Large-scale unmanned aerial vehicle task planning method and system
Optimizing the UAV mission sequence through spectral clustering and adaptive variable neighborhood search algorithms, solving the problem of complex computing and convergence difficulties in large-scale UAV mission planning, achieving rapid acquisition of global optimal solutions, and improving post-disaster materials Efficiency of delivery.
Patent Information
- Application Number
- CN202310607922.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-26
- Publication Date
- 2024-11-29
AI Technical Summary
The existing UAV mission planning methods are complex in computing and convergence when facing large-scale problems, and cannot quickly obtain the optimal task sequence, especially in post-disaster material distribution scenarios, which are difficult to meet the needs of rapid response.
The spectral clustering algorithm is used to divide the task sequence into groups, and the adaptive variable neighborhood search algorithm is used to optimize each group of task sequences, combined with local search algorithms to reduce computational complexity and accelerate convergence, so as to achieve the acquisition of global optimal solutions.
Through the partitioning and governance strategy and adaptive search mechanism, the computational complexity is significantly reduced, the convergence speed and resolution of the algorithm are improved, and the optimal drone mission sequence can be quickly planned and the waiting time for material delivery is shortened.
Smart Images

Figure CN119045505A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular to a large-scale unmanned aerial vehicle mission planning method and system. Background Art
[0002] With the rapid development of drone technology, the characteristics of drones such as flexibility, speed and low cost provide them with great application value in many industries. Especially in post-disaster rescue scenarios, drones are quickly deployed and flexible in action. They are not restricted by terrain and can quickly reach the disaster site to complete the material delivery mission. In disaster scenarios such as wildfires and earthquakes, how to reduce the waiting time for material delivery at the disaster site is the key to reducing losses. Therefore, when there are multiple drone departure points and drones have limited capacity and flight distance, how to plan the customer service sequence of each drone so that all customers are served and only served once, and the material delivery waiting time of all customers is minimized is a cumulative drone task planning problem with capacity constraints at multiple stations that deserves close attention.
[0003] The solutions of domestic and foreign researchers are mainly divided into deterministic methods and heuristic methods. Deterministic methods include nonlinear integer programming solvers, dynamic programming, etc. The advantage of this type of method is that the problem to be solved is modeled as a mixed integer linear programming problem, which can obtain the optimal or approximately optimal solution. However, this type of method does not have a unified treatment method for different problems, and needs to be analyzed and processed according to the specific properties of different problems. In addition, when the scale of the problem increases, the computational cost of the solution rises sharply. On large-scale data sets, it is impossible to obtain a solution in a short time. These defects are in great conflict with the requirements of post-disaster material distribution scenarios. Heuristic methods include intelligent optimization methods such as genetic algorithms, ant colony algorithms, and local search. Compared with deterministic methods, this type of algorithm can find the approximate optimal solution in the current situation in a shorter time by designing a reasonable solution structure and search strategy.
[0004] However, when the number of customers to be served is greater than 250, which can be considered as a large-scale UAV mission planning problem, the existing heuristic methods lack a suitable divide-and-conquer strategy, which makes the algorithm complex to solve and has low global convergence ability, affecting the quality of the task sequence solution; the existing local search method has a high overhead of traversing all neighborhoods, and the performance of a randomly selected neighborhood is relatively poor, and there is a lack of methods to select a suitable search neighborhood. Summary of the invention
[0005] In view of the above analysis, an embodiment of the present invention aims to provide a large-scale UAV task planning method and system to solve the problem that the optimal task sequence cannot be obtained due to the complex and time-consuming calculation method and the difficulty in algorithm convergence when facing large-scale problems.
[0006] On the one hand, an embodiment of the present invention provides a large-scale UAV mission planning method comprising the following steps:
[0007] Construct the objective function and constraints of UAV mission planning;
[0008] Initialize the task sequence, and insert each customer into the task sequence with the smallest objective function value according to the change in the objective function value after the customer is inserted into the task sequence, so as to obtain multiple initial task sequences;
[0009] The following update operations are iteratively performed on multiple initial task sequences: the spectral clustering algorithm is used to divide the multiple initial task sequences into multiple groups; according to the preset neighborhood operator, the adaptive variable neighborhood search algorithm is used to optimize the task sequence of each group in turn, and then merge them to obtain multiple global task sequences; if the total objective function value of the multiple global task sequences is less than the total objective function value of the multiple initial task sequences, the multiple global task sequences are used to update the multiple initial task sequences; the next round of update operations is performed until the stop condition is reached, and the updated multiple initial task sequences constitute the optimal UAV task.
[0010] Based on the further improvement of the above method, after merging multiple global task sequences, a local search algorithm is used to iteratively repair each global task sequence using each neighborhood operator in turn, and the repaired task sequence with the smallest objective function value is taken to update the corresponding global task sequence; and multiple repaired global task sequences are obtained.
[0011] Based on the further improvement of the above method, the objective function and constraints of UAV mission planning are constructed, including:
[0012] The objective function is constructed by taking the time it takes for each drone to reach a customer as the cumulative time and the goal of minimizing the total cumulative time for each drone to depart from the warehouse and reach each customer in the task sequence.
[0013] Constraints are constructed based on the number of times each customer is visited, the time it takes for the drone to reach the customer in the same mission sequence, the maximum capacity of the drone to carry supplies, and the remaining power of the drone when it reaches each customer;
[0014] When calculating the objective function value of any task sequence, if the current task sequence meets all constraints, the objective function value is calculated according to the objective function, otherwise, the objective function value is set to the preset maximum function value.
[0015] Based on the further improvement of the above method, the task sequence is initialized, and according to the change value of the objective function value after the customer is inserted into the task sequence, each customer is inserted into the task sequence with the smallest objective function value, and multiple initial task sequences are obtained, including:
[0016] For each drone, take its warehouse as the starting point of the task sequence and build multiple first task sequences; put all customers into the set to be inserted;
[0017] Iteratively perform the following update operation on multiple first task sequences: insert each customer in the set to be inserted into each first task sequence in turn, calculate the change value of the objective function value of each first task sequence before and after the insertion, and obtain the difference between the second minimum change value and the minimum change value; compare the difference value corresponding to each customer, insert the customer with the largest difference value to the end of the first task sequence with the minimum change value corresponding to the customer, and delete the customer from the set to be inserted, and perform the next update operation until the set to be inserted is empty;
[0018] From the updated multiple first task sequences, the first task sequences containing customers are screened out, and the warehouse closest to the last customer is inserted at the end of each of the screened first task sequences as the task sequence end point to obtain multiple initial task sequences.
[0019] Based on the further improvement of the above method, the spectral clustering algorithm is used to divide multiple initial task sequences into multiple groups, including:
[0020] According to the longitude and latitude coordinates of each node in each initial task sequence, and the number of customers and warehouses, the centroid coordinates of each initial task sequence are calculated by the following formula as input samples:
[0021]
[0022] in, Represents the mth task sequence V m The center of gravity b m The longitude and latitude coordinates of |V m | represents the number of nodes in the mth task sequence, including customers and warehouses; x i Represents the longitude coordinate of the i-th node; y i Represents the latitude coordinate of the i-th node;
[0023] The input samples, the preset feature dimensions and the number of customers in each group are input into the spectral clustering algorithm to cluster multiple initial task sequences into multiple groups.
[0024] Based on the further improvement of the above method, according to the preset neighborhood operator, an adaptive variable neighborhood search algorithm is used to optimize the task sequence of each group in turn, including:
[0025] Perform the following iterative optimization operations on the task sequence of each group: initialize the global optimal solution, the optimal solution of the previous iteration, and the weight of the neighborhood operator, and set the maximum number of executions;
[0026] In each execution, a neighborhood operator is selected by the roulette wheel selection algorithm according to the weight of each neighborhood operator; the first solution of each group is calculated according to the selected neighborhood operator; the variable neighborhood descent algorithm is used for the first solution to obtain the second solution; the performance record value of the selected neighborhood operator, the global optimal solution and the optimal solution of the current iteration are updated according to the global optimal solution, the optimal solution of the current iteration, the optimal solution of the previous iteration and the second solution, and the next execution is performed after each group is updated with the global optimal solution until the maximum number of executions is reached, and the optimal solution of the previous iteration is updated with the optimal solution of the current iteration for each group, and the weight of each neighborhood operator is updated according to the preset weight update rate and the performance record value of each neighborhood operator, and the next round of iterative optimization is performed until the maximum number of iterations of the group is reached, and the global optimal solution of each group is obtained, which is the task sequence of each group after optimization.
[0027] Based on the further improvement of the above method, the performance record value of the selected neighborhood operator, the global optimal solution and the optimal solution of the current iteration are updated according to the global optimal solution, the optimal solution of the current iteration, the optimal solution of the previous iteration and the second solution, including:
[0028] If the optimal solution of the current iteration is empty, the second solution is used to update the optimal solution of the current iteration;
[0029] If the objective function value of the second solution is less than the objective function value of the global optimal solution, then according to the value of the first weight update factor, the performance record value of the selected neighborhood operator is increased, and the global optimal solution is updated using the second solution;
[0030] If the objective function value of the second solution is less than the objective function value of the optimal solution of the current iteration, then the performance record value of the selected neighborhood operator is increased according to the value of the second weight update factor, and the second solution is used to update the optimal solution of the current iteration;
[0031] If the objective function value of the second solution is less than the objective function value of the optimal solution of the previous iteration, the performance record value of the selected neighborhood operator is increased according to the value of the third weight update factor;
[0032] The value of the first weight update factor is greater than the value of the second weight update factor, and the value of the second weight update factor is greater than the value of the third weight update factor.
[0033] Based on the further improvement of the above method, according to the preset weight update rate and the performance record value of each neighborhood operator, the weight of each neighborhood operator is updated by the following formula:
[0034] ω p =ρ*ω p +(1-ρ)*σ ρ
[0035] Among them, ω p represents the weight of the pth neighborhood operator, σ irepresents the performance record value of the pth neighborhood operator; ρ represents the weight update rate, 0≤ρ≤1.
[0036] Based on the further improvement of the above method, the neighborhood operators include:
[0037] The Or-Opt operator is used to randomly select a segment containing three consecutive customers and reinsert it into another position;
[0038] The fragment exchange operator is used to exchange fragments of arbitrary length between two task sequences;
[0039] The three-node position exchange operator is used to randomly select two customers c1 and c2 on a task sequence and a customer c3 on another task sequence, insert customer c3 into the position of customer c1, insert customer c1 into the position of customer c2, and insert customer c2 into the position of customer c3.
[0040] On the other hand, an embodiment of the present invention provides a large-scale UAV mission planning system, including:
[0041] The target building module is used to build the objective function and its constraints for UAV mission planning;
[0042] A task construction module is used to initialize the task sequence. According to the change of the objective function value after inserting the customer into the task sequence, each customer is inserted into the task sequence with the smallest objective function value, so as to obtain multiple initial task sequences.
[0043] The task update module is used to iteratively perform the following update operations on multiple initial task sequences: use the spectral clustering algorithm to divide the multiple initial task sequences into multiple groups; according to the preset neighborhood operator, use the adaptive variable neighborhood search algorithm to optimize the task sequence of each group in turn, and merge them to obtain multiple global task sequences; if the total objective function value of the multiple global task sequences is less than the total objective function value of the multiple initial task sequences, use the multiple global task sequences to update the multiple initial task sequences; perform the next round of update operations until the stop condition is reached, and the updated multiple initial task sequences constitute the optimal UAV task.
[0044] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:
[0045] 1. Use spectral clustering method to extract the correlation knowledge of each task sequence, map the correlation knowledge to the feature space, and classify similar task sequences into the same group, so as to achieve the division and conquest of large-scale problems, reduce the scale of the problem, reduce the computational complexity, and improve the computational speed;
[0046] 2. Use the adaptive variable neighborhood search algorithm to solve each group. The adaptive mechanism selects the neighborhood for the next search based on the knowledge of the correlation between the neighborhood and the solution, making the neighborhood search more efficient. While accelerating the convergence of the algorithm, it avoids falling into the local optimum and obtains the global optimal solution.
[0047] 3. Starting from minimizing the cumulative time of drone delivery, taking into account the drone's battery power and the capacity of the materials it carries, the optimal task sequence can be quickly planned, which has important practical significance for shortening the waiting time for material delivery and efficiently planning drone delivery tasks.
[0048] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] The drawings are only for the purpose of illustrating specific embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.
[0050] Figure 1 This is a flow chart of a large-scale UAV mission planning method in Example 1 of the present invention;
[0051] Figure 2 is an update flow chart of the initial task sequence in Embodiment 1 of the present invention;
[0052] Figure 3 is an optimized flow chart of the task sequence of each group in Example 1 of the present invention;
[0053] Figure 4 A distribution diagram of warehouses and customers in the example scenario in Embodiment 1 of the present invention;
[0054] Figure 5 The initial task sequence distribution diagram for the example scenario planned in Embodiment 1 of the present invention;
[0055] Figure 6 This is the optimal task sequence distribution diagram planned for the example scenario in Example 1 of the present invention. DETAILED DESCRIPTION
[0056] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.
[0057] Example 1
[0058] A specific embodiment of the present invention discloses a large-scale UAV mission planning method, such as Figure 1 As shown, the following steps are included:
[0059] S1: Construct the objective function and constraints of UAV mission planning.
[0060] It should be noted that the large-scale drone scenario of this embodiment includes: there are multiple warehouses and customers, and each warehouse has multiple drones. Drone task planning is to plan a task sequence for each drone in each warehouse, so that each drone departs from the warehouse and achieves the shortest total cumulative time for all task sequences while ensuring that each customer is served only once. Among them, the total cumulative time is the total waiting time of all customers in the task sequence before receiving service. The drone departs from the warehouse, visits only one customer, and returns to any warehouse. The drone's flight speed is uniform, and the drone's flight distance and carrying capacity are limited.
[0061] Based on the above scenario, let C represent the customer point set, D represent the warehouse point set, V=C∪D represent the union of the customer and warehouse point sets, R represent the set of available drones, Q represent the capacity of the drone to carry materials, and E represent the battery capacity of the drone.
[0062] Taking the time it takes for each drone to reach a customer as the cumulative time, and taking the shortest total cumulative time for each drone to reach each customer in the task sequence from the warehouse as the goal, the objective function of drone task planning is constructed, which is expressed by the following formula:
[0063]
[0064] in, It represents the time when UAV k arrives at the i-th customer node, that is, the cumulative time before the i-th customer node receives the service.
[0065] It should be noted that large-scale drone missions include multiple drone mission sequences; the mission sequence of each drone starts from the warehouse and returns to the warehouse, but the total accumulated time does not take into account the time it takes for the drone to return to the warehouse from the last customer in the mission sequence.
[0066] Furthermore, constraints are constructed based on the number of times each customer is visited, the time it takes for the drone to reach the customer in the same mission sequence, the maximum capacity of the drone to carry materials, and the remaining power of the drone when it reaches each customer. Constraints include:
[0067] 1) There can be only one task sequence from customer i to customer j, which can be expressed by the following formula:
[0068]
[0069] in, Indicates whether drone k starts from node i and visits node j. If yes, it takes 1, otherwise it takes 0. It is expressed by the following formula:
[0070]
[0071] 2) Any drone must leave after visiting a customer, expressed as follows:
[0072]
[0073] 3) Any drone must start and end from the warehouse, which can be expressed by the following formula:
[0074]
[0075] 4) The demand for customers served by any drone cannot exceed the maximum capacity of the supplies it carries, which can be expressed by the following formula:
[0076]
[0077] Among them, q j represents the demand of the jth node;
[0078] 5) If a drone visits customer j from customer i, then the time it takes for the drone to reach customer i must be less than the time it takes to reach customer j;
[0079]
[0080] Among them, t ij represents the time taken by the drone to reach the jth node from the i-th node, M is the preset maximum function value, illustratively, M=99999999;
[0081] 6) If the drone visits customer j from customer i, then the remaining battery of the drone at customer i must be greater than the remaining battery at customer j, expressed by the following formula:
[0082]
[0083] in, represents the remaining power of UAV k when it reaches the i-th node; α ij It represents the amount of electricity consumed by the drone when it departs from the i-th node and reaches the j-th node;
[0084] 7) After visiting a customer point, the drone must have enough power left to return to the warehouse, which is expressed by the following formula:
[0085]
[0086] 8) The time when the drone arrives at the warehouse is 0. Since the return trip to the warehouse is not considered, this constraint indicates that the time when the drone departs from the warehouse is 0, which is expressed by the following formula:
[0087]
[0088] 9) The time it takes for a drone to reach any node should be greater than 0, expressed as follows:
[0089]
[0090] 10) The drone is fully charged when it departs, and the starting power is its battery capacity, which is expressed by the following formula:
[0091]
[0092] 11) The remaining power of the drone at any node should be less than its battery capacity, which can be expressed by the following formula:
[0093]
[0094] It is worth noting that when calculating the objective function value of any task sequence, if the current task sequence meets all constraints, the objective function value is calculated according to the objective function of formula (1); otherwise, the objective function value is set to the preset maximum function value, such as 99999999.
[0095] Preprocess the acquired drone, warehouse and customer information to obtain the information in the objective function and its constraints. For example, construct a set of customers, warehouses, drones, and customers and warehouses; obtain the time required for the drone to fly between any two nodes, the power consumed, and the material capacity based on the location information of the warehouse and customers, the flight speed of the drone, and the customer needs.
[0096] It should be noted that this embodiment starts from minimizing the cumulative time of drone delivery, takes into account the drone's battery power and the capacity of the materials it carries, and is more suitable for existing practical application scenarios.
[0097] S2: Initialize the task sequence. According to the change in the objective function value after inserting the customer into the task sequence, insert each customer into the task sequence with the smallest objective function value, and obtain multiple initial task sequences.
[0098] Specifically, the following steps are included:
[0099] ① For each drone, take its warehouse as the starting point of the task sequence and build multiple first task sequences; put all customers into the set to be inserted.
[0100] It should be noted that, if there are N warehouses, each of which contains M drones, then M×N first task sequences are initially constructed, and each first task sequence initially only includes the warehouse to which the corresponding drone belongs. For example, the first task sequence corresponding to drone A in warehouse d0 is {d0}.
[0101] ② Iteratively perform the following update operations on multiple first task sequences: insert each customer in the set to be inserted into each first task sequence in turn, calculate the change in the objective function value of each first task sequence before and after the insertion, and obtain the difference between the second minimum change value and the minimum change value; compare the difference corresponding to each customer, insert the customer with the largest difference to the end of the first task sequence with the minimum change value corresponding to the customer, and delete the customer from the set to be inserted, and perform the next update operation until the set to be inserted is empty.
[0102] It should be noted that when there is no sub-minimum change value, the minimum change value is taken as the difference. The sub-minimum change value is the second smallest change value. The change value of the objective function value of each first task sequence before and after the insertion is calculated, that is, the customer's insertion cost. This step first finds the difference in insertion cost corresponding to each customer from all task sequences, and then finds the customer with the largest insertion cost difference through the difference, and inserts it to the end of the task sequence with the smallest insertion cost for the customer.
[0103] Exemplarily, after customer c1 is inserted into the first task sequence {d0}, it is updated to {d0, c1}, and after customer c2 is inserted into the first task sequence, it is updated to {d0, c1, c2}.
[0104] ③ Filter out the first task sequences that contain customers from the updated multiple first task sequences, and insert the warehouse closest to the last customer at the end of each of the filtered first task sequences as the task sequence end point, thereby obtaining multiple initial task sequences.
[0105] Exemplarily, at the end of the first task sequence {d0, c1, c2}, the warehouse d1 closest to the last customer c2 is inserted to obtain {d0, c1, c2, d1} as the initial task sequence.
[0106] It should be noted that the warehouse at the end of the task sequence can be the same as the warehouse at the start of the task sequence.
[0107] S3: Iteratively perform the following update operations on multiple initial task sequences: use the spectral clustering algorithm to divide the multiple initial task sequences into multiple groups; according to the preset neighborhood operator, use the adaptive variable neighborhood search algorithm to optimize the task sequence of each group in turn, and then merge them to obtain multiple global task sequences; if the total objective function value of the multiple global task sequences is less than the total objective function value of the multiple initial task sequences, use the multiple global task sequences to update the multiple initial task sequences; perform the next round of update operations until the stop condition is reached, and the updated multiple initial task sequences constitute the optimal UAV task.
[0108] It should be noted that this step is a global iterative process, which continuously optimizes the initial task sequence sint, such as Figure 2 As shown, the following steps are included:
[0109] S30, setting the global maximum number of iterations and the global maximum number of iterations without improvement.
[0110] S31. Using a spectral clustering algorithm, multiple initial task sequences are divided into multiple groups.
[0111] Compared with the prior art, this step uses a spectral clustering algorithm to extract the correlation knowledge of each task sequence, maps the correlation knowledge to the feature space, and groups similar task sequences into the same group, thus achieving problem division and reducing the scale of the problem.
[0112] S32. According to a preset neighborhood operator, an adaptive variable neighborhood search algorithm is used to optimize the task sequence of each group in turn, and then multiple global task sequences are obtained by merging.
[0113] Compared with the prior art, this step is used to adopt an adaptive variable neighborhood search algorithm for each group. The adaptive mechanism selects the neighborhood for the next search based on the knowledge of the correlation between the neighborhood and the solution, making the neighborhood search more efficient, while accelerating the convergence of the algorithm and preventing the result from falling into the local optimum.
[0114] S33: If the total objective function value of the multiple global task sequences is less than the total objective function value of the multiple initial task sequences, the multiple global task sequences are used to update the multiple initial task sequences.
[0115] Preferably, a local search algorithm is adopted to iteratively repair each global task sequence using each neighborhood operator in turn, and the repaired task sequence with the smallest total objective function value is taken to update the corresponding global task sequence; after obtaining multiple repaired global task sequences, the total objective function values of the multiple global task sequences are compared with the total objective function values of the multiple initial task sequences.
[0116] If the objective function values of multiple global task sequences are greater than or equal to the objective function values of multiple initial task sequences, it means that the performance has not been improved, and the number of iterations without performance improvement increases by 1; otherwise, the performance has been improved, and the number of iterations without performance improvement is 0.
[0117] S34, determine whether the stop condition is met, if not, return to step S31 to perform the next round of update steps, otherwise, obtain the updated multiple initial task sequences to form the optimal drone task.
[0118] It should be noted that the stopping conditions include: the number of iterations reaches the global maximum number of iterations, or the number of no-improvement iterations reaches the global maximum number of no-improvement iterations.
[0119] Specifically, step S31 includes:
[0120] ① According to the longitude and latitude coordinates of each node in each initial task sequence, and the number of customers and warehouses, the centroid coordinates of each initial task sequence are calculated by the following formula as input samples:
[0121]
[0122] in, Represents the mth task sequence V m The center of gravity b m The longitude and latitude coordinates of |V m | represents the number of nodes in the mth task sequence, including customers and warehouses; x i Represents the longitude coordinate of the i-th node; y i Represents the latitude coordinate of the i-th node.
[0123] The input samples corresponding to the K initial task sequences are: {b1,b2,…,b m ,…,b K}.
[0124] ②Construct similarity matrix W and degree matrix D based on the input samples.
[0125] It should be noted that each element in the similarity matrix W is calculated by the following formula:
[0126]
[0127] Among them, ||b m -b n ||2 represents the distance between the center of gravity of the mth task sequence and the center of gravity of the nth task sequence, and σ is an adjustable parameter.
[0128] Degree matrix D = diag(d1, d2, ..., d m ,…,dK ),in That is, each element in the degree matrix is defined as the sum of the weights of all edges connected to it.
[0129] ③Calculate the Laplace matrix L = DW, L∈R K×K .
[0130] ④ Calculate the eigenvalues of the Laplace matrix L. According to the preset eigenvalue number lk, select the eigenvectors f corresponding to the lk eigenvalues starting from the minimum eigenvalue to form the eigenmatrix F, F∈R K×lk ; Each row in F represents a data point of lk dimension.
[0131] ⑤ Use the k-means clustering method to cluster the feature matrix F and divide the K initial task sequences into H groups {g1, g2, …, g H}, each group contains multiple initial task sequences.
[0132] It should be noted that there is no duplication of task sequences between groups and they are independent of each other. C represents the total number of customers, B represents the number of customers in each preset group, To round up.
[0133] Step S31 is to divide the large-scale problem into multiple small problems and cluster them in units of task sequences. The number of customers included in each task sequence is uncertain. The actual number of customers included in each group obtained after clustering is approximately B. The task sequences of each group are independent of each other and have no overlap.
[0134] It should be noted that step S32 utilizes a preset neighborhood operator to iteratively optimize the task sequence of each group obtained in step S31 through a double-layer iterative process.
[0135] It should be noted that the neighborhood operators in this embodiment include 8 types, which are described as follows:
[0136] The node position reset operator is used to delete a customer node from its original position and randomly select another task sequence or a position on the task sequence to insert the customer;
[0137] The node position exchange operator is used to exchange the positions of two customer nodes in different task sequences;
[0138] 2-2 Exchange operator, used to exchange two segments containing two consecutive customers in different task sequences;
[0139] The segment and node exchange operator is used to randomly select a customer in one task sequence and two adjacent customers in another task sequence and exchange their positions;
[0140] The Or-Opt operator is used to randomly select a segment containing three consecutive customers and reinsert it into another position;
[0141] The fragment exchange operator is used to exchange fragments of arbitrary length between two task sequences;
[0142] The three-node position exchange operator is used to randomly select two customers c1 and c2 on one task sequence and a customer c3 on another task sequence, insert customer c3 into the position of customer c1, insert customer c1 into the position of customer c2, and insert customer c2 into the position of customer c3;
[0143] The node reinsertion operator is used to calculate the change ε of the objective function of the solution after each customer node is deleted from its original position, select the first three customer nodes with the largest ε values, and insert these three customer nodes in sequence to the position that increases the objective function of the solution the least.
[0144] Further, step S32 uses an adaptive variable neighborhood search algorithm based on a preset neighborhood operator to optimize the task sequence of each group in turn, that is, traverse each group and use the following method for each group: Figure 3 The steps S320 to S322 shown are optimized to obtain the optimized task sequence of each group, which are merged to obtain multiple global task sequences as the output of step S32.
[0145] Specifically, step S32 performs the following iterative optimization operations on the task sequence of each group:
[0146] S320, initializing the data of each group, including: the global optimal solution gbest, the optimal solution lbest of the previous iteration, the neighborhood operator weight, the maximum number of executions and the maximum number of group iterations.
[0147] It should be noted that the first-layer iterative optimization operation is performed on each group in turn according to the maximum number of group iterations; the second-layer neighborhood search operation is performed in each round of iterative optimization operation in each group according to the maximum number of executions. The maximum number of executions in each round of iterative optimization operation is set according to the number of preset neighborhood operators.
[0148] At the same time, the task sequence of each group obtained in step S31 is used as the initial global optimal solution and the optimal solution of the previous round of iteration, and the weight of each neighborhood operator is initialized to 1. The optimal solution of the previous round of iteration refers to the optimal solution of the previous round of iteration of each group in the iterative optimization operation of the first layer; the global optimal solution is the task sequence of each group after the task sequence is optimized in step S32, that is, the task sequence obtained after reaching the maximum number of iterations of the group.
[0149] S321. Perform an iterative adaptive variable neighborhood search operation on the task sequence of each group until the maximum number of executions is reached. For each group, use the optimal solution of the current iteration to update the optimal solution of the previous iteration. Update the weight of each neighborhood operator according to the preset weight update rate and the performance record value of each neighborhood operator.
[0150] Specifically, each time an adaptive variable neighborhood search is performed, it includes:
[0151] S321-0, initialize the optimal solution currentbest of the current iteration to be empty, and the performance record value of each neighborhood operator to be 0.
[0152] S321-1. According to the weight of each neighborhood operator, a neighborhood operator is selected by using a roulette wheel selection algorithm, wherein the roulette wheel selection probability of each neighborhood operator is the ratio of the weight of each neighborhood operator to the sum of the weights of all neighborhood operators.
[0153] S321-2. Calculate the first solution of each group according to the selected neighborhood operator; that is, according to the selected neighborhood operator, randomly select a position in the neighborhood operator that meets the constraint conditions, update the task sequence of each group, and obtain the first solution.
[0154] S321-3. Apply the variable neighborhood descent (VND) algorithm to the first solution to obtain the second solution. The neighborhood operators used in VND are all neighborhood operators, which are used to find the solution with the highest performance than the input solution. If the optimal solution of the current iteration is empty, the second solution is used to update the optimal solution of the current iteration.
[0155] It can be understood that the first solution refers to the task sequence obtained by adjusting the position of the customer node in the task sequence of each group according to the selected neighborhood operator, and the second solution refers to the task sequence obtained by adjusting the position of the customer node in the task sequence of the first solution according to the VND algorithm. The total objective function value of the task sequence of the second solution is less than or equal to the total objective function value of the task sequence of the first solution.
[0156] S321-4. Update the performance record value of the selected neighborhood operator, the global optimal solution and the optimal solution of the current iteration according to the global optimal solution, the optimal solution of the current iteration, the optimal solution of the previous iteration and the second solution.
[0157] Specifically, the update is performed according to the following judgments:
[0158] If the objective function value of the second solution is less than the objective function value of the global optimal solution, then according to the value of the first weight update factor, the performance record value of the selected neighborhood operator is increased, and the global optimal solution is updated using the second solution;
[0159] If the objective function value of the second solution is less than the objective function value of the optimal solution of the current iteration, then the performance record value of the selected neighborhood operator is increased according to the value of the second weight update factor, and the second solution is used to update the optimal solution of the current iteration;
[0160] If the objective function value of the second solution is less than the objective function value of the optimal solution of the previous iteration, the performance record value of the selected neighborhood operator is increased according to the value of the third weight update factor;
[0161] The value of the first weight update factor is greater than the value of the second weight update factor, and the value of the second weight update factor is greater than the value of the third weight update factor.
[0162] It should be noted that the value of the weight update factor is preset and is used to increase the weight of the selected neighborhood operator according to the performance improvement of the second solution.
[0163] S321-5. Determine whether the maximum number of executions has been reached. If not, use the global optimal solution to update each group and return to S321-1 for the next execution. Otherwise, complete a round of iterative optimization and execute step S322.
[0164] S322. Determine whether the maximum number of iterations of the group has been reached. If not, use the optimal solution of the current iteration to update the optimal solution of the previous iteration. Update the weight of each neighborhood operator according to the preset weight update rate and the performance record value of each neighborhood operator. Return to step S321 and perform the next round of iterative optimization. Otherwise, obtain the global optimal solution gbest for each group, which is the optimized task sequence of each group.
[0165] Specifically, the weight of each neighborhood operator is updated by the following formula, so that the neighborhood with greater performance improvement has more chances to be selected.
[0166] ω p =ρ*ω p +(1-ρ)*σ p Formula (16)
[0167] Among them, ω p represents the weight of the pth neighborhood operator, σ i represents the performance record value of the pth neighborhood operator; ρ represents the weight update rate, 0≤ρ≤1.
[0168] It should be noted that the adaptive variable neighborhood search in step S321 mainly dynamically selects neighborhood operators through the weights of neighborhood operators and the roulette wheel selection probability, and updates different performance record values for neighborhood operators by identifying the performance level of the solution, thereby increasing the weights of neighborhood operators that can improve performance, making the neighborhood search more efficient, and preventing the results from falling into local optimality while accelerating the convergence of the algorithm.
[0169] In step S33, a local search algorithm is used to iteratively repair each global task sequence using each neighborhood operator in turn. For each neighborhood operator, all feasible positions are traversed, and from all feasible positions of all neighborhood operators, the position that minimizes the objective function value of the task sequence (i.e., the maximum performance improvement) is selected to update the global task sequence. Then, the objective function value is compared with all the initial task sequences obtained in step S2. If the objective function values of multiple global task sequences are smaller than the objective function values of multiple initial task sequences, multiple global task sequences are used to update multiple initial task sequences. Otherwise, it indicates that the performance has not been improved and the next round of global iteration is required.
[0170] The following takes a specific application scenario as an example to use the large-scale UAV mission planning method of this embodiment to obtain the optimal UAV mission. Figure 4 The figure shows the distribution of warehouses and customers in the example scenario, where the center symbol represents customers, the number is 50, and the triangle symbol represents warehouses, the number is 4. Each warehouse contains 4 drones, each drone has a payload carrying capacity of 80, a battery capacity of 14, and the drone battery consumption rate is 1 / 5 of the distance, that is, one unit of power is consumed for every 5 units of flight distance. The average demand of 50 customers is 15.41, the minimum demand is 3, and the maximum demand is 41.
[0171] In step S2, 16 initial task sequences are obtained, such as Figure 5 As shown, the objective function value is 1077.26.
[0172] The global maximum number of iterations in the iteration stop condition set in step S3 is 50, and the global maximum number of iterations without improvement is 5; in the global iteration process, the preset feature dimension number lk in step S31 is 20, and the preset number of customers B in each group is 20, and the 16 initial task sequences are divided into 3 groups. In step S32, the adaptive weight update rate ρ is set to 0.95, the values of the first, second and third weight update factors are 30, 10 and 5 respectively, and the maximum number of group iterations is 20. The algorithm convergence time of step S3 is about 40 seconds, and the total running time is about 95 seconds. After the operation, 16 updated task sequences are obtained to form the optimal drone task, as shown in Figure 6 As shown, the objective function value is 863.40. Figure 5 and Figure 6 It can be seen that there are fewer intersections between task sequences in the optimal UAV mission, and the task sequences are more independent and reasonable.
[0173] Compared with the prior art, the large-scale UAV task planning method provided in this embodiment starts from minimizing the cumulative time of UAV delivery, takes into account the power of the UAV and the capacity of the materials it carries, and quickly plans the optimal task sequence, which has important practical significance for shortening the waiting time for material delivery and efficiently planning UAV delivery tasks. Among them, the spectral clustering method is used to extract the correlation knowledge of each task sequence, and the correlation knowledge is mapped to the feature space, so that similar task sequences are divided into the same group, realizing the division and control of large-scale problems, reducing the scale of the problem, reducing the computational complexity, and improving the computational speed; an adaptive variable neighborhood search algorithm is used to solve each group, and the adaptive mechanism selects the neighborhood for the next search based on the correlation knowledge of the neighborhood and the solution, making the neighborhood search more efficient, while accelerating the convergence of the algorithm, avoiding falling into the local optimum, and obtaining the global optimal solution.
[0174] Example 2
[0175] Another embodiment of the present invention discloses a large-scale UAV mission planning system, thereby implementing the large-scale UAV mission planning method in Example 1. The specific implementation of each module refers to the corresponding description in Example 1. The system includes:
[0176] The target building module is used to build the objective function and its constraints for UAV mission planning;
[0177] A task construction module is used to initialize the task sequence. According to the change of the objective function value after inserting the customer into the task sequence, each customer is inserted into the task sequence with the smallest objective function value, so as to obtain multiple initial task sequences.
[0178] The task update module is used to iteratively perform the following update operations on multiple initial task sequences: use the spectral clustering algorithm to divide the multiple initial task sequences into multiple groups; according to the preset neighborhood operator, use the adaptive variable neighborhood search algorithm to optimize the task sequence of each group in turn, and merge them to obtain multiple global task sequences; if the total objective function value of the multiple global task sequences is less than the total objective function value of the multiple initial task sequences, use the multiple global task sequences to update the multiple initial task sequences; perform the next round of update operations until the stop condition is reached, and the updated multiple initial task sequences constitute the optimal UAV task.
[0179] Since the present embodiment and the aforementioned large-scale UAV mission planning method embodiment are related, they can be mutually referenced, and are repeated here, so they will not be repeated here. Since the present system embodiment and the aforementioned method embodiment have the same principle, the present system embodiment also has the corresponding technical effects of the aforementioned method embodiment.
[0180] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.
[0181] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
Claims
1. A large-scale UAV mission planning method, characterized in that: The following steps are involved: Construct the objective function and constraints of UAV mission planning; Initialize the task sequence, and insert each customer into the task sequence with the smallest objective function value according to the change in the objective function value after the customer is inserted into the task sequence, so as to obtain multiple initial task sequences; The following update operations are iteratively performed on multiple initial task sequences: the spectral clustering algorithm is used to divide the multiple initial task sequences into multiple groups; according to the preset neighborhood operator, the adaptive variable neighborhood search algorithm is used to optimize the task sequence of each group in turn, and then merge them to obtain multiple global task sequences; if the total objective function value of the multiple global task sequences is less than the total objective function value of the multiple initial task sequences, the multiple global task sequences are used to update the multiple initial task sequences; the next round of update operations is performed until the stop condition is reached, and the updated multiple initial task sequences constitute the optimal UAV task.
2. The large-scale UAV mission planning method according to claim 1, characterized in that: After the merging obtains multiple global task sequences, a local search algorithm is used to iteratively repair each global task sequence using each neighborhood operator in turn, and the repair task sequence with the smallest objective function value is taken to update the corresponding global task sequence; Get the repaired multiple global task sequences.
3. The large-scale UAV mission planning method according to claim 1 or 2, characterized in that: The objective function and constraint conditions for constructing the UAV mission planning include: The objective function is constructed by taking the time it takes for each drone to reach a customer as the cumulative time and the goal of minimizing the total cumulative time for each drone to depart from the warehouse and reach each customer in the task sequence. Constraints are constructed based on the number of times each customer is visited, the time it takes for the drone to reach the customer in the same mission sequence, the maximum capacity of the drone to carry supplies, and the remaining power of the drone when it reaches each customer; When calculating the objective function value of any task sequence, if the current task sequence meets all constraints, the objective function value is calculated according to the objective function, otherwise, the objective function value is set to the preset maximum function value.
4. The large-scale UAV mission planning method according to claim 3 is characterized in that: The initialization task sequence inserts each customer into the task sequence with the smallest objective function value according to the change value of the objective function value after the customer is inserted into the task sequence, and obtains multiple initial task sequences, including: For each drone, take its warehouse as the starting point of the task sequence and build multiple first task sequences; put all customers into the set to be inserted; Iteratively perform the following update operation on multiple first task sequences: insert each customer in the set to be inserted into each first task sequence in turn, calculate the change value of the objective function value of each first task sequence before and after the insertion, and obtain the difference between the second minimum change value and the minimum change value; compare the difference value corresponding to each customer, insert the customer with the largest difference value to the end of the first task sequence with the minimum change value corresponding to the customer, and delete the customer from the set to be inserted, and perform the next update operation until the set to be inserted is empty; From the updated multiple first task sequences, the first task sequences containing customers are screened out, and the warehouse closest to the last customer is inserted at the end of each of the screened first task sequences as the task sequence end point to obtain multiple initial task sequences.
5. The large-scale UAV mission planning method according to claim 1, characterized in that: The spectral clustering algorithm is used to divide multiple initial task sequences into multiple groups, including: According to the longitude and latitude coordinates of each node in each initial task sequence, and the number of customers and warehouses, the centroid coordinates of each initial task sequence are calculated by the following formula as input samples: in, Represents the mth task sequence V m The center of gravity b m The longitude and latitude coordinates of |V m | represents the number of nodes in the mth task sequence, and the nodes include customers and warehouses; x i Represents the longitude coordinate of the i-th node; y i Represents the latitude coordinate of the i-th node; The input samples, the preset feature dimensions and the number of customers in each group are input into the spectral clustering algorithm to cluster multiple initial task sequences into multiple groups.
6. The large-scale UAV mission planning method according to claim 1, characterized in that: The method of optimizing the task sequence of each group in turn by using an adaptive variable neighborhood search algorithm according to a preset neighborhood operator includes: Perform the following iterative optimization operations on the task sequence of each group: initialize the global optimal solution, the optimal solution of the previous iteration, and the weight of the neighborhood operator, and set the maximum number of executions; In each execution, a neighborhood operator is selected by the roulette wheel selection algorithm according to the weight of each neighborhood operator; the first solution of each group is calculated according to the selected neighborhood operator; the variable neighborhood descent algorithm is used for the first solution to obtain the second solution; the performance record value of the selected neighborhood operator, the global optimal solution and the optimal solution of the current iteration are updated according to the global optimal solution, the optimal solution of the current iteration, the optimal solution of the previous iteration and the second solution, and the next execution is performed after each group is updated with the global optimal solution until the maximum number of executions is reached, and the optimal solution of the previous iteration is updated with the optimal solution of the current iteration for each group, and the weight of each neighborhood operator is updated according to the preset weight update rate and the performance record value of each neighborhood operator, and the next round of iterative optimization is performed until the maximum number of iterations of the group is reached, and the global optimal solution of each group is obtained, which is the task sequence of each group after optimization.
7. The large-scale UAV mission planning method according to claim 6, characterized in that: The updating of the performance record value of the selected neighborhood operator, the global optimal solution and the optimal solution of the current round of iteration according to the global optimal solution, the optimal solution of the current round of iteration, the optimal solution of the previous round of iteration and the second solution comprises: If the optimal solution of the current iteration is empty, the second solution is used to update the optimal solution of the current iteration; If the objective function value of the second solution is less than the objective function value of the global optimal solution, then according to the value of the first weight update factor, the performance record value of the selected neighborhood operator is increased, and the global optimal solution is updated using the second solution; If the objective function value of the second solution is less than the objective function value of the optimal solution of the current iteration, then the performance record value of the selected neighborhood operator is increased according to the value of the second weight update factor, and the second solution is used to update the optimal solution of the current iteration; If the objective function value of the second solution is less than the objective function value of the optimal solution of the previous iteration, the performance record value of the selected neighborhood operator is increased according to the value of the third weight update factor; The value of the first weight update factor is greater than the value of the second weight update factor, and the value of the second weight update factor is greater than the value of the third weight update factor.
8. The large-scale UAV mission planning method according to claim 6, characterized in that: According to the preset weight update rate and the performance record value of each neighborhood operator, the weight of each neighborhood operator is updated by the following formula: oh p =r*ω p +(1-ρ)*σ p Among them, ω p represents the weight of the pth neighborhood operator, σ i represents the performance record value of the pth neighborhood operator; ρ represents the weight update rate, 0≤ρ≤1.
9. The large-scale UAV mission planning method according to claim 1, characterized in that: The neighborhood operator includes: The Or-Opt operator is used to randomly select a segment containing three consecutive customers and reinsert it into another position; The fragment exchange operator is used to exchange fragments of arbitrary length between two task sequences; The three-node position exchange operator is used to randomly select two customers c1 and c2 on a task sequence and a customer c3 on another task sequence, insert customer c3 into the position of customer c1, insert customer c1 into the position of customer c2, and insert customer c2 into the position of customer c3.
10. A large-scale UAV mission planning system, characterized in that: include: The target building module is used to build the objective function and its constraints for UAV mission planning; A task construction module is used to initialize the task sequence. According to the change of the objective function value after inserting the customer into the task sequence, each customer is inserted into the task sequence with the smallest objective function value, so as to obtain multiple initial task sequences. The task update module is used to iteratively perform the following update operations on multiple initial task sequences: A spectral clustering algorithm is used to divide multiple initial task sequences into multiple groups; according to the preset neighborhood operator, an adaptive variable neighborhood search algorithm is used to optimize the task sequence of each group in turn, and then merge them into multiple global task sequences; if the total objective function value of multiple global task sequences is less than the total objective function value of multiple initial task sequences, multiple global task sequences are used to update multiple initial task sequences; the next round of update operation is performed until the stop condition is reached, and the updated multiple initial task sequences constitute the optimal UAV task.
Citation Information
Cited By
Intelligent noise detection method based on multiple parameters
CN119860840A
Carrier delivery point and target grouping collaborative decision-making method for task planning in carrier delivery mode
CN120278557A
A collaborative decision-making method for vehicle delivery point and target grouping for task planning in vehicle delivery mode
CN120278557B
Rapid arrangement and insertion method for newly added tasks of unmanned aerial vehicle
CN122066194A
