A Multi-UAV Intelligent Target Avoidance Search Method Guided by Prior Probability Information

Abstract

This invention belongs to the field of multi-UAV cooperative planning technology, and specifically to a multi-UAV intelligent target avoidance search method guided by prior probability information. The method includes the following steps: Step 1: Establishing a UAV platform model, sensor model, and target motion model; Step 2: Establishing an environmental information graph model, including a target search probability graph, an environmental uncertainty information graph, and a digital pheromone graph; Step 3: Using a distributed model predictive control (DMPC) framework, combined with particle swarm optimization (PSO) and Dubins curves for path planning. The UAVs dynamically track the target based on its avoidance behavior. Once a target is captured by the UAV, it stops moving until all targets are captured. This invention establishes a motion prediction model and updates the probability map in real time, achieving efficient multi-target search given the known number of targets, initial positions, and motion states. This solves the problems of low efficiency in cooperative search algorithms and resource waste in multi-UAV operations.

Description

Technical Field

[0001] This invention relates to the field of multi-UAV collaborative planning technology, specifically to a multi-UAV intelligent target avoidance search method guided by prior probability information. Background Technology

[0002] With the rapid development of unmanned aerial vehicle (UAV) technology, it is now widely used in search and rescue, agricultural monitoring, and urban planning. Target search is the first step in many missions; only after the target is located can subsequent tracking and engagement operations be carried out. Focusing on multi-UAV target search in dynamic environments, compared to the limited coverage and response speed of a single UAV, multi-UAV collaborative search significantly improves search efficiency and success rate through task allocation, path planning, and communication coordination. Its application is becoming increasingly widespread, especially in adversarial missions such as military reconnaissance and border patrol. In these missions, targets often possess a certain degree of intelligence and adopt strategies to evade reconnaissance, such as changing their movement paths, hiding signals, and interfering with reconnaissance signals, making traditional collaborative search methods difficult to effectively track targets. Therefore, how to respond quickly and adjust search strategies in real time over large areas is a highly challenging task.

[0003] Cooperative search for moving targets in wide-area environments is an extremely challenging task, as the uncertainty of target location directly complicates the search path planning process. Current methods mostly focus on static or general dynamic targets, with limited coverage of intelligently evasive dynamic targets. In such environments, traditional target search methods struggle to adapt to real-time target avoidance behavior, leading to reduced search efficiency. Therefore, this paper proposes a multi-UAV intelligent target avoidance search method guided by prior probability information to address these issues. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a multi-UAV intelligent target avoidance search method guided by prior probability information, which solves the problems mentioned in the background.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention specifically adopts the following technical solution:

[0008] A multi-UAV intelligent target avoidance search method based on prior probability information includes the following steps:

[0009] Step 1: Establish the UAV platform model, sensor model, and target motion model;

[0010] Step 2: Establish an environmental information graph model, including a target search probability graph, an environmental uncertainty information graph, and a digital pheromone graph;

[0011] Step 3: Path planning is performed using a distributed model predictive control (DMPC) framework, combined with particle swarm optimization (PSO) and Dubins curves. The UAV tracks the target dynamically based on its avoidance behavior. Once a target is captured by the UAV, it stops moving until all targets are captured.

[0012] Furthermore, the specific content of establishing the UAV platform model in step 1 is as follows: a small fixed-wing UAV with constant speed and fixed altitude is used for cruise. When planning the search path for the UAV, its flight performance and related constraints need to be fully considered. The UAV flies forward at a fixed speed and is limited by the maximum yaw angle.

[0013] Furthermore, the specific content of establishing the sensor model in step 1 is as follows: During the search process, the UAV uses sensors to detect targets and continuously updates the target existence probability based on the sensor detection information. However, due to the uncertainty of sensor measurements and environmental occlusion, targets may be missed or misjudged. Assuming the UAV sensor detection probability p... d ∈[0,1], representing the probability that the drone also detects the target given its presence; the sensor false alarm probability p f ∈[0,1] represents the probability that the UAV detects the target even when the target does not exist. During the search mission, the UAV considers the sensor detection probability and the false alarm probability, and uses the Bayesian criterion to dynamically update the target existence probability in the mission area based on the detection information of its own sensors.

[0014] Furthermore, the specific content of establishing the target motion model in step 1 is as follows: simplifying the target's structure and motion attitude to a point mass; for dynamic targets with intelligent evasion and reconnaissance characteristics, considering that the actual target being searched has a certain anti-reconnaissance capability, the target can sense the presence of nearby drones. When a drone enters the sensing range, the target will take evasive action. When the target senses the drone, it will adjust its motion direction away from the drone. The target will update its position along the evasion direction. The target motion equation is designed as follows:

[0015] Where: p t It is the position of the target at time t, p t+1 It is the target position at the next moment, v t Δt is the velocity of the target, and Δt is the time step. It is a unit vector representing the evasion direction, calculated using the relative positions of the target and the drone: r = p t -puav =(x t -x uav ,y t -y uav );

[0016] To avoid the drone, the target will move in the opposite direction of its position to evade it. It is the unit vector in the opposite direction of the relative position vector:

[0017] Considering multiple drones, the target avoids obstacles by comprehensively considering the positions of all drones, and then weights the impact of the drone positions to an average: Among them, w i It is the evasion intensity, defined as The closer the drone is, the greater its influence on the evasion direction.

[0018] Furthermore, the specific content of establishing the target search probability map in step 2 is as follows: Before the multi-UAV system searches the mission area, each UAV initializes its own search information map. During the search and tracking of dynamic targets, the following three situations should be considered: I: when the approximate initial position of the target is known; II: when the position, velocity, and direction of the target are all known; III: when the position, velocity, and direction of the target are all known; The target position is updated using a probability-based dynamic model, through the probability density function p. xy The probability of a target being at position (x,y) at time t is described by (t). Considering that the target may move to adjacent grid cells within each time step, this dynamic update is described by the probability change between grid cells Δp(m,l,k,i): Δp(m,l,k,i)=p(m+Δ,l,k)-p(m,l,k); where Δ corresponds to adjacent grid points in different directions, and Δp(m,l,k,i) represents the spatial gradient of the probability of target l's existence in a certain direction i of grid m, calculated by the UAV k. To further correct for the uncertainty during target movement, a scattering intensity coefficient d(l,i) is introduced, resulting in the following update formula:

[0019]

[0020] Furthermore, in step 2, an environmental uncertainty information map is established. As the collaborative search of the region by multiple UAVs deepens, the UAVs' understanding of the task area gradually increases, and the uncertainty decreases; when a grid (x,y) is detected, χ... xy (t+1)=η·χ xy (t); when the grid (x,y) is not detected, χ xy (t)+0.5(1-χ xy(t)); where η is the decay factor of information uncertainty in the environment, η∈[0,1],χ xy (t=0)=1.

[0021] Furthermore, the specific content of establishing the digital pheromone map in step 2 is as follows: a digital pheromone map is used to assist decision control. By constructing the pheromone map, changes within the UAV's detection area are quantified, guiding the UAV to reduce repeated searches of the same area and revisit areas where targets have already been detected; attracting and repelling pheromones in the digital pheromone map are established.

[0022] Furthermore, in step 3, within the DMPC framework, the system transforms the large-scale search planning problem into a series of short-time domain planning problems, establishing a multi-UAV cooperative search rolling planning model. Let x(k+q|k) and u(k+q|k) be the predictions of the system state and control input at time k+q, respectively. For simplicity, let the N-step predicted state at time k be X(k) and the control input be U(k). Then, the search efficiency of the system after N-step prediction is denoted as: Where P is the length of the rolling optimization prediction window, let P = 3, then at time k, the optimization model for solving the optimal task decision of the system can be described as: The overall search performance function J of a multi-UAV system after applying control input in a given state is derived from the regional target discovery gain J during the target search process. T Expected detection benefits of multiple UAVs J F Benefits of multi-drone collaboration J C The cost of drone execution is C1, the cost of drone collision avoidance is C2, and the cost of threat is C3.

[0023] Furthermore, the specific content of trajectory planning combining particle swarm optimization (PSO) and Dubins path in step 3 is as follows: During the DMPC iterative solution process, the system uses the PSO algorithm to optimize the control signal to obtain the control input sequence U(k) of the subsystem; within the control time domain, each UAV calculates its initial heading based on the predicted target position and direction, and while detecting the target and updating environmental information, adjusts the control signal according to the decision results to calculate the new position and heading; in addition, the system optimizes the UAV's trajectory by generating a Dubins curve path, optimizes the UAV's heading angle adjustment by particle swarm optimization, and then generates a Dubins path that satisfies the minimum turning radius constraint; the entire process is executed cyclically in each control cycle to ensure that the UAV can adapt to environmental changes in real time and complete the target search;

[0024] Consider N v There are 10 drones, and the initial position of each drone is... The initial heading angle is Where j = 1, 2, ..., N vThe goal is to plan the control sequence for each UAV within the rolling time domain N to maximize the overall mission benefit while satisfying the minimum turning radius R of the UAV. min Isodynamic constraints;

[0025] PSO is used to optimize the change in heading angle of a UAV and determine its trajectory at each prediction time step. The fitness function J of PSO is... i Combining multiple factors such as target detection, path smoothness, and collision avoidance, PSO searches for the optimal solution by iteratively updating the position and velocity of particles.

[0026] In each iteration, the fitness function J is used. i Evaluate the performance of particle i and update the individual optimal solution p of particle i. i,best and the global optimal solution g best Once the algorithm converges, the globally optimal particle g is used. best The control signals of the drone are decoded.

[0027] (III) Beneficial Effects

[0028] Compared with existing technologies, this invention provides a multi-UAV intelligent target avoidance search method guided by prior probability information, which has the following beneficial effects:

[0029] This invention addresses real-world scenarios where the trajectories of targets to be searched are random and variable, and they possess reconnaissance-avoidance characteristics. It fully utilizes environmental detection information to establish a motion prediction model and updates the probability map in real time. This enables efficient searching of multiple targets given the known number of targets, their initial positions, and their motion states, thus solving the problems of low efficiency in collaborative search algorithms and wasted resources from multiple UAVs. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method of the present invention;

[0031] Figure 2 This is a block diagram of the method system of the present invention;

[0032] Figure 3 This is a diagram showing the path taken using a traditional search method.

[0033] Figure 4 This is a diagram illustrating the search path effect of the present invention.

[0034] Figure 5 This is a comparison chart of the search efficiency between the traditional method and the method proposed in this invention;

[0035] Figure 6 This is a comparison chart of the coverage rates of the traditional method and the method proposed in this invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Example

[0038] like Figure 1-6 As shown, an embodiment of the present invention proposes a multi-UAV intelligent target avoidance search method guided by prior probability information, which includes the following steps:

[0039] Step 1: Establish the UAV platform model, sensor model, and target motion model;

[0040] The specific content of establishing the UAV platform model is as follows: A small fixed-wing UAV is used, cruising at a constant speed and fixed altitude. When planning the search path for the UAV, its flight performance and related constraints need to be fully considered. The UAV flies forward at a fixed speed and is limited by a maximum yaw angle. Let the state vector of the i-th UAV at time k be:

[0041] s i (k)=[P uav (k),ψ i (k)]

[0042] In the formula: P uav (k)=(x i (k),y i (k) represents the position coordinates of the UAV, ψ i (k) represents the yaw angle of the UAV. The discrete system motion model of the i-th UAV is as follows:

[0043]

[0044] |u i |≤ω max

[0045] At any given moment, the UAV can only acquire the target grid position and heading information for the next moment. In actual search operations, the UAV maneuvers continuously, therefore, it is necessary to plan the flight path from the current grid to the target grid online based on the target grid position determined by the decision results at each step.

[0046] In the formula: v is the UAV flight speed; Δt is the time step; u i (k)=Δψ i (t) represents the decision variable of the UAV, namely the change in yaw angle.

[0047] The specific content of establishing the sensor model is as follows: During the search process, the UAV uses sensors to detect targets and continuously updates the target presence probability based on the sensor detection information. However, due to uncertainties in sensor measurements and environmental occlusion, targets may be missed or misjudged. Assume the UAV sensor detection probability is p. d ∈[0,1], representing the probability that the drone also detects the target given its presence; the sensor false alarm probability p f ∈[0,1] represents the probability that the UAV detects a target even though the target does not exist. During the search mission, the UAV dynamically updates the probability p of the target's presence in the mission area based on the detection information from its own sensors. xy (t). Considering the sensor detection probability and the false alarm probability, the Bayesian criterion is used to update the target presence probability, as described below:

[0048]

[0049] The above formula represents the relationship between the probability of a target existing in the grid at the next moment and the probability of detection by the sensor at the previous moment, where: p d =0.8,p f =0.2, b(n)=0 no target found; b(n)=1 target found.

[0050] The specific content of establishing the target motion model is as follows: the structure and motion attitude of the target are simplified to a point mass, and the motion of the UAV in the Z-axis direction is ignored. Let the state vector of the j-th target at time k be:

[0051] s j (k)=[p t (k),φ j (k)]

[0052] In the formula: p t (k)=(x target (k),y target Let (k) be the target's position coordinates and φ(t) be the target's yaw angle. Then, the discrete system motion model of the j-th target is expressed as follows:

[0053]

[0054]

[0055] p t It is the position of the target at time t, p t+1 It is the target location at the next moment. t Δt is the velocity of the target, and Δt is the time step. It is the unit vector of the target's current direction of motion.

[0056] For dynamic targets with intelligent reconnaissance evasion capabilities, considering that the actual target possesses a certain degree of counter-reconnaissance capability, it can sense the presence of nearby drones. When a drone enters this sensing range, the target will take evasive action. When the target senses the drone, it will adjust its direction of motion, moving away from the drone. The target will update its position along the evasion direction, and the new equation of motion becomes:

[0057]

[0058] in: This is the unit vector of the evasion direction. The evasion direction is calculated using the relative positions of the target and the drone.

[0059] r = p t -p uav =(x t -x uav ,y t -y uav )

[0060] To avoid the drone, the target will move in the opposite direction of its position to evade it. It is the unit vector in the opposite direction of the relative position vector:

[0061]

[0062] Considering multiple drones, the target avoids obstacles based on a comprehensive analysis of the positions of all drones. The influence of all drone positions is weighted and averaged:

[0063]

[0064] w i It is the evasion intensity, defined as The closer the drone is, the greater its influence on the evasion direction.

[0065] Step 2: Establish an environmental information graph model, including an environmental uncertainty information graph, a digital pheromone graph, and a target search probability graph;

[0066] The specific steps for establishing the target search probability map are as follows: Before the multi-UAV system searches the mission area, each UAV initializes its own search information map. During the search and tracking of dynamic targets, a probability-based dynamic model is used to update the target position. This is achieved through the following probability density function p. xy Let (t) describe the probability that a target is located at position (x,y) at time t. By fully utilizing the detected environmental information to update the probability map, we can better predict the target distribution, reduce search time, and better solve the target search problem in time-varying environments.

[0067] When the approximate initial position of the target is known, in the search environment, each target exists independently. Since the target's direction of motion and speed are unknown, its position variables (x, y) in the X and Y axes are also unknown. t ,y t They are also independent of each other, and therefore each follows a two-dimensional Gaussian distribution: The probability distribution information in the entire search environment is N. t The summation of the independent target probability distribution values ​​is used to initialize the target search probability map by employing a multi-peaked normal distribution.

[0068] r(i,j) 2 =(x i -x t ) 2 +(y j -y t ) 2

[0069]

[0070] Where r(i,j) represents the distance of the target from position (x,y).

[0071] When the target's position and velocity are known, the probability density function considers the target's expected position after movement, assuming the target moves in a straight line:

[0072]

[0073] Among them, v t t0 is the target's velocity, and t0 is the time the target's motion is delayed, t0 = 10s.

[0074] When the target's position, velocity, and direction are known, the probability density function is further refined, taking into account not only changes in the target's expected position but also orientation error:

[0075]

[0076] Where θ(i,j) refers to the relative angle between any point on the map and the center of the peak of the Gaussian distribution, θ t (i,j) refers to the initial direction of the dynamic target, δ θ denoted as the standard deviation of the angular distribution.

[0077] Within each time step, the target may move to an adjacent grid cell. This dynamic update is described by the probability change Δp(m,l,k,i) between grid cells:

[0078] Δp(m,l,k,i)=p(m+Δ,l,k)-p(m,l,k)

[0079] Where Δ corresponds to adjacent grid points in different directions. Δp(m,l,k,i) represents the spatial gradient of the probability of the existence of target l in a certain direction i of grid m, calculated by UAV k. This update method considers the possible directions and velocities of the target within the time step, ensuring dynamic correction of the target position at different time steps.

[0080] To further correct for the uncertainties during the target's movement, a scattering intensity coefficient d(l,i) is introduced, and the following updated formula is finally obtained:

[0081]

[0082] This formula allows the target's probability to be adjusted in different directions to better reflect its actual motion, ensuring accurate tracking of the target during the search process.

[0083] The specific content of establishing the environmental uncertainty infographic is as follows: In order not to miss any possible targets within the mission area, the UAV's search path should cover as many unknown areas as possible, that is, areas with high uncertainty in the search probability map. As the degree of collaborative search of the area by multiple UAVs deepens, the UAV's understanding of the mission area gradually increases, and the uncertainty decreases.

[0084]

[0085] Where η is the attenuation factor of information uncertainty in the environment, η∈[0,1],χ xy (t=0)=1.

[0086] The specific content of establishing a digital pheromone map is as follows: A digital pheromone map is used to assist decision-making and control. By constructing the pheromone map, changes within the UAV's detection area are quantified. This guides the UAV to reduce repeated searches of the same area and to revisit areas where targets have already been detected, achieving effective collaborative searching. The update methods for attracting and repelling pheromones in establishing the digital pheromone map are represented as follows:

[0087]

[0088] E a G represents the pheromone evaporation coefficient. a D represents the propagation coefficient of the attracted pheromone to surrounding grid cells. a This indicates the amount of pheromones released spontaneously to attract pheromones. E r G represents the pheromone evaporation coefficient. r D represents the repulsive pheromone propagation coefficient. r This represents the amount of repulsive pheromones released spontaneously. Define the pheromone switching matrix K. s for:

[0089]

[0090] Where Δt is the time interval between two probe visits to the grid cell (x,y), δ T The time threshold for follow-up visits is Δt ≤ δ. T k x,y =0 indicates that a UAV has detected this cell in a short period of time and does not need to be revisited; otherwise, the grid cell needs to be revisited. and Let represent the amount of attractive pheromone and repulsive pheromone transmitted from neighboring cells during the search period from t-1 to t, respectively. The propagation range of the pheromone is defined as the 8 neighboring cells N surrounding the current cell. s =8.

[0091]

[0092] Step 3: Path planning is performed using a distributed model predictive control (DMPC) framework, combining particle swarm optimization (PSO) and Dubins curves;

[0093] During the search process, a distributed target search method based on Distributed Model Predictive Control (DMPC) was adopted. To improve the solution efficiency of the multi-UAV cooperative search planning problem, under the DMPC framework, the system transforms the large-scale search planning problem into a series of short-time domain planning problems, establishing a multi-UAV cooperative search rolling planning model. Let x(k+q|k) and u(k+q|k) be the predictions of the system state and control input at time k+q, respectively. For simplicity, let the predicted state at time k N steps be X(k) and the control input be U(k). Then, the search efficiency of the system after N steps of prediction is denoted as:

[0094]

[0095] Where P is the length of the rolling optimization prediction window, let P = 3. Then, at time k, the optimization model for solving the optimal task decision of the system can be described as follows:

[0096]

[0097] Based on the actual scenario requirements and the probability map model, an objective function for UAV search was designed. By solving the objective function, the UAV can make reasonable decisions.

[0098] Regional target discovery benefit J during the target search process T The target detection function refers to the probability that a UAV will detect a target during its flight from the current time t to the next time t+1. This metric is described by the cumulative target detection probability of the UAV swarm, defined as follows:

[0099]

[0100] Expected Detection Benefits of Multiple Unmanned Aerial Vehicles J F Since the uncertainty of the environmental grid decreases with the number of times the UAV visits, in order to guide the UAV to search for areas with high uncertainty that have not been visited for a long time and to reduce the uncertainty of the entire search environment as soon as possible, it is assumed that the range of the area that can be detected within the planning time domain is S. k Expected detection benefit J F Defined as:

[0101]

[0102] By using this method, the UAV path planning system can prioritize areas where targets are likely to be detected but where there is significant uncertainty, thereby maximizing overall search efficiency and effectiveness.

[0103] Multi-UAV Collaboration Benefits J C During collaborative area search, multiple UAVs should strive to ensure safe mission execution, minimize redundant searches of the same area, and guide UAVs to revisit already detected areas. Therefore, the collaborative benefit can be defined as the environmental search state S on the UAV search track. xy The function of (t):

[0104]

[0105] Execution cost C1: To reduce the number of turns the UAV makes during the cooperative search process, effectively reduce energy consumption, and minimize the execution cost of guiding the UAV to complete the cooperative search task, we use the change in the UAV's heading angle to represent the execution cost. The execution cost C1 within the detection area at time t is defined as:

[0106]

[0107] Where θ(t) represents the heading of the UAV at time t, and N represents the control window for rolling optimization, N=3.

[0108] Drone Collision Avoidance C2: To avoid collisions between drones during cooperative search, a virtual repulsive force is introduced to construct a drone-to-drone collision avoidance mechanism. The repulsive force exerted on drone i by other drones at time t is:

[0109]

[0110] Among them: F ij k represents the repulsive force exerted by drone j on drone i. f ω represents the repulsion coefficient. f D represents the repulsion factor.ij E represents the distance between drones. ij φ represents the unit vector indicating the direction of the force exerted by UAV i on UAV j. ij Indicates the drone's i-heading and E-heading ij The angle between them. β represents the position of the drone at time t+1 and the repulsive force F. i The angle between (t) and (t).

[0111] Threat Cost C3: This refers to assessing the risk that a drone might pass through or remain in a threat zone. Each no-fly zone is defined as a hemisphere, and collision avoidance benefits are calculated by checking whether the shortest distance from the current position of each trajectory segment to the last waypoint and the center of the hemisphere is less than the radius.

[0112]

[0113] Where X(t+1|t) represents the trajectory of the UAV from time t to t+1. f Indicates the central assembly point of the no-fly zone, O r This represents the set of radii for no-fly zones.

[0114] In summary, the overall search efficiency of a multi-UAV system in state x(k), after adopting control input u(k), is defined as:

[0115] J k (x(k),u(k))=ω1J T +ω2J F +ω3J C -ω4C1-ω5C2-ω6C3

[0116] in: These are weighting coefficients, with different coefficients indicating the degree of importance placed on the benefits. They should be set appropriately according to the needs of the actual application scenario.

[0117] The trajectory planning method combining Particle Swarm Optimization (PSO) and Dubins path optimization is as follows: During the DMPC iterative solution process, a crucial issue is solving the local optimization of each UAV subsystem to obtain the control input sequence U(k). Considering the unique advantages of PSO in solving nonlinear optimization problems, the system employs PSO for control signal optimization. In the control time domain, each UAV calculates its initial heading based on the predicted target position and direction. Simultaneously, while detecting the target and updating environmental information, the system adjusts the control signals based on the decision results, calculating new positions and headings. Furthermore, the system optimizes the UAV's trajectory by generating a Dubins curve path. The core idea of ​​this method is to optimize the UAV's heading angle adjustment through PSO, thereby generating a Dubins path that satisfies the minimum turning radius constraint. The entire process is executed cyclically in each control cycle, ensuring that the UAV can adapt to environmental changes in real time and complete target search.

[0118] Consider N v There are 10 drones, and the initial position of each drone is... The initial heading angle is Where j = 1, 2, ..., N v The goal is to plan the control sequence for each UAV within the rolling time domain N to maximize the overall mission benefit while satisfying the minimum turning radius R of the UAV. min Isodynamic constraints.

[0119] In this method, PSO is used to optimize the change in the UAV's heading angle, thereby determining its trajectory at each prediction time step. The fitness function J of PSO incorporates multiple factors, such as target detection, path smoothness, and collision avoidance. PSO searches for the optimal solution by iteratively updating the particle's position and velocity.

[0120] v i (t+1)=w(t)·v i (t)+c1r1(p i,best (t)-x i (t))+c2r2(g best -x i (t))

[0121] x i (t+1)=x i (t)+v i (t+1)

[0122] Where, x i (t) represents the solution for the i-th particle in the t-th iteration, v i (t) is the velocity of the particle, p i,best It is the best historical solution for particles, g bestThis is the globally optimal solution. c1 and c2 are learning factors, usually set to 2. r1 and r2 are random numbers between [0-1]. w(t) is the inertia weight, which is dynamically adjusted with the number of iterations.

[0123]

[0124] Among them, w max =0.9 is the initial inertia weight, used to speed up the search in the initial stage. w min =0.4 is the minimum inertia weight, used in later stages to allow the algorithm to converge more accurately to the optimal solution. NC max =50 is the maximum number of iterations.

[0125] In each iteration, the fitness function J is used. i Evaluate the performance of particle i and update the individual optimal solution p of particle i. i,best and the global optimal solution g best Once the algorithm converges, the globally optimal particle g is used. best The control signals of the drone were decoded:

[0126]

[0127] Among them, V j =V max It is a fixed speed.

[0128] The algorithm proposed in this invention was compared with standard heuristic search methods based on information graphs. The essential difference between the two algorithms is that the standard method does not establish a target motion model. Therefore, we designed scenarios with random and variable target trajectories, especially those with reconnaissance evasion characteristics, to verify the effectiveness of the target motion model. Figure 3 This is a diagram showing the path taken using a traditional search method. Figure 4 The diagram shows the path rendering of the search method of this invention. The initial positions of the UAV are represented by solid blue dots T1-T4, and the target's trajectory is represented by red dots. The initial target information is the same in all four algorithms, as shown in Table 1.

[0129] Table 1. Initial state of the target.

[0130]

[0131] In the initial stages, when the target moves slowly, is close to the drone's initial position, or the drone's detection range is sufficient to cover the target's movement distance, motion prediction has little impact on search efficiency. Both algorithms find the first two targets within almost 25 iterations. However, when the target moves quickly or exhibits evasive behavior, models lacking motion prediction struggle to adapt to dynamic changes due to relying solely on the current position, leading to frequent drone repositioning, increased energy consumption and maneuverability, and a decreased capture success rate. In contrast, systems with motion prediction can adjust in advance, maintaining high tracking accuracy. When searching for T3, the comparison... Figure 3 and Figure 4 It can be seen that algorithms incorporating motion prediction allow UAVs to predict the target's upcoming location. Especially in adversarial environments, this enables proactive responses to rapid target movement and turns, quickly "encircling" fast-moving or turning targets. This effectively avoids search delays caused by dynamic target changes and reduces the need to search almost the entire area to capture a target. To further analyze and compare the search capabilities of different algorithms for moving targets, we conducted 30 repeated simulation experiments on different algorithms. Figure 5 This is a comparison chart of the search efficiency between the traditional method and the method proposed in this invention; Figure 6 This chart compares the coverage rates of traditional methods and the method proposed in this invention. In a given search environment, the algorithm without motion prediction completes the search for all targets in 77 iterations, achieving a coverage rate of 83.97%. The algorithm with motion prediction completes the search for all targets in 75 iterations, achieving a coverage rate of 69.61%. Compared to not considering the motion prediction algorithm, the search efficiency is improved by 2.60%, while the coverage rate decreases by 17.09%. This indicates that intelligent target avoidance motion prediction provides UAVs with an efficient and accurate strategy for searching for targets in dynamic environments, covering a smaller area while achieving higher efficiency in capturing all targets.

[0132] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-UAV intelligent target avoidance search method based on prior probability information, characterized in that: Includes the following steps: Step 1: Establish the UAV platform model, sensor model, and target motion model; Step 2: Establish an environmental information graph model, including a target search probability graph, an environmental uncertainty information graph, and a digital pheromone graph; Step 3: Path planning is performed using a distributed model predictive control (DMPC) framework, combined with particle swarm optimization (PSO) and Dubins curves. The UAV tracks the target dynamically based on its avoidance behavior. Once a target is captured by the UAV, it stops moving until all targets are captured. The specific content of establishing the target motion model in step 1 is as follows: The target's structure and motion attitude are simplified to a point mass; for dynamic targets with intelligent evasion and reconnaissance characteristics, considering that the actual target being searched has a certain anti-reconnaissance capability, the target can sense the presence of nearby drones. When a drone enters the sensing range, the target will take evasive action. When the target senses the drone, it will adjust its motion direction away from the drone. The target will update its position along the evasion direction. The target motion equation is designed as follows: Where: p t It is the position of the target at time t, p t+1 It is the target position at the next moment, v t Δt is the velocity of the target, and Δt is the time step. It is a unit vector representing the evasion direction, calculated using the relative positions of the target and the drone: r = p t -p uav =(x t -x uav ,y t -y uav ); To distance itself from the drone, the target will move in the opposite direction of its position to evade it. It is the unit vector in the opposite direction of the relative position vector: Considering multiple drones, the target avoids obstacles by comprehensively considering the positions of all drones, and then weights the impact of the drone positions to an average: Among them, w i It is the evasion intensity, defined as The closer the drone is, the greater its influence on the evasion direction; The specific content of establishing the target search probability map in step 2 is as follows: Before the multi-UAV system searches the mission area, each UAV initializes its own search information map. During the search and tracking of dynamic targets, the following three situations should be considered: I: When the approximate initial position of the target is known; II: When the position and velocity of the target are known; III: When the position, velocity, and direction of the target are all known; The target position is updated using a probability-based dynamic model, through the probability density function p. xy The probability of a target being at position (x,y) at time t is described by (t). Considering that the target may move to adjacent grid cells within each time step, this dynamic update is described by the probability change between grid cells Δp(m,l,k,i): Δp(m,l,k,i)=p(m+Δ,l,k)-p(m,l,k); where Δ corresponds to adjacent grid points in different directions, and Δp(m,l,k,i) represents the spatial gradient of the probability of target l's existence in a certain direction i of grid m, calculated by the UAV k. To further correct for the uncertainty during target movement, a scattering intensity coefficient d(l,i) is introduced, resulting in the following update formula: In step 2, an environmental uncertainty information map is established. As the collaborative search of the region by multiple UAVs deepens, the UAVs' understanding of the mission area gradually increases, and the uncertainty decreases. When a grid (x,y) is detected, χ... xy (t+1)=η·χ xy (t); when the grid (x,y) is not detected, χ xy (t)+0.5(1-χ xy (t)); where η is the decay factor of information uncertainty in the environment, η∈[0,1],χ xy (t=0)=1.

2. The multi-UAV intelligent target avoidance search method based on prior probability information as described in claim 1, characterized in that: The specific content of establishing the UAV platform model in step 1 is as follows: a small fixed-wing UAV with constant speed and fixed altitude is used. When planning the search path for the UAV, its flight performance and related constraints need to be fully considered. The UAV flies forward at a fixed speed and is limited by the maximum yaw angle.

3. The multi-UAV intelligent target avoidance search method based on prior probability information as described in claim 1, characterized in that: The specific content of establishing the sensor model in step 1 is as follows: During the search process, the UAV uses sensors to detect targets and continuously updates the target presence probability based on the sensor detection information. However, due to the uncertainty of sensor measurements and environmental occlusion factors, targets may be missed or misjudged. Assume that the UAV sensor detection probability p d ∈[0,1], representing the probability that the drone also detects the target given its presence; the sensor false alarm probability p f ∈[0,1] represents the probability that the UAV detects the target even when the target does not exist. During the search mission, the UAV considers the sensor detection probability and the false alarm probability, and uses the Bayesian criterion to dynamically update the target existence probability in the mission area based on the detection information of its own sensors.

4. The multi-UAV intelligent target avoidance search method based on prior probability information as described in claim 1, characterized in that: The specific content of establishing the digital pheromone map in step 2 is as follows: a digital pheromone map is used to assist decision control. By constructing the pheromone map, the changes in the UAV detection area are quantified, the UAV is guided to reduce repeated searches of the same area, and the area where the target has been detected is revisited. Construct a digital pheromone map of attractive and repulsive pheromones.

5. The multi-UAV intelligent target avoidance search method based on prior probability information as described in claim 1, characterized in that: In step 3, within the DMPC framework, the system transforms the large-scale search planning problem into a series of short-time domain planning problems, establishing a multi-UAV cooperative search rolling planning model. Let x(k+q|k) and u(k+q|k) be the predictions of the system state and control input at time k+q, respectively. For simplicity, let the predicted state at time k N steps be X(k), and the control input be U(k). Then, the search efficiency of the system after N steps of prediction is denoted as: Where P is the length of the rolling optimization prediction window, let P = 3, then at time k, the optimization model for solving the optimal task decision of the system can be described as: The overall search performance function J of a multi-UAV system after applying control input in a given state is derived from the regional target discovery gain J during the target search process. T Expected detection benefits of multiple UAVs J F Benefits of multi-drone collaboration J C The cost of drone execution is C1, the cost of drone collision avoidance is C2, and the cost of threat is C3.

6. The multi-UAV intelligent target avoidance search method based on prior probability information as described in claim 5, characterized in that: The specific content of trajectory planning combining particle swarm optimization (PSO) and Dubins path in step 3 is as follows: During the DMPC iterative solution process, the system uses the PSO algorithm to optimize the control signal to obtain the control input sequence U(k) of the subsystem; in the control time domain, each UAV calculates its initial heading based on the predicted target position and direction, and while detecting the target and updating environmental information, adjusts the control signal according to the decision results to calculate the new position and heading; in addition, the system optimizes the UAV's trajectory by generating a Dubins curve path, optimizes the UAV's heading angle adjustment by particle swarm optimization, and then generates a Dubins path that meets the minimum turning radius constraint; the entire process is executed cyclically in each control cycle to ensure that the UAV can adapt to environmental changes in real time and complete the target search; Consider N v There are 10 drones, and the initial position of each drone is... The initial heading angle is Where j = 1, 2, ..., N v The goal is to plan the control sequence for each UAV within the rolling time domain N to maximize the overall mission benefit while satisfying the minimum turning radius R of the UAV. min Dynamic constraints; PSO is used to optimize the change in heading angle of a UAV and determine its trajectory at each prediction time step. The fitness function J of PSO is... i Combining multiple factors such as target detection, path smoothness, and collision avoidance, PSO searches for the optimal solution by iteratively updating the position and velocity of particles. In each iteration, the fitness function J is used. i Evaluate the performance of particle i and update the individual optimal solution p of particle i. i,best and the global optimal solution g best Once the algorithm converges, the globally optimal particle g is used. best The control signals of the drone are decoded.

Images