A multi-unmanned aerial vehicle hybrid intelligent path planning method for real urban airspace

Abstract

The application discloses a multi-unmanned aerial vehicle hybrid intelligent path planning method for real city airspace, relates to the technical field of path planning, and is characterized in that the method comprises the following steps: S1, three-dimensional environment modeling, constructing a city low-altitude unmanned aerial vehicle scene; S2, unmanned aerial vehicle flight model modeling, the multi-unmanned aerial vehicle path planning problem involves multi-objective and multi-constraint optimization, a cost function model of comprehensive evaluation is established in combination with the three-dimensional environment model, and a constraint model of unmanned aerial vehicle operation planning is constructed. The technical problem of the application is to provide a multi-unmanned aerial vehicle hybrid intelligent path planning method for real city airspace, which is suitable for the multi-unmanned aerial vehicle hybrid intelligent path planning method under the conditions of real city three-dimensional space and multi-constraints. Through comprehensive optimization of factors such as ground risk, wind field disturbance, flight efficiency and multi-unmanned aerial vehicle cooperative avoidance, the multi-unmanned aerial vehicle can realize safe, stable and efficient task execution in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and more specifically, to a multi-UAV hybrid intelligent path planning method for real urban airspace. Background Technology

[0002] Currently, in the field of multi-UAV flight path planning technology, existing solutions mainly include path planning methods based on traditional heuristic algorithms and search algorithms based on intelligent optimization algorithms. Traditional heuristic algorithms, such as A*, Dijkstra, and RRT, can generate feasible paths in regularized environments, but they are easily limited by computational complexity and scalability in high-dimensional continuous spaces and large-scale multi-UAV tasks, making them difficult to directly apply to real urban scenarios. To improve planning efficiency and enhance the applicability of algorithms in complex environments, various algorithms based on swarm intelligence and metaheuristics have been proposed in recent years, such as Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Grey Wolf Optimization (GWO), and Genetic Algorithm (GA). These methods can balance global search and local convergence capabilities through iterative optimization. However, when facing complex urban scenarios with multiple constraints and objectives, single algorithms are still prone to premature convergence, insufficient search capabilities, and a lack of adaptive adjustment capabilities to environmental changes. To overcome these shortcomings, some studies have proposed hybrid intelligent optimization schemes, combining PSO, simulated annealing, genetic operators, and local perturbation strategies to improve the stability and global exploration capabilities of the algorithms. However, such solutions are mostly designed based on ideal simulation environments or single constraints, lacking the ability to comprehensively model factors such as building occlusion, multi-UAV conflict avoidance, wind field disturbances, and ground safety risks in real urban three-dimensional spatial environments.

[0003] Therefore, existing technologies still struggle to simultaneously meet the requirements for path planning accuracy, execution efficiency, safety, and environmental adaptability when dealing with engineering-level multi-drone collaborative tasks, and cannot effectively support multi-drone task planning in real urban environments.

[0004] Existing multi-UAV path planning technologies still have significant shortcomings in real-world urban scenarios. On the one hand, many planning methods rely on idealized or rule-based environmental models for design, lacking validation in real, high-resolution urban environments. This makes it difficult for planning results to maintain effectiveness and reliability when facing complex terrain, building structures, and dynamic risk factors. On the other hand, when path planning involves multiple objectives such as task time windows, ground risk control, wind field disturbances, energy consumption constraints, and flight safety, existing technologies are insufficient in multi-objective coordination, and planning performance is prone to instability with environmental changes. At the optimization level, traditional single-intelligent algorithms (such as PSO and ACO) are prone to getting trapped in local optima in high-dimensional, multi-constraint search spaces, heavily reliant on initial parameters and environmental models, making it difficult to guarantee the stability and consistency of planning results. Furthermore, existing solutions primarily focus on single-UAV path planning, lacking sufficient support for trajectory conflict avoidance, behavioral coupling relationships, and task scaling capabilities in multi-UAV collaborative tasks, making it difficult to meet the application requirements of large-scale, multi-UAV collaborative tasks. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a multi-UAV hybrid intelligent path planning method for real urban airspace. This method is applicable to multi-UAV hybrid intelligent path planning under multiple constraints in real urban three-dimensional space. By comprehensively optimizing factors such as ground risk, wind field disturbance, flight efficiency and multi-UAV collaborative avoidance, it enables multi-UAVs to achieve safe, stable and efficient mission execution in complex environments.

[0006] The present invention achieves its objective by employing the following technical solution:

[0007] A multi-UAV hybrid intelligent path planning method for real urban airspace includes the following steps:

[0008] S1: 3D environment modeling to construct urban low-altitude drone scenarios;

[0009] S2: UAV flight model modeling. The multi-UAV path planning problem involves multi-objective and multi-constraint optimization. Combining the three-dimensional environment model, a comprehensive evaluation cost function model is established, and a UAV operation planning constraint model is constructed.

[0010] S3: Multi-UAV path planning; Hybrid intelligent optimization strategy for solving multi-UAV path planning.

[0011] As a further limitation of this technical solution, in S1, high-precision geographic information system data and building distribution data are used to construct a three-dimensional model of the multi-UAV mission execution scenario.

[0012] As a further limitation of this technical solution, in S2, the fitness function JThis is used to measure the feasibility and merits of each candidate route, and is defined as follows:

[0013] (1);

[0014] in: λ k These are the weighting coefficients;

[0015] J k This indicates the cost of flight.

[0016] As a further limitation of this technical solution, S3 adopts a multi-stage hybrid optimization strategy consisting of a region adaptive initialization mechanism, an improved sparrow search algorithm, a multi-operator cooperative evolution strategy, and an adaptive simulated annealing local optimization operator.

[0017] As a further limitation of this technical solution, the specific steps of S3 are as follows:

[0018] S31: Based on the drone's starting point S ( x _ s , y _ s , z _ s ) and endpoint G ( x _ g , y _ g , z _ g Construct a candidate task cube region Ω and introduce a dynamic expansion factor Δ to increase the size of the search space;

[0019] (2);

[0020] Random initialization within the region Candidate path particles:

[0021] (3);

[0022] in: Indicates the first The complete path solution representation of each candidate path particle;

[0023] Calculate the fitness value of each particle along its path and record its historical best position. P best and the global optimal position of the population G best :

[0024] (4);

[0025] S32: The global search stage of the improved sparrow search algorithm based on regional adaptive control;

[0026] The improved sparrow search algorithm is used to update the positions of particles located in the discoverer set within the population. The update formula is as follows:

[0027] (5);

[0028] in: X t For the particle in the first t The current position vector of the generation;

[0029] X t+1 The updated position;

[0030] T This represents the maximum number of iterations.

[0031] ε To prevent tiny constants with a denominator of 0;

[0032] α These are the control parameters for the amplitude of the normal disturbance.

[0033] (0,1) represents a random variable that follows a standard normal distribution with a mean of 0 and a variance of 1;

[0034] The proportion of dynamic discoverers introduced R t Defined as:

[0035] (6);

[0036] in: R min and R max These represent the lower and upper limits of the proportion of dynamic discoverers, respectively.

[0037] Then, update the positions of the particles of the joining character:

[0038] (7)

[0039] in: Q A random perturbation factor that follows a uniform distribution;

[0040] This is the position with the worst fitness in the population;

[0041] N This represents the total number of particles in the current population;

[0042] β To develop step size weights;

[0043] By escaping the local optimum through large displacement:

[0044] (8);

[0045] in: γ The jump intensity coefficient for the vigilant;

[0046] For random disturbance factors;

[0047] S33: Perform a convergence stagnation judgment. When the algorithm meets the convergence stagnation judgment condition, the multi-operator cooperative evolution mechanism is triggered.

[0048] First, the particles are sorted according to their population fitness, and then divided into parent and child generations proportionally. X dad With the mother generation X mum Subsequently, chaotic weight parameters based on chaotic mapping are introduced to generate fusion coefficients with unpredictable and jump-like characteristics:

[0049] (9);

[0050] Finally, the child solutions are generated using a continuously weighted method:

[0051] (10);

[0052] S34: After completing the global search and multi-operator evolution, an adaptive annealing algorithm is used to perform local slight perturbations on some representative candidate solutions;

[0053] First, the particles in the population are sorted from best to worst fitness, and the top R = 0.2M particles are selected as elite particles, where M is the population size. The remaining particles are ordinary particles. For elite particles, their local perturbation probability is defined using an adaptive function:

[0054] (11);

[0055] in; Indicates the first The probability of triggering a local perturbation in the current iteration for each particle;

[0056] This represents the fixed low-perturbation probability value used for ordinary particles or the undisturbed phase;

[0057] Indicates the number of elite particles;

[0058] This represents the ranking index value of the particles in the current iteration after sorting them by fitness from best to worst;

[0059] iter Indicates the number of iterations of the current algorithm;

[0060] When the perturbation triggering condition is met, a new candidate solution is generated using a Gaussian perturbation model:

[0061] (12);

[0062] in: This represents the original position vector of the particle before the disturbance;

[0063] Repeat the above steps until the preset maximum number of iterations is reached to obtain the global optimal solution and draw the optimal path on the 3D map.

[0064] Compared with the prior art, the advantages and positive effects of the present invention are:

[0065] 1. The core innovation of this invention lies in proposing a multi-UAV hybrid intelligent path planning method suitable for real urban airspace environments. Its key protection points are mainly reflected in the overall design of the algorithm framework and solution mechanism.

[0066] 2. This invention dynamically prunes the search space through a region adaptive initialization mechanism, introduces an improved sparrow search algorithm (ISSA) to achieve global optimization that combines global exploration with the ability to escape local optima, and combines a multi-operator cooperative evolution strategy (EMA) based on fitness stagnation detection to effectively restore population diversity. At the same time, an adaptive simulated annealing mechanism is used to fine-tune the elite solution locally, thus forming a continuous optimization system of "global search - evolution enhancement - local optimization".

[0067] 3. Regarding cost modeling, this invention uses a multi-objective fitness function to uniformly represent various real-world environmental constraints such as urban wind field, risk distribution, building obstacles, cooperative collision avoidance, and time windows, making the planning results feasible. Therefore, the scope of protection of this invention includes not only the aforementioned modules themselves, but also their cooperative operation mode and overall optimization framework, which is a key technical solution for achieving safe, efficient, and multi-UAV path planning in complex urban environments. Attached Figure Description

[0068] Figure 1 This is a system structure block diagram of the present invention.

[0069] Figure 2 This invention relates to a multi-UAV mission execution scenario.

[0070] Figure 3The present invention constructs a three-dimensional model of a multi-UAV mission execution scenario based on high-precision GIS data and building distribution data.

[0071] Figure 4 This is a flowchart of the multi-UAV path planning process of the present invention.

[0072] Figure 5 This is a schematic diagram of the experimental results of the method of the present invention.

[0073] Figure 6 This is a schematic diagram of the SSA experimental results.

[0074] Figure 7 This is a schematic diagram of the EMA experiment results. Detailed Implementation

[0075] The following detailed description of a specific embodiment of the present invention is provided in conjunction with the accompanying drawings. However, it should be understood that the scope of protection of the present invention is not limited to the specific embodiment.

[0076] Compared with existing multi-UAV path planning technologies, this invention has significant advantages in adaptability, solution quality, and convergence stability in complex real-world urban environments. Traditional methods often rely on rule-based or simplified scenarios, making it difficult to simultaneously handle multiple constraints such as building obstacles, wind disturbances, third-party risks, multi-UAV collaborative collision avoidance, and mission time windows. This invention, however, constructs a comprehensive risk-environment-dynamics integrated cost model, making the planning results more closely aligned with practical application needs. Regarding the optimization mechanism, compared to traditional algorithms prone to premature convergence or getting trapped in local optima, this invention employs an improved sparrow search algorithm for efficient global exploration, combined with a chaos-driven multi-operator collaborative evolution strategy and an adaptive simulated annealing local optimization mechanism. This ensures the algorithm maintains search activity throughout the process and possesses a stronger ability to escape local optima, thereby significantly improving solution stability and path continuity. Furthermore, this invention maintains good computational efficiency and collaborative planning capabilities even when the scale of multi-UAV operations expands, generating safer, more reasonable, and executable multi-UAV 3D paths. Therefore, it demonstrates superior overall performance compared to existing best-in-class technologies in real-world urban airspace missions.

[0077] This invention includes the following steps;

[0078] S1: 3D environment modeling to construct urban low-altitude drone scenarios;

[0079] First, a typical urban area was selected as the scenario for multi-drone mission execution. The coverage area of ​​this area is approximately 5km × 5km. Figure 2 As shown.

[0080] Subsequently, a 3D urban model was constructed based on high-precision Geographic Information System (GIS) data and building distribution data. The model's matrix size was 500×500, and building heights were visualized and encoded using pseudo-color, ranging from 0 to 37 meters. Figure 3 As shown.

[0081] S2: UAV flight model modeling. The multi-UAV path planning problem involves multi-objective and multi-constraint optimization. Combining a three-dimensional environment model, a comprehensive evaluation cost function model is established, and a UAV operation planning constraint model is constructed.

[0082] In S2, the fitness function J This is used to measure the feasibility and merits of each candidate route, and is defined as follows:

[0083] (1)

[0084] in: λ k This is a weighting coefficient that can be dynamically adjusted according to task requirements;

[0085] J k This indicates the cost of flight.

[0086] J 1~ J 9 represents the flight path cost, comprehensive risk cost, wind disturbance cost, building collision avoidance cost, collision avoidance cost between UAVs, flight altitude constraint cost, climb / descent angle constraint cost, heading angle turning constraint cost, and mission time synchronization cost, respectively.

[0087] S3: Multi-UAV path planning; Hybrid intelligent optimization strategy for solving multi-UAV path planning.

[0088] In S3, a multi-stage hybrid optimization strategy is adopted, consisting of a regional adaptive initialization mechanism, an improved sparrow search algorithm (ISSA), a multi-operator collaborative evolution strategy (EMA), and an adaptive simulated annealing local optimization operator (SA).

[0089] The specific steps of S3 are as follows:

[0090] S31: Based on the drone's starting point S ( x _ s , y _s , z _ s ) and endpoint G ( x _ g , y _ g , z _ g Construct a candidate task cube region Ω and introduce a dynamic expansion factor Δ to increase the size of the search space;

[0091] (2)

[0092] Random initialization within the region Candidate path particles:

[0093] (3);

[0094] in: Indicates the first The complete path solution representation of each candidate path particle;

[0095] Calculate the fitness value of each particle along its path and record its historical best position. P best and the global optimal position of the population G best :

[0096] (4);

[0097] S32: The global search stage of the improved sparrow search algorithm based on regional adaptive control;

[0098] The improved sparrow search algorithm is used to update the positions of particles located in the discoverer set in the population. The update formula is as follows:

[0099] (5);

[0100] in: X t For the particle in the first t The current position vector of the generation;

[0101] X t+1 The updated position;

[0102] T This represents the maximum number of iterations.

[0103] ε To prevent tiny constants with a denominator of 0;

[0104] α These are the control parameters for the amplitude of the normal disturbance.

[0105] (0,1) represents a random variable that follows a standard normal distribution with a mean of 0 and a variance of 1;

[0106] The proportion of dynamic discoverers introduced R t Defined as:

[0107] (6);

[0108] in: R min and R max These represent the lower and upper limits of the proportion of dynamic discoverers, respectively, used to limit the range of discoverer particles participating in the global search;

[0109] Then, update the positions of the particles of the joining character:

[0110] (7);

[0111] in: Q A random perturbation factor that follows a uniform distribution;

[0112] This is the position with the worst fitness in the population;

[0113] N This represents the total number of particles in the current population (i.e., the population size).

[0114] β To develop step size weights;

[0115] When a particle's individual fitness is significantly worse than the population average or close to the worst fitness value, it is considered to be in a dangerous state, triggering the vigilant jump mechanism, which causes it to jump out of the local optimum by making a large displacement:

[0116] (8);

[0117] in: γ The jump intensity coefficient for the vigilant;

[0118] For random disturbance factors;

[0119] S3: Perform a convergence stagnation judgment. When the algorithm meets the convergence stagnation judgment condition, the multi-operator cooperative evolution mechanism is triggered.

[0120] First, the particles are sorted according to their population fitness, and then divided into parent and child generations proportionally. X dad With the mother generation Xmum Subsequently, chaotic weight parameters based on Logistic Chaotic Mapping are introduced to generate fusion coefficients with unpredictable and jump-like characteristics:

[0121] (9);

[0122] Finally, the child solutions are generated using a continuously weighted method:

[0123] (10);

[0124] S4: After completing the global search and multi-operator evolution, in order to further improve the fineness of the solution and enhance the local region development capability, an adaptive annealing algorithm is used to perform local slight perturbations on some representative candidate solutions.

[0125] First, the particles in the population are sorted from best to worst fitness, and the top R = 0.2M particles are selected as elite particles, where M is the population size. The remaining particles are ordinary particles. For elite particles, their local perturbation probability is defined using an adaptive function:

[0126] (11);

[0127] in; Indicates the first The probability of triggering a local perturbation in the current iteration for each particle;

[0128] This represents the fixed low-perturbation probability value used for ordinary particles or the undisturbed phase;

[0129] Indicates the number of elite particles;

[0130] This represents the ranking index value of the particles in the current iteration after sorting them by fitness from best to worst;

[0131] iter represents the current iteration number of the algorithm;

[0132] When the perturbation triggering condition is met, a new candidate solution is generated using a Gaussian perturbation model:

[0133] (12);

[0134] in: This represents the original position vector of the particle before the disturbance;

[0135] Repeat the above steps until the preset maximum number of iterations is reached to obtain the global optimal solution and draw the optimal path on the 3D map.

[0136] The experimental results of comparing this invention with the traditional SSA and EMA algorithms are shown in Table 1. This invention demonstrates the best performance in terms of fitness value, obstacle avoidance capability, path length, risk cost, wind interference suppression, and flight altitude control.

[0137] Table 1 Algorithm Comparison:

[0138] ;

[0139] From the three-dimensional trajectory diagram ( Figures 5 to 7 As can be seen, the path generated by this invention is smoother and the flight selection is more reasonable.

[0140] The above-disclosed embodiments are merely specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A multi-UAV hybrid intelligent path planning method for real urban airspace, characterized in that, Comprise the following steps: S1: three-dimensional environment modeling, build city low-altitude unmanned airport scene; S2: unmanned aerial vehicle flight model modeling, multi-unmanned aerial vehicle path planning problem involves multi-objective, multi-constraint optimization, combined with three-dimensional environment model, the cost function model of comprehensive evaluation is established, and the unmanned aerial vehicle operation planning constraint model is constructed; S3: multi-unmanned aerial vehicle path planning; mixed intelligent optimization strategy is used to solve multi-unmanned aerial vehicle path planning; In the S3, a multi-stage mixed optimization strategy composed of regional adaptive initialization mechanism, improved sparrow search algorithm, multi-operator collaborative evolution strategy and adaptive simulated annealing local optimization operator is adopted; The specific steps of the S3 are: S31: according to the starting point S(x_s, y_s, z_s) and the terminal point G(x_g, y_g, z_g) of the unmanned aerial vehicle, a task candidate cubic region Omega is constructed, and a dynamic expansion factor Delta is introduced to improve the search space scale; （2） Randomly initialize N candidate path particles in the region: （3） wherein: represents the complete path solution of the candidate path particle; Calculate fitness value of each path particle and record its own history optimal position and population global optimal position : （4） S32: the improved sparrow search algorithm based on regional adaptive regulation is used for global search stage; The improved sparrow search algorithm is used to update the position of the particles in the finder set in the population, and the update formula is: （5） wherein: is the current position vector of the particle at the tth generation; updated position; T is the maximum number of iterations; Epsilon is a small constant to prevent the denominator from being zero; Alpha is the normal disturbance amplitude control parameter; N(0, 1) indicates a standard normal distribution random variable with mean 0 and variance 1; introduced dynamic discoverer ratio defined as: （6） wherein: and respectively represent lower and upper limit values of the dynamic discoverer ratio. Then, the position of the particle in the joiner role is updated: （7） Wherein: Q is a random disturbance factor obeying uniform distribution; is the worst fitness position in the population; N indicates the total number of particles in the current population; Beta is the development step weight; A large displacement is used to jump out of the local optimum: （8） wherein: is the alert jumper intensity coefficient; is a random disturbance factor; S33: convergence stagnation judgment is carried out, and when the algorithm meets the convergence stagnation judgment condition, the multi-operator collaborative evolution mechanism is triggered; First, the particles are sorted according to the population fitness and proportionally divided into parents With the parents ; subsequently, a chaotic weight parameter based on chaotic mapping is introduced to generate a fusion coefficient with unpredictability and jumping characteristics: （9） Finally, the offspring solution is generated in a continuous weighted manner: （10） S34: after global search and multi-operator evolution, an adaptive annealing algorithm is used to perform local slight disturbance on part of the representative candidate solutions; Firstly, the population particles are sorted from good to bad according to the fitness, and the first R=0.2M particles are selected as elite particles, wherein M is the population size, and the remaining particles are ordinary particles. For elite particles, the local disturbance probability is defined by an adaptive function: （11） in, Indicates the first The probability of triggering a local perturbation in the current iteration for each particle; represents a fixed low disturbance probability value employed for the ordinary particle or non-disturbed phase; Nel represents the number of elite particles; represents the rank index value of the particle after sorting by fitness from best to worst in the current iteration; represents the number of iterations of the current algorithm; When the disturbance trigger condition is met, a new candidate solution is generated by using a Gaussian disturbance model: （12） wherein: represents the original position vector of the current particle before the disturbance; Repeat the above steps until the number of times reaches the preset maximum number of iterations, obtain the global optimal solution, and draw the optimal path in the three-dimensional map.

2. The multi-UAV hybrid intelligent path planning method for real urban airspace according to claim 1, characterized in that: In the S1, high-precision geographic information system data and building distribution data are used to construct a three-dimensional model of multi-unmanned aerial vehicle task execution scene.

3. The multi-UAV hybrid intelligent path planning method for real urban airspace according to claim 1, characterized in that: In the S2, the fitness function J is used to measure the feasibility and degree of excellence of each candidate track, and is defined as: （1） wherein: is a weight coefficient, denotes the flight cost.

Images