A closed-loop control method and system for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input.

By employing a tiered risk safety buffer and a full-link delay compensation mechanism, combined with a lightweight BP neural network and three-level multi-parameter collaborative control, the problem of accurate prediction and real-time control of droplet drift in agricultural drones has been solved, improving operational efficiency and safety.

CN123067548APending Publication Date: 2026-11-10SHANDONG UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611261069.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-19
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing agricultural drone droplet drift prediction models have low accuracy, statically fixed safety distances, limited spraying control methods, and poor system integration. These issues result in large droplet drift prediction errors, insufficient or excessively restrictive safety distances, low operational efficiency, and an inability to control spraying in real time, posing a risk of pesticide damage.

Method used

Employing a graded risk safety buffer mechanism, a lightweight BP neural network, and a three-level multi-parameter collaborative control strategy, combined with a full-link delay compensation mechanism, closed-loop control is achieved through ground-based graded configuration terminals, field wireless meteorological acquisition units, airborne RTK sensing modules, and centrifugal nozzle control units to dynamically adjust droplet size and spraying parameters.

Benefits of technology

It enables accurate prediction and real-time control of droplet drift, avoids the risk of pesticide damage, improves operational efficiency and safety, and adapts to differentiated protection in different risk scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123067548A_ABST
    Figure CN123067548A_ABST
Patent Text Reader

Abstract

This invention proposes a closed-loop control method and system for droplet drift of agricultural drones with end-to-end delay compensation and tiered risk input. The method includes: matching a preset safety buffer distance based on the downwind risk level of the plot; collecting real-time parameters from multiple sensors; obtaining a compensated predicted wind speed through linear extrapolation based on the pre-calibrated end-to-end total delay and real-time wind speed; predicting the cumulative droplet drift distance using a pre-trained droplet cumulative drift distance prediction model; executing a three-level multi-parameter collaborative control strategy based on the predicted cumulative droplet drift distance and tiered safety buffer distance parameters, determining the target particle size and remaining control commands, and issuing them; calculating speed control commands based on the target particle size using a particle size-PWM fitting equation, executing the speed control commands and feeding back the actual rotational speed, and correcting the speed control commands based on the feedback of the actual rotational speed to achieve closed-loop correction. This invention maximizes field operation efficiency while eliminating cross-border pesticide damage caused by droplets.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the technical field of pesticide spraying operations, and in particular to a closed-loop control method and system for droplet drift of a plant protection drone. Background Technology

[0002] In the current low-altitude, low-volume spraying mode of agricultural drones, droplets are highly susceptible to drift due to the combined effects of environmental factors such as crosswinds, rotor airflow, operating altitude, and droplet size. Droplet drift significantly reduces pesticide target deposition and utilization rates, causing pesticide damage to nearby sensitive crops, water sources, and residential areas. It also poses risks such as ecological pollution and agricultural disputes, representing a core pain point that urgently needs to be addressed in aerial agricultural spraying. Existing technologies suffer from the following problems:

[0003] (1) Low accuracy and lack of interpretability of drift prediction models: Existing drift prediction models mostly use CFD physical mechanism models, simple empirical formulas or traditional machine learning models such as SVM and random forests. They can only represent the influence of a single variable in a one-sided way and cannot fully depict the diffusion law of fog droplets under the coupling of multiple factors such as wind speed, height, particle size, temperature and humidity. The prediction error is too large. Moreover, most of the models are "black box" structures, which cannot clearly define the contribution of each parameter to drift and make it difficult to guide the optimization of field operation parameters.

[0004] (2) The safety distance is statically fixed and there is no quantitative dynamic inversion mechanism: The existing operation buffer safety distance is mostly fixed by industry experience. For example, the invention patent with publication number CN117806349B discloses a method for reducing the drift risk of plant protection drone spraying operations. The boundary distance x is only a single fixed value globally. It cannot distinguish different risk scenarios such as water sources, sensitive crops, and contiguous fields. The unified standard is either insufficient in protection or excessively restricts the efficiency of operation. There is a lack of a quantitative inversion method for safety distance based on the automatic classification of drift prediction results.

[0005] (3) The spraying control methods are simple and there is no real-time closed-loop active control: the mainstream drones adopt a fixed droplet size spraying strategy, and the operation cannot be dynamically adjusted according to the wind and boundary distance. A few variable nozzle schemes only make simple particle size switching. For example, the invention patent with publication number CN118778688A discloses a method and system for preventing droplet drift control for plant protection drones. The compensation method only defines general means such as nozzle forward movement and tilt angle compensation in a macroscopic way. It does not link the liquid flow and flight speed to control drift in sync. The risk disposal only has the binary logic of "increase the particle size / stop the machine directly". Frequent interruption of operation in windy weather seriously reduces the daily operation area. Moreover, most of them only stay at the theoretical simulation and have not built a complete hardware closed-loop system from perception to calculation to execution, so they cannot implement real-time active prevention and control.

[0006] (4) Each module is independent and isolated, the system integration is poor, and the end-to-end delay is ignored: Existing studies separate meteorological perception, drift prediction, safety decision-making and nozzle execution, and there is no integrated hardware and software solution; there is a millisecond-level system delay in the entire process of meteorological data wireless transmission, model inference and motor speed regulation, the control action is delayed when the gust suddenly changes, and the droplet cross-border pesticide damage occurs instantaneously. Existing technologies have not designed a compensation and correction mechanism for the end-to-end delay. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention proposes a closed-loop control method and system for droplet drift of agricultural drones with end-to-end delay compensation and tiered risk input. It solves the problem of statically fixed safety distances and the lack of a quantitative dynamic inversion mechanism by employing a tiered risk safety buffer mechanism and a lightweight BP neural network S90 quantification drift criterion. Furthermore, it addresses the issue of singular spraying control methods and the lack of real-time closed-loop active control by using a three-level multi-parameter collaborative control strategy and centrifugal nozzle particle size closed-loop correction. Finally, it solves the problem of ignored end-to-end delay by employing an end-to-end delay feedforward compensation mechanism. The lightweight integrated hardware and software system ultimately maximizes field operation efficiency while eliminating cross-border pesticide damage from droplets.

[0008] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0009] A closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input includes:

[0010] S1: Import the operation route, set and identify the downwind risk level of the plot, match the preset safety buffer distance according to the downwind risk level of the plot, and issue the graded safety buffer distance parameters, pre-trained droplet cumulative drift distance prediction model, and particle size-PWM fitting equation.

[0011] S2: Collects real-time meteorological data including wind speed, distance from the UAV to the downwind base boundary, real-time flight control operation parameters including droplet size, and spraying parameters;

[0012] S3: Based on the pre-calibrated total delay of the entire link and the real-time wind speed, linear extrapolation is used to perform feedforward compensation to obtain the compensated predicted wind speed;

[0013] S4: Based on the compensated predicted wind speed, other real-time meteorological data, and flight control real-time operation parameters including droplet size, the cumulative drift distance of droplets is predicted using a pre-trained cumulative drift distance prediction model.

[0014] S5: Based on the predicted cumulative drift distance of fog droplets, the distance from the UAV to the downwind basic boundary, the graded safety buffer distance parameters, and the real-time flight control operation parameters, a three-level multi-parameter collaborative control strategy is executed to determine the target particle size and the remaining control commands and issue them.

[0015] S6: Calculate the speed control command based on the target particle size using the particle size-PWM fitting equation, execute the speed control command and feed back the actual speed, and correct the speed control command based on the actual speed feedback to achieve closed-loop correction.

[0016] Furthermore, the downwind risk level of the plot includes at least Level 1 risk, Level 2 risk, and Level 3 risk; Level 1 risk includes at least water sources, residential areas, or sensitive economic crops; Level 2 risk includes at least ordinary adjacent fields; Level 3 risk includes at least contiguous farmland or wasteland; and the preset safety buffer distance is the distance from the plot boundary to the target point.

[0017] Furthermore, the pre-trained cumulative drift distance prediction model for fog droplets adopts a pre-trained lightweight BP model. The pre-trained lightweight BP model takes meteorological data, the distance from the UAV to the downwind baseline boundary, real-time flight control operation parameters including fog droplet size, and spraying parameters as inputs, and outputs 90% cumulative drift distance S90.

[0018] Furthermore, in this embodiment, the particle size-PWM fitting equation adopts a nonlinear fitting equation:

[0019] The real-time flight control parameters include at least the operating altitude. Flight speed Drug flow rate Current droplet size The spraying parameters include the number of nozzles and the type of nozzles.

[0020] Furthermore, the pre-calibrated total latency across the entire link includes at least the meteorological data acquisition latency, transmission latency, model inference time, and nozzle speed adjustment latency.

[0021] Furthermore, the linear extrapolation formula is:

[0022] Furthermore, in step S4, the accompanying SHAP algorithm calculates the contribution of each input feature to S90, quantifies the influence of each input parameter on drift, and is used to guide the optimization of field operation parameters in reverse.

[0023] Furthermore, the three-level multi-parameter coordinated regulation strategy includes:

[0024] Low risk: To maintain constant droplet size, drug flow rate, and flight velocity, ensuring target coverage;

[0025] Medium risk Calculate the minimum safe particle size at the corresponding wind speed. Increase droplet size to Simultaneously reduce the spray height by a preset ratio;

[0026] High risk: Regardless of particle size, spraying should be suspended and the area should be left in place to wait for the wind speed to decrease.

[0027] A closed-loop control system for droplet drift of an agricultural drone with end-to-end delay compensation and tiered risk input is provided to implement the aforementioned closed-loop control method for droplet drift of an agricultural drone with end-to-end delay compensation and tiered risk input. The system is characterized by comprising:

[0028] Ground-based hierarchical configuration terminal: used to import operation routes, set and identify the downwind risk level of the plot, match the preset safety buffer distance according to the downwind risk level of the plot, send hierarchical safety buffer distance parameters and pre-trained droplet cumulative drift distance prediction model to the embedded edge decision unit, and send the particle size-PWM fitting equation to the centrifugal nozzle control and execution unit.

[0029] Field wireless weather data acquisition unit: Collects real-time weather data, including wind speed, and transmits it wirelessly to the embedded edge decision unit via LoRa.

[0030] Airborne RTK perception module: used to acquire the drone's centimeter-level coordinates in real time, combine them with the pre-stored field boundary coordinates, calculate the distance from the drone to the downwind basic boundary in real time, and transmit it to the embedded edge decision unit;

[0031] Embedded edge decision unit: Used to acquire real-time flight control operation parameters, including droplet size, and spraying parameters; based on the pre-calibrated total latency of the entire link and real-time wind speed, linear extrapolation is used for feedforward compensation to obtain the compensated predicted wind speed; based on the compensated predicted wind speed, other meteorological data, and real-time flight control operation parameters, including droplet size, a pre-trained droplet cumulative drift distance prediction model is used to predict the cumulative drift distance of droplets; based on the predicted cumulative drift distance of droplets, the distance from the UAV to the downwind base boundary, the graded safety buffer distance parameters, and the real-time flight control operation parameters, a three-level multi-parameter collaborative control strategy is executed to determine the target particle size and the remaining control commands and issue them; when the distance from the boundary UAV to the downwind base boundary is identified as zero, a stop spraying command is issued to the centrifugal nozzle control execution unit;

[0032] Centrifugal nozzle control and execution unit: used to calculate speed control commands based on the target particle size using a particle size-PWM fitting equation, execute the speed control commands and feed back the actual rotational speed, and correct the speed control commands based on the actual rotational speed feedback to achieve closed-loop correction.

[0033] The beneficial effects of this invention are as follows:

[0034] A tiered risk safety buffer mechanism was proposed: abandoning the traditional management method of uniformly fixed safety distances, a three-level protection buffer standard was divided according to the types of sensitive objects in the downwind direction of the field; using the 90% cumulative drift distance S90 as a unified risk quantification benchmark, it can provide reliable protection for high-risk areas such as water sources, residential areas, and sensitive economic crops, while relaxing operation restrictions in low-risk areas such as contiguous fields and wastelands, thus balancing spraying safety and field operation efficiency.

[0035] A full-link delay feedforward compensation mechanism is proposed: the total system delay formed by the superposition of meteorological data collection, wireless data transmission, model inference calculation, and centrifugal nozzle motor speed regulation is fully calculated; the actual wind speed at the time of completion of the control action is predicted by the linear extrapolation algorithm, so as to completely eliminate the problem of control action lag under the condition of sudden wind change and avoid the risk of pesticide damage caused by instantaneous fog droplets drifting out of the operation buffer from the source.

[0036] A lightweight BP neural network coupled with S90 quantification judgment system is proposed: the lightweight BP neural network is trained based on field multi-scenario measured datasets, and the 90% cumulative drift distance S90 is selected as the unified quantification index of fog droplet drift risk; the SHAP feature attribution analysis method is used to quantify the influence weight of each operation and environmental parameter on drift diffusion. The network model can complete millisecond-level local inference on airborne embedded edge hardware, and adapt to the airborne computing power constraints of UAVs while ensuring drift prediction accuracy.

[0037] A three-level, multi-parameter coordinated control strategy was proposed: unlike the existing binary extensive control mode that only adjusts droplet size or directly stops the machine, this strategy divides the working conditions into low, medium, and high risk levels based on the comparison results of S90 and graded safety buffer distance; it simultaneously links multiple spraying parameters such as droplet size, spraying height, and liquid flow rate to jointly suppress droplet drift, and only suspends spraying when the drift risk exceeds the critical threshold, which significantly reduces the frequency of operation interruption in windy weather and increases the effective operating area per day.

[0038] A complete closed-loop system integrating software and hardware is proposed: integrating five major hardware modules, namely ground-based hierarchical configuration terminal, field wireless meteorological acquisition unit, airborne RTK sensing module, embedded edge decision unit, and centrifugal nozzle control and execution unit, to build a complete physical control link covering pre-configuration of operational risks, real-time perception of field environment, airborne edge intelligent decision-making, and precise execution and control of nozzles. The whole solution has the conditions for whole-machine integration and implementation, and is not limited to simulation theoretical verification. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of the closed-loop control method for droplet drift of agricultural drones with full-link delay compensation and graded risk input according to the present invention.

[0041] Figure 2 This is a schematic diagram of the closed-loop control system for droplet drift of agricultural drones with full-link delay compensation and graded risk input according to the present invention.

[0042] Figure 3 This is a schematic diagram of the lightweight BP model structure in an embodiment of the present invention.

[0043] Figure 4 This is a schematic diagram of the overall assembly in an embodiment of the present invention.

[0044] Among them, 1 is the embedded edge decision unit (Jetson Nano), 2 is the RTK centimeter-level positioning module, 3 is the LoRa wireless meteorological data receiving module, 4 is the brushless ESC driver (Arduino lower-level machine), 5 is the centrifugal atomizing nozzle execution unit, 6 is the ground-based graded configuration terminal, 7 is the liquid tank, 8 is the liquid flow sensor, and 9 is the independent wireless weather station. Detailed Implementation

[0045] 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.

[0046] A closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input. For example... Figure 1 As shown, the steps include:

[0047] S1: Import the operation route, set and identify the downwind risk level of the plot, match the preset safety buffer distance according to the downwind risk level of the plot, and issue the graded safety buffer distance parameters, pre-trained droplet cumulative drift distance prediction model, and particle size-PWM fitting equation.

[0048] In this embodiment, the operator imports the operational flight path and sets the downwind risk level of the plot based on the surrounding environment:

[0049] Level 1 Risk (Water Source / Residential Area / Sensitive Economic Crops): The safe buffer distance from the field boundary to the target point is Ssafe=15m;

[0050] Level 2 risk (ordinary adjacent fields): The safe buffer distance from the field boundary to the target point is Ssafe=10m;

[0051] Level 3 risk (contiguous farmland / wasteland): The safe buffer distance from the boundary of the field to the target point is Ssafe=5m;

[0052] Match a preset safety buffer distance based on the downwind risk level of the plot;

[0053] In this embodiment of the application, the pre-trained droplet cumulative drift distance prediction model adopts a pre-trained lightweight BP model, such as... Figure 3 As shown, the pre-trained lightweight BP model takes meteorological data, the distance from the UAV to the downwind base boundary, real-time flight control parameters including droplet size, and spraying parameters as inputs, and outputs 90% cumulative drift distance S90.

[0054] In this embodiment of the application, the particle size-PWM fitting equation is:

[0055] ;

[0056] in, Target droplet size , PWM signal pulse width , In this embodiment of the application, as fitting parameters, , c=1247.81.

[0057] Current technologies generally employ uniform and fixed safety distances, which are insufficient in high-risk scenarios and can easily lead to pesticide-related disputes, while in low-risk scenarios, they excessively restrict the operating range and reduce efficiency. Tiered configuration can match differentiated protection standards to different risk scenarios, avoiding unnecessary operational restrictions while ensuring safety, and achieving a balance between safety and efficiency.

[0058] S2: Collects real-time meteorological data including wind speed, distance from the UAV to the downwind base boundary, real-time flight control operation parameters including droplet size, and spraying parameters;

[0059] In this embodiment of the application, the meteorological data includes at least lateral wind speed, ambient temperature, ambient humidity, and wind direction angle.

[0060] In this embodiment, the real-time flight control parameters include at least the operating altitude, flight speed, liquid flow rate, and current droplet size. .

[0061] In this embodiment of the application, the spraying parameters include the number of nozzles and the type of nozzle.

[0062] S3: Based on the pre-calibrated total delay of the entire link and the real-time wind speed, linear extrapolation is used to perform feedforward compensation to obtain the compensated predicted wind speed;

[0063] In this embodiment of the application, the pre-calibrated total latency of the entire link includes at least the meteorological acquisition latency, transmission latency, model inference time, and nozzle speed adjustment latency;

[0064] In this embodiment of the application, the linear extrapolation formula is:

[0065]

[0066] in, To compensate for the predicted wind speed, for Crosswind speed at all times, The sampling period is This represents the total latency across the entire link.

[0067] Existing technologies do not consider end-to-end system latency. When gusts suddenly change, control actions lag behind wind speed changes, making it highly susceptible to instantaneous fog droplet drift out of bounds. By predicting future wind speed through linear extrapolation and issuing commands in advance, end-to-end system latency can be offset at its source, avoiding the risk of short-term drift caused by gusts. At the same time, linear extrapolation requires very little computation, consuming almost no onboard computing power, making it very suitable for embedded real-time scenarios.

[0068] S4: Based on the compensated predicted wind speed, other meteorological data, and real-time flight control operation parameters including droplet size, the cumulative drift distance of droplets is predicted using a pre-trained cumulative drift distance prediction model.

[0069] In the embodiments of this application, to compensate for predicted wind speed Ambient temperature Ambient humidity Wind direction angle and working height Flight speed Drug flow rate Current droplet size The system takes the number and type of nozzles as inputs and the cumulative droplet drift distance S90 as output, using the pre-trained lightweight BP model for prediction. Simultaneously, the SHAP algorithm can be used to calculate the contribution of each input feature to S90, quantifying the influence of each parameter on drift, and providing reverse guidance for optimizing field operation parameters. The cumulative droplet drift distance S90 represents the distance from the downwind side of the spraying area when the droplet deposition reaches 90% of the total deposition. It is a common core indicator in plant protection for measuring droplet drift range, representing the diffusion boundary of the vast majority of droplets.

[0070] Compared to traditional CFD physics models, which are slow to compute and have low accuracy due to empirical formulas, lightweight BP neural networks can achieve millisecond-level inference on Jetson Nano embedded devices, meeting real-time requirements. Ten feature inputs can fully characterize the drift patterns coupled by multiple factors such as wind speed, altitude, and particle size, resulting in higher prediction accuracy. Using the 90% cumulative drift distance S90 as a unified risk quantification index, the minimum safe droplet size under various wind speeds can be inverted, without requiring massive historical operational data or relying on graph convolution / knowledge graphs, thus exhibiting stronger generalization capabilities. SHAP interpretability ensures that the model is no longer a "black box," allowing it to guide the optimization of operational parameters and enhance its practical engineering value.

[0071] S5: Based on the predicted cumulative drift distance of fog droplets, the distance from the UAV to the downwind basic boundary, the graded safety buffer distance parameters, and the real-time flight control operation parameters, a three-level multi-parameter collaborative control strategy is executed to determine the target particle size and the remaining control commands and issue them.

[0072] In this embodiment of the application, the three-level multi-parameter coordinated control strategy includes:

[0073] Low risk: To maintain constant droplet size, drug flow rate, and flight velocity, ensuring target coverage;

[0074] Medium risk Calculate the minimum safe particle size at the corresponding wind speed. ( (The solution is obtained by inversion using a pre-trained droplet cumulative drift distance prediction model). In this embodiment, it is... Increase droplet size to Simultaneously reduce the spraying height by 10%, while keeping other control parameters unchanged;

[0075] High risk: Regardless of particle size, spraying should be suspended and the area should be left in place to wait for the wind speed to decrease.

[0076] The final control commands, including the target particle size, were determined. (unchanged or) ), height adjustment, liquid flow rate, and stop / re-spray command.

[0077] S6: Calculate the speed control command based on the target particle size using the particle size-PWM fitting equation, execute the speed control command and feed back the actual speed, and correct the speed control command based on the actual speed feedback to achieve closed-loop correction;

[0078] In this embodiment of the application, the specific process includes:

[0079] Target particle size D targetSubstitute the particle size-PWM fitting equation to calculate the corresponding PWM signal pulse width y. Adjust the actual particle size according to the PWM signal. If the relative error between the actual particle size and the target particle size is >5%, correct the PWM pulse width through the PID closed-loop control algorithm to form closed-loop control.

[0080] S7: When the distance between the boundary drone and the downwind base boundary is detected to be zero, a stop spraying command is issued. The system stops spraying, and the current operation process ends.

[0081] A closed-loop control system for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input, such as... Figure 2 As shown, it includes:

[0082] Ground-based hierarchical configuration terminal 6: Used to import operation routes, set and identify the downwind risk level of the plot, match the preset safety buffer distance according to the downwind risk level of the plot, send hierarchical safety buffer distance parameters and pre-trained droplet cumulative drift distance prediction model to the embedded edge decision unit, and send the particle size-PWM fitting equation to the centrifugal nozzle control and execution unit.

[0083] Field wireless meteorological data acquisition unit (LoRa wireless meteorological data receiving module 3): collects real-time meteorological data including wind speed and transmits it wirelessly to the embedded edge decision unit via LoRa; in this embodiment, an ultrasonic integrated meteorological station (485 type) is selected as the meteorological data acquisition unit of the system; an STM32F407 development board is selected as the main controller of the meteorological station; and the ATK-LORA-01 LoRa module of Zhengdian Atom is selected for meteorological data transmission and reception.

[0084] Airborne RTK perception module (RTK centimeter-level positioning module 2): used to acquire the centimeter-level coordinates of the UAV in real time, combine them with the pre-stored field boundary coordinates, calculate the distance from the UAV to the downwind basic boundary in real time, and transmit it to the embedded edge decision unit.

[0085] Embedded edge decision unit 1: This unit acquires real-time flight control parameters, including droplet size, and spraying parameters. Based on the pre-calibrated total latency of the entire link and real-time wind speed, it uses linear extrapolation for feedforward compensation to obtain the compensated predicted wind speed. Based on the compensated predicted wind speed, other meteorological data, and real-time flight control parameters, including droplet size, it uses a pre-trained droplet cumulative drift distance prediction model to predict the cumulative drift distance of droplets. Based on the predicted cumulative drift distance of droplets, the distance from the UAV to the downwind base boundary, the graded safety buffer distance parameters, and the real-time flight control parameters, it executes a three-level multi-parameter collaborative control strategy to determine the target particle size and remaining control commands and issues them. When the distance from the boundary UAV to the downwind base boundary is identified as zero, a stop spraying command is issued to the centrifugal nozzle control execution unit. In this embodiment, the NVIDIA Jetson Nano B01 is selected as the computing and control center. This development board uses a quad-core ARM processor, integrates a 128-core GPU, and is equipped with 4GB of memory. It can support lightweight neural network models running locally and connects to a LoRa wireless meteorological data receiving module to receive real-time meteorological data.

[0086] The centrifugal nozzle control and execution unit (brushless ESC drive 4 + centrifugal atomizing nozzle execution unit 5) is used to calculate the speed control command based on the target particle size using a particle size-PWM fitting equation, execute the speed control command and provide feedback on the actual rotation speed, and correct the speed control command based on the feedback actual rotation speed to achieve closed-loop correction. In this embodiment, an Arduino Nano is selected as the main control development board of the control unit, which can meet the operation requirements of the PWM control program in this study; a Hobbywing Lotte 40A brushless electronic speed controller is selected as the speed control drive component of the centrifugal nozzle control unit; and a general-purpose C2808 brushless atomizing centrifugal nozzle for agricultural plant protection drones is selected as the droplet particle size execution component. By adjusting the speed of the brushless motor, the liquid breaking process is changed, thereby achieving droplet particle size adjustment.

[0087] In this embodiment, to verify the application effect of the fog droplet drift risk decision and control system constructed in this study under actual operating conditions and to conduct subsequent field trials, the DJI T30 agricultural drone was selected as the test platform. This model has stable flight performance and good platform integration, meeting the system requirements. A schematic diagram of the overall assembly is shown below. Figure 4 As shown.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input, characterized in that, include: S1: Import the operation route, set and identify the downwind risk level of the plot, match the preset safety buffer distance according to the downwind risk level of the plot, and issue the graded safety buffer distance parameters, pre-trained droplet cumulative drift distance prediction model, and particle size-PWM fitting equation. S2: Collects real-time meteorological data including wind speed, distance from the UAV to the downwind base boundary, real-time flight control parameters including droplet size, and spraying parameters; S3: Based on the pre-calibrated total delay of the entire link and the real-time wind speed, linear extrapolation is used to perform feedforward compensation to obtain the compensated predicted wind speed; S4: Based on the compensated predicted wind speed, other real-time meteorological data, and flight control real-time operation parameters including droplet size, the cumulative drift distance of droplets is predicted using a pre-trained cumulative drift distance prediction model. S5: Based on the predicted cumulative drift distance of fog droplets, the distance from the UAV to the downwind basic boundary, the graded safety buffer distance parameters, and the real-time flight control operation parameters, a three-level multi-parameter collaborative control strategy is executed to determine the target particle size and the remaining control commands and issue them. S6: Calculate the speed control command based on the target particle size using the particle size-PWM fitting equation, execute the speed control command and feed back the actual speed, and correct the speed control command based on the actual speed feedback to achieve closed-loop correction.

2. The closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input as described in claim 1, characterized in that, The downwind risk level of the land parcel includes at least Level 1, Level 2, and Level 3 risks; Level 1 risks include at least water sources, residential areas, or sensitive economic crops; Level 2 risks include at least adjacent ordinary farmland; and Level 3 risks include at least contiguous farmland or wasteland. A preset safety buffer distance is provided. This represents the distance from the field boundary to the target point.

3. The closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input as described in claim 2, characterized in that, The pre-trained cumulative drift distance prediction model for fog droplets adopts a pre-trained lightweight BP model. The pre-trained lightweight BP model takes meteorological data, the distance from the UAV to the downwind baseline boundary, real-time flight control parameters including fog droplet size, and spraying parameters as inputs, and outputs 90% cumulative drift distance S90.

4. The closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input according to any one of claims 1-3, characterized in that, In this embodiment, the particle size-PWM fitting equation is a nonlinear fitting equation: ; in, For the target droplet size, for Signal pulse width, These are the fitting parameters.

5. The closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input as described in claim 4, characterized in that, The meteorological data includes at least crosswind speed, ambient temperature, ambient humidity, and wind direction angle; The real-time flight control parameters include at least the operating altitude, flight speed, liquid flow rate, and current droplet size. The spraying parameters include the number of nozzles and the type of nozzles.

6. The closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input as described in claim 5, characterized in that, The pre-calibrated total latency across the entire link includes at least the meteorological data acquisition latency, transmission latency, model inference time, and nozzle speed adjustment latency.

7. The closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input as described in claim 6, characterized in that, The linear extrapolation formula is: ; in, To compensate for the predicted wind speed, for Crosswind speed at all times, The sampling period is This represents the total latency across the entire link.

8. The closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input as described in claim 7, characterized in that, In step S4, the accompanying SHAP algorithm calculates the contribution of each input feature to S90, quantifies the influence of each input parameter on drift, and is used to guide the optimization of field operation parameters in reverse.

9. The closed-loop control method for droplet drift of agricultural drones with end-to-end delay compensation and graded risk input as described in claim 8, characterized in that, The three-level multi-parameter coordinated regulation strategy includes: Low risk: , To ensure a safe buffer distance, droplet size, drug flow rate, and flight speed are kept constant to guarantee target coverage. Medium risk Calculate the minimum safe particle size at the corresponding wind speed. Increase droplet size to Simultaneously reduce the spray height by a preset ratio; High risk: Regardless of particle size, spraying should be suspended and the area should be left in place to wait for the wind speed to decrease.

10. A closed-loop control system for droplet drift of an agricultural drone with end-to-end delay compensation and graded risk input, used to implement the closed-loop control method for droplet drift of an agricultural drone with end-to-end delay compensation and graded risk input as described in any one of claims 1-9, characterized in that, include: Ground-based hierarchical configuration terminal: used to import operation routes, set and identify the downwind risk level of the plot, match the preset safety buffer distance according to the downwind risk level of the plot, send hierarchical safety buffer distance parameters and pre-trained droplet cumulative drift distance prediction model to the embedded edge decision unit, and send the particle size-PWM fitting equation to the centrifugal nozzle control and execution unit. Field wireless weather data acquisition unit: Collects real-time weather data, including wind speed, and transmits it wirelessly to the embedded edge decision unit via LoRa. Airborne RTK perception module: used to acquire the drone's centimeter-level coordinates in real time, combine them with the pre-stored field boundary coordinates, calculate the distance from the drone to the downwind basic boundary in real time, and transmit it to the embedded edge decision unit; Embedded edge decision unit: used to acquire real-time flight control operation parameters, including droplet size, and spraying parameters. Based on the pre-calibrated total delay of the entire link and the real-time wind speed, linear extrapolation is used for feedforward compensation to obtain the compensated predicted wind speed. Based on the compensated predicted wind speed, other meteorological data, and real-time flight control parameters including droplet size, the cumulative drift distance of the droplets is predicted using a pre-trained cumulative drift distance prediction model. Based on the predicted cumulative drift distance of droplets, the distance between the UAV and the downwind base boundary, the graded safety buffer distance parameters, and the real-time flight control operation parameters, a three-level multi-parameter collaborative control strategy is implemented to determine the target particle size and the remaining control commands and issue them; when it is detected that the distance between the boundary UAV and the downwind base boundary is zero, a stop spraying command is issued to the centrifugal nozzle control execution unit. Centrifugal nozzle control and execution unit: used to calculate speed control commands based on the target particle size using a particle size-PWM fitting equation, execute the speed control commands and feed back the actual rotational speed, and correct the speed control commands based on the actual rotational speed feedback to achieve closed-loop correction.

Citation Information

Patent Citations

  • A method for reducing drift risk in plant protection drone spraying operations

    CN117806349B

  • Anti-fogdrop drifting regulation and control method and system for plant protection unmanned aerial vehicle

    CN118778688A