An Online Planning Method for Multi-UAV Cooperative Ground Search under Communication Constrained Conditions

Abstract

The present invention relates to a multi-UAV collaborative ground search online planning method under communication restricted conditions. Aiming at the limitation of intermittent communication, the Voronoi diagram partitioning method is adopted to solve the problem of keeping UAVs within a region during the search process in complex scenarios. A distributed model predictive control framework based on the Voronoi diagram is used to solve the data interaction and control optimization problems of UAVs when communication is intermittently available. According to the characteristics of UAV search tasks, the ant colony algorithm is improved by using the attention mechanism to enhance the search efficiency of the ant colony algorithm during search. At the same time, a search factor is added to strengthen the search effectiveness of UAVs for unsearched regions and accelerate the convergence of the algorithm. The method of the present invention aims at the UAV collaborative ground search scenario under complex conditions and uses the ant colony algorithm based on distributed model predictive control and attention mechanism to solve the UAV collaborative online ground search planning and control problem under multiple constraint conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to an online planning method for multi-unmanned aerial vehicle collaborative ground search under communication constraints. Background Art

[0002] Unmanned aerial vehicles have characteristics such as simple airframe structure, relatively low cost, and long hovering time. With these advantages, unmanned aerial vehicles are gradually playing a huge role in fields such as forest fire prevention, disaster area monitoring, search and rescue in uninhabited areas, and wildlife monitoring, and have broad application prospects. To address the problem of limited detection range of a single unmanned aerial vehicle during large-area detection, a method of using multiple unmanned aerial vehicles to form a search formation through a communication network can break through the performance limitations of a single unmanned aerial vehicle and effectively expand the search range and duration. However, in application scenarios such as mountain search and rescue and forest fire prevention, the task execution area is inaccessible, cellular data base stations are missing, and factors such as tree and mountain occlusion are likely to occur, which may lead to frequent communication restrictions of the unmanned aerial vehicle formation, making it impossible to effectively and timely coordinate the search ranges of each unmanned aerial vehicle in the formation, resulting in problems such as overlapping search areas and collision losses of unmanned aerial vehicle units. How to effectively plan the search area of the unmanned aerial vehicle formation, dynamically adjust according to local terrain, obstacles, threats and other information as well as task requirements, and at the same time automatically guide each unmanned aerial vehicle unit in the formation to conduct effective and continuous search for the planned area is of great significance for improving the monitoring and search efficiency.

[0003] Existing UAV collaborative search methods can be categorized into two main types: offline and online. Offline methods manually or automatically plan the search area and trajectory before a mission and transmit it to the UAV unit performing the mission. The UAV unit then performs search and surveillance tasks within a fixed area or route. These methods are not suitable for complex and changing mission scenarios and lack the flexibility to dynamically adjust to unexpected events. Online methods involve collaborative UAV units planning the mission area and search path online and in real time during the mission, allowing for flexible changes in search strategies based on mission requirements. Based on the collaborative architecture, these methods can be categorized as centralized or distributed. Centralized methods include one or more decision-making units (DMUs). These DMUs calculate a plan based on the status of all collaborating UAVs and direct their flight in real time. If a DMU fails or a UAV is unable to receive its control plan instructions due to interference, each UAV unit in the swarm loses guidance and control, ultimately impairing its ability to execute the mission. Furthermore, since the DMU requires mission planning data from all UAVs in the pen, the DMUs' computational workload is high, placing high demands on data link bandwidth and real-time performance. In a distributed architecture, UAVs exchange real-time information about adjacent or all collaborating UAV units via a data link to independently plan their own search plans. Distributed methods offer greater robustness than centralized approaches, but existing distributed search technologies still rely heavily on inter-machine communication data links. This leads to a series of issues when communication fails, as information from other machines cannot be obtained. While some methods address communication interference, these often rely on data compensation to achieve planning. They fail to consider the challenges of multi-UAV coordinated ground search planning in situations where communication is limited, such as prolonged or even partial UAV outages. This makes it impossible to meet the requirements for multi-UAV coordinated search in complex areas with poor communication reliability, such as mountainous areas. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides an online planning method for collaborative ground search by multiple UAVs under communication-restricted conditions. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0005] The present invention provides an online planning method for collaborative ground search by multiple UAVs under limited trust conditions, comprising:

[0006] Step 1: Establish a search map and initial uncertainty information map for the mission area, and build a UAV motion model and airborne sensor model;

[0007] Step 2: Determine the current Whether communication is available at the moment,

[0008] When communication is available, obtain the position information of all currently communicable drones, divide the area using the Voronoi diagram partitioning method based on the position information of all drones, and update the area partitioning result. According to the updated area partitioning result, each drone executes an ant colony algorithm based on the attention mechanism to obtain the corresponding first control decision sequence;

[0009] When communication is unavailable, according to the area partitioning result of the previous moment, each drone executes an ant colony algorithm based on the attention mechanism to obtain the corresponding second control decision sequence;

[0010] Step 3: When communication is unavailable, control the drones to execute the corresponding second control decision sequence. After execution, update the target presence probability map and the uncertainty information map, and determine whether the updated uncertainty information map meets the preset uncertainty requirement. If it meets, end the search task; if not, return to Step 2 for the next round of search planning until the search task ends;

[0011] When communication is available, each drone obtains the corresponding first control decision sequence of other drones to achieve information sharing. After information sharing, update the target presence probability map and the uncertainty information map of the task area, and determine whether the updated uncertainty information map meets the preset uncertainty requirement. If it meets, end the search task; if not, continue with Step 4;

[0012] Step 4: Calculate the corresponding benefit value for each drone according to the first control decision sequence corresponding to the drone and the current state of the drone, and add it to the benefit value set;

[0013] Step 5: According to the preset prediction step T p , perform a predictive control decision generation operation on each drone to obtain the corresponding predictive control decision sequence for each drone;

[0014] The predictive control decision generation operation includes: when 0 ≤ t < T p , according to the control decision sequence of the drone at the moment, obtain the area partitioning result at the moment. According to this area partitioning result, execute an ant colony algorithm based on the attention mechanism to generate the control decision sequence at the moment, and use it as the predictive control decision sequence of the drone;

[0015] Step 6: Calculate the corresponding new benefit value for each drone according to the predictive control decision sequence corresponding to the drone and the current state of the drone. Compare the new benefit value of each drone with the benefit value in the benefit value set, and update the benefit value set according to the comparison result. Repeat Step 5 - Step 6 until the preset benefit value set update termination condition is reached to obtain the final benefit value set;

[0016] Step 7: Use the decision sequence corresponding to the benefit value in the final benefit value set as the control decision sequence of the corresponding UAV, control the UAV to execute the control decision sequence, and return to step 2 after execution to perform the next round of search planning until the search mission is completed.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] The present invention provides an online planning method for collaborative ground search by multiple UAVs under communication-restricted conditions, which solves the problem of a UAV formation performing a search task under communication-restricted conditions where data exchange within a UAV cluster is only carried out within a limited time window in a complex environment. The problem of poor communication is fully taken into account, and the search area is divided based on the Voronoi diagram within a limited time window, which effectively solves the problem of overlapping search areas and UAV collisions caused by the inability of units within the cluster to exchange data in real time. The AACO algorithm proposed in the present invention adaptively updates the search strategy by introducing a self-attention mechanism for conventional instruction information, prior information, and burst information of the task, so as to meet different search expectations to the greatest extent. The constructed DMPC framework based on the Voronoi diagram division method and the AACO algorithm can effectively solve the ground search problem of UAV formations under communication-restricted conditions.

[0019] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following preferred embodiments are specifically cited and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of a partitioning process based on a Voronoi diagram partitioning method provided by an embodiment of the present invention;

[0021] Figure 2 This is a DMPC-AACO control framework diagram provided by an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of a UAV ground search path based on AACO provided by an embodiment of the present invention;

[0023] Figure 4 This is a post-search environment uncertainty information graph provided by an embodiment of the present invention;

[0024] Figure 5 This is a comparison chart of the effects of the AACO method provided by an embodiment of the present invention and a common method;

[0025] Figure 6It is a schematic diagram of the search trajectory based on the DMPC-AACO method provided by an embodiment of the present invention;

[0026] Figure 7 It is an uncertainty map of the task area environment after searching based on the DMPC-AACO method provided by an embodiment of the present invention. Detailed implementation manners

[0027] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following combines the accompanying drawings and specific implementation manners to detail a multi-UAV collaborative ground search online planning method under communication constraints proposed according to the present invention.

[0028] The foregoing and other technical contents, features and effects of the present invention can be clearly presented in the following detailed description in conjunction with the accompanying drawings. Through the description of the specific implementation manners, a more in-depth and specific understanding of the technical means and effects adopted by the present invention to achieve the predetermined purpose can be obtained. However, the attached drawings are only for reference and illustration, and are not used to limit the technical solution of the present invention.

[0029] Embodiment 1

[0030] A multi-UAV collaborative ground search online planning method provided by this embodiment is aimed at the UAV collaborative ground search scenario under complex conditions. The scenario includes factors such as emergency detection, key area monitoring, coverage search, and intermittent availability of communication. The ant colony algorithm based on distributed model predictive control and attention mechanism is used to solve the problem of UAV collaborative online ground search planning control under multiple constraint conditions.

[0031] The method of this embodiment mainly includes the following three parts: (1) Aiming at the limitations of intermittent communication, the Voronoi diagram partitioning method is adopted to solve the problem that the UAVs stay within a region during the search process in complex scenarios. When communication is available, data exchange is carried out within the UAV swarm, and the search areas of each UAV are re-planned in real time according to the current situation information to prevent the UAV swarm from repeatedly searching the areas that have been searched by other units during the independent search stage. At the same time, the problem of collision loss caused by the inability of the UAV swarm to interact their positions in real time is solved; (2) A distributed model predictive control (DMPC) framework based on the Voronoi diagram is proposed to solve the data interaction and control optimization problems of UAVs when communication is intermittently available; (3) Aiming at the UAV search control planning problem, due to communication limitations, there is no real-time global information available to optimize the solution of the task allocation problem, so there is a high probability that the best global solution cannot be obtained. Considering the limited search range of UAVs and the requirements of real-time processing, an attention mechanism ant colony optimization (AACO) algorithm is proposed. According to the characteristics of the UAV search task, the attention mechanism is used to improve the ant colony algorithm, enhance the search efficiency of the ant colony algorithm during search, and at the same time increase the search factor to strengthen the search effectiveness of UAVs for unsearched areas and accelerate the algorithm convergence.

[0032] First, the Voronoi diagram partitioning method and the attention mechanism ant colony algorithm provided in this implementation are described.

[0033] Please refer to Figure 1 The schematic diagram of the partitioning process of a partitioning method based on the Voronoi diagram provided by the embodiment of the present invention as shown. Based on the Voronoi diagram partitioning method, it specifically includes the following steps:

[0034] (1) Based on the position information of all UAV units at the partitioning moment As Figure 1 shown in subfigure (a) of Figure 1 , construct a Delaunay triangle, as

[0035] (2) Record each scatter point P Tri in the triangle linked list;

[0036] (3) Insert the scatter points in the point set in sequence, traverse the triangle linked list, and find all adjacent triangles T Tri of the triangle whose circumcircle O Tri contains the inserted point;

[0037] (4) Delete the common edges that affect the triangle, connect the insertion point to all the vertices of the affected triangle, and record them in the list of Voronoi edges. The result is shown in figure (c) of Figure 1 ;

[0038] (5) Generate the partition area V(p i ) corresponding to each drone i according to the list of Voronoi edges. The result is shown in figure (d) of Figure 1 ;

[0039] In the collaborative search task, there is often one or more areas with the probability of target existence obtained by other preliminary reconnaissance means in the task area that need to be searched preferentially. When an emergency occurs, it is necessary to search and investigate the location of the emergency. For such problems, this embodiment proposes an attention mechanism. The pheromone matrix is processed through the attention mechanism. The focus points in the attention mechanism are divided into three types for three situations, including: key focus points, emergency focus points, and general focus points. The search is guided by these three types of focus points to improve the search efficiency.

[0040] The ant colony algorithm based on the attention mechanism (Attention Mechanism Ant Colony Optimization, AACO) of this embodiment specifically includes the following steps:

[0041] Step a: For each drone, according to different focus points (i.e., key focus points, emergency focus points, and general focus points), use the attention mechanism to update the target existence probability map and the uncertainty information map;

[0042] Among them, if there are key focus points in the search area of the drone, the uncertainty information map of the surrounding grids (x, y) of the key focus points is updated as follows:

[0043]

[0044] In the formula, (x important , y important ) represents the grid coordinates where the key focus point is located, P prior (x important , y important ) represents the prior existence probability of the target of the grid where the key focus point is located, σ 2 is used to control the probability density diffusion range of the key search area, and N important represents the total number of key focus points;

[0045] If there are emergency focus points in the search area of the drone, the uncertainty information map of the surrounding grids (x, y) of the emergency focus points is updated as follows:

[0046]

[0047] where (x ∈ , y ∈ ) represents the grid coordinates where the sudden focus point is located, and the movement of the sudden focus point target follows a two-dimensional standard normal distribution

[0048] If there are no key focus points and sudden focus points in the search area of the UAV, the uncertainty centroid is used as the ordinary focus point, and the calculation method of the uncertainty centroid is as follows:

[0049]

[0050] Step b: Generate an expected search area based on the focus point, and the calculation formula of the expected search area is as follows:

[0051]

[0052] where (x AttentionPoint , y AttentionPoint ) represents the grid coordinates where the focus point is located, represents the grid coordinates where the UAV is located, and (x0, y0) represents the grid coordinates in the expected search area.

[0053] Step c: Execute the ant colony algorithm in the expected search area, and initialize the parameters of the ant colony algorithm. The parameters include the number of ants M, the number of iterations n, the pheromone heuristic factor α, the expected heuristic factor β, and the search factor detect j ;

[0054] Step d: For ant k, according to the motion model of the UAV, restrict the optional path points for the next moment. The probability P of selecting the path point n at the next moment from the current path point m where the UAV is located mn is calculated as shown in the following formula:

[0055]

[0056]

[0057] where τ mn (t) represents the pheromone matrix from path point m to path point n at the current moment t, η mn (t) represents the heuristic information from path point m to path point n at the current moment t, A n represents the total number of grids that can be detected when the UAV is located at path point n, D n represents the total number of grids that have been detected within the detectable range when the UAV is located at path point n, ∈ represents the coefficient of the search factor; allow represents the allowable flight area of the UAV;

[0058] Step e: According to the airborne sensor model and the selected path points at the next moment, update the uncertainty information map. The change in the uncertainty of the grid with a distance of r around the UAV path is updated by the following formula:

[0059] χ(x,y,t + 1) = (1 - f(r))χ(x,y,t) (7);

[0060] Step f: Repeat Step d - Step e until the number of ant movement steps meets the preset ant step threshold to obtain the path trajectory of this ant k;

[0061] Step g: Repeat Step d - Step f until the path trajectories of all ants are obtained;

[0062] Step h: Calculate the corresponding dominance value J according to the path trajectory of each ant k , the dominance value J k The calculation formula is as follows:

[0063]

[0064] In the formula, ω1 represents the weight of the information gain J χ (t), ω2 represents the weight of the detection gain J D (t), ω3 represents the weight of the total gain J E (t), ω1 + ω2 + ω3 = 1, detectedMap(x,y) represents the search state indication, detectedMap(x,y) = 1 means the grid (x,y) has been searched, detectedMap(x,y) = 0 means the grid (x,y) has not been searched, represents the search area of UAV i, and ExceptedArea represents the expected search area;

[0065] Step i: Select the path trajectory of the ant corresponding to the maximum dominance value. According to the path trajectory of this ant, update the pheromone matrix. Repeat Step d - Step h until n iterations are completed, and output the path trajectory corresponding to the maximum dominance value in all iterations as the control decision sequence.

[0066] The AACO algorithm in this embodiment adapts the search strategy by introducing the self-attention mechanism for the conventional instruction information, prior information, and sudden information of the task, so as to meet different search expectations to the greatest extent.

[0067] Further, the specific process of the multi-UAV collaborative ground search online planning method under communication constraints in this embodiment is described. In this embodiment, based on the idea of distributed model predictive control, the search task planning problem for the entire mission cycle is transformed into a planning problem within a short time domain. At the same time, the planning strategy can be adjusted in real time during the communication available phase, and a distributed model predictive control framework based on Voronoi diagram partitioning is established. As Figure 2 shown, all the UAVs performing the task can be regarded as a large system. Then the state of the UAVs in this system is X(t) = {X1(t), X2(t), …, X n (t)}, and the control input of the system is U(t) = {U1(t), U2(t), …, U n (t)}. Then the state equation of the entire collaborative system is denoted as: X(t + 1) = f(X(t), U(t)).

[0068] Among them, f is the state transition function of the UAV system. Denote the state of the i-th UAV State i (t) as X i (t) = [p i (t), h i (t)] T , where p i (t) represents the current position of the i-th UAV, and h i (t) represents the current heading of the i-th UAV. U i (t) = [Δh i (t)] is the control input of the UAV, and Δh i (t) represents the next forward direction of the i-th UAV. Then, within the search area, the state equation of the i-th UAV can be expressed as:

[0069] [[ID=3,2]]

[0070] Under this framework, the multi-UAV collaborative ground search online planning method under communication constraints in this embodiment includes:

[0071] Step 1: Establish a search graph and an initial uncertain information graph for the mission area, and construct a motion model and an on-board sensor model of the UAVs;

[0072] In this embodiment, Step 1 includes:

[0073] Step 1.1: Perform grid processing on the mission area to establish a search graph for the mission area;

[0074] Assume that the mission area is a rectangle with length L x and width L yThe rectangular area, i.e., Environment = {(x,y)|x = 1,2,...,L x , y = 1,2,...L y}, and the grid is identified by (x,y).

[0075] Step 1.2: According to the prior existence probability P initial of the search target, determine the target existence probability of each grid at the initial moment. Based on the target existence probability of each grid at the initial moment, construct the initial target existence probability map;

[0076] In the actual search task, there is often a prior existence probability P initial of the search target in a specific area. Therefore, the target existence probability of each grid at the initial moment t0 is P(x,y,t0) = P initial , and then an initial target existence probability map of the task area composed of these grids with prior probabilities is constructed.

[0077] Step 1.3: Generate an initial uncertainty information map according to the initial target existence probability map. The calculation formula for the uncertainty of each grid is as follows:

[0078]

[0079] In the formula, χ(x,y,t) represents the uncertainty of the grid (x,y) at time t, (x,y) represents the coordinates of the grid, and P(x,y,t) represents the target existence probability of the grid (x,y) at time t;

[0080] Step 1.4: Set the state of the UAV i at time t as: represents the current position information of the UAV i, represents the current heading information of the UAV i. According to the state representation of each UAV, construct the motion model of the UAV;

[0081] Among them, the current position information of the UAV i is represented as:

[0082]

[0083] In the formula, represents the grid coordinates where the UAV i is located, L x represents the length of the x-axis of the task area, L y represents the length of the y-axis of the task area;

[0084] The current heading information of the UAV i is represented as:

[0085]

[0086] In the formula, H represents the heading, and h represents the heading angle of the area where the UAV can go to in the next moment in the grid. Among them, 1 represents directly ahead, 2 represents 45° to the right front, 3 represents directly to the right, 4 represents 135° to the right rear, 5 represents 135° to the left rear, 6 represents directly to the left, and 7 represents 45° to the left front.

[0087] Step 1.5: In the grid environment, the detection efficiency of the on-board sensor is p d representing the detection probability of the on-board sensor for the current grid where it is located, r represents the distance between the detected cell and the cell where the on-board sensor is located, and a represents the sensor adjustment parameter. According to the detection efficiency of each on-board sensor, an on-board sensor model is constructed.

[0088] Step 2: Determine whether the communication at the current moment is available;

[0089] Among them, when the communication is available, obtain the position information of all the UAVs that can communicate currently, perform area division using the Voronoi diagram division method according to the position information of all the UAVs and update the area division result. According to the updated area division result, each UAV executes the ant colony algorithm based on the attention mechanism to obtain the corresponding first control decision sequence;

[0090] When the communication is unavailable, according to the area division result of the previous moment, each UAV executes the ant colony algorithm based on the attention mechanism to obtain the corresponding second control decision sequence;

[0091] Step 3: When the communication is unavailable, control the UAV to execute the corresponding second control decision sequence. After execution, update the target existence probability map and the uncertainty information map, and judge whether the updated uncertainty information map meets the preset uncertainty requirement. If it meets, end the search task; if not, return to Step 2 for the next round of search planning until the search task ends;

[0092] When the communication is available, each UAV obtains the corresponding first control decision sequence of other UAVs to achieve information sharing. After information sharing, update the target existence probability map and the uncertainty information map of the task area, and judge whether the updated uncertainty information map meets the preset uncertainty requirement. If it meets, end the search task; if not, continue with Step 4;

[0093] In this embodiment, the preset uncertainty requirement is that the average uncertainty of the task area is less than 0.1.

[0094] Step 4: According to the first control decision sequence corresponding to the UAV and the current state of the UAV, calculate the corresponding benefit value for each UAV and add it to the benefit value set;

[0095] Among them, the calculation method of the profit value Profit is as follows:

[0096]

[0097] s.t.

[0098]

[0099] In the formula, represents the current state of the UAV at the moment, represents the current decision sequence of the UAV at the moment, N v represents the number of UAVs.

[0100] Step 5: According to the preset prediction step length T p , perform the operation of generating predictive control decisions for each UAV to obtain the predictive control decision sequence corresponding to each UAV;

[0101] The operation of generating predictive control decisions includes: when 0 ≤ t < T p , according to the control decision sequence of the UAV at the moment, obtain the area division result at the moment, and according to this area division result, execute the ant colony algorithm based on the attention mechanism to generate the control decision sequence at the moment, and use it as the predictive control decision sequence of this UAV;

[0102] Step 6: According to the predictive control decision sequence corresponding to the UAV and the current state of the UAV, calculate the new profit value corresponding to each UAV, compare the new profit value of each UAV with the profit value in the profit value set, update the profit value set according to the comparison result, and repeat Step 5 - Step 6 until the preset update termination condition of the profit value set is reached to obtain the final profit value set;

[0103] In this embodiment, comparing the new profit value of each UAV with the profit value in the profit value set and updating the profit value set according to the comparison result includes:

[0104] Solve the absolute value of the difference between the new profit value of the UAV and the profit value in the profit value set. If the absolute value of this difference is greater than the preset difference threshold, then retain the larger profit value in the profit value set.

[0105] Among them, the preset update termination condition of the profit value set is:

[0106] The absolute value of the above difference is less than or equal to the preset difference threshold, or the number of updates of the profit value set reaches the preset number of times.

[0107] Step 7: Use the decision sequence corresponding to the benefit value in the final benefit value set as the control decision sequence for the corresponding UAV, control the UAV to execute this control decision sequence, and after execution, return to Step 2 to perform the next round of search planning until the search task ends.

[0108] The online planning method for multi-UAV cooperative ground search under communication constraints in the embodiments of the present invention solves the problem of UAV formations performing search tasks under communication constraints where data interaction is only carried out within a limited time window in a UAV cluster in a complex environment. The problem of poor communication is fully considered, and the search area is divided based on the Voronoi diagram within a limited time window, effectively solving the problems of search area overlap and UAV collision caused by the inability of each unit in the cluster to interact real-time data.

[0109] Embodiment 2

[0110] This embodiment verifies and illustrates the effect of the method proposed in Embodiment 1 through simulation experiments.

[0111] Simulation 1: Simulation analysis of the AACO method:

[0112] Assume that at the initial moment of the simulation, a UAV searches a 30 km × 30 km area, and there are two key targets in the scene. Their target parameters are set as shown in Table 1 below.

[0113] Table 1 Target parameter settings

[0114]

[0115] The initialization parameters of the ant colony algorithm are shown in Table 2 below.

[0116] Table 2 Parameter settings of the AACO method

[0117]

[0118] First, initialize the scene to determine the probability map and environmental uncertainty map before the search. As the UAV continuously searches the area, the environmental uncertainty of the area will gradually decrease. When the UAV reaches the 38th step, and the environmental uncertainty of the two key areas drops to 0, completing the priority search task for the key areas. After that, the UAV will use the centroid of the environmental uncertainty as heuristic information to search the task area. The subsequent search trajectory is as shown in Figure 3 , and after completing all 200 steps of the search, the environmental uncertainty information map after the search is as shown in Figure 4As shown. The comparison of its search efficiency with the three common methods (RS greedy search algorithm, stander - ACO standard ant colony algorithm, PSS parallel line method) is as follows Figure 5 As shown

[0119] Simulation 2: DMPC - AACO simulation analysis based on Voronoi diagram partitioning method

[0120] Establish a distributed UAV cluster search mission planning platform for large - scale scenarios, and set the scenario as follows

[0121] (1) There are 4 UAVs at different positions in the cluster

[0122] (2) The size of the mission area is $500km\times 500km$

[0123] (3) There are 10 mission areas, and the target location data, prior probability information of the existence of targets and area properties are shown in Table 3

[0124] (4) In this example, the limitation of no - fly zones is added, and the no - fly zone positions are shown in Table 3

[0125] (5) For silent communication, the UAVs can communicate once every 80 decision steps; after the communication window at 960 steps, electromagnetic interference appears, the data link fails, and the electromagnetic interference lasts until 1440 steps

[0126] Table 3 Environmental parameter settings

[0127]

[0128] The parameters of the DMPC - AACO method are initialized as shown in Table 4. The prediction step of model predictive control is 15, the execution step is 10, and the termination condition of the final task is that the environmental uncertainty of the entire mission area is less than 0.1

[0129] Table 4 DMPC - AACO method parameter settings

[0130]

[0131] The results are as Figure 6 and Figure 7 shown Figure 6 In, the circular area is the diffusion result of the focus area, and the rectangular area is the no - fly zone Figure 7 In, the rectangular column is the no - fly zone. As shown in the result diagram, the four UAVs cooperate to reduce the environmental uncertainty to the specified threshold. The average environmental uncertainty of the final mission area is <0.05, and the UAVs have completed the search for key focus points and emergency mission points

[0132] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant is intended to cover non-exclusive inclusion, so that an article or device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the article or device comprising the said element. Similar words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The orientation or positional relationship indicated by "above", "below", "left", "right", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be construed as a limitation on the present invention.

[0133] The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for online planning of multi-UAV coordinated ground search under communication-restricted conditions, characterized by: include: Step 1: Establish a search map and initial uncertainty information map for the mission area, and build a UAV motion model and airborne sensor model; Step 2: Determine the current Whether communication is available at the moment, When communication is available, the location information of all currently communicable UAVs is obtained. The Voronoi diagram partitioning method is used to divide the regions according to the location information of all UAVs and the region partitioning results are updated. Based on the updated region partitioning results, each UAV executes the ant colony algorithm based on the attention mechanism to obtain the corresponding first control decision sequence; When communication is unavailable, each UAV executes an ant colony algorithm based on the attention mechanism according to the area division result at the previous moment to obtain the corresponding second control decision sequence; Step 3: When communication is unavailable, the UAV is controlled to execute the corresponding second control decision sequence. After execution, the target existence probability map and uncertainty information map are updated. It is determined whether the updated uncertainty information map meets the preset uncertainty requirements. If so, the search mission is terminated. If not, the search plan is returned to step 2 for the next round until the search mission is completed. When communication is available, each UAV obtains the first control decision sequence corresponding to the other UAVs to achieve information sharing. After information sharing, the target existence probability map and uncertainty information map of the mission area are updated. It is determined whether the updated uncertainty information map meets the preset uncertainty requirements. If so, the search mission is terminated. If not, the search continues to step 4. Step 4: Calculate the benefit value corresponding to each drone based on the first control decision sequence corresponding to the drone and the current state of the drone, and add it to the benefit value set; Step 5: According to the preset prediction step T p , perform predictive control decision generation operations on each UAV and obtain the predictive control decision sequence corresponding to each UAV; The predictive control decision generation operation includes: <T p When the drone The control decision sequence at time The regional division result at the moment, according to the regional division result, executes the ant colony algorithm based on the attention mechanism to generate The control decision sequence at each moment is used as the predicted control decision sequence of the UAV; Step 6: Based on the predictive control decision sequence corresponding to the UAV and the current state of the UAV, calculate the new benefit value corresponding to each UAV, compare the new benefit value of each UAV with the benefit value in the benefit value set, and update the benefit value set based on the comparison result. Repeat steps 5-6 until the preset benefit value set update cutoff condition is met to obtain the final benefit value set; Step 7: The decision sequence corresponding to the benefit value in the final benefit value set is used as the control decision sequence of the corresponding UAV. The UAV is controlled to execute the control decision sequence. After execution, it returns to step 2 and performs the next round of search planning until the search mission is completed. The ant colony algorithm based on the attention mechanism includes the following steps: Step a: For each drone, the attention mechanism is used to update the target existence probability map and uncertainty information map according to different attention points, where the attention points include: key attention points, sudden attention points, and general attention points; Step b: Generate the desired search area based on the focus point; Step c: Execute the ant colony algorithm in the desired search area and initialize the parameters of the ant colony algorithm, which include the number of ants M, the number of iterations n, the pheromone heuristic factor α, the expected heuristic factor β and the search factor detect j ; Step d: For ant k, according to the UAV's motion model, limit the optional path points to be selected at the next moment, and the probability P of selecting the next moment's path point n from the current moment t's path point m is: mn The calculation is as follows: Where, τ mn (t) represents the pheromone matrix from path point m to path point n at the current time t, η mn (t) represents the heuristic information from path point m to path point n at the current time t, A n It represents the total number of grids that can be detected when the UAV is at the path point n, D n It represents the total number of grids that have been detected within the detectable range when the UAV is located at the path point n, ∈ represents the coefficient of the search factor; allow represents the UAV’s allowed flight area; Step e: Update the uncertainty information graph based on the onboard sensor model and the selected path point at the next moment. The uncertainty change of the grid at a distance r around the drone path is updated by the following formula: χ(x,y,t+1)=(1-f(r))χ(x,y,t); Where χ(x,y,t) represents the uncertainty of the grid (x,y) at time t, (x,y) represents the coordinates of the grid; f(r) is the detection efficiency of the airborne sensor, and r represents the distance between the detected cell and the cell where the airborne sensor is located; Step f: Repeat steps d to e until the number of ant steps meets the preset ant step threshold, and obtain the path trajectory of ant k; Step g: Repeat steps d to f until the paths of all ants are obtained; Step h: Calculate the corresponding advantage value J based on the path trajectory of each ant k , the advantage value J k The calculation formula is as follows: Where ω1 represents the information benefit J χ (t), ω2 represents the detection benefit J D (t), ω3 represents the total benefit J E The weight of (t), ω1+ω2+ω3=1, detectedMap(x,y) represents the search status, detectedMap(x,y)=1 means that the grid (x,y) has been searched, detectedMap(x,y)=0 means that the grid (x,y) has not been searched, represents the search area of UAV i, ExceptedArea represents the expected search area, and P(x,y,t) represents the probability of the target existing in the grid (x,y) at time t; Step i: Select the path trajectory of the ant corresponding to the maximum advantage value, update the pheromone matrix according to the path trajectory of the ant, repeat steps d to h until n iterations are completed, and output the path trajectory corresponding to the maximum advantage value in all iterations as the control decision sequence.

2. The online planning method for multi-UAV coordinated ground search under communication-restricted conditions according to claim 1 is characterized in that: The step 1 comprises: Step 1.1: Grid the task area and create a search map for the task area. Step 1.2: Based on the prior probability P of the search target initial , determining the target existence probability of each grid at the initial moment, and constructing an initial target existence probability map based on the target existence probability of each grid at the initial moment; Step 1.3: Generate an initial uncertainty information map based on the initial target existence probability map. The calculation formula for the uncertainty of each grid is as follows: χ(x,y,t)=-P(x,y,t)log2(P(x,y,t))-(1-P(x,y,t))log2(1-P(x,y,t)); Where χ(x,y,t) represents the uncertainty of the grid (x,y) at time t, (x,y) represents the coordinates of the grid, and P(x,y,t) represents the probability of the target existing in the grid (x,y) at time t. Step 1.4: Set the state of drone i at time t to be: Indicates the current location information of drone i. Represents the current heading information of UAV i. Based on the state representation of each UAV, the motion model of the UAV is constructed. Step 1.5: In a gridded environment, the detection efficiency of the airborne sensor is p d Represents the detection probability of the airborne sensor for the current grid, r represents the distance between the detected cell and the cell where the airborne sensor is located, and a represents the sensor adjustment parameter. According to the detection efficiency of each airborne sensor, an airborne sensor model is constructed.

3. The online planning method for collaborative ground search by multiple UAVs under communication-restricted conditions according to claim 2 is characterized in that: Drone's current location information Expressed as: Where, Indicates the grid coordinates where UAV i is located, L x Indicates the x-axis length of the task area, L y Indicates the y-axis length of the task area; Drone's current heading information Expressed as: Where H represents the heading, h represents the heading angle of the area that the UAV can go to at the next moment in the grid, where 1 represents the front, 2 represents 45° to the right front, 3 represents the right, 4 represents 135° to the right rear, 5 represents 135° to the left rear, 6 represents the left, and 7 represents 45° to the left front.

4. The online planning method for collaborative ground search by multiple UAVs under communication-restricted conditions according to claim 1 is characterized in that: In step a, if there is a focus point in the drone's search area, the uncertainty information map of the grid (x, y) surrounding the focus point is updated as follows: In the formula, (x important ,y important ) represents the grid coordinates of the focus point, P prior (x important ,y important ) represents the prior probability of the target in the grid where the focus point is located, σ 2 Used to control the probability density diffusion range of the key search area, N important Indicates the total number of key focus points; If there is a sudden point of interest in the drone's search area, the uncertainty information map of the surrounding grid (x, y) of the sudden point of interest is updated as follows: In the formula, (x ∈ ,y ∈ ) represents the grid coordinates of the sudden focus point, the movement of the sudden focus point target obeys the two-dimensional standard normal distribution If there are no key points of interest or sudden points of interest within the drone's search area, the uncertainty centroid is used as the normal point of interest. The uncertainty centroid is calculated as follows:

5. The online planning method for collaborative ground search by multiple UAVs under communication-restricted conditions according to claim 4 is characterized in that: In step b, the calculation formula for the expected search area is as follows: In the formula, (x AttentionPoint ,y AttentionPoint ) represents the grid coordinates of the point of interest. Indicates the grid coordinates of the drone, and (x0, y0) indicates the grid coordinates within the desired search area.

6. The online planning method for multi-UAV coordinated ground search under communication-restricted conditions according to claim 1 is characterized in that: The profit value Profit is calculated as follows: st In the formula, Profit represents the profit value, Indicates the current state of the drone The state of the moment, Indicates the current state of the drone The decision sequence at the moment, N v Indicates the number of drones.

7. The online planning method for collaborative ground search by multiple UAVs under communication-restricted conditions according to claim 6 is characterized in that: The preset benefit value set update deadline condition is: The absolute value of the difference between the new benefit value of the UAV and the benefit value in the benefit value set is less than or equal to the preset difference threshold, or the number of times the benefit value set is updated reaches the preset number.

Images