An improved GBNN algorithm-based path planning method for unmanned aerial vehicle cluster in complex environment

Abstract

This invention discloses a path planning method for UAV swarms in complex environments based on an improved GBNN algorithm. It constructs a 3D grid map of the complex environment by quantifying environmental complex factors such as restricted areas and performance degradation areas. Combining the flight characteristics of UAV swarms, it sets individual UAV constraints and collision avoidance constraints within the swarm, establishing a multi-objective optimization function with total path length, risk, and safety as its core. Based on the improved GBNN algorithm, it filters candidate neighborhood subsets and performs multi-step linear prediction. The optimal path is selected through path intersection detection and length-turning cost filtering. After cubic B-spline curve smoothing, the optimal flight path that satisfies the constraints is output. This invention can effectively reduce the path risk of UAV swarms in complex environments, improve path safety and real-time planning, while simultaneously considering path length optimization and collision avoidance capabilities, thereby improving the efficiency of swarm task execution.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) path planning technology, and in particular to a UAV swarm path planning method based on an improved GBNN algorithm in complex environments. Background Technology

[0002] The path planning method based on the discrete biologically inspired neural network algorithm GBNN learns environmental information and the motion characteristics of intelligent devices through iterative optimization of neuron activity values, thereby generating a path that meets the constraints, and is suitable for autonomous navigation scenarios in complex environments.

[0003] Currently, common optimization strategies for GBNN path planning in complex environments include neighborhood search optimization and single-target weight allocation. However, these strategies still have the following core problems in practical applications: (1) Insufficient real-time performance: Traditional and existing improved GBNN algorithms often need to traverse all three-dimensional neighborhood grids of the current position, resulting in a large amount of computation and making it difficult to meet the strong real-time requirements of UAV swarm path planning; (2) Prone to getting trapped in local optima: The lack of a global path prediction mechanism makes it easy for the planned path to make redundant turns, increasing flight energy consumption and reducing flight safety; (3) Poor multi-target adaptability: The evaluation system of existing algorithms is not adaptable to complex scenarios and it is difficult to simultaneously take into account the comprehensive optimization requirements of path length, risk and safety. The above problems make it impossible for existing GBNN algorithms to fully meet the requirements of efficient and safe path planning for UAV swarms in complex environments, thus limiting their engineering application in complex scenarios. Summary of the Invention

[0004] Purpose of the invention: The present invention aims to provide a method for UAV swarm path planning in complex environments based on the improved discrete biologically inspired neural network algorithm GBNN, which takes into account low risk, high security and strong real-time performance.

[0005] Technical solution: The UAV swarm path planning method based on the improved GBNN algorithm in complex environments described in this invention includes the following steps: (1) Considering restricted areas and performance degradation areas, quantify the accessibility of the grid and the impact of performance degradation areas to construct a three-dimensional complex environmental threat grid map; (2) Establish single-aircraft flight constraints and collision avoidance constraints within the cluster; (3) Based on the total path length, path risk, and path safety, construct a multi-objective optimization function; (4) Introduce the Euclidean distance from the passable grid to the next task target point and the influence of the performance degradation region, calculate the neuron activity value of the passable grid, and filter the neighboring grids to obtain the candidate neighborhood subset in combination with the global neuron activity value distribution. Perform multi-step path prediction for each candidate grid in the candidate neighborhood subset. For the multi-step path prediction results, perform preliminary optimization based on the intersection point for path length cost. When the path length cost is the same, introduce the direction continuity optimization strategy to perform secondary optimization and obtain the optimized multi-step path prediction results. After path smoothing and constraint verification, complete the UAV swarm path planning. (5) Solve for the optimal flight path based on the multi-objective optimization function.

[0006] Furthermore, the degree of impact of the performance degradation region. for ; In the formula, The maximum impact intensity in the performance degradation region. This is the performance degradation coefficient. The straight-line distance from the current grid cell to the center of the sphere in the performance degradation region. Let be the radius of the sphere in the performance degradation region.

[0007] Furthermore, the multi-objective optimization function is: ; ; ; ; In the formula, For comprehensive optimization, The total path length is used to evaluate the function value. This is the value of the path risk assessment function. The value of the path security evaluation function. , , These are the weighting coefficients; For the number of drone swarms, This represents the total number of waypoints for a single drone. For the first The drone in From the path point to the first Flight distance of each waypoint For the first The first drone Path points, path point The number of obstacle grids in the neighborhood; For the first The complete path of the drone For the first Drone route The number of times the course changes.

[0008] Furthermore, the activity value of grid neurons for ; In the formula, The target incentive coefficient, The performance degradation suppression coefficient, The distance from the current grid to the next target point of the UAV is the Euclidean distance. This indicates the degree of impact of the performance degradation region.

[0009] Furthermore, the multi-step path prediction formula is as follows: ; In the formula, For the first One predicted path point, For the next step of selecting grid cells, For neuron indexing, To predict the number of steps, To predict the step size, This is the heading vector.

[0010] Furthermore, the steps to determine the intersection point are as follows: Based on the multi-step path prediction formula, for the next candidate grid , generate subsequent Step path point set For each candidate location within the candidate neighborhood subset , generate subsequent Step path point set Intersection grid : ; In the formula, For path set The first in Path points, For path set The first in There are path points.

[0011] Furthermore, based on the initial optimization of path length cost at the intersection point, the next candidate grid is selected. for ; In the formula, , These represent the candidate positions. Next step: candidate grid To the intersection grid Euclidean distance, Indicates if intersecting grids exist , Indicates that there are no intersecting grids. .

[0012] Furthermore, the secondary selection process is as follows: ; ; In the formula, , These represent the candidate positions. Next step: candidate grid The corresponding turning cost, This refers to the turning angle at the turning point of the path.

[0013] Furthermore, single-aircraft flight constraints include range constraints, altitude constraints, pitch angle constraints, and turn angle constraints.

[0014] Beneficial Effects: Compared with existing technologies, the significant advantages of this invention are: 1. This invention introduces prohibited areas and performance degradation areas to construct a three-dimensional grid map of complex environments (such as mountainous terrain), quantifying the traversability attributes of the grid and the environmental complexity factors of the impact of performance degradation areas; 2. This invention combines the flight characteristics of UAV swarms to set individual constraints such as range, altitude, pitch angle, and turning angle, as well as collision avoidance constraints within the swarm; 3. This invention introduces the Euclidean distance from traversable grids to the next task target point and the impact of performance degradation areas, calculating the neuron activity value of traversable grids. The closer the grid is to the next task target point, the higher the neuron activity value, guiding the UAV to fly towards the target. At the same time, the greater the impact of performance degradation areas, the higher the corresponding path cost, and the lower the neuron activity value, inhibiting UAVs from entering high-cost areas; 4. This invention uses path intersection detection and length-turning cost screening to select the optimal path, and after curve smoothing, outputs the optimal flight path that meets the constraints; 5. This invention can effectively reduce the path risk of UAV swarms in complex environments, improve path safety and real-time planning, while taking into account path length optimization and collision avoidance capabilities, thereby improving the efficiency of swarm task execution. Attached Figure Description

[0015] Figure 1 To improve the GBNN flowchart; Figure 2 This is a top-down view of the simulation scene; Figure 3 A schematic diagram for candidate neighborhood selection; Figure 4 This is a topographic elevation profile. Figure 5 This is a comparison chart of the path planning performance of the present invention with that of the E²QRSA algorithm, the AEGWCPO algorithm, and the traditional GBNN algorithm; Figure 5 (a) in the figure is the effect diagram of the algorithm of the present invention; Figure 5 (b) in the figure shows the effect of the E²QRSA algorithm; Figure 5 (c) in the figure is the effect diagram of the AEGWCPO algorithm; Figure 5 (d) in the figure shows the effect of the traditional GBNN algorithm; Figure 6 A comparison chart of total path lengths; Figure 7 A comparison chart of path risk; Figure 8 This is a comparison chart of path threat levels. Detailed Implementation

[0016] The UAV swarm path planning method based on the improved GBNN algorithm in complex environments, as described in this invention, includes the following steps: S1. Construct a 3D raster map of a complex environment and complete the quantitative processing of terrain obstacles and multi-source regional information.

[0017] like Figure 2 The complex environment shown in the top view, such as Figure 4 The terrain elevation profile shown is used to complete raster modeling and quantization of restricted areas and performance degradation areas; S11. Using the grid method, the complex environmental area of ​​1000m×1000m×500m is discretized into a three-dimensional grid map of 50×50×25. The physical size of a single grid is 20m×20m×20m, and the three-dimensional coordinates of the grid correspond one-to-one with the actual spatial location. S12. Areas with elevations exceeding the maximum or below the minimum flight altitude of the UAV are marked as terrain obstacle grids. The formula for determining terrain obstacle grids is: ; In the formula, This is a terrain obstacle grid identifier (0 represents an obstacle, 1 represents a non-obstacle). For grid elevation, This is the maximum flight altitude of the drone. This represents the minimum flight altitude for the drone. The formula quickly filters out terrain obstacle areas that the drone cannot pass through using an elevation threshold, establishing basic passage constraints from a spatial height dimension to prevent the drone from colliding with terrain or exceeding flight altitude limits. In this embodiment, Take 500m, A grid with an elevation of 50m or less is designated as a terrain obstacle grid.

[0018] The restricted area is a spherical region, and its formula is: ; In the formula, This is a function indicating a restricted area; 0 indicates absolute prohibition (drones are prohibited), and 1 indicates passage is permitted. The straight-line distance from the current grid cell to the center of the no-entry zone is given. Let be the radius of the sphere representing the restricted area. This formula means that drones are prohibited from entering any grid cell whose distance from the center of the restricted area is less than or equal to the radius. In this embodiment, two restricted areas are deployed.

[0019] The performance degradation region is also a spherical region, and its degree of influence is calculated using the following formula: ; In the formula, The degree of impact of the performance degradation region, The maximum impact intensity in the performance degradation region. This is the performance degradation coefficient. The straight-line distance from the current grid cell to the center of the sphere in the performance degradation region. Let be the radius of the sphere representing the performance degradation region. This formula means that, with the center of the performance degradation region as the sphere's radius... Within a spherical region of radius , the closer the grid is to the center of the sphere, the less affected it is. The larger the radius, the greater the corresponding path cost; beyond this radius, the impact is zero. During path planning, grids with low impact and low path cost are prioritized for passage. In this embodiment, one performance degradation region is deployed. Based on the experience of those skilled in the art, the performance degradation coefficient... Take 0.003 as the maximum regional influence intensity Take 1.

[0020] Comprehensive flight constraints are determined as follows: like =0 or =0, then the grid is an absolutely no-fly grid, and the path planning algorithm will prevent drones from entering; like =1 and =1, then the degree of performance degradation As a reference for grid constraints, the path planning algorithm prioritizes passing through grids with minimal impact.

[0021] S13. Preprocess the 3D raster map, remove invalid rasters, label each valid raster with a tuple (coordinate information, status identifier), complete the environment model initialization, and import it into the MATLAB simulation platform.

[0022] S2. Combining the characteristics of UAV swarm flight, construct individual flight constraints and collision avoidance constraints within the swarm.

[0023] S21. Based on the flight performance parameters of the UAV, set single-aircraft flight constraints, including four types of constraints: range, altitude, pitch angle, and turn angle, as the basic basis for judging the single-aircraft flight of the UAV. Flight range constraints: The total length of the planned path must not exceed the maximum range of the drone; in, This is the total path length. For the maximum range of the drone, Given a distance of 2000m, the total length of the planned path is required. m; Height constraints: The drone's flight altitude must be between the minimum and maximum flight altitude; in, The altitude at which the drone flies. Minimum flight altitude for drones This is the maximum flight altitude of the drone. Take 50m, Given a target altitude of 500m, the required flight altitude is... ; Pitch angle constraint: The pitch angle of the drone must not exceed the maximum pitch angle; in, The pitch angle, This is the maximum pitch angle. Take 15°, and the required pitch angle is... ; Turning angle constraints: The turning angle of the drone must not exceed the maximum turning angle; in, For turning angle, Maximum turning angle Take 30°, and the required turning angle is... ; To avoid collisions during drone swarm flight, a minimum safe distance is set between adjacent drones within the swarm. The distance between any two drones must not be less than this minimum safe distance. Swarm collision avoidance constraints: ; in, For the first frame and the first Spacing between drones This is the minimum safe distance. Given a distance of 5m, find the distance between any two drones. When the distance to the drone is detected to be less than 5m during the simulation, the collision avoidance adjustment mechanism is automatically triggered.

[0024] S3. Establish a multi-objective optimization function with total path length, risk, and safety as the core.

[0025] S31. Construct a comprehensive optimization function with the core optimization objectives of minimizing total path length, minimizing path risk, and optimizing path safety; Multi-objective optimization function formula: ; In the formula, For comprehensive optimization, The total path length is used to evaluate the function value. This is the value of the path risk assessment function. The value of the path security evaluation function. , , Let be the weight coefficient, and satisfy... Considering that the focus of this invention is obstacle avoidance and safe flight in complex environments, namely, improving the path risk and safety indicators of the UAV, the weight allocation is adjusted. In this embodiment, based on the experience of those skilled in the art, the weight coefficient is set to... =0.2、 =0.4、 =0.4.

[0026] S32. Quantify the total path length, risk, and safety indicators, and clarify the calculation logic; Formula for total path length: In the formula, This is the evaluation function for the total path length. For the number of drone swarms, This represents the total number of waypoints for a single drone. For the first The drone in From the path point to the first The flight distance to each waypoint. This metric measures flight energy consumption and mission efficiency by the total path length of the drone swarm; a smaller value indicates a shorter path and lower energy consumption.

[0027] Path risk formula: In the formula, This is the path risk assessment function. For the number of drone swarms, This represents the total number of waypoints for a single drone. For the first The first drone Path points, path point The number of obstacle grids in the neighborhood. This metric measures the risk of a path by the number of obstacles in the surrounding grids; a smaller value indicates fewer obstacles in the surrounding grids and a lower path risk.

[0028] Path security formula: In the formula, This is a path safety evaluation function. For the number of drone swarms, For the first The complete path of the drone For the first Drone route The number of course changes during flight. This indicator measures path safety by the number of times the drone turns; a smaller value means fewer turns and higher safety.

[0029] S4. Based on the improved GBNN algorithm, perform UAV swarm path planning, and introduce multi-step path prediction, direction continuity optimization and adaptive real-time control mechanism to generate candidate paths.

[0030] To address complex environments with restricted and performance degradation zones, this invention presents a drone swarm path planning task. Each drone in the swarm must sequentially reach multiple target points along its path. The overall optimal flight path for the drone swarm is planned using the total path length, total path risk, and total path safety as core optimization indicators. The scenario studied in this invention has a solid application foundation, such as in emergency rescue scenarios involving drone swarms during natural disasters. Collapsed buildings and landslide-prone areas in the disaster zone are restricted zones, while communication blind spots or weak areas are performance degradation zones. Simultaneously, there are locations within the disaster zone that are difficult to reach quickly, requiring the drone swarm to deliver relief supplies to these locations. Furthermore, due to the priority differences between these locations, each drone in the swarm must arrive at its respective target point sequentially according to priority. This step implements the above path planning by improving the GBNN algorithm. The overall algorithm flow is as follows: Figure 1 As shown.

[0031] S41. Initialize algorithm parameters. In this embodiment, four UAVs are configured for swarm path planning. The initial positions of each UAV are (15,30,10), (200,15,15), (77,25,70), and (142,15,125), respectively. The flight speed of each UAV is 0.045 km / s, and the maximum path length is 15 km. Initialize the starting position of the UAV swarm and eight task target points. The target position coordinates are shown in Table 1. Set a real-time constraint threshold of 5 seconds and an early stopping threshold of 5 consecutive steps without a valid path update.

[0032] Table 1 Target Attributes

[0033] An improved formula for calculating the neuron activity value of GBNN was developed, and the global neuron activity value distribution was calculated based on this formula.

[0034] For passable grids (unobstructed terrain and not in restricted areas, i.e., satisfying the condition) =1 and =1) Calculate the neuron activity value using the improved formula: ; In the formula, This represents the activity value of grid neurons. The target incentive coefficient, The performance degradation suppression coefficient, The distance from the current grid to the next target point of the UAV is the Euclidean distance. This represents the degree of influence of the performance degradation region. The formula indicates that the closer the grid is to the next task target point, the higher the neuron activity value, guiding the UAV towards the target. Simultaneously, the greater the influence of the performance degradation region, the higher the corresponding path cost, and the lower the neuron activity value, inhibiting the UAV from entering the high-cost region. In this embodiment, based on the experience of those skilled in the art, Take 1.1, Take 0.8, This indicates the degree of impact of the performance degradation region.

[0035] Based on the aforementioned neuron activity value formula A, a unified calculation is performed on all accessible grates across the entire region to obtain a global neuron activity value distribution covering the entire planning environment. This global activity value distribution is the core basis for subsequent path planning, grate priority ranking, and waypoint decision-making; the higher the grate activity value, the higher the passage priority.

[0036] The height constraint set by S2 must be satisfied simultaneously during the global activity value calculation. Pitch angle constraint Flight range constraints And only retain those that satisfy , Accessible grid cells are included in the calculation to ensure that all initial grid cells are within a safe flying range.

[0037] S42. First, obtain the neuron activity value A corresponding to the current position. Combined with the global neuron activity value distribution, select grids with higher activity values ​​to form a candidate neighborhood subset based on the neuron activity value ranking. Perform multi-step linear prediction for each candidate raster in the subset.

[0038] like Figure 3As shown, 26 3D neighborhood grids are selected centered on the current location. Grids without terrain obstacles, with a comprehensive influence intensity less than a threshold, and meeting the constraints are selected. The six grids closest to the target are then chosen to form a candidate subset. For each candidate grid, multi-step linear prediction is performed at a neighborhood depth of 5 to generate a path segment containing 5 path points. The multi-step path prediction formula is: ; In the formula, For the first One predicted path point, For the next step of selecting grid cells, For neuron indexing, To predict the number of steps, To predict the step size, This is the heading vector. The formula generates multi-step look-ahead path points based on the current heading, avoiding getting trapped in local optima through single-step decision-making. In this embodiment, The prediction step count is set to 5. The prediction step length is set to 20m.

[0039] During the screening process, the performance degradation region grid... The higher the value, the lower the corresponding neuron activity value and the lower the passage priority, while ensuring that each candidate grid satisfies the following conditions. (i.e., within a non-restricted area), the path satisfies altitude constraints, pitch angle constraints, range constraints, and swarm safety distance constraints. ( For any two drones, (Minimum safe distance).

[0040] S43, Targeting candidate neighborhood subsets Based on the multi-step path prediction results of S42, subsequent steps corresponding to each candidate location are generated. Step path point set, detect the intersection grid between different path sets, and select the best path length cost based on the intersection points; Based on the S42 prediction formula, for the next candidate grid , generate subsequent Step path point set For candidate neighborhood subsets Each candidate position within Generate subsequent products using the same method. Step path point set .

[0041] Intersection detection logic and formula: ; In the formula, For path set The first in Path points, For path set The first in Each path point is checked to ensure that the following conditions are met. grid position That is, the set of multi-step paths of different grids. and The formula is used to achieve global intersection detection between different multi-step predicted paths, providing a basis for subsequent optimization based on the path length cost of intersection points.

[0042] Preliminary optimization based on the path length cost at the intersection point: ; In the formula, , Representing candidate positions , To the junction Euclidean distance, This is the next grid position output after the intersection point length cost optimization in this step. The formula prioritizes the intersection point path length cost, selecting shorter candidate paths to achieve preliminary optimization from a global perspective. Both intersection detection and length optimization are performed within the grid range initially optimized by the neuron activity value A. The calculation is performed internally, and all participating grids satisfy the following conditions: , The basic access criteria are that all grid cells are passable and all path points involved meet the constraints of altitude, pitch angle, range, and cluster safety distance.

[0043] S44. When the path length costs are the same, calculate the turning cost and select candidate optimal path segments through the direction continuity optimization strategy. Steering cost formula: ; In the formula, The turning angle at the turning point of the path satisfies , This is the maximum turning angle.

[0044] If the intersection path length costs are equal in the initial selection based on the intersection path length cost of S43, then a secondary selection based on the turning cost is required. The secondary selection formula is as follows: ; ; In the formula, , Candidate positions , The corresponding turning cost, This outputs the final updated grid position for the next step after a two-layer optimization process using intersection point length cost and turning cost. The formula indicates that when path length costs are equal, a secondary optimization is performed using turning cost as a secondary indicator to select the path with less heading change, thus improving flight safety and stability. During the optimization process, it is ensured that the turning angle does not exceed the maximum limit, the flight altitude is within a safe range, and the swarm safety distance constraints are met between UAVs. Under the same conditions, priority is given to selecting... Smaller grids that are less affected by performance degradation.

[0045] S45. Use a cubic B-spline curve to smooth the optimal path obtained after the above selection, and eliminate path inflection points. Path smoothing formula: In the formula The three-dimensional coordinates of the path points after smoothing. The total number of path points. For the first Path points, For cubic B-spline basis functions, represents the spline parameters. This formula is used to smooth the path.

[0046] S46. Update the drone cluster position and determine whether the target position has been reached. If not, return to S42 and repeat the process. Target arrival determination formula:

[0047] ; In the formula, The current coordinates of the drone. The coordinates are the target location coordinates. This formula is used to determine whether the UAV has reached the current mission target point. If the coordinates of the current grid are exactly the same as the target grid coordinates, it is determined that the target point has been reached, the entire algorithm process is terminated, and the pathfinding task is completed.

[0048] If the target position is not reached, return to S42 to start a new round of iterative optimization. At the same time, outside the entire iterative optimization cycle, a globally independent adaptive real-time convergence control mechanism is added.

[0049] Adaptive real-time control accuracy formula: ; In the formula For the first in S31 The comprehensive optimization value of the next iteration For the first The comprehensive optimization value of the next iteration This is the accuracy threshold. This formula is used to control the convergence of the algorithm iterations. After each iteration, the multi-objective comprehensive optimization function constructed by S3 is called. A comprehensive evaluation of the current global path scheme is performed. The evaluation is considered complete when the difference between the comprehensive evaluation functions of two consecutive iterations is less than the accuracy threshold. When the path has converged and stabilized and subsequent iterations offer no significant optimization benefits, the entire iteration process can be terminated early, even if the target position has not yet been reached. This eliminates the need to run the remaining iteration loops, reduces ineffective computing power consumption, and improves the overall real-time performance of path planning.

[0050] This adaptive real-time control mechanism is an iterative optimization enhancement mechanism added globally to the algorithm. It is independent of the main process of single path node selection and operates on the entire cycle of S4 overall iterative optimization. Therefore, no separate execution step is set in the flowchart.

[0051] S5. Solve for the optimal flight path based on the multi-objective optimization function.

[0052] The solution to the optimal flight path is verified by simulation, and four indicators are monitored: total path length, risk, safety, and algorithm running time. The verification results are as follows: total path length 1599.20m, path risk 37, path threat level 24, and running time 4.20s. All indicators meet the constraints and are determined to be the optimal path.

[0053] In this embodiment, under the same environment and parameters, the algorithm of this invention (i.e., the improved discrete biologically inspired neural network GBNN algorithm), the entropy-enhanced quantum ripple collaborative optimization algorithm (E²QRSA algorithm), the adaptive elite gray wolf guided porcupine optimization algorithm (AEGWCPO algorithm), and the traditional discrete biologically inspired neural network GBNN algorithm were used to perform the same planning task, and the three core indicators were compared.

[0054] To unify the evaluation metrics, path threat level is introduced to characterize path safety. This ensures that the three metrics—total path length, path risk, and path threat level—all follow a trend of decreasing as much as possible, facilitating comprehensive comparative analysis. The conversion formula between path threat level and path safety is as follows: ; in, For path threat assessment function, This is a path safety evaluation function.

[0055] Test results are as follows Figure 5 As shown, the total path length of the algorithm of this invention is 1519.92m, the path risk is 32, and the path threat level is 9; the total path length of the E²QRSA algorithm is 1836.16m, the path risk is 44, and the path threat level is 34; the total path length of the AEGWCPO algorithm is 1772.01m, the path risk is 41, and the path threat level is 37; and the total path length of the traditional GBNN algorithm is 1622.55m, the path risk is 42, and the path threat level is 31.

[0056] Based on the above test data, the performance differences between the algorithm of this invention and the other three algorithms were quantitatively compared: Compared with the E²QRSA algorithm, the total path length was shortened by 17.22%, the path risk was reduced by 27.27%, and the path threat level was reduced by 73.53%; compared with the AEGWCPO algorithm, the total path length was shortened by 14.23%, the path risk was reduced by 21.95%, and the path threat level was reduced by 75.68%; compared with the traditional GBNN algorithm, the total path length was shortened by 6.33%, the path risk was reduced by 23.81%, and the path threat level was reduced by 70.97%.

[0057] The comparison results clearly show that the improved GBNN algorithm of this invention outperforms other algorithms in the three core indicators of total path length, path risk, and path security. It has the best overall path performance and can better meet the path planning needs of UAV swarms in complex environments.

[0058] Path planning adaptability verification for scaling up cluster

[0059] To verify the robustness and applicability of this invention under expanded cluster size and complex task scenarios, this embodiment selects the aforementioned mainstream advanced algorithms and the traditional GBNN algorithm for large-scale scenario verification. The experiment scale is expanded to 18 UAVs and 60 target points, performing path planning tasks in the same complex environment. By comparing the path planning performance of the algorithm of this invention, the E²QRSA algorithm, the AEGWCPO algorithm, and the traditional GBNN algorithm, the feasibility of the algorithm is evaluated. The test indicators include total path length, path risk, and path threat level. The data results are as follows.

[0060] The total path length of the algorithm in this invention is 5037.16 m, the path risk is 79.70, and the path threat level is 14.45; the total path length of the E²QRSA algorithm is 5465.42 m, the path risk is 143.85, and the path threat level is 61.00; the total path length of the AEGWCPO algorithm is 5225.75 m, the path risk is 153.75, and the path threat level is 54.95; the total path length of the traditional GBNN algorithm is 5801.51 m, the path risk is 270.10, and the path threat level is 87.30. The running times of each algorithm are as follows: the algorithm in this invention is 4.20 s, the E²QRSA algorithm is 4.97 s, the AEGWCPO algorithm is 4.55 s, and the traditional GBNN algorithm is 4.89 s.

[0061] like Figure 6 , 7Figures 8 and 9 show the comparison results of the total path length, path risk, path threat level, and running time of the four algorithms for the first 20 target points. The comparison shows that the algorithm of this invention has a shorter total path length, lower risk, lower threat level, and better real-time performance. It can effectively shorten the path, avoid obstacles, and reduce the number of turns. Even when the scale of the UAV swarm increases, it can still significantly improve the path planning quality and has good swarm adaptability.

[0062] The comparison shows that the planning method of this invention can better meet the collaborative flight requirements of large-scale UAV swarms, better balance the comprehensive optimization goals of path length, risk and safety, and the generated path can effectively avoid collisions within the swarm. Moreover, it maintains optimal real-time performance in large-scale scenarios, significantly improving the execution efficiency and reliability of the task.

Claims

1. A method for UAV swarm path planning in complex environments based on an improved GBNN algorithm, characterized in that, Includes the following steps: (1) Considering restricted areas and performance degradation areas, quantify the accessibility of the grid and the impact of performance degradation areas to construct a three-dimensional complex environmental threat grid map; (2) Establish single-aircraft flight constraints and collision avoidance constraints within the cluster; (3) Based on the total path length, path risk, and path safety, construct a multi-objective optimization function; (4) Introduce the Euclidean distance from the passable grid to the next task target point and the influence of the performance degradation region, calculate the neuron activity value of the passable grid, and combine the global neuron activity value distribution to screen the neighboring grids to obtain the candidate neighborhood subset. Perform multi-step path prediction for each candidate grid in the candidate neighborhood subset. Based on the multi-step path prediction results, the path length cost is initially selected based on the intersection point. When the path length cost is the same, the direction continuity optimization strategy is introduced to perform a second selection, and the multi-step path prediction results are obtained after selection. After path smoothing and constraint verification, the path planning of the UAV swarm is completed. (5) Solve for the optimal flight path based on the multi-objective optimization function.

2. The method for UAV swarm path planning in complex environments based on the improved GBNN algorithm according to claim 1, characterized in that, Degree of impact of performance degradation region for ； In the formula, The maximum impact intensity in the performance degradation region. This is the performance degradation coefficient. The straight-line distance from the current grid cell to the center of the sphere in the performance degradation region. Let be the radius of the sphere in the performance degradation region.

3. The method for UAV swarm path planning in complex environments based on the improved GBNN algorithm according to claim 1, characterized in that, The multi-objective optimization function is ； ； ； In the formula, For comprehensive optimization, The total path length is used to evaluate the function value. This is the value of the path risk assessment function. The value of the path security evaluation function. , , These are the weighting coefficients; For the number of drone swarms, This represents the total number of waypoints for a single drone. For the first The first drone Path points, path point The number of obstacle grids in the neighborhood; For the first The complete path of the drone For the first Drone route The number of times the course changes.

4. The method for UAV swarm path planning in complex environments based on the improved GBNN algorithm according to claim 2, characterized in that, Grid neuron activity value for ； In the formula, The target incentive coefficient, The performance degradation suppression coefficient, The distance from the current grid to the next target point of the UAV is the Euclidean distance. This indicates the degree of impact of the performance degradation region.

5. The method for UAV swarm path planning in complex environments based on the improved GBNN algorithm according to claim 1, characterized in that, The multi-step path prediction formula is as follows: ； In the formula, For the first One predicted path point, For the next step of selecting grid cells, For neuron indexing, To predict the number of steps, To predict the step size, This is the heading vector.

6. The method for UAV swarm path planning in complex environments based on the improved GBNN algorithm according to claim 5, characterized in that, The steps to determine the intersection point are as follows: Based on the multi-step path prediction formula, for the next candidate grid , generate subsequent Step path point set For each candidate location within the candidate neighborhood subset , generate subsequent Step path point set ; Intersection Grid : ； In the formula, For path set The first in Path points, For path set The first in There are path points.

7. The method for UAV swarm path planning in complex environments based on the improved GBNN algorithm according to claim 6, characterized in that, The next candidate grid after preliminary optimization based on path length cost at the intersection point. for ； In the formula, , These represent the candidate positions. Next step: candidate grid To the intersection grid Euclidean distance, Indicates if intersecting grids exist , Indicates that there are no intersecting grids. The situation.

8. The method for UAV swarm path planning in complex environments based on the improved GBNN algorithm according to claim 7, characterized in that, The second-order selection process is as follows: ； ； In the formula, , These represent the candidate positions. Next step: candidate grid The corresponding turning cost, The turning angle at the turning point of the path; The final updated grid position is output after a two-layer optimization of intersection length cost and turning cost.

9. The method for UAV swarm path planning in complex environments based on the improved GBNN algorithm according to claim 1, characterized in that, Single-aircraft flight constraints include range constraints, altitude constraints, pitch angle constraints, and turn angle constraints.

Images