task priority potential field;

[0051] A plurality of task priority potential fields are combined by superimposition according to the spatial position to obtain a multi-scene task priority distribution;

[0052] Based on the scene type, a radius parameter is set and a matrix processing is performed on the multi-scene task priority distribution to obtain a dynamic potential field matrix.

[0053] Specifically, based on the spatial coordinate information in the multi-scene task constraint data, the coordinates of each task point are taken as the center position of the Gaussian function, and the spatial influence distribution centered on the task point is constructed by calculating the Euclidean distance from any point in space to the task center point. The Gaussian potential field contribution function adopts a two-dimensional Gaussian distribution model, in which the task point coordinates are taken as the mean parameter, and the standard deviation parameter controls the diffusion degree of the influence range. The closer the spatial position to the task center, the higher the potential field strength value. The farther the position, the potential field strength decays exponentially. The single-point potential field distribution is formed by calculating the Gaussian function value on the preset spatial grid point by point. Each grid point corresponds to a potential field strength value, forming a continuous distribution surface with the task point as the peak value. The calculation process of the single-point potential field distribution divides the space into a regular grid structure, and the grid accuracy determines the spatial resolution of the potential field calculation. Higher grid accuracy can more accurately describe the spatial variation characteristics of the potential field, but at the same time increases the calculation complexity.

[0054] The priority weighting process applies the task weight as a multiplication coefficient to each numerical point of the single-point potential field distribution. The task weight value is directly derived from the normalized weight value in the multi-scene task constraint data. The larger the weight value, the higher the task priority, and the larger the corresponding potential field strength. The weighting process is realized through matrix element-level multiplication operation, which multiplies the task weight with each element in the single-point potential field distribution matrix to obtain the adjusted potential field strength distribution. The task priority potential field maintains the spatial shape characteristics of the original single-point potential field distribution, but the overall intensity level is proportionally adjusted according to the task importance. The peak value of the potential field of high-priority tasks is higher, and the potential field strength in the influence range is also correspondingly enhanced. The priority weighting process solves the problem of being unable to distinguish the importance of tasks in the prior art, and realizes the quantitative expression and spatial mapping of task priority through the numerical weight mechanism.

[0055] The superposition combination of multiple task priority potential fields is realized by point-by-point numerical addition. When multiple tasks have overlapping influence ranges in space, the potential field strength in the overlapping area is equal to the sum of the numerical values of each task priority potential field at that position. The superposition process is realized by iterating all grid points and summing the potential field contributions of all tasks at each position to form a comprehensive multi-scene task priority distribution. The multi-scene task priority distribution reflects the comprehensive importance level of different regions in the entire work space. Regions with high task density and priority present high potential field strength, while regions with low task density or priority present low potential field strength. The spatial competition relationship between tasks is also considered in the superposition process. When multiple high-priority tasks are adjacent in space, their potential field superposition will form a stronger attraction area, guiding more unmanned aerial vehicle resources to concentrate in that area.

[0056] The influence radius parameter is set based on the different characteristics of different scenario types. In agricultural scenarios, due to the need for large-scale continuous operations, the influence radius is set to a larger value to cover a wide planting area. In logistics scenarios, which emphasize precise point-to-point delivery, the influence radius is set to a smaller value to achieve accurate positioning. In emergency scenarios, involving emergency rescue, the influence radius is dynamically adjusted according to the disaster area and rescue needs. The influence radius parameter directly affects the diffusion degree of the Gaussian potential field contribution function; a larger influence radius makes the potential field distribution more gradual and widespread, while a smaller influence radius makes the potential field distribution steeper and more concentrated. Matrix processing converts the continuous multi-scenario task priority distribution into a discrete numerical matrix format. The rows and columns of the matrix correspond to the coordinate indices of the spatial grid, and the matrix element values ​​correspond to the potential field intensity at the corresponding grid location. The dynamic potential field matrix, as the final output, contains the potential field distribution information of the entire workspace. The matrix dimension is determined by the precision of the spatial grid division, and the range of element values ​​reflects the distribution characteristics of task priorities.

[0057] When handling multi-scenario operations involving both orchard pest and disease monitoring and logistics delivery, the multi-scenario task constraint data provides the central coordinates and high-priority weights for the orchard monitoring task, and the target coordinates and medium-priority weights for the delivery task. During the construction of the Gaussian potential field contribution function, the orchard monitoring task constructs a Gaussian distribution with its central coordinates as the peak value. The influence radius is set to a relatively large value based on the agricultural scenario type to cover the entire orchard area, forming a broad and gentle single-point potential field distribution. The delivery task constructs another Gaussian distribution with its target coordinates as the center, and the influence radius is set to a smaller value based on the logistics scenario type to achieve precise positioning, forming a concentrated and steep single-point potential field distribution. Priority weighting multiplies the high-weight value of the orchard monitoring task with its single-point potential field distribution to obtain a high-intensity task priority potential field, while the medium-weight value of the delivery task is multiplied with its single-point potential field distribution to obtain a medium-intensity task priority potential field. During the superposition and combination process, the potential fields of the two tasks are numerically accumulated according to their spatial locations. In the orchard area, a high-intensity distribution dominated by the monitoring task is formed, while a medium-intensity distribution dominated by the delivery task is formed at the delivery target point. The potential field intensity in other areas of the space exhibits gradient changes according to the distance decay law. The final generated dynamic potential field matrix integrates the spatial distribution and priority information of the two different scenario tasks.

[0058] In one specific embodiment, step S3 includes:

[0059] The gradient components of the potential field are calculated based on the dynamic potential field matrix to obtain the task driving force vector pointing to the high-priority task region.

[0060] Gradient calculation of the pressure field is performed based on the spatial distribution density of the drone swarm to obtain the pressure gradient term that prevents excessive concentration of drones.

[0061] The velocity field is diffused according to the continuity requirement of the UAV group motion to obtain a viscosity diffusion term for keeping the velocity field smooth transition;

[0062] The task driving force vector, the pressure gradient term and the viscosity diffusion term are solved by a dynamic equation to obtain a UAV group continuous motion velocity field describing the coordinated motion of the UAV group in a multi-scene operation environment.

[0063] Specifically, spatial derivative operation is performed on the dynamic potential field matrix, and the gradient is calculated by calculating the difference between the potential field intensity of each grid point in the matrix and its adjacent grid points, and then dividing the grid spacing to obtain the gradient vector at the position. The task driving force vector is obtained by taking the negative value of the potential field gradient, because the gradient points to the direction of the fastest growth of the potential field, and the driving force needs to point to the high potential field area, that is, to the high priority task area. During the gradient calculation process, for a two-dimensional potential field matrix, the x-direction gradient of each internal grid point is obtained by subtracting the potential field value of the left grid point from the potential field value of the right grid point and then dividing by twice the grid spacing, and the y-direction gradient is obtained by subtracting the potential field value of the lower grid point from the potential field value of the upper grid point and then dividing by twice the grid spacing. The gradient calculation of the boundary grid point uses the forward difference or backward difference method to ensure that all grid points can obtain effective gradient values. The size of the task driving force vector reflects the attraction strength of the position to the high priority task area, and the direction points to the nearest high potential field area, forming a force field distribution guiding the UAV to move to the task center.

[0064] The UAV group spatial distribution density calculation is based on the real-time position information of all UAVs at the current time, and the kernel density estimation algorithm is used to calculate the UAV aggregation degree at each position on the spatial grid. The kernel density estimation uses a Gaussian kernel function to construct a Gaussian distribution with each UAV position as the center, and the superposition of the Gaussian distributions of all UAVs obtains the density distribution of the entire space. The pressure field gradient calculation takes the density distribution as the pressure distribution, and obtains the pressure gradient term by gradient operation on the density matrix. The pressure gradient points to the direction of reduced density, generating a repulsive force away from the high-density area. The calculation process of the pressure gradient term is similar to that of the potential field gradient, but the sign is opposite, because the pressure gradient needs to produce an outward diffusion effect. When the UAV density of a certain area is too high, the pressure gradient vector of the area points to the surrounding, pushing the UAV to diffuse to the area with lower density, preventing the UAV from over-aggregating in the same position and causing collision risk or resource waste. The strength of the pressure gradient term is proportional to the local UAV density, and the higher the density of the area, the stronger the repulsive force.

[0065] The velocity field diffusion process uses a Laplacian operator to smooth the current velocity field. The Laplacian operator calculates the velocity difference between each grid point and its surrounding neighbor grid points, and eliminates the mutations and discontinuities in the velocity field by weighted average. The calculation of the viscosity diffusion term is performed by performing a Laplacian operation on each component of the velocity field. The Laplacian value of the x-direction velocity component is equal to the x-direction velocity of the point minus the average value of the x-direction velocities of the four adjacent points. The y-direction velocity component is processed in the same way. The diffusion process ensures that drones at adjacent spatial positions have similar motion trends, avoiding unreasonable situations where adjacent drones move in completely opposite directions. The strength of the viscosity diffusion term is controlled by the viscosity coefficient. The larger the viscosity coefficient, the stronger the diffusion effect, and the smoother the velocity field. However, it will also reduce the response sensitivity of the velocity field to local task changes. The viscosity diffusion term simulates the viscous resistance effect in fluid, making the motion of the drone swarm exhibit continuous and coordinated characteristics similar to fluid.

[0066] The dynamics equation solving vector combines the task driving force vector, the pressure gradient term, and the viscosity diffusion term, and uses a simplified form based on the Navier-Stokes equation for numerical solution. The solving process of the dynamics equation first performs vector addition operation on the three force components at each grid point to obtain the resultant force vector at that position. Then, according to Newton's second law, the resultant force is divided by the equivalent mass to obtain the acceleration vector. Finally, the velocity vector is obtained by numerical integration. Numerical integration uses the Euler method or the Runge-Kutta method to calculate the velocity value at the next time based on the current velocity and acceleration. Boundary conditions need to be set during the solving process. Typically, a zero velocity boundary condition is set at the boundary of the work area to prevent drones from flying out of the work range. The continuous motion velocity field of the drone swarm is the solving result, which contains the velocity vector information at each position in the entire work space. The direction of the velocity vector indicates the motion direction of the drone at that position, and the size of the velocity vector indicates the speed of the motion.

[0067] When processing multi-scene tasks containing orchard monitoring and emergency distribution, the dynamic potential field matrix shows that the orchard area has a high potential field strength, and the distribution target point has a medium potential field strength. The potential field gradient calculation generates a task-driven force vector pointing to the center of the orchard in the surrounding area of the orchard, and a driving force vector pointing to the target point around the distribution target point. When multiple drones gather at the entrance of the orchard, the spatial distribution density of the drone swarm reaches a peak at this location, and the density gradient calculation generates a pressure gradient term that spreads to the interior and periphery of the orchard, pushing some drones to disperse to different areas of the orchard to perform monitoring tasks. The velocity field diffusion process ensures that adjacent monitoring drones maintain a coordinated cruising speed and flight direction, avoiding motion trajectories that interfere with each other between fruit tree rows. The distribution drone has a higher task urgency and a clear target, so it receives a stronger task-driven force vector and a relatively smaller pressure gradient term, and mainly moves along a straight path pointing to the distribution target. During the solution process of the kinetic equation, the vector composition of the three force components causes the monitoring drones to form an orderly cruising pattern in the orchard, and the distribution drones form a fast straight-through motion trajectory. The motion velocity field of the entire drone swarm reflects the coordinated execution of multi-scene tasks and the optimized allocation of resources.

[0068] In a specific embodiment, step S4 comprises:

[0069] Based on the weighted product operation of the UAV capability evaluation value and the current position potential field strength, the vortex influence coefficient of each UAV is obtained;

[0070] According to the vortex influence coefficient and the vortex influence radius, a local vortex field is constructed to obtain the rotational speed distribution of each UAV;

[0071] The tangential velocity component calculation is performed on the rotational speed distribution to obtain the local vortex velocity component for preventing UAV collision and overlap;

[0072] The local vortex velocity component is vector superimposed and fused with the continuous motion velocity field of the UAV swarm to obtain a multi-scene adaptive control velocity field with global coordination and local avoidance functions.

[0073] Specifically, based on the comprehensive calculation of multi-dimensional capability indicators, the capability evaluation value is obtained by weighted summation of key parameters such as residual power, load capacity, sensor accuracy, flight stability, etc. The residual power value is directly obtained from the unmanned aerial vehicle battery management sensor, the load capacity is calculated according to the ratio of the current carrying equipment weight to the maximum load, the sensor accuracy is determined by the equipment specification parameters, and the flight stability is evaluated by the GPS positioning accuracy and attitude sensor data. The current position potential field strength is directly obtained from the dynamic potential field matrix according to the real-time coordinates of the unmanned aerial vehicle, and the coordinate index is realized by converting the geographic coordinates into matrix row and column indexes. The vortex influence coefficient is obtained by multiplying the unmanned aerial vehicle capability evaluation value with the current position potential field strength, and then dividing by the sum of the products of all unmanned aerial vehicle capability evaluation values and their corresponding potential field strengths for normalization processing, ensuring that the sum of the vortex influence coefficients of all unmanned aerial vehicles is equal to one. The unmanned aerial vehicle with high capability and located in the high potential field strength position obtains a larger vortex influence coefficient, indicating that it has stronger influence and priority execution in the local area. The numerical range of the vortex influence coefficient is between zero and one, and the numerical distribution reflects the capability difference and location advantage within the unmanned aerial vehicle group.

[0074] The local vortex field is constructed based on the vortex influence coefficient and the preset vortex influence radius parameter, and the vortex influence radius is determined according to the physical size of the unmanned aerial vehicle, the safety distance requirement and the operation accuracy requirement. The local vortex field adopts a vortex function model, taking the current position of the unmanned aerial vehicle as the center, and constructing a rotationally symmetric velocity distribution within the vortex influence radius. The vortex intensity is zero at the center position, increases with the increase of the radial distance, reaches a peak at a certain radius, and then decays with the increase of the distance. The rotational velocity distribution is obtained by calculating the angular velocity and radial distance of each spatial point in the vortex field. The angular velocity is determined by the vortex influence coefficient and the vortex intensity function, and the radial distance is calculated by the Euclidean distance from the point to the center of the unmanned aerial vehicle. The rotational velocity distribution presents a ring structure, the area close to the center of the unmanned aerial vehicle has smaller rotational velocity, the medium distance area has the maximum rotational velocity, and the rotational velocity of the far distance area gradually decays to zero. The direction of the rotational velocity distribution follows the right-hand rule, forming a clockwise or counterclockwise rotation mode.

[0075] The tangential velocity component calculation converts the rotational velocity distribution into a velocity vector component in rectangular coordinates, and the tangential direction is determined by the perpendicular direction of the radial vector. For any point in the vortex field, first calculate the radial vector of the point relative to the vortex center, then rotate the radial vector counterclockwise by ninety degrees to get the tangential unit vector, and the tangential velocity size is equal to the rotational velocity value of the point. The local vortex velocity component is obtained by multiplying the tangential velocity size with the tangential unit vector, forming a two-dimensional velocity vector with x and y components. The calculation of the vortex velocity component ensures that the adjacent UAVs produce a mutual circling motion trend, and when the distance between the two UAVs is too close, the interaction of their respective vortex fields produces an avoidance effect. The size of the tangential velocity component is proportional to the vortex influence coefficient, and the stronger the UAV, the stronger the vortex field it produces, and the greater the repulsive effect on the surrounding UAVs. The vortex velocity component also has a distance attenuation characteristic, and the influence on the position away from the vortex center gradually weakens, ensuring that the vortex effect only plays a role in a local range.

[0076] The vector superposition fusion performs point-by-point vector addition operation on the local vortex velocity component and the continuous motion velocity field of the UAV group, and the final velocity of each spatial grid point is equal to the continuous motion velocity vector of the point plus the vortex velocity component vector generated by all UAVs at the point. The superposition process needs to traverse all the vortex fields of the UAVs, calculate the contribution of each vortex field at the target grid point, and then add it to the corresponding component of the continuous motion velocity field. The multi-scene adaptive control velocity field, as the fusion result, has both global coordination function and local avoidance function. The global coordination function comes from the response of the continuous motion velocity field to the task priority, and the local avoidance function comes from the processing of the conflict between UAVs by the vortex velocity component. The fused velocity field maintains a strong attractive effect in the high-priority task area, produces appropriate dispersion effect in the UAV dense area, and produces circling effect in the UAV proximity area. The numerical distribution of the velocity field reflects the balance between multi-scene task demand and UAV group coordination constraint.

[0077] In a specific embodiment, the local vortex field is constructed according to the vortex influence coefficient and the vortex influence radius, and the rotational velocity distribution of each UAV is obtained, including:

[0078] The radial distance is normalized based on the vortex influence coefficient to obtain a distance weight factor;

[0079] The vortex action range is set according to the vortex influence radius, and the distance weight factor is calculated by exponential decay to obtain a spatial attenuation coefficient;

[0080] The vortex influence coefficient and the spatial attenuation coefficient are multiplied to obtain the vortex intensity value of each spatial position;

[0081] The radial velocity component and the tangential velocity component of the surrounding space points are decomposed based on the vortex intensity value, to obtain a rotational velocity distribution describing the rotational motion mode around the UAV.

[0082] Specifically, based on the vortex influence coefficient as a standardized benchmark, the actual distance from each space point around the UAV to the center of the UAV is divided by the vortex influence coefficient to perform a standardization operation. The radial distance is obtained by calculating the Euclidean distance between the coordinates of the space grid points and the current position coordinates of the UAV. The Euclidean distance calculation uses the distance formula between two points, which adds the square of the x-coordinate difference and the square of the y-coordinate difference and then takes the square root. The actual radial distance is divided by the vortex influence coefficient in the normalization process. The larger the vortex influence coefficient of the UAV, the smaller the normalized distance value of the same actual distance point around the UAV. The smaller the vortex influence coefficient of the UAV, the larger the normalized distance value of the same actual distance point around the UAV. The distance weight factor is calculated by the reciprocal of the normalized distance. The closer the point, the larger the weight factor, and the farther the point, the smaller the weight factor. The calculation of the distance weight factor ensures that the UAV with strong ability and located in the high potential field region has a stronger influence on the surrounding space points, reflecting the adjusting effect of the UAV ability difference on the vortex field distribution.

[0083] The vortex action range is set based on the vortex influence radius parameter, and the vortex influence radius is determined comprehensively according to the safe flight distance of the UAV, the sensor detection range and the operation accuracy requirement. The exponential decay calculation adopts an exponential function to perform decay processing on the distance weight factor. The exponential decay function is constructed in the form of a negative exponential, which ensures that the distance weight factor quickly decays with the increase of the radial distance. The base of the exponential decay is usually the base of the natural logarithm, and the parameter of the exponential is determined by the ratio of the radial distance to the vortex influence radius. When the radial distance is equal to the vortex influence radius, the decay coefficient decreases to a certain proportion of the original value. When the radial distance exceeds the vortex influence radius, the decay coefficient tends to zero. The spatial decay coefficient is obtained by multiplying the distance weight factor by the exponential decay function. The decay coefficient remains a high value near the center of the UAV, sharply decreases at the boundary of the vortex influence radius, and tends to zero outside the boundary. The distribution characteristics of the spatial decay coefficient ensure that the vortex effect only works within the predetermined range, avoiding unnecessary mutual influence at a long distance.

[0084] The vortex intensity value calculation performs point-by-point multiplication operation of the vortex influence coefficient and the spatial decay coefficient. The multiplication operation is independently performed at each spatial grid point. The vortex influence coefficient is applied as a global coefficient to all grid points within the influence range of the UAV. The spatial decay coefficient is adjusted as a position-dependent local coefficient to adjust the vortex intensity at different positions. The numerical distribution of the vortex intensity value presents a ring structure centered on the UAV. The vortex intensity value at the central position is zero. As the radial distance increases, the vortex intensity value first increases and then decreases. At a certain radius, the vortex intensity value reaches a peak value, and then decays to zero as the distance increases. The peak position of the vortex intensity value is determined by the vortex influence coefficient and the decay function parameters. Generally, it is located at a certain proportion of the vortex influence radius. The size of the vortex intensity value directly affects the rotational speed at that position. The larger the intensity value, the faster the rotational motion at that position. The position with zero intensity value does not produce rotational effect. The distribution of the vortex intensity value at each spatial position forms a two-dimensional intensity field describing the rotational motion intensity around the UAV.

[0085] The velocity component decomposition converts the vortex intensity at each spatial point into radial and tangential velocity components based on the vortex intensity value and spatial geometric relationship. The radial velocity component calculates the projection of the vortex intensity value in the radial direction, which is defined as the direction from the center of the UAV to the target spatial point. The radial velocity component is usually set to zero because the vortex motion mainly manifests as tangential rotation rather than radial diffusion or contraction. The tangential velocity component calculates the projection of the vortex intensity value in the tangential direction, which is defined as the direction rotated ninety degrees counterclockwise from the radial direction. The tangential velocity is equal to the vortex intensity value at that point. The direction of the tangential velocity component follows the right-hand rule, forming a rotational velocity field around the center of the UAV. The rotational velocity distribution is formed by combining the velocity components of all spatial points. The velocity distribution presents a spiral or annular streamline structure. The streamline density reflects the size of the rotational velocity, and the streamline direction reflects the direction of the rotational motion. The calculation of the rotational velocity distribution solves the problem of rigidity in the existing UAV avoidance mechanism, and generates a smooth bypass trajectory through the continuous rotational velocity field.

[0086] In a specific embodiment, step S5 comprises:

[0087] Based on the multi-scene adaptive control velocity field and the scene type, a feature vector is extracted to obtain a task type feature vector;

[0088] According to the task type feature vector, a task allocation weight calculation is performed on each point in the space to obtain a task allocation vector field;

[0089] The current position and capability evaluation value of the UAV are input into the task allocation vector field to perform probability matrix operation, and the allocation probability value of each UAV corresponding to the task area is obtained;

[0090] Based on the allocation probability value, the region attribution judgment and the control instruction generation are performed to obtain a UAV operation region configuration scheme containing a flight path and operation parameters, and the scheme is converted into real-time flight control instructions.

[0091] Specifically, based on the numerical distribution characteristics of the multi-scenario adaptive control velocity field and the semantic identification of the scenario type, the feature vector extraction adopts a method combining principal component analysis and feature encoding. The multi-scenario adaptive control velocity field contains the velocity vector information of each grid point in the space. The feature extraction first performs statistical analysis on the velocity field, calculates the mean, variance, maximum and minimum of the velocity size, and then calculates the distribution characteristics of the velocity direction, including the main flow direction, divergence degree and rotation intensity. The feature encoding of the scenario type encodes the agricultural scenario as a feature vector with area coverage priority, encodes the logistics scenario as a feature vector with point-to-point accuracy priority, and encodes the emergency scenario as a feature vector with time sensitivity priority. The task type feature vector is obtained by weighted fusion of the velocity field statistical characteristics and the scenario type encoding. The weighted coefficients are determined according to the importance of the velocity field characteristics in different scenarios. The agricultural scenario pays more attention to the uniformity of the velocity field coverage, the logistics scenario pays more attention to the convergence of the velocity field, and the emergency scenario pays more attention to the rapid response of the velocity field. The dimensions of the feature vector include key indicators such as spatial coverage, motion coordination, task responsiveness and resource utilization, each of which is quantified by specific numerical values.

[0092] The task allocation weight calculation is based on the adaptive evaluation of each point in the space by the task type feature vector. The weight calculation adopts an algorithm combining distance weighting and feature matching. The task allocation weight of each grid point in the space is obtained by calculating the similarity between the velocity characteristics of the point and the task type feature vector. The similarity calculation adopts cosine similarity or the reciprocal of Euclidean distance as the measurement standard. The distance weighting considers the physical distance from the space point to the task center. The closer the point, the higher the basic weight, and the farther the point, the lower the basic weight. The distance weight is calculated by a Gaussian decay function. The feature matching weight is based on the matching degree of the velocity characteristics of the point and the task demand. The matching of velocity size and task intensity demand, the matching of velocity direction and task execution direction, and the matching of velocity change rate and task dynamics all affect the feature matching weight. The task allocation vector field is obtained by multiplying the distance weight and the feature matching weight. The value of each point in the vector field represents the suitability of the location for executing the corresponding task. The higher the value, the more suitable the location for assigning to the UAV to execute the task. The distribution of the task allocation vector field reflects the adaptability differences of different regions in the space to different types of tasks.

[0093] The probability matrix operation takes the current position and capability evaluation value of the UAV as input parameters, and queries and calculates in the task allocation vector field. The current position of the UAV is obtained through the GPS coordinates, the coordinates are converted into the grid index of the task allocation vector field, and the task suitability value of the position is extracted from the vector field. The capability evaluation value includes the comprehensive indexes of the remaining power, load capacity, sensor accuracy and flight stability of the UAV, and a single capability evaluation value is obtained by weighted summation. The probability calculation adopts the softmax function to perform exponential transformation and normalization processing on the suitability values of the UAV in each task area, so as to ensure that the sum of the allocation probabilities of all task areas is equal to one. The calculation formula of the allocation probability value is that the suitability value of a certain task area is multiplied by the capability evaluation value, and then the exponential is taken, and then divided by the sum of the exponents of the corresponding values of all task areas. The rows of the probability matrix correspond to different UAVs, the columns correspond to different task areas, and the matrix elements represent the probability value of the corresponding UAV allocated to the corresponding task area. The probability matrix operation solves the problem that the task allocation in the prior art lacks quantitative basis, and quantifies the rationality and priority of the task allocation through the probability.

[0094] The region attribution determination adopts the maximum probability principle to select the task area with the highest probability value as the allocation target for each UAV. The determination process traverses each row of the probability matrix, finds the element with the maximum value in the row, and the corresponding column index is the allocation task area of the UAV. The control instruction generation is based on the coordinates of the allocated task area and the scene type parameter to generate a complete instruction set including the waypoint sequence, flight height, flight speed and operation parameter. The waypoint sequence is generated by a path planning algorithm, and the optimal flight path from the current position of the UAV to the target task area is considered in the path planning. The flight height is automatically set according to the scene type, the agricultural scene is set to a safe height above the crop canopy, the logistics scene is set to a transport height avoiding buildings, and the emergency scene is set to a suitable height for observation and rescue. The operation parameters include specific operation settings such as sensor configuration, data acquisition frequency and flight mode, which are matched and configured according to the task type and UAV capability. The UAV operation area configuration scheme is the final output, which integrates the task allocation results and execution parameters of all UAVs. The real-time flight control instruction is sent to the UAV flight control through a standardized communication protocol, and the instruction format conforms to the mainstream UAV control interface specifications such as MAVSDK or ROS.

[0095] The above describes the multi-scene-oriented UAV operation resource dynamic configuration and scheduling method in the embodiments of the application, and the following describes the multi-scene-oriented UAV operation resource dynamic configuration and scheduling system in the embodiments of the application. Please refer to Figure 2 The multi-scene-oriented UAV operation resource dynamic configuration and scheduling system in the embodiments of the application includes one embodiment:

[0096] The generating module is configured to perform semantic analysis and constraint formalization processing on the natural language instruction input by the user, and generate multi-scene task constraint data, wherein the multi-scene task constraint data comprises spatial coordinates, task weights and scene types.

[0097] The constructing module is configured to construct a dynamic potential field matrix reflecting multi-scene task priority distribution by Gaussian function combination based on the spatial coordinates and the task weights in the multi-scene task constraint data.

[0098] The analyzing module is configured to take the dynamic potential field matrix as a driving source to establish a continuous motion velocity field of the UAV group containing a pressure gradient term and a viscosity diffusion term.

[0099] The fusion module is configured to calculate a vortex influence coefficient according to a UAV capability evaluation value and a current position potential field intensity, generate a local vortex velocity component, and fuse the local vortex velocity component with the continuous motion velocity field to obtain a multi-scene adaptive control velocity field.

[0100] The conversion module is configured to construct a task allocation vector field based on the multi-scene adaptive control velocity field and scene type characteristics, generate a UAV operation area configuration scheme by dynamic probability calculation, and convert the UAV operation area configuration scheme into real-time flight control instructions to complete closed-loop resource scheduling.

[0101] The above Figure 2 The multi-scene UAV operation resource dynamic configuration and scheduling system in the embodiment of the present application is described in detail from the perspective of a modular functional entity, and the multi-scene UAV operation resource dynamic configuration and scheduling device in the embodiment of the present application is described in detail from the perspective of hardware processing.

[0102] Referring to Figure 3 In the embodiment of the present application, a multi-scene UAV operation resource dynamic configuration and scheduling device is also provided, which can be a server, and the internal structure thereof can be as shown in Figure 3The multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling device is used to store the corresponding data in this embodiment. The network interface of the multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to realize the above method.

[0103] Those skilled in the art can understand that, Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the present application scheme, and does not constitute a limitation on the multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling device to which the present application scheme is applied.

[0104] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium, and can also be a volatile computer readable storage medium. The computer readable storage medium stores instructions, and when the instructions run on the computer, the computer executes the steps of the multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method.

[0105] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-mentioned system, system and unit can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0106] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the entire or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0107] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-scenario oriented dynamic configuration and scheduling method for UAV operation resources, characterized in that, The method comprises: Step S1: semantic analysis and constraint formalization processing are performed on a natural language instruction input by a user, multi-scene task constraint data are generated, and the multi-scene task constraint data comprise spatial coordinates, task weights and scene types; Step S2: a dynamic potential field matrix reflecting multi-scene task priority distribution is constructed by means of Gaussian function combination based on the spatial coordinates and the task weights in the multi-scene task constraint data; Step S3: the dynamic potential field matrix is taken as a driving source to establish a continuous motion velocity field of a UAV group containing a pressure gradient term and a viscosity diffusion term, comprising: a task driving force vector pointing to a high-priority task area is obtained by calculating a potential field gradient component according to the dynamic potential field matrix; the pressure gradient term for preventing excessive concentration of the UAV group is obtained by gradient calculation of a pressure field based on spatial distribution density of the UAV group; the viscosity diffusion term for keeping the velocity field smooth transition is obtained by diffusion processing of the velocity field according to the continuity requirement of the UAV group motion; the task driving force vector, the pressure gradient term and the viscosity diffusion term are solved by means of a dynamics equation to obtain the continuous motion velocity field of the UAV group describing the coordinated motion of the UAV group in a multi-scene operation environment; Step S4: a vortex influence coefficient is calculated according to a UAV capability evaluation value and a current position potential field intensity, a local vortex velocity component is generated and fused with the continuous motion velocity field to obtain a multi-scene adaptive control velocity field, comprising: the vortex influence coefficient of each UAV is obtained by weighted product operation based on the UAV capability evaluation value and the current position potential field intensity; the rotation velocity distribution of each UAV is obtained by constructing a local vortex field according to the vortex influence coefficient and a vortex influence radius; the local vortex velocity component for preventing collision and overlap of the UAVs is obtained by tangential velocity component calculation of the rotation velocity distribution; the local vortex velocity component is fused with the continuous motion velocity field of the UAV group by vector superposition to obtain the multi-scene adaptive control velocity field having both global coordination and local avoidance functions; Wherein, the UAV capability evaluation value is obtained by weighted summation based on residual power, load capacity and sensor accuracy indicators; the vortex influence coefficient is obtained by multiplying the UAV capability evaluation value with the current position potential field intensity, and then normalized by dividing by the sum of the products of all UAV capability evaluation values and their corresponding potential field intensities; Step S5: a task allocation vector field is constructed based on the multi-scene adaptive control velocity field and scene type characteristics, a UAV operation area configuration scheme is generated by dynamic probability calculation, and real-time flight control instructions are converted to complete closed-loop resource scheduling.

2. The multi-scene oriented dynamic configuration and scheduling method of UAV operation resources according to claim 1, characterized in that, The step S1 comprises: The natural language instruction is subjected to morphological analysis and syntactic analysis to obtain a semantic element set containing scene identifiers, task type descriptors and constraint condition descriptors; The semantic element set is input into a pre-trained domain knowledge base for matching and mapping to obtain a standardized task parameter combination; The spatial coordinates and the task weights are extracted based on the standardized task parameter combination to obtain coordinate values and weight values. According to the scene identifier, a scene classification process is performed to obtain the scene type, and the scene type is combined with the spatial coordinates and the task weight to obtain the multi-scene task constraint data.

3. The multi-scene oriented dynamic configuration and scheduling method of UAV operation resources according to claim 1, characterized in that, The step S2 comprises: Based on the spatial coordinates in the multi-scene task constraint data, a Gaussian potential field contribution function of each task point is constructed to obtain a single-point potential field distribution; According to the task weight, a priority weighting process is performed on the single-point potential field distribution of each task point to obtain a task priority potential field; A plurality of task priority potential fields are superimposed and combined according to spatial positions to obtain a multi-scene task priority distribution; Based on the scene type, an influence radius parameter is set, and a matrix processing is performed on the multi-scene task priority distribution to obtain the dynamic potential field matrix.

4. The multi-scene oriented dynamic configuration and scheduling method of UAV operation resources according to claim 1, characterized in that, The construction of the local vortex field according to the vortex influence coefficient and the vortex influence radius comprises: Based on the vortex influence coefficient, a radial distance is normalized to obtain a distance weight factor; According to the vortex influence radius, a vortex action range is set, and an exponential decay calculation is performed on the distance weight factor to obtain a spatial decay coefficient; The vortex influence coefficient and the spatial decay coefficient are multiplied to obtain a vortex intensity value of each spatial position; Based on the vortex intensity value, a radial velocity component and a tangential velocity component of the surrounding space points are decomposed to obtain the rotational velocity distribution describing the rotational motion mode around the unmanned aerial vehicle.

5. The multi-scene oriented dynamic configuration and scheduling method of UAV operation resources according to claim 1, characterized in that, The step S5 comprises: Based on the multi-scene adaptive control velocity field and the scene type, a feature vector is extracted to obtain a task type feature vector; According to the task type feature vector, a task allocation weight calculation is performed on each point in the space to obtain the task allocation vector field; The current position and the capability evaluation value of the unmanned aerial vehicle are input into the task allocation vector field to perform a probability matrix operation to obtain an allocation probability value of each unmanned aerial vehicle corresponding to a task region; Based on the allocation probability value, a region attribution judgment and a control instruction generation are performed to obtain the unmanned aerial vehicle operation region configuration scheme including a flight path and an operation parameter, and the unmanned aerial vehicle operation region configuration scheme is converted into the real-time flight control instruction. 6.A multi-scenario oriented unmanned aerial vehicle (UAV) operation resource dynamic configuration and scheduling system, characterized in that, The multi-scene oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling system for implementing the multi-scene oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling method comprises: A generation module is configured to perform semantic analysis and constraint formalization processing on a natural language instruction input by a user to generate multi-scene task constraint data, wherein the multi-scene task constraint data comprises spatial coordinates, a task weight, and a scene type; A construction module is configured to construct a dynamic potential field matrix reflecting a multi-scene task priority distribution by Gaussian function combination based on the spatial coordinates and the task weight in the multi-scene task constraint data; An analysis module is configured to take the dynamic potential field matrix as a driving source to establish a continuous motion velocity field of a swarm of unmanned aerial vehicles including a pressure gradient term and a viscosity diffusion term; A fusion module is configured to calculate a vortex influence coefficient according to a capability evaluation value and a current position potential field intensity of an unmanned aerial vehicle, generate a local vortex velocity component, and fuse the local vortex velocity component with the continuous motion velocity field to obtain a multi-scene adaptive control velocity field. The conversion module is used for constructing a task allocation vector field based on the multi-scene adaptive control speed field and the scene type feature, generating a UAV operation area configuration scheme through dynamic probability calculation, and converting the UAV operation area configuration scheme into real-time flight control instructions to complete closed-loop resource scheduling.

7. A multi-scene-oriented unmanned aerial vehicle operation resource dynamic configuration and scheduling device, characterized in that, The computer program is stored in the memory and can be run on the processor, and the processor implements the multi-scene-oriented UAV operation resource dynamic configuration and scheduling method in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is stored in the memory and can be run on the processor, and the processor implements the multi-scene-oriented UAV operation resource dynamic configuration and scheduling method in any one of claims 1 to 5 when executing the computer program.

Images