A large-scale unmanned aerial vehicle reconnaissance and attack cooperative task planning method based on divide-and-conquer idea

Abstract

A large-scale unmanned aerial vehicle reconnaissance and attack cooperative task planning method based on the divide and conquer idea belongs to the field of unmanned aerial vehicle task planning. The present application is based on the acquired composite task set of each target point, the configuration information of the heterogeneous unmanned aerial vehicle type and quantity, and the base position and capacity information, based on the time sequence weighted fuzzy C means clustering decomposition method, and the weighted distance measurement of time sequence correlation and geographical distance are fused; the initial task allocation scheme is obtained. Based on the initial task allocation scheme, a two-stage cooperative optimization framework is iterated; stage one generates candidate schemes through three disturbance operators to avoid cycles by tabu table, calls stage two to solve the optimal sequence and calculates the completion time, and updates the scheme according to the probability rule; stage two constructs an initial sequence based on the comprehensive score, generates candidates by using six neighborhood operators, and iteratively outputs the local optimal sequence after constraint inspection and completion time calculation; the two stages are alternately converged, and the global optimal reconnaissance and attack cooperative task planning scheme is output.

Description

Technical Field

[0001] This invention relates to a large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach, belonging to the field of unmanned mission planning. Background Technology

[0002] This invention belongs to the field of unmanned aerial vehicle (UAV) mission planning technology, specifically relating to a collaborative mission planning method for large-scale heterogeneous UAV swarms performing various tasks such as reconnaissance, strike, and assessment. With the maturity of UAV technology, the application of UAV swarms in complex mission scenarios such as reconnaissance and strike is becoming increasingly widespread. How to efficiently and rationally plan mission sequences for large-scale UAV swarms to minimize mission completion time and satisfy multiple constraints is a typical NP-hard combinatorial optimization problem.

[0003] Existing technical solutions can be mainly divided into the following three categories:

[0004] (1) Methods based on exact optimization models. These methods construct the task planning problem into mathematical models such as mixed integer programming (MIP) and dynamic programming (DP), and use commercial solvers such as Gurobi and CPLEX for exact solution.

[0005] These methods can obtain theoretically optimal solutions for small-scale problems, but when the scale of drones and missions increases, their computational complexity grows exponentially, making it difficult to find feasible solutions within a finite time.

[0006] (2) Methods based on classical heuristic algorithms: In order to overcome the shortcomings of poor scalability of exact algorithms, researchers have widely adopted metaheuristic algorithms such as genetic algorithm (GA), particle swarm optimization (PSO), and ant colony optimization (ACO).

[0007] These methods improve the solution speed, but they usually rely heavily on empirical parameter settings. When solving large-scale, strongly constrained reconnaissance and strike tasks, they suffer from insufficient convergence stability and are prone to getting trapped in local optima, making it difficult to guarantee the quality of the solution.

[0008] (3) Decomposition framework method based on the idea of ​​"divide and conquer": Some studies have attempted to reduce complexity by decomposing problems. However, the sub-problem division and coordination mechanism of such methods are often relatively simple, ignoring the complex temporal dependencies between tasks, such as the strict order of "reconnaissance-strike-assessment" and load balancing among UAVs, resulting in poor global coordination after decomposition and limited improvement in overall system performance. Summary of the Invention

[0009] The purpose of this invention is to provide a large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach. This method, known as the Divide-and-Conquerand Iterative Optimization based Mission Planning Method (DCIOMP), decomposes the original large-scale planning problem into several solvable subproblems. Each subproblem corresponds to a sub-task set, significantly reducing the problem size and computational complexity. The iterative optimization mechanism enables coordination and information feedback between subproblems at the global level, ensuring that task allocation and path planning results gradually converge under global constraints. This invention fully combines the global nature of centralized optimization with the efficiency of distributed computing to achieve hierarchical solution and dynamic collaboration of complex systems, thereby obtaining the optimal scheme for reconnaissance and strike collaborative mission planning of large-scale heterogeneous UAV swarms within a reasonable timeframe.

[0010] The objective of this invention is achieved through the following technical solution:

[0011] This invention discloses a large-scale UAV reconnaissance and strike cooperative mission planning method based on the divide-and-conquer approach, comprising the following steps:

[0012] Step 1: Based on the target point task decomposition and task sample generation method, the composite task set contained in each target point is decomposed into task samples with a single task as the basic unit; based on the obtained composite task set of each target point, heterogeneous UAV type and quantity configuration information, and base location and capacity information, the weighted distance metric of time-series association and geographical distance is integrated using the time-series weighted fuzzy C-means clustering decomposition method; the initial task allocation scheme is obtained by using the defuzzification and initial allocation scheme generation method.

[0013] Step Two: Based on the initial task allocation scheme, an iterative two-stage collaborative optimization framework is adopted. Phase One generates candidate schemes using three perturbation operators, avoids loops with a tabu list, and calls Phase Two to solve for the optimal sequence machine-by-machine, summarizing completion times and updating the scheme according to probability rules. Phase Two constructs an initial sequence based on the comprehensive score, generates candidates using six neighborhood operators, and iteratively outputs the locally optimal sequence after constraints such as range, time window, and time sequence dependency, and calculation of completion time. The two phases converge alternately, outputting the globally optimal reconnaissance and strike collaborative task planning scheme. The three perturbation operators are the tabu operator, the transition operator, and the exchange operator. The six neighborhood operators are random exchange, greedy removal-reinsertion, 2-opt inversion, cross-segment insertion, insertion of unscheduled tasks, and exchange of unscheduled tasks.

[0014] It also includes step three, which uses the globally optimal reconnaissance and strike coordination task planning scheme output in step two to improve the accuracy and efficiency of reconnaissance and strike coordination in large-scale heterogeneous UAV swarms.

[0015] Furthermore, the method for target point task decomposition and task sample generation described in step one is as follows:

[0016] The set of composite tasks contained at each target point is split into task samples with a single task as the basic unit.

[0017] After expanding the task sets for each target point sequentially, the total number of task samples is obtained:

[0018] (1)

[0019] in, This refers to the total number of target points within the mission area.

[0020] Furthermore, the implementation method of the weighted distance metric that integrates temporal correlation and geographic distance described in step one is as follows:

[0021] Geographic coordinate attributes and temporal logical identifier attributes are extracted from the task samples to construct a weighted comprehensive distance metric between the task samples:

[0022] (2)

[0023] in, The geographic distance is obtained through the task's geographic coordinates attribute; The timing logic distance is obtained through the timing logic identifier attribute.

[0024] . That is, the time-series correlation weighting factor, when It degenerates into pure geographical clustering; when At that time, the temporal logical associations can affect the clustering results.

[0025] Furthermore, the time-weighted fuzzy C-means clustering decomposition method described in step one is implemented using the TS-FCM clustering iterative method.

[0026] The TS-FCM clustering iterative method is as follows:

[0027] Based on distance metric Run TS-FCM clustering iterations, the goal of which is to... The task samples are divided into A fuzzy cluster is formed such that the sum of the weighted distances from the task samples within the cluster to the cluster center is minimized.

[0028] Upon first execution, the membership matrix is ​​randomly initialized. , dimension ,in Represents task samples Belonging to cluster The degree to which matrix U satisfies:

[0029] (3)

[0030] in, This indicates that the condition holds true for all task samples i, and [0,1] represents the range of membership levels, where 0 indicates no membership at all and 1 indicates full membership.

[0031] Through the geographic coordinate attributes of (2) (1) The total number of task samples and the current membership matrix Update cluster center:

[0032] (4)

[0033] in, It is a fuzzy index.

[0034] For each task sample and each cluster Weighted composite distance from task sample to cluster center for:

[0035] (5)

[0036] in, For task samples Geographic coordinates to cluster center The Euclidean distance; For task samples To the cluster center The timing logic distance is calculated using the following formula:

[0037] (6)

[0038] Update the membership matrix. Use the standard fuzzy C-means algorithm, combined with weighted comprehensive distance. Update the membership degree:

[0039] (7)

[0040] in, For task samples Belonging to cluster The membership degree satisfies ; This represents the number of clusters, and this value is equal to the total number of drones.

[0041] Convergence condition:

[0042] (8)

[0043] in, It is the convergence threshold, usually set to a value of [value missing]. If the updated If equation (7) is satisfied, the iteration terminates; otherwise, equations (4) to (6) are repeated. After the iteration converges, the updated fuzzy membership matrix is ​​obtained. .

[0044] Furthermore, the method for deblurring and generating the initial allocation scheme in step one is as follows:

[0045] Based on the updated matrix Each task sample is deblurred, that is, for each task sample... Choose the cluster with the highest membership degree. ,cluster One-to-one allocation to drones The initial task allocation scheme is as follows:

[0046] (9)

[0047] The total number of drones is Among them, reconnaissance drones Frame, strike drone shelf, to meet For each drone Give the set of tasks assigned to it. The allocation must satisfy base capacity constraints and mission-drone type matching constraints.

[0048] Furthermore, the base capacity constraint and the mission-UAV type matching constraint are as follows:

[0049] Node set ,in For drone base assembly, ; For a set of tasks, ,in For reconnaissance mission assembly, For the purpose of combating the mission, To evaluate the task set, let the set of drones be... ,in For the reconnaissance drone assembly, To combat drone swarms, a 0-1 decision variable is introduced. If and only if drone When flying directly from node i to node j, the value is 1, where .

[0050] The base capacity constraint is:

[0051] (10)

[0052] in, For drone base The number of reconnaissance drones; For drone base The number of drones used for strikes.

[0053] The task-drone type matching constraint is:

[0054] Combating drones that do not perform reconnaissance and assessment missions:

[0055] (11)

[0056] (12)

[0057] Reconnaissance drones do not perform strike missions:

[0058] (13)

[0059] (14)

[0060] All strike missions must be carried out by strike drones:

[0061] (15)

[0062] in, This represents the total number of strike missions.

[0063] Furthermore, the parameter setting method is as follows:

[0064] ① Initial temperature ,in The maximum target increment of the initial random perturbation. Let the initial acceptance probability satisfy 0.80. Initialize the current temperature .

[0065] ② The temperature decay factor satisfies .

[0066] ③ This is the preset termination temperature.

[0067] ④ Initialize the round improvement flag, let Initialize the unimproved round counter to satisfy... .

[0068] ⑤K is the preset outer loop control judgment threshold, let K=5 or K=10.

[0069] ⑥ The maximum number of iterations in the inner loop, satisfying... Inner loop counter .

[0070] ⑦ Maximum length of the taboo table .

[0071] ⑧ For each drone Perform temperature updates and calculate the initial mission sequence for each UAV. and completion time ; Make the globally optimal task sequence Global current target value Global optimal objective value .

[0072] Furthermore, the outer loop control method is as follows:

[0073] S1. Determine whether any of the following termination conditions are met.

[0074] Condition 1 (Temperature Condition): The current temperature drops to the preset termination temperature. The following is:

[0075] (16)

[0076] Condition 2 (Solution Quality Condition): After each inner cycle is completed, improve the indicator according to the round number. Update the unimproved round counter:

[0077] ·like This indicates that a better solution was found in this round of temperature iteration, and the temperature was reset. ;

[0078] ·like This indicates that the global optimal solution was not improved in this round of temperature iteration. .

[0079] The unimproved round counter reaches the threshold, i.e.: This indicates that the algorithm failed to find a better solution in the Kth round of the outer loop and is considered to have converged to the vicinity of the global optimum.

[0080] If any of the above conditions are met, the outer loop exits, the entire optimization process terminates, and the globally optimal solution is output. Otherwise, the inner loop counter is reset. =1, execute the inner loop.

[0081] S2. Inner Loop

[0082] when At that time, the execution phase one: balanced allocation of tasks among multiple drones; when At that time, perform a temperature update;

[0083] S2.1 Phase 1: Balanced task allocation among multiple drones

[0084] S2.1.1. Generate candidate solutions

[0085] S2.1.1.1. Randomly select a perturbation operator to perform task transfer. During each perturbation, one of the following three dedicated perturbation operators is randomly selected for execution. All perturbation operations are pre-checked for the task-UAV type matching constraints before execution to ensure that a reconnaissance task is not assigned to a strike UAV or vice versa.

[0086] • Taboo operator: Randomly select task i and its current drone. Search for the task in the taboo list. The set of drones that have not received this type of task in this iteration. Randomly select one from them Then task i will be transferred from... Transfer to .

[0087] • Transfer operator: Randomly select a drone From its task set Randomly select task i; then randomly select another drone capable of performing task i. ( ), will the task from Transfer to .

[0088] • Swap operator: Randomly select two drones and ( ), respectively from their task sets , Randomly select task i and task j from each; pre-check bidirectional type matching constraints (task i can be...). Execution, and task j can be... Once the task is executed and satisfied, the two tasks will be swapped.

[0089] S2.1.1.2. Use a taboo list to manage task transfer records. After any of the above-mentioned disturbance operations are executed, add the task type (UAV) involved in the operation to the taboo list; if the taboo list reaches its maximum length... If the first element to enter is removed, then the first element to enter will be removed.

[0090] After applying the above perturbation to the initial task allocation scheme or the current task allocation scheme, candidate task allocation schemes are obtained. Proceed to the next step.

[0091] S2.1.2. Calculate the completion time of candidate solutions

[0092] for Each drone in Phase Two: Single UAV Task Chain Optimization, calculating the optimal task sequence for the UAV. and precise task completion time The optimal mission sequences of all UAVs are merged to obtain the complete global mission sequence. Calculate the global target value for all drones:

[0093] (17)

[0094] S2.1.3 Update Plan

[0095] Calculate the difference in the objective function:

[0096] (18)

[0097] like Then, accept the candidate solution unconditionally and update the current solution. Current target value ;like Synchronously update the globally optimal objective solution , , ;like Then based on probability Accept the inferior solution.

[0098] S2.2. Update the inner loop counter

[0099] Update the inner loop counter = Return to S2 and continue the next inner loop.

[0100] S3. Temperature Update

[0101] Reset round improvement flag Lower the temperature according to the annealing plan:

[0102] (19)

[0103] Updated temperature Assign to Return to S1 and continue the outer loop for the next iteration.

[0104] Furthermore, the method for optimizing the single UAV mission chain in stage two is as follows:

[0105] For each UAV currently assigned a set of tasks, deep local optimization is performed with the goal of generating the optimal task execution sequence that minimizes task completion time while satisfying all constraints. and its precise task completion time The specific implementation method is as follows:

[0106] S1. Initially sort the set of single-UAV missions based on their scores.

[0107] For current drones Each task to be assigned Calculate a comprehensive score:

[0108] Among them, five indicators They are respectively:

[0109] The closer the mission is to the base, the higher the score.

[0110] The duration of the task time window; the wider the window, the higher the score.

[0111] The urgency of the task is indicated by a smaller lower bound of the time window, which indicates greater urgency and a higher score.

[0112] Geographical advantages, such as being located in the center of the region, are preset according to the scenario;

[0113] Task reward value: Tasks with high-value objectives receive higher scores.

[0114] Initialize the current sequence S={}, weights The task set is selected in descending order of comprehensive score. The task in the sequence is inserted into the current sequence. The initial feasible task sequence is obtained by inserting tasks into positions that minimize the increase in total completion time while satisfying the constraints, until all tasks have been inserted. and the task completion time .

[0115] S2. Perform depth sorting of the single UAV mission sequence using a neighborhood iterative search algorithm.

[0116] Based on the current neighborhood operator The transformation rule iterates through all possible neighborhood moves to generate a set of candidate sequences. Among them, the current neighborhood operator index during the first run. The six neighborhood operators are described in detail below:

[0117] (Random Swap): Randomly select two tasks at different positions in the sequence, swap their positions, and keep the sequence length unchanged to obtain... ;

[0118] ② (Greedy Removal-Reinsertion): Remove each task from its current position in the sequence and try to reinsert it into each other feasible position in the sequence, choosing the reinsertion method that reduces the objective function the most.

[0119] ③ (2-opt reverse): Select any two positions in the sequence and reverse the entire sub-path between the two positions;

[0120] ④ (Inter-segment insertion): Removes a continuous segment of tasks from the sequence and attempts to insert it into another position in the sequence;

[0121] ⑤ (Insert unscheduled task): If there are tasks that have not been included in the current sequence due to constraint conflicts, try to insert them into each feasible position in the sequence;

[0122] ⑥ (Swap Unscheduled Tasks): Select a low-yield task in the sequence and attempt to swap its position with a high-yield task in the set of unscheduled tasks, provided that the sequence still satisfies the constraints after the swap.

[0123] right Each candidate sequence in Perform the following assessments in sequence:

[0124] S2.1. Feasibility Check

[0125] verify Do they satisfy the following conditions simultaneously:

[0126] • The range constraint: the sum of the flight distances between all adjacent nodes in the sequence. .in, For drones The sequence of flight path nodes, including base and mission nodes. Let be the flight distance from node i to node j; For drones Maximum range;

[0127] • The time window constraint: calculates the start time of each task. ,satisfy .in, This is the lower bound of the time window for task n; This is the upper bound of the time window for task n; This represents the actual start time of task n;

[0128] • The temporal dependency constraint between tasks: The reconnaissance → strike → evaluation chain of tasks at the same target point must appear in a strictly logical order. Let the target point be... The reconnaissance mission on the ground began at the following time. The end time is The strike mission will begin at [time]. The end time is The assessment task start time is .

[0129] The reconnaissance mission shall end no later than the strike mission begins.

[0130] (20)

[0131] The strike mission shall end no later than the assessment mission start time.

[0132] (twenty one)

[0133] in, ; ; .

[0134] For target locations that only involve strike and assessment missions, the assessment mission start time must not be earlier than the strike mission end time.

[0135] (twenty two)

[0136] in, .

[0137] like If any of the above constraints are not met, discard the candidate and continue the check. The next candidate is selected; if none of the candidate sets satisfy the above constraints, the neighborhood operator index is incremented by one, and then... Return to S2 to continue the iterative search.

[0138] S2.2. Calculation of the objective function

[0139] Candidates that pass the feasibility check The mission completion time is calculated according to formula (23), which is the sum of flight time and mission execution time:

[0140] (twenty three)

[0141] in, For drones The sequence of flight path nodes, including base and mission nodes; Let be the flight distance from node i to node j; For drones Flight speed; Let n be the execution time of task n.

[0142] S3. Update the candidate task sequence for a single drone and switch the neighborhood operator.

[0143] In operators Among all feasible candidate tasks, traverse All feasible candidates Calculate the completion time of each candidate according to formula (23). ;Pick Minimum candidate As Calculate the improvement amount Determine the current operator The search results. Among them, The task completion time for the initial feasible task sequence S;

[0144] If an improved solution is found :accept As the new current solution, let , And reset the neighborhood operator index to Return to S2 to continue the iterative search;

[0145] If no improved solution is found, i.e., the current solution... All candidates within the neighborhood fail to decrease f(S): Increment the neighborhood operator index by one, let Return to S2 to continue the iterative search.

[0146] S4. Output the local optimal task sequence for a single UAV.

[0147] When the neighborhood operator index When all six neighborhood operators are exceeded, it indicates that the current solution cannot be further improved, and the algorithm terminates. At this point, the locally optimal task sequence for the UAV under the current task assignment is output. and its precise task completion time .

[0148] Beneficial effects:

[0149] A Large-Scale Unmanned Aerial Vehicle (UAV) Reconnaissance and Strike Cooperative Mission Planning Method Based on Divide and Conquer and Iterative Optimization

[0150] 1. Compared to directly applying genetic algorithms and particle swarm optimization, this invention achieves a better balance between efficiency and quality. The invention discloses a large-scale UAV reconnaissance and strike collaborative task planning method based on the divide-and-conquer approach. This strategy significantly reduces the solution scale for each optimization, making local VND search more efficient. Simultaneously, the global exploration capability of the SATL framework avoids the pitfalls of traditional heuristic algorithms that easily get trapped in local optima. This invention (SATL-VND) outperforms the comparative algorithms in both convergence speed and final solution quality.

[0151] 2. Compared to existing simple divide-and-conquer methods, this invention offers stronger collaborative processing capabilities for complex constraints. The large-scale UAV reconnaissance and strike collaborative task planning method disclosed in this invention, based on the divide-and-conquer approach, considers task logic constraints through time-weighted clustering during the decomposition phase and strictly guarantees constraints such as time windows and task sequences through VND during the optimization phase. This mechanism of "pre-coordination during decomposition and strong constraints during optimization" addresses the problem of weak global collaboration in existing methods, generating highly feasible planning schemes.

[0152] 3. Improved adaptability and robustness: This invention discloses a large-scale UAV reconnaissance and strike collaborative task planning method based on the divide-and-conquer approach. It integrates multiple optimization strategies such as SA, TS, and VND, and enhances adaptability to different scales and distributions of task scenarios by designing various targeted neighborhood operators. This invention can stably output high-quality solutions and has advantages such as low fluctuation and good robustness.

[0153] Attached Description

[0154] Figure 1 A flowchart of a large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach of the present invention.

[0155] Figure 2 The detailed workflow and data interaction diagram of the two-stage algorithm of this invention. Detailed Implementation

[0156] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0157] Example: Large-scale reconnaissance and strike coordination mission planning in border areas

[0158] Taking a joint strike mission in a border area as an example, a total of 10 target points were set up within the mission area. Each target point includes three types of composite tasks: reconnaissance, strike, and assessment, totaling 30 task samples. Two drone bases are deployed. Equipped with 6 heterogeneous drones: 3 reconnaissance drones Maximum range ,speed 3 strike drones Maximum range ,speed The geographical coordinates of each target point are distributed across three sectors: east, south, and west of the base. (Mission time window) Set as minute.

[0159] like Figure 1 As shown in the figure, this embodiment discloses a large-scale UAV reconnaissance and strike cooperative mission planning method based on the divide-and-conquer approach. The specific implementation steps are as follows:

[0160] Step 1 (Problem Breakdown):

[0161] Based on S1.1, the composite task for each target point is broken down into 30 independent task samples, and geographic coordinates and temporal logical identifiers are extracted. A weighted comprehensive distance metric is constructed according to S1.2, and... The temporal logical distance is assigned as follows: 0 for direct pre- and post-order, 0.3 for indirect temporal, 0.5 for the same type, and 1 for different target points. Run TS-FCM clustering in S1.3, with... (Total number of drones) is the number of clusters iterated until convergence. ), to obtain the fuzzy membership matrix After defuzzing in S1.4, an initial allocation scheme is generated, satisfying the base capacity constraint (3 aircraft deployed from each base) and the mission-UAV type matching constraint. Reconnaissance drones They are respectively responsible for reconnaissance and assessment missions in the eastern, southern, and western sectors; and combating drones. To undertake strike missions in the corresponding sectors.

[0162] Step Two Execution (Two-Phase Collaborative Optimization):

[0163] Initial temperature (Depend on (Calculation), attenuation factor Termination temperature Inner loop Taboo table length .

[0164] Phase 1 (S2.2.1.1): Based on the current scheme, candidate schemes are generated through tabu, transition, and exchange operators. For example, will A reconnaissance mission in the mission sequence was transferred to (Pre-check to ensure type matching), update the taboo table.

[0165] Phase Two (S2.3): For Each drone in the process utilizes individual optimization. For example, the initial sequence of its task set after sorting by comprehensive score is as follows: Six neighborhood operators are executed sequentially: exchange and Location; Greedy re-insertion After the strike mission, a range constraint check (total path length) was performed. km), time window constraints (start time of each task) After verifying the time-series dependency constraints of reconnaissance → strike → assessment, the completion time is calculated. Minutes. Iterate through all candidates and select... make Reduced to 172 minutes, accepted the improved solution and jumped back to .

[0166] The global solution is obtained by summarizing the optimal sequence of six drones. ,calculate Minutes. Compared to the current target value. Minute comparison, Unconditional acceptance; synchronized updates , .

[0167] Temperature decay after outer circulation and unimproved cycle control ( The algorithm will run as the temperature drops to... The following will terminate.

[0168] Final output:

[0169] Global optimal solution Includes the precise mission execution sequence of 6 drones, the start / end time of each mission, and path nodes. Total completion time. The plan reduced the time by 18 minutes compared to the initial plan, balanced the load of each UAV, satisfied all constraints, and achieved the optimal solution for planning reconnaissance and strike collaborative missions of a large-scale heterogeneous UAV swarm.

[0170] The overall solution process of this invention consists of two steps: problem decomposition, two-stage collaborative optimization, and iterative convergence.

[0171] S1. Problem Breakdown

[0172] Based on the acquired composite task set, UAV type and quantity information, and base information of each target point, an initial task allocation scheme is obtained using the time-series weighted fuzzy C-means clustering decomposition method.

[0173] This step mainly includes four sub-processes: target point task splitting and task sample generation, weighted distance measurement that integrates temporal correlation and geographical distance, TS-FCM clustering iteration, defuzzification, and initial task allocation scheme generation.

[0174] S1.1. Target Point Task Decomposition and Task Sample Generation

[0175] For the target point Its task set is denoted as ,in This indicates the number of tasks at that target point. The set... Each task in Expand into an independent task sample, the attributes of which include: geographic coordinates, task type code, time sequence logical identifier, and task time.

[0176] After expanding the task sets for each target point sequentially, the total number of task samples is obtained:

[0177] (1)

[0178] in, This refers to the total number of target points within the mission area.

[0179] S1.2. Weighted distance metric that integrates temporal correlation and geographic distance

[0180] Geographic coordinate attributes and temporal logical identifier attributes are extracted from the task samples to construct a weighted comprehensive distance metric between the task samples:

[0181] (2)

[0182] in, The geographic distance is obtained through the task's geographic coordinates attribute; The timing logic distance is obtained through the timing logic identifier attribute. The rule for determining the value is: for the same target point, the direct sequential relationship is taken as follows: Indirect temporal relationships at the same target point are taken For the same target point and the same type of task, take Different target points .

[0183] . This is the time-series correlation weighting factor. When It degenerates into pure geographical clustering; when At that time, temporal logical relationships can affect the clustering results. In this invention, we typically take... , .

[0184] S1.3.TS-FCM Clustering Iteration

[0185] Based on distance metric Run TS-FCM clustering iterations, the goal of which is to... The task samples are divided into A fuzzy cluster is formed such that the sum of the weighted distances from the task samples within the cluster to the cluster center is minimized.

[0186] Upon first execution, the membership matrix is ​​randomly initialized. , dimension ,in Represents task samples Belonging to cluster To what extent, Let U be the number of clusters, which is equal to the total number of drones. Matrix U satisfies:

[0187] (3)

[0188] in, This indicates that the condition holds true for all task samples i, and [0,1] represents the range of membership levels, where 0 indicates no membership at all and 1 indicates full membership.

[0189] Geographic coordinate attributes of S1.2 The total number of task samples in S1.1 and the current membership matrix. Update cluster center:

[0190] (4)

[0191] in, It is the fuzzy index (usually) ).

[0192] Secondly, for each task sample and each cluster Define the weighted composite distance from the task sample to the cluster center. for:

[0193] (5)

[0194] in, For task samples Geographic coordinates to cluster center The Euclidean distance; For task samples To the cluster center The timing logic distance. The calculation formula is as follows:

[0195] (6)

[0196] Indicates that it belongs to the cluster in the current iteration. The task sample set is usually determined by membership degree. The task sample, or a weighted average of all task samples.

[0197] Next, the membership matrix is ​​updated. The standard fuzzy C-means algorithm is used, combined with a weighted comprehensive distance. Update the membership degree:

[0198] (7)

[0199] in, For task samples Belonging to cluster The membership degree satisfies .

[0200] Convergence condition:

[0201] (8)

[0202] in, and Let represent the membership degree of the same task sample i to the same cluster k in two adjacent iterations, and let represent the absolute value of the difference between the two values, which reflects the magnitude of the change in membership degree. It is the convergence threshold, usually set to a value of [value missing]. If the above conditions are met, the iteration terminates; otherwise, the TS-FCM clustering iteration continues. After the iteration converges, the fuzzy membership matrix is ​​obtained. .

[0203] S1.4. Defuzzification and Initial Task Assignment Scheme Generation

[0204] Based on the updated matrix Each task sample is deblurred, that is, for each task sample... Choose the cluster with the highest membership degree. ,cluster One-to-one allocation to drones The initial task allocation scheme is as follows:

[0205] (9)

[0206] The total number of drones is Among them, reconnaissance drones Frame, strike drone shelf, to meet For each drone Give the set of tasks assigned to it. The allocation must satisfy base capacity constraints and mission-drone type matching constraints:

[0207] Let the set of nodes be ,in For drone base assembly, ; For a set of tasks, ,in For reconnaissance mission assembly, For the purpose of combating the mission, To evaluate the task set, let the set of drones be... ,in For the reconnaissance drone assembly, To combat drone swarms, a 0-1 decision variable is introduced. If and only if drone When flying directly from node i to node j, the value is 1, where .

[0208] The base capacity constraint is:

[0209] (10)

[0210] in, For drone base The number of reconnaissance drones; For drone base The number of drones used for strikes.

[0211] The task-drone type matching constraint is:

[0212] Combating drones that do not perform reconnaissance and assessment missions:

[0213] (11)

[0214] (12)

[0215] Reconnaissance drones do not perform strike missions:

[0216] (13)

[0217] (14)

[0218] All strike missions must be carried out by strike drones:

[0219] (15)

[0220] in, This represents the total number of strike missions.

[0221] S2. Step Two: Two-Stage Collaborative Optimization and Iterative Convergence

[0222] Based on the initial task allocation scheme A, this step employs a two-stage collaborative optimization framework to execute the following double-loop iterative optimization process:

[0223] S2.1. Parameter Settings:

[0224] ① Initial temperature ,in The maximum target increment of the initial random perturbation satisfies , Let the initial acceptance probability satisfy 0.80. Initialize the current temperature .

[0225] ② The temperature decay factor satisfies .

[0226] ③ The preset termination temperature is usually taken as... .

[0227] ④ Initialize the round improvement flag, let Initialize the unimproved round counter to satisfy... .

[0228] ⑤K is the preset outer loop control judgment threshold, let K=5 or K=10.

[0229] ⑥ The maximum number of iterations in the inner loop, satisfying... Inner loop counter =1.

[0230] ⑦ Maximum length of the taboo table .

[0231] ⑧ For each drone Call S2.3 to calculate the initial mission sequence for each UAV. and completion time ; Make the globally optimal task sequence Global current target value Global optimal objective value .

[0232] S2.2. Outer Loop Control

[0233] Determine whether any of the following termination conditions are met:

[0234] Condition 1 (Temperature Condition): The current temperature drops to the preset termination temperature. The following is:

[0235] (16)

[0236] Condition 2 (Solution Quality Condition): After each completion of the inner cycle of S2.2.1, improve the indicator according to the cycle. Update the unimproved round counter:

[0237] ·like This indicates that a better solution was found in this round of temperature iteration, and the temperature was reset. ;

[0238] ·like This indicates that the global optimal solution was not improved in this round of temperature iteration. .

[0239] The unimproved round counter reaches the threshold, i.e.: This indicates that the algorithm failed to find a better solution in the Kth round of the outer loop and is considered to have converged to the vicinity of the global optimum.

[0240] If any of the above conditions are met, the outer loop exits, the entire optimization process terminates, and the globally optimal solution is output. Otherwise, the inner loop counter is reset. =1, execute S2.2.1. Inner loop.

[0241] S2.2.1. Inner Loop

[0242] when S2.2.1.1. Execution Phase 1: Balanced task allocation among multiple drones.

[0243] S2.2.1.1. Execution Phase 1: Balanced Task Allocation Among Multiple Unmanned Units

[0244] S2.2.1.1.1. Generate candidate solutions

[0245] 1) Randomly select a perturbation operator to perform task transfer. During each perturbation, one of the following three dedicated perturbation operators is randomly selected for execution. All perturbation operations are pre-checked for the task-UAV type matching constraints before execution to ensure that reconnaissance tasks are not assigned to strike UAVs or strike tasks are not assigned to reconnaissance UAVs.

[0246] • Taboo operator: Randomly select task i and its current drone. Search for the task in the taboo list. The set of drones that have not received this type of task in this iteration. Randomly select one from them Then task i will be transferred from... Transfer to .

[0247] • Transfer operator: Randomly select a drone From its task set Randomly select task i; then randomly select another drone capable of performing task i. ( ), will the task from Transfer to .

[0248] • Swap operator: Randomly select two drones and ( ), respectively from their task sets , Randomly select task i and task j from each; pre-check bidirectional type matching constraints (task i can be...). Execute, and task j can be Once the task is executed and satisfied, the two tasks will be swapped.

[0249] 2) Use a taboo list to manage task transfer records. After any of the above-mentioned disturbance operations are executed, add the task type (UAV) involved in the operation to the taboo list; if the taboo list reaches its maximum length... If the first element to enter is removed, then the first element to enter will be removed.

[0250] After performing the above perturbation, a candidate task allocation scheme is obtained. Proceed to the next step.

[0251] S2.2.1.1.2. Calculate the completion time of candidate solutions

[0252] for Each drone in Call S2.3. Define Phase Two: Single UAV Task Chain Optimization, and calculate the optimal task sequence for the UAV. and precise task completion time All drones Merge to obtain the complete global task sequence And calculate the global target value:

[0253] (17)

[0254] S2.2.1.1.3 Update Scheme

[0255] Calculate the difference in the objective function:

[0256] (18)

[0257] like Then, the candidate solution will be accepted unconditionally, and the current task solution will be updated. Current target value ;like Synchronously update the globally optimal objective solution , , ;like Then based on probability Accept the worse solution.

[0258] S2.2.1.2. Update the inner loop counter

[0259] Update the inner loop counter = Return to S2.2.1 and continue the next inner loop.

[0260] S2.2.2. Temperature Update

[0261] Reset round improvement flag Lower the temperature according to the annealing plan:

[0262] (19)

[0263] Updated temperature Assign to Return to S2.2 and continue the outer loop for the next iteration.

[0264] S2.3. Definition Phase Two: Single UAV Mission Chain Optimization

[0265] For each UAV currently assigned a set of tasks, deep local optimization is performed with the goal of generating the optimal task execution sequence that minimizes task completion time while satisfying all constraints. and its precise task completion time The specific calculation process is as follows:

[0266] S2.3.1. Initially sort the set of single UAV missions based on scores.

[0267] For current drones Each task to be assigned Calculate a comprehensive score:

[0268] Among them, five indicators They are respectively:

[0269] The closer the mission is to the base, the higher the score.

[0270] The duration of the task time window; the wider the window, the higher the score.

[0271] The urgency of the task is indicated by a smaller lower bound of the time window, which indicates greater urgency and a higher score.

[0272] Geographical advantages, such as being located in the center of the region, are preset according to the scenario;

[0273] Task reward value: Tasks with high-value objectives receive higher scores.

[0274] Initialize the current sequence S={}, weights The task set is selected in descending order of comprehensive score. The task in the sequence is inserted into the current sequence. The initial feasible task sequence is obtained by inserting tasks into positions that minimize the increase in total completion time while satisfying the constraints, until all tasks have been inserted. and the task completion time .

[0275] S2.3.2. Perform depth sorting of the single UAV mission sequence using a neighborhood iterative search algorithm.

[0276] In the current neighborhood operator Next, according to the transformation rules defined by the operator, all possible neighborhood moves are traversed to generate a candidate sequence set. Among them, the current neighborhood operator index during the first run. The six neighborhood operators are described in detail below:

[0277] ① (Random Swap): Randomly select two tasks at different positions in the sequence and swap their positions while keeping the sequence length unchanged;

[0278] ② (Greedy Removal-Reinsertion): Remove each task from its current position in the sequence and try to reinsert it into each other feasible position in the sequence, choosing the reinsertion method that reduces the objective function the most.

[0279] ③ (2-opt reverse): Select any two positions in the sequence and reverse the entire sub-path between the two positions;

[0280] ④ (Inter-segment insertion): Removes a continuous segment of tasks from the sequence and attempts to insert it into another position in the sequence;

[0281] ⑤ (Insert unscheduled task): If there are tasks that have not been included in the current sequence due to constraint conflicts, try to insert them into each feasible position in the sequence;

[0282] ⑥ (Swap Unscheduled Tasks): Select a low-yield task in the sequence and attempt to swap its position with a high-yield task in the set of unscheduled tasks, provided that the sequence still satisfies the constraints after the swap.

[0283] right Each candidate sequence in Perform the following assessments in sequence:

[0284] S2.3.2.1. Feasibility Check

[0285] verify Do they satisfy the following conditions simultaneously:

[0286] • The range constraint: the sum of the flight distances between all adjacent nodes in the sequence. .in, For drones The sequence of flight path nodes, including base and mission nodes. Let be the flight distance from node i to node j; For drones Maximum range;

[0287] • The time window constraint: calculates the start time of each task. ,satisfy .in, This is the lower bound of the time window for task n; This is the upper bound of the time window for task n; The actual start time of task n;

[0288] • The temporal dependency constraint between tasks: The reconnaissance → strike → evaluation chain of tasks at the same target point must appear in a strictly logical order. Let the target point be... The reconnaissance mission on the ground began at the following time. The end time is The strike mission will begin at [time]. The end time is The assessment task start time is .

[0289] The reconnaissance mission shall end no later than the strike mission begins.

[0290] (20)

[0291] The strike mission shall end no later than the assessment mission start time.

[0292] (twenty one)

[0293] in, ; ; .

[0294] For target locations that only involve strike and assessment missions, the assessment mission start time must not be earlier than the strike mission end time.

[0295] (twenty two)

[0296] in, .

[0297] like If any of the above constraints are not met, discard the candidate and continue the check. Next candidate; if none of the candidate sets satisfy the above constraints, increment the neighborhood operator index by one, and let... Return to S2.3.2 and continue the iterative search.

[0298] S2.3.2.2. Calculation of the objective function

[0299] Candidates that pass the feasibility check The mission completion time is calculated according to formula (23), which is the sum of flight time and mission execution time:

[0300] (twenty three)

[0301] in, For drones The sequence of flight path nodes, including base and mission nodes; Let be the flight distance from node i to node j; For drones Flight speed; Let n be the execution time of task n.

[0302] S2.3.3. Update the candidate task sequence for a single UAV and the switching neighborhood operator

[0303] In operators Among all feasible candidate tasks, traverse All feasible candidates Calculate the completion time of each candidate according to formula (23). ;Pick Minimum candidate As Calculate the improvement amount Determine the current operator The search results. Among them, The task completion time for the initial feasible task sequence S:

[0304] If an improved solution is found ( ):accept As the new current solution, let , And reset the neighborhood operator index to Return to S2.3.2 and continue the iterative search;

[0305] If no improved solution is found, i.e., the current solution... All candidates within the neighborhood fail to decrease f(S): Increment the neighborhood operator index by one, let Return to S2.3.2 and continue the iterative search.

[0306] S2.3.4 Output the local optimal mission sequence for a single UAV

[0307] When the neighborhood operator index When all six neighborhood operators are exceeded, it indicates that the current solution cannot be further improved, and the algorithm terminates. At this point, the locally optimal task sequence for the UAV under the current task assignment is output. and its precise task completion time .

[0308] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach, characterized in that: Includes the following steps, Step 1: Based on the target point task decomposition and task sample generation method, the composite task set contained in each target point is decomposed into task samples with a single task as the basic unit; based on the obtained composite task set of each target point, heterogeneous UAV type and quantity configuration information, and base location and capacity information, the weighted distance metric of time-series association and geographical distance is integrated using the time-series weighted fuzzy C-means clustering decomposition method; the initial task allocation scheme is obtained by using the defuzzification and initial allocation scheme generation method. Step 2: Based on the initial task allocation scheme, a two-stage collaborative optimization framework is adopted for iteration. In Stage 1, candidate schemes are generated using three perturbation operators, loop avoidance is achieved using a tabu list, and Stage 2 is called to solve for the optimal sequence machine by machine and summarize the completion time, updating the scheme according to probability rules. In Stage 2, an initial sequence is constructed based on the comprehensive score, and candidates are generated using six neighborhood operators. After constraint verification and completion time calculation, a locally optimal sequence is iteratively output. The two stages converge alternately, outputting the globally optimal reconnaissance and strike collaborative task planning scheme. The three perturbation operators are tabu operator, transition operator, and exchange operator. The six neighborhood operators are random exchange, greedy removal-reinsertion, 2-opt inversion, cross-segment insertion, insertion of unscheduled tasks, and exchange of unscheduled tasks. The constraint verification includes flight distance, time window, and time sequence dependency.

2. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 1, characterized in that: It also includes step three, which uses the globally optimal reconnaissance and strike coordination task planning scheme output in step two to improve the accuracy and efficiency of reconnaissance and strike coordination in large-scale heterogeneous UAV swarms.

3. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 2, characterized in that: The method for target point task decomposition and task sample generation described in step one is as follows: The set of composite tasks contained at each target point is divided into task samples with a single task as the basic unit; After expanding the task sets for each target point sequentially, the total number of task samples is obtained: （1） in, This refers to the total number of target points within the mission area.

4. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 3, characterized in that: The implementation method of the weighted distance metric that integrates temporal correlation and geographic distance described in step one is as follows: Geographic coordinate attributes and temporal logical identifier attributes are extracted from the task samples to construct a weighted comprehensive distance metric between the task samples: （2） in, The geographic distance is obtained through the task's geographic coordinates attribute; The timing logic distance is obtained through the timing logic identifier attribute; That is, the time-series correlation weighting factor, when It degenerates into pure geographical clustering; when At that time, the temporal logical associations can affect the clustering results.

5. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 4, characterized in that: The time-weighted fuzzy C-means clustering decomposition method described in step one is implemented using the TS-FCM clustering iterative method. The TS-FCM clustering iterative method is as follows: Based on distance metric Run TS-FCM clustering iterations, the goal of which is to... The task samples are divided into A fuzzy cluster is formed such that the sum of the weighted distances from the task samples within the cluster to the cluster center is minimized; Upon first execution, the membership matrix is ​​randomly initialized. , dimension ,in Represents task samples Belonging to cluster The degree to which matrix U satisfies: （3） in, This indicates that the condition holds true for all task samples i, and [0,1] represents the range of membership degree, where 0 indicates no membership at all and 1 indicates full membership. Through the geographic coordinate attributes of (2) (1) The total number of task samples and the current membership matrix Update cluster center: （4） in, It is a fuzzy index; For each task sample and each cluster Weighted composite distance from task sample to cluster center for: （5） in, For task samples Geographic coordinates to cluster center The Euclidean distance; For task samples To the cluster center The timing logic distance is calculated using the following formula: （6） Update the membership matrix; use the standard fuzzy C-means algorithm combined with weighted comprehensive distance. Update the membership degree: （7） in, For task samples Belonging to cluster The membership degree satisfies ; This represents the number of clusters, and this value is equal to the total number of drones. Convergence condition: （8） in, It is the convergence threshold, usually set to a value of [value missing]. If the updated If equation (7) is satisfied, the iteration terminates; otherwise, equations (4) to (6) are repeated. After the iteration converges, the updated fuzzy membership matrix is ​​obtained. .

6. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 5, characterized in that: The method for deblurring and generating the initial allocation scheme in step one is as follows: Based on the updated matrix Each task sample is deblurred, that is, for each task sample... Choose the cluster with the highest membership degree. ,cluster One-to-one allocation to drones The initial task allocation scheme is as follows: （9） The total number of drones is Among them, reconnaissance drones Frame, strike drone shelf, to meet For each drone Give the set of tasks assigned to it. The allocation must satisfy base capacity constraints and mission-drone type matching constraints.

7. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 6, characterized in that: The base capacity constraint and the mission-UAV type matching constraint are as follows: Node set ,in For drone base assembly, ; For a set of tasks, ,in For reconnaissance mission assembly, For the purpose of combating the mission, To evaluate the task set; let the set of drones be... ,in For the reconnaissance drone assembly, To combat drone swarms; introduce 0-1 decision variables. If and only if drone When flying directly from node i to node j, the value is 1, where ; The base capacity constraint is: （10） in, For drone base The number of reconnaissance drones; For drone base The number of drones used for strikes; The task-drone type matching constraint is: Combating drones that do not perform reconnaissance and assessment missions: （11） （12） Reconnaissance drones do not perform strike missions: （13） （14） All strike missions must be carried out by strike drones: （15） in, This represents the total number of strike missions.

8. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 7, characterized in that: The parameter setting method is as follows: ① Initial temperature ,in The maximum target increment of the initial random perturbation. Let the initial acceptance probability satisfy 0.

80. Initialize the current temperature ; ② The temperature decay factor satisfies ; ③ The preset termination temperature; ④ Initialize the round improvement flag, let ; Initialize the unimproved round counter to satisfy... ; ⑤K is the preset outer loop control judgment threshold, let K=5 or K=10; ⑥ The maximum number of iterations in the inner loop, satisfying... Inner loop counter ; ⑦ Maximum length of the taboo table ; ⑧ For each drone Perform temperature updates and calculate the initial mission sequence for each UAV. and completion time ; Make the globally optimal task sequence Global current target value Global optimal objective value .

9. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 8, characterized in that: The outer loop control method is as follows: S1. Determine whether any of the following termination conditions are met. Condition 1: The current temperature drops to the preset termination temperature. The following is: （16） Condition 2: After each inner loop is completed, improve the flag according to the round number. Update the unimproved round counter: ·like This indicates that a better solution was found in this round of temperature iteration, and the temperature was reset. ; ·like This indicates that the global optimal solution was not improved in this round of temperature iteration. ; The unimproved round counter reaches the threshold, i.e.: This indicates that the algorithm failed to find a better solution in the K outer loops and is considered to have converged to the vicinity of the global optimum. If any of the above conditions are met, the outer loop exits, the entire optimization process terminates, and the globally optimal solution is output. ; Otherwise, the inner loop counter is reset. =1, execute the inner loop; S2. Inner Loop when At that time, the execution phase one: balanced allocation of tasks among multiple drones; when At that time, perform a temperature update; S2.1 Phase 1: Balanced task allocation among multiple drones S2.1.

1. Generate candidate solutions S2.1.1.

1. Randomly select a perturbation operator to perform task transfer; each time a perturbation occurs, randomly select one of the following three dedicated perturbation operators to execute; all perturbation operations are pre-checked for the task-UAV type matching constraints before execution to ensure that reconnaissance tasks are not assigned to strike UAVs or strike tasks are not assigned to reconnaissance UAVs. • Taboo operator: Randomly select task i and its current drone. Search for the task in the taboo list. The set of drones that have not received this type of task in this iteration. Randomly select one from them Then task i will be transferred from... Transfer to ; • Transfer operator: Randomly select a drone From its task set Randomly select task i; then randomly select another drone capable of performing task i. ( ), will the task from Transfer to ; • Swap operator: Randomly select two drones and ( ), respectively from their task sets , Randomly select task i and task j from each; pre-check the bidirectional type matching constraint, and swap the two tasks if the constraint is satisfied. S2.1.1.

2. Use a taboo list to manage task transfer records; after any of the above-mentioned disturbance operations are executed, add the task type-UAV pair involved in the operation to the taboo list; if the taboo list reaches its maximum length... If so, then remove the earliest element that entered the element; After applying the above perturbation to the initial task allocation scheme or the current task allocation scheme, candidate task allocation schemes are obtained. Proceed to the next step; S2.1.

2. Calculate the completion time of candidate solutions for Each drone in Phase Two: Single UAV Task Chain Optimization, calculating the optimal task sequence for the UAV. and precise task completion time Merge the optimal mission sequences of all UAVs to obtain the complete global mission sequence. Calculate the global target value for all drones: （17） S2.1.3 Update Plan Calculate the difference in the objective function: （18） like Then, accept the candidate solution unconditionally and update the current solution. Current target value ;like Synchronously update the globally optimal objective solution , , ;like Then based on probability Accept the inferior solution; S2.

2. Update the inner loop counter Update the inner loop counter = Return to S2 and continue the next inner loop; S3. Temperature Update Reset round improvement flag Lower the temperature according to the annealing plan: （19） Updated temperature Assign to Return to S1 and continue the outer loop for the next iteration.

10. The large-scale UAV reconnaissance and strike collaborative mission planning method based on the divide-and-conquer approach as described in claim 9, characterized in that: The method for optimizing the single UAV mission chain in Phase Two is as follows: For each UAV, a deep local optimization is performed on the current set of tasks assigned to it, with the goal of generating the optimal task execution sequence that minimizes task completion time while satisfying all constraints. and its precise task completion time The specific implementation method is as follows: S1. Initially sort the set of single-UAV missions based on their scores. For current drones Each task to be assigned Calculate a comprehensive score: Among them, five indicators They are respectively: The closer the mission is to the base, the higher the score. The duration of the task time window; the wider the window, the higher the score. The urgency of the task is indicated by a smaller lower bound of the time window, which indicates greater urgency and a higher score. Geographical advantages, such as being located in the center of the region, are preset according to the scenario; Task reward value: Tasks with high-value objectives receive higher scores. Initialize the current sequence S={}, weights Task sets are selected in descending order of comprehensive scores. The task in the sequence is inserted into the current sequence. The initial feasible task sequence is obtained by inserting tasks into positions that minimize the increase in total completion time while satisfying the constraints, until all tasks have been inserted. and the task completion time ; S2. Perform depth sorting of the single UAV mission sequence using a neighborhood iterative search algorithm. Based on the current neighborhood operator The transformation rule iterates through all possible neighborhood moves to generate a set of candidate sequences. Among them, the current neighborhood operator index during the first run. The six neighborhood operators are described in detail below: : Randomly select two tasks at different positions in the sequence, swap their positions, and keep the sequence length unchanged to obtain ; ② Remove each task from its current position in sequence and try to re-insert it into each other feasible position in the sequence, selecting the re-insertion method that reduces the objective function the most. ③ : Select any two positions in the sequence and reverse the entire sub-path between the two positions; ④ Remove a continuous task segment from the sequence and attempt to insert it into another position in the sequence; ⑤ If there are tasks that are not included in the current sequence due to constraint conflicts, try to insert them into each feasible position in the sequence; ⑥ Select a low-yield task in the sequence and attempt to swap its position with a high-yield task in the set of unscheduled tasks, provided that the sequence still satisfies the constraints after the swap. right Each candidate sequence in Perform the following assessments in sequence: S2.

1. Feasibility Check verify Do they satisfy the following conditions simultaneously: • The range constraint: the sum of the flight distances between all adjacent nodes in the sequence. ;in, For drones The sequence of flight path nodes, including base and mission nodes; Let be the flight distance from node i to node j; For drones Maximum range; • The time window constraint: calculates the start time of each task. ,satisfy ;in, This is the lower bound of the time window for task n; This is the upper bound of the time window for task n; The actual start time of task n; • The temporal dependency constraint between tasks: the reconnaissance → strike → evaluation chain tasks at the same target point must appear in a strictly logical order; let the target point be... The reconnaissance mission on the ground began at the following time. The end time is The strike mission will begin at [time]. The end time is The assessment task start time is ; The reconnaissance mission shall end no later than the strike mission begins. （20） The strike mission shall end no later than the assessment mission start time. （21） in, ; ; ; For target locations that only involve strike and assessment missions, the assessment mission start time must not be earlier than the strike mission end time. （22） in, ; like If any of the above constraints are not met, discard the candidate and continue the check. Next candidate; if none of the candidate sets satisfy the above constraints, increment the neighborhood operator index by one, and let... Return to S2 to continue the iterative search; S2.

2. Calculation of the objective function For candidates that have passed the feasibility check The mission completion time is calculated according to formula (23), which is the sum of flight time and mission execution time: （23） in, For drones The sequence of flight path nodes, including base and mission nodes; Let be the flight distance from node i to node j; For drones Flight speed; Let n be the execution time of task n; S3. Update the candidate task sequence for a single drone and switch the neighborhood operator. In operators Among all feasible candidate tasks, traverse All feasible candidates Calculate the completion time of each candidate according to formula (23). ;Pick Minimum candidate As ; Calculate the improvement amount Determine the current operator The search results; among them, The task completion time for the initial feasible task sequence S; If an improved solution is found :accept As the new current solution, let , And reset the neighborhood operator index to Return to S2 to continue the iterative search; If no improved solution is found, i.e., the current solution... All candidates within the neighborhood fail to decrease f(S): Increment the neighborhood operator index by one, let Return to S2 to continue the iterative search; S4. Output the local optimal task sequence for a single UAV. When the neighborhood operator index When all six neighborhood operators are exceeded, it indicates that the current solution cannot be further improved, and the algorithm terminates; at this point, the locally optimal task sequence for the UAV under the current task assignment is output. and its precise task completion time .

Images