Unmanned aerial vehicle cluster path planning method based on improved artificial travel mouse optimization algorithm

Abstract

The invention discloses an unmanned aerial vehicle cluster path planning method based on an improved artificial travel mouse optimization algorithm, and the method comprises the steps: obtaining environment data, and building an environment model comprising a threat region; constructing an evaluation function of an unmanned aerial vehicle cluster path; an evaluation function of the unmanned aerial vehicle cluster path is solved based on an improved artificial travel mouse optimization algorithm, and an optimal path planning result is obtained; the improvement on the artificial travel mouse optimization algorithm comprises the following steps: improving an initial point of a random population by adopting chaotic mapping; in iteration, a differential evolution variation strategy is adopted to carry out variation on individuals, and the updating mode of the individuals is controlled by controlling variables. Firstly, a chaotic mapping method is adopted to initialize populations, population diversity and space coverage can be enhanced, and the utilization rate of the whole search space of the algorithm is improved; secondly, a differential evolution variation strategy and a hybrid updating mechanism are introduced, and the jumping ability and the global search performance of the population are enhanced;

Description

Technical Field

[0001] The present invention relates to unmanned aerial vehicle (UAV) cluster safety path planning, and in particular to a UAV cluster path planning method based on an improved artificial lemming optimization algorithm. Background Art

[0002] Drone path planning aims to prioritize drone safety while ensuring compliance with performance and environmental constraints, and to find the most cost-effective route in the shortest possible time. As a critical step in mission execution, safe path planning is crucial in drone swarm applications. In swarm mission scenarios facing complex environments and dynamic obstacles, efficiently planning safe and effective paths for each drone while maintaining the effectiveness of swarm coordinated operations remains a key area of current academic research. The complexity and challenges of such path planning remain significant.

[0003] It's important to emphasize that safe path planning permeates the entire mission cycle, making in-depth research on the path planning problem for drone swarms crucial. In complex and ever-changing battlefield environments, optimizing the overall efficiency of the swarm while ensuring individual safety has long been a research focus. The drone swarm problem is essentially an extension of the single-machine problem. By applying uniform constraints to each drone, the goal is to minimize the overall cost of the swarm.

[0004] Existing evaluation functions for drone path planning primarily consist of four components: fuel consumption constraints, flight altitude constraints, threat avoidance constraints, and time synchronization constraints. These functions generally fail to consider inter-drone collision safety. The path planning problem for drone swarms is essentially a multi-objective optimization problem. The Artificial Lemmings Algorithm (ALA), a bio-inspired optimization algorithm inspired by the long-distance migration, burrowing, foraging, and predator avoidance behaviors of lemmings, demonstrates excellent exploration capabilities in complex, high-dimensional environments. However, in practical applications, the original ALA still suffers from the following issues: First, random initialization leads to uneven population spatial distribution and low search space utilization, limiting exploration capabilities. Second, the traditional update mechanism relies on a fixed strategy, making the switching between exploration and exploitation inflexible. This results in difficulty in rapidly expanding the search range in the early stages of the algorithm and a tendency to suffer from local precision in the later stages. Furthermore, once trapped in a local optimum in multimodal optimization problems, individuals lack effective jumping capabilities, making it difficult to escape the local trap. Summary of the Invention

[0005] Purpose of the invention: In view of the above shortcomings, the present invention provides a UAV cluster path planning method based on an improved artificial lemming optimization algorithm with high exploration capability and high precision.

[0006] Technical solution: To solve the above problems, the present invention adopts a UAV cluster path planning method based on an improved artificial lemming optimization algorithm, which includes the following steps:

[0007] Step 1: Obtain environmental data and build an environmental model including the threat area;

[0008] Step 2: Construct the evaluation function of the UAV cluster path;

[0009] Step 3: Solve the evaluation function of the UAV cluster path based on the improved artificial lemming optimization algorithm to obtain the optimal path planning result; the improvements to the artificial lemming optimization algorithm include: using chaotic mapping to improve the initial point of the random population; using differential evolution mutation strategy to mutate individuals that meet the conditions during iteration, and controlling the update method of individuals that do not meet the conditions by controlling variables.

[0010] Furthermore, the evaluation function of the UAV cluster path is:

[0011] F=w1f o +w2f h +w3f m +w4f t +w5f c

[0012] Among them, F is the evaluation function, f o is the fuel consumption constraint, f h is the flight altitude constraint, f m To avoid the constraint of threat, f t is the time synchronization constraint, f c For collision avoidance constraints, w1, w2, w3, w4, and w5 are weighted coefficients.

[0013] Furthermore, the collision avoidance constraint is expressed as:

[0014] f c =p5·C total

[0015] Among them, p5 is the single collision penalty coefficient, C total is the total number of collisions of all track points in the entire track.

[0016] Furthermore, the specific steps of solving the evaluation function of the drone cluster path based on the improved artificial lemming optimization algorithm include:

[0017] Step 3.1: Population initialization;

[0018] Step 3.2: Fitness evaluation: For each individual in the population, perform fitness evaluation based on the evaluation function of the drone cluster path to obtain the fitness value;

[0019] Step 3.3: Optimal solution update; according to the fitness value, sort the individuals, find the optimal position of the individuals, and determine the global optimal solution of the population in the current iteration;

[0020] Step 3.4: Iterative update: In each iteration, introduce the differential evolution mutation strategy to mutate the individuals that meet the conditions; for the individuals that do not meet the conditions, judge whether it is in the global exploration stage or the local development stage by controlling variables; update the position of each individual according to the judgment result;

[0021] Step 3.5: Judge the number of iterations. If the maximum number of iterations is reached or other termination conditions are met, output the optimal solution, otherwise return to Step 3.2.

[0022] Furthermore, when initializing the population in Step 3.1, use the Logistic chaotic mapping to generate uniformly distributed initial points, and the formula is:

[0023] z n+1 = μz n (1 - z n ), μ = 4, z0 ∈ (0, 1)

[0024] Map the initial point of the chaotic sequence into the solution space:

[0025]

[0026] where, z n represents the chaotic variable obtained in the nth iteration, z0 is the initial chaotic variable within the range of (0, 1), μ represents the control parameter of the chaotic system, z n+1 represents the chaotic variable obtained in the (n + 1)th iteration, represents the initial position of the i-th search agent in the j-th dimension, lb j represents the lower boundary of the search space in the j-th dimension, represents the chaotic variable generated by the i-th individual in the j-th dimension in the chaotic mapping, ub j represents the upper boundary of the search space in the j-th dimension.

[0027] Furthermore, in Step 3.4, introduce the differential evolution mutation strategy to mutate the individuals that meet the conditions. Specifically:

[0028] If the chaotic variable z n of the individual satisfies z n < CR, trigger the differential evolution mutation strategy:

[0029]

[0030] where, is the position vector X of the current individual ii The position vector after differential evolution, r1 and r2≠i are the indices of two different individuals selected by chaotic sequence control, and CR is the crossover probability that adapts with iteration;

[0031]

[0032] Among them, it is the number of iterations and MaxIt is the maximum number of iterations.

[0033] Furthermore, the control variable E is:

[0034]

[0035] Updating the position of each individual based on the judgment results specifically includes:

[0036] When E>1, it is in the global exploration stage, guiding the individual to jump in the global range. The update formula is as follows:

[0037]

[0038] Among them, X * is the global optimal solution of the population in the current iteration, F i represents the directional disturbance factor of individual i, RB i represents the Brownian motion perturbation vector of individual i, α1 and α2 represent uniform random numbers in [0,1], X rand represents the position of an individual randomly selected from the population;

[0039] When E≤1, it is in the local development stage, and the Lévy flight and spiral mechanism are introduced for perturbation update:

[0040]

[0041] Wherein, spiral represents the simulation of the spiral migration behavior of lemmings, Levy(d) is the disturbance vector generated by the Levy distribution, and G represents the contraction control factor in the development stage, which controls the amplitude of the Levy flight disturbance.

[0042] Furthermore, after determining the number of iterations in step 3.6, when the first preset condition is met, a random restart mechanism model is introduced. The random restart mechanism model is expressed as:

[0043]

[0044] Among them, unifrnd() means uniform random sampling between the upper and lower limits of each dimensional variable to generate a new solution vector, VarMin represents the upper limit of the variable, and VarMax represents the lower limit of the variable.

[0045] Furthermore, after determining the number of iterations in step 3.6, when the second preset condition is met, an elite local search mechanism is introduced. The elite local search mechanism is to find the optimal individual X in the current * On each dimension variable j, try to apply a small perturbation ∈:

[0046]

[0047] in, Represents the optimal individual X * The new position obtained by perturbing the j-th dimension of , ∈ represents the local perturbation step constant.

[0048] Furthermore, the first preset condition is: t>0.7·MaxIt and z n <0.1; the second preset condition is: it>0.5·MaxIt.

[0049] Beneficial effects: Compared with the existing technology, the significant advantages of the present invention are that it first uses the Logistic chaos mapping method to initialize the population, which can enhance the population diversity and spatial coverage, and improve the utilization rate of the entire search space of the algorithm; secondly, it introduces the differential evolution mutation strategy and hybrid update mechanism, triggers the differential perturbation operation with adaptive probability during iteration, enhances the population's jumping ability and global search performance, and regulates the search behavior through the linear growth crossover probability, so that the algorithm pays more attention to wide-area exploration in the early stage and more inclined to local development in the later stage, realizing dynamic adjustment of the search process. Further, in order to deal with the problem of population convergence stagnation in the late search stage, a random restart mechanism is introduced to break the population convergence inertia in the late search stage; finally, an elite local search strategy is introduced to break the local optimal trap, activate population diversity, enhance the local accuracy of the solution, and improve the performance of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of the path planning method in the present invention.

[0051] Figure 2 It is a schematic diagram of the flight altitude range in the present invention.

[0052] Figure 3 Schematic diagram of the threat zone in the present invention.

[0053] Figure 4 It is a schematic diagram of collision avoidance in the present invention.

[0054] Figure 5 It is a schematic diagram of the three-dimensional space trajectory of the UAV cluster in the present invention.

[0055] Figure 6 This is a schematic diagram of the comparison of the two-dimensional spatial paths of the UAV cluster in the present invention.

[0056] Figure 7 It is a comparison chart of the optimization algorithm iteration in the present invention. DETAILED DESCRIPTION

[0057] like Figure 1 As shown, in this embodiment, a drone swarm path planning method based on an improved artificial lemming optimization algorithm is provided, including introducing collision avoidance constraints for each drone in the drone swarm, and improving the artificial lemming optimization algorithm. Logistic chaotic mapping population initialization is introduced instead of randomly generating initial positions to enhance population diversity and spatial coverage, thereby improving space utilization; differential evolution mutation strategy and hybrid update mechanism are introduced in iteration to enhance the population's jumping ability and global search performance, and the search behavior is regulated by linearly growing crossover probability, so that the algorithm is more biased towards local development in the later stage; a random restart mechanism is introduced to solve the problem of population convergence stagnation in the late search stage to break the population convergence inertia in the late search stage; and finally, an elite local search strategy is introduced to improve the overall performance of the algorithm.

[0058] 1. UAV track quality evaluation model

[0059] The quality of UAV trajectory planning is directly related to the efficiency and safety of UAV mission execution. In order to accurately evaluate the effect of path planning, a set of path planning evaluation function systems covering multiple aspects is constructed by comprehensively considering flight scenarios and various actual constraints. It consists of five parts: the first part is fuel consumption constraint f o , the second part of the flight height constraint f h , Part III Threat Avoidance Constraints f m , Part IV Time Synchronization Constraint f t , Part V Collision Avoidance Constraint f c .

[0060] Therefore, the total evaluation function F is defined as the weighted sum of the five sub-goal constraints, and the formula is:

[0061] F=w1f o +w2f h +w3f m +w4f t +w5f c

[0062] 1.1 Fuel consumption constraint f o

[0063] In order to ensure that the UAV can complete the scheduled mission and return, and avoid accidents such as UAV crashes due to fuel exhaustion. At the same time, reducing fuel consumption can reduce operating costs and improve the economy and efficiency of mission execution. The fuel consumption constraint is set as:

[0064]

[0065] Among them, L i represents the actual flight distance of the i-th UAV; is the sum of the maximum allowed ranges of all drones; p1 is the normalization coefficient, and its value is set to 1.

[0066] 1.2 Flight altitude constraint f h

[0067] Flying a drone too low may cause it to be affected by ground obstacles, increasing the risk of collision, and may also be affected by complex terrain and weather conditions; flying too high may exceed the drone's performance range, increase fuel consumption, and may be restricted by air traffic control. Figure 2 As shown, set the height constraint as:

[0068]

[0069] Among them, z i,k H represents the height of the kth waypoint of the i-th UAV; max,i ,H min,i The maximum and minimum values of the altitude range allowed for UAV i; p 21 、p 22 It is the penalty coefficient for exceeding the limit, and its value is set to 1.

[0070] 1.3 Threat Avoidance Constraints m

[0071] In order to ensure the safety of drones when performing missions, they must stay away from various threat sources, such as Figure 3 For radar threats, the fourth power inverse strategy is used, as shown below:

[0072]

[0073] This way, when the drone is close to the radar threat, the penalty value will increase dramatically, forcing the drone to maintain a safe distance;

[0074] For other threats (such as missiles, weather, etc.), a linear inverse strategy is used, as shown below:

[0075]

[0076] The penalty value decreases as the distance increases. This method can effectively guide the drone to avoid different types of threats.

[0077] Therefore, the threat avoidance constraint is set as:

[0078]

[0079] Among them, Pi,k represents the k-th waypoint coordinate of UAV i; O rader,m ,O othre,m Represents the location of radar and other threats; p 31 、p 32 are the threat type sensitivity coefficients, and their values are set to 1.2 and 1.1 respectively.

[0080] 1.4 Time synchronization constraints f t

[0081] In a multi-UAV collaborative mission, each UAV needs to reach the designated location within the specified time to achieve mission synchronization. c If the time difference exceeds the adjustable range of UAV i, the absolute value of the time difference is penalized to ensure that each UAV can reach the target location on time and achieve task coordination. Therefore, the following settings are set:

[0082]

[0083] Among them, t i =L i / v i is the actual flight time of UAV i; t c represents the global coordination time; δ i =[L i / v max,i ,L i / v min,i ] is the acceptable time window based on speed constraint; p4 is the time deviation penalty coefficient, and its value is set to 1.2.

[0084] 1.5 Collision Avoidance Constraint f c

[0085] like Figure 4 As shown in Figure 2, when a swarm of drones is flying, the relative speed and flight trajectory between the drones may lead to collision accidents. d and R si There are two fixed circular areas, namely the collision detection circle and safe flight circle of the i-th UAV. Therefore, setting collision avoidance constraints is to ensure the safety of the UAV swarm flight, avoid losses caused by collisions, and ensure the smooth progress of the mission. Set:

[0086]

[0087] Among them, C i Indicates that the i-th UAV has collided, d i-j Represents the distance between the i-th UAV and the j-th UAV. If the distance is less than the minimum safe flight distance d s, then record one collision, count the collisions of all track points of the entire track in turn, and finally sum to get C total .

[0088] The collision avoidance constraints are thus:

[0089] f c =p5·C total

[0090] p5 is the single collision penalty coefficient, and its value is set to 1.

[0091] 2. Improve the fitness function solution algorithm

[0092] When planning the path of an unmanned swarm, the choice of a fitness function is crucial. This embodiment uses the optimal overall evaluation function for the swarm when navigating a complex environment with multiple threats as the fitness function. The improved artificial lemmings algorithm (IALA) is then used to solve this fitness function.

[0093] The Artificial Lemming Algorithm (ALA), as a biologically inspired optimization algorithm inspired by the long-distance migration, burrowing, foraging and predator avoidance behaviors of lemmings, has good exploration capabilities in high-dimensional complex environments. This embodiment makes the following improvements to the original Artificial Lemming Algorithm (ALA). First, a Logistic chaotic map is used to generate the initial population to enhance population diversity and spatial coverage; second, a differential evolution mutation strategy and a hybrid update mechanism are introduced to enhance the population's jumping ability and global search performance, and the search behavior is regulated by a linearly growing crossover probability to achieve dynamic regulation of the search process; further, a random restart mechanism is introduced to address the problem of population convergence stagnation in the late search stage to break the population convergence inertia in the late search stage; finally, an elite local search strategy is introduced to enhance the local accuracy of the solution. Through the above four improvement strategies, the Artificial Lemming Algorithm is improved.

[0094] The first step is population initialization, which is done using the improved artificial lemming algorithm (IALA).

[0095] Traditional randomly generated populations are unevenly distributed across the solution space, being overly dense in some areas and sparse in others. This not only reduces the utilization of the solution space but also reduces population diversity, making it easy for the algorithm to get stuck in local optima early on. To address this issue, a chaotic initialization strategy was introduced. Chaotic initialization exploits the properties of chaotic mapping—both randomness and uniform distribution—to generate an initial population that is more evenly distributed across the search space.

[0096] Logistic chaotic mapping is used to generate uniformly distributed initial points using the formula:

[0097] zn+1 =μz n (1-z n ),μ=4,z0∈(0,1)

[0098] Map the initial point of the chaotic sequence to the solution space:

[0099]

[0100] where z n represents the chaotic variable obtained by the nth iteration, z0 is the initial chaotic variable in the range of (0,1), μ represents the control parameter of the chaotic system, z n+1 represents the chaotic variable obtained in the n+1th iteration, represents the initial position of the i-th search agent in the j-th dimension, lb j represents the lower boundary of the search space in the jth dimension, Represents the chaotic variable generated by the i-th individual in the chaotic map in the j-th dimension, ub j Represents the upper boundary of the search space in the jth dimension.

[0101] This approach ensures the uniformity of the initial population in the multidimensional space, improving the utilization of the learning space and the initial search capability.

[0102] The second step is fitness evaluation. For each individual in the population, the fitness is evaluated based on the comprehensive evaluation function of the unmanned cluster path.

[0103] The third step is to update the optimal solution. According to the fitness value, the individuals are sorted, the optimal position of the individuals is found, and the global optimal solution X of the population in the current iteration is determined. * =arg minF(i).

[0104] Step 4: Differential evolution mutation and hybrid update mechanism

[0105] To enhance the original artificial lemming algorithm's ability to escape and search flexibility in complex, high-dimensional problems, and to address its vulnerability to local extrema and lack of diversity in population updates during the search process, a differential evolution strategy was introduced into the algorithm framework. An adaptive crossover probability control mechanism was designed based on the number of iterations, effectively integrating global search driven by differential mutation with local development and updates driven by the original lemming behavior. The core idea of this strategy is to ensure that individuals have a certain probability of prioritizing differential jumps in each iteration, thereby significantly improving the population's ability to escape local optimal solutions and global search efficiency while maintaining the stability of the original algorithm structure.

[0106] The differential evolution (DE) strategy is introduced, and the crossover probability CR is set to adapt with iteration:

[0107]

[0108] In each iteration, if the chaotic variable z of a certain individual n satisfies z n <CR, then the differential mutation strategy is triggered:

[0109]

[0110] where r1, r2 ≠ i are two different individual indices selected by controlling the chaotic sequence, X i is the position vector of the current individual i, is the position vector after differential mutation.

[0111] If this condition is not met, then it reverts to the migration update strategy in the original ALA. This strategy is based on the exploration and exploitation mechanism of "lemming migration behavior", and in each iteration, it determines whether it is in the global exploration stage or the local exploitation stage by controlling the variable E:

[0112]

[0113] When E > 1 (exploration stage), it guides the individual to jump in the global range, and the update formula is as follows:

[0114]

[0115] where, X * is the global optimal solution of the population in the current iteration, F i represents the direction perturbation factor of individual i, RB i represents the Brownian motion perturbation vector of individual i, α1 and α2 represent uniform random numbers within [0, 1], X rand represents the position of a randomly selected individual from the population;

[0116] When E ≤ 1 (exploitation stage), the Lévy flight and spiral mechanism are introduced for perturbation update:

[0117]

[0118] where, spiral = ‖X * - X i ‖·[sin(2πr3) + cos(2πr3)] represents simulating the vortex-like migration behavior of lemmings, Levy(d) is the perturbation vector generated by the Lévy distribution, and G represents the contraction control factor in the exploitation stage, which controls the perturbation amplitude of the Lévy flight.

[0119] Step 5, Random restart mechanism

[0120] In the later iterations of the original artificial lemming algorithm, although there may still be room for optimization for the globally optimal individual, due to the lack of a disturbance mechanism in the population as a whole, other individuals continue to gather towards it, resulting in the localization of the search space. The algorithm falls into a state of "premature convergence", stagnation, and even "misjudgment of the optimality."

[0121] To effectively address this issue, a random restart mechanism is introduced as a means of "escaping the local trap," specifically designed to break the population's convergence inertia in the late stages of the search. The core idea is that when the algorithm enters the late stages (i.e., when the number of iterations it exceeds 70% of the maximum number of iterations MaxIt), the current individual position vector after differential mutation or the position vector updated by the migration update strategy in the original ALA is reinitialized with a certain small probability, randomly redistributing its position across the entire search space, thereby adding new potential search directions.

[0122] The specific operation forms are as follows:

[0123]

[0124] Among them, unifrnd() means to perform uniform random sampling between the upper and lower limits of each dimension variable to generate a new solution vector. The triggering conditions for this operation are: when it>0.7·MaxIt and z n <0.1.

[0125] Step 6: Elite Individual Local Search Mechanism

[0126] Based on the original artificial lemming algorithm (ALA), an elite local search mechanism is introduced, which targets the current optimal individual X * , actively perform perturbation search on the algorithm when it enters the middle and late stages (i.e., it>0.5·MaxIt), thereby further improving the local accuracy and stability of the final solution.

[0127] At the current optimal individual X * Try to apply a small perturbation to each dimension variable j:

[0128]

[0129] in, Represents the optimal individual X * The new position obtained by perturbing the j-th dimension of , ∈ represents the local perturbation step constant, which is set to 0.01.

[0130] The new solution after the disturbance will be re-evaluated through the fitness function. If its fitness is better than the original optimal solution, it will be updated as a new elite individual.

[0131] Step 7: Termination Condition and Output

[0132] The entire algorithm repeats the updating process from step 3 to step 6 until the maximum number of iterations is reached or the termination condition is met.

[0133] 3. Experimental simulation analysis

[0134] This example simulates drone path planning in the MATLAB R2024b environment and verifies the effectiveness of the proposed algorithm. The experiment uses a swarm of four drones, and an optimization algorithm is used to simultaneously plan the paths of the four drones. The drone evaluation function consists of five components: fuel consumption constraints, flight altitude constraints, threat avoidance constraints, time synchronization constraints, and collision avoidance constraints.

[0135] A randomized experimental environment model is constructed, and threats of radius r are established within these scenarios. The number and location of threats (represented by cylinders) are selected based on the complexity of the environment. Two algorithms are used to plot the flight paths of the drone swarm within the environment. The starting coordinates of the drone swarm are UAV1 (0, 0, 10), UAV2 (0, 100, 10), UAV3 (300, 0, 10), and UAV4 (0, 300, 10), and the target coordinates are UAV1 (875, 875, 10), UAV2 (800, 875, 10), UAV3 (875, 800, 10), and UAV4 (800, 875, 10). Path planning for the drone swarm requires flying from the starting coordinates to the target coordinates, bypassing the threat area. Taking into account the randomness of the heuristic algorithm, the population size of the artificial lemming algorithm and the improved artificial lemming optimization algorithm are both set to 30, and the maximum number of iterations is set to 200. The performance of the two algorithms is evaluated and compared through experimental results.

[0136] After experimental testing, the results show the three-dimensional spatial schematic diagram, two-dimensional spatial path comparison diagram, and optimization algorithm iteration schematic diagram of the optimal drone cluster generated by the two algorithms in 200 iterations. Figure 5 、 Figure 6 、 Figure 7 The results show that the drone paths generated by the two optimization algorithms can both achieve the requirement of bypassing threats, and each path meets the expected requirements in terms of length, altitude, turning angle, climb and dive angle, ultimately showing that the overall effect of the IALA algorithm is better.

[0137] according to Figure 5 By analyzing the performance of drone path planning using the two optimization algorithms in a 3D environment, the following differences can be observed:

[0138] 1. Path length and complexity

[0139] ALA: The planned path is relatively tortuous. Tracks 2 and 4, in particular, contain numerous turns and detours. This complex path design may increase the drone's energy consumption and extend mission execution time during actual flight, reflecting the algorithm's shortcomings in path simplification.

[0140] IALA: All path designs are highly intuitive and concise. Each path has relatively few turns, maintaining a clear and direct overall path, and minimizing unnecessary detours, demonstrating the algorithm's strength in path optimization. In particular, the IALA algorithm effectively avoids complex turns and maintains smooth paths in the path planning for Track 2 and Track 4.

[0141] 2. Obstacle avoidance capability

[0142] ALA: When encountering obstacles, the path changes in various ways. The ALA algorithm plans a complex obstacle avoidance path near the obstacle, increasing the path's tortuosity to circumvent it. While this design effectively avoids collisions, the relatively complex path may affect flight efficiency. Furthermore, in areas with dense obstacles, the planned path may appear congested, potentially posing a risk.

[0143] IALA demonstrates excellent obstacle avoidance capabilities, with simple and effective path planning. When faced with obstacles, the IALA algorithm directly circumvents them, resulting in clear and concise paths. As can be seen in the images, each path quickly returns to a straight line after making appropriate adjustments near obstacles, effectively controlling the overall path length.

[0144] 3. Overall path planning effect

[0145] ALA: The path planning is complex and lacks overall intuitiveness. The images show a chaotic distribution of multiple paths in three-dimensional space, particularly in areas with dense obstacles, where paths intersect and detour. This complex path layout may lead to more navigation challenges in actual flight and increase the difficulty of flight control.

[0146] IALA: Path planning is effective, providing direct and efficient path design. Observing the image, we can see that the paths are arranged in an orderly fashion in three-dimensional space, avoiding unnecessary intersections and detours. The IALA algorithm optimizes the connection between path start and end points, reducing the complexity of intermediate links and making the entire path network clearer and easier to understand. This planning effect helps improve drone flight efficiency and mission success rates.

[0147] The above comparison demonstrates that the IALA algorithm outperforms the ALA algorithm in terms of intuitive path planning, simplicity, and obstacle avoidance capabilities. In this 3D environment, the IALA algorithm not only effectively avoids obstacles but also plans a shorter, more direct flight path, significantly reducing flight energy consumption and flight time. Therefore, the IALA algorithm demonstrates superior overall performance. Its advantages in path optimization and obstacle avoidance efficiency make it more practical and reliable in real-world applications.

[0148] Figure 6 The trajectory diagrams of the two algorithms for path planning in a two-dimensional plane are shown. By comparing the two algorithms, we can find that:

[0149] The ALA paths are quite volatile, with several paths clearly bypassing large areas of the threat zone. This results in large differences in the paths of the various UAVs and a longer overall path, increasing fuel consumption costs and the risk of time losses.

[0150] As can be seen from the figure, the overall performance of the IALA algorithm is better than that of the ALA algorithm. It can effectively avoid threat areas while avoiding obstacles in a smoother and more direct path. Compared with the ALA algorithm, it is more efficient and stable.

[0151] Figure 7 The following are the cost function images of the two algorithms after 200 iterations. We can clearly see that the performance of the two algorithms is:

[0152] 1. Objective function value

[0153] ALA algorithm: The objective function value is relatively high during the iteration process and decreases slowly. The objective function value at the final stabilization is also higher than that of the IALA algorithm, indicating that its optimization effect is inferior to that of the IALA algorithm.

[0154] 2. Convergence speed

[0155] IALA algorithm: The objective function value fluctuates greatly in the early stage of iteration, but quickly stabilizes and eventually stabilizes at a lower objective function value, indicating that it can more effectively optimize the objective function and find a better solution.

[0156] ALA algorithm: The convergence speed is relatively slow. The objective function value does not decrease significantly in the early stage of iteration. It requires more iterations to gradually stabilize, indicating that it approaches the optimal solution relatively slowly during the search process.

[0157] IALA algorithm: The objective function value is quickly reduced when the number of iterations is small, and it basically stabilizes after about 50 iterations, indicating that it converges quickly and can quickly find a more ideal solution.

[0158] 3. Stability

[0159] ALA algorithm: The objective function value fluctuates significantly during the entire iteration process, especially in the later stages of the iteration. This indicates that its stability is inferior to that of the IALA algorithm and there may be some uncertainty.

[0160] IALA algorithm: The objective function value fluctuates slightly during the iteration process, and remains basically stable once it stabilizes, indicating that it has good stability and exhibits reliable performance during the optimization process.

Claims

1. A UAV cluster path planning method based on an improved artificial lemming optimization algorithm, characterized in that: The following steps are involved: Step 1: Obtain environmental data and build an environmental model including the threat area; Step 2: Construct the evaluation function of the UAV cluster path; Step 3: Solve the evaluation function of the UAV cluster path based on the improved artificial lemming optimization algorithm to obtain the optimal path planning result; the improvement of the artificial lemming optimization algorithm includes: using chaos mapping to improve the initial point of the random population; During the iteration, the differential evolution mutation strategy is used to mutate the individuals that meet the conditions, and the update method of the individuals that do not meet the conditions is controlled by controlling variables.

2. The UAV cluster path planning method according to claim 1, characterized in that: The evaluation function of the UAV cluster path is: F=w1f o +w2f h +w3f m +w4f t +w5f c Among them, F is the evaluation function, f o is the fuel consumption constraint, f h is the flight altitude constraint, f m To avoid the constraint of threat, f t is the time synchronization constraint, f c For collision avoidance constraints, w1, w2, w3, w4, and w5 are weighted coefficients.

3. The UAV cluster path planning method according to claim 2, characterized in that: The collision avoidance constraint is expressed as: F c =p5·C total Among them, p5 is the single collision penalty coefficient, C total is the total number of collisions of all track points in the entire track.

4. The UAV cluster path planning method according to claim 3, characterized in that: The specific steps of solving the evaluation function of the drone cluster path based on the improved artificial lemming optimization algorithm include: Step 3.1: Population initialization; Step 3.2: Fitness evaluation: For each individual in the population, perform fitness evaluation based on the evaluation function of the drone cluster path to obtain the fitness value; Step 3.3: Update the optimal solution; sort the individuals according to their fitness values, find the optimal position of the individuals and determine the global optimal solution of the population in the current iteration; Step 3.4: Iterative update: In each iteration, introduce the differential evolution mutation strategy to mutate individuals that meet the conditions; for individuals that do not meet the conditions, use control variables to determine whether they are currently in the global exploration stage or the local development stage; and update the position of each individual based on the judgment result; Step 3.5: Determine the number of iterations. If the maximum number of iterations is reached or other termination conditions are met, output the optimal solution. Otherwise, return to step 3.

2.

5. The UAV cluster path planning method according to claim 3, characterized in that: When the population is initialized in step 3.1, the Logistic chaotic map is used to generate uniformly distributed initial points. The formula is: z n+1 =μz n (1-z n ),z0∈(0,1) Map the initial points generated by the chaotic map into the solution space: Among them, z n represents the chaotic variable obtained by the nth iteration, z0 is the initial chaotic variable in the range of (0,1), μ represents the control parameter of the chaotic system, z n+1 represents the chaotic variable obtained in the n+1th iteration, represents the initial position of the i-th search agent in the j-th dimension, lb j represents the lower boundary of the search space in the jth dimension, Represents the chaotic variable generated by the i-th individual in the chaotic map in the j-th dimension, ub j Represents the upper boundary of the search space in the jth dimension.

6. The UAV cluster path planning method according to claim 5, characterized in that: In step 3.4, the differential evolution mutation strategy is introduced to mutate the individuals that meet the conditions, specifically: If the chaotic variable z of an individual n satisfies z n <CR, then trigger the differential evolution mutation strategy: in, is the position vector X of the current individual i i The position vector after differential evolution, r1 and r2≠i are the indices of two different individuals selected by chaotic sequence control, and CR is the crossover probability that adapts with iteration; Among them, it is the number of iterations and Maxlt is the maximum number of iterations.

7. The UAV cluster path planning method according to claim 6, characterized in that: The control variable E is: Updating the position of each individual based on the judgment results specifically includes: When E>1, it is in the global exploration stage, guiding the individual to jump in the global range. The update formula is as follows: Among them, X * is the global optimal solution of the population in the current iteration, F i Represents the directional disturbance factor of individual i, RB i represents the Brownian motion perturbation vector of individual i, α1 and α2 represent uniform random numbers in [0,1], X rand represents the position of an individual randomly selected from the population; When E≤1, it is in the local development stage, and the Lévy flight and spiral mechanism are introduced for perturbation update: Wherein, spiral represents the simulation of the spiral migration behavior of lemmings, Levy(d) is the disturbance vector generated by the Levy distribution, and G represents the contraction control factor in the development stage, which controls the amplitude of the Levy flight disturbance.

8. The UAV cluster path planning method according to claim 7, characterized in that: After determining the number of iterations in step 3.6, when the first preset condition is met, a random restart mechanism model is introduced. The random restart mechanism model is expressed as: Among them, unifrnd() means uniform random sampling between the upper and lower limits of each dimensional variable to generate a new solution vector, VarMin represents the upper limit of the variable, and VarMax represents the lower limit of the variable.

9. The UAV cluster path planning method according to claim 8, characterized in that: After determining the number of iterations in step 3.6, when the second preset condition is met, the elite local search mechanism is introduced. The elite local search mechanism is to find the optimal individual X in the current * On each dimension variable j, try to apply a small perturbation ∈: in, Represents the optimal individual X * The new position obtained by perturbing the j-th dimension of , ∈ represents the local perturbation step constant.

10. The UAV cluster path planning method according to claim 9, characterized in that: The first preset condition is: t>0.7·MaxIt and z n <0.1; the second preset condition is: it>0.5·MaxIt.

Images