Unmanned aerial vehicle cluster reconnaissance task planning method and equipment based on hierarchical model

Abstract

The invention provides an unmanned aerial vehicle cluster reconnaissance task planning method and device based on a hierarchical model, and the method comprises the following steps: pre-collecting a battlefield situation, carrying out the pre-task planning according to the pre-collected battlefield situation, and enabling each unmanned aerial vehicle in an unmanned aerial vehicle cluster to execute a task according to the pre-task planning; in the task execution process, if a new battlefield situation occurs to cause that the pre-task planning scheme is not suitable any more, dynamic task planning is triggered; after dynamic task planning is triggered, alliance establishment is carried out, unmanned aerial vehicle clusters form a new formation model according to a hierarchical model, a new battlefield situation is collected, and task planning is carried out on the unmanned aerial vehicle clusters again; in the new formation model, the unmanned aerial vehicle cluster is divided into reconnaissance unmanned aerial vehicles and task unmanned aerial vehicles, a new battlefield situation is collected, and tasks are distributed to the unmanned aerial vehicle cluster again. The dynamic instant planning problem under the combat condition is considered, and the task completion efficiency is higher.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned aerial vehicle mission planning, and in particular relates to a method and device for planning an unmanned aerial vehicle cluster reconnaissance mission based on a hierarchical model. Background Art

[0002] With the progress and development of science and technology, intelligent and unmanned weapons and equipment are increasingly being used in modern battlefields, and the application of unmanned aerial vehicle (UAV) swarms has become one of the hot research areas. UAV swarms combine the characteristics of drones and swarms, and have the advantages of high self-organization, strong adaptability, and strong group stability. In modern warfare, drone swarm reconnaissance is a hot area for the application of unmanned equipment. However, the complex environment of the battlefield puts forward higher requirements on the timeliness and mobility of drone swarm reconnaissance.

[0003] Considering the limited resources and the diversity of combat objectives, the problem is generally abstracted as a multi-objective constraint problem. Traditional solution methods include enumeration method, dynamic programming method, etc. However, in large-scale drone swarm combat, there are many parameters that need to be coordinated, and the dimension of the solution space increases geometrically with the number of drones and the number of mission objectives. The low efficiency of traditional solutions makes it difficult to meet the timeliness requirements under battlefield conditions.

[0004] In the process of drone swarms performing reconnaissance missions, drone swarms may encounter emergencies such as drone damage and the emergence of new mission targets, which require the pre-planned plan to be updated. How to achieve resource integration, information processing, task re-planning and assignment in a short time requires not only the drone swarm to have the ability to plan tasks, but also the drones to achieve consistent situational awareness and stable communication of the cluster. The above situations are collectively referred to as dynamic real-time planning problems. The mainstream solution is to abstract the process of cluster situational awareness and planning into the process of alliance formation. However, in existing studies on the scenario of sudden failure of some drones, the alliance formation strategy remains centered on the faulty drone. Due to communication limitations, it is neither practical nor effective in exerting the collaboration of the cluster. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method and device for planning a UAV cluster reconnaissance mission based on a hierarchical model.

[0006] The present invention includes: a method for planning a UAV cluster reconnaissance mission based on a hierarchical model, which includes the following steps:

[0007] Collect battlefield situations in advance, carry out pre-task planning based on the pre-collected battlefield situations, and each drone in the drone cluster executes the task according to the pre-task planning;

[0008] During the mission execution, if a new battlefield situation arises and the pre-planned mission plan is no longer applicable, dynamic mission planning will be triggered;

[0009] After the dynamic mission planning is triggered, an alliance is formed, and the drone cluster forms a new formation model according to the hierarchical model, collects new battlefield situations, and re-plans tasks for the drone cluster; in the new formation model, the drone cluster is divided into two types of drones, including reconnaissance drones and mission drones; the reconnaissance drone is responsible for monitoring the battlefield environment, conducting situation fusion and mission planning, and as a communication hub, establishing a communication network with the mission drone, collecting and processing information uploaded by the mission drone; the mission drone is responsible for enemy reconnaissance, performing tasks according to the pre-mission plan or the tasks and tracks planned by the reconnaissance drone, and at the same time providing real-time feedback on the mission completion to the reconnaissance drone.

[0010] According to the above technical solution of the present invention, the following improvements can also be made:

[0011] Optionally, pre-mission planning includes the following steps:

[0012] A combat environment parameter model is established based on the battlefield situation collected in advance. The combat environment parameter model has several restrictions, including at least no-fly zones and weapon threats. The combat parameter model avoids the restrictions to plan the trajectory of the drone cluster.

[0013] Then, a multi-objective constraint model is established according to the constraints existing in the task execution process, and the task allocation is performed based on the multi-objective constraint model. The constraints include the UAV self-constraints and task constraints. The UAV self-constraints are that the total fuel consumption of a single UAV must be less than the maximum fuel capacity when the UAV cluster performs the task. The task constraint is that when the UAV cluster performs the task, each task point is completed by a certain UAV from the starting point.

[0014] The pre-task planning is formed by coupling the trajectory planning and the task allocation. If there is no restriction between the two task targets in the process, the straight line between the two task targets is taken as the trajectory. If there is a restriction between the two task targets, the trajectory is corrected and coupled with the task allocation. When correcting the trajectory, for the case where the straight track passes through the no-fly zone, a point is selected outside the no-fly zone so that the lines connecting the point with the starting point and the end point are tangent to the no-fly zone, and the two corrected trajectories are taken as the trajectory between the two task targets.

[0015] Optionally, the multi-objective constraint model is as follows:

[0016] P1:

[0017] P2:

[0018]

[0019] Among them, P1 and P2 are the no-fly zone sets, n = cu{0,n+1}, Q is the full fuel load of the drone, C ij is the distance between tasks i and j, x ij Indicates whether the arc from task i to j is selected, is the track correction operator, is the vertical distance from the center of the no-fly zone n to the line connecting the two mission points i and j, r P is the radius of the no-fly zone P; is the constraint of the drone itself; k is the number of drones, r n is the radius of the no-fly zone n.

[0020]

[0021]

[0022]

[0023] is the task constraint, where K represents the maximum number of arcs from the starting point, and x ik ,x kj Respectively indicate whether the outgoing arc and incoming arc of task point k are selected, x oj Indicates whether the arc from the starting point o to other task points is selected;

[0024] The total path distance is expressed as:

[0025]

[0026] x ij ∈{0,1},

[0027] Where C ij Represents the distance between two target points;

[0028] The total path risk is expressed as:

[0029]

[0030]

[0031] Where f ij represents the quantitative risk index between two target points, x ij Indicates whether the route from task i to j is selected, M is the air defense weapon set, r P is the radius of the no-fly zone p, r m is the radius of the mth air defense zone, is the vertical distance from the air defense zone m to the line connecting the two mission points i and j, and o is the unit fuel consumption.

[0032] Optionally, in the multi-objective constraint model, all drones in the drone cluster are divided into different levels, there is no domination between drones at the same level, and high-level drones dominate low-level drones. Each drone i is set to have parameters n(i) and s(i), where n(i) is the number of solutions that dominate drone i in the drone cluster, and s(i) is the set of solutions dominated by drone i; then the non-dominated sorting of drones is performed according to the following steps:

[0033] S001. Store all drones with n(i)=0 in the drone cluster into set F1, and assign the same non-dominated order i (rank) to the drones in the set;

[0034] S002, the set of statistically dominant drones S = {S i |i∈F1};

[0035] S003, reduce n(i) of all drones in set S by 1;

[0036] S004. Repeat steps S001 to S003 until all drones are classified.

[0037] Optionally, when all drones are classified, the hierarchical relationship within the drone cluster is obtained. For the drones in each level, the drones in this level and the drones in the previous level are composed of a drone sub-population. Each drone in the drone cluster represents an individual in the population. The two drones with the closest distance are selected to establish an intra-layer or inter-layer relationship. The above operations are repeated until all drones complete the establishment of intra-layer or inter-layer relationships, and a drone cluster super network model is obtained.

[0038] Optionally, the congestion degree is calculated based on the obtained drone cluster super network model. The congestion degree is calculated in the following way:

[0039]

[0040] Among them, f m is the objective function of the mth dimension, rank is the level corresponding to the drone, and the weights of distance calculation between different levels are assigned by the level difference. i-1 and i+1 correspond to the two nodes closest to drone i in the drone sub-population composed of the current level and the previous level, respectively.

[0041] Optionally, the genetic operator of the drone swarm hypernetwork model is as follows:

[0042] Selection operator: Perform non-dominated sorting on the population after the merger of the father and son generations, add individuals from each layer in turn, and select several individuals with the lowest crowding degree from the critical layer to be retained for the next generation;

[0043] Crossover operator: Perform a two-point crossover operation on individuals, and then modify the genes outside the crossover area according to the mapping relationship of each gene in the crossover area to eliminate conflicts;

[0044] Mutation operator: The random exchange operator is used for mutation operation, and two genes in an individual are randomly selected for exchange to obtain the mutated individual.

[0045] Optionally, when dynamic mission planning is triggered, a new drone formation is established, the new drone formation is a three-dimensional drone formation, and then an alliance is formed; the alliance formation includes the following steps:

[0046] Tender Release: When UAV U i When the drone is destroyed, the drone cluster enters the bidding release state. After the reconnaissance drone obtains the information, it becomes the leader of the alliance and establishes a local communication network with the UAVs in the neighborhood to release the bidding information P i :

[0047] Pi={U i ,T};

[0048] Where U i It represents the basic information of the UAV, and T represents the remaining mission objectives of the UAV;

[0049] Bidding application: After the bidding is released, all drones in the local communication network receive the bidding information. The drones first judge the bidding information. If a drone meets the conditions in the multi-objective constraint model, it becomes a bidding drone and then sends a bid to the alliance leader.

[0050] Bi={U i ,T i};

[0051] Where Ti represents the target to be auctioned;

[0052] Alliance construction: After the bidding application, the alliance leader receives the bids from the bidding drones in the communication network. The alliance leader needs to sort the bids, select the drone with the highest performance, and publish the winning bid information. The sorting of the bids takes into account the risk and distance cost. The alliance leader constructs an effective set according to the Pareto criterion, randomly selects a bid from the effective set and publishes the winning bid information F i Feedback to all bidding drones, F i Can be represented by a binary group;

[0053] F i={S, TR};

[0054] Where S represents the updated task sequence and TR represents the updated flight path;

[0055] Execution phase: The UAVs that win the bid will update their task sequences and flight paths to complete dynamic task planning.

[0056] Optionally, during the coalition formation process, the flight path planning method for reconnaissance UAVs is as follows:

[0057] Collect relevant information: The flight path of the reconnaissance UAV is planned in segments, changing the heading at fixed time intervals and maintaining a fixed heading and speed within the interval. When a heading change is required, the reconnaissance UAV requests relevant information from each UAV within its communication range;

[0058] Calculate the heading: The artificial potential field method is used to plan the flight path of the reconnaissance UAV. The specific steps are as follows:

[0059] Establish the gravitational field function:

[0060]

[0061] Where d(p, p i ) represents the distance between the mission UAV i and the reconnaissance aircraft, and r i represents the risk index of the mission UAV i. When d < d*, the magnitude of the gravitational potential energy is proportional to the square of the distance from the current position to the target position; when d > d*, the value of the gravitational calculation function is reduced to avoid the problem of excessive gravity when far from the target position. d* represents the communication radius of the reconnaissance UAV,

[0062] Establish the repulsive field function:

[0063]

[0064] Where d(p, p i ) represents the distance between the reconnaissance UAV itself and other reconnaissance UAVs i, α is the repulsive gain constant; q* is the action threshold range of the repulsion force, and the repulsion force will only be generated within this threshold range;

[0065] Calculate the heading based on the artificial force field:

[0066] Superimpose the gravitational field and the repulsive field to obtain the artificial force field U:

[0067] U(p) = U att (q) + U rep (q);

[0068] Take the gradient of the artificial potential field to obtain the potential field force F:

[0069]

[0070] Decompose the potential field force into two directions, X and Y, and normalize them to get the heading;

[0071] Execution of the voyage:

[0072] According to the obtained heading, keep flying at a certain speed.

[0073] According to another aspect of the present invention, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any one of the above methods when executing the program.

[0074] The present invention provides a method and device for planning a UAV swarm reconnaissance mission based on a hierarchical model. It focuses on the task of UAV swarms performing target strikes in a battlefield environment. It proposes a new solution to the problem that traditional models cannot meet the time delay requirements when forming an alliance. The technical effects include:

[0075] (1) For the UAV swarm mission planning problem, this method is superior to the traditional UAV swarm mission planning method in terms of timeliness and solution space.

[0076] (2) Hypernetwork modeling was performed to introduce a new measurement perspective for determining the crowding degree and priority evaluation of genetic individuals.

[0077] (3) Considering the dynamic planning problem under the condition of UAV damage, a new UAV formation model is proposed. By moving the planning and communication center to the command UAV, a multi-level formation of command and task UAVs is established to solve the problem of cluster task replanning when the UAV is damaged. The new formation model can effectively enhance the cooperation ability of UAVs and improve the task completion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a schematic diagram of the UAV cluster collaborative task execution of the present invention.

[0079] Figure 2 It is a schematic diagram of the no-fly zone correction of the present invention.

[0080] Figure 3 It is a coding schematic diagram of the present invention.

[0081] Figure 4 This is a schematic diagram of the hypernetwork modeling of the present invention.

[0082] Figure 5 A schematic diagram for selecting offspring of the operator of the present invention.

[0083] Figure 6 The figure is a flowchart of the task planning method of the genetic method of the present invention.

[0084] Figure 7 This is a schematic diagram of the UAV cluster communication network of the present invention.

[0085] Figure 8 It is a schematic diagram of the traditional formation model of the present invention.

[0086] Fig. 9 It is a schematic diagram of the new formation model of the present invention.

[0087] Fig.10 It is a schematic diagram of simulation scenario three of the present invention.

[0088] Fig.11 It is a schematic diagram of simulation scenario 1 of the present invention.

[0089] Fig.12 It is a schematic diagram of simulation scenario 2 of the present invention.

[0090] Fig.13 Schematic diagram of the results of different methods for simulation scenario 1 of the present invention.

[0091] Fig.14 It is a schematic diagram of the results of different methods of simulation scenario 2 of the present invention.

[0092] Fig.15 Schematic diagram of the results of different methods for simulation scenario three of the present invention.

[0093] Fig.16 It is a schematic diagram of the execution time of the method of the present invention.

[0094] Fig.17 It is a schematic diagram of the formation track of the present invention.

[0095] Fig.18 Schematic diagram of the impact of the formation model of the present invention on the task completion rate.

[0096] Fig.19 It is a schematic diagram of the influence of the formation model of the present invention on the total sailing distance. DETAILED DESCRIPTION

[0097] like Figure 1-Figure 19 As shown, the present invention provides a method and device for planning a UAV cluster reconnaissance mission based on a hierarchical model, the method comprising:

[0098] Collect battlefield situations in advance, carry out pre-task planning based on the pre-collected battlefield situations, and each drone in the drone cluster executes the task according to the pre-task planning;

[0099] During the mission execution, if a new battlefield situation arises and the pre-planned mission plan is no longer applicable, dynamic mission planning will be triggered;

[0100] After dynamic mission planning is triggered, an alliance is formed, new battlefield situations are collected, and tasks are reassigned to the drone cluster.

[0101] In this embodiment, the mission objective can be described as sending m drones to n target points for detection. After collecting target information and returning to the base, the mission is considered completed. First, according to the pre-collected battlefield situation, that is, the enemy and our situation, battlefield environment, etc., pre-task planning is carried out, and each drone sets out to perform the mission according to the planning results; during the mission execution process, if a new battlefield situation such as drone damage occurs, the pre-task planning scheme is no longer applicable, then dynamic mission planning is triggered, and through alliance formation, new battlefield situation is collected, and tasks are re-assigned to the drone cluster.

[0102] Figure 1 An example of a mission scenario with 4 mission points and 2 drones is shown.

[0103] Optionally, the drone swarm is divided into two types of drones, including reconnaissance drones and mission drones, where

[0104] The reconnaissance drone is responsible for monitoring the battlefield environment, conducting situation integration and mission planning, and as a communication hub, establishing a communication network with the mission drones, collecting and processing information uploaded by the mission drones, and playing the role of a communication center and planning center;

[0105] The function of the mission drone is to conduct reconnaissance against the enemy. It does not have autonomy and executes according to the tasks and tracks planned in advance or by the reconnaissance drone. At the same time, it provides real-time feedback on the mission completion to the reconnaissance drone.

[0106] In this embodiment, the drone formation is composed of two types of drones, namely reconnaissance drones and mission drones. The reconnaissance drone is responsible for monitoring the battlefield environment, situation fusion, mission planning, and as a communication hub, establishing a communication network with the mission drone, collecting and processing information uploaded by the mission drone, and playing the role of a communication center and planning center; the mission drone's function is to conduct reconnaissance against the enemy, and it does not have autonomy. It executes according to the tasks and tracks planned in advance or by the reconnaissance drone, and at the same time, it feeds back the completion of the task to the reconnaissance drone in real time. The execution of cluster tasks needs to consider constraints such as its own flight performance, communication range, enemy fire threat, battlefield environment, etc., and plan with the goal of low-risk, low-consumption mission planning. When the drone cluster completes all n tasks or cannot complete the remaining tasks due to damage, etc., the operation ends.

[0107] Optionally, based on the pre-collected battlefield situation, a battlefield environment model is established as follows: E = {Z, W}; where E is the battlefield environment, Z = {Z i |i=1,2,…n2} is the terrain set, W={W i |i=1,2,…nW} for the enemy's air defense weapons assembly;

[0108] Pre-mission planning includes the following steps:

[0109] A combat environment parameter model is established based on the battlefield situation collected in advance. The combat environment parameter model has several restrictions, including at least no-fly zones and weapon threats. The combat parameter model avoids the restrictions to plan the trajectory of the drone cluster.

[0110] Then, a multi-objective constraint model is established according to the constraints existing in the task execution process, and the task allocation is performed based on the multi-objective constraint model. The constraints include the UAV self-constraints and task constraints. The UAV self-constraints are that the total fuel consumption of a single UAV must be less than the maximum fuel capacity when the UAV cluster performs the task. The task constraint is that when the UAV cluster performs the task, each task point is completed by a certain UAV from the starting point.

[0111] The pre-task planning is formed by coupling the trajectory planning and the task allocation. If there is no restriction between the two task targets in the process, the straight line between the two task targets is taken as the trajectory. If there is a restriction between the two task targets, the trajectory is corrected and coupled with the task allocation. When correcting the trajectory, for the case where the straight track passes through the no-fly zone, a point is selected outside the no-fly zone so that the lines connecting the point with the starting point and the end point are tangent to the no-fly zone, and the two corrected trajectories are taken as the trajectory between the two task targets.

[0112] Optionally, the multi-objective constraint model is as follows:

[0113] P1:

[0114] P2:

[0115]

[0116] Among them, P1 and P2 are the no-fly zone sets, n = cu{0,n+1}, Q is the full fuel load of the drone, C ij is the distance between tasks i and j, x ij Indicates whether the arc from task i to j is selected, is the track correction operator, is the vertical distance from the center of the no-fly zone n to the line connecting the two mission points i and j, r P is the radius of the no-fly zone P; is the constraint of the drone itself; k is the number of drones, r n is the radius of the no-fly zone n.

[0117]

[0118]

[0119]

[0120] is the task constraint, where K represents the maximum number of arcs from the starting point, and x ik ,x kj Respectively indicate whether the outgoing arc and incoming arc of task point k are selected, x oj Indicates whether the arc from the starting point o to other task points is selected;

[0121] The total path distance is expressed as:

[0122]

[0123] x ij ∈{0,1},

[0124] Where C ij Represents the distance between two target points;

[0125] The total path risk is expressed as:

[0126]

[0127]

[0128] Where f ij represents the quantitative risk index between two target points, x ij Indicates whether the route from task i to j is selected, M is the air defense weapon set, r P is the radius of the no-fly zone p, r m is the radius of the mth air defense zone, is the vertical distance from the air defense zone m to the line connecting the two mission points i and j, and o is the unit fuel consumption.

[0129] In this embodiment, the mission planning includes two parts: mission allocation and track planning. In the embodiment, the two parts are coupled. If there is no obstacle between the two targets, a straight line is taken as the track. If there is an obstacle between the two targets, correction is performed according to the following method:

[0130] like Figure 2 , introducing the distance correction operator For a straight line trajectory passing through a no-fly zone, select a point outside the zone so that the lines connecting the point and the starting and ending points of the trajectory are tangent to the no-fly zone, and take the sum of the two corrected segments as the distance between the two points. The constraints of the drone itself mainly consider the fuel constraints when the drone performs a mission, that is, the total fuel consumption of a single drone must be less than the maximum fuel capacity. Mission constraints describe the restrictions on the execution of missions by drone clusters. The drone cluster starts from the starting point, and each mission point is completed by a certain drone; the first two of the mission constraint formulas indicate that each target point must be reached and only reached once, and the latter formula indicates that there are at most K outgoing arcs from the starting point.

[0131] In this embodiment, based on NSGA-Ⅱ and combined with the idea of ​​super network modeling, the N2SGA-Ⅱ method is proposed, and the N2SGA-Ⅱ method is defined as a genetic method based on super network modeling.

[0132] Variable coding design, the solution code W consists of two parts: W = {M, S}; where M = {m1, m2, …, m n} represents the number of tasks assigned to each drone, and S represents the execution sequence of the tasks. Taking two drones performing four tasks as an example, the solution is encoded as Figure 3 As shown, Figure 3 It means that UAV 1 performs two tasks, and the task sequence is to perform task 2 first and then task 3. UAV 2 performs two tasks, and the task sequence is to perform task 1 first and then task 4.

[0133] The standard NSGA-Ⅱ method only considers the solution set of the same level in the congestion calculation, lacks global consideration of the solution set, and in the context of the problem in this embodiment, due to the discreteness of the solution space, clustered duplicate solutions are prone to occur, affecting the efficiency of iterative improvement. In view of the above situation, a genetic method based on hypernetwork modeling is proposed, which is called N2SGA-Ⅱ in this embodiment:

[0134] In the multi-objective constraint model, all drones in the drone cluster are divided into different levels. There is no domination between drones at the same level. High-level drones dominate low-level drones. Each drone i is set to have parameters n(i) and s(i), where n(i) is the number of solutions that dominate drone i in the drone cluster, and s(i) is the set of solutions dominated by drone i. Then, the non-dominated sorting of drones is performed according to the following steps:

[0135] S001. Store all drones with n(i)=0 in the drone cluster into set F1, and assign the same non-dominated order i (rank) to the drones in the set;

[0136] S002, the set of statistically dominant drones S = {S i |i∈F1};

[0137] S003, reduce n(i) of all drones in set S by 1;

[0138] S004. Repeat steps S001 to S003 until all drones are classified.

[0139] Optionally, when all drones are classified, the hierarchical relationship within the drone cluster is obtained. For the drones in each level, the drones in this level and the drones in the previous level are composed of a drone sub-population. Each drone in the drone cluster represents an individual in the population. The two drones with the closest distance are selected to establish an intra-layer or inter-layer relationship. The above operations are repeated until all drones complete the establishment of intra-layer or inter-layer relationships, and a drone cluster super network model is obtained.

[0140] Optionally, the congestion degree is calculated based on the obtained drone cluster super network model. The congestion degree is calculated in the following way:

[0141]

[0142] Among them, f m is the objective function of the mth dimension, rank is the level corresponding to the drone, and the weights of distance calculation between different levels are assigned by the level difference. i-1 and i+1 correspond to the two nodes closest to drone i in the drone sub-population composed of the current level and the previous level, respectively.

[0143] like Figure 4 As shown, F1, F2, and F3 represent the first, second, and third levels obtained by non-dominated sorting, respectively, and the connecting lines represent the relationships established according to the above conditions.

[0144] When calculating the crowding degree, the standard NSGA method only considers the distance between individuals on the same layer. Considering the iterative improvement of the solution, a solution close to the Pareto solution set surface can often be converted into a Pareto solution in the next iteration. Therefore, the crowding degree is calculated based on the population super network model, and the crowding degree is defined as follows:

[0145]

[0146] Among them, f m is the objective function of the mth dimension, rank is the level corresponding to the individual, and the weights of distance calculation between different levels are assigned by the level difference. i-1 and i+1 correspond to the two nodes closest to individual i in the subpopulation composed of the current level and the previous level, respectively.

[0147] exist Figure 4 In the figure, the crowdedness of solution i is the weighted sum of the perimeters of the cuboid enclosed by the dotted lines.

[0148] The specific operation operators of the individual are defined as follows (the specific process is as follows Figure 6 shown):

[0149] like Figure 5 As shown, the standard NSGA-Ⅱ method based on non-dominated sorting and crowding selection is retained, and the population after the merger of the father and son generations is stratified by non-dominated sorting. Individuals from each layer are added in turn, and several individuals with the lowest crowding are selected from the critical layer and retained to the next generation.

[0150] The standard NSGA-II method uses a simulated binary crossover operator, that is, a gene of two parents is selected and weighted summed according to a certain formula. However, in this problem, since the gene adopts integer coding, if a simulated binary crossover operation is performed, the solution obtained does not meet the coding specification, so it is not applicable. This embodiment uses a partial implicit crossover operator (PMX), and the crossover operation is completed in two steps. The first step is to perform a conventional two-point crossover operation on the individual; the second step is to modify the genes outside the crossover area according to the mapping relationship of each gene in the crossover area to eliminate conflicts.

[0151] The mutation operator of the standard NSGA-II method is a polynomial mutation operator, that is, through formula calculation, a random number between 0 and 1 is added or subtracted to a gene in the code. This embodiment adopts a random exchange operator to perform a mutation operation, randomly selects two genes in an individual to exchange, and obtains a mutated individual.

[0152] UAVs can achieve two-way communication through communication ad hoc networks, and can transmit tasks, their own status and other information through the network. The maximum communication distance of UAV networking is h. max and the maximum number of communication nodes a max Considering the time delay and transmission loss of information dissemination in the actual mission process, information is not allowed to be chained through adjacent drones. The adjacent drone group is called the domain of the central drone, and its information can be shared with drones in the neighborhood.

[0153] like Figure 7 As shown, Figure 7 There are 6 UAVs in China. The dotted circle represents the communication range of U1, and the dotted line represents the network connection between UAVs. The communication ad hoc network formed by U1 is Ls = {U2, U3, U4}. When U1 finds new situations or is damaged, it can send its information to the UAVs in Ls.

[0154] The drones formed by the alliance form a three-dimensional drone formation, such as Fig. 9 The new formation shown divides drones into mission drones and reconnaissance drones. Mission drones only perform tasks and upload their own location, status and other basic information to reconnaissance drones. Reconnaissance drones are responsible for mission planning, alliance formation and other tasks, and transmit the planning results to each mission drone. Figure 8This is a traditional formation model. Alliance formation specifically includes the following steps:

[0155] Tender Release: When UAV U i When the swarm system enters the bidding release state, the reconnaissance drone becomes the leader of the alliance after observing this phenomenon, and establishes a local communication network with the UAVs in the neighborhood to release the bidding information. i :

[0156] Pi={U i ,T};

[0157] Where U i It represents the basic information of the UAV, and T represents the remaining mission objectives of the UAV;

[0158] Bidding application: After the above steps, all drones in the local communication network have received the bidding information. First, the bidding information is judged. If the constraints such as fuel are met, the drone becomes a bidding drone and sends a bid Bi to the leader:

[0159] Bi={U i ,T i};

[0160] Where Ti represents the target to be auctioned;

[0161] Alliance construction: After the bidding application in the above steps, the alliance leader receives the bids from the bidding drones in the communication network. The leader needs to sort the bids, select the drone with the highest performance, and publish the winning bid information; the sorting of the bids takes into account the risk and distance cost. The leader constructs an effective set according to the Pareto criterion, randomly selects a bid from the effective set and publishes the winning bid information F i Feedback to all bidding drones, F i Can be represented by a binary group;

[0162] F i ={S,TR};

[0163] Where S represents the updated task sequence, and TR represents the updated track;

[0164] Execution phase: The successful UAV will update its mission sequence and trajectory to complete dynamic mission planning.

[0165] Reconnaissance UAVs play the role of communication relays and computing nodes in combat missions. In order to give full play to their effectiveness and ensure the efficient completion of missions, the following two requirements are met.

[0166] (1) The reconnaissance drone should cover as many mission drones as possible within its communication range so that more resources can be mobilized to deal with emergencies.

[0167] (2) The reconnaissance UAV should give priority to ensuring that the UAVs at risk are within the communication range.

[0168] Based on this, this embodiment draws on the idea of the artificial potential field method and proposes a trajectory planning method for reconnaissance UAVs based on gravity. The trajectory planning method for reconnaissance UAVs is as follows:

[0169] Collect relevant information: The trajectory of the reconnaissance UAV is planned in segments, changing the heading at fixed time intervals and maintaining a fixed heading and speed within the interval. When triggering a heading change, the reconnaissance aircraft requests relevant information from each UAV within the communication range;

[0170] Calculate the heading: The artificial potential field method is used to plan the trajectory of the reconnaissance UAV. The specific steps are as follows:

[0171] Establish the gravitational field function:

[0172]

[0173] In the formula, d(p, p i ) represents the distance between the mission UAV i and the reconnaissance aircraft, and r i represents the risk index of the mission UAV i. When d < d*, the magnitude of the gravitational potential energy is proportional to the square of the distance from the current position to the target position; when d > d*, the value of the gravitational calculation function is reduced to avoid the problem of excessive gravity when far from the target position. d* represents the communication radius of the reconnaissance aircraft.

[0174] Establish the repulsive field function:

[0175]

[0176] In the formula, d(p, p i ) represents the distance between the reconnaissance aircraft itself and other reconnaissance aircraft i. α is the repulsive gain constant; q* is the action threshold range of the repulsion force, and the repulsion force will only be generated within this threshold range;

[0177] Calculate the heading based on the artificial force field:

[0178] Superimpose the gravitational field and the repulsive field to obtain the artificial force field U:

[0179] U(p) = U att (q) + U rep (q);

[0180] Take the gradient of the artificial potential field to obtain the potential field force F:

[0181]

[0182] Decompose the potential field force into the X and Y directions and normalize it to obtain the heading;

[0183] Execution of the voyage:

[0184] According to the obtained heading, keep flying at a certain speed.

[0185] Optionally, the UAV swarm dynamic mission planning includes the following steps:

[0186] Mission re-planning: After receiving the re-planning command, the drone uses its current location as the starting point, its remaining missions and the accepted mission set as the target set, and uses the above method to re-plan;

[0187] Calculation of bidding cost: The drone calculates the fitness F1 after accepting the task and the fitness F2 after rejecting the task, and takes F1-F2 as the cost;

[0188] The winning drone is selected based on the bidding price, and the winning drone will carry out drone swarm mission re-planning.

[0189] Simulation and result analysis:

[0190] Depending on the number of task objectives, the design Figure 10-12 In the figure, the regular octagon represents the target, the triangle represents the starting point, the dot is the location of the enemy weapon, the surrounding black circle dotted line is the effective strike range of the weapon in the coverage area, and the gray circle coverage area is the no-fly zone. The number of task targets from scene one to scene three is 5, 10, and 20 respectively. The target points are randomly assigned in the task scenes to analyze the performance of the method under different task requirements. The experimental environment and parameters in the figure are shown in the following table:

[0191]

[0192] The performance of the method is tested from the solution quality and running time. The parameter settings are shown in the following table:

[0193]

[0194]

[0195] Calculate the Pareto solution surface of the above three methods in each scenario, and the results are as follows Figure 13-Figure 15As shown, in this embodiment, the solution space is the allocation and sorting of tasks, so the solution space is discrete and grows geometrically with the increase of the task volume. For multi-objective problems, the comparison of the advantages and disadvantages of the solution set first compares the fitness of each objective, which is the smallest total distance and risk value in this embodiment. In addition, the breadth and diversity of the solution set are compared. The results show that the improved NSGA-Ⅱ method performs well in all scenarios. The particle swarm method has poor performance for scenarios with relatively large task volumes, that is, complex solution spaces. The MOEAD method is not as robust as the improved NSGA-Ⅱ, and performs worst in scenarios with medium task volumes. In scenarios with high task volumes, the solution set is not as good as the improved NSGA-Ⅱ, but is worse than the particle swarm method. The improved NSGA-Ⅱ method performs close to the traditional NSGA-Ⅱ in scenarios one and two, and has better diversity in scenario three.

[0196] The running time of each method in different scenarios is counted, and the results are as follows Fig.16 As shown in the figure, in terms of time, the running time of the improved NSGA-Ⅱ method is significantly better than that of the other three methods. The improved NSGA-Ⅱ method improves the mutation operator and makes the offspring generated by mutation more effective, so it improves the running time of the method to a certain extent. The more complex the task scenario, the more variable dimensions, and the larger the solution space, the better the improvement effect.

[0197] The simulation results of the UAV swarm dynamic planning for damage are shown as follows: Fig.17 As shown, Fig.17 The whole process track of two drones and a reconnaissance aircraft performing the mission is shown. Route 1 and Route 2 are the initial planning paths of the drones, and Route 3 is the activity track of the reconnaissance aircraft. The drones first perform the combat mission according to the pre-planned results. When drone 1 navigates to the coordinates (-0.85, -1.8), it is destroyed by enemy weapons. At this time, the reconnaissance drone receives the information that drone 1 is destroyed, integrates the remaining tasks uploaded by drone 1 and the available resources-drone 2, and immediately uses the above method to re-plan (route 4), and sends the planned track to drone 2, which takes over the remaining tasks of drone 1. The results show that the method can quickly respond to battlefield changes and coordinate the remaining resources, which illustrates the feasibility of the method for solving the drone cluster planning problem in the embodiment.

[0198] The new formation model improves the efficiency of drone swarm mission planning. There are many angles to evaluate the efficiency of drone swarm reconnaissance missions, and the main angles to consider are the degree of mission completion and the cost of mission completion. The implementation example starts from the above two angles and selects an important indicator respectively - the improvement of the new formation model on the total flight range required for the drone formation to complete the mission under the premise of mission completion and ensuring mission completion, and compares the new formation model with the traditional model.

[0199] Improvement of task completion by the new formation model: Assuming that the enemy has an air defense weapon and there are ten target tasks to be achieved, the improved NSGA-Ⅱ method is used for planning and solving. The new formation model and the traditional formation model are used for simulation experiments. The specific scenario is shown in the above scenario 2. Different numbers of drones are selected in each round of simulation. The experiment is repeated 10 times, and the number of completed tasks is counted and the average value is taken. The results are as follows: Fig.18 As shown in the figure, with the increase in the number of drones, the number of tasks that can be accomplished by drones in both modes has increased, but the increase in the task completion rate of the drone group under the new formation model is significantly faster than that of the traditional formation model. In addition, the performance difference between the two models is closely related to the number of drones. When the number of drones is large or small, the difference between the two is not large; the gain in task completion rate is most obvious when the number is moderate. This is because the new formation model can plan more drones when dealing with drone damage, which can better ensure the completion of tasks; when the number of drones is small or large, the cluster planning faces too few or too many resources, and it is difficult to reflect the advantages of the formation model in resource planning, but overall, the new formation model can effectively improve the task completion rate.

[0200] Improvement of the total flight range by the new formation model under the premise of ensuring mission completion: To further test the performance of the new formation model on the key data of total flight range, this embodiment designs mission scenarios with different numbers of missions, randomly changes the positions of mission target points in the scenarios, controls the number of drones to be 10, conducts repeated experiments, and statistically analyzes the experimental results of the two models in ensuring that the mission completion rate is higher than 80%, and calculates the average value of the total flight track. The results are as follows: Fig.19 As shown: Regardless of the scenario with a small number of targets or a large number of targets, the new formation model has a significant improvement in the total navigation distance because the leader aircraft has a wider communication range, has more options for dealing with drone damage, and can better mobilize the superior drones in the cluster to take over the mission.

[0201] Conclusion, the embodiment reflects the realization of UAV mission planning in a combat environment. First, according to the combat assumptions, the mission planning problem is abstracted into a mathematical problem based on VPR, and the goals and constraints of the problem are sorted out, and a multi-objective constraint model is established for solution. In terms of model solution, this embodiment is based on NSGA-Ⅱ, combined with hypernetwork modeling, and a new N2SGA-Ⅱ method is designed. Three scenarios are designed according to the relative task volume, and compared with other three popular modern optimization methods: traditional NSGA-Ⅱ, MOEAD, and particle swarm method. The results show that the improved NSGA-Ⅱ method is more efficient and has better robustness, and can obtain a better solution set. The calculation of congestion in the NSGA-Ⅱ method is essentially to measure the quality of a solution and evaluate the priority of the solution in the cluster accordingly. The purpose is to obtain a broad, uniform and high-quality Pareto solution set surface. The introduction of hypernetwork to model the population provides a new evaluation method, such as based on node centrality, betweenness, core, etc., considering the dynamic real-time planning problem under combat conditions, that is, the reallocation of cluster tasks after the UAV is damaged under limited communication, and proposes a new formation model to address the shortcomings of traditional UAV formations. The results show that the new formation model has a significant improvement on the overall effectiveness of UAV cluster combat, can improve the upper limit of UAV task completion to a certain extent, reduce the loss caused by formation navigation, and complete tasks more efficiently.

[0202] According to another aspect of the present invention, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any one of the above methods when executing the program.

[0203] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0204] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A UAV swarm reconnaissance mission planning method based on a hierarchical model is characterized by: The following steps are involved: Collect battlefield situations in advance, carry out pre-task planning based on the pre-collected battlefield situations, and each drone in the drone cluster executes the task according to the pre-task planning; During the mission execution, if a new battlefield situation arises and the pre-planned mission plan is no longer applicable, dynamic mission planning will be triggered; After the dynamic mission planning is triggered, the alliance is formed, and the drone cluster forms a new formation model according to the hierarchical model, collects new battlefield situations, and re-plans the mission for the drone cluster; In the new formation model, the drone cluster is divided into two types of drones, including reconnaissance drones and mission drones; reconnaissance drones are responsible for monitoring the battlefield environment, conducting situation fusion and mission planning, and acting as a communication hub to establish a communication network with mission drones, collect and process information uploaded by mission drones; mission drones are responsible for enemy reconnaissance, performing tasks according to pre-mission plans or tasks and tracks planned by reconnaissance drones, and providing real-time feedback on mission completion to reconnaissance drones.

2. The method for planning a UAV cluster reconnaissance mission based on a hierarchical model as claimed in claim 1, characterized in that: Pre-mission planning includes the following steps: A combat environment parameter model is established based on the battlefield situation collected in advance. The combat environment parameter model has several restrictions, including at least no-fly zones and weapon threats. The combat parameter model avoids the restrictions to plan the trajectory of the drone cluster. Then, a multi-objective constraint model is established according to the constraints existing in the task execution process, and the task allocation is performed based on the multi-objective constraint model. The constraints include the UAV self-constraints and task constraints. The UAV self-constraints are that the total fuel consumption of a single UAV must be less than the maximum fuel capacity when the UAV cluster performs the task. The task constraint is that when the UAV cluster performs the task, each task point is completed by a certain UAV from the starting point. The pre-task planning is formed by coupling the trajectory planning and the task allocation. If there is no restriction between the two task targets in the process, the straight line between the two task targets is taken as the trajectory. If there is a restriction between the two task targets, the trajectory is corrected and coupled with the task allocation. When correcting the trajectory, for the case where the straight track passes through the no-fly zone, a point is selected outside the no-fly zone so that the lines connecting the point with the starting point and the end point are tangent to the no-fly zone, and the two corrected trajectories are taken as the trajectory between the two task targets.

3. The method for planning a UAV cluster reconnaissance mission based on a hierarchical model as claimed in claim 2, characterized in that: The multi-objective constraint model is as follows: Among them, P1 and P2 are the no-fly zone sets, n = cu{0,n+1}, Q is the full fuel load of the drone, C ij is the distance between tasks i and j, x ij Indicates whether the arc from task i to j is selected, is the track correction operator, is the vertical distance from the center of the no-fly zone n to the line connecting the two mission points i and j, r P is the radius of the no-fly zone P; is the constraint of the drone itself; k is the number of drones, r n is the radius of the no-fly zone n; The task constraints are expressed as follows: Among them, K represents the maximum number of arcs from the starting point, x ik 、x kj Respectively indicate whether the outgoing arc and incoming arc of task point k are selected, x oj Indicates whether the arc from the starting point o to other task points is selected; The total path distance is expressed as: x ij ∈{0,1}， Where C ij Represents the distance between two target points, x ij Indicates whether the route from task i to j is selected; The total path risk is expressed as: Where f ij represents the quantitative risk index between two target points, M is the air defense weapon set, r P is the radius of the no-fly zone p, r m is the radius of the mth air defense zone, is the vertical distance from the air defense zone m to the line connecting the two mission points i and j, and o is the unit fuel consumption.

4. The method for planning a UAV cluster reconnaissance mission based on a hierarchical model as claimed in claim 3, characterized in that: In the multi-objective constraint model, all drones in the drone cluster are divided into different levels. There is no domination between drones at the same level. High-level drones dominate low-level drones. Each drone i is set to have parameters n(i) and s(i), where n(i) is the number of solutions that dominate drone i in the drone cluster, and s(i) is the set of solutions dominated by drone i. Then follow the steps below to perform non-dominated sorting of drones: S001. Store all drones with n(i)=0 in the drone cluster into set F1, and assign the same non-dominated order i (rank) to the drones in the set; S002, the set of statistically dominant drones S = {S i |i∈F1}; S003, reduce n(i) of all drones in set S by 1; S004. Repeat steps S001 to S003 until all drones are classified.

5. The method for planning a UAV cluster reconnaissance mission based on a hierarchical model as claimed in claim 4, characterized in that: When all drones are classified, the hierarchical relationship within the drone cluster is obtained. For drones in each level, the drones in this level and the drones in the previous level form a drone sub-population. Each drone in the drone cluster represents an individual in the population. The two drones with the closest distance are selected to establish an intra-layer or inter-layer relationship. The above operations are repeated until all drones complete the establishment of intra-layer or inter-layer relationships, and the drone cluster super network model is obtained.

6. The method for planning a UAV cluster reconnaissance mission based on a hierarchical model as claimed in claim 5, characterized in that: The congestion degree is calculated based on the obtained drone cluster super network model. The congestion degree is calculated in the following way: Among them, f m is the objective function of the mth dimension, rank is the level corresponding to the drone, and the weights of distance calculation between different levels are assigned by the level difference. i-1 and i+1 correspond to the two nodes closest to drone i in the drone sub-population composed of the current level and the previous level, respectively.

7. The method for planning a UAV cluster reconnaissance mission based on a hierarchical model as claimed in claim 6, characterized in that: The genetic operators of the drone swarm hypernetwork model are as follows: Selection operator: Perform non-dominated sorting on the population after the merger of the father and son generations, add individuals from each layer in turn, and select several individuals with the lowest crowding degree from the critical layer to be retained for the next generation; Crossover operator: Perform a two-point crossover operation on individuals, and then modify the genes outside the crossover area according to the mapping relationship of each gene in the crossover area to eliminate conflicts; Mutation operator: The random exchange operator is used for mutation operation, and two genes in an individual are randomly selected for exchange to obtain the mutated individual.

8. The method for planning a UAV cluster reconnaissance mission based on a hierarchical model as claimed in claim 4, characterized in that: When dynamic mission planning is triggered, a new drone formation is established. The new drone formation is a three-dimensional drone formation, and then an alliance is formed; Alliance formation includes the following steps: Tender Release: When UAV U i When the drone is destroyed, the drone cluster enters the bidding release state. After the reconnaissance drone obtains the information, it becomes the leader of the alliance and establishes a local communication network with the UAVs in the neighborhood to release the bidding information P i : Pi={U i ,T}; Where U i It represents the basic information of the UAV, and T represents the remaining mission objectives of the UAV; Bidding application: After the bidding is released, all drones in the local communication network receive the bidding information. The drones first judge the bidding information. If a drone meets the conditions in the multi-objective constraint model, it becomes a bidding drone and then sends a bid to the alliance leader. Bi={U i ,T i }; Where Ti represents the target to be auctioned; Alliance construction: After the bidding application, the alliance leader receives the bids from the bidding drones in the communication network. The alliance leader needs to sort the bids, select the drone with the highest performance, and publish the winning bid information. The sorting of the bids takes into account the risk and distance cost. The alliance leader constructs an effective set according to the Pareto criterion, randomly selects a bid from the effective set and publishes the winning bid information F i Feedback to all bidding drones, F i Can be represented by a binary group; F i ={S,TR}; Where S represents the updated task sequence, and TR represents the updated track; Execution phase: The successful UAV will update its mission sequence and trajectory to complete dynamic mission planning.

9. The method for planning a UAV cluster reconnaissance mission based on a hierarchical model as claimed in claim 8, characterized in that: During the alliance formation process, the reconnaissance drone trajectory planning method is as follows: Collect relevant information: The reconnaissance drone's track is planned in sections, and the course is changed at fixed intervals. The fixed course and speed are maintained during the intervals. If the course needs to be changed, the reconnaissance drone will request relevant information from all drones within the communication range. Calculate heading: Use the artificial potential field method to plan the trajectory of the reconnaissance drone. The specific steps are as follows: Establish the gravitational field function: where d(p, p i ) represents the distance between mission UAV i and the reconnaissance aircraft, and r i represents the risk index of mission UAV i. When d < d*, the magnitude of the gravitational potential energy is proportional to the square of the distance from the current position to the target position; when d > d*, the value of the gravitational calculation function is reduced to avoid the problem of excessive gravity when far from the target position. d* represents the communication radius of the reconnaissance UAV. Establish a repulsive field function: Where d(p,p i ) represents the distance between the reconnaissance drone itself and other reconnaissance drones i, α is the repulsion gain constant; q* is the threshold range of the repulsion force, within which the repulsion force will be generated; Calculate heading based on artificial force field: Superimposing the gravitational field and the repulsive field, we get the artificial force field U: U(p)=U att (q)+U rep (q); Find the gradient of the artificial potential field and get the potential field force F: Decompose the potential field force into two directions, X and Y, and normalize them to get the heading; Execution of the voyage: According to the obtained heading, keep flying at a certain speed.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 9 is implemented.

Images