A multi-unmanned aerial vehicle distributed energy-aware task auction and four-dimensional space-time corridor cooperative planning method and system for farmland operation

Abstract

This invention relates to the field of multi-UAV cooperative control, and discloses a distributed energy-sensing task auction and four-dimensional spatiotemporal corridor cooperative planning method for farmland operations. It proposes the DEATA distributed energy-sensing auction algorithm, which designs a multi-dimensional bidding vector including energy prediction margin and path conflict cost, and achieves energy load balancing task allocation through consensus bundle iteration. The CAST-DP four-dimensional spatiotemporal corridor dynamic planning algorithm describes dedicated flight corridors using capsules and time windows, supporting rapid conflict resolution through local pairwise negotiation. When a UAV deviates from the corridor or has insufficient power, it only performs millisecond-level local renegotiation or task reassignment with affected neighbors. Simulation and field tests show that this invention improves the task completion rate to over 97%, compresses the renegotiation response time to 143ms, and achieves collision-free operation, significantly improving the continuity, safety, and real-time response capability of multi-UAV farmland operations.

Description

Technical Field

[0001] This invention relates to the field of multi-UAV cooperative control, specifically to a method and system for multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor cooperative planning for farmland operations. Background Technology

[0002] Multi-UAV collaborative operations are an important development direction for precision agriculture. Existing technologies neglect real-time energy state changes during flight in the task allocation phase, leading to a high rate of mid-flight return-to-home interruptions. Furthermore, in the path planning phase, task allocation and conflict resolution are executed sequentially, resulting in slow response to sudden disturbances. Specifically, existing methods have the following shortcomings: 1. Task allocation relies on initial power estimation, failing to adapt to dynamic energy consumption factors such as sudden wind speed changes and load variations, causing UAVs to return mid-flight and disrupting coverage continuity; 2. Path planning and conflict resolution are executed sequentially, requiring global replanning in the event of sudden disturbances, resulting in a response delay exceeding 1 second and insufficient safety margin during high-speed swarm flight. Therefore, this invention proposes a multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning method and system for farmland operations. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for auctioning multi-UAV distributed energy sensing tasks and co-planning four-dimensional spatiotemporal corridors for farmland operations, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning for farmland operations, comprising the following steps:

[0005] Step S1: Divide the work area into multiple rectangular work blocks to form a task pool; each drone broadcasts a state vector containing real-time remaining battery power and wind speed and direction at a fixed period.

[0006] Step S2: Execute the Distributed Energy Sensing Task Auction Algorithm (DEATA). Each UAV independently calculates a multi-dimensional bidding vector for unallocated blocks and performs distributed iterations through an improved consensus bundle algorithm to complete task allocation and initial path conflict resolution. The bidding vector includes at least an energy prediction margin component and a path segment conflict cost component.

[0007] Step S3: Receive the DEATA allocation scheme and execute the four-dimensional spatiotemporal corridor dynamic planning algorithm CAST-DP to plan a dedicated flight corridor for each UAV in the four-dimensional spatiotemporal space composed of several corridor segments in the three-dimensional spatial coordinates and time t; the corridor segments are described by three-dimensional convex hull time intervals.

[0008] Step S4: Each drone flies and operates within its own corridor. When the deviation between the actual position and the nominal position of the corridor exceeds the threshold, a local corridor renegotiation process is triggered. Only the affected neighboring drones negotiate with each other to adjust the corridor time window or perform altitude avoidance.

[0009] Step S5: When the remaining battery power of the drone is lower than the safe return threshold, the task is automatically reassigned. The drone releases the unfinished blocks back to the task pool and returns autonomously. The adjacent drones perform a local re-auction to inherit the task.

[0010] Preferably: the bidding vector in step S2 Represented as:

[0011] (1)

[0012] Where ϕE(i,j) is the energy prediction margin component, defined as:

[0013] (2)

[0014] in To estimate the total power consumption throughout the process; For safe return to base, Given a fully charged battery, φ_D(i,j) represents the wind direction correction flight cost component. The component lost due to work interruption. This represents the path segment conflict cost component. These are the corresponding weighting coefficients.

[0015] Preferably: the corridor section in step S3 Described using a three-dimensional convex hull time interval:

[0016]

[0017] in Based on the start and end points of the path segment and Central axis, safety isolation radius The capsule body, For the projected flight time window of this segment, the corridor conflict cost is defined as the product of the duration of temporal overlap and the spatial penetration depth:

[0018] (3).

[0019] Preferably, the local corridor renegotiation process in step S4 includes: broadcasting a deviation notification message to the drone i and identifying the set of affected neighbors. For each affected neighbor, both parties calculate the concession cost. :

[0020] (4)

[0021] in , , As weight, The total job delay caused by time shifting, The extra power consumption is due to the detour. The percentage of the current block completed is used as the threshold; the party with the lower cost will either have its corridor time window moved back or its height layer avoided.

[0022] A multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning system for farmland operations, used to execute the methods described above, includes: a ground control station for inputting the coordinates of the operation area and the operation type, dividing the operation into blocks, and monitoring cluster performance indicators in real time; an airborne module deployed on each UAV, including: a DEATA bidding calculation unit for calculating multi-dimensional bidding vectors and participating in distributed auction iterations; a CAST-DP corridor planning unit for generating and managing four-dimensional spatiotemporal corridors; a local renegotiation unit for detecting corridor deviations and performing pairwise negotiations with neighboring UAVs; a power warning and task reallocation unit for monitoring remaining power and triggering task release and re-auction; and an inter-UAV communication network for supporting direct exchange of bidding messages, corridor descriptions, and negotiation information between UAVs.

[0023] Preferably, the DEATA bidding calculation unit and the CAST-DP corridor planning unit have a coupling interface: the bidding message output by the DEATA unit contains path segment polygons, and the CAST-DP unit directly reuses this information to generate an initial corridor segment sequence; when the CAST-DP unit detects a corridor deviation caused by abnormal power levels, it triggers the DEATA unit to perform a local re-auction.

[0024] Compared with the prior art, the beneficial effects of this invention are as follows:

[0025] (1) Significantly improved operational continuity: DEATA energy-aware allocation enables precise matching of block allocation with the actual endurance of each aircraft. In the 8-aircraft 400ha simulation, the mission completion rate (without mid-flight return interruption) increased from 84.7% to 97.2% in the baseline method without energy awareness, and the number of mid-flight return interruptions decreased from 3.0 times / hour to 0.22 times / hour (a reduction of 93%), corresponding to a reduction of approximately 18% in the total cluster operation time (avoiding extra time in the area of ​​flight interruption). The "boundary effect" of the components ensures that infeasible allocations (unable to return safely) are automatically excluded in the auction without the need for additional feasibility constraints, making the design simple and effective.

[0026] (2) Fast task allocation and support for real-time reallocation: DEATA distributed iteration converges on average in 6.7±1.8 rounds (comparable to 7.1±2.3 rounds of standard CBBA, and the addition of four-component bidding does not significantly affect the convergence speed). The total allocation delay is about 335ms, which is about 4.2 times faster than centralized integer programming (1412ms). It meets the real-time reallocation requirements (the local re-auction is triggered on average 4.2 times / hour throughout the process, all of which are completed automatically without manual intervention). The coupling interface between DEATA and CAST-DP (the bidding message contains the path segment polygon) eliminates the repeated path calculation in the initial planning stage of CAST-DP, further reducing the total system startup time.

[0027] (3) The corridor replanning response is fast and has little impact on the cluster: the average response time of CAST-DP local negotiation is 143ms (maximum 217ms), which is about 1 / 12 of the global replanning scheme (IP-GPP: 1847ms, CBBA-GPP: 1623ms). At the cluster's flight speed of 7m / s, the movement is about 1m within 143ms, with sufficient safety margin. Each renegotiation involves an average of 1.8 UAVs (global replanning involves all 8 UAVs), which has an impact of about 0.7% on the overall operation efficiency of the cluster. The minimum MSD value throughout the process is 4.7m in simulation and 5.2m in actual measurement, both of which are higher than the safe isolation distance of 3m, and there are no collision events.

[0028] (4) Real-world test to verify reliability: In an 8-aircraft formation test in a 40-ha sub-area of ​​rice planting base in Heilongjiang Province, the average deviation of the flight trajectory from the planned corridor was 0.82m (1-sigma), and the maximum deviation was 2.1m (successfully triggered renegotiation recovery); the operation coverage rate was 99.6%, the spraying overlap rate was 4.8% (meeting the design target of ≤5%), there were no collision events throughout the process, and the number of operation interruptions was 0 (compared to the historical record of about 0.3 times / hour for a single aircraft), which verified the effectiveness and reliability of the system in a real agricultural environment. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention;

[0030] Figure 2 This is a diagram of the overall system architecture.

[0031] Figure 3 This is a schematic diagram of the DEATA four-component bidding vector;

[0032] Figure 4 A graph showing the convergence process and performance comparison of DEATA distributed auction;

[0033] Figure 5 This is a schematic diagram illustrating the CAST-DP corridor description and overlap detection principle.

[0034] Figure 6A sequence diagram of the local corridor renegotiation process;

[0035] Figure 7 Comparison of renegotiation response time and full MSD timing curve. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Example

[0038] Please see Figures 1-7 The diagram illustrates a multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning method for farmland operations. This method operates in a two-stage cycle: pre-operation preparation and operation execution, requiring no central scheduling node. Specifically, it includes:

[0039] Step 1: Task preprocessing and UAV status awareness. The operator inputs the GIS coordinates of the work area and the task type (spraying / inspection / surveying) at the ground control station. The system then processes the data based on the UAV specifications (battery capacity). Spray width Work speed The system automatically calculates the maximum flight distance for each plot of farmland, dividing it into M rectangular task strips. The strips are preferentially aligned with the ridge direction to minimize turning. Adjacent strips maintain a 1.5 times spray width overlap to ensure complete coverage, forming a task pool. After each drone is powered on and performs a self-test, it will periodically... Broadcast state vector to the system ,in The current three-dimensional position, The remaining battery capacity (mAh) is displayed in real time. This represents the wind speed and direction measurements at the current location. The current block number is 0 (0 indicates idle). Inter-machine communication supports two modes: 802.11p (communication range of about 500m, single hop delay of about 5ms) and 4GLTE (delay of about 20 to 50ms). Bidding messages are about 200 bytes each. The total communication time for all 8 drones to broadcast in one round is about 0.8ms. The communication overhead is negligible.

[0040] Step two, Distributed Energy Sensing Task Auction (DEATA): When there are unallocated blocks in the task pool, a DEATA distributed auction is executed. Each UAV independently calculates the allocation of all unallocated blocks. Bidding vector The vector is composed of a weighted combination of four components:

[0041] (1)

[0042] The first component, ϕE(i,j), is the energy prediction margin component, a core innovative component of DEATA, reflecting the completion of the block by UAV i. The remaining battery capacity after the entire journey (to the starting point, performing the operation, and returning to the nearest charging point) and safe return is defined in normalized form:

[0043] (2)

[0044] in To predict power consumption throughout the entire process, a flight dynamics power model was used. Calculate the power consumption of each section ( The drag coefficient, (Lift coefficient, all specified in the UAV specifications), and corrected for wind speed and direction (wind speed correction factor). , (Fitted from measured data) For safe return to base, Fully charged This indicates that there is sufficient energy and the mission can be completed safely. If a drone cannot safely return after completing a task, its allocation is automatically eliminated in the auction without further feasibility constraints, naturally creating an energy load balancing effect—drones with ample energy will bid higher when competing for the same task, prioritizing high-power-consuming segments. For example, if drone i has 3200mAh remaining, 800mAh for a safe return, 5000mAh on a full charge, and consumes 2350mAh to complete a certain segment, then… The bidding contribution was extremely low; while the other drone with a remaining 4800mAh battery... With a significant competitive advantage in energy, the auction results naturally allocated the block to drones with sufficient power, effectively avoiding mid-journey return interruptions caused by energy prediction errors.

[0045] The second component φ_D(i,j) is the wind direction correction flight cost component, which corrects the flight path of drone i to the current wind speed and direction. The effective flight cost at the starting point decreases with a tailwind and increases with a headwind: Equivalent flight distance , The flight direction angle, The current wind direction angle, third component For the loss component due to job interruption: if drone i is currently executing a block Switch to The loss of the completed work is ( This represents the percentage of the current block that has been completed. for (Relative priority); Idle time The fourth component For the path segment conflict cost component: taking the drone i flying to The sum of the collision probabilities of the path segment and the path segments already announced by other machines measures the risk of path conflict. This reduces the bidding price of blocks with high path conflict risk, creating a path separation effect and adjusting the weight of each component. Determined by ablation experiments (energy component is the most important).

[0046] Once the bidding package is determined, a distributed iterative protocol is executed using an improved consensus bundle algorithm (CBBA improved version): in each iteration, each drone starts from the highest bid. Select the local bidding packet from each block and broadcast a bidding message containing the machine ID, block ID, bidding vector, estimated arrival time, and path segment polygon; upon receiving bidding messages from other machines, if the other machine's bid is higher, release the block and simultaneously update the path conflict cost incrementally (only recalculating bidding components that conflict with the updated path segment, approximately 0.5 to 2 ms); continuously When the local bidding package remains unchanged during rounds of iteration, the local allocation converges. When all drones converge, the global allocation is stable. In an 8-drone 400ha simulation, DEATA converges on average after 6.7±1.8 rounds of iteration (corresponding to an allocation delay of about 335ms), which is comparable to the convergence speed of the standard CBBA in 7.1±2.3 rounds. This indicates that adding four-component bidding does not significantly affect the convergence speed, but significantly improves the allocation quality. When any drone i detects that its remaining battery is lower than the safe return battery plus 5% margin during execution, it immediately broadcasts a task release message, re-enters the incomplete block into the pool, and sends a local re-auction request to drones within a 2-hop range. Only adjacent drones participate (without disturbing the normal operation of the remote drones). Local convergence usually takes 2 to 4 rounds (about 100 to 200ms).

[0047] Step 3, Initial Planning of the 4D Spacetime Corridor (CAST-DP): After receiving the DEATA allocation scheme, CAST-DP plans a dedicated flight corridor for each UAV in 4D spacetime (x,y,z,t). The corridor consists of several continuous corridor segments, each of which... Described using a three-dimensional convex hull time interval: ,in For capsule body (based on the start and end points of the path segment) and Central axis, safety isolation radius (sweeping shape) For the estimated flight time window of this segment, the capsule-based GJK nearest point detection has an analytical form (no iteration required), with a single detection time of approximately 0.005 ms, about 4 times faster than the general GJK. The false alarm collision rate introduced by the capsule approximation is verified to be less than 2.3% through simulation, which is within an acceptable range of safety margin. CAST-DP directly reuses the "estimated arrival time" and "path segment polygon" from the DEATA bidding message to generate the initial corridor segment sequence, eliminating the step of repeatedly calculating the path during the corridor planning stage.

[0048] After each drone broadcasts its complete corridor description, the system performs full corridor overlap detection, checking corridor segments for overlap. A two-level fast filtering method is used: the first level is time detection, which directly determines no conflict if the time intervals do not overlap (approximately 70% of segment pairs are filtered out here); the second level is spatial detection, which calculates the result using the GJK algorithm when the time intervals overlap. and nearest point distance There is no conflict. Corridor conflict, the cost of which is defined as the product of the duration of temporal overlap and the depth of spatial penetration:

[0049] (3)

[0050] Corridor planning aims to minimize the total conflict cost across the entire cluster. To optimize objectives, for corridor pairs where conflicts are detected, the lower-priority drones (those with lower total DEATA bids) will have their conflict time window shifted backwards. (Minimum separation time, corresponding to a flight distance of approximately 35m, covering GPS error), repeated detection until no conflict, full detection of 8 aircraft scenarios (approximately 6400 capsule body detections) takes approximately 32ms, requiring an average of 2 to 3 rounds of time translation iteration, with a total corridor planning time of approximately 800 to 1200ms. After planning is completed, each aircraft obtains a complete corridor description without conflict, and is ready for takeoff. A typical corridor includes: 1 to 2 starting transition sections, approximately 20 to 30 operation sections (each line of operation is a corridor section), several inter-block transfer sections, and 1 return section. The time window for each section is calculated based on flight speed and section length, with a turn time (approximately 2 to 4 seconds) reserved between adjacent sections.

[0051] Step four: Corridor execution and dynamic local renegotiation. Each UAV strictly flies within its own corridor to perform operations. The onboard renegotiation module detects its own actual position in real time at a 100ms cycle. relative to the corridor's nominal location The deviation, when Exceed The local corridor renegotiation process is triggered at times, and the reasons for deviation are divided into four categories: (a) sudden change in wind speed (breakthrough) (a) Threshold); (b) Avoid sudden obstacles (field machinery, birds, etc.); (c) Power consumption drops rapidly (exceeding the predicted power consumption rate by 20%, which triggers DEATA local re-auction to ensure that task allocation and corridor adjustment are synchronized); (d) Sensor disturbances cause attitude deviation, of which (a) and (b) trigger high-priority renegotiation, and (c) and (d) trigger low-priority renegotiation.

[0052] Drone i broadcasts a deviation notification message (including its current location, deviation amount, and predicted new trajectory interval); only when the corridor segment of another drone j potentially overlaps with the new trajectory interval of i (overlapping time intervals and spatial distance less than 9m) is drone j included in the set of participants for this renegotiation. Unaffected drones will not participate in the renegotiation and will continue to operate normally. Drone i sends a corridor concession request to drone j (including the type of deviation reason, the magnitude of the deviation direction, and the expected negotiation solution); drone j calculates its own concession cost. The cost of concessions to i The party with the lower cost will make a concession—the party with the lower cost will either postpone its own corridor time window or implement a height-level avoidance (moving it up by r_h=5m):

[0053] (4)

[0054] in =0.4 (time delay weight) =0.4 (weight of power consumption during detours) =0.2 (interruption loss weight), all determined by ablation experiments; The total job delay caused by time shifting, The extra power consumption is due to the detour. The current segment completion percentage (representing the switching loss) is used as the basis for determining the completion rate. After the negotiation result is confirmed, both parties update their own corridor segment descriptions and broadcast the corridor update message. Other drones, upon receiving this message, only re-inspect segments related to the updated corridor (rather than performing full re-inspection) and perform CRC32 hash verification to ensure global consistency of the corridor database. When the negotiation rounds exceed [a certain number], [the process continues]. If conflicts still occur in a round, an arbitration mechanism is implemented: the drone with the smaller number remains in its corridor, while the drone with the larger number performs altitude layer separation (moves up 10m and flies to the end of the conflict area), ensuring convergence within a limited number of rounds. The typical renegotiation sequence is as follows: T=0ms detect deviation → T=5ms broadcast deviation notification → T=10ms neighbor identification → T=15ms send concession request → T=25ms receive concession response → T=30ms broadcast corridor update → T=40ms complete local conflict re-detection and hash verification → T=143ms (including communication round-trip delay) renegotiation is completed. The entire process involves an average of only 1.8 drones, and the remaining drones in the cluster continue to execute the original corridor without being aware of it.

[0055] Step 5: Battery warning and automatic task reallocation. The ground monitoring module updates the real-time performance indicators of the entire cluster every 500ms: cluster operation coverage progress (completed area / total area), and estimated remaining flight time for each UAV (based on current data). And real-time power consumption), nearest proximity distance MSD (minimum instantaneous distance of the entire cluster, alarm threshold 6m, emergency braking threshold 3m), corridor deviation event frequency (number of renegotiations every 10 minutes), task completion rate (the proportion of completed blocks without interruption), and the remaining battery power of drone i. Below safe return-to-home battery level add When there is an additional margin, the system triggers a task interruption: UAV i broadcasts a task release message, adds its incomplete blocks back to the task pool, and notifies UAVs within two adjacent hops to perform local redistribution (2 to 4 iterations, completed in about 100 to 200 ms); UAV i autonomously returns to charge according to the pre-stored shortest path, and after charging, it rebroadcasts its full charge status and participates in the bidding for subsequent blocks to continue working. The system continuously loops the above steps two to five until all blocks are completed. The entire process of charging and task inheritance is transparent to the ground operator and requires no manual intervention. In actual testing, local redistribution was triggered 4.2 times / hour, and the task completion rate (without mid-way return interruption) reached 97.2%, which is 12.5 percentage points higher than the baseline method without energy awareness (task completion rate of 84.7%).

[0056] The effectiveness of the method of the present invention will be verified below with reference to specific implementation methods. In this embodiment, approximately 400 hectares of farmland and 8 drones in a rice-growing area of ​​Heilongjiang Province are used as demonstration scenarios (simulation: Gazebo+PX4 flight control simulation; actual test: DJI Agras T40 improved version, airborne Jetson Nano 4G, 4G LTE communication).

[0057] In the pre-operation preparation phase, step one is task segmentation. The operator inputs the GIS coordinates of the operation area (including farmland boundary polygons) and the operation type (spraying) at the ground control station. The system automatically calculates the maximum flight distance for each segment according to the UAV specifications. ( =0.8, For hovering power, To provide extra power for spraying, the farmland is divided into M rectangular strips. The direction of the strips is preferably aligned with the direction of the ridges (to reduce turns). Adjacent strips are kept with a 1.5 times spray width buffer overlap. Irregular edges are filled with triangular patches. In this example, 400 hectares are divided into M=40 strips of about 10 hectares each. The operation time for each strip is about 12 to 15 minutes.

[0058] Step 2: Powering on and broadcasting status of the drones. After powering on and performing a self-test, each drone broadcasts its status vector at a period of T_s=1s. Initial state: Fully charged (5000mAh), current location on the tarmac. (Idle), Communication mode: This embodiment uses 4G LTE (base station signal strength > -80dBm), with a single-hop latency of about 20 to 50ms. The total communication time for all 8 drones to broadcast for one round is about 0.4ms.

[0059] DEATA initial task auction execution, step three: Each drone independently calculates its bidding vector. For all M=40 unallocated blocks in the task pool, each drone calculates its bidding vector. (Formula (1)), power consumption estimate Calculation by position To strips Equivalent flight distance at the starting point (wind direction correction in formula (2)), optimal energy-efficient flight speed calculate ; based on the area of ​​the strip Work speed Spray width Calculate the time of the operation Multiply by the operating power have to The distance from the end of the block to the nearest charging station is calculated in the same way. In this embodiment, it takes about 8ms for each drone to calculate the bidding vector of 40 blocks (ARM Cortex-A72, 1.5GHz).

[0060] Step 4: Auction iterative convergence, each drone executes bidding-negotiation iteration according to the CBBA improved protocol: each round broadcasts a bidding package (including the highest bid). The bidding vector and path segment polygon of each block (approximately 600 bytes) are used to update the path conflict cost increment after receiving bidding messages from other machines (only the affected components are recalculated, approximately 1ms). The bidding size is compared to determine whether to retain or release the block. When the number of wheels remains unchanged, the machine converges. In this embodiment, it takes an average of 6.7 ± 1.8 wheels (approximately 335 ms) to reach a stable allocation, with each machine obtaining an average of 5 blocks (balanced allocation). The energy load balancing effect of the drive: the ratio of the number of allocation blocks for full-load motors to low-load motors is approximately 6:4, which is close to the energy ratio.

[0061] CAST-DP initial corridor planning execution, step five: initial corridor generation. After each UAV receives the DEATA allocation scheme (directly reusing the path segment polygon in the bidding message), it discretizes and generates a corridor segment sequence with a time slice of δt=0.5s. A typical corridor includes: 1 to 2 starting transition segments, about 20 to 30 operation segments (one segment per row), 4 inter-block transfer segments (4 transfers in 5 blocks), and 1 return segment, for a total of about 26 to 37 corridor segments. It broadcasts its own complete corridor description (about 1.2 to 1.8KB / UAV, about 10KB total broadcast for 8 UAVs).

[0062] Step Six: Full-pair corridor overlap detection and time shift resolution. Perform full-pair capsule overlap detection on 8 machines × 37 segments (approximately 6400 times, total time approximately 32ms). Conflicting corridor pairs are detected (in this embodiment, there are initially about 3 to 5 pairs). The lower priority side shifts the time window of the conflict segment to the later stage. The process is repeated until there are no conflicts (an average of 2 to 3 rounds of iteration, with a total corridor planning time of about 800 to 1200 ms). Finally, each aircraft obtains a complete corridor description without conflicts and is ready for takeoff.

[0063] During the operation execution phase, step seven: corridor execution and real-time monitoring. Each aircraft takes off and executes its assigned blocks according to its own corridor. The airborne local renegotiation module detects corridor deviation at a 100ms cycle. The ground monitoring module updates cluster performance indicators every 500ms: operation coverage progress, estimated remaining flight time for each aircraft, MSD (minimum instantaneous distance of the entire cluster, alarm threshold 6m, emergency braking threshold 3m), and corridor deviation event frequency. In this embodiment, the average deviation during actual operation was 0.82m, and the maximum deviation was 2.1m (caused by gusts of wind, which successfully triggered renegotiation and recovery).

[0064] Step 8: Local renegotiation of corridor deviation; when the corridor deviation exceeds... At the same time, identify the deviation type (sudden wind speed change / obstacle / abnormal power / sensor disturbance), broadcast deviation notification, and identify the set of affected neighbors. The corridor adjustment is performed according to the pairwise negotiation agreement (formula (4) concession cost comparison) and the CRC32 hash consistency verification is completed. The typical complete renegotiation process takes about 143ms (including about 40 to 80ms of 4GLTE communication round trip). In this embodiment, renegotiation is triggered about 11 times in total (about 2.3 times / hour), and each time it affects an average of 1.8 drones.

[0065] Step Nine: Battery Warning and Automatic Task Reassignment. When the remaining battery power of drone i is lower than the safe return battery power plus 5% margin, a task release message is immediately broadcast. Unfinished blocks (approximately 0 to 2 unfinished blocks in this embodiment) are re-entered into the pool, and a local re-auction request is sent to neighbors within a 2-hop range. Neighboring drones evaluate themselves. After deciding whether to bid, the local redistribution is usually completed in 2 to 4 iterations (about 100 to 200 ms). After the drone returns to its home base to recharge, it rebroadcasts its full charge status and participates in subsequent block bidding to continue operations. In this embodiment, a total of 4.2 local redistributions were triggered per hour, all of which were completed automatically without the need for ground human intervention, and the number of mission interruptions was 0.

[0066] Table 1 Summary of Key Algorithm Parameters

[0067]

[0068] The experimental verification results, evaluated on 8 machines over 400 hectares (Gazebo + PX4, constant bottom wind 3 m / s + random gusts 0 to 6 m / s) and field measurements (approximately 40 hectares of paddy field in Heilongjiang Province, DJI Agras T40 + Jetson Nano + 4G LTE) are as follows:

[0069] (a) Performance comparison of task allocation

[0070] The DEATA invention achieves an initial allocation time of 335ms, a task completion rate (without interruptions) of 97.2%, an average mid-journey return time of 0.22 times / hour, and a total cluster operation time reduction of 18% compared to the centralized IP-GPP benchmark. Compared to IP-GPP (task completion rate 81.3%, interruptions 3.4 times / hour, allocation time 1412ms) and CBBA-GPP (84.7%, 3.0 times / hour, 524ms), DEATA demonstrates a significant improvement in task completion rate (+12.5 percentage points vs. IP-GPP, +12.5 percentage points vs. CBBA-GPP), proving the key value of the energy-aware component phi_E.

[0071] (II) Comparison of Corridor Safety Performance

[0072] The present invention, CAST-DP, has an average renegotiation response time of 143ms (maximum 217ms), a minimum MSD of 4.7m in simulation and 5.2m in actual measurement, zero collision events, and an average impact of 1.8 drones per renegotiation. Compared with IP-GPP (global replanning 1847ms, minimum MSD 3.8m) and CBBA-GPP (1623ms, minimum MSD 3.5m), CAST-DP's response time is approximately 1 / 12 of global replanning, and the interference surface of each renegotiation on the cluster is extremely small (1.8 / 8 drones), verifying the efficiency and security of the local corridor negotiation architecture.

[0073] (III) Actual Measurement Verification

[0074] In a real 40-ha sub-area of ​​farmland, the average deviation of the flight trajectory from the planned corridor was 0.82m (1-sigma), the operation coverage was 99.6%, the spraying overlap rate was 4.8% (meeting the design target of ≤5%), and there were zero collisions and zero interruptions throughout the process. The measured minimum MSD of 5.2m was higher than the safety distance of 3m, proving the reliability of CAST-DP under real GPS noise (about 0.5m position error) and 4G LTE communication delay (20 to 50ms).

[0075] Figure 2 This is a system architecture diagram of the present invention, showing the three-layer architecture and information flow of the ground control station (mission preprocessing and monitoring), each UAV airborne module (DEATA bidding calculation, CAST-DP corridor planning, local renegotiation), and the inter-UAV communication network (802.11p / 4GLTE). The two coupling interfaces between DEATA and CAST-DP are marked (DEATA→CAST-DP: path segment announcement; CAST-DP→DEATA: power abnormality triggers local re-auction).

[0076] Figure 3 This is a diagram illustrating the four-component bid vector of DEATA. (Energy forecast margin) (Flight cost for wind direction correction) (Losses due to work interruption) The calculation logic of the four components of (path segment conflict cost) and examples of the values ​​of each component as the scene changes (including...). The load balancing diagram shows that fully charged drones are given priority in bidding for high power-consuming blocks.

[0077] Figure 4 The diagram shows the convergence process and performance comparison of DEATA distributed auction, illustrating the round-by-round evolution of the auction of 16 blocks by 8 drones from initial random allocation to stable allocation (task-drone allocation matrix iteration heatmap), as well as the comparison curves of task completion rate (without mid-course interruption) between DEATA and standard CBBA (without phi_E) under different wind speed conditions (0 to 10 m / s).

[0078] Figure 5 This diagram illustrates the principle of CAST-DP corridor description and overlap detection, showcasing the visualization of a single UAV corridor segment sequence (capsule + time window) in 3D space. It also demonstrates a two-stage rapid filtering process: interval arithmetic time detection (time axis interval overlap judgment) and GJK space detection (3D capsule nearest point distance calculation) for two corridor segments, with conflict costs clearly labeled. The geometric meaning (duration of time overlap × depth of spatial penetration).

[0079] Figure 6This is a sequence diagram of the local corridor renegotiation process, showing the complete message sequence of pairwise negotiation triggered by a sudden wind speed event (drone 3 deviating from the corridor) (at each time from T=0 to 143ms: deviation detection → broadcast notification → neighbor identification → concession request → concession response → corridor update → conflict re-detection → negotiation completion), and a comparison of the spatial layout before and after the corridor update. It also marks the time translation concession operation of drone 5 and the imperceptible execution state of drones 1 / 2 / 4 / 6 / 7 / 8.

[0080] Figure 7 The comparison of renegotiation response time and the time series curve of MSD throughout the process show the box plot comparison of the response time of CAST-DP local negotiation (about 100 to 200 ms) and IP-GPP / CBBA-GPP global replanning (about 1200 to 2800 ms) in 20 random corridor deviation events; the time series curve of MSD during the entire operation of 8 machines in 400ha simulation (marking the corridor renegotiation trigger event point and MSD recovery process, with the minimum value of 4.7m being higher than the safety distance of 3m).

[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning for farmland operations, characterized in that, Includes the following steps: Step S1: Divide the work area into multiple rectangular work blocks to form a task pool; each drone broadcasts a state vector containing real-time remaining battery power and wind speed and direction at a fixed period. Step S2: Execute the Distributed Energy Sensing Task Auction Algorithm (DEATA). Each UAV independently calculates a multi-dimensional bidding vector for unallocated blocks and performs distributed iterations through an improved consensus bundle algorithm to complete task allocation and initial path conflict resolution. The bidding vector includes at least an energy prediction margin component and a path segment conflict cost component. Step S3: Receive the DEATA allocation scheme and execute the four-dimensional spatiotemporal corridor dynamic planning algorithm CAST-DP to plan a dedicated flight corridor for each UAV in the four-dimensional spatiotemporal space composed of several corridor segments in the three-dimensional spatial coordinates and time t; the corridor segments are described by three-dimensional convex hull time intervals. Step S4: Each drone flies and operates within its own corridor. When the deviation between the actual position and the nominal position of the corridor exceeds the threshold, a local corridor renegotiation process is triggered. Only the affected neighboring drones negotiate with each other to adjust the corridor time window or perform altitude avoidance. Step S5: When the remaining battery power of the drone is lower than the safe return threshold, the task is automatically reassigned. The drone releases the unfinished blocks back to the task pool and returns autonomously. The adjacent drones perform a local re-auction to inherit the task.

2. The method for multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning for farmland operations according to claim 1, characterized in that: The bidding vector in step S2 Represented as: (1) Where ϕE(i,j) is the energy prediction margin component, defined as: (2) in To estimate the total power consumption throughout the process; For safe return to base, Given a fully charged battery, φ_D(i,j) represents the wind direction correction flight cost component. The component lost due to work interruption. This is the path segment conflict cost component. These are the corresponding weighting coefficients.

3. The method for multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning for farmland operations according to claim 2, characterized in that: The corridor section in step S3 Described using a three-dimensional convex hull time interval: in Based on the start and end points of the path segment and Central axis, safety isolation radius The capsule body, For the projected flight time window of this segment, the corridor conflict cost is defined as the product of the duration of temporal overlap and the spatial penetration depth: (3)。 4. The method for multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning for farmland operations according to claim 3, characterized in that: The local corridor renegotiation process in step S4 includes: broadcasting a deviation notification message from the drone and identifying the set of affected neighbors. For each affected neighbor, both parties calculate the concession cost. : (4) in , , As weight, The total job delay caused by time shifting, The extra power consumption is due to the detour. The percentage of the current block completed is used as the threshold; the party with the lower cost will either have its corridor time window moved back or its height layer avoided.

5. A multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning system for farmland operations, used to execute the method of any one of claims 1-4, characterized in that, include: The ground control station is used to input the coordinates of the work area and the type of work, divide the work into blocks, and monitor the cluster performance indicators in real time. The airborne module, deployed on each UAV, includes: a DEATA bidding calculation unit for calculating multi-dimensional bidding vectors and participating in distributed auction iterations; a CAST-DP corridor planning unit for generating and managing four-dimensional spatiotemporal corridors; a local renegotiation unit for detecting corridor deviations and performing pairwise negotiations with neighboring UAVs; a power warning and task reallocation unit for monitoring remaining power and triggering task release and re-auction; and an inter-UAV communication network that supports direct exchange of bidding messages, corridor descriptions, and negotiation information between UAVs.

6. A multi-UAV distributed energy sensing task auction and four-dimensional spatiotemporal corridor collaborative planning system for farmland operations according to claim 5, characterized in that: The DEATA bidding calculation unit and the CAST-DP corridor planning unit have a coupling interface: the bidding message output by the DEATA unit contains path segment polygons, and the CAST-DP unit directly reuses this information to generate the initial corridor segment sequence; when the CAST-DP unit detects a corridor deviation caused by abnormal power, it triggers the DEATA unit to perform a local re-auction.

Images