Farm heterogeneous unmanned aerial vehicle collaborative planning method and system based on double-layer auction

Abstract

The invention provides a farm heterogeneous unmanned aerial vehicle cooperative planning method based on double-layer auction, and the method comprises the steps: 1, dividing heterogeneous unmanned aerial vehicles into different types based on an improved k-means clustering algorithm; step 2, adaptive task clusters are generated for different types of heterogeneous unmanned aerial vehicles based on an improved k-means clustering algorithm, and the scale of the task clusters is dynamically adjusted to match the types of the heterogeneous unmanned aerial vehicles; step 3, tasks are allocated for each unmanned aerial vehicle through a two-stage mechanism of same-type auction and cross-type compensation auction, the same-type auction only allows the heterogeneous unmanned aerial vehicles matched with the task cluster type to bid, and the cross-type compensation auction allows other types of heterogeneous unmanned aerial vehicles to participate in the remaining task bid; and step 4, optimizing the unmanned aerial vehicle path in each unmanned aerial vehicle distribution task based on a simulated annealing algorithm, and generating a global optimal path. The resource utilization rate and the task completion rate of the unmanned aerial vehicle are effectively improved, and a more intelligent and efficient solution is provided for intelligent agricultural data monitoring.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication technology, and in particular relates to a farm heterogeneous UAV collaborative planning method and system based on double-layer auction. Background Art

[0002] With the accelerated development of agricultural modernization, medium- and large-scale smart agriculture scenarios are becoming increasingly prevalent. These smart farms are not only vast but also feature detailed functional zoning. Different regions cultivate diverse crop types and significant variations in cultivated area. Soil characteristics vary significantly depending on factors such as geography and historical farming practices. For example, some areas have acidic soils, making them suitable for cash crops like blueberries, while others have fertile soils, making them more suitable for food crops.

[0003] In this complex and ever-changing agricultural environment, data monitoring is crucial for achieving precision agriculture. Drones, with their efficiency, flexibility, and multi-sensor fusion capabilities, have become a crucial tool for agricultural data monitoring. Compared to traditional manual monitoring methods, drones can efficiently cover the entire farm area in a short period of time, quickly completing initial data collection and significantly reducing time costs. Furthermore, drones can flexibly adjust flight paths and monitoring strategies based on topographical features and crop distribution, enabling precise data collection. Furthermore, drones equipped with advanced equipment such as high-resolution cameras and multispectral sensors can capture multi-dimensional data on crop growth, soil fertility, and pest and disease occurrence, providing accurate and forward-looking decision support for agricultural production.

[0004] However, when allocating tasks to heterogeneous drones, traditional auction algorithms often suffer from problems such as unbalanced resource utilization, insufficient task allocation, and partial task retention, making it difficult to meet the requirements of smart agriculture for efficient, comprehensive, and secure data monitoring. Summary of the Invention

[0005] This paper proposes a collaborative planning method and system for heterogeneous drones on farms based on a two-tier auction. First, an improved k-means clustering algorithm is used to classify the heterogeneous drones into three types, generating adaptive task clusters for each type. Subsequently, a two-tier auction mechanism is introduced for task allocation, ensuring that tasks are appropriately matched to each type of drone. Finally, a simulated annealing algorithm is used to optimize flight paths, achieving local obstacle avoidance and global path coordination, thereby improving drone resource utilization, task completion rate, and monitoring safety.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: A collaborative planning method for heterogeneous drones in farms based on a double-layer auction includes the following steps: Step 1. Based on the improved k-means clustering algorithm, heterogeneous UAVs are divided into different types; Step 2 generates adaptive task clusters for different types of heterogeneous UAVs based on the improved k-means clustering algorithm, and dynamically adjusts the task cluster size to match the heterogeneous UAV types; Step 3. Assign tasks to each heterogeneous UAV through a two-stage mechanism of same-type bidding and cross-type compensatory auction. The same-type auction only allows heterogeneous UAVs matching the task cluster type to bid, while the cross-type compensatory auction allows heterogeneous UAVs of other types to bid for the remaining tasks. Step 4. Optimize the UAV paths in each heterogeneous UAV assignment task based on the simulated annealing algorithm to generate the global optimal path.

[0007] Furthermore, the types of heterogeneous drones in step 1 include: high-range-low-capacity type, balanced type, and low-range-high-capacity type.

[0008] Furthermore, the step 1 includes: Based on the normalized range and capacity feature vectors, the initial centers of different types of heterogeneous UAVs are selected. Calculate the range-capacity weighted distance of each heterogeneous UAV to different types of initial centers, and assign each heterogeneous UAV to the center cluster with the smallest range-capacity weighted distance; After each iteration, the central cluster properties are checked and type constraints are enforced by removing or adjusting members until the clustering converges.

[0009] Furthermore, the calculation formula of the range-capacity weighted distance is:

[0010] in, represents the range-capacity weighted distance of UAV individual i to the initial center of the kth heterogeneous UAV, Represents individual drones, Represents three types of heterogeneous drones, represents the range weight of the kth heterogeneous UAV, represents the capacity weight of the kth type of heterogeneous UAVs, and represents the mean range and capacity of the kth type of heterogeneous UAVs, and represents the normalized range and normalized capacity, is the range of drone individual i, is the data capacity of drone individual i; is the maximum range among all heterogeneous UAVs, It is the maximum data capacity among all heterogeneous drones.

[0011] Furthermore, the step 2 includes: The weighted features of the task cluster are set according to the types of heterogeneous UAVs, and the initial center of the task is selected as the center of the task cluster; Calculate the distance between the task point and the task cluster center - the data volume weighted distance, and assign each task point to the task cluster with the smallest distance - data volume weighted distance; After each iteration, the task cluster attributes are checked and the type constraints are enforced by removing or absorbing task points until the clustering converges.

[0012] Furthermore, the distance-data volume weighted distance is calculated as follows:

[0013] in, Indicates that task point j is The distance between the center of the task cluster and the weighted distance of the data volume, is the distance feature weight, is the data feature weight, is the original distance between mission point j and the UAV base, is the data size of task point j, Indicates the The original mean distance between the centers of the task clusters, Indicates the The mean amount of data at the center of the task cluster.

[0014] Furthermore, the step 3 includes: In the same-type bidding stage, the bid prices of heterogeneous drones matching the task cluster type are calculated, and the bidder with the lowest bid price among the heterogeneous drones matching the task cluster type is selected as the winner; In the cross-type compensation auction stage, the remaining tasks are bid by other types of heterogeneous drones. The bid prices of other types of heterogeneous drones are calculated, and the bidder with the lowest bid price among the bid prices of other types of heterogeneous drones is selected as the winner.

[0015] Furthermore, the calculation formula for the bidding price of heterogeneous UAVs matching the task cluster type is:

[0016] in, represents the bidding price of heterogeneous UAVs matching the task cluster type, Heterogeneous UAVs matching the mission cluster type Execute the task The cost, is a random disturbance term, and denote the cost weights of range and capacity respectively, is the random disturbance threshold; The formula for calculating the bidding price of other types of heterogeneous drones is:

[0017] in, represents the bidding price of other types of heterogeneous drones, For other types of heterogeneous drones Execute the task The cost, The constraints are: Maximum range constraint ; Maximum capacity constraint .

[0018] Furthermore, the step 4 includes: The initial path is a closed route, and candidate paths are generated by inserting, deleting or moving detour points; The energy of candidate new paths generated based on annealing energy function , the annealing energy function includes a length term, an obstacle avoidance penalty term, and a range constraint penalty term; The path is updated according to the energy difference and probability acceptance criteria until the temperature drops to a threshold or the maximum number of iterations is reached.

[0019] In another aspect, the present invention provides a farm heterogeneous drone collaborative planning system based on a double-layer auction, comprising: UAV pre-classification module, which is used to classify heterogeneous UAVs into different types based on the improved k-means clustering algorithm; Adaptive task cluster generation module, which is used to generate adaptive task clusters for different types of heterogeneous UAVs based on the improved k-means clustering algorithm, and dynamically adjust the task cluster size to match the heterogeneous UAV types; A two-tiered auction task allocation module, which is used to assign tasks to each drone through a two-stage mechanism of same-type bidding and cross-type compensatory auction. The same-type auction only allows heterogeneous drones matching the task cluster type to bid, while the cross-type compensatory auction allows other heterogeneous drones to participate in the bidding for the remaining tasks. The simulated annealing path optimization module is used to optimize the drone paths in each drone allocation task based on the simulated annealing algorithm to generate the global optimal path.

[0020] Compared with the prior art, the present invention has the following beneficial effects: 1. Compared with the traditional k-means clustering algorithm that randomly selects initial centers and processes without constraints, this method selects the initial centers of different types based on the characteristics of heterogeneous drones and task points with different data volumes, and uses hard constraints to force adjustments after each round of iteration to accurately represent the core features of different types.

[0021] 2. Compared to traditional k-means clustering, which cannot distinguish feature importance when using Euclidean distance, this invention defines different weights for each type of drone and task cluster, calculates the weighted Euclidean distance, amplifies the influence of the core features of the type, and guides the clustering results to accelerate towards the optimal solution.

[0022] 3. Compared with traditional auction algorithms that are prone to falling into local optimality, resulting in waste of some drone resources and missed tasks, the present invention uses a clustering algorithm and a double-layer auction deep coordination mechanism to force tasks to be assigned to adaptive drones first, and allows other types of drones to take over unassigned tasks when their capability constraints are met.

[0023] 4. Use the simulated annealing algorithm to plan the path, introduce path length indicators, obstacle avoidance penalties, and range constraint penalties to optimize the path length while balancing obstacle avoidance requirements and global path consistency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0025] Figure 1 This is a flow chart of an embodiment of the present invention; Figure 2 This is a schematic diagram of the drone pre-classification process in an embodiment of the present invention; Figure 3 A schematic diagram of the process of generating an adaptation task cluster in an embodiment of the present invention; Figure 4 A schematic diagram of the double-tier auction mechanism process in an embodiment of the present invention; Figure 5 This is a schematic diagram of the simulated annealing optimization path flow in an embodiment of the present invention; Figure 6 Schematic diagram of the system architecture in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0027] Example 1 like Figure 1As shown, this embodiment provides a farm heterogeneous UAV collaborative planning method based on double-layer auction. Assume that heterogeneous UAVs are in a The drones are dispatched from farms to different areas to complete data monitoring tasks. This task is divided into three stages: first, the heterogeneous drones need to be pre-classified, then each drone needs to be assigned a task, and finally, each drone's path needs to be planned. During this process, the drones need to optimize the task allocation mechanism to improve resource utilization and task completion rate.

[0028] The specific steps include: Step 1. Based on the improved k-means clustering algorithm, heterogeneous UAVs are divided into different types; In this embodiment, heterogeneous drones are classified into three categories: Type 1 (high range, low capacity), Type 2 (balanced), and Type 3 (low range, high capacity). This process mainly uses the improved k-means clustering algorithm for pre-classification. The specific implementation method is as follows: First determine the task point set , where each task point attribute is expressed as , containing the coordinates of the mission point And the amount of data at the task point .

[0029] Drone collection , containing the starting coordinates of the drone , the maximum range of the drone and the maximum data capacity of the drone . Define the type of eigenvector , representing the normalized range and normalized capacity, ensuring that the range and capacity are comparable in numerical range to avoid a single dimension dominating the clustering results. Next, we use the improved k-means clustering algorithm to classify heterogeneous drones. The algorithm is implemented as follows: (1) First, select the initial centers for the three types of drone orientations:

[0030] in is the normalized mean distance, is the normalized capacity mean. The above formula indicates that for Type1 ( ),choose (range-to-capacity ratio) is the largest and The smallest drone as the initial center ;For Type2( ), select the drone with the range and capacity closest to the mean as the initial center ; For Type3 ( ),choose Largest and The smallest drone as the initial center ; (2) Calculate the range-capacity weighted distance of each drone to different types of centers:

[0031] In the above formula: Represents the range weights of three types of drones and capacity weight different, and represents the mean range and capacity of the k-th category UAV.

[0032] (3) According to the range-capacity weighted distance, each drone is assigned to the cluster with the smallest weighted distance: , Indicates that the drone Assign the type label to the drone cluster with the smallest distance, and dynamically update the mean of the drone features in the cluster as the new center: in Indicates the Class drone collection.

[0033] (4) Next, we need to ensure that the characteristics of each class conform to the predefined type, which requires checking the drone type constraints: (In the following formula 、 、 、 Indicates the threshold value)

[0034] If the average capacity of Type 1 exceeds the standard, the UAV with the largest capacity in the cluster will be moved to Type 3; if 3's average range exceeds the standard, so the drone with the longest range in the cluster is moved to 1. If 2. If the clustering is unbalanced, remove the drone with the largest deviation from the mean and reallocate it to another cluster. After the clustering converges or the maximum number of iterations is reached, the iteration is terminated, completing the drone pre-classification.

[0035] like Figure 2 As shown, the specific steps are: Step 101: Process starts, input heterogeneous drone data .

[0036] Step 102: Calculate the heterogeneous UAV feature vector.

[0037] Step 103: Directionally select the initial centers of three types of drones.

[0038] Step 104: Calculate the dynamic weighted distance between each UAV and the initial center.

[0039] Step 105: Assign the drone to the type with the closest dynamic weighted distance.

[0040] Step 106: Update the cluster center.

[0041] Step 107: Check the type constraints of the drone; if the type constraints are met, go to step 109.

[0042] Step 108: Modify the drone members of this type.

[0043] Step 109: Determine whether convergence has occurred or the maximum number of iterations has been reached.

[0044] Step 110: Output the heterogeneous drone classification results.

[0045] Step 111: The process ends.

[0046] Adaptive task cluster generation: Generate three adaptive task clusters for the three types of drones that have been classified. This process mainly uses the improved k-means clustering algorithm to generate the three task clusters.

[0047] Step 2 generates adaptive task clusters for different types of heterogeneous UAVs based on the improved k-means clustering algorithm, and dynamically adjusts the task cluster size to match the heterogeneous UAV types; After pre-classifying the drones into three types, it is necessary to generate an adaptive task cluster for each type of drone. First, for each task point , according to its coordinates and the initial coordinates of the drone , calculate the distance between the mission point and the UAV base , and according to the type of drone Calculate weighted features: ,in is the distance feature weight, is the feature weight of data volume. Next, we use the improved k-means clustering algorithm to generate three adaptive task clusters. The algorithm is implemented as follows: (1) First, select the initial mission center for different types of UAVs :

[0048] For Type 1 UAVs, the features of long-distance mission points are directly amplified by high distance weights to ensure that the first mission center represents long-distance characteristics; for Type 2 UAVs, the mission point with the highest comprehensive score of distance and data volume is selected; for Type 3 UAVs, the features of high-data volume mission points are directly amplified by high data volume weights to ensure that the first mission center represents high-data volume characteristics.

[0049] (2) Calculation task points To cluster center The weighted original distance - data volume distance, Respectively represent the mean of the k-th task cluster center in the original distance and data volume dimensions:

[0050] (3) According to the original distance minus the weighted distance of the data volume, each task point is assigned to the cluster with the smallest weighted distance: , Indicates that the task Assign the type label to the task cluster with the smallest distance, and dynamically update the center of each cluster to the feature mean of the task points in the cluster: ,in Indicates the Class task cluster.

[0051] (4) After each iteration, it is necessary to check whether the task cluster attributes conform to the drone type definition, where the type constraint is: (In the following formula 、 、 、 is the threshold)

[0052] Describe the low k task clusters, for (1) Task cluster: if the distance does not meet the standard, the 10% task points closest to the cluster will be removed; if the data volume exceeds the standard, the 10% task points with the largest data volume will be removed; (2) The task cluster needs to remove the task point that deviates the most from the balance; (3) If the data volume of a task cluster does not meet the requirements, it absorbs high-data-volume task points from other clusters. After the clustering converges or the maximum number of iterations is reached, the iteration is terminated, and the generation of adaptive task clusters for different types of drones is completed.

[0053] Further introduction to step 2, such as Figure 3 As shown, the specific steps are: Step 201: Process starts, enter task point data .

[0054] Step 202: Calculate weighted features for the mission point data based on the three types of drones .

[0055] Step 203: Select three initial centers of the task clusters based on the characteristics of the corresponding types of UAVs.

[0056] Step 204: Calculate the weighted distance between each task point and the task cluster center.

[0057] Step 205: Assign the task point to the task cluster center with the smallest weighted distance.

[0058] Step 206: Update the task cluster center.

[0059] Step 207: Check the type constraints of different task clusters. If they meet the type constraints, go to step 209; if not, go to the next step.

[0060] Step 208: Modify cluster members.

[0061] Step 209: Determine whether convergence has occurred or the maximum number of iterations has been reached.

[0062] Step 210: The process ends.

[0063] Step 3. Assign tasks to each UAV through a two-stage mechanism of same-type bidding and cross-type compensatory auction. The same-type auction only allows heterogeneous UAVs matching the task cluster type to bid, while the cross-type compensatory auction allows heterogeneous UAVs of other types to bid for the remaining tasks. The two-tier auction algorithm used is as follows: A. Same type of bidding In this phase, only drones that match the mission cluster type are allowed to participate in the bidding, balancing the amount of data and the range. Execute the task The cost is calculated by the following formula:

[0064] and Represents the cost weight of range and capacity, satisfying ; Indicates drone The maximum range, Indicates drone The bid price submitted by the drone is the cost plus the random disturbance term : The constraints are: Maximum range constraint ; Maximum capacity constraint .

[0065] For each task , select the drone with the lowest bid price as the winning bid: , Indicates a task The successful bidder pays the bid price .

[0066] B. Cross-Type Compensation Auctions This auction is designed to handle tasks that have not been assigned to drones of the same type, allowing drones of other types to participate in the bidding and improve the task completion rate. , other types of drones ( ) Submit bid price: The constraints are: Maximum range constraint ; Maximum capacity constraint The drone with the lowest bid price is selected as the winner. .

[0067] Further introduction to step 3, such as Figure 4 As shown, the specific steps are: Step 301: The process starts.

[0068] Step 302: UAVs of the same type submit bids.

[0069] Step 303: Determine whether the range and capacity restrictions are met. If not, return to step 302; if satisfied, proceed to the next step.

[0070] Step 304: Select the lowest bidder as the winning bidder.

[0071] Step 305: Determine whether there are any drones of this type left. If yes, return to step 302; if not, proceed to the next step.

[0072] Step 306 : Determine whether all tasks have been assigned; if all tasks have been assigned, go to step 311 .

[0073] Step 307: Other types of drones submit bids.

[0074] Step 308: Determine whether the range and capacity constraints are met.

[0075] Step 309: Select the lowest bidder as the winning bidder.

[0076] Step 310: Determine whether there are any remaining drones.

[0077] Step 311: If there are no drones left, the process ends.

[0078] Simulated Annealing Path Planning and Obstacle Avoidance: After each UAV is assigned a mission, a path needs to be planned and obstacle avoidance implemented. This process primarily involves optimizing global path length and coordinating local obstacle avoidance.

[0079] Step 4. Optimize the UAV paths in each UAV assignment task based on the simulated annealing algorithm to generate the global optimal path.

[0080] After assigning tasks to each drone, we use the simulated annealing algorithm to optimize the path and achieve obstacle avoidance. The simulated annealing algorithm used is as follows: (1) The initial path is a closed path , where B represents the coordinates of the drone base, Indicates the coordinates of the mission point that the drone needs to perform.

[0081] (2) During the iterative optimization process, candidate paths are generated by inserting / deleting / moving detour points through domain operations, and the annealing energy function is used to calculate and , is the annealing energy of the generated candidate path, is the annealing energy of the initial path and the difference is The annealing energy function includes path length term, obstacle avoidance penalty term and range constraint penalty term:

[0082] In the above formula: Represents a path segment The Euclidean distance of To avoid obstacles, the penalty weight is set to force the path to avoid obstacles. Represents each edge of the obstacle polygon, is the weight of the voyage constraint penalty item Represents the obstacle indicator function, which calculates the intersection of the path segment and each edge of the obstacle polygon. If there is an intersection, the path is marked as conflicting with the obstacle and the indicator function value is 1, otherwise it is 0. Indicates the maximum range of the drone.

[0083] Update the path according to the following rules and cool down after the update .

[0084] (3) When the temperature Or when the maximum number of iterations is reached, the optimal path is output as For drone bases, The detour point inserted in the optimal path, For the mission point.

[0085] Further introduction to step 4, such as Figure 5 As shown, the specific steps are: Step 401: The process starts.

[0086] Step 402: Plan the initial path as a straight line path.

[0087] Step 403: Setting the initial annealing temperature.

[0088] Step 404: Apply domain operations to generate new paths.

[0089] Step 405: Calculate the energy difference between the new path and the original path.

[0090] Step 406 : Determine whether the energy difference is less than 0; if the energy difference is less than 0, go to step 409 .

[0091] Step 407: If the energy difference is greater than 0, the new path is accepted with probability. If the new path is accepted, go to step 409.

[0092] Step 408 : If the new path is not accepted, retain the original path and go to step 410 .

[0093] Step 409: Accept the new path.

[0094] Step 410: Update the current latest path.

[0095] Step 411: Cooling down.

[0096] Step 412 : Determine whether the temperature meets the termination condition (drops to the minimum temperature or reaches the maximum number of iterations). If not, return to step 404 .

[0097] Step 413: Output the optimal path.

[0098] Step 414: The process ends.

[0099] The final allocation and path are then output.

[0100] Example 2 like Figure 6 The present embodiment provides a heterogeneous drone collaborative planning system based on a double-layer auction farm, including: UAV pre-classification module, which is used to classify heterogeneous UAVs into different types based on the improved k-means clustering algorithm; Adaptive task cluster generation module, which is used to generate adaptive task clusters for different types of heterogeneous UAVs based on the improved k-means clustering algorithm, and dynamically adjust the task cluster size to match the heterogeneous UAV types; A two-tiered auction task allocation module, which is used to assign tasks to each drone through a two-stage mechanism of same-type bidding and cross-type compensatory auction. The same-type auction only allows heterogeneous drones matching the task cluster type to bid, while the cross-type compensatory auction allows other heterogeneous drones to participate in the bidding for the remaining tasks. The simulated annealing path optimization module is used to optimize the drone paths in each drone allocation task based on the simulated annealing algorithm to generate the global optimal path.

[0101] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0102] It should be understood that the above description of the preferred embodiments is relatively detailed and cannot be considered as limiting the scope of protection of the present invention. It is not necessary and impossible to list all embodiments here. Under the guidance of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which fall within the scope of protection of the present invention. The scope of protection of the present invention shall be based on the attached claims.

Claims

1. A collaborative planning method for heterogeneous drones in farms based on double-layer auction, characterized by: The following steps are involved: Step 1. Based on the improved k-means clustering algorithm, heterogeneous UAVs are divided into different types; Step 2 generates adaptive task clusters for different types of heterogeneous UAVs based on the improved k-means clustering algorithm, and dynamically adjusts the task cluster size to match the heterogeneous UAV types; Step 3. Assign tasks to each heterogeneous UAV through a two-stage mechanism of same-type bidding and cross-type compensatory auction. The same-type auction only allows heterogeneous UAVs matching the task cluster type to bid, while the cross-type compensatory auction allows heterogeneous UAVs of other types to bid for the remaining tasks. Step 4. Optimize the UAV paths in each heterogeneous UAV assignment task based on the simulated annealing algorithm to generate the global optimal path.

2. The method for collaborative planning of heterogeneous drones in farms based on double-layer auction according to claim 1 is characterized in that: The heterogeneous UAV types in step 1 include: high-range-low-capacity type, balanced type, and low-range-high-capacity type.

3. The method for collaborative planning of heterogeneous drones in farms based on double-layer auction according to claim 2 is characterized in that: The step 1 comprises: Based on the normalized range and capacity feature vectors, the initial centers of different types of heterogeneous UAVs are selected. Calculate the range-capacity weighted distance of each heterogeneous UAV to different types of initial centers, and assign each heterogeneous UAV to the center cluster with the smallest range-capacity weighted distance; After each iteration, the central cluster properties are checked and type constraints are enforced by removing or adjusting members until the clustering converges.

4. The method for collaborative planning of heterogeneous drones in farms based on double-layer auction according to claim 3 is characterized in that: The calculation formula of the range-capacity weighted distance is: in, represents the range-capacity weighted distance of UAV individual i to the initial center of the kth heterogeneous UAV, Represents individual drones, Represents three types of heterogeneous drones, represents the range weight of the kth heterogeneous UAV, represents the capacity weight of the kth type of heterogeneous UAVs, and represents the mean range and capacity of the kth type of heterogeneous UAVs, and represents the normalized range and normalized capacity, is the range of drone individual i, is the data capacity of drone individual i; is the maximum range among all heterogeneous UAVs, It is the maximum data capacity among all heterogeneous drones.

5. The method for collaborative planning of heterogeneous drones in farms based on double-layer auction according to claim 2 is characterized in that: The step 2 includes: The weighted features of the task cluster are set according to the types of heterogeneous UAVs, and the initial center of the task is selected as the center of the task cluster; Calculate the distance between the task point and the task cluster center - the data volume weighted distance, and assign each task point to the task cluster with the smallest distance - data volume weighted distance; After each iteration, the task cluster attributes are checked and the type constraints are enforced by removing or absorbing task points until the clustering converges.

6. The method for collaborative planning of heterogeneous drones in farms based on double-layer auction according to claim 5 is characterized in that: The calculation formula of the distance-data volume weighted distance is: in, Indicates that task point j is The distance between the center of the task cluster and the weighted distance of the data volume, is the distance feature weight, is the data feature weight, is the original distance between mission point j and the UAV base, is the data size of task point j, Indicates the The original mean distance between the centers of the task clusters, Indicates the The mean amount of data at the center of the task cluster.

7. The method for collaborative planning of heterogeneous drones in farms based on double-layer auction according to claim 6 is characterized in that: The step 3 comprises: In the same-type bidding stage, the bid prices of heterogeneous drones matching the task cluster type are calculated, and the bidder with the lowest bid price among the heterogeneous drones matching the task cluster type is selected as the winner; In the cross-type compensation auction stage, the remaining tasks are bid by other types of heterogeneous drones. The bid prices of other types of heterogeneous drones are calculated, and the bidder with the lowest bid price among the bid prices of other types of heterogeneous drones is selected as the winner.

8. The method for collaborative planning of heterogeneous drones in farms based on double-layer auction according to claim 7 is characterized in that: The calculation formula for the bidding price of heterogeneous UAVs matching mission cluster types is: in, represents the bidding price of heterogeneous UAVs matching the task cluster type, Heterogeneous UAVs matching the mission cluster type Execute the task The cost, is a random disturbance term, and denote the cost weights of range and capacity respectively, is the random disturbance threshold; The formula for calculating the bidding price of other types of heterogeneous drones is: in, represents the bidding price of other types of heterogeneous drones, For other types of heterogeneous drones Execute the task The cost, The constraints are: Maximum range constraint ; Maximum capacity constraint .

9. The method for collaborative planning of heterogeneous drones in farms based on double-layer auction according to claim 2 is characterized in that: The step 4 comprises: The initial path is a closed route, and candidate paths are generated by inserting, deleting or moving detour points; The energy of candidate new paths generated based on annealing energy function , the annealing energy function includes a length term, an obstacle avoidance penalty term, and a range constraint penalty term; The path is updated according to the energy difference and probability acceptance criteria until the temperature drops to a threshold or the maximum number of iterations is reached.

10. A farm heterogeneous drone collaborative planning system based on double-layer auction, characterized by: include: UAV pre-classification module, which is used to classify heterogeneous UAVs into different types based on the improved k-means clustering algorithm; Adaptive task cluster generation module, which is used to generate adaptive task clusters for different types of heterogeneous UAVs based on the improved k-means clustering algorithm, and dynamically adjust the task cluster size to match the heterogeneous UAV types; A two-tiered auction task allocation module, which is used to assign tasks to each drone through a two-stage mechanism of same-type bidding and cross-type compensatory auction. The same-type auction only allows heterogeneous drones matching the task cluster type to bid, while the cross-type compensatory auction allows other heterogeneous drones to participate in the bidding for the remaining tasks. The simulated annealing path optimization module is used to optimize the UAV paths in each UAV assignment task based on the simulated annealing algorithm to generate the global optimal path; The farm heterogeneous drone collaborative planning system based on double-layer auction is used to execute the steps in the farm heterogeneous drone collaborative planning method based on double-layer auction as described in any one of claims 1-9.

Images