Multi-plant protection unmanned aerial vehicle path optimization method
By optimizing the path planning of multiple agricultural drones using an improved simulated annealing algorithm, the problems of local optima and load impact in decentralized farmland operations were solved, enabling efficient and stable multi-drone collaborative operations and improving resource utilization and operational efficiency.
Patent Information
- Application Number
- CN202510946252.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-09
- Publication Date
- 2025-10-28
AI Technical Summary
Existing multi-drone path planning algorithms are prone to getting trapped in local optima, failing to effectively handle the needs of scattered farmland operations, and do not consider the impact of drone load, resulting in low resource utilization and low operational efficiency.
An improved simulated annealing algorithm is adopted, combined with K-means++ clustering, greedy algorithm and upper confidence boundary (UCB) algorithm, and constraints such as energy and drug quantity are introduced. The path planning is optimized by modeling in the form of fully connected weighted graph. Neighborhood and perturbation operator strategies are used to dynamically adjust the perturbation selection and a constraint repair algorithm based on the minimum energy increment criterion is introduced.
It improves the efficiency of multi-drone path planning, extends equipment lifespan, saves labor costs, enables efficient operation in scattered farmland, and improves resource utilization and operational stability.
Smart Images
Figure CN120848581A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of drone path planning technology, specifically a method for optimizing the paths of multiple agricultural drones. Background Technology
[0002] Pests and diseases are one of the main factors restricting agricultural production in my country, characterized by their wide impact, numerous species, and rapid spread, making them difficult to eradicate completely over a long period. Simultaneously, during crop growth, timely application of chemical fertilizers and nutrient solutions is necessary to promote growth, accelerate maturation, and increase yield. With the continuous advancement of urbanization and the intensification of population aging in my country, rural labor resources are becoming increasingly scarce, further restricting the improvement of agricultural production efficiency. Therefore, traditional agricultural practices are no longer sufficient to meet the demands of modern, efficient, and intelligent agriculture.
[0003] In recent years, the rapid development of drone technology has brought new solutions to the field of agricultural plant protection. Multi-drone collaborative operations can significantly improve plant protection efficiency, but how to rationally plan the flight paths of multiple drones and reduce path length has become a key issue. Traditional path planning methods often easily get trapped in local optima when dealing with large-scale, complex terrain, leading to low path planning efficiency.
[0004] Current path planning algorithms for agricultural drones have the following drawbacks: 1. Currently, path planning for agricultural drones mainly focuses on full-coverage path planning for single areas. However, in actual agricultural production, crop planting is often characterized by small, scattered areas rather than large, contiguous fields. This scattered planting pattern requires drones to operate continuously between multiple scattered fields. Due to the limited payload and endurance of drones, they need to frequently land during operations to replenish batteries and pesticides. Research on path planning for this situation is relatively limited. 2. Some current path planning algorithms are prone to getting trapped in local optima, which leads to decreased resource utilization and affects overall operational efficiency and economic benefits. 3. Many algorithms neglect to assess the workload of drones during path planning, resulting in individual drones being overloaded and affecting the stability of drone swarm collaborative operations. Summary of the Invention
[0005] The purpose of this invention is to provide a path optimization method for multiple agricultural drones, which can solve the technical problems of existing multiple agricultural drones in path planning that are not realistic, are prone to getting trapped in local optima, and do not consider the impact of drone load.
[0006] To achieve the above objectives, the present invention adopts the following technical solution.
[0007] A method for optimizing the paths of multiple agricultural drones includes the following steps: S1. Combining the parameters of agricultural drones and the actual pesticide dosage requirements in the fields, construct a farmland dataset and a path planning model for collaborative operations of multiple agricultural drones. S2. The K-means++ clustering algorithm, greedy algorithm, and upper confidence boundary (UCB) algorithm are introduced into the simulated annealing algorithm (SA). Combined with the path planning model in S1, an improved simulated annealing algorithm is generated. S3. Input the farmland dataset into the improved simulated annealing algorithm to realize path planning for multiple agricultural drones in a dispersed farmland scenario.
[0008] Furthermore, in S1, when constructing the farmland dataset, the quantity of farmland, geographical distribution, pesticide dosage requirements, and drone parameters are adjusted according to the actual needs of plant protection operations to construct a simulation dataset that fits the actual agricultural operation scenario.
[0009] Furthermore, in S1, the path planning model is configured with constraints on power consumption, pesticide dosage, task integrity, path uniqueness, and task balance. The relationship between farmland nodes and service points is modeled using a fully connected weighted graph.
[0010] Furthermore, in S2, during the path initialization phase, the K-means++ clustering algorithm is used to cluster the farmland operation area, and the greedy algorithm is combined to construct the initial operation path in each sub-region. The K-means++ clustering algorithm selects the initial cluster center through the D2 weight method, and the greedy algorithm generates the operation order according to the distance priority rule between nodes.
[0011] Furthermore, in S2, an upper confidence boundary algorithm is introduced as an adaptive selection mechanism for the perturbation operator, which dynamically adjusts the selection probability of the perturbation operator based on the historical average reward, selection count, and total number of experiments.
[0012] Furthermore, in S2, a constraint repair algorithm based on the minimum energy increment criterion is introduced, ensuring that the path after calling the constraint repair algorithm is always the current shortest path.
[0013] Furthermore, the location of service points is optimized through insertion and movement operations, thereby optimizing the path after calling the constraint repair algorithm.
[0014] Furthermore, in S3, the farmland dataset is input into an improved simulated annealing algorithm to derive multi-UAV planned paths.
[0015] By adopting the above technical solution, the present invention has the following beneficial effects: 1. Based on the traditional SA, this invention introduces the K-means++ clustering algorithm and the greedy algorithm, which improves the quality of the initial solution; 2. This invention enhances the algorithm's optimization capability by combining neighborhood and perturbation operators to generate new perturbation strategies, and introduces the UCB algorithm to dynamically select perturbation operators, achieving efficient local search; 3. This invention introduces a constraint repair algorithm based on the minimum energy increment criterion. The path after calling the constraint repair algorithm is always the current shortest path, which significantly reduces the path length of the improved simulated annealing algorithm. 4. This method can effectively plan multiple drones for plant protection operations in scattered farmland, improve drone operation efficiency and extend equipment lifespan, save labor costs, and provide a feasible solution to the practical problems of scattered farmland and high cost and low efficiency of small farmers' self-management in my country. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating the improved simulated annealing algorithm of this invention.
[0017] Figure 2 This is a schematic diagram of the node selection strategy based on neighborhood constraints in the invention.
[0018] Figure 3 This is a diagram showing the performance comparison of the UCB-SA, SA, GA and LNS algorithms in Case 1.
[0019] Figure 4 This is a diagram showing the performance comparison of the UCB-SA, SA, GA and LNS algorithms in Case 2.
[0020] Figure 5 This is a diagram showing the performance comparison of the UCB-SA, SA, GA and LNS algorithms in Case 3.
[0021] Figure 6 This is a diagram showing the performance comparison of the UCB-SA, SA, GA and LNS algorithms in Case 4.
[0022] Figure 7 This is a diagram showing the performance comparison of the UCB-SA, SA, GA and LNS algorithms in Case 5.
[0023] Figure 8 This is a diagram showing the performance comparison of the UCB-SA, SA, GA and LNS algorithms in Case 6.
[0024] Figure 9 This is a diagram showing the effect comparison of the improved simulated annealing algorithm after changing the number of drones in Case 6. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the features and performance of a multi-drone path optimization method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0026] Please see the appendix Figures 1-9 A method for optimizing the path of multiple agricultural drones includes the following steps.
[0027] S1. Combining the parameters of agricultural drones and the actual pesticide dosage requirements in the fields, construct a farmland dataset and a path planning model for collaborative operations of multiple agricultural drones.
[0028] When constructing the farmland dataset, the quantity of farmland, geographical distribution, pesticide dosage requirements, and drone parameters are adjusted according to the actual needs of plant protection operations to build a simulation dataset that fits the actual agricultural operation scenario.
[0029] In the path planning model, constraints such as power consumption, pesticide dosage, task integrity, path uniqueness, and task balance are set, and the relationship between farmland nodes and service points is modeled in the form of a fully connected weighted graph.
[0030] S2. The K-means++ clustering algorithm, greedy algorithm, and upper confidence boundary (UCB) algorithm are introduced into the simulated annealing algorithm (SA). Combined with the path planning model in S1, an improved simulated annealing algorithm is generated.
[0031] In the path initialization phase, the K-means++ clustering algorithm is used to cluster the farmland operation area, and the greedy algorithm is combined to construct the initial operation path in each sub-region. The K-means++ clustering algorithm selects the initial cluster center through the D2 weight method, and the greedy algorithm generates the operation order according to the distance priority rule between nodes.
[0032] An upper confidence boundary algorithm is introduced as an adaptive selection mechanism for the perturbation operator, which dynamically adjusts the selection probability of the perturbation operator based on the historical average reward, selection count, and total number of experiments.
[0033] A constraint repair algorithm based on the minimum energy increment criterion is introduced to ensure that the path after invoking the constraint repair algorithm is always the current shortest path. The path after invoking the constraint repair algorithm is optimized by using insertion and movement operations to improve the location of service points.
[0034] S3. Input the farmland dataset into the improved simulated annealing algorithm to realize path planning for multiple agricultural drones in a dispersed farmland scenario.
[0035] By inputting the farmland dataset into the improved simulated annealing algorithm, multi-UAV planned paths are obtained.
[0036] In practical implementation, this invention provides a path optimization method for multiple agricultural drones based on an improved simulated annealing algorithm, taking scattered farmland in different regions as the research object.
[0037] S1. Combining the key performance parameters of agricultural drones and the pesticide requirements of the actual fields, a path planning model for collaborative operation of multiple agricultural drones is constructed. This model comprehensively considers multiple physical and operational parameters during its design, including the drone's battery capacity, maximum payload, pesticide consumption per unit area, spraying row spacing, and flight altitude. It also comprehensively evaluates indicators such as path feasibility and task balance by constructing multi-dimensional evaluation functions. Specifically, it includes the following sub-steps.
[0038] S11. Conduct a survey on the performance parameters of mainstream agricultural drones on the market to obtain key indicators such as battery capacity, maximum payload, flight speed, flight time, and spraying width. Use these parameters as the basic physical parameters of the agricultural drones used in this method to guide subsequent modeling and optimization work.
[0039] S12. To better reflect the needs of actual agricultural operations, the CEVRP instance proposed in the IEEE WCCI 2020 competition was adapted to suit the task scenario. This included: adjusting the number and distribution of client nodes to simulate field areas of different sizes; setting the pesticide dosage requirements for each field based on the actual pesticide application rate per unit area and the geographical characteristics of the application area; and adjusting the power limit parameters of the agricultural drone based on its flight radius and battery endurance. Ultimately, a simulation dataset more closely resembling real-world agricultural operations was constructed.
[0040] S14. In the path planning model, to ensure the generated work path is realistically feasible, several constraints are further introduced, including: power constraints, drug quantity constraints, task integrity constraints, path uniqueness constraints, and task balance constraints. The path planning problem is mathematically modeled using a fully connected weighted graph to effectively represent complex path planning scenarios. It is a graph theory structure used to describe all possible connections in the system, where: The vertex set represents all nodes in the problem, where D represents a warehouse, S represents a service point, and F represents farmland. These nodes form the basic framework of the graph. This represents the edge set, containing all direct edges connecting vertices, with each edge having a weight. This is used to calculate the transportation cost between two nodes.
[0041] The constraints include: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Equation (1) defines the range of values for the variables; equation (2) ensures that each farmland is operated on only once; and equation (3) allows the drone to visit the service point multiple times. Equation (4) establishes flow conservation by ensuring that the number of input arcs at each node equals the number of flow-to arcs. Equations (5), (6), and (7) are capacity constraints to ensure that the drone's load is non-negative when it reaches any node. This indicates the amount of pesticide remaining when the drone arrives at the farmland. This indicates the amount of pesticide consumed by the drone in the farmland. This represents the maximum capacity of the medicine box. Equations (8), (9), and (10) are used to ensure that the battery charge never drops below 0, where This indicates the remaining battery power of the drone when it arrives at the farmland. This indicates the drone's battery consumption rate. This indicates the amount of electricity consumed by the drone in the farmland. Indicates battery level.
[0042] S2. This invention proposes an improved simulated annealing algorithm, such as... Figure 1 As shown, traditional SA (Self-Assessing) is selected as the basic algorithm, and K-means++ clustering and a greedy algorithm are introduced into SA to improve the quality of the initial solution. To efficiently select the perturbation strategy, the UCB (Unified Perturbation Principle) algorithm is introduced. A constraint repair algorithm based on the minimum energy increment criterion is proposed to improve solution efficiency and reduce computational complexity during optimization. Furthermore, to further improve solution quality, a service point optimization module is introduced after repair, which adjusts the location of service points to compress the path structure and improve task efficiency. Specifically, this includes the following sub-steps.
[0043] S21. SA is selected as the basic algorithm.
[0044] S22. Initialize the parameters of the traditional SA, initial temperature. Set to 30, termination temperature Set to 0.01, cooling coefficient Set to 0.99, maximum number of iterations Based on changes in the amount of farmland.
[0045] S23. In the initialization phase, traditional Optimization Strategies (SA) typically generate initial solutions using a completely random method. However, this method has certain limitations; the quality of the generated solutions may be poor, thus affecting the overall optimization effect and convergence efficiency. To overcome these problems, this invention introduces the K-means++ clustering algorithm and a greedy algorithm in the path initialization phase to improve the feasibility and quality of the initial solutions, providing a better starting point for subsequent iterative searches.
[0046] First, the K-means++ clustering algorithm was used to perform cluster analysis on the farmland area, dividing the farmland into multiple sub-regions. Each sub-region was managed by a drone. The initial centroids of the K-means++ clustering algorithm were selected using the D2 weighting method to improve the clustering effect.
[0047] Next, within each sub-region, a greedy algorithm is used to sort the nodes and construct initial job paths. The greedy algorithm's strategy is to select the nearest unvisited point as the next target point each time, until all points have been visited. This strategy, through a local optimum construction mechanism, can quickly generate initial solutions with shorter path lengths.
[0048] S24. After constructing the initial operation path, to ensure its feasibility, this invention performs constraint checks and corrections on the initial operation path. Based on the battery life and pesticide load of the agricultural drone, each path node is analyzed to determine whether the task can be completed within the resource limits. If a risk of insufficient energy is detected on the path, i.e., the power or pesticide load may run out before the task is completed, a service point will be automatically inserted at a suitable location to replenish resources. After the service point is inserted, the feasibility verification module is invoked to verify the corrected path, ensuring that the corrected path meets the requirements.
[0049] S25. A neighborhood-constrained node perturbation strategy is proposed, defining two types of neighborhoods: large neighborhoods for global exploration and small neighborhoods for local optimization. For example... Figure 2As shown, node 9 is our randomly selected initial node. For it, the purple area represents a relatively small neighborhood, while the blue area represents a relatively large neighborhood. We employed four perturbation operators: insertion, swap, inversion, and 2-opt. 2-opt reverses the path between two nodes. By combining the two neighborhoods (small and large) and the four different perturbation operators, we obtained eight perturbation strategies: insertion, swap, inversion, and 2-opt based on the small neighborhood, and insertion, swap, inversion, and 2-opt based on the large neighborhood, thereby improving search efficiency.
[0050] S26. In traditional Search Algorithms (SA), the selection of perturbation operators typically relies on fixed probabilities or preset strategies. This static approach lacks adaptability and is difficult to adjust flexibly according to the current search state, easily leading to the algorithm getting trapped in local optima or overexploring invalid neighborhoods, thus significantly reducing overall optimization efficiency. To improve the algorithm's search capability in the neighborhood space and the adaptability of the perturbation strategy, this invention introduces the UCB algorithm as a perturbation operator selection mechanism into traditional SA, achieving a dynamic balance between "exploration and exploitation." Its core idea is that in each perturbation strategy selection, not only is the historical success rate of a perturbation operator considered, but also the degree of exploration reflected by the number of times it has been selected.
[0051] Furthermore, the UCB algorithm prioritizes two types of perturbation operators: first, operators that have performed well in historical operations and have a high success rate, i.e., those with high "utilization value"; and second, operators that have not been fully tried and have been selected less frequently, to enhance the algorithm's "exploration ability." In this way, the UCB algorithm dynamically adjusts the selection probability of the perturbation strategy, enabling it to broadly explore potentially high-return operators in the early stages and gradually converge to more effective local regions in the later stages, thus balancing global search and local optimization. The formula for the UCB algorithm is as follows: , in, This indicates the optimal action selected at time step t. This represents the historical average reward of action 'a'. It is the number of times action 'a' has been selected. As an exploratory term, it decays with the number of trials, reflecting the uncertainty about action a.
[0052] S27. For infeasible paths, this invention employs a constraint repair algorithm based on the minimum energy increment criterion. This method adjusts and optimizes the overall solution by reassigning nodes that violate constraints to other feasible paths. Compared to directly inserting charging stations or adding new paths to violating paths, this strategy effectively avoids the disorderly expansion of path structure or infrastructure, thereby improving the compactness and convergence speed of the solution while controlling computational complexity.
[0053] S28. To optimize the location of service points, this invention employs two operations: insertion and movement. The algorithm first divides the overall route into independent paths for individual drones. For each path, the algorithm performs an insertion or movement operation with a preset probability. Insertion adds the nearest charging station to a randomly selected feasible location, while movement relocates an existing charging station. If a movement operation causes a flight segment to no longer meet the constraints, an additional charging station is inserted to restore feasibility. This optimization strategy dynamically optimizes the spatial distribution of charging stations while strictly satisfying the drone's endurance constraints, thus reducing the overall path length.
[0054] S29. According to the basic principles of Search Algorithm (SA), after each new solution is generated, the Metropolis acceptance criterion must be used to determine whether to accept the solution as the current solution. This criterion not only allows the acceptance of better solutions but also accepts inferior solutions with a certain probability, thereby avoiding the algorithm from getting trapped in local optima and enhancing global search capabilities. The specific formula is as follows: , in, This represents the probability of accepting a new state in the k-th iteration, where T is the current temperature parameter. This is the difference in the objective function. The acceptance probability is negatively correlated with the difference in the objective function and positively correlated with the temperature parameter. Initially, when the temperature is high, the algorithm tends to accept inferior solutions to achieve a broad search. As the temperature gradually decreases, the algorithm becomes more conservative, accepting only better solutions or slightly worse new solutions, gradually converging to the optimal region.
[0055] S3. Experiment with the farmland dataset created in S1 using an improved simulated annealing algorithm to derive the planned path. This includes the following sub-steps.
[0056] S32. A weighted sum of total path length and task balance is used as the comprehensive evaluation function, where total path length measures overall operational efficiency, and task balance reflects the rationality of task allocation among various UAVs. The specific formula is as follows: , Where K represents the number of drones. This represents the number of missions performed by the k-th drone. This represents the average number of missions performed by all drones.
[0057] To evaluate the performance of the modified simulated annealing algorithm in this invention, it is compared with traditional SA, genetic algorithms (GA), and large neighborhood search (LNS).
[0058] like Figures 3-8 As shown, six specific cases were constructed: Case 1 (2 drones, 27 farmlands), Case 2 (2 drones, 29 farmlands), Case 3 (3 drones, 36 farmlands), Case 4 (4 drones, 50 farmlands), Case 5 (5 drones, 80 farmlands), and Case 6 (6 drones, 105 farmlands). Experimental results of different path planning algorithms were compared within these specific cases.
[0059] We can see that the GA algorithm exhibits a stable but extremely slow convergence speed, mainly because the perturbations introduced by its exchange and mutation operators are relatively small, resulting in insufficient exploration capability during the search process. For large-scale problems, this inefficient exploration method further limits the algorithm's optimization efficiency.
[0060] Compared to other algorithms, LNS converges faster and runs in less time because it only modifies a portion of the solution rather than recalculating the entire solution. However, the rapid convergence of LNS can lead to premature stagnation in local optima, reducing its global search capability.
[0061] Compared to the traditional Simulated Annealing (SA), the improved Simulated Annealing (UCB-SA) algorithm consistently outperforms in initial solution quality, primarily due to the effectiveness of the initial clustering algorithm. While both algorithms converge at nearly the same speed, the improved SA, through a more effective perturbation selection strategy, is able to find superior solutions, achieving an average performance improvement of 2%–11% compared to the traditional SA.
[0062] Therefore, the improved simulated annealing algorithm proposed in this invention outperforms the comparative algorithms in terms of path optimization, exhibiting higher stability and overall performance, thus verifying the effectiveness and applicability of the method in real-world scenarios.
[0063] To further verify the effectiveness of the algorithm of this invention in unmanned aerial vehicle (UAV) task load balancing, for Case 6 above, the number of UAVs was changed to 4, and path planning was performed again.
[0064] like Figure 9 As shown in the figure, the pink squares represent warehouses, the yellow triangles represent service points, the green circles represent farmland, and the connecting lines of different colors represent the operating trajectories of different drones. It can be seen that regardless of the number of drones, the method of this invention can allocate tasks to each drone relatively evenly, with the task load fluctuation range controlled within 3, the standard deviation not exceeding 1.090, and the coefficient of variation always below 0.038.
[0065] In summary, this invention systematically studies a novel UAV path planning problem, which aims to find the optimal path for multiple UAVs while maintaining task balance. To address this problem, this invention proposes an improved simulated annealing algorithm. This method employs an initialization strategy combining k-means++ clustering and a greedy algorithm, a perturbation strategy combining neighborhood and perturbation operators, a selection strategy based on the UCB algorithm, and a constraint repair algorithm based on the minimum energy increment criterion. To demonstrate the performance of the proposed method, we evaluate it on multiple case studies designed according to the problem scenario. The results prove that the proposed algorithm can plan better paths for multiple agricultural UAVs while satisfying the constraints, providing a direct application reference for multiple agricultural UAVs operating in dispersed farmland.
[0066] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for optimizing the path of multiple agricultural drones, characterized in that: Includes the following steps, S1. Combining the parameters of agricultural drones and the actual pesticide dosage requirements in the fields, construct a farmland dataset and a path planning model for collaborative operations of multiple agricultural drones. S2. Introduce the K-means++ clustering algorithm, greedy algorithm, and upper confidence boundary algorithm into the simulated annealing algorithm, and combine them with the path planning model in S1 to generate an improved simulated annealing algorithm; S3. Input the farmland dataset into the improved simulated annealing algorithm to realize path planning for multiple agricultural drones in a dispersed farmland scenario.
2. The method for optimizing the path of multiple agricultural drones as described in claim 1, characterized in that: In S1, when constructing the farmland dataset, the quantity of farmland, geographical distribution, pesticide dosage requirements, and drone parameters are adjusted according to the actual plant protection operation needs to construct a simulation dataset that fits the actual agricultural operation scenario.
3. The method for optimizing the path of multiple agricultural drones as described in claim 1, characterized in that: In S1, the path planning model sets constraints on power consumption, pesticide dosage, task integrity, path uniqueness, and task balance. The relationship between farmland nodes and service points is modeled in the form of a fully connected weighted graph.
4. The method for optimizing the path of multiple agricultural drones as described in claim 1, characterized in that: In S2, during the path initialization phase, the K-means++ clustering algorithm is used to cluster the farmland operation area, and the greedy algorithm is combined to construct the initial operation path in each sub-region. The K-means++ clustering algorithm selects the initial cluster center through the D2 weight method, and the greedy algorithm generates the operation order according to the distance priority rule between nodes.
5. The method for optimizing the path of multiple agricultural drones as described in claim 1, characterized in that: In S2, the upper confidence boundary algorithm is introduced as an adaptive selection mechanism for the perturbation operator, which dynamically adjusts the selection probability of the perturbation operator based on the historical average reward, selection count, and total number of experiments.
6. The method for optimizing the path of multiple agricultural drones as described in claim 1, characterized in that: In S2, a constraint repair algorithm based on the minimum energy increment criterion is introduced, so that the path after calling the constraint repair algorithm is always the current shortest path.
7. The method for optimizing the path of multiple agricultural drones as described in claim 6, characterized in that: The location of service points is optimized by using insertion and movement operations, thus optimizing the path after calling the constraint repair algorithm.
8. The method for optimizing the path of multiple agricultural drones as described in claim 1, characterized in that: In S3, the farmland dataset is input into the improved simulated annealing algorithm to derive multi-UAV planned paths.