Unmanned aerial vehicle flight path planning method based on air-space linkage

Abstract

The invention provides an unmanned aerial vehicle flight path planning method based on air-space linkage, and the method comprises the steps: carrying out the multi-time-domain air-space data fusion, double-layer asynchronous iterative planning, multi-constraint joint optimization, distributed computing resource scheduling, task perception path generation, and other innovative technologies; the problems of limited environment perception capability, high calculation complexity, low resource allocation efficiency, insufficient real-time track adjustment capability and the like in the prior art are solved, and the flight performance and task execution efficiency of the unmanned aerial vehicle in a complex environment are improved.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a UAV flight trajectory planning method based on air-space linkage. Background Art

[0002] With the rapid development of drone technology, it is widely used in disaster monitoring, search and rescue, border patrol, and precision agriculture. In complex environments and under changing mission conditions, flight trajectory planning has become a key technology to ensure safe and efficient operation. Existing technologies face the following major challenges: inefficient multi-source information fusion, which fails to fully utilize the complementary nature of airborne and space-based information; trajectory planning methods lack adaptability, making it difficult to balance real-time performance and multi-constraint optimization; inefficient computational resource allocation and insufficient multi-platform collaborative scheduling; insufficient multi-objective collaborative optimization, making it difficult to achieve global optimal trajectory; and limited real-time adjustment capabilities in complex environments, making it difficult to respond to emergencies.

[0003] Therefore, there is an urgent need to develop a UAV trajectory planning method that can integrate aerospace information, take into account multi-constraint optimization, and support multi-task adaptation, so as to improve flight performance and mission execution efficiency in complex environments. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method for planning the flight trajectory of a UAV based on air-space linkage, which specifically includes the following steps:

[0005] Step S1, system initialization, including: loading mission parameters and constraints, initializing the sensing units of the air-based platform and the space-based platform, establishing a computing resource pool and performing initial resource allocation based on resource capabilities and communication bandwidth;

[0006] Step S2, multi-temporal air-space data acquisition and fusion, includes: collecting multi-source heterogeneous data including terrain data, obstacle information, and meteorological data from air-based and space-based platforms, decoupling the data in the temporal domain according to the frequency of change, processing the long-temporal domain data, medium-temporal domain data, and short-temporal domain data separately, and generating a unified environmental perception model using a multi-temporal domain fusion algorithm based on spatiotemporal consistency assessment;

[0007] Step S3, environmental analysis and risk assessment: Based on the environmental perception model generated in step S2, an environmental model including a terrain model, an obstacle model, and a weather model is constructed to perform static and dynamic risk assessments, generating a risk heat map representing the spatial risk distribution and an analysis result of the traversable area representing the flyable space;

[0008] Step S4, two-layer asynchronous iterative trajectory planning, using the risk assessment results of step S3, uses an improved A* algorithm that takes into account distance, risk, and energy consumption to generate a global path and flight corridor at the macro level. At the micro level, the trajectory is represented by a Bézier curve within the flight corridor and local trajectory optimization is performed through nonlinear model predictive control. Inter-layer information exchange and asynchronous iterative updates are achieved through an event-driven message passing mechanism.

[0009] Step S5, multi-constraint joint optimization trajectory adjustment, based on the initial trajectory generated in step S4, constructs a multi-objective optimization model that takes into account safety, energy efficiency, communication quality, and mission requirements. A modified non-dominated sorting genetic algorithm is used to solve the Pareto optimal trajectory set, and the final trajectory solution is selected based on the decision preference according to the task priority.

[0010] Step S6, dynamic allocation of computing resources, representing the computing tasks from step S2 to step S5 as a directed acyclic graph with dependencies, assigning tasks based on task computational complexity, timeliness requirements, resource processing capabilities, and energy status, and implementing an adaptive load balancing strategy based on load balancing metrics and task migration judgment;

[0011] Step S7, trajectory execution and real-time adjustment: Execute the final trajectory generated in step S5. Monitor the flight status and environmental changes in real time through onboard sensors. Based on the degree of deviation and the magnitude of environmental changes, a hierarchical adjustment strategy, including trajectory fine-tuning, local obstacle avoidance, local replanning, and global replanning, is adopted. If necessary, a replanning mechanism is triggered to return to step S4 or step S5 to update the trajectory.

[0012] The above steps form a complete closed-loop system. There is a tight data flow and control flow between each step, which cooperate with each other to achieve efficient planning and dynamic optimization of the UAV trajectory.

[0013] Beneficial effects of the present invention:

[0014] 1. Improve environmental perception capabilities: Through a multi-temporal decoupled aerospace data fusion preprocessing mechanism, heterogeneous data in the long, medium and short time domains can be processed simultaneously to improve the accuracy and timeliness of environmental perception and provide a reliable information basis for trajectory planning.

[0015] 2. Reduce computational complexity: A two-layer asynchronous iterative trajectory planning algorithm is adopted. The macro layer is responsible for global planning, and the micro layer is responsible for local optimization. Through efficient collaboration through asynchronous iteration, planning quality is guaranteed while improving computational efficiency.

[0016] 3. Optimize resource utilization: A distributed computing framework for dynamic allocation of aerospace resources is proposed, which dynamically allocates tasks to air-based and space-based resources. Combined with an adaptive load balancing strategy, it achieves efficient resource utilization and reduces system power consumption.

[0017] 4. Achieve multi-constraint balance: Based on a multi-constraint joint optimization trajectory fine-tuning method, a comprehensive optimization model is constructed to balance safety, energy consumption, communication quality, and mission effectiveness, comprehensively improving the UAV's flight performance.

[0018] 5. Enhanced mission adaptability: Based on task-aware adaptive trajectory generation methods, corresponding optimization objectives and trajectory patterns are adopted for different task types, significantly improving mission execution efficiency and success rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flow chart showing the steps of the method of the present invention is shown; DETAILED DESCRIPTION

[0020] Exemplary embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0021] Combine Figure 1 The present invention provides a method for planning the flight trajectory of a UAV based on air-space linkage, which specifically includes the following steps:

[0022] Step S1, system initialization, including: loading mission parameters and constraints, initializing the sensing units of the air-based platform and the space-based platform, establishing a computing resource pool and performing initial resource allocation based on resource capabilities and communication bandwidth;

[0023] Step S2, multi-temporal air-space data acquisition and fusion, includes: collecting multi-source heterogeneous data including terrain data, obstacle information, and meteorological data from air-based and space-based platforms, decoupling the data in the temporal domain according to the frequency of change, processing the long-temporal domain data, medium-temporal domain data, and short-temporal domain data separately, and generating a unified environmental perception model using a multi-temporal domain fusion algorithm based on spatiotemporal consistency assessment;

[0024] Step S3, environmental analysis and risk assessment: Based on the environmental perception model generated in step S2, an environmental model including a terrain model, an obstacle model, and a weather model is constructed to perform static and dynamic risk assessments, generating a risk heat map representing the spatial risk distribution and an analysis result of the traversable area representing the flyable space;

[0025] Step S4, two-layer asynchronous iterative trajectory planning, using the risk assessment results of step S3, uses an improved A* algorithm that takes into account distance, risk, and energy consumption to generate a global path and flight corridor at the macro level. At the micro level, the trajectory is represented by a Bézier curve within the flight corridor and local trajectory optimization is performed through nonlinear model predictive control. Inter-layer information exchange and asynchronous iterative updates are achieved through an event-driven message passing mechanism.

[0026] Step S5, multi-constraint joint optimization trajectory adjustment, based on the initial trajectory generated in step S4, constructs a multi-objective optimization model that takes into account safety, energy efficiency, communication quality, and mission requirements. A modified non-dominated sorting genetic algorithm is used to solve the Pareto optimal trajectory set, and the final trajectory solution is selected based on the decision preference according to the task priority.

[0027] Step S6, dynamic allocation of computing resources, representing the computing tasks from step S2 to step S5 as a directed acyclic graph with dependencies, assigning tasks based on task computational complexity, timeliness requirements, resource processing capabilities, and energy status, and implementing an adaptive load balancing strategy based on load balancing metrics and task migration judgment;

[0028] Step S7, trajectory execution and real-time adjustment: Execute the final trajectory generated in step S5. Monitor the flight status and environmental changes in real time through onboard sensors. Based on the degree of deviation and the magnitude of environmental changes, a hierarchical adjustment strategy, including trajectory fine-tuning, local obstacle avoidance, local replanning, and global replanning, is adopted. If necessary, a replanning mechanism is triggered to return to step S4 or step S5 to update the trajectory.

[0029] The above steps form a complete closed-loop system. There is a tight data flow and control flow between each step, which cooperate with each other to achieve efficient planning and dynamic optimization of the UAV trajectory.

[0030] Preferably, in step S2, the multi-time domain decoupled air-space data fusion preprocessing mechanism specifically includes the following implementation methods:

[0031] First, the data is divided into three temporal layers: long-term data (changes over a period of more than one hour, such as topography and static obstacles), medium-term data (changes over a period of one to sixty minutes, such as weather changes and regional activities), and short-term data (changes over a period of less than one minute, such as dynamic obstacles and temporary no-fly zones). This layered processing strategy allows for different processing methods to be used for environmental information with different rates of change, improving data processing efficiency.

[0032] Secondly, establish the data characteristic model matrix M(s) = [ρ s ,τ s ,σ s ,δ s ] T , where ρ s Indicates the spatial resolution characteristics of the data source s, in meters / pixel; τ s Indicates the temporal resolution characteristics of the data source s, in seconds / frame; σ s represents the noise characteristics of the data source s, and represents the signal-to-noise ratio; δ sRepresents the latency characteristics of data source s, in seconds. Through this characteristic model matrix, the system can objectively evaluate the characteristics of each data source, providing a basis for subsequent data fusion.

[0033] Then, different processing strategies are adopted for different time domain data: For long time domain data, multi-resolution analysis and sparse representation are used to extract stable environmental features, using the formula:

[0034]

[0035] Among them, F l is the feature extracted from long-term data, Φ is the feature extraction operator, D l is a long-term dataset, φ i is the i-th basis function, α i is the coefficient of the i-th basis function, and N is the total number of basis functions. This processing method can effectively extract stable features from long-term data and provide a reliable basis for environmental modeling.

[0036] For mid-time domain data, variational mode decomposition (VMD) is used to extract time-varying features, and its optimization objective is:

[0037]

[0038] and

[0039] Among them, u k is the kth modal component, ω k is the center frequency of the kth modal component, δ(t) is the impulse function, f is the original signal, K is the number of modal components, * is the convolution operator, is the time partial derivative, j is the imaginary unit, ||·|| 2 Represents the L2 norm (norm of a square integrable function). The VMD algorithm can effectively separate different modal components in mid-time domain data and identify time-varying features in the environment.

[0040] For short-time domain data, adaptive sliding window processing is used, and the window size calculation formula is:

[0041] W(t)=W b ·(1+β·V(D s (t)))

[0042] Where W(t) is the window size at time t, in seconds; W b is the basic window size, in seconds; β is the adjustment coefficient, dimensionless; V(D s (t)) is the short-time domain data D s The rate of change at time t is calculated as:

[0043]

[0044] Where Δt is the time interval in seconds, and |·| represents the vector norm. The adaptive sliding window can dynamically adjust the window size based on the data change rate, quickly responding to environmental changes.

[0045] Next, we design a spatiotemporal consistency evaluation function to evaluate the consistency between different data sources:

[0046] E(D)=α·E sp (D)+β·E tp (D)+γ·E rl (D)

[0047] Where E(D) is the spatiotemporal consistency evaluation value of dataset D; E sp (D) is the spatial consistency assessment, E tp (D) is the time consistency evaluation, E rl (D) represents the reliability assessment; α, β, and γ are weight coefficients, satisfying α + β + γ = 1. This function comprehensively considers the spatial consistency, temporal consistency, and reliability of the data, and can effectively identify and handle outliers and contradictory information in the data.

[0048] The calculation formula for spatial consistency evaluation is:

[0049]

[0050] Among them, w ij is the weight between data sources i and j, |D i -D j | is the difference measure between data sources i and j, f(d ij ) is based on the spatial distance d ij The decay function is defined as:

[0051]

[0052] Among them, σ d is the spatial attenuation coefficient. Spatial consistency assessment ensures the spatial consistency of information from different data sources.

[0053] The calculation formula for time consistency evaluation is:

[0054]

[0055] Among them, w it is the weight of data source i at time t, |D i (t)-D i(t-1)| is the difference measure of data source i at adjacent time points, and g(Δt) is the decay function based on the time interval Δt, which is defined as:

[0056]

[0057] Among them, T c is the time-correlation characteristic constant. Temporal consistency evaluates the continuity and stability of monitoring data in the time dimension.

[0058] The reliability evaluation calculation formula is:

[0059]

[0060] Among them, c i is the credibility coefficient of data source i, SNRi is the signal-to-noise ratio of data source i, and Pm,i is the probability of missing data for data source i. Reliability assessment reflects the quality and reliability of the data source itself.

[0061] Finally, the adaptive information entropy weighted fusion algorithm is used to fuse the multi-source data:

[0062]

[0063] Among them, D f is the fused data, D i is the data of the i-th data source, w i is the weight, calculated based on information entropy:

[0064]

[0065] Among them, H i is the information entropy of data source i, and the calculation formula is:

[0066]

[0067] Among them, p i (k) is the probability distribution of the value k in data source i. The information entropy weighting method can adaptively assign weights to different data sources. Data sources with more information receive higher weights, thereby improving the information value of the fused data.

[0068] The above-mentioned multi-time-domain decoupled aerospace data fusion preprocessing mechanism can comprehensively and accurately perceive complex dynamic environments through targeted processing and fusion of data from different time domains, providing reliable environmental information support for subsequent trajectory planning.

[0069] Preferably, in step S3, the environmental analysis and risk assessment specifically include the following implementation methods:

[0070] First, a comprehensive environmental model is constructed based on the multi-temporal aerospace data fused in step S2. This environmental model consists of three key components: a terrain model, an obstacle model, and a meteorological model. The terrain model is represented using a multi-resolution grid, with fine-grained grids used in high-risk areas and coarse-grained grids in low-risk areas, to achieve a rational allocation of computing resources. The obstacle model distinguishes between static and dynamic obstacles, establishing an accurate geometric representation for static obstacles and a motion state model for dynamic obstacles, including position, velocity, and acceleration, and performing trajectory prediction. The meteorological model integrates meteorological data such as wind, air pressure, and temperature to construct a four-dimensional meteorological field, providing a basis for energy consumption estimation and safety analysis.

[0071] Secondly, based on the constructed environmental model, static and dynamic risk assessments are performed. The static risk assessment is based on terrain and fixed obstacles, and the calculation formula is:

[0072]

[0073] Among them, R static (x) is the static risk value of position x, ranging from [0,1]; R terrain (x) is the terrain risk, which is related to factors such as slope and roughness, and its value range is [0,1]; R i is the inherent risk value of the i-th static obstacle, ranging from [0,1]; d(x,o i ) is the position x to the obstacle o i The distance in meters; i is the distance attenuation coefficient, in units of 1 / meter 2 ; α t and α o is the weight coefficient, satisfying α t +α o =1;N s is the number of static obstacles. This formula comprehensively considers terrain risk and static obstacle risk, and uses an inverse square decay model to describe how the impact of obstacles changes with distance.

[0074] Terrain risk R terrain (x) is further defined as:

[0075] R terrain (x) = w s ·S(x)+w r ·R(x)+w e ·E(x)

[0076] Where S(x) is the slope risk at position x, ranging from [0, 1]; R(x) is the roughness risk, ranging from [0, 1]; E(x) is the elevation risk, ranging from [0, 1]; w s 、wr and w e is the weight coefficient, satisfying w s +w r +w e = 1. Slope risk reflects the degree of terrain inclination, roughness risk indicates the degree of surface unevenness, and elevation risk considers the impact of altitude on flight.

[0077] Dynamic risk assessment takes into account moving obstacles and weather changes, and the calculation formula is:

[0078]

[0079] Among them, R dynamic (x, t) is the dynamic risk value of position x at time t, ranging from [0, 1]; R j is the risk coefficient of the jth dynamic obstacle, ranging from [0,1]; P coll (x,t,d j ) is the position x at time t and the dynamic obstacle d j The collision probability ranges from [0,1]; R weather (x, t) is the meteorological risk, ranging from [0, 1], and is related to factors such as wind speed and turbulence; β d and β w is the weight coefficient, satisfying β d +β w =1;N d is the number of dynamic obstacles. This formula comprehensively considers the collision risk of dynamic obstacles and meteorological risk, providing a risk assessment for dynamic environments for trajectory planning.

[0080] The collision probability is calculated based on Monte Carlo simulation:

[0081]

[0082] Where M is the number of Monte Carlo simulations, ranging from 100 to 1000; T k (·) is the trajectory of the kth simulation; d min (·,·) is the minimum distance between two trajectories, in meters; d safe is the safety distance in meters; I(·) is an indicator function that takes the value 1 if the condition is true and 0 otherwise. This method evaluates the collision risk by repeatedly simulating the future trajectories of the drone and dynamic obstacles and counting the frequency of possible collisions.

[0083] Weather risk R weather (x,t) is defined as:

[0084] R weather (x, t) = γ v ·R wind(x,t)+γ t ·R turb (x,t)+γ p ·R precip (x,t)

[0085] Among them, R wind (x, t) is the wind speed risk, ranging from [0, 1]; R turb (x, t) is the turbulence risk, ranging from [0, 1]; R precip (x, t) is the precipitation risk, ranging from [0, 1]; γ v , γ t and γ p is the weight coefficient, satisfying γ v +γ t +γ p = 1. Wind speed risk is related to wind speed, turbulence risk is related to turbulence intensity, and precipitation risk is related to precipitation intensity.

[0086] The wind speed risk calculation formula is:

[0087] Among them, |v wind (x,t)| is the wind speed at position x at time t, in meters per second; v wind,max The maximum wind speed at which the drone can safely fly, in meters per second.

[0088] Next, based on the static and dynamic risk assessment results, a risk heat map and traversable area analysis are generated. The risk heat map uses a layered coloring scheme to divide the space into high-risk areas (red), medium-risk areas (yellow), and low-risk areas (green), making the risk distribution intuitive. The total risk calculation formula is:

[0089] R total (x,t)=w s ·R static (x)+w d ·R dynamic (x,t)

[0090] Among them, R total (x,t) is the total risk value of position x at time t, ranging from [0,1]; w s and w d is the weight of static risk and dynamic risk, satisfying w s +w d = 1. This formula combines static and dynamic risks to provide a comprehensive risk assessment.

[0091] The traversable area analysis is conducted based on risk thresholds, dividing the space into no-fly zones, restricted zones, and free zones: total (x,t)>Rhigh For a no-fly zone; when R low <R total (x,t)≤R high is the restricted area; when R total (x,t)≤R low is a free zone. low 、R high These are low-risk and high-risk thresholds, respectively. Drones are prohibited from entering no-fly zones, restricted zones allow passage under certain conditions but are not recommended for extended stays, and free zones allow for free flight. This regional division provides clear spatial constraints for trajectory planning.

[0092] To improve computational efficiency, an adaptive mesh refinement technique is used, using finer meshes in areas with large risk gradients and coarser meshes in areas with uniform risk. The refinement criteria are:

[0093]

[0094] Where c is the grid unit; Indicates the maximum difference in risk value within a grid cell; δ refine To refine the threshold, a value of 0.1-0.2 is usually used. This technique can reduce computational complexity while ensuring accuracy.

[0095] Finally, the risk heat map and the traversable area analysis results are output to step S4 to provide a decision basis for trajectory planning. At the same time, a risk field gradient map is constructed to guide trajectory planning in the direction of risk reduction:

[0096]

[0097] in, is the risk gradient vector at position x at time t; and are the partial derivatives of the risk value in the x, y, and z directions, respectively. The risk gradient provides the direction and rate of change of the risk field, which helps the trajectory planning algorithm find the safest path.

[0098] To support the two-layer trajectory planning in step S4, it is also necessary to calculate the safe flight corridor (SFC), which is defined as a continuous low-risk area:

[0099]

[0100] in, represents the set of real numbers, R corridor is the corridor risk threshold, usually R corridor =0.5; t0 and t fThe safe flight corridor provides spatial constraints for macro-level path planning, ensuring that the planned path is within the safe area.

[0101] The above environmental analysis and risk assessment method conducts in-depth risk analysis and assessment based on the environmental perception model generated in step S2, generates an intuitive risk heat map and traversable area analysis results, and provides a key decision-making basis for subsequent trajectory planning.

[0102] Preferably, in step S4, the dual-layer asynchronous iterative trajectory planning algorithm specifically includes the following implementation methods:

[0103] First, global path generation is performed at the macro level: an improved spatial path graph G = (V, E, W) is constructed, where V is the node set representing spatial sampling points; E is the edge set representing feasible connections between nodes; and W is the weight matrix defining the cost of edges. This spatial path graph, constructed based on the environmental analysis results from step S3, effectively represents the topological structure and traffic characteristics of the flight space.

[0104] On this basis, a heuristic function is designed to guide the global path search:

[0105] h(n)=λ1·d(n,g)+λ2·r(n)+λ3·e(n,g)

[0106] Where h(n) is the estimated cost from node n to the target; d(n,g) is the Euclidean distance from node n to target g, in meters; r(n) is the risk assessment function for node n, ranging from 0 to 1; e(n,g) is the estimated energy consumption from node n to target g, in joules; and λ1, λ2, and λ3 are weight coefficients such that λ1+λ2+λ3=1. This heuristic function comprehensively considers distance, risk, and energy consumption, guiding the search algorithm to find a global path that balances safety and energy efficiency.

[0107] The risk assessment function is further defined as:

[0108]

[0109] Where M is the number of obstacles in the environment; r i is the risk factor of the ith obstacle; o i is the position of the i-th obstacle; ε is a small positive number to prevent the denominator from being zero. This function models the impact of obstacles as an inverse square decay, which can accurately reflect the risk level of the node.

[0110] The energy consumption estimation function is defined as:

[0111] e(n,g)=k e ·m·g r·|h n -h g |+k d d(n,g)

[0112] Among them, k e is the vertical motion energy consumption coefficient; k d is the energy consumption coefficient of horizontal motion; m is the mass of the drone in kilograms; g r is the acceleration due to gravity, 9.8m / s 2 ;h n 、h g is the height of node n and target g, in meters. This function takes into account the difference in energy consumption between vertical and horizontal motion and provides a reasonable estimate of energy consumption.

[0113] Based on the above definition, the improved A* algorithm is used to generate a global waypoint sequence and construct a flight corridor. The evaluation function of A* is:

[0114] f(n)=g(n)+h(n)

[0115] Where f(n) is the total evaluation value of node n; g(n) is the actual cost from the starting point to node n; and h(n) is the estimated cost from node n to the goal. The improved A* algorithm can efficiently generate global paths in large-scale spaces and, through the construction of flight corridors, provide constraints for trajectory optimization at the micro level.

[0116] Secondly, the trajectory is refined at the micro level: the trajectory is represented using piecewise Bézier curves:

[0117]

[0118] Where P(t) is the curve point under parameter t; is the Bernstein basis function, defined as:

[0119]

[0120] Among them, P i is the i-th control point; n is the order of the Bézier curve. The Bézier curve has good mathematical properties and can generate smooth and continuous trajectories, which facilitates subsequent optimization and control.

[0121] Based on the Bézier curve representation, nonlinear model predictive control (NMPC) is used to optimize the trajectory parameters, and the optimization objective is:

[0122]

[0123] Where J is the optimization objective function; a(t) is the acceleration vector at time t; j(t) is the jerk vector at time t; R(x(t)) is the risk function of position x(t); E(v(t),a(t)) is the energy consumption function under velocity v(t) and acceleration a(t); λ1, λ2, λ3, and λ4 are weight coefficients that satisfy This optimization objective comprehensively considers the smoothness, safety, and energy efficiency of the trajectory and can generate high-quality local trajectories.

[0124] The energy consumption function is further defined as:

[0125] E(v,a)=P h +P f (v)+P a (a)

[0126] Among them, P h is the hovering power; P f (v) is the forward flight power, calculated as:

[0127] P f (v)=k1·‖v‖ 3 +k2·‖v‖ 2 +k3·‖v‖

[0128] P a (a) is the acceleration power, which is calculated as follows:

[0129] P a (a)=k4·m·‖a‖·‖v‖

[0130] Where k1, k2, k3, and k4 are energy consumption model parameters, and m is the mass of the drone. This energy consumption model takes into account the energy consumption characteristics of the drone under different flight states and provides an accurate energy consumption assessment for trajectory optimization.

[0131] Then, establish an asynchronous iteration mechanism to achieve the collaborative work of the macro layer and the micro layer: define the trigger set T t Deciding when to replan:

[0132] T t ={t|[ΔR(t)>τ R ]∨[ΔE(t)>τ E ]∨[d(x(t),x p (t))>τ d ]}

[0133] Where ΔR(t) is the risk change at time t; ΔE(t) is the energy consumption deviation at time t; d(x(t),x p (t)) is the deviation between the actual position and the planned position; τR , τ E , τ d This trigger mechanism can trigger re-planning in a timely manner when the environment changes significantly or the execution state deviates from the plan, ensuring the real-time adjustment capability of the trajectory.

[0134] An event-driven two-layer communication protocol is used to achieve efficient information exchange between layers. The message passing model is:

[0135] M i,j ={ID,Type,Priority,Content,Timestamp}

[0136] Where ID is the unique identifier of the message; Type is the message type; Priority is the message priority; Content is the message content; and Timestamp is the message timestamp. This communication protocol ensures timely synchronization of information between the macro and micro layers, enabling the two layers to work together efficiently.

[0137] The above two-layer asynchronous iterative trajectory planning algorithm balances planning efficiency and accuracy through decoupling and coordination of the macro and micro layers, and can efficiently generate trajectories that meet multiple constraints in complex environments.

[0138] Preferably, in step S5, the trajectory fine adjustment method of multi-constraint joint optimization specifically includes the following implementation methods:

[0139] First, conduct security risk field modeling and construct a hybrid risk potential field model:

[0140] R t (x,t)=R s (x)+R d (x,t)+R u (x,t)+R m (x,t)

[0141] Among them, R t (x,t) is the total risk value of position x at time t, ranging from [0,1]; R s (x) is the static obstacle risk, reflecting the impact of fixed obstacles; R d (x, t) is the dynamic obstacle risk, which indicates the threat of moving obstacles; R u (x, t) is the uncertainty risk, taking into account the uncertainty of environmental perception; R m (x, t) is the mission-related risk, which is related to the specific mission requirements. This hybrid risk model comprehensively considers various risk factors in the flight environment and provides a risk assessment basis for trajectory optimization.

[0142] The static obstacle risk calculation formula is:

[0143]

[0144] Among them, N s is the number of static obstacles; s i is the position of the i-th static obstacle; α i is the risk coefficient of the i-th static obstacle; σ i is the influence range parameter of the i-th static obstacle. This formula models the impact of static obstacles as a Gaussian function, and the risk decays with increasing distance.

[0145] Dynamic barrier risk adopts probabilistic model:

[0146]

[0147] Among them, N d is the number of dynamic obstacles; d i (t) is the position of the i-th dynamic obstacle at time t; β i is the risk coefficient of the i-th dynamic obstacle; σ i is the influence range parameter of the i-th dynamic obstacle; P c (x,d i ,t) is the collision probability estimation function, and the calculation formula is:

[0148] P c (x,d i ,t)=∫ t t+Δt p(‖x(τ)-d i (τ) <d safe )dτ

[0149] Among them, d safe is the safety distance; p(·) is the probability density function. This model not only considers the current position of dynamic obstacles, but also predicts the possible collision risk in the future.

[0150] The formula for calculating uncertainty risk is:

[0151]

[0152] in, is the variance of the location estimate; is a normalization parameter. This formula reflects the impact of environmental perception uncertainty on risk. The greater the position estimation error, the higher the risk.

[0153] Secondly, energy-aware trajectory optimization is performed: an accurate energy consumption model is established that takes into account terrain, wind field, and flight status:

[0154] E=k1·m·g r·v c +k2·ρ(h)·S·C d ·v 3 +k3·‖a‖ 2 +k4·f(w,v)

[0155] Where E is the energy consumption under time t, speed v, acceleration a, height h and wind field w; v c is the climbing velocity component; ρ(h) is the air density at height h, and the calculation formula is:

[0156]

[0157] Where ρ0 is the air density at sea level; M is the molar mass of air; R is the universal gas constant; T0 is the standard temperature; S is the characteristic area of ​​the drone; C d is the drag coefficient; f(w,v) is a function that characterizes the effect of wind field w on energy consumption, which is defined as:

[0158] f(w,v)=‖v‖·‖w‖·cos(θ v,w )

[0159] Among them, θ v,w is the angle between the velocity vector v and the wind vector w;

[0160] k1, k2, k3, and k4 are parameters of the energy consumption model. This energy consumption model comprehensively considers factors such as vertical motion, air resistance, acceleration changes, and wind field effects, and can accurately evaluate energy consumption under different flight conditions.

[0161] Then, realize communication quality assurance: design communication quality evaluation function:

[0162]

[0163] Among them, Q c (x) is the communication quality evaluation value of position x; N c is the number of communication nodes; w j is the weight of communication node j; SNR j (x) is the signal-to-noise ratio between position x and node j, and is calculated as:

[0164] SNR j (x) = P t +G t +G r -PL(d(x,x j ))-N0

[0165] Among them, P t is the transmission power; G t is the transmission gain; G ris the receiving gain; PL(d) is the path loss, and the simplified model is:

[0166]

[0167] Where PL0 is the path loss at the reference distance d0; n is the path loss exponent; N0 is the noise power; SN R min,j is the minimum acceptable signal-to-noise ratio of node j; λ is the distance attenuation coefficient; d(x,x j ) is the position x and the communication node x j This function evaluates the communication quality of the UAV at different locations and ensures the reliability of the communication link.

[0168] Next, multi-objective trajectory optimization is performed: the modified non-dominated sorting genetic algorithm (NSGA-III) is used to solve the multi-objective optimization problem:

[0169] minF(τ)=[F1(τ),F2(τ),F3(τ),F4(τ)] T

[0170] Among them, F1(τ) is the security risk function, which is defined as:

[0171]

[0172] F2(τ) is the total energy consumption function, defined as:

[0173]

[0174] F3(τ) is the communication quality loss function, defined as:

[0175]

[0176] F4(τ) is the total flight time function, defined as:

[0177] F4(τ)=t f -t0

[0178] Constraints include:

[0179]

[0180] g4(τ):F2(τ)≤E budget

[0181]

[0182] Among them, v min 、v max is the minimum and maximum speed limit; a max is the maximum acceleration limit; R thresholdis the risk threshold; E budget is the energy consumption budget; Q min is the minimum communication quality requirement. This multi-objective optimization problem comprehensively considers safety, energy efficiency, communication quality, and flight time, and can find a Pareto optimal trajectory set that meets various constraints.

[0183] Finally, perform trajectory smoothing and timing adjustment: Use variational B-spline technology to smooth the trajectory:

[0184] min∫0 T [α·‖p ″ (t)‖ 2 +β·‖p″′(t)‖ 2 ]dt

[0185] Among them, p ″ (t) is the second-order derivative (acceleration) of the trajectory; p″′(t) is the third-order derivative (jerk) of the trajectory; α and β are smoothing weight coefficients. This optimization problem minimizes the acceleration and jerk of the trajectory, generating a smoother trajectory.

[0186] Adjust timing via optimal time reparameterization:

[0187]

[0188] And s(0)=0,s(T)=1,s'(t)>0

[0189] Among them, s(t) is the time reparameterization function, which maps [0,T] to [0,1]; s ′ (t), s ″ (t) is the first and second order derivatives of the function s(t); γ and η are weight coefficients; the trajectory after reparameterization is: p new (t) = p(s * (t)). Time reparameterization can adjust the execution timing of the trajectory and optimize the velocity and acceleration distribution.

[0190] This multi-constraint joint optimization trajectory fine-tuning method optimizes all aspects of the trajectory by comprehensively considering multiple constraints, including safety, energy efficiency, and communication quality, to generate a high-quality flight trajectory. This method further optimizes the initial trajectory generated in step S4 and closely integrates the trajectory execution and adjustment in step S7, ensuring the practicality and feasibility of the trajectory.

[0191] Preferably, in step S6, the distributed computing framework for dynamic allocation of aerospace resources specifically includes the following implementation methods:

[0192] First, the task is decomposed and represented, and the trajectory planning task is represented as a directed acyclic graph G t=(V t ,E t ), where V t is a set of task nodes, representing the computing tasks that need to be executed; E t The task dependency edge set represents the dependency relationship between tasks. This graph representation method can clearly describe the structure and dependency relationship of complex trajectory planning tasks, providing a basis for subsequent task allocation.

[0193] Each task node v∈V t With attribute vector:

[0194] attr(v)=[c v ,m v ,d v ,p v ]

[0195] Among them, c v is the computational complexity, in FLOPS (floating-point operations per second), which indicates the computational requirements of the task; m v Memory requirement, in bytes, represents the memory space required for task execution; d v is the deadline, in seconds, indicating the time point when the task must be completed; p v is the priority, ranging from [1,10], indicating the importance of the task. This attribute vector fully describes the characteristics of the task and provides a basis for task allocation decisions.

[0196] Secondly, perform resource perception and modeling: establish a heterogeneous computing resource model R = {r i |i∈[1,n]}, where r i represents the i-th computing resource, including air-based resources (ground control stations, edge computing nodes) or space-based resources (satellites, high-altitude computing platforms). This model encompasses all available computing resources in the system and provides a resource pool for task allocation.

[0197] Each resource i With a dynamic attribute vector:

[0198] attr(r i )=[p i (t),m i (t),e i (t),b i (t)]

[0199] Among them, p i (t) is the processing capacity at time t, in FLOPS, which represents the computing power of the resource; m i (t) is the available memory at time t, in bytes, indicating the memory capacity of the resource; e i(t) is the energy state at time t, in joules, representing the energy reserve of the resource; b i (t) is the communication bandwidth at time t, in bits per second, representing the communication capability of the resource. This attribute vector reflects the state and capability of the resource in real time, enabling the system to allocate tasks based on the actual availability of the resource.

[0200] Then, we design a task allocation decision mechanism: construct a multi-objective utility function to evaluate task-resource matching:

[0201]

[0202] Where U(i,j) is the utility value assigned to resource j by task i; P(j) is the processing capacity of resource j; C(i) is the computational complexity of task i; T(i) is the timeliness requirement of task i; L(i,j) is the expected execution delay of task i on resource j; E(j) is the energy state of resource j; E c (i, j) is the energy consumption of task i executed on resource j; B(i) is the data volume of task i; C c (i, j) is the communication cost between task i and resource j; α, β, γ, and δ are weight coefficients, satisfying α + β + γ + δ = 1. This utility function comprehensively considers computational efficiency, timeliness, energy consumption, and communication cost, and can comprehensively evaluate the quality of task-resource matching.

[0203] Introduce reinforcement learning methods for task scheduling and design reward functions:

[0204]

[0205] Among them, R(s,a) is the reward for performing action a in state s; t e is the execution time; t d is the deadline; e u is the energy consumption used; e b is the energy consumption budget; p c is the communication penalty, and the calculation formula is:

[0206]

[0207] Where B(i) is the data volume of task i; d(i,j) is the communication distance between task i and resource j; b avg is the average communication bandwidth; r l is the load imbalance penalty, and the calculation formula is:

[0208]

[0209] Among them, σ L is the standard deviation of resource load, μ Lis the average value of resource load;

[0210] w1, w2, w3, w4 are weight coefficients, satisfying This reinforcement learning method can adaptively adjust the task scheduling strategy according to environmental feedback, thereby improving scheduling efficiency and quality.

[0211] Finally, implement adaptive load balancing and define a system load imbalance metric:

[0212]

[0213] Where I(L) is the load imbalance metric; L i is the load of resource i; L a is the average load, and the calculation formula is:

[0214]

[0215] Where n is the number of resources. This metric can quantify the distribution of resource load in the system and provide a basis for load balancing.

[0216] Design potential-based task transfer criteria:

[0217] ΔU(i,j→k)=φ(L j )-φ(L j -w i )+φ(L k +w i )-φ(L k )-C m (i,j→k)

[0218] Where ΔU(i,j→k) is the potential energy change of migrating task i from resource j to resource k; φ(L) is the load potential energy function, defined as φ(L)=L 2 / 2;L j 、L k is the load of resources j and k; w i is the workload of task i; C m (i, j→k) is the migration cost, which is calculated as follows:

[0219]

[0220] Where B(i) is the data volume of task i; d(j,k) is the distance between resources j and k; b j,k is the communication bandwidth between resources j and k; S i The state saving cost for task i; α m , β mis the migration cost weight coefficient. This criterion can evaluate the benefits and costs of task migration, perform task migration at the appropriate time, and achieve dynamic balance of system load.

[0221] The above-mentioned distributed computing framework for dynamic allocation of aerospace resources achieves efficient utilization of computing resources through mechanisms such as task decomposition, resource modeling, task allocation and load balancing, providing computing support for other steps.

[0222] Preferably, in step S7, the trajectory execution and real-time adjustment specifically include the following implementation methods:

[0223] First, to achieve precise execution of the planned trajectory, a hierarchical control architecture is employed, with a high-level trajectory tracking controller and a low-level attitude controller working in tandem. The high-level trajectory tracking controller is responsible for generating reference attitudes and throttle commands to ensure the drone follows the planned trajectory; the low-level attitude controller is responsible for achieving the reference attitude and ensuring flight stability. The trajectory tracking controller uses a nonlinear model predictive control (NMPC) algorithm with the following optimization objective:

[0224]

[0225] Where J is the cost function of the controller; e p (τ)=p(τ)-p d (τ) is the position error in meters; e v (τ)=v(τ)-v d (τ) is the velocity error in meters per second; p(τ) and v(τ) are the predicted position and velocity; p d (τ) and v d (τ) is the desired position and velocity on the reference trajectory; u(τ) is the control input; Q p , Q v and R is the weight matrix; T p is the prediction time domain length in seconds. NMPC solves the above optimization problem in each control cycle through a rolling optimization method to generate the optimal control input.

[0226] The dynamic constraints during trajectory tracking are:

[0227]

[0228] Where p(t) and v(t) are the position and velocity vectors of the drone, respectively; g is the gravitational acceleration vector; F(t) is the total thrust; m is the mass of the drone; R(q(t)) is the rotation matrix represented by the quaternion q(t); e z =[0,0,1] T is the vertical unit vector in the body coordinate system; F d (v(t)) is the air resistance vector.

[0229] Secondly, it realizes real-time status monitoring and environmental perception, including the drone's own status monitoring and environmental change perception. The drone's own status monitoring collects position, speed, attitude, and energy status information through onboard sensors, and calculates the deviation from the reference trajectory:

[0230] e p (t)=||p(t)-p d (t)||

[0231] e v (t)=||v(t)-v d (t)||

[0232]

[0233] Among them, e p (t) is the position deviation, in meters; e v (t) is the velocity deviation in meters per second; e a (t) is the attitude deviation, in radians; a(t) and a d (t) represents the axis angle of the actual and desired attitudes, respectively. The condition monitoring data is transmitted to the control station in real time via a wireless link and recorded in an onboard storage device for subsequent analysis and optimization.

[0234] Environmental change perception acquires real-time environmental data through onboard sensors and aerospace information links, compares it with the environmental model fused in step S2, and detects environmental changes:

[0235] ΔE(t)=||E real (t)-E model (t)|| F

[0236] Among them, ΔE(t) is the environmental change measure; E real (t) is the real-time perceived environmental state; E model (t) is the state predicted by the environment model; ||·|| F is the Frobenius norm. When the environment changes beyond the threshold, it triggers the environment model update and potential trajectory replanning.

[0237] In particular, an obstacle detection and tracking algorithm is used to identify and track dynamic obstacles in real time, update their motion status, and calculate collision risks:

[0238] P coll (t) = P(d min (p(t:t+Δt),p obs (t:t+Δt)) <d safe )

[0239] Among them, P coll (t) is the collision probability at the current moment; d min (·,·) is the minimum distance between the two trajectories; p(t:t+Δt) is the predicted trajectory of the UAV; p obs (t:t+Δt) is the predicted trajectory of the obstacle; d safe is the safety distance; P(·) is the probability function. When the collision probability exceeds the safety threshold, the obstacle avoidance mechanism is triggered.

[0240] Then, we implement real-time trajectory adjustment based on multi-level strategies, using different levels of adjustment strategies according to the severity of environmental changes and state deviations:

[0241] Level 1: Trajectory fine-tuning. When the state deviation is small or the environmental change is not significant, the trajectory is fine-tuned by adjusting the controller parameters without changing the original planned trajectory:

[0242] Q p ′ =Q p +ΔQ p (e p (t))

[0243] Q v ′ =Q v +ΔQ v (e v (t))

[0244] Among them, Q p ′ and Q v ′ is the adjusted weight matrix; ΔQ p (e p (t)) and ΔQ v (e v (t)) is the weight adjustment amount based on the bias. The fine-tuning process does not trigger replanning, saving computing resources.

[0245] Level 2: Local obstacle avoidance. When a dynamic obstacle is detected and the collision risk is high, the local obstacle avoidance mechanism is activated, temporarily deviating from the original planned trajectory, avoiding the obstacle, and then returning to the original trajectory. Local obstacle avoidance uses an improved artificial potential field method:

[0246]

[0247] Among them, F total (p(t)) is the resultant force; k att is the attraction coefficient; p d (t) is the target point on the original planned trajectory; F rep,i(p(t)) is the repulsive force generated by the i-th obstacle, when ||p(t)-p obs,i (t)||≤d obs , When ||p(t)-p obs,i (t)||>d obs , F rep,i (p(t))=0. Where, k rep,i is the repulsive force coefficient; p obs,i (t) is the obstacle position; d obs is the influence range of the obstacle. In order to solve the local minimum problem of the traditional artificial potential field method, random perturbation is introduced:

[0248] F′ total (p(t))=F total (p(t))+ξ(t)

[0249] Among them, ξ(t) is a random perturbation vector, which is activated when the drone is detected to be trapped in a local minimum.

[0250] Level 3: Local replanning. When the state deviation is large or the environment changes significantly but is still within the controllable range, local replanning is performed between the current position and a future point on the original trajectory, using a fast replanning algorithm:

[0251]

[0252] Among them, τ local is the local replanning trajectory; T local is the local planning time domain; c smooth 、c risk and c energy are smoothness, risk and energy cost functions respectively; T delay A delay factor is used to adjust the connection point between the replanned trajectory and the original trajectory. Local replanning does not change the global path, but only modifies the local trajectory, which reduces the computational burden.

[0253] Level 4: Global replanning. When the environment changes significantly or local adjustments cannot ensure safety, global replanning is triggered and the complete trajectory is regenerated back to step S4. The triggering conditions are:

[0254]

[0255] Among them, δ E , δ P and δ p are the thresholds for environment change, collision probability, and position deviation, respectively. Global replanning requires high computing resources and should be triggered with caution.

[0256] Finally, implement mission completion evaluation and flight log recording to regularly evaluate mission completion:

[0257]

[0258] Among them, C mission (t) is the total completion rate of the task; C i (t) is the completion degree of the i-th subtask; w i is the weight coefficient; N task The number of subtasks. Task completion evaluation provides a basis for adjusting subsequent tasks.

[0259] The above trajectory execution and real-time adjustment method ensures that the UAV can accurately execute the planned trajectory and respond promptly to environmental changes through a hierarchical control architecture, multi-level adjustment strategy and a complete monitoring mechanism.

[0260] Preferably, during the execution of steps S2 to S6, a task-aware trajectory adaptive generation method is further included, which specifically includes the following implementation methods:

[0261] First, perform task space representation: construct the task-related spatiotemporal information gain field:

[0262]

[0263] Among them, G(x,t) is the information gain value of position x at time t; N i is the number of information types; Q i (x, t) is the information gain of the i-th category at position x at time t, and the calculation method depends on the specific task; T i (x, t) is the information timeliness function, which indicates the degree of outdatedness of information; γ i is the weight coefficient of the i-th type of information; ξ i is the time-effectiveness attenuation coefficient of the i-th type of information. This information gain field quantifies the task value of different positions in space and provides a basis for task-oriented trajectory generation.

[0264] Secondly, different task-adaptive trajectory modes are selected according to the task type. For reconnaissance and surveillance tasks, the maximum information entropy reduction trajectory is adopted:

[0265]

[0266] Among them, H(X|Z0) is the conditional entropy of state X under the given initial observation Z0; H(X|Z0,Z τ ) is a given initial observation Z0 and the observation Z on the trajectory τ τ The conditional entropy of the state X under the condition. This optimization objective maximizes the amount of information obtained on the trajectory and is suitable for reconnaissance and surveillance tasks.

[0267] For target tracking tasks, the minimum estimation error trajectory is used:

[0268]

[0269] where P(t|τ) is the state estimate covariance matrix at time t under trajectory τ, and tr[·] is the trace of the matrix. This optimization objective minimizes the uncertainty of the target state estimate and is suitable for target tracking tasks.

[0270] State estimation is performed using the Extended Kalman Filter (EKF), whose prediction and update equations are:

[0271] Prediction steps:

[0272]

[0273] Update steps:

[0274]

[0275] Where f(·) is the state transfer function; h(·) is the observation function; F k is the state transfer Jacobian matrix; H k is the observation Jacobian matrix; Q k is the process noise covariance matrix; R k is the observation noise covariance matrix; K k is the Kalman gain. The extended Kalman filter can handle the state estimation problem of nonlinear systems and provides an effective tool for target tracking.

[0276] For search and rescue missions, the maximum detection probability trajectory is used:

[0277]

[0278] Among them, P d (x(t)) is the probability of target detection at position x(t), and the calculation formula is:

[0279]

[0280] Among them, P d,max is the maximum detection probability; x target is the target position (unknown, expressed using probability distribution); σ d is the detection range parameter. This optimization objective maximizes the probability of detecting the target on the trajectory and is suitable for search and rescue tasks.

[0281] Next, the task constraint transformation is performed to transform the task requirements into the trajectory constraint mapping function Φ:T req →C traj, where T req is the task requirement space, C traj is the trajectory constraint space. This mapping function transforms abstract task requirements into specific trajectory constraints, making trajectory planning adaptable to various task requirements.

[0282] Design task completion evaluation function:

[0283]

[0284] Among them, C task (τ) is the task completion degree of trajectory τ; N t is the number of subtasks; w i is the weight of subtask i; C i (τ) is the completion function of subtask i. This evaluation function quantifies the contribution of the trajectory to the task and provides a basis for trajectory selection.

[0285] The above-mentioned task-aware trajectory adaptive generation method runs through the execution process of steps S2 to S6. According to different task types and requirements, the trajectory generation and optimization strategies are adjusted to improve the task execution efficiency.

[0286] In summary, this invention provides a method for UAV flight trajectory planning based on air-space collaboration. By integrating airborne and space-based platforms, it achieves efficient fusion of multi-temporal data, hierarchical path planning, multi-constrained trajectory optimization, distributed computing resource scheduling, and task-based trajectory adaptive generation. This method has broad practical application in disaster monitoring, search and rescue, and other fields, demonstrating its high practicality and potential for widespread adoption.

[0287] While this document illustrates specific applications of the present invention through specific embodiments, these embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims, and any appropriate modifications, equivalent substitutions, or improvements based on the principles of the present invention should be considered to fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be given the broadest interpretation to encompass all such modifications, equivalent structures, and functions.

Claims

1. A UAV flight trajectory planning method based on air-space linkage, characterized in that: Including steps: Step S1, system initialization, including: loading mission parameters and constraints, initializing the sensing units of the air-based platform and the space-based platform, establishing a computing resource pool and performing initial resource allocation based on resource capabilities and communication bandwidth; Step S2, multi-temporal air-space data acquisition and fusion, includes: collecting multi-source heterogeneous data including terrain data, obstacle information, and meteorological data from air-based and space-based platforms, decoupling the data in the temporal domain according to the frequency of change, processing the long-temporal domain data, medium-temporal domain data, and short-temporal domain data separately, and generating a unified environmental perception model using a multi-temporal domain fusion algorithm based on spatiotemporal consistency assessment; Step S3, environmental analysis and risk assessment: Based on the environmental perception model generated in step S2, an environmental model including a terrain model, an obstacle model, and a weather model is constructed to perform static and dynamic risk assessments, generating a risk heat map representing the spatial risk distribution and an analysis result of the traversable area representing the flyable space; Step S4, two-layer asynchronous iterative trajectory planning, using the risk assessment results of step S3, uses an improved A* algorithm that takes into account distance, risk, and energy consumption to generate a global path and flight corridor at the macro level. At the micro level, the trajectory is represented by a Bézier curve within the flight corridor and local trajectory optimization is performed through nonlinear model predictive control. Inter-layer information exchange and asynchronous iterative updates are achieved through an event-driven message passing mechanism. Step S5, multi-constraint joint optimization trajectory adjustment, based on the initial trajectory generated in step S4, constructs a multi-objective optimization model that takes into account safety, energy efficiency, communication quality, and mission requirements. A modified non-dominated sorting genetic algorithm is used to solve the Pareto optimal trajectory set, and the final trajectory solution is selected based on the decision preference according to the task priority. Step S6, dynamic allocation of computing resources, representing the computing tasks from step S2 to step S5 as a directed acyclic graph with dependencies, assigning tasks based on task computational complexity, timeliness requirements, resource processing capabilities, and energy status, and implementing an adaptive load balancing strategy based on load balancing metrics and task migration judgment; Step S7, trajectory execution and real-time adjustment: Execute the final trajectory generated in step S5. Monitor the flight status and environmental changes in real time through onboard sensors. Based on the degree of deviation and the magnitude of environmental changes, a hierarchical adjustment strategy, including trajectory fine-tuning, local obstacle avoidance, local replanning, and global replanning, is adopted. If necessary, a replanning mechanism is triggered to return to step S4 or step S5 to update the trajectory.

2. The method for planning the flight trajectory of a UAV based on air-space linkage according to claim 1, characterized in that: In step S2, the multi-time domain decoupled air-space data fusion preprocessing mechanism includes: The data is divided into long-term data (change period greater than 1 hour), medium-term data (change period 1-60 minutes) and short-term data (change period less than 1 minute); Establish a data characteristic model matrix to represent the spatial resolution, temporal resolution, noise characteristics, and delay characteristics of the data source; Multi-resolution analysis and sparse representation are used to extract stable environmental features for long-time domain data, variational mode decomposition is used to extract time-varying features for medium-time domain data, and adaptive sliding window processing is used to update the local environment in real time for short-time domain data. Design a spatiotemporal consistency evaluation function to evaluate the spatial consistency, temporal consistency and reliability of data; Adaptive information entropy weighted fusion algorithm is used to fuse multi-source data.

3. The method for planning the flight trajectory of a UAV based on air-space linkage according to claim 1, characterized in that: In step S3, the environmental analysis and risk assessment include: Build an environmental model based on the fused data, including terrain model, obstacle model and weather model; Conduct static risk assessment, taking into account terrain risk and static obstacle risk; Conduct dynamic risk assessments, taking into account dynamic obstacle collision probabilities and meteorological influences; Generate a risk heat map and divide the space into high-risk areas, medium-risk areas, and low-risk areas; Conduct traversable area analysis and divide the space into no-fly zones, restricted zones, and free zones based on risk thresholds; Adaptive mesh refinement technology is used to use finer meshes in areas with large risk gradients to improve computational efficiency.

4. The method for planning the flight trajectory of a UAV based on air-space linkage according to claim 1, characterized in that: In step S4, the dual-layer asynchronous iterative trajectory planning algorithm includes: Macro-level global path generation: Construct an improved spatial path map, design a heuristic function that considers distance, risk, and energy consumption, use an improved A* algorithm to generate a global waypoint sequence, and construct a flight corridor; Micro-level trajectory refinement: Use piecewise Bézier curves to represent the trajectory and optimize trajectory parameters based on nonlinear model predictive control; Asynchronous iteration mechanism: Define trigger sets to determine when to replan, and use an event-driven two-layer communication protocol to achieve efficient information exchange between layers.

5. The method for planning the flight trajectory of a UAV based on air-space linkage according to claim 1, characterized in that: In step S5, the trajectory fine adjustment method of the multi-constraint joint optimization includes: Safety risk field modeling: Constructing a hybrid risk potential field model that considers static obstacle risk, dynamic obstacle risk, uncertainty risk, and task-related risk; Energy-aware trajectory optimization: Build an energy consumption model that takes terrain, wind field, and flight status into account; Communication quality assurance: Design a communication quality assessment function to evaluate the signal-to-noise ratio and distance between the drone and the communication node; Multi-objective trajectory optimization: A modified non-dominated sorting genetic algorithm is used to solve a multi-objective optimization problem that comprehensively considers safety risk, total energy consumption, communication quality, and total flight time. Trajectory smoothing and timing adjustment: Use variational B-spline technology for trajectory smoothing and adjust the timing through optimal time reparameterization.

6. The method for planning the flight trajectory of a UAV based on air-space linkage according to claim 1, characterized in that: In step S6, the distributed computing framework for dynamic allocation of aerospace resources includes: Task decomposition and representation: The trajectory planning task is represented as a directed acyclic graph, where each task node has an attribute vector representing computational complexity, memory requirements, deadline, and priority; Resource awareness and modeling: Build a heterogeneous computing resource model where each resource has a dynamic attribute vector representing processing power, available memory, energy status, and communication bandwidth; Task allocation decision mechanism: Design a multi-objective utility function to evaluate task-resource matching and introduce reinforcement learning methods for task scheduling; Adaptive load balancing: Define the system load imbalance metric and design a potential-based task migration decision criterion.

7. The method for planning the flight trajectory of a UAV based on air-space linkage according to claim 1, characterized in that: In step S7, the trajectory execution and real-time adjustment include: A hierarchical control architecture is used to achieve accurate execution of the planned trajectory, including a high-level trajectory tracking controller and a low-level attitude controller; Realize real-time status monitoring and environmental perception, and monitor UAV status deviations and environmental changes; Adopt multi-level adjustment strategies for real-time trajectory adjustment, including trajectory fine-tuning, local obstacle avoidance, local replanning, and global replanning; Realize mission completion evaluation and flight log recording to provide a basis for system optimization.

8. The method for planning the flight trajectory of a UAV based on air-space linkage according to claim 1, characterized in that: The method further includes a task-aware trajectory adaptive generation mechanism, which runs through steps S2 to S6 and includes: Task space representation: Constructing a task-related spatiotemporal information gain field to quantify the task value at different locations in space; Mission-adaptive trajectory mode: Selects different trajectory optimization objectives based on the mission type, including information entropy reduction optimization for reconnaissance and surveillance missions, state estimation error optimization for target tracking missions, and target detection probability optimization for search and rescue missions. Task constraint transformation: A mapping function that transforms task requirements into trajectory constraints and designs a task completion evaluation function.

Images