A method and system for planning a low-altitude flight route in a plateau mountain area based on improved A-star and particle swarm optimization

Abstract

This invention discloses a method and system for low-altitude flight path planning in high-altitude mountainous areas based on an improved A-satellite algorithm and particle swarm optimization, relating to the field of UAV flight path planning technology. The method collects data on aircraft status, 3D terrain, airspace control, meteorology, and mission priors to construct a 3D grid model. Based on this model, a multi-objective comprehensive cost function is established, and adaptive weights are loaded according to the mission type. Before adjacent candidate nodes enter the open list, the improved A-satellite algorithm sequentially performs four layers of hard constraint pruning: terrain clearance, airspace control, mountain occlusion angle, and strong convection probability. This is combined with a lower bound heuristic function to output candidate flight paths. The candidate flight paths are resampled into particle positions, and particle velocities are initialized with negative gradients. The target flight path is obtained through particle swarm optimization iterations and multi-constraint validity repair. This method can reduce the size of the open list, improve search efficiency in high-density infeasible areas, and enhance the comprehensive adaptability of flight paths to terrain, meteorology, occlusion, and emergency risks.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) route planning technology, specifically to the field of low-altitude intelligent transportation and autonomous navigation technology for unmanned aerial vehicles, and particularly to a low-altitude route planning method and system for complex terrain, complex weather, and communication obstruction conditions in plateau and mountainous areas. Background Technology

[0002] Low-altitude flight scenarios in plateau and mountainous areas are characterized by strong terrain undulations, dense valleys and ridges, frequent local weather changes, communication or navigation links being easily affected by mountain obstruction, and a scarcity of emergency landing sites. When performing tasks such as tourism, agricultural and forestry operations, emergency rescue, power line inspection, and disaster reconnaissance, drones and other low-altitude aircraft need to achieve a comprehensive balance among multiple objectives, including terrain airspace safety, climb and descent capabilities, weather avoidance, communication obstruction risks, and emergency accessibility.

[0003] Among existing route planning methods, the A* algorithm is widely used for UAV path planning due to its advantages such as high heuristic search efficiency and ease of integration with grid maps. However, the traditional A* algorithm suffers from problems such as a large number of expanded nodes and a large open list size during the path search process. Some improved methods perform feasibility assessments only after nodes are popped from the open list, or only perform geometric smoothing on the generated path, failing to effectively intercept obviously infeasible nodes before adjacent nodes enter the open list. In three-dimensional grid space, the number of nodes increases cubically with increasing spatial resolution. If a large number of nodes with insufficient airspace, located in no-fly zones, exceeding shielding limits, or with excessively high weather risks enter the open list first, it will lead to an expansion of the priority queue size and a significant increase in heap sorting overhead.

[0004] On the other hand, existing technologies still have shortcomings in modeling the multi-source risks of low-altitude flights in plateau and mountainous areas. Some methods only treat terrain shading as a binary visibility result, making it difficult to express the continuous risk changes when nodes approach the critical shading angle; some methods only consider static terrain constraints and fail to incorporate dynamic meteorological factors such as the probability of short-term severe convection and instantaneous wind speed into the predicted arrival time of path nodes; some methods lack systematic cost modeling in terms of emergency landing point considerations. In addition, existing hybrid methods of A* algorithm and subsequent optimizers (such as particle swarm optimization) are mostly loosely coupled structures—the A* stage only outputs a single path as the initial solution for subsequent optimization, and the particle initialization in the subsequent optimization stage does not fully utilize the cost field information formed in the A* stage, resulting in insufficient information transfer between the two stages and limited convergence speed and search quality.

[0005] Therefore, there is a need for a low-altitude flight path planning method and system that can uniformly express terrain, weather, shelter and emergency risks in a three-dimensional terrain-weather coupled environment, directly prune infeasible nodes before they are enlisted, and tightly couple the cost field information of the A-satellite stage into the particle swarm optimization stage. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for low-altitude flight path planning in high-altitude and mountainous areas based on an improved A-satellite and particle swarm optimization, addressing the aforementioned problems.

[0007] The technical solution of the present invention is as follows: A method for low-altitude flight path planning in high-altitude and mountainous areas based on improved A-satellite and particle swarm optimization includes the following steps: Collect aircraft status parameters, 3D terrain data, airspace control data, meteorological coupling data, and mission prior configuration data; Three-dimensional terrain data, airspace control data and meteorological coupled data are integrated into a three-dimensional raster model, so that each raster node carries information on terrain height, airspace attributes, mountain shading angle, strong convection probability and wind speed. A multi-objective comprehensive cost function is constructed based on a 3D raster model, which includes terrain clearance safety, slope and elevation gradient, emergency accessibility, weather and shelter avoidance. Load the task adaptive weight vector into the multi-objective comprehensive cost function according to the task type; The improved A* algorithm is used for the first stage of search. Before adjacent candidate nodes enter the open list, the improved A* algorithm performs multi-layer hard constraint pruning with at least terrain clearance constraints and mountain shading angle constraints. The lower bound heuristic function is used for search and outputs at least one candidate route. The particle positions for particle swarm optimization are initialized using candidate routes as seeds, and the particle velocities are initialized based on the gradient information of the multi-objective comprehensive cost function. The target route is obtained and output through particle swarm optimization iteration and multi-constraint legality repair.

[0008] By employing a tightly coupled architecture of "multi-layer pruning before queuing + negative gradient velocity initialization," the three levels of problems pointed out in the background technology are addressed simultaneously: at the algorithm level, the pruning mechanism reduces the size of the open list from the source, avoiding the expansion of invalid nodes; at the modeling level, the multi-objective cost function uniformly quantifies terrain, weather, occlusion, and emergency risks; and at the hybrid architecture level, gradient information initialization enables the transfer of cost field knowledge from the A-stage to PSO, achieving a deep integration of discrete search and continuous optimization, accelerating convergence, and improving the quality of the solution.

[0009] Furthermore, the multi-layer hard constraint pruning sequentially includes: Terrain clearance constraint: Determine whether the flight altitude of adjacent candidate nodes is greater than or equal to the sum of the terrain height and the minimum terrain clearance safety threshold; Airspace control constraints: Determine whether adjacent candidate nodes are located outside no-fly zones or temporary control zones; Mountain shading angle constraint: Determine whether the mountain shading angle of adjacent candidate nodes is less than or equal to a preset hard threshold; Strong convection high incidence probability constraint: Determine whether the probability of strong convection high incidence of adjacent candidate nodes at the predicted arrival time is less than or equal to a preset probability hard threshold. If any constraint is not met, the adjacent candidate node is blocked so that it is not included in the open list.

[0010] The pruning process is concretized into four constraint decisions executed sequentially. Its "pre-interception" mechanism allows infeasible nodes to completely bypass the priority queue, directly reducing heap adjustment overhead. The progressive order of constraints (geometry → regulations → communication → meteorology) prioritizes low-computational-cost constraints, reducing the number of calls to high-cost constraints. Compared to the traditional A-satellite approach of "enqueuing first, dequeuing then eliminating," this solution offers a particularly significant efficiency advantage in high-density infeasible regions (such as densely populated canyon areas).

[0011] Furthermore, the mountain's shielding angle Calculate using the following formula: , in, For nodes The horizontal projection is centered and has a radius of The surface sampling set, Sampling points The terrain elevation at that location For nodes Flight altitude For nodes With sampling points Horizontal distance, To prevent the minimum distance constant from having a denominator of zero; The mountain shading angle is used for both: hard threshold determination in the mountain shading angle constraint and soft penalty term in the meteorological and shading avoidance cost.

[0012] Replacing binary visibility judgment with continuous angle values ​​can express the continuous process of communication link quality gradually degrading as the aircraft approaches the ridge, making the cost function more refined in characterizing the risk of obstruction; the same obstruction angle is used as both the hard threshold (absolute exclusion) before joining the formation and the soft penalty (gradual guidance) of the cost function, forming a two-layer risk control mechanism of "hard boundary interception + soft gradient early warning".

[0013] Furthermore, the aforementioned weather and shelter avoidance costs for:

[0014] in, Candidate routes The number of nodes, For the first 1 path node For nodes The mountain's sheltered angle, For nodes The predicted arrival time, It is a monotonically non-decreasing normalized function. Soft threshold for the mountain shading angle, The safe wind speed threshold for aircraft, Path node At its predicted arrival time The probability of strong convection is high. Path node At its predicted arrival time Instantaneous wind speed, , , Internal weights and satisfying .

[0015] By normalizing and weighting three heterogeneous risks—shading angle, strong convection probability, and instantaneous wind speed—the algorithm can comprehensively balance communication quality, flight safety, and attitude energy consumption within the same framework. The "dead zone + progressive penalty" design of the soft threshold ensures flight path smoothness, while the time index τ(p_i) transforms meteorological risk into a time-varying field that dynamically changes with flight progress, enabling predictive avoidance.

[0016] Furthermore, the high probability of strong convection It is obtained by weighted fusion of historical statistical terms and real-time radar extrapolation terms: , in, , For nodes Historical statistical probability of strong convection, For nodes At any moment Real-time strong convection prediction probability.

[0017] By weighted fusion of historical statistics and real-time radar extrapolation, the probability of severe convection is made to take into account both long-term climate patterns (stability) and short-term weather changes (timeliness). In view of the characteristics of severe convection in plateau and mountainous areas, which are characterized by strong suddenness and short life cycle, the prediction accuracy and reliability are optimized.

[0018] Furthermore, the terrain clearance safety cost in the multi-objective integrated cost function for: , in, The minimum terrain clearance safety threshold, For nodes Flight altitude Indicate extraction The clearance at the location is insufficient. This indicates the minimum safe flight altitude. When the flight altitude is equal to or higher than the minimum safe flight altitude, ; Slope and elevation gradient costs for: , in, The maximum allowable climb or descent gradient for the aircraft. For nodes With nodes Horizontal distance.

[0019] The emergency accessibility cost is calculated as the ratio of the distance from the path node to the nearest preset emergency landing point to the emergency accessibility distance threshold: , in, For nodes Distance to the nearest pre-designated emergency landing point This is the emergency reachability threshold.

[0020] The clearance cost only penalizes the insufficient portion (the cost is 0 when the threshold is met), avoiding imposing unnecessary costs on safe areas; the slope cost is symmetrically constrained for the ascent and descent sections, and through a normalized comparison of the actual slope and the maximum allowable slope, the cost has a clear physical meaning and is comparable under different terrain conditions.

[0021] Furthermore, the initialization of particle velocity based on gradient information from the multi-objective integrated cost function is specifically as follows: , in, Initialize the step size for the velocity. For random disturbance terms, For the first The position vector of each particle at the initial moment. This is the finite difference approximation gradient calculated by applying a preset perturbation to the intermediate nodes of the particle path; 60% to 80% of the particles are generated by selecting and resampling the candidate routes in sequence, while the remaining particles are generated by randomly selecting the candidate routes and superimposing Gaussian noise. The resampling process keeps the starting point and target point fixed, and performs legality repair on interpolation points that fall into illegal grids.

[0022] The global cost field information accumulated during the A-star stage is encoded as the initial velocity of PSO particles in the negative gradient direction, transforming PSO from "blind random search" to "directed optimization guided by cost field", which significantly shortens the convergence path. At the same time, Top-K resampling preserves a variety of excellent solution structures, and Gaussian noise supplements the population diversity, effectively avoiding premature convergence.

[0023] Furthermore, the multi-constraint legality repair includes: Raise the height of nodes with insufficient clearance to be greater than or equal to the sum of the terrain height and the minimum terrain clearance safety threshold; Nodes that have entered no-fly zones or controlled airspace will be moved back to the point outside the nearest legal airspace boundary; Replace nodes with excessive mountain shading angles with candidate points with low shading angles within the neighborhood. Replace the nodes with the highest probability of severe convection with the nearest meteorological safety candidate points; When the height difference between adjacent nodes exceeds the upper limit of the climbing or descending gradient, it can be repaired by inserting intermediate transition nodes or adjusting the node height; If no valid candidate point is found within the preset neighborhood radius, the neighborhood is expanded to the preset upper limit. If no valid candidate point is found, the search is reverted to the previous valid node and a new search is triggered or the current particle is deemed infeasible.

[0024] The method implements classified repair for different failure types such as airspace clearance, airspace, obstruction, strong convection, and slope. Compared with the general penalty function method, it has a higher success rate and less loss of flight path quality. The mechanism of "neighborhood expansion step by step → backtracking and re-searching" ensures that particle paths can still be restored to legality in high-density infeasible areas, and makes PSO iteration based on feasible solutions throughout the process, avoiding fitness landscape distortion.

[0025] Furthermore, the task types include at least one of tourism, agricultural and forestry operations, emergency rescue, power transmission line inspection, and disaster survey. Task adaptive weight vector satisfy: , Weight vectors for different task types are loaded into the multi-objective integrated cost function via configuration injection before the search begins.

[0026] By configuring weights, the same algorithm pipeline can be adapted to five types of tasks: tourism, agricultural and forestry operations, emergency rescue, power transmission line inspection, and disaster investigation. This eliminates the need to modify the core code or redevelop the algorithm, significantly reducing engineering deployment costs and on-site parameter tuning time.

[0027] This application also includes a high-altitude, low-altitude flight path planning system for plateau and mountainous terrain based on an improved A-satellite and particle swarm optimization approach, which executes a high-altitude, low-altitude flight path planning method for plateau and mountainous terrain based on an improved A-satellite and particle swarm optimization approach, including: The data acquisition module is used to collect aircraft status parameters, 3D terrain data, airspace control data, meteorological coupling data, and mission prior configuration data. The 3D raster model building module is used to fuse 3D terrain data, airspace control data, and meteorological coupled data into a 3D raster model. The integrated cost function construction module is used to construct a multi-objective integrated cost function that includes terrain clearance safety, slope and elevation gradient, emergency accessibility, weather and shelter avoidance. The task adaptive weight mapping module is used to load task adaptive weight vectors according to task type; Improve the A-star search module to perform multi-layer hard constraint pruning with at least terrain clearance constraints and mountain shading angle constraints before adjacent candidate nodes enter the open list. Use a lower bound heuristic function for search and output at least one candidate route. The particle swarm optimization module is used to initialize particle positions with candidate routes as seeds and initialize particle velocities based on gradient information of the multi-objective comprehensive cost function. The target route is obtained through particle swarm optimization iteration and multi-constraint legality repair. The route output module is used to output the target route.

[0028] Compared with existing technologies, the advantages of this invention are: 1. By employing multi-layered hard constraint pruning before queuing, the search space is reduced from the source, significantly improving path search efficiency in a 3D grid environment. This application performs four layers of hard checks before adjacent candidate nodes enter the open list: terrain clearance constraints, airspace control constraints, mountain shading angle constraints, and strong convection high probability constraints. Nodes that do not meet any of these constraints are directly intercepted and do not participate in priority queue sorting or subsequent expansion. Compared to the traditional A algorithm's approach of only performing feasibility checks after nodes are popped from the open list, this invention's "pre-queue interception" mechanism reduces the peak size of the open list from the source, reducing the number of comparisons and adjustments in heap sort. In a 3D grid space, the number of nodes increases cubically with increasing spatial resolution. This mechanism ensures that the actual number of nodes entering the open list during the search process is only a fraction to a few tens of times that of traditional methods (depending on terrain complexity), thus significantly reducing the algorithm's time and space overhead. More importantly, this pruning strategy works synergistically with the lower bound heuristic function—pruning ensures that all nodes entering the open list are potentially feasible nodes, and the lower bound heuristic function ensures the strong directionality of A search. Together, they ensure that the algorithm can still efficiently find feasible solutions in extremely dense infeasible regions (such as dense canyon regions), which is an effect that is difficult to achieve by using any strategy alone.

[0029] 2. Establish a unified quantitative framework for multi-source risks to achieve continuous and fine-grained expression of topographic, meteorological, sheltering, and emergency risks. This application integrates six factors—topographic clearance safety, slope and elevation gradient, emergency accessibility, mountain sheltering angle, strong convection probability, and instantaneous wind speed—into a unified cost function, overcoming the shortcomings of existing technologies where multi-source risks are "modeled independently and difficult to quantify uniformly." Specifically: On the one hand, this application replaces the binary visibility judgment in the prior art with a continuous mountain occlusion angle. Existing methods usually treat occlusion as a binary result of "visibility / no visibility," failing to distinguish between different states such as "complete exposure," "critical occlusion," and "deep occlusion." The continuous occlusion angle definition of this invention can express the continuous risk change of the communication or navigation link quality gradually decreasing when a node approaches the critical occlusion state, making the cost function more precise in characterizing the risk. This design is particularly suitable for plateau, mountain, and canyon terrain—in such terrain, the relative positional relationship between the aircraft and the mountain continuously changes, and binary models cannot capture the gradual degradation of link quality, while continuous models can guide the flight path to actively maintain a safe angular distance from the ridgeline.

[0030] On the other hand, this application weightedly fuses historical severe convective statistical probabilities with real-time radar extrapolation probabilities to form a time-varying probability of frequent severe convective events, and incorporates the predicted arrival time of path nodes as a time index into cost calculation. This allows flight route planning to simultaneously consider both the long-term climate statistical characteristics of the region and short-term real-time weather change trends, achieving true "dynamic weather avoidance" rather than relying solely on static weather risk layers for planning. As is well known to those skilled in the art, local severe convective events in plateau and mountainous areas are characterized by their suddenness, short lifespan, and small spatial scale. Static risk layers cannot reflect these time-varying characteristics, and the fusion prediction mechanism of this invention effectively compensates for this deficiency.

[0031] 3. By using the tightly coupled design of A-star-PSO, we can achieve a deep integration of discrete search and continuous optimization, which can accelerate convergence and improve the quality of the solution.

[0032] This application breaks through the "loosely coupled" technical paradigm of existing hybrid A and PSO methods. In existing methods, the A stage typically outputs only one path as the initial solution for the PSO stage, and the information transfer between the two stages is extremely limited. The PSO stage actually starts searching from scratch after "forgetting" the cost field information explored in the A stage. This application achieves "tight coupling" in two aspects: First, the A-stage outputs Top-K candidate routes instead of a single path, which are then resampled and used as seeds for PSO particle positions. The Top-K strategy preserves the various excellent solutions with different structures found in the A-stage search, giving the initial population of the particle swarm structural diversity and avoiding premature convergence caused by a single optimal solution. This design allows particle swarm optimization to stand on the "shoulders of giants" of the A-stage in the early stages of iteration, rather than starting from random positions.

[0033] Secondly, this application uses the negative gradient of the comprehensive cost function at the particle path as the initial velocity of the particle. The core idea of ​​this design is that the A-stage explores and accumulates the distribution information of the global cost field in the three-dimensional grid space. This information is approximated by finite difference to the cost descent direction of each node on the particle path, and explicitly encoded as the initial velocity vector of the PSO particle. Compared with the traditional method of initializing with a zero vector or random vector, the "cost field-guided" initial velocity of this invention gives the particle a clear cost descent direction in the first iteration, significantly shortening the convergence path of the PSO stage. Experiments show that this tightly coupled initialization strategy can reduce the number of iterations required for PSO to reach the same target cost value by approximately 100%, and the final convergence comprehensive cost is lower than that of the random initialization scheme.

[0034] As is well known to those skilled in the art, Discrete Search (A*) and Continuous Optimization (PSO) belong to two different types of optimization paradigms, and the information transfer and fusion between them is not an obvious conventional design. This invention achieves deep fusion of the two algorithms through a tightly coupled scheme of "Top-K resampling + negative gradient initialization." The resulting accelerated convergence and performance improvement exceed the reasonable expectations of those skilled in the art, constituting an "unexpected technical effect."

[0035] 4. This application utilizes a task-adaptive weight mapping table to load corresponding weight vectors via configuration injection before the search begins, based on different task types such as tourism, agricultural and forestry operations, emergency rescue, power transmission line inspection, and disaster investigation. This design allows the same algorithm pipeline to adapt to multiple operational scenarios without modifying the core code, significantly reducing the time cost of algorithm engineering deployment and on-site parameter tuning. Those skilled in the art typically need to maintain an independent planning algorithm for each type of task or perform extensive parameter retuning when dealing with multiple task scenarios. This invention solves this engineering pain point through weight configuration and has strong practical value.

[0036] 5. This application performs multi-constraint legality repair on particle paths during PSO iteration, including operations such as insufficient airspace adjustment, no-fly zone regression, masking angle exceeding limits replacement, strong convection exceeding limits replacement, and gradient exceeding limits interpolation repair. This effectively solves the hard constraint failure problem caused by continuous perturbation in particle swarm optimization. Simultaneously, this invention performs a global feasibility check on remaining battery power before outputting the flight path, incorporating remaining battery power, minimum return battery power threshold, full-charge range, and high-altitude range reduction coefficient into the feasibility judgment, ensuring the final output flight path is executable in actual flight. The above design makes the flight path planning results of this invention not only theoretically multi-objective optimal but also possess engineering-level flightability. Attached Figure Description

[0037] Figure 1 This is a structural block diagram of the plateau and mountain low-altitude air route planning system of this application.

[0038] Figure 2 This is a flowchart of the low-altitude air route planning method for plateau and mountainous areas proposed in this application.

[0039] Figure 3 The flowchart for the improved A-star queuing process with hard constraints pruning in the first four layers is provided for this application.

[0040] Figure 4 The flowchart for A-Star-PSO tight-coupled population initialization and particle swarm optimization provided in this application. Detailed Implementation

[0041] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0042] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0043] Please see Figure 1-4 A method for low-altitude flight path planning in high-altitude and mountainous areas based on improved A-satellite and particle swarm optimization, such as... Figure 2 As shown, it includes the following steps: Collect aircraft status parameters, 3D terrain data of plateau and mountainous areas, airspace control data, meteorological coupling data, and mission prior configuration data; Three-dimensional terrain data, airspace control data and meteorological coupled data are integrated into a three-dimensional terrain-meteorological coupled raster model, so that each raster node can simultaneously carry fields of terrain height, slope, airspace attributes, mountain shading angle, probability of severe convection and wind speed prediction. Candidate routes Represented as a sequence of nodes: , in, Starting from, For the target point, For nodes The predicted arrival time.

[0044] Constructing a multi-objective integrated cost function based on a 3D terrain-meteorology coupled raster model: , in, To pay the price for terrain clearance safety, The cost is the slope and elevation gradient. For the cost of emergency accessibility, The costs are related to weather and shelter avoidance, and the external weights corresponding to each cost satisfy the following: , , This represents the external weight vector of the multi-objective integrated cost function. Read the task adaptive weight vector based on the current task type. The task adaptive weight vector is loaded into the multi-objective comprehensive cost function; the task type includes at least one of tourism, agricultural and forestry operations, emergency rescue, power transmission line inspection and disaster investigation; the task adaptive weight vector is loaded into the multi-objective comprehensive cost function by configuration injection before the search starts.

[0045] In one implementation, the task adaptive weight mapping table is shown in Table 1: Table 1 Task Adaptive Weight Mapping Table

[0046] The above weight values ​​are exemplary configurations. In practical applications, adjustments can be made based on aircraft type, mission level, operating area, airspace requirements, and meteorological risk level.

[0047] The first stage of the search is performed using an improved A* algorithm, which includes the following improvements: For the current node, generate adjacent candidate nodes, and before each adjacent candidate node enters the open list, sequentially determine the terrain clearance constraints, airspace control constraints, mountain shielding angle constraints, and strong convection high probability constraints; if any constraint is not satisfied, block the adjacent candidate node so that the adjacent candidate node does not enter the open list. The A* search is performed using a lower bound heuristic function, which is: , in, For the target node, For the current node To the target node The lower bound of the three-dimensional Euclidean distance or grid distance. This is the lower bound of the minimum feasible cost per unit distance.

[0048] The improved A* algorithm uses the following evaluation function: , in, From the starting point to the current node The accumulated costs, The heuristic cost from the current node to the target node.

[0049] The actual cost of improving the A* algorithm is recursively calculated as follows: , in, For the current node, For adjacent candidate nodes, the edge cost is: , in, The distance between adjacent nodes. For nodes The local comprehensive risk cost, This is a scaling factor for risk cost. If a neighboring node cannot obtain terrain, airspace, or weather fields, then that neighboring node is considered an infeasible node and will not be added to the open list.

[0050] Output the Top-K candidate route set The improved A-star search phase can output not only a single optimal path, but also a set of Top-K candidate routes: , Among them, the candidate routes are selected according to the comprehensive cost function. Sort the candidate routes from lowest to highest. This Top-K candidate route set provides a high-quality initial population structure for the subsequent particle swarm optimization stage. If the number of feasible candidate routes is less than... The number of actual feasible candidate routes is used as the seed number, and the particle swarm size is supplemented by perturbation replication.

[0051] like Figure 4 As shown, the A-star-PSO tightly coupled population initialization involves resampling the candidate routes in the Top-K candidate route set to a fixed length. 3D node sequence: , in: , , The starting and target points remain fixed. If a resampled interpolation point falls into an invalid grid, it is projected to the nearest valid grid or a multi-constraint validity correction is triggered. Indicates the starting point of the flight mission. This is a special notation introduced in the context of particle swarm optimization to highlight the boundary condition of "keeping the starting and target points fixed," avoiding association with intermediate nodes on the path. Confusing.

[0052] Using the Top-K candidate route set as the seed set for particle swarm optimization, the candidate routes are resampled into a fixed-length 3D node sequence as particle positions. The negative gradient of the multi-objective integrated cost function at the particle path is used as the initial velocity of the particle. , in, Initialize the step size for the velocity. For random disturbance terms, This is the finite difference approximation gradient calculated by applying a preset perturbation to the intermediate nodes of the particle path. For the first The position vector of each particle at the initial moment; the finite difference approximation of the negative gradient is calculated as follows: for each intermediate node in the particle path, respectively at... , , Apply a preset perturbation in the direction and calculate the comprehensive cost function. The increment, and constructed from the direction of the increment. Endpoint and It does not participate in velocity disturbances.

[0053] The target route is obtained through particle swarm optimization iteration and multi-constraint legality repair, and the waypoint sequence corresponding to the target route is output.

[0054] In one implementation, approximately 70% of the particles are generated by sequentially selecting and resampling Top-K candidate routes to preserve the excellent solution structure of the A-star stage; approximately 30% of the particles are generated by randomly selecting from the Top-K candidate routes and superimposing Gaussian noise to maintain population diversity.

[0055] When resampling the Top-K candidate route set into particle positions, 60% to 80% of the particles are generated by sequentially selecting from the Top-K candidate route set, while the remaining particles are generated by randomly selecting from the Top-K candidate route set and superimposing Gaussian noise. The resampling process keeps the starting point and target point fixed, and performs legality repair on interpolation points that fall into illegal grids.

[0056] Mountain shading angle For nodes The horizontal projection is centered and has a radius of surface sampling set Relative nodes of each sampling point The maximum elevation angle is calculated as follows: , in, Sampling points The height of the terrain or obstacles at that location For nodes Flight altitude For nodes With sampling points Horizontal distance, To prevent the minimum distance constant from having a denominator of zero. Can be excluded from satisfying The sampling points, or through A lower bound is provided for the denominator. This definition uses the maximum elevation angle as an indicator of continuous shading risk. Compared to binary models that only determine whether the line of sight is blocked, this application can characterize the continuous risk changes when a node approaches a critical shading state.

[0057] Weather and the Cost of Sheltering The following results were obtained by normalizing the combination of mountain shielding angle, probability of high incidence of severe convection, and instantaneous wind speed:

[0058] in, Candidate routes The number of nodes, For the first 1 path node For nodes The predicted arrival time, It is a monotonically non-decreasing normalized function. Soft threshold for the mountain shading angle, The safe wind speed threshold for aircraft, Path node At its predicted arrival time The probability of strong convection is high. Path node At its predicted arrival time Instantaneous wind speed, , and Let the internal weights be the weights, and satisfy the following: , .

[0059] The probability of frequent severe convective weather is obtained by weighted fusion of historical statistical terms and real-time radar extrapolation terms: , in, In one implementation, The value is set between 0.3 and 0.5 to take into account both long-term climate statistics and short-term real-time weather changes. For nodes Historical statistical probability of strong convection, For nodes At any moment Real-time strong convection prediction probability.

[0060] The terrain clearance safety cost only penalizes the portion of the clearance deficiency, and its calculation method is as follows: , in, The minimum terrain clearance safety threshold, For nodes Flight altitude Indicate extraction The clearance at the location is insufficient. This indicates the minimum safe flight altitude. When the flight altitude is equal to or higher than the minimum safe flight altitude, This expression sets the penalty for nodes where the net clearance meets the threshold as follows: Nodes with insufficient clearance will be penalized proportionally to the deficiency, and through... Limit the upper bound of single-node penalty.

[0061] The costs of slope and elevation gradient are calculated from the height difference between adjacent path nodes, the horizontal distance, and the maximum allowable climb or descent gradient of the aircraft: , in, The maximum allowable climb or descent gradient for the aircraft. For nodes With nodes The horizontal distance. Using absolute height difference can simultaneously constrain both the ascent and descent sections. This can prevent division by zero from occurring when the horizontal distance between adjacent nodes is too small.

[0062] The emergency accessibility cost is calculated as the ratio of the distance from the path node to the nearest preset emergency landing point to the emergency accessibility distance threshold: , in, For nodes Distance to the nearest pre-designated emergency landing point This is the emergency reachability threshold.

[0063] The particle swarm optimization iteration uses the following update formula: , , in, For the first Replace inertial weights, and As a learning factor, and For interval Internal random number, For the first The historical best position of each particle The optimal position for the entire group. Indicates the first The particle in the first The position vector of the substitute, Indicates the first The particle in the first The velocity vector of the generation.

[0064] Inertia weight By the following method linearly decreasing to : , in, The maximum number of iterations, This represents the maximum value of the inertia weight in the Particle Swarm Optimization (PSO) algorithm. This represents the minimum value of the inertia weight in the Particle Swarm Optimization (PSO) algorithm. Let represent the current iteration number of the particle swarm optimization. This allows for a wider exploration range in the early stages and faster convergence in the later stages. Iteration termination conditions include reaching the maximum number of iterations. If the decrease in global optimal cost is less than a preset threshold for several consecutive generations, or if the target route has simultaneously met the hard constraints and the global feasibility verification of remaining power capacity.

[0065] The multi-constraint legality repair includes: raising nodes with insufficient airspace to a level greater than or equal to the sum of the terrain height and the minimum terrain airspace safety threshold; reverting nodes that have entered no-fly zones or controlled airspace to the nearest legal airspace boundary; replacing nodes with excessive mountain shielding angles with candidate points with low shielding angles in the neighborhood; replacing nodes with excessive probability of severe convection with the nearest meteorological safety candidate point; and repairing nodes whose height difference exceeds the upper limit of the ascent or descent gradient by inserting intermediate transition nodes or adjusting node heights when the height difference between adjacent nodes exceeds the upper limit of the ascent or descent gradient; if no legal candidate point exists within the preset neighborhood radius, expanding the neighborhood to the preset upper limit, and if no legal candidate point still exists, reverting to the previous legal node and triggering a re-search or determining that the current particle is infeasible.

[0066] Multi-constraint legality repair specifically includes: After each generation of particles is updated, the validity of intermediate nodes in the particle path is corrected: If the node has insufficient clearance, raise the node height to: , in, This represents an additional safety margin beyond the minimum terrain clearance safety threshold. Represents a node The height of the terrain or obstacles at the location.

[0067] If a node enters a no-fly zone or controlled airspace, it will be moved back to the nearest point outside the legal airspace boundary. If the shading angle of a node exceeds the hard threshold, it is replaced with a candidate point with a lower shading angle in the neighborhood. If the probability of a node experiencing severe convection exceeds the hard threshold, it will be replaced with the nearest meteorological safety candidate point. If the height difference between adjacent nodes exceeds the upper limit of the climbing or descending gradient, it can be repaired by inserting an intermediate transition node or adjusting the node height.

[0068] When the aforementioned replacement-type repair fails to find a valid candidate point within a preset neighborhood radius, the system expands the neighborhood to a preset upper limit. If no valid candidate point is found, the system either reverts to the previous valid node and triggers a local re-search, or marks the current particle as infeasible and imposes a penalty in the current generation's fitness evaluation. Through these repairs, the particle swarm optimization phase can maintain the validity of the hard constraints of the flight path after continuous path perturbations.

[0069] This further includes a global feasibility check of the remaining battery power, which includes: Before outputting the target flight route, a global feasibility check of the remaining battery power can be performed. The current maximum flyable distance is: , in This represents the current remaining battery percentage. Minimum safe return-to-home battery threshold Maximum range on a full charge High-altitude range reduction factor.

[0070] The current route mileage is: , in This represents the Euclidean distance between adjacent waypoints. These are the waypoint coordinates.

[0071] If the following conditions are met: , If the target route meets the global feasibility requirement for remaining battery power, the system will determine that it does not. Otherwise, the system will return to the A-satellite search phase to regenerate candidate routes, or increase the emergency reachability weight and re-execute PSO optimization.

[0072] Construction of a 3D terrain-meteorology coupled raster model: In this embodiment, the planning area is discretized into a three-dimensional grid. Each grid node It should include at least the following fields: , in, For spatial coordinates, The height of the terrain or obstacle. This is information related to terrain slope. For airspace attributes, To shield the corner of the mountain, This indicates a high probability of severe convection. This refers to the instantaneous wind speed.

[0073] 3D terrain data can be derived from SRTM, ASTER GDEM, ALOS World 3D, national DEM, DSM, or airborne LiDAR mapping data. Meteorological coupled data can be derived from radar echoes, short-term forecasts, wind field grid data, and airborne meteorological sensor data. Airspace control data includes no-fly zones, temporary control zones, low-altitude routes, and altitude restrictions.

[0074] like Figure 3 As shown, the improved hard constraint pruning of the first four layers before queuing A-star includes: For the current node Generate 26 candidate neighbor nodes. For each candidate neighbor node... Before it enters the open list, the following checks are performed in sequence: First, terrain clearance constraints: ; Second, airspace control constraints: nodes It is not located in a no-fly zone or temporary control zone, and meets the altitude restrictions; Third, the constraint of the mountain's shading angle: , The hard threshold for the mountain shading angle.

[0075] Fourth, the constraint of the high probability of severe convection: , The hard threshold is set for the probability of strong convection.

[0076] If any constraint is not satisfied, the adjacent candidate node is skipped and not added to the open list. This avoids obviously infeasible nodes from participating in priority queue sorting and subsequent expansion. If all adjacent candidate nodes of the current node have been pruned, the search continues to the next candidate node in the open list; if the open list is empty, it is determined that there is no feasible route that satisfies the constraints under the current parameters or airspace conditions, and a failure flag is returned.

[0077] This application also includes a high-altitude, low-altitude flight path planning system for mountainous terrain based on an improved A-satellite and particle swarm optimization approach, such as... Figure 1 As shown, it includes: The aircraft status parameter acquisition module is used to collect the aircraft's attitude, speed, and remaining battery power. The 3D terrain data acquisition module is used to acquire or access DEM, DSM, or LiDAR terrain data. The airspace control data acquisition module is used to access data from no-fly zones, temporary control zones, low-altitude routes, and altitude layers. The meteorological coupled data acquisition module is used to access radar echoes, short-term forecasts, wind field grids, and airborne meteorological sensor data. The task prior configuration module is used to access the whitelist of task types, starting points, target points, and emergency landing points; The 3D terrain-meteorology coupled raster model construction module is used to construct 3D terrain-meteorology coupled raster models. The multi-objective integrated cost function construction module is used to construct multi-objective integrated cost functions; The task adaptive weight mapping module is used to load task adaptive weight vectors according to task type; Improve the A-star search module to perform hard constraint pruning and lower bound heuristic search before queuing and output a Top-K candidate route set; A-Star-PSO tightly coupled population initialization module is used to resample the Top-K candidate routes into particle positions and initialize particle velocities with the negative gradient of the comprehensive cost function. The particle swarm optimization refinement module is used to perform particle swarm optimization iterations and multi-constraint validity repair. The route output module is used to output the target route.

[0078] Examples of key parameters are shown in Table 2: Table 2 Examples of Key Parameters

[0079] The parameters described above are only typical values ​​used in the embodiments and are not intended to limit the scope of protection of this invention.

[0080] In an emergency rescue scenario involving a landslide in the mountainous region of Yunnan Province, the starting point is the mountain rescue base, and the target point is the landslide site. This area has a high altitude, steep slopes, and narrow nearby valleys. The system loads the emergency rescue task weights as follows: , Among these, emergency accessibility has a higher weight, making flight routes more likely to pass through or remain close to pre-designated emergency landing points. The parameter can be set as follows: , , , , , , , .

[0081] During planning, the system first constructs a 3D terrain-meteorological coupled grid model and performs hard constraint pruning of the first four layers before queuing in Phase A, preventing nodes with insufficient airspace, entering controlled airspace, exceeding shielding limits, or exceeding the probability of strong convection from entering the open list. After outputting the Top-K candidate routes in Phase A, the system resamples the candidate routes into fixed-length particles based on cumulative arc length and initializes the particle velocities using finite difference approximation of negative gradients. During PSO iteration, the system performs airspace, airspace, shielding, strong convection, and slope legality repair on the particle paths; if there are no legal replacement points in the neighborhood, a local re-search is triggered. The final output waypoint sequence satisfies the global feasibility check of remaining battery power and forms explicit passages near multiple candidate emergency landing points, reducing the cost of emergency accessibility.

[0082] In the inspection of power transmission lines along mountain ridges, the system loads the inspection task weights: , This weighting setting prioritizes terrain clearance safety and the costs associated with slope and elevation gradients, making it suitable for maintaining stable clearance above the route and controlling the magnitude of ascents and descents. The parameters can be set as follows: , , , , , , .

[0083] During the planning process, the Improved A phase prunes pre-entry routes based on terrain clearance, no-fly zones, mountain shielding angles, and the probability of severe convection, automatically bypassing severe convection sections marked by short-term forecasts and selecting low-risk nodes that meet shielding angle constraints between ridges and valleys. The PSO phase further smooths the node sequence and optimizes the overall cost, inserting transition nodes or adjusting node heights in sections where the height difference between adjacent nodes exceeds the upper limit of the climb or descent gradient, so that the final route is superior to the initial A candidate route in terms of airspace safety, climb and descent smoothness, and weather shielding risk.

[0084] In high-altitude mountainous scenic route missions, the system input includes the sightseeing route's origin, destination, permitted flight area, scenic route preferences, DEM topographic data, wind field prediction data, and communication obstruction risk sampling data. The system loads tourism and sightseeing mission weights: , This weighting setting increases the proportion of weather and obstruction avoidance costs, preventing flight paths from entering canyon areas with high obstruction angles for extended periods, and favoring ridges or open areas with more stable communication and navigation links and wider visibility. The initial speed step size can be set to... To reduce the magnitude of negative gradient velocity and improve the diversity of particle swarm search, the system performs a global feasibility check of remaining battery power before outputting the target waypoint sequence. If the check fails, it returns to the candidate route generation stage to search again, thus ensuring that the final output route simultaneously meets the sightseeing mission preferences, weather obstruction avoidance, and endurance constraints.

[0085] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. A highland mountain low-altitude air route planning method based on improved A-star and particle swarm optimization, characterized in that, Includes the following steps: Collect aircraft status parameters, 3D terrain data, airspace control data, meteorological coupling data, and mission prior configuration data; Three-dimensional terrain data, airspace control data and meteorological coupled data are integrated into a three-dimensional raster model, so that each raster node carries information on terrain height, airspace attributes, mountain shading angle, strong convection probability and wind speed. A multi-objective comprehensive cost function is constructed based on a 3D raster model, which includes terrain clearance safety, slope and elevation gradient, emergency accessibility, weather and shelter avoidance. Load the task adaptive weight vector into the multi-objective comprehensive cost function according to the task type; The improved A* algorithm is used for the first stage of search. Before adjacent candidate nodes enter the open list, the improved A* algorithm performs multi-layer hard constraint pruning with at least terrain clearance constraints and mountain shading angle constraints. The lower bound heuristic function is used for search and outputs at least one candidate route. The particle positions for particle swarm optimization are initialized using candidate routes as seeds, and the particle velocities are initialized based on the gradient information of the multi-objective comprehensive cost function. The target route is obtained and output through particle swarm optimization iteration and multi-constraint legality repair.

2. The method for low-altitude flight path planning in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in claim 1, characterized in that, The multi-layer hard constraint pruning includes, in sequence: Terrain clearance constraint: Determine whether the flight altitude of adjacent candidate nodes is greater than or equal to the sum of the terrain height and the minimum terrain clearance safety threshold; Airspace control constraints: Determine whether adjacent candidate nodes are located outside no-fly zones or temporary control zones; Mountain shading angle constraint: Determine whether the mountain shading angle of adjacent candidate nodes is less than or equal to a preset hard threshold; Strong convection high incidence probability constraint: Determine whether the probability of strong convection high incidence of adjacent candidate nodes at the predicted arrival time is less than or equal to a preset probability hard threshold. If any constraint is not met, the adjacent candidate node is blocked so that it is not included in the open list.

3. The method for low-altitude flight path planning in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in claim 2, characterized in that, The mountain's shading angle Calculate using the following formula: ， in, For nodes The horizontal projection is centered and has a radius of The surface sampling set, Sampling points The terrain elevation at that location For nodes Flight altitude For nodes With sampling points Horizontal distance, To prevent the minimum distance constant from having a denominator of zero; The mountain shading angle is used for both: hard threshold determination in the mountain shading angle constraint and soft penalty term in the meteorological and shading avoidance cost.

4. The method for low-altitude flight path planning in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in claim 3, characterized in that, The costs of weather and shelter avoidance for: in, Candidate routes The number of nodes, For the first 1 path node For nodes The mountain's sheltered angle, For nodes The predicted arrival time, It is a monotonically non-decreasing normalized function. Soft threshold for the mountain shading angle, The safe wind speed threshold for aircraft, Path node At its predicted arrival time The probability of strong convection is high. Path node At its predicted arrival time Instantaneous wind speed, , , Internal weights and satisfying .

5. The method for low-altitude flight path planning in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in claim 4, characterized in that, The high probability of severe convection It is obtained by weighted fusion of historical statistical terms and real-time radar extrapolation terms: ， in, , For nodes Historical statistical probability of strong convection, For nodes At any moment Real-time strong convection prediction probability.

6. The method for low-altitude flight path planning in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in claim 1, characterized in that, The terrain clearance safety cost in the multi-objective comprehensive cost function for: ， in, The minimum terrain clearance safety threshold, For nodes Flight altitude Indicate extraction The clearance at the location is insufficient. This indicates the minimum safe flight altitude. When the flight altitude is equal to or higher than the minimum safe flight altitude, ; Slope and elevation gradient costs for: ， in, The maximum allowable climb or descent gradient for the aircraft. For nodes With nodes Horizontal distance; The emergency accessibility cost is calculated as the ratio of the distance from the path node to the nearest preset emergency landing point to the emergency accessibility distance threshold: ， in, For nodes Distance to the nearest pre-designated emergency landing point This is the emergency reachability threshold.

7. The method for low-altitude flight path planning in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in claim 1, characterized in that, The initialization of particle velocity based on gradient information from the multi-objective comprehensive cost function is specifically as follows: ， in, Initialize the step size for the velocity. For random disturbance terms, For the first The position vector of each particle at the initial moment. This is the finite difference approximation gradient calculated by applying a preset perturbation to the intermediate nodes of the particle path; 60% to 80% of the particles are generated by selecting and resampling the candidate routes in sequence, while the remaining particles are generated by randomly selecting the candidate routes and superimposing Gaussian noise. The resampling process keeps the starting point and target point fixed, and performs legality repair on interpolation points that fall into illegal grids.

8. The method for low-altitude flight path planning in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in claim 1, characterized in that, The multi-constraint legality repair includes: Raise the height of nodes with insufficient clearance to be greater than or equal to the sum of the terrain height and the minimum terrain clearance safety threshold; Nodes that have entered no-fly zones or controlled airspace will be moved back to the point outside the nearest legal airspace boundary; Replace nodes with excessive mountain shading angles with candidate points with low shading angles within the neighborhood. Replace the nodes with the highest probability of severe convection with the nearest meteorological safety candidate points; When the height difference between adjacent nodes exceeds the upper limit of the climbing or descending gradient, it can be repaired by inserting intermediate transition nodes or adjusting the node height; If no valid candidate point is found within the preset neighborhood radius, the neighborhood is expanded to the preset upper limit. If no valid candidate point is found, the search is reverted to the previous valid node and a new search is triggered or the current particle is deemed infeasible.

9. The method for low-altitude flight path planning in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in claim 1, characterized in that, The task types include at least one of tourism, agricultural and forestry operations, emergency rescue, power transmission line inspection, and disaster survey. Task adaptive weight vector satisfy: ， Weight vectors for different task types are loaded into the multi-objective integrated cost function via configuration injection before the search begins.

10. A low-altitude flight path planning system for plateau and mountainous terrain based on improved A-satellite and particle swarm optimization, characterized in that, The method for planning low-altitude flight paths in high-altitude mountainous areas based on improved A-satellite and particle swarm optimization as described in any one of claims 1-9 includes: The data acquisition module is used to collect aircraft status parameters, 3D terrain data, airspace control data, meteorological coupling data, and mission prior configuration data. The 3D raster model building module is used to fuse 3D terrain data, airspace control data, and meteorological coupled data into a 3D raster model. The integrated cost function construction module is used to construct a multi-objective integrated cost function that includes terrain clearance safety, slope and elevation gradient, emergency accessibility, weather and shelter avoidance. The task adaptive weight mapping module is used to load task adaptive weight vectors according to task type; Improve the A-star search module to perform multi-layer hard constraint pruning with at least terrain clearance constraints and mountain shading angle constraints before adjacent candidate nodes enter the open list. Use a lower bound heuristic function for search and output at least one candidate route. The particle swarm optimization module is used to initialize particle positions with candidate routes as seeds and initialize particle velocities based on gradient information of the multi-objective comprehensive cost function. The target route is obtained through particle swarm optimization iteration and multi-constraint legality repair. The route output module is used to output the target route.

Images